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 },
748 * trace_parser_get_init - gets the buffer for trace parser
750 int trace_parser_get_init(struct trace_parser *parser, int size)
752 memset(parser, 0, sizeof(*parser));
754 parser->buffer = kmalloc(size, GFP_KERNEL);
763 * trace_parser_put - frees the buffer for trace parser
765 void trace_parser_put(struct trace_parser *parser)
767 kfree(parser->buffer);
771 * trace_get_user - reads the user input string separated by space
772 * (matched by isspace(ch))
774 * For each string found the 'struct trace_parser' is updated,
775 * and the function returns.
777 * Returns number of bytes read.
779 * See kernel/trace/trace.h for 'struct trace_parser' details.
781 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
782 size_t cnt, loff_t *ppos)
789 trace_parser_clear(parser);
791 ret = get_user(ch, ubuf++);
799 * The parser is not finished with the last write,
800 * continue reading the user input without skipping spaces.
803 /* skip white space */
804 while (cnt && isspace(ch)) {
805 ret = get_user(ch, ubuf++);
812 /* only spaces were written */
822 /* read the non-space input */
823 while (cnt && !isspace(ch)) {
824 if (parser->idx < parser->size - 1)
825 parser->buffer[parser->idx++] = ch;
830 ret = get_user(ch, ubuf++);
837 /* We either got finished input or we have to wait for another call. */
839 parser->buffer[parser->idx] = 0;
840 parser->cont = false;
843 parser->buffer[parser->idx++] = ch;
853 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
861 if (s->len <= s->readpos)
864 len = s->len - s->readpos;
867 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
877 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
881 if (s->len <= s->readpos)
884 len = s->len - s->readpos;
887 memcpy(buf, s->buffer + s->readpos, cnt);
894 * ftrace_max_lock is used to protect the swapping of buffers
895 * when taking a max snapshot. The buffers themselves are
896 * protected by per_cpu spinlocks. But the action of the swap
897 * needs its own lock.
899 * This is defined as a arch_spinlock_t in order to help
900 * with performance when lockdep debugging is enabled.
902 * It is also used in other places outside the update_max_tr
903 * so it needs to be defined outside of the
904 * CONFIG_TRACER_MAX_TRACE.
906 static arch_spinlock_t ftrace_max_lock =
907 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
909 unsigned long __read_mostly tracing_thresh;
911 #ifdef CONFIG_TRACER_MAX_TRACE
912 unsigned long __read_mostly tracing_max_latency;
915 * Copy the new maximum trace into the separate maximum-trace
916 * structure. (this way the maximum trace is permanently saved,
917 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
920 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
922 struct trace_buffer *trace_buf = &tr->trace_buffer;
923 struct trace_buffer *max_buf = &tr->max_buffer;
924 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
925 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
928 max_buf->time_start = data->preempt_timestamp;
930 max_data->saved_latency = tracing_max_latency;
931 max_data->critical_start = data->critical_start;
932 max_data->critical_end = data->critical_end;
934 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
935 max_data->pid = tsk->pid;
937 * If tsk == current, then use current_uid(), as that does not use
938 * RCU. The irq tracer can be called out of RCU scope.
941 max_data->uid = current_uid();
943 max_data->uid = task_uid(tsk);
945 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
946 max_data->policy = tsk->policy;
947 max_data->rt_priority = tsk->rt_priority;
949 /* record this tasks comm */
950 tracing_record_cmdline(tsk);
954 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
956 * @tsk: the task with the latency
957 * @cpu: The cpu that initiated the trace.
959 * Flip the buffers between the @tr and the max_tr and record information
960 * about which task was the cause of this latency.
963 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
965 struct ring_buffer *buf;
970 WARN_ON_ONCE(!irqs_disabled());
972 if (!tr->allocated_snapshot) {
973 /* Only the nop tracer should hit this when disabling */
974 WARN_ON_ONCE(tr->current_trace != &nop_trace);
978 arch_spin_lock(&ftrace_max_lock);
980 buf = tr->trace_buffer.buffer;
981 tr->trace_buffer.buffer = tr->max_buffer.buffer;
982 tr->max_buffer.buffer = buf;
984 __update_max_tr(tr, tsk, cpu);
985 arch_spin_unlock(&ftrace_max_lock);
989 * update_max_tr_single - only copy one trace over, and reset the rest
991 * @tsk - task with the latency
992 * @cpu - the cpu of the buffer to copy.
994 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
997 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1004 WARN_ON_ONCE(!irqs_disabled());
1005 if (!tr->allocated_snapshot) {
1006 /* Only the nop tracer should hit this when disabling */
1007 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1011 arch_spin_lock(&ftrace_max_lock);
1013 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
1015 if (ret == -EBUSY) {
1017 * We failed to swap the buffer due to a commit taking
1018 * place on this CPU. We fail to record, but we reset
1019 * the max trace buffer (no one writes directly to it)
1020 * and flag that it failed.
1022 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1023 "Failed to swap buffers due to commit in progress\n");
1026 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1028 __update_max_tr(tr, tsk, cpu);
1029 arch_spin_unlock(&ftrace_max_lock);
1031 #endif /* CONFIG_TRACER_MAX_TRACE */
1033 static void default_wait_pipe(struct trace_iterator *iter)
1035 /* Iterators are static, they should be filled or empty */
1036 if (trace_buffer_iter(iter, iter->cpu_file))
1039 ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file);
1042 #ifdef CONFIG_FTRACE_STARTUP_TEST
1043 static int run_tracer_selftest(struct tracer *type)
1045 struct trace_array *tr = &global_trace;
1046 struct tracer *saved_tracer = tr->current_trace;
1049 if (!type->selftest || tracing_selftest_disabled)
1053 * Run a selftest on this tracer.
1054 * Here we reset the trace buffer, and set the current
1055 * tracer to be this tracer. The tracer can then run some
1056 * internal tracing to verify that everything is in order.
1057 * If we fail, we do not register this tracer.
1059 tracing_reset_online_cpus(&tr->trace_buffer);
1061 tr->current_trace = type;
1063 #ifdef CONFIG_TRACER_MAX_TRACE
1064 if (type->use_max_tr) {
1065 /* If we expanded the buffers, make sure the max is expanded too */
1066 if (ring_buffer_expanded)
1067 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1068 RING_BUFFER_ALL_CPUS);
1069 tr->allocated_snapshot = true;
1073 /* the test is responsible for initializing and enabling */
1074 pr_info("Testing tracer %s: ", type->name);
1075 ret = type->selftest(type, tr);
1076 /* the test is responsible for resetting too */
1077 tr->current_trace = saved_tracer;
1079 printk(KERN_CONT "FAILED!\n");
1080 /* Add the warning after printing 'FAILED' */
1084 /* Only reset on passing, to avoid touching corrupted buffers */
1085 tracing_reset_online_cpus(&tr->trace_buffer);
1087 #ifdef CONFIG_TRACER_MAX_TRACE
1088 if (type->use_max_tr) {
1089 tr->allocated_snapshot = false;
1091 /* Shrink the max buffer again */
1092 if (ring_buffer_expanded)
1093 ring_buffer_resize(tr->max_buffer.buffer, 1,
1094 RING_BUFFER_ALL_CPUS);
1098 printk(KERN_CONT "PASSED\n");
1102 static inline int run_tracer_selftest(struct tracer *type)
1106 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1109 * register_tracer - register a tracer with the ftrace system.
1110 * @type - the plugin for the tracer
1112 * Register a new plugin tracer.
1114 int register_tracer(struct tracer *type)
1120 pr_info("Tracer must have a name\n");
1124 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1125 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1129 mutex_lock(&trace_types_lock);
1131 tracing_selftest_running = true;
1133 for (t = trace_types; t; t = t->next) {
1134 if (strcmp(type->name, t->name) == 0) {
1136 pr_info("Tracer %s already registered\n",
1143 if (!type->set_flag)
1144 type->set_flag = &dummy_set_flag;
1146 type->flags = &dummy_tracer_flags;
1148 if (!type->flags->opts)
1149 type->flags->opts = dummy_tracer_opt;
1150 if (!type->wait_pipe)
1151 type->wait_pipe = default_wait_pipe;
1153 ret = run_tracer_selftest(type);
1157 type->next = trace_types;
1161 tracing_selftest_running = false;
1162 mutex_unlock(&trace_types_lock);
1164 if (ret || !default_bootup_tracer)
1167 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1170 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1171 /* Do we want this tracer to start on bootup? */
1172 tracing_set_tracer(type->name);
1173 default_bootup_tracer = NULL;
1174 /* disable other selftests, since this will break it. */
1175 tracing_selftest_disabled = true;
1176 #ifdef CONFIG_FTRACE_STARTUP_TEST
1177 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1185 void tracing_reset(struct trace_buffer *buf, int cpu)
1187 struct ring_buffer *buffer = buf->buffer;
1192 ring_buffer_record_disable(buffer);
1194 /* Make sure all commits have finished */
1195 synchronize_sched();
1196 ring_buffer_reset_cpu(buffer, cpu);
1198 ring_buffer_record_enable(buffer);
1201 void tracing_reset_online_cpus(struct trace_buffer *buf)
1203 struct ring_buffer *buffer = buf->buffer;
1209 ring_buffer_record_disable(buffer);
1211 /* Make sure all commits have finished */
1212 synchronize_sched();
1214 buf->time_start = ftrace_now(buf->cpu);
1216 for_each_online_cpu(cpu)
1217 ring_buffer_reset_cpu(buffer, cpu);
1219 ring_buffer_record_enable(buffer);
1222 void tracing_reset_current(int cpu)
1224 tracing_reset(&global_trace.trace_buffer, cpu);
1227 void tracing_reset_all_online_cpus(void)
1229 struct trace_array *tr;
1231 mutex_lock(&trace_types_lock);
1232 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1233 tracing_reset_online_cpus(&tr->trace_buffer);
1234 #ifdef CONFIG_TRACER_MAX_TRACE
1235 tracing_reset_online_cpus(&tr->max_buffer);
1238 mutex_unlock(&trace_types_lock);
1241 #define SAVED_CMDLINES 128
1242 #define NO_CMDLINE_MAP UINT_MAX
1243 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1244 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
1245 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
1246 static int cmdline_idx;
1247 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1249 /* temporary disable recording */
1250 static atomic_t trace_record_cmdline_disabled __read_mostly;
1252 static void trace_init_cmdlines(void)
1254 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
1255 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
1259 int is_tracing_stopped(void)
1261 return global_trace.stop_count;
1265 * ftrace_off_permanent - disable all ftrace code permanently
1267 * This should only be called when a serious anomally has
1268 * been detected. This will turn off the function tracing,
1269 * ring buffers, and other tracing utilites. It takes no
1270 * locks and can be called from any context.
1272 void ftrace_off_permanent(void)
1274 tracing_disabled = 1;
1276 tracing_off_permanent();
1280 * tracing_start - quick start of the tracer
1282 * If tracing is enabled but was stopped by tracing_stop,
1283 * this will start the tracer back up.
1285 void tracing_start(void)
1287 struct ring_buffer *buffer;
1288 unsigned long flags;
1290 if (tracing_disabled)
1293 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1294 if (--global_trace.stop_count) {
1295 if (global_trace.stop_count < 0) {
1296 /* Someone screwed up their debugging */
1298 global_trace.stop_count = 0;
1303 /* Prevent the buffers from switching */
1304 arch_spin_lock(&ftrace_max_lock);
1306 buffer = global_trace.trace_buffer.buffer;
1308 ring_buffer_record_enable(buffer);
1310 #ifdef CONFIG_TRACER_MAX_TRACE
1311 buffer = global_trace.max_buffer.buffer;
1313 ring_buffer_record_enable(buffer);
1316 arch_spin_unlock(&ftrace_max_lock);
1320 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1323 static void tracing_start_tr(struct trace_array *tr)
1325 struct ring_buffer *buffer;
1326 unsigned long flags;
1328 if (tracing_disabled)
1331 /* If global, we need to also start the max tracer */
1332 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1333 return tracing_start();
1335 raw_spin_lock_irqsave(&tr->start_lock, flags);
1337 if (--tr->stop_count) {
1338 if (tr->stop_count < 0) {
1339 /* Someone screwed up their debugging */
1346 buffer = tr->trace_buffer.buffer;
1348 ring_buffer_record_enable(buffer);
1351 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1355 * tracing_stop - quick stop of the tracer
1357 * Light weight way to stop tracing. Use in conjunction with
1360 void tracing_stop(void)
1362 struct ring_buffer *buffer;
1363 unsigned long flags;
1366 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1367 if (global_trace.stop_count++)
1370 /* Prevent the buffers from switching */
1371 arch_spin_lock(&ftrace_max_lock);
1373 buffer = global_trace.trace_buffer.buffer;
1375 ring_buffer_record_disable(buffer);
1377 #ifdef CONFIG_TRACER_MAX_TRACE
1378 buffer = global_trace.max_buffer.buffer;
1380 ring_buffer_record_disable(buffer);
1383 arch_spin_unlock(&ftrace_max_lock);
1386 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1389 static void tracing_stop_tr(struct trace_array *tr)
1391 struct ring_buffer *buffer;
1392 unsigned long flags;
1394 /* If global, we need to also stop the max tracer */
1395 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1396 return tracing_stop();
1398 raw_spin_lock_irqsave(&tr->start_lock, flags);
1399 if (tr->stop_count++)
1402 buffer = tr->trace_buffer.buffer;
1404 ring_buffer_record_disable(buffer);
1407 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1410 void trace_stop_cmdline_recording(void);
1412 static void trace_save_cmdline(struct task_struct *tsk)
1416 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1420 * It's not the end of the world if we don't get
1421 * the lock, but we also don't want to spin
1422 * nor do we want to disable interrupts,
1423 * so if we miss here, then better luck next time.
1425 if (!arch_spin_trylock(&trace_cmdline_lock))
1428 idx = map_pid_to_cmdline[tsk->pid];
1429 if (idx == NO_CMDLINE_MAP) {
1430 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1433 * Check whether the cmdline buffer at idx has a pid
1434 * mapped. We are going to overwrite that entry so we
1435 * need to clear the map_pid_to_cmdline. Otherwise we
1436 * would read the new comm for the old pid.
1438 pid = map_cmdline_to_pid[idx];
1439 if (pid != NO_CMDLINE_MAP)
1440 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1442 map_cmdline_to_pid[idx] = tsk->pid;
1443 map_pid_to_cmdline[tsk->pid] = idx;
1448 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1450 arch_spin_unlock(&trace_cmdline_lock);
1453 void trace_find_cmdline(int pid, char comm[])
1458 strcpy(comm, "<idle>");
1462 if (WARN_ON_ONCE(pid < 0)) {
1463 strcpy(comm, "<XXX>");
1467 if (pid > PID_MAX_DEFAULT) {
1468 strcpy(comm, "<...>");
1473 arch_spin_lock(&trace_cmdline_lock);
1474 map = map_pid_to_cmdline[pid];
1475 if (map != NO_CMDLINE_MAP)
1476 strcpy(comm, saved_cmdlines[map]);
1478 strcpy(comm, "<...>");
1480 arch_spin_unlock(&trace_cmdline_lock);
1484 void tracing_record_cmdline(struct task_struct *tsk)
1486 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1489 if (!__this_cpu_read(trace_cmdline_save))
1492 __this_cpu_write(trace_cmdline_save, false);
1494 trace_save_cmdline(tsk);
1498 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1501 struct task_struct *tsk = current;
1503 entry->preempt_count = pc & 0xff;
1504 entry->pid = (tsk) ? tsk->pid : 0;
1506 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1507 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1509 TRACE_FLAG_IRQS_NOSUPPORT |
1511 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1512 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1513 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1515 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1517 struct ring_buffer_event *
1518 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1521 unsigned long flags, int pc)
1523 struct ring_buffer_event *event;
1525 event = ring_buffer_lock_reserve(buffer, len);
1526 if (event != NULL) {
1527 struct trace_entry *ent = ring_buffer_event_data(event);
1529 tracing_generic_entry_update(ent, flags, pc);
1537 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1539 __this_cpu_write(trace_cmdline_save, true);
1540 ring_buffer_unlock_commit(buffer, event);
1544 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1545 struct ring_buffer_event *event,
1546 unsigned long flags, int pc)
1548 __buffer_unlock_commit(buffer, event);
1550 ftrace_trace_stack(buffer, flags, 6, pc);
1551 ftrace_trace_userstack(buffer, flags, pc);
1554 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1555 struct ring_buffer_event *event,
1556 unsigned long flags, int pc)
1558 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1560 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1562 struct ring_buffer_event *
1563 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1564 struct ftrace_event_file *ftrace_file,
1565 int type, unsigned long len,
1566 unsigned long flags, int pc)
1568 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1569 return trace_buffer_lock_reserve(*current_rb,
1570 type, len, flags, pc);
1572 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1574 struct ring_buffer_event *
1575 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1576 int type, unsigned long len,
1577 unsigned long flags, int pc)
1579 *current_rb = global_trace.trace_buffer.buffer;
1580 return trace_buffer_lock_reserve(*current_rb,
1581 type, len, flags, pc);
1583 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1585 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1586 struct ring_buffer_event *event,
1587 unsigned long flags, int pc)
1589 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1591 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1593 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1594 struct ring_buffer_event *event,
1595 unsigned long flags, int pc,
1596 struct pt_regs *regs)
1598 __buffer_unlock_commit(buffer, event);
1600 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1601 ftrace_trace_userstack(buffer, flags, pc);
1603 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1605 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1606 struct ring_buffer_event *event)
1608 ring_buffer_discard_commit(buffer, event);
1610 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1613 trace_function(struct trace_array *tr,
1614 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1617 struct ftrace_event_call *call = &event_function;
1618 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1619 struct ring_buffer_event *event;
1620 struct ftrace_entry *entry;
1622 /* If we are reading the ring buffer, don't trace */
1623 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1626 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1630 entry = ring_buffer_event_data(event);
1632 entry->parent_ip = parent_ip;
1634 if (!filter_check_discard(call, entry, buffer, event))
1635 __buffer_unlock_commit(buffer, event);
1638 #ifdef CONFIG_STACKTRACE
1640 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1641 struct ftrace_stack {
1642 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1645 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1646 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1648 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1649 unsigned long flags,
1650 int skip, int pc, struct pt_regs *regs)
1652 struct ftrace_event_call *call = &event_kernel_stack;
1653 struct ring_buffer_event *event;
1654 struct stack_entry *entry;
1655 struct stack_trace trace;
1657 int size = FTRACE_STACK_ENTRIES;
1659 trace.nr_entries = 0;
1663 * Since events can happen in NMIs there's no safe way to
1664 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1665 * or NMI comes in, it will just have to use the default
1666 * FTRACE_STACK_SIZE.
1668 preempt_disable_notrace();
1670 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1672 * We don't need any atomic variables, just a barrier.
1673 * If an interrupt comes in, we don't care, because it would
1674 * have exited and put the counter back to what we want.
1675 * We just need a barrier to keep gcc from moving things
1679 if (use_stack == 1) {
1680 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1681 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1684 save_stack_trace_regs(regs, &trace);
1686 save_stack_trace(&trace);
1688 if (trace.nr_entries > size)
1689 size = trace.nr_entries;
1691 /* From now on, use_stack is a boolean */
1694 size *= sizeof(unsigned long);
1696 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1697 sizeof(*entry) + size, flags, pc);
1700 entry = ring_buffer_event_data(event);
1702 memset(&entry->caller, 0, size);
1705 memcpy(&entry->caller, trace.entries,
1706 trace.nr_entries * sizeof(unsigned long));
1708 trace.max_entries = FTRACE_STACK_ENTRIES;
1709 trace.entries = entry->caller;
1711 save_stack_trace_regs(regs, &trace);
1713 save_stack_trace(&trace);
1716 entry->size = trace.nr_entries;
1718 if (!filter_check_discard(call, entry, buffer, event))
1719 __buffer_unlock_commit(buffer, event);
1722 /* Again, don't let gcc optimize things here */
1724 __this_cpu_dec(ftrace_stack_reserve);
1725 preempt_enable_notrace();
1729 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1730 int skip, int pc, struct pt_regs *regs)
1732 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1735 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1738 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1741 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1744 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1747 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1750 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1754 * trace_dump_stack - record a stack back trace in the trace buffer
1755 * @skip: Number of functions to skip (helper handlers)
1757 void trace_dump_stack(int skip)
1759 unsigned long flags;
1761 if (tracing_disabled || tracing_selftest_running)
1764 local_save_flags(flags);
1767 * Skip 3 more, seems to get us at the caller of
1771 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
1772 flags, skip, preempt_count(), NULL);
1775 static DEFINE_PER_CPU(int, user_stack_count);
1778 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1780 struct ftrace_event_call *call = &event_user_stack;
1781 struct ring_buffer_event *event;
1782 struct userstack_entry *entry;
1783 struct stack_trace trace;
1785 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1789 * NMIs can not handle page faults, even with fix ups.
1790 * The save user stack can (and often does) fault.
1792 if (unlikely(in_nmi()))
1796 * prevent recursion, since the user stack tracing may
1797 * trigger other kernel events.
1800 if (__this_cpu_read(user_stack_count))
1803 __this_cpu_inc(user_stack_count);
1805 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1806 sizeof(*entry), flags, pc);
1808 goto out_drop_count;
1809 entry = ring_buffer_event_data(event);
1811 entry->tgid = current->tgid;
1812 memset(&entry->caller, 0, sizeof(entry->caller));
1814 trace.nr_entries = 0;
1815 trace.max_entries = FTRACE_STACK_ENTRIES;
1817 trace.entries = entry->caller;
1819 save_stack_trace_user(&trace);
1820 if (!filter_check_discard(call, entry, buffer, event))
1821 __buffer_unlock_commit(buffer, event);
1824 __this_cpu_dec(user_stack_count);
1830 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1832 ftrace_trace_userstack(tr, flags, preempt_count());
1836 #endif /* CONFIG_STACKTRACE */
1838 /* created for use with alloc_percpu */
1839 struct trace_buffer_struct {
1840 char buffer[TRACE_BUF_SIZE];
1843 static struct trace_buffer_struct *trace_percpu_buffer;
1844 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1845 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1846 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1849 * The buffer used is dependent on the context. There is a per cpu
1850 * buffer for normal context, softirq contex, hard irq context and
1851 * for NMI context. Thise allows for lockless recording.
1853 * Note, if the buffers failed to be allocated, then this returns NULL
1855 static char *get_trace_buf(void)
1857 struct trace_buffer_struct *percpu_buffer;
1860 * If we have allocated per cpu buffers, then we do not
1861 * need to do any locking.
1864 percpu_buffer = trace_percpu_nmi_buffer;
1866 percpu_buffer = trace_percpu_irq_buffer;
1867 else if (in_softirq())
1868 percpu_buffer = trace_percpu_sirq_buffer;
1870 percpu_buffer = trace_percpu_buffer;
1875 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1878 static int alloc_percpu_trace_buffer(void)
1880 struct trace_buffer_struct *buffers;
1881 struct trace_buffer_struct *sirq_buffers;
1882 struct trace_buffer_struct *irq_buffers;
1883 struct trace_buffer_struct *nmi_buffers;
1885 buffers = alloc_percpu(struct trace_buffer_struct);
1889 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1893 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1897 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1901 trace_percpu_buffer = buffers;
1902 trace_percpu_sirq_buffer = sirq_buffers;
1903 trace_percpu_irq_buffer = irq_buffers;
1904 trace_percpu_nmi_buffer = nmi_buffers;
1909 free_percpu(irq_buffers);
1911 free_percpu(sirq_buffers);
1913 free_percpu(buffers);
1915 WARN(1, "Could not allocate percpu trace_printk buffer");
1919 static int buffers_allocated;
1921 void trace_printk_init_buffers(void)
1923 if (buffers_allocated)
1926 if (alloc_percpu_trace_buffer())
1929 pr_info("ftrace: Allocated trace_printk buffers\n");
1931 /* Expand the buffers to set size */
1932 tracing_update_buffers();
1934 buffers_allocated = 1;
1937 * trace_printk_init_buffers() can be called by modules.
1938 * If that happens, then we need to start cmdline recording
1939 * directly here. If the global_trace.buffer is already
1940 * allocated here, then this was called by module code.
1942 if (global_trace.trace_buffer.buffer)
1943 tracing_start_cmdline_record();
1946 void trace_printk_start_comm(void)
1948 /* Start tracing comms if trace printk is set */
1949 if (!buffers_allocated)
1951 tracing_start_cmdline_record();
1954 static void trace_printk_start_stop_comm(int enabled)
1956 if (!buffers_allocated)
1960 tracing_start_cmdline_record();
1962 tracing_stop_cmdline_record();
1966 * trace_vbprintk - write binary msg to tracing buffer
1969 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1971 struct ftrace_event_call *call = &event_bprint;
1972 struct ring_buffer_event *event;
1973 struct ring_buffer *buffer;
1974 struct trace_array *tr = &global_trace;
1975 struct bprint_entry *entry;
1976 unsigned long flags;
1978 int len = 0, size, pc;
1980 if (unlikely(tracing_selftest_running || tracing_disabled))
1983 /* Don't pollute graph traces with trace_vprintk internals */
1984 pause_graph_tracing();
1986 pc = preempt_count();
1987 preempt_disable_notrace();
1989 tbuffer = get_trace_buf();
1995 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1997 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
2000 local_save_flags(flags);
2001 size = sizeof(*entry) + sizeof(u32) * len;
2002 buffer = tr->trace_buffer.buffer;
2003 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
2007 entry = ring_buffer_event_data(event);
2011 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
2012 if (!filter_check_discard(call, entry, buffer, event)) {
2013 __buffer_unlock_commit(buffer, event);
2014 ftrace_trace_stack(buffer, flags, 6, pc);
2018 preempt_enable_notrace();
2019 unpause_graph_tracing();
2023 EXPORT_SYMBOL_GPL(trace_vbprintk);
2026 __trace_array_vprintk(struct ring_buffer *buffer,
2027 unsigned long ip, const char *fmt, va_list args)
2029 struct ftrace_event_call *call = &event_print;
2030 struct ring_buffer_event *event;
2031 int len = 0, size, pc;
2032 struct print_entry *entry;
2033 unsigned long flags;
2036 if (tracing_disabled || tracing_selftest_running)
2039 /* Don't pollute graph traces with trace_vprintk internals */
2040 pause_graph_tracing();
2042 pc = preempt_count();
2043 preempt_disable_notrace();
2046 tbuffer = get_trace_buf();
2052 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
2053 if (len > TRACE_BUF_SIZE)
2056 local_save_flags(flags);
2057 size = sizeof(*entry) + len + 1;
2058 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
2062 entry = ring_buffer_event_data(event);
2065 memcpy(&entry->buf, tbuffer, len);
2066 entry->buf[len] = '\0';
2067 if (!filter_check_discard(call, entry, buffer, event)) {
2068 __buffer_unlock_commit(buffer, event);
2069 ftrace_trace_stack(buffer, flags, 6, pc);
2072 preempt_enable_notrace();
2073 unpause_graph_tracing();
2078 int trace_array_vprintk(struct trace_array *tr,
2079 unsigned long ip, const char *fmt, va_list args)
2081 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
2084 int trace_array_printk(struct trace_array *tr,
2085 unsigned long ip, const char *fmt, ...)
2090 if (!(trace_flags & TRACE_ITER_PRINTK))
2094 ret = trace_array_vprintk(tr, ip, fmt, ap);
2099 int trace_array_printk_buf(struct ring_buffer *buffer,
2100 unsigned long ip, const char *fmt, ...)
2105 if (!(trace_flags & TRACE_ITER_PRINTK))
2109 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2114 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2116 return trace_array_vprintk(&global_trace, ip, fmt, args);
2118 EXPORT_SYMBOL_GPL(trace_vprintk);
2120 static void trace_iterator_increment(struct trace_iterator *iter)
2122 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2126 ring_buffer_read(buf_iter, NULL);
2129 static struct trace_entry *
2130 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2131 unsigned long *lost_events)
2133 struct ring_buffer_event *event;
2134 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2137 event = ring_buffer_iter_peek(buf_iter, ts);
2139 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2143 iter->ent_size = ring_buffer_event_length(event);
2144 return ring_buffer_event_data(event);
2150 static struct trace_entry *
2151 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2152 unsigned long *missing_events, u64 *ent_ts)
2154 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2155 struct trace_entry *ent, *next = NULL;
2156 unsigned long lost_events = 0, next_lost = 0;
2157 int cpu_file = iter->cpu_file;
2158 u64 next_ts = 0, ts;
2164 * If we are in a per_cpu trace file, don't bother by iterating over
2165 * all cpu and peek directly.
2167 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2168 if (ring_buffer_empty_cpu(buffer, cpu_file))
2170 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2172 *ent_cpu = cpu_file;
2177 for_each_tracing_cpu(cpu) {
2179 if (ring_buffer_empty_cpu(buffer, cpu))
2182 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2185 * Pick the entry with the smallest timestamp:
2187 if (ent && (!next || ts < next_ts)) {
2191 next_lost = lost_events;
2192 next_size = iter->ent_size;
2196 iter->ent_size = next_size;
2199 *ent_cpu = next_cpu;
2205 *missing_events = next_lost;
2210 /* Find the next real entry, without updating the iterator itself */
2211 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2212 int *ent_cpu, u64 *ent_ts)
2214 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2217 /* Find the next real entry, and increment the iterator to the next entry */
2218 void *trace_find_next_entry_inc(struct trace_iterator *iter)
2220 iter->ent = __find_next_entry(iter, &iter->cpu,
2221 &iter->lost_events, &iter->ts);
2224 trace_iterator_increment(iter);
2226 return iter->ent ? iter : NULL;
2229 static void trace_consume(struct trace_iterator *iter)
2231 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2232 &iter->lost_events);
2235 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2237 struct trace_iterator *iter = m->private;
2241 WARN_ON_ONCE(iter->leftover);
2245 /* can't go backwards */
2250 ent = trace_find_next_entry_inc(iter);
2254 while (ent && iter->idx < i)
2255 ent = trace_find_next_entry_inc(iter);
2262 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2264 struct ring_buffer_event *event;
2265 struct ring_buffer_iter *buf_iter;
2266 unsigned long entries = 0;
2269 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2271 buf_iter = trace_buffer_iter(iter, cpu);
2275 ring_buffer_iter_reset(buf_iter);
2278 * We could have the case with the max latency tracers
2279 * that a reset never took place on a cpu. This is evident
2280 * by the timestamp being before the start of the buffer.
2282 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2283 if (ts >= iter->trace_buffer->time_start)
2286 ring_buffer_read(buf_iter, NULL);
2289 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2293 * The current tracer is copied to avoid a global locking
2296 static void *s_start(struct seq_file *m, loff_t *pos)
2298 struct trace_iterator *iter = m->private;
2299 struct trace_array *tr = iter->tr;
2300 int cpu_file = iter->cpu_file;
2306 * copy the tracer to avoid using a global lock all around.
2307 * iter->trace is a copy of current_trace, the pointer to the
2308 * name may be used instead of a strcmp(), as iter->trace->name
2309 * will point to the same string as current_trace->name.
2311 mutex_lock(&trace_types_lock);
2312 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2313 *iter->trace = *tr->current_trace;
2314 mutex_unlock(&trace_types_lock);
2316 #ifdef CONFIG_TRACER_MAX_TRACE
2317 if (iter->snapshot && iter->trace->use_max_tr)
2318 return ERR_PTR(-EBUSY);
2321 if (!iter->snapshot)
2322 atomic_inc(&trace_record_cmdline_disabled);
2324 if (*pos != iter->pos) {
2329 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2330 for_each_tracing_cpu(cpu)
2331 tracing_iter_reset(iter, cpu);
2333 tracing_iter_reset(iter, cpu_file);
2336 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2341 * If we overflowed the seq_file before, then we want
2342 * to just reuse the trace_seq buffer again.
2348 p = s_next(m, p, &l);
2352 trace_event_read_lock();
2353 trace_access_lock(cpu_file);
2357 static void s_stop(struct seq_file *m, void *p)
2359 struct trace_iterator *iter = m->private;
2361 #ifdef CONFIG_TRACER_MAX_TRACE
2362 if (iter->snapshot && iter->trace->use_max_tr)
2366 if (!iter->snapshot)
2367 atomic_dec(&trace_record_cmdline_disabled);
2369 trace_access_unlock(iter->cpu_file);
2370 trace_event_read_unlock();
2374 get_total_entries(struct trace_buffer *buf,
2375 unsigned long *total, unsigned long *entries)
2377 unsigned long count;
2383 for_each_tracing_cpu(cpu) {
2384 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2386 * If this buffer has skipped entries, then we hold all
2387 * entries for the trace and we need to ignore the
2388 * ones before the time stamp.
2390 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2391 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2392 /* total is the same as the entries */
2396 ring_buffer_overrun_cpu(buf->buffer, cpu);
2401 static void print_lat_help_header(struct seq_file *m)
2403 seq_puts(m, "# _------=> CPU# \n");
2404 seq_puts(m, "# / _-----=> irqs-off \n");
2405 seq_puts(m, "# | / _----=> need-resched \n");
2406 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2407 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2408 seq_puts(m, "# |||| / delay \n");
2409 seq_puts(m, "# cmd pid ||||| time | caller \n");
2410 seq_puts(m, "# \\ / ||||| \\ | / \n");
2413 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2415 unsigned long total;
2416 unsigned long entries;
2418 get_total_entries(buf, &total, &entries);
2419 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2420 entries, total, num_online_cpus());
2424 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2426 print_event_info(buf, m);
2427 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2428 seq_puts(m, "# | | | | |\n");
2431 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2433 print_event_info(buf, m);
2434 seq_puts(m, "# _-----=> irqs-off\n");
2435 seq_puts(m, "# / _----=> need-resched\n");
2436 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2437 seq_puts(m, "# || / _--=> preempt-depth\n");
2438 seq_puts(m, "# ||| / delay\n");
2439 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2440 seq_puts(m, "# | | | |||| | |\n");
2444 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2446 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2447 struct trace_buffer *buf = iter->trace_buffer;
2448 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2449 struct tracer *type = iter->trace;
2450 unsigned long entries;
2451 unsigned long total;
2452 const char *name = "preemption";
2456 get_total_entries(buf, &total, &entries);
2458 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2460 seq_puts(m, "# -----------------------------------"
2461 "---------------------------------\n");
2462 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2463 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2464 nsecs_to_usecs(data->saved_latency),
2468 #if defined(CONFIG_PREEMPT_NONE)
2470 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2472 #elif defined(CONFIG_PREEMPT)
2477 /* These are reserved for later use */
2480 seq_printf(m, " #P:%d)\n", num_online_cpus());
2484 seq_puts(m, "# -----------------\n");
2485 seq_printf(m, "# | task: %.16s-%d "
2486 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2487 data->comm, data->pid,
2488 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2489 data->policy, data->rt_priority);
2490 seq_puts(m, "# -----------------\n");
2492 if (data->critical_start) {
2493 seq_puts(m, "# => started at: ");
2494 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2495 trace_print_seq(m, &iter->seq);
2496 seq_puts(m, "\n# => ended at: ");
2497 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2498 trace_print_seq(m, &iter->seq);
2499 seq_puts(m, "\n#\n");
2505 static void test_cpu_buff_start(struct trace_iterator *iter)
2507 struct trace_seq *s = &iter->seq;
2509 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2512 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2515 if (cpumask_test_cpu(iter->cpu, iter->started))
2518 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2521 cpumask_set_cpu(iter->cpu, iter->started);
2523 /* Don't print started cpu buffer for the first entry of the trace */
2525 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2529 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2531 struct trace_seq *s = &iter->seq;
2532 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2533 struct trace_entry *entry;
2534 struct trace_event *event;
2538 test_cpu_buff_start(iter);
2540 event = ftrace_find_event(entry->type);
2542 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2543 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2544 if (!trace_print_lat_context(iter))
2547 if (!trace_print_context(iter))
2553 return event->funcs->trace(iter, sym_flags, event);
2555 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2558 return TRACE_TYPE_HANDLED;
2560 return TRACE_TYPE_PARTIAL_LINE;
2563 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2565 struct trace_seq *s = &iter->seq;
2566 struct trace_entry *entry;
2567 struct trace_event *event;
2571 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2572 if (!trace_seq_printf(s, "%d %d %llu ",
2573 entry->pid, iter->cpu, iter->ts))
2577 event = ftrace_find_event(entry->type);
2579 return event->funcs->raw(iter, 0, event);
2581 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2584 return TRACE_TYPE_HANDLED;
2586 return TRACE_TYPE_PARTIAL_LINE;
2589 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2591 struct trace_seq *s = &iter->seq;
2592 unsigned char newline = '\n';
2593 struct trace_entry *entry;
2594 struct trace_event *event;
2598 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2599 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2600 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2601 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2604 event = ftrace_find_event(entry->type);
2606 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2607 if (ret != TRACE_TYPE_HANDLED)
2611 SEQ_PUT_FIELD_RET(s, newline);
2613 return TRACE_TYPE_HANDLED;
2616 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2618 struct trace_seq *s = &iter->seq;
2619 struct trace_entry *entry;
2620 struct trace_event *event;
2624 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2625 SEQ_PUT_FIELD_RET(s, entry->pid);
2626 SEQ_PUT_FIELD_RET(s, iter->cpu);
2627 SEQ_PUT_FIELD_RET(s, iter->ts);
2630 event = ftrace_find_event(entry->type);
2631 return event ? event->funcs->binary(iter, 0, event) :
2635 int trace_empty(struct trace_iterator *iter)
2637 struct ring_buffer_iter *buf_iter;
2640 /* If we are looking at one CPU buffer, only check that one */
2641 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2642 cpu = iter->cpu_file;
2643 buf_iter = trace_buffer_iter(iter, cpu);
2645 if (!ring_buffer_iter_empty(buf_iter))
2648 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2654 for_each_tracing_cpu(cpu) {
2655 buf_iter = trace_buffer_iter(iter, cpu);
2657 if (!ring_buffer_iter_empty(buf_iter))
2660 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2668 /* Called with trace_event_read_lock() held. */
2669 enum print_line_t print_trace_line(struct trace_iterator *iter)
2671 enum print_line_t ret;
2673 if (iter->lost_events &&
2674 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2675 iter->cpu, iter->lost_events))
2676 return TRACE_TYPE_PARTIAL_LINE;
2678 if (iter->trace && iter->trace->print_line) {
2679 ret = iter->trace->print_line(iter);
2680 if (ret != TRACE_TYPE_UNHANDLED)
2684 if (iter->ent->type == TRACE_BPUTS &&
2685 trace_flags & TRACE_ITER_PRINTK &&
2686 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2687 return trace_print_bputs_msg_only(iter);
2689 if (iter->ent->type == TRACE_BPRINT &&
2690 trace_flags & TRACE_ITER_PRINTK &&
2691 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2692 return trace_print_bprintk_msg_only(iter);
2694 if (iter->ent->type == TRACE_PRINT &&
2695 trace_flags & TRACE_ITER_PRINTK &&
2696 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2697 return trace_print_printk_msg_only(iter);
2699 if (trace_flags & TRACE_ITER_BIN)
2700 return print_bin_fmt(iter);
2702 if (trace_flags & TRACE_ITER_HEX)
2703 return print_hex_fmt(iter);
2705 if (trace_flags & TRACE_ITER_RAW)
2706 return print_raw_fmt(iter);
2708 return print_trace_fmt(iter);
2711 void trace_latency_header(struct seq_file *m)
2713 struct trace_iterator *iter = m->private;
2715 /* print nothing if the buffers are empty */
2716 if (trace_empty(iter))
2719 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2720 print_trace_header(m, iter);
2722 if (!(trace_flags & TRACE_ITER_VERBOSE))
2723 print_lat_help_header(m);
2726 void trace_default_header(struct seq_file *m)
2728 struct trace_iterator *iter = m->private;
2730 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2733 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2734 /* print nothing if the buffers are empty */
2735 if (trace_empty(iter))
2737 print_trace_header(m, iter);
2738 if (!(trace_flags & TRACE_ITER_VERBOSE))
2739 print_lat_help_header(m);
2741 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2742 if (trace_flags & TRACE_ITER_IRQ_INFO)
2743 print_func_help_header_irq(iter->trace_buffer, m);
2745 print_func_help_header(iter->trace_buffer, m);
2750 static void test_ftrace_alive(struct seq_file *m)
2752 if (!ftrace_is_dead())
2754 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2755 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2758 #ifdef CONFIG_TRACER_MAX_TRACE
2759 static void show_snapshot_main_help(struct seq_file *m)
2761 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2762 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2763 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2764 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate)\n");
2765 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2766 seq_printf(m, "# is not a '0' or '1')\n");
2769 static void show_snapshot_percpu_help(struct seq_file *m)
2771 seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2772 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
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 for this cpu.\n");
2776 seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n");
2777 seq_printf(m, "# Must use main snapshot file to allocate.\n");
2779 seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2780 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2781 seq_printf(m, "# is not a '0' or '1')\n");
2784 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2786 if (iter->tr->allocated_snapshot)
2787 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2789 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2791 seq_printf(m, "# Snapshot commands:\n");
2792 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2793 show_snapshot_main_help(m);
2795 show_snapshot_percpu_help(m);
2798 /* Should never be called */
2799 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2802 static int s_show(struct seq_file *m, void *v)
2804 struct trace_iterator *iter = v;
2807 if (iter->ent == NULL) {
2809 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2811 test_ftrace_alive(m);
2813 if (iter->snapshot && trace_empty(iter))
2814 print_snapshot_help(m, iter);
2815 else if (iter->trace && iter->trace->print_header)
2816 iter->trace->print_header(m);
2818 trace_default_header(m);
2820 } else if (iter->leftover) {
2822 * If we filled the seq_file buffer earlier, we
2823 * want to just show it now.
2825 ret = trace_print_seq(m, &iter->seq);
2827 /* ret should this time be zero, but you never know */
2828 iter->leftover = ret;
2831 print_trace_line(iter);
2832 ret = trace_print_seq(m, &iter->seq);
2834 * If we overflow the seq_file buffer, then it will
2835 * ask us for this data again at start up.
2837 * ret is 0 if seq_file write succeeded.
2840 iter->leftover = ret;
2847 * Should be used after trace_array_get(), trace_types_lock
2848 * ensures that i_cdev was already initialized.
2850 static inline int tracing_get_cpu(struct inode *inode)
2852 if (inode->i_cdev) /* See trace_create_cpu_file() */
2853 return (long)inode->i_cdev - 1;
2854 return RING_BUFFER_ALL_CPUS;
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[tr->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 static 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;
2987 static int tracing_release(struct inode *inode, struct file *file)
2989 struct seq_file *m = file->private_data;
2990 struct trace_iterator *iter;
2991 struct trace_array *tr;
2994 /* Writes do not use seq_file, need to grab tr from inode */
2995 if (!(file->f_mode & FMODE_READ)) {
2996 struct trace_cpu *tc = inode->i_private;
2998 trace_array_put(tc->tr);
3005 mutex_lock(&trace_types_lock);
3007 for_each_tracing_cpu(cpu) {
3008 if (iter->buffer_iter[cpu])
3009 ring_buffer_read_finish(iter->buffer_iter[cpu]);
3012 if (iter->trace && iter->trace->close)
3013 iter->trace->close(iter);
3015 if (!iter->snapshot)
3016 /* reenable tracing if it was previously enabled */
3017 tracing_start_tr(tr);
3019 __trace_array_put(tr);
3021 mutex_unlock(&trace_types_lock);
3023 mutex_destroy(&iter->mutex);
3024 free_cpumask_var(iter->started);
3026 kfree(iter->buffer_iter);
3027 seq_release_private(inode, file);
3032 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
3034 struct trace_array *tr = inode->i_private;
3036 trace_array_put(tr);
3040 static int tracing_single_release_tr(struct inode *inode, struct file *file)
3042 struct trace_array *tr = inode->i_private;
3044 trace_array_put(tr);
3046 return single_release(inode, file);
3049 static int tracing_open(struct inode *inode, struct file *file)
3051 struct trace_cpu *tc = inode->i_private;
3052 struct trace_array *tr = tc->tr;
3053 struct trace_iterator *iter;
3056 if (trace_array_get(tr) < 0)
3059 /* If this file was open for write, then erase contents */
3060 if ((file->f_mode & FMODE_WRITE) &&
3061 (file->f_flags & O_TRUNC)) {
3062 if (tc->cpu == RING_BUFFER_ALL_CPUS)
3063 tracing_reset_online_cpus(&tr->trace_buffer);
3065 tracing_reset(&tr->trace_buffer, tc->cpu);
3068 if (file->f_mode & FMODE_READ) {
3069 iter = __tracing_open(tr, tc, inode, file, false);
3071 ret = PTR_ERR(iter);
3072 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
3073 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3077 trace_array_put(tr);
3083 t_next(struct seq_file *m, void *v, loff_t *pos)
3085 struct tracer *t = v;
3095 static void *t_start(struct seq_file *m, loff_t *pos)
3100 mutex_lock(&trace_types_lock);
3101 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
3107 static void t_stop(struct seq_file *m, void *p)
3109 mutex_unlock(&trace_types_lock);
3112 static int t_show(struct seq_file *m, void *v)
3114 struct tracer *t = v;
3119 seq_printf(m, "%s", t->name);
3128 static const struct seq_operations show_traces_seq_ops = {
3135 static int show_traces_open(struct inode *inode, struct file *file)
3137 if (tracing_disabled)
3140 return seq_open(file, &show_traces_seq_ops);
3144 tracing_write_stub(struct file *filp, const char __user *ubuf,
3145 size_t count, loff_t *ppos)
3150 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
3152 if (file->f_mode & FMODE_READ)
3153 return seq_lseek(file, offset, origin);
3158 static const struct file_operations tracing_fops = {
3159 .open = tracing_open,
3161 .write = tracing_write_stub,
3162 .llseek = tracing_seek,
3163 .release = tracing_release,
3166 static const struct file_operations show_traces_fops = {
3167 .open = show_traces_open,
3169 .release = seq_release,
3170 .llseek = seq_lseek,
3174 * Only trace on a CPU if the bitmask is set:
3176 static cpumask_var_t tracing_cpumask;
3179 * The tracer itself will not take this lock, but still we want
3180 * to provide a consistent cpumask to user-space:
3182 static DEFINE_MUTEX(tracing_cpumask_update_lock);
3185 * Temporary storage for the character representation of the
3186 * CPU bitmask (and one more byte for the newline):
3188 static char mask_str[NR_CPUS + 1];
3191 tracing_cpumask_read(struct file *filp, char __user *ubuf,
3192 size_t count, loff_t *ppos)
3196 mutex_lock(&tracing_cpumask_update_lock);
3198 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
3199 if (count - len < 2) {
3203 len += sprintf(mask_str + len, "\n");
3204 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3207 mutex_unlock(&tracing_cpumask_update_lock);
3213 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3214 size_t count, loff_t *ppos)
3216 struct trace_array *tr = filp->private_data;
3217 cpumask_var_t tracing_cpumask_new;
3220 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3223 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3227 mutex_lock(&tracing_cpumask_update_lock);
3229 local_irq_disable();
3230 arch_spin_lock(&ftrace_max_lock);
3231 for_each_tracing_cpu(cpu) {
3233 * Increase/decrease the disabled counter if we are
3234 * about to flip a bit in the cpumask:
3236 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
3237 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3238 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3239 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3241 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
3242 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3243 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3244 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3247 arch_spin_unlock(&ftrace_max_lock);
3250 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
3252 mutex_unlock(&tracing_cpumask_update_lock);
3253 free_cpumask_var(tracing_cpumask_new);
3258 free_cpumask_var(tracing_cpumask_new);
3263 static const struct file_operations tracing_cpumask_fops = {
3264 .open = tracing_open_generic,
3265 .read = tracing_cpumask_read,
3266 .write = tracing_cpumask_write,
3267 .llseek = generic_file_llseek,
3270 static int tracing_trace_options_show(struct seq_file *m, void *v)
3272 struct tracer_opt *trace_opts;
3273 struct trace_array *tr = m->private;
3277 mutex_lock(&trace_types_lock);
3278 tracer_flags = tr->current_trace->flags->val;
3279 trace_opts = tr->current_trace->flags->opts;
3281 for (i = 0; trace_options[i]; i++) {
3282 if (trace_flags & (1 << i))
3283 seq_printf(m, "%s\n", trace_options[i]);
3285 seq_printf(m, "no%s\n", trace_options[i]);
3288 for (i = 0; trace_opts[i].name; i++) {
3289 if (tracer_flags & trace_opts[i].bit)
3290 seq_printf(m, "%s\n", trace_opts[i].name);
3292 seq_printf(m, "no%s\n", trace_opts[i].name);
3294 mutex_unlock(&trace_types_lock);
3299 static int __set_tracer_option(struct tracer *trace,
3300 struct tracer_flags *tracer_flags,
3301 struct tracer_opt *opts, int neg)
3305 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
3310 tracer_flags->val &= ~opts->bit;
3312 tracer_flags->val |= opts->bit;
3316 /* Try to assign a tracer specific option */
3317 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
3319 struct tracer_flags *tracer_flags = trace->flags;
3320 struct tracer_opt *opts = NULL;
3323 for (i = 0; tracer_flags->opts[i].name; i++) {
3324 opts = &tracer_flags->opts[i];
3326 if (strcmp(cmp, opts->name) == 0)
3327 return __set_tracer_option(trace, trace->flags,
3334 /* Some tracers require overwrite to stay enabled */
3335 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3337 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3343 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3345 /* do nothing if flag is already set */
3346 if (!!(trace_flags & mask) == !!enabled)
3349 /* Give the tracer a chance to approve the change */
3350 if (tr->current_trace->flag_changed)
3351 if (tr->current_trace->flag_changed(tr->current_trace, mask, !!enabled))
3355 trace_flags |= mask;
3357 trace_flags &= ~mask;
3359 if (mask == TRACE_ITER_RECORD_CMD)
3360 trace_event_enable_cmd_record(enabled);
3362 if (mask == TRACE_ITER_OVERWRITE) {
3363 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3364 #ifdef CONFIG_TRACER_MAX_TRACE
3365 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3369 if (mask == TRACE_ITER_PRINTK)
3370 trace_printk_start_stop_comm(enabled);
3375 static int trace_set_options(struct trace_array *tr, char *option)
3382 cmp = strstrip(option);
3384 if (strncmp(cmp, "no", 2) == 0) {
3389 mutex_lock(&trace_types_lock);
3391 for (i = 0; trace_options[i]; i++) {
3392 if (strcmp(cmp, trace_options[i]) == 0) {
3393 ret = set_tracer_flag(tr, 1 << i, !neg);
3398 /* If no option could be set, test the specific tracer options */
3399 if (!trace_options[i])
3400 ret = set_tracer_option(tr->current_trace, cmp, neg);
3402 mutex_unlock(&trace_types_lock);
3408 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3409 size_t cnt, loff_t *ppos)
3411 struct seq_file *m = filp->private_data;
3412 struct trace_array *tr = m->private;
3416 if (cnt >= sizeof(buf))
3419 if (copy_from_user(&buf, ubuf, cnt))
3424 ret = trace_set_options(tr, buf);
3433 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3435 struct trace_array *tr = inode->i_private;
3438 if (tracing_disabled)
3441 if (trace_array_get(tr) < 0)
3444 ret = single_open(file, tracing_trace_options_show, inode->i_private);
3446 trace_array_put(tr);
3451 static const struct file_operations tracing_iter_fops = {
3452 .open = tracing_trace_options_open,
3454 .llseek = seq_lseek,
3455 .release = tracing_single_release_tr,
3456 .write = tracing_trace_options_write,
3459 static const char readme_msg[] =
3460 "tracing mini-HOWTO:\n\n"
3461 "# echo 0 > tracing_on : quick way to disable tracing\n"
3462 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3463 " Important files:\n"
3464 " trace\t\t\t- The static contents of the buffer\n"
3465 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3466 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3467 " current_tracer\t- function and latency tracers\n"
3468 " available_tracers\t- list of configured tracers for current_tracer\n"
3469 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3470 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3471 " trace_clock\t\t-change the clock used to order events\n"
3472 " local: Per cpu clock but may not be synced across CPUs\n"
3473 " global: Synced across CPUs but slows tracing down.\n"
3474 " counter: Not a clock, but just an increment\n"
3475 " uptime: Jiffy counter from time of boot\n"
3476 " perf: Same clock that perf events use\n"
3477 #ifdef CONFIG_X86_64
3478 " x86-tsc: TSC cycle counter\n"
3480 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3481 " tracing_cpumask\t- Limit which CPUs to trace\n"
3482 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3483 "\t\t\t Remove sub-buffer with rmdir\n"
3484 " trace_options\t\t- Set format or modify how tracing happens\n"
3485 "\t\t\t Disable an option by adding a suffix 'no' to the option name\n"
3486 #ifdef CONFIG_DYNAMIC_FTRACE
3487 "\n available_filter_functions - list of functions that can be filtered on\n"
3488 " set_ftrace_filter\t- echo function name in here to only trace these functions\n"
3489 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3490 " modules: Can select a group via module\n"
3491 " Format: :mod:<module-name>\n"
3492 " example: echo :mod:ext3 > set_ftrace_filter\n"
3493 " triggers: a command to perform when function is hit\n"
3494 " Format: <function>:<trigger>[:count]\n"
3495 " trigger: traceon, traceoff\n"
3496 " enable_event:<system>:<event>\n"
3497 " disable_event:<system>:<event>\n"
3498 #ifdef CONFIG_STACKTRACE
3501 #ifdef CONFIG_TRACER_SNAPSHOT
3504 " example: echo do_fault:traceoff > set_ftrace_filter\n"
3505 " echo do_trap:traceoff:3 > set_ftrace_filter\n"
3506 " The first one will disable tracing every time do_fault is hit\n"
3507 " The second will disable tracing at most 3 times when do_trap is hit\n"
3508 " The first time do trap is hit and it disables tracing, the counter\n"
3509 " will decrement to 2. If tracing is already disabled, the counter\n"
3510 " will not decrement. It only decrements when the trigger did work\n"
3511 " To remove trigger without count:\n"
3512 " echo '!<function>:<trigger> > set_ftrace_filter\n"
3513 " To remove trigger with a count:\n"
3514 " echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3515 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3516 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3517 " modules: Can select a group via module command :mod:\n"
3518 " Does not accept triggers\n"
3519 #endif /* CONFIG_DYNAMIC_FTRACE */
3520 #ifdef CONFIG_FUNCTION_TRACER
3521 " set_ftrace_pid\t- Write pid(s) to only function trace those pids (function)\n"
3523 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3524 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3525 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3527 #ifdef CONFIG_TRACER_SNAPSHOT
3528 "\n snapshot\t\t- Like 'trace' but shows the content of the static snapshot buffer\n"
3529 "\t\t\t Read the contents for more information\n"
3531 #ifdef CONFIG_STACK_TRACER
3532 " stack_trace\t\t- Shows the max stack trace when active\n"
3533 " stack_max_size\t- Shows current max stack size that was traced\n"
3534 "\t\t\t Write into this file to reset the max size (trigger a new trace)\n"
3535 #ifdef CONFIG_DYNAMIC_FTRACE
3536 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace traces\n"
3538 #endif /* CONFIG_STACK_TRACER */
3542 tracing_readme_read(struct file *filp, char __user *ubuf,
3543 size_t cnt, loff_t *ppos)
3545 return simple_read_from_buffer(ubuf, cnt, ppos,
3546 readme_msg, strlen(readme_msg));
3549 static const struct file_operations tracing_readme_fops = {
3550 .open = tracing_open_generic,
3551 .read = tracing_readme_read,
3552 .llseek = generic_file_llseek,
3556 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3557 size_t cnt, loff_t *ppos)
3566 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3570 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3578 for (i = 0; i < SAVED_CMDLINES; i++) {
3581 pid = map_cmdline_to_pid[i];
3582 if (pid == -1 || pid == NO_CMDLINE_MAP)
3585 trace_find_cmdline(pid, buf_comm);
3586 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3591 len = simple_read_from_buffer(ubuf, cnt, ppos,
3600 static const struct file_operations tracing_saved_cmdlines_fops = {
3601 .open = tracing_open_generic,
3602 .read = tracing_saved_cmdlines_read,
3603 .llseek = generic_file_llseek,
3607 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3608 size_t cnt, loff_t *ppos)
3610 struct trace_array *tr = filp->private_data;
3611 char buf[MAX_TRACER_SIZE+2];
3614 mutex_lock(&trace_types_lock);
3615 r = sprintf(buf, "%s\n", tr->current_trace->name);
3616 mutex_unlock(&trace_types_lock);
3618 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3621 int tracer_init(struct tracer *t, struct trace_array *tr)
3623 tracing_reset_online_cpus(&tr->trace_buffer);
3627 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3631 for_each_tracing_cpu(cpu)
3632 per_cpu_ptr(buf->data, cpu)->entries = val;
3635 #ifdef CONFIG_TRACER_MAX_TRACE
3636 /* resize @tr's buffer to the size of @size_tr's entries */
3637 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3638 struct trace_buffer *size_buf, int cpu_id)
3642 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3643 for_each_tracing_cpu(cpu) {
3644 ret = ring_buffer_resize(trace_buf->buffer,
3645 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3648 per_cpu_ptr(trace_buf->data, cpu)->entries =
3649 per_cpu_ptr(size_buf->data, cpu)->entries;
3652 ret = ring_buffer_resize(trace_buf->buffer,
3653 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3655 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3656 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3661 #endif /* CONFIG_TRACER_MAX_TRACE */
3663 static int __tracing_resize_ring_buffer(struct trace_array *tr,
3664 unsigned long size, int cpu)
3669 * If kernel or user changes the size of the ring buffer
3670 * we use the size that was given, and we can forget about
3671 * expanding it later.
3673 ring_buffer_expanded = true;
3675 /* May be called before buffers are initialized */
3676 if (!tr->trace_buffer.buffer)
3679 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3683 #ifdef CONFIG_TRACER_MAX_TRACE
3684 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3685 !tr->current_trace->use_max_tr)
3688 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3690 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3691 &tr->trace_buffer, cpu);
3694 * AARGH! We are left with different
3695 * size max buffer!!!!
3696 * The max buffer is our "snapshot" buffer.
3697 * When a tracer needs a snapshot (one of the
3698 * latency tracers), it swaps the max buffer
3699 * with the saved snap shot. We succeeded to
3700 * update the size of the main buffer, but failed to
3701 * update the size of the max buffer. But when we tried
3702 * to reset the main buffer to the original size, we
3703 * failed there too. This is very unlikely to
3704 * happen, but if it does, warn and kill all
3708 tracing_disabled = 1;
3713 if (cpu == RING_BUFFER_ALL_CPUS)
3714 set_buffer_entries(&tr->max_buffer, size);
3716 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3719 #endif /* CONFIG_TRACER_MAX_TRACE */
3721 if (cpu == RING_BUFFER_ALL_CPUS)
3722 set_buffer_entries(&tr->trace_buffer, size);
3724 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3729 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3730 unsigned long size, int cpu_id)
3734 mutex_lock(&trace_types_lock);
3736 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3737 /* make sure, this cpu is enabled in the mask */
3738 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3744 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3749 mutex_unlock(&trace_types_lock);
3756 * tracing_update_buffers - used by tracing facility to expand ring buffers
3758 * To save on memory when the tracing is never used on a system with it
3759 * configured in. The ring buffers are set to a minimum size. But once
3760 * a user starts to use the tracing facility, then they need to grow
3761 * to their default size.
3763 * This function is to be called when a tracer is about to be used.
3765 int tracing_update_buffers(void)
3769 mutex_lock(&trace_types_lock);
3770 if (!ring_buffer_expanded)
3771 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3772 RING_BUFFER_ALL_CPUS);
3773 mutex_unlock(&trace_types_lock);
3778 struct trace_option_dentry;
3780 static struct trace_option_dentry *
3781 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3784 destroy_trace_option_files(struct trace_option_dentry *topts);
3786 static int tracing_set_tracer(const char *buf)
3788 static struct trace_option_dentry *topts;
3789 struct trace_array *tr = &global_trace;
3791 #ifdef CONFIG_TRACER_MAX_TRACE
3796 mutex_lock(&trace_types_lock);
3798 if (!ring_buffer_expanded) {
3799 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3800 RING_BUFFER_ALL_CPUS);
3806 for (t = trace_types; t; t = t->next) {
3807 if (strcmp(t->name, buf) == 0)
3814 if (t == tr->current_trace)
3817 trace_branch_disable();
3819 tr->current_trace->enabled = false;
3821 if (tr->current_trace->reset)
3822 tr->current_trace->reset(tr);
3824 /* Current trace needs to be nop_trace before synchronize_sched */
3825 tr->current_trace = &nop_trace;
3827 #ifdef CONFIG_TRACER_MAX_TRACE
3828 had_max_tr = tr->allocated_snapshot;
3830 if (had_max_tr && !t->use_max_tr) {
3832 * We need to make sure that the update_max_tr sees that
3833 * current_trace changed to nop_trace to keep it from
3834 * swapping the buffers after we resize it.
3835 * The update_max_tr is called from interrupts disabled
3836 * so a synchronized_sched() is sufficient.
3838 synchronize_sched();
3842 destroy_trace_option_files(topts);
3844 topts = create_trace_option_files(tr, t);
3846 #ifdef CONFIG_TRACER_MAX_TRACE
3847 if (t->use_max_tr && !had_max_tr) {
3848 ret = alloc_snapshot(tr);
3855 ret = tracer_init(t, tr);
3860 tr->current_trace = t;
3861 tr->current_trace->enabled = true;
3862 trace_branch_enable(tr);
3864 mutex_unlock(&trace_types_lock);
3870 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3871 size_t cnt, loff_t *ppos)
3873 char buf[MAX_TRACER_SIZE+1];
3880 if (cnt > MAX_TRACER_SIZE)
3881 cnt = MAX_TRACER_SIZE;
3883 if (copy_from_user(&buf, ubuf, cnt))
3888 /* strip ending whitespace. */
3889 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3892 err = tracing_set_tracer(buf);
3902 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3903 size_t cnt, loff_t *ppos)
3905 unsigned long *ptr = filp->private_data;
3909 r = snprintf(buf, sizeof(buf), "%ld\n",
3910 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3911 if (r > sizeof(buf))
3913 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3917 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3918 size_t cnt, loff_t *ppos)
3920 unsigned long *ptr = filp->private_data;
3924 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3933 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3935 struct trace_array *tr = inode->i_private;
3936 struct trace_iterator *iter;
3939 if (tracing_disabled)
3942 if (trace_array_get(tr) < 0)
3945 mutex_lock(&trace_types_lock);
3947 /* create a buffer to store the information to pass to userspace */
3948 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3951 __trace_array_put(tr);
3956 * We make a copy of the current tracer to avoid concurrent
3957 * changes on it while we are reading.
3959 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3964 *iter->trace = *tr->current_trace;
3966 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3971 /* trace pipe does not show start of buffer */
3972 cpumask_setall(iter->started);
3974 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3975 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3977 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3978 if (trace_clocks[tr->clock_id].in_ns)
3979 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3982 iter->trace_buffer = &tr->trace_buffer;
3983 iter->cpu_file = tracing_get_cpu(inode);
3984 mutex_init(&iter->mutex);
3985 filp->private_data = iter;
3987 if (iter->trace->pipe_open)
3988 iter->trace->pipe_open(iter);
3990 nonseekable_open(inode, filp);
3992 mutex_unlock(&trace_types_lock);
3998 __trace_array_put(tr);
3999 mutex_unlock(&trace_types_lock);
4003 static int tracing_release_pipe(struct inode *inode, struct file *file)
4005 struct trace_iterator *iter = file->private_data;
4006 struct trace_array *tr = inode->i_private;
4008 mutex_lock(&trace_types_lock);
4010 if (iter->trace->pipe_close)
4011 iter->trace->pipe_close(iter);
4013 mutex_unlock(&trace_types_lock);
4015 free_cpumask_var(iter->started);
4016 mutex_destroy(&iter->mutex);
4020 trace_array_put(tr);
4026 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
4028 /* Iterators are static, they should be filled or empty */
4029 if (trace_buffer_iter(iter, iter->cpu_file))
4030 return POLLIN | POLLRDNORM;
4032 if (trace_flags & TRACE_ITER_BLOCK)
4034 * Always select as readable when in blocking mode
4036 return POLLIN | POLLRDNORM;
4038 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
4043 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
4045 struct trace_iterator *iter = filp->private_data;
4047 return trace_poll(iter, filp, poll_table);
4051 * This is a make-shift waitqueue.
4052 * A tracer might use this callback on some rare cases:
4054 * 1) the current tracer might hold the runqueue lock when it wakes up
4055 * a reader, hence a deadlock (sched, function, and function graph tracers)
4056 * 2) the function tracers, trace all functions, we don't want
4057 * the overhead of calling wake_up and friends
4058 * (and tracing them too)
4060 * Anyway, this is really very primitive wakeup.
4062 void poll_wait_pipe(struct trace_iterator *iter)
4064 set_current_state(TASK_INTERRUPTIBLE);
4065 /* sleep for 100 msecs, and try again. */
4066 schedule_timeout(HZ / 10);
4069 /* Must be called with trace_types_lock mutex held. */
4070 static int tracing_wait_pipe(struct file *filp)
4072 struct trace_iterator *iter = filp->private_data;
4074 while (trace_empty(iter)) {
4076 if ((filp->f_flags & O_NONBLOCK)) {
4080 mutex_unlock(&iter->mutex);
4082 iter->trace->wait_pipe(iter);
4084 mutex_lock(&iter->mutex);
4086 if (signal_pending(current))
4090 * We block until we read something and tracing is disabled.
4091 * We still block if tracing is disabled, but we have never
4092 * read anything. This allows a user to cat this file, and
4093 * then enable tracing. But after we have read something,
4094 * we give an EOF when tracing is again disabled.
4096 * iter->pos will be 0 if we haven't read anything.
4098 if (!tracing_is_on() && iter->pos)
4109 tracing_read_pipe(struct file *filp, char __user *ubuf,
4110 size_t cnt, loff_t *ppos)
4112 struct trace_iterator *iter = filp->private_data;
4113 struct trace_array *tr = iter->tr;
4116 /* return any leftover data */
4117 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4121 trace_seq_init(&iter->seq);
4123 /* copy the tracer to avoid using a global lock all around */
4124 mutex_lock(&trace_types_lock);
4125 if (unlikely(iter->trace->name != tr->current_trace->name))
4126 *iter->trace = *tr->current_trace;
4127 mutex_unlock(&trace_types_lock);
4130 * Avoid more than one consumer on a single file descriptor
4131 * This is just a matter of traces coherency, the ring buffer itself
4134 mutex_lock(&iter->mutex);
4135 if (iter->trace->read) {
4136 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
4142 sret = tracing_wait_pipe(filp);
4146 /* stop when tracing is finished */
4147 if (trace_empty(iter)) {
4152 if (cnt >= PAGE_SIZE)
4153 cnt = PAGE_SIZE - 1;
4155 /* reset all but tr, trace, and overruns */
4156 memset(&iter->seq, 0,
4157 sizeof(struct trace_iterator) -
4158 offsetof(struct trace_iterator, seq));
4161 trace_event_read_lock();
4162 trace_access_lock(iter->cpu_file);
4163 while (trace_find_next_entry_inc(iter) != NULL) {
4164 enum print_line_t ret;
4165 int len = iter->seq.len;
4167 ret = print_trace_line(iter);
4168 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4169 /* don't print partial lines */
4170 iter->seq.len = len;
4173 if (ret != TRACE_TYPE_NO_CONSUME)
4174 trace_consume(iter);
4176 if (iter->seq.len >= cnt)
4180 * Setting the full flag means we reached the trace_seq buffer
4181 * size and we should leave by partial output condition above.
4182 * One of the trace_seq_* functions is not used properly.
4184 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4187 trace_access_unlock(iter->cpu_file);
4188 trace_event_read_unlock();
4190 /* Now copy what we have to the user */
4191 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4192 if (iter->seq.readpos >= iter->seq.len)
4193 trace_seq_init(&iter->seq);
4196 * If there was nothing to send to user, in spite of consuming trace
4197 * entries, go back to wait for more entries.
4203 mutex_unlock(&iter->mutex);
4208 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
4209 struct pipe_buffer *buf)
4211 __free_page(buf->page);
4214 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4217 __free_page(spd->pages[idx]);
4220 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4222 .map = generic_pipe_buf_map,
4223 .unmap = generic_pipe_buf_unmap,
4224 .confirm = generic_pipe_buf_confirm,
4225 .release = tracing_pipe_buf_release,
4226 .steal = generic_pipe_buf_steal,
4227 .get = generic_pipe_buf_get,
4231 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4236 /* Seq buffer is page-sized, exactly what we need. */
4238 count = iter->seq.len;
4239 ret = print_trace_line(iter);
4240 count = iter->seq.len - count;
4243 iter->seq.len -= count;
4246 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4247 iter->seq.len -= count;
4251 if (ret != TRACE_TYPE_NO_CONSUME)
4252 trace_consume(iter);
4254 if (!trace_find_next_entry_inc(iter)) {
4264 static ssize_t tracing_splice_read_pipe(struct file *filp,
4266 struct pipe_inode_info *pipe,
4270 struct page *pages_def[PIPE_DEF_BUFFERS];
4271 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4272 struct trace_iterator *iter = filp->private_data;
4273 struct splice_pipe_desc spd = {
4275 .partial = partial_def,
4276 .nr_pages = 0, /* This gets updated below. */
4277 .nr_pages_max = PIPE_DEF_BUFFERS,
4279 .ops = &tracing_pipe_buf_ops,
4280 .spd_release = tracing_spd_release_pipe,
4282 struct trace_array *tr = iter->tr;
4287 if (splice_grow_spd(pipe, &spd))
4290 /* copy the tracer to avoid using a global lock all around */
4291 mutex_lock(&trace_types_lock);
4292 if (unlikely(iter->trace->name != tr->current_trace->name))
4293 *iter->trace = *tr->current_trace;
4294 mutex_unlock(&trace_types_lock);
4296 mutex_lock(&iter->mutex);
4298 if (iter->trace->splice_read) {
4299 ret = iter->trace->splice_read(iter, filp,
4300 ppos, pipe, len, flags);
4305 ret = tracing_wait_pipe(filp);
4309 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4314 trace_event_read_lock();
4315 trace_access_lock(iter->cpu_file);
4317 /* Fill as many pages as possible. */
4318 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
4319 spd.pages[i] = alloc_page(GFP_KERNEL);
4323 rem = tracing_fill_pipe_page(rem, iter);
4325 /* Copy the data into the page, so we can start over. */
4326 ret = trace_seq_to_buffer(&iter->seq,
4327 page_address(spd.pages[i]),
4330 __free_page(spd.pages[i]);
4333 spd.partial[i].offset = 0;
4334 spd.partial[i].len = iter->seq.len;
4336 trace_seq_init(&iter->seq);
4339 trace_access_unlock(iter->cpu_file);
4340 trace_event_read_unlock();
4341 mutex_unlock(&iter->mutex);
4345 ret = splice_to_pipe(pipe, &spd);
4347 splice_shrink_spd(&spd);
4351 mutex_unlock(&iter->mutex);
4356 tracing_entries_read(struct file *filp, char __user *ubuf,
4357 size_t cnt, loff_t *ppos)
4359 struct inode *inode = file_inode(filp);
4360 struct trace_array *tr = inode->i_private;
4361 int cpu = tracing_get_cpu(inode);
4366 mutex_lock(&trace_types_lock);
4368 if (cpu == RING_BUFFER_ALL_CPUS) {
4369 int cpu, buf_size_same;
4374 /* check if all cpu sizes are same */
4375 for_each_tracing_cpu(cpu) {
4376 /* fill in the size from first enabled cpu */
4378 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4379 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4385 if (buf_size_same) {
4386 if (!ring_buffer_expanded)
4387 r = sprintf(buf, "%lu (expanded: %lu)\n",
4389 trace_buf_size >> 10);
4391 r = sprintf(buf, "%lu\n", size >> 10);
4393 r = sprintf(buf, "X\n");
4395 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10);
4397 mutex_unlock(&trace_types_lock);
4399 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4404 tracing_entries_write(struct file *filp, const char __user *ubuf,
4405 size_t cnt, loff_t *ppos)
4407 struct inode *inode = file_inode(filp);
4408 struct trace_array *tr = inode->i_private;
4412 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4416 /* must have at least 1 entry */
4420 /* value is in KB */
4422 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
4432 tracing_total_entries_read(struct file *filp, char __user *ubuf,
4433 size_t cnt, loff_t *ppos)
4435 struct trace_array *tr = filp->private_data;
4438 unsigned long size = 0, expanded_size = 0;
4440 mutex_lock(&trace_types_lock);
4441 for_each_tracing_cpu(cpu) {
4442 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
4443 if (!ring_buffer_expanded)
4444 expanded_size += trace_buf_size >> 10;
4446 if (ring_buffer_expanded)
4447 r = sprintf(buf, "%lu\n", size);
4449 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
4450 mutex_unlock(&trace_types_lock);
4452 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4456 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4457 size_t cnt, loff_t *ppos)
4460 * There is no need to read what the user has written, this function
4461 * is just to make sure that there is no error when "echo" is used
4470 tracing_free_buffer_release(struct inode *inode, struct file *filp)
4472 struct trace_array *tr = inode->i_private;
4474 /* disable tracing ? */
4475 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4477 /* resize the ring buffer to 0 */
4478 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4480 trace_array_put(tr);
4486 tracing_mark_write(struct file *filp, const char __user *ubuf,
4487 size_t cnt, loff_t *fpos)
4489 unsigned long addr = (unsigned long)ubuf;
4490 struct trace_array *tr = filp->private_data;
4491 struct ring_buffer_event *event;
4492 struct ring_buffer *buffer;
4493 struct print_entry *entry;
4494 unsigned long irq_flags;
4495 struct page *pages[2];
4505 if (tracing_disabled)
4508 if (!(trace_flags & TRACE_ITER_MARKERS))
4511 if (cnt > TRACE_BUF_SIZE)
4512 cnt = TRACE_BUF_SIZE;
4515 * Userspace is injecting traces into the kernel trace buffer.
4516 * We want to be as non intrusive as possible.
4517 * To do so, we do not want to allocate any special buffers
4518 * or take any locks, but instead write the userspace data
4519 * straight into the ring buffer.
4521 * First we need to pin the userspace buffer into memory,
4522 * which, most likely it is, because it just referenced it.
4523 * But there's no guarantee that it is. By using get_user_pages_fast()
4524 * and kmap_atomic/kunmap_atomic() we can get access to the
4525 * pages directly. We then write the data directly into the
4528 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4530 /* check if we cross pages */
4531 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4534 offset = addr & (PAGE_SIZE - 1);
4537 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4538 if (ret < nr_pages) {
4540 put_page(pages[ret]);
4545 for (i = 0; i < nr_pages; i++)
4546 map_page[i] = kmap_atomic(pages[i]);
4548 local_save_flags(irq_flags);
4549 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4550 buffer = tr->trace_buffer.buffer;
4551 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4552 irq_flags, preempt_count());
4554 /* Ring buffer disabled, return as if not open for write */
4559 entry = ring_buffer_event_data(event);
4560 entry->ip = _THIS_IP_;
4562 if (nr_pages == 2) {
4563 len = PAGE_SIZE - offset;
4564 memcpy(&entry->buf, map_page[0] + offset, len);
4565 memcpy(&entry->buf[len], map_page[1], cnt - len);
4567 memcpy(&entry->buf, map_page[0] + offset, cnt);
4569 if (entry->buf[cnt - 1] != '\n') {
4570 entry->buf[cnt] = '\n';
4571 entry->buf[cnt + 1] = '\0';
4573 entry->buf[cnt] = '\0';
4575 __buffer_unlock_commit(buffer, event);
4582 for (i = 0; i < nr_pages; i++){
4583 kunmap_atomic(map_page[i]);
4590 static int tracing_clock_show(struct seq_file *m, void *v)
4592 struct trace_array *tr = m->private;
4595 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4597 "%s%s%s%s", i ? " " : "",
4598 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4599 i == tr->clock_id ? "]" : "");
4605 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4606 size_t cnt, loff_t *fpos)
4608 struct seq_file *m = filp->private_data;
4609 struct trace_array *tr = m->private;
4611 const char *clockstr;
4614 if (cnt >= sizeof(buf))
4617 if (copy_from_user(&buf, ubuf, cnt))
4622 clockstr = strstrip(buf);
4624 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4625 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4628 if (i == ARRAY_SIZE(trace_clocks))
4631 mutex_lock(&trace_types_lock);
4635 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4638 * New clock may not be consistent with the previous clock.
4639 * Reset the buffer so that it doesn't have incomparable timestamps.
4641 tracing_reset_online_cpus(&global_trace.trace_buffer);
4643 #ifdef CONFIG_TRACER_MAX_TRACE
4644 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4645 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4646 tracing_reset_online_cpus(&global_trace.max_buffer);
4649 mutex_unlock(&trace_types_lock);
4656 static int tracing_clock_open(struct inode *inode, struct file *file)
4658 struct trace_array *tr = inode->i_private;
4661 if (tracing_disabled)
4664 if (trace_array_get(tr))
4667 ret = single_open(file, tracing_clock_show, inode->i_private);
4669 trace_array_put(tr);
4674 struct ftrace_buffer_info {
4675 struct trace_iterator iter;
4680 #ifdef CONFIG_TRACER_SNAPSHOT
4681 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4683 struct trace_cpu *tc = inode->i_private;
4684 struct trace_array *tr = tc->tr;
4685 struct trace_iterator *iter;
4689 if (trace_array_get(tr) < 0)
4692 if (file->f_mode & FMODE_READ) {
4693 iter = __tracing_open(tr, tc, inode, file, true);
4695 ret = PTR_ERR(iter);
4697 /* Writes still need the seq_file to hold the private data */
4699 m = kzalloc(sizeof(*m), GFP_KERNEL);
4702 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4710 iter->trace_buffer = &tc->tr->max_buffer;
4711 iter->cpu_file = tc->cpu;
4713 file->private_data = m;
4717 trace_array_put(tr);
4723 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4726 struct seq_file *m = filp->private_data;
4727 struct trace_iterator *iter = m->private;
4728 struct trace_array *tr = iter->tr;
4732 ret = tracing_update_buffers();
4736 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4740 mutex_lock(&trace_types_lock);
4742 if (tr->current_trace->use_max_tr) {
4749 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4753 if (tr->allocated_snapshot)
4757 /* Only allow per-cpu swap if the ring buffer supports it */
4758 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4759 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4764 if (!tr->allocated_snapshot) {
4765 ret = alloc_snapshot(tr);
4769 local_irq_disable();
4770 /* Now, we're going to swap */
4771 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4772 update_max_tr(tr, current, smp_processor_id());
4774 update_max_tr_single(tr, current, iter->cpu_file);
4778 if (tr->allocated_snapshot) {
4779 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4780 tracing_reset_online_cpus(&tr->max_buffer);
4782 tracing_reset(&tr->max_buffer, iter->cpu_file);
4792 mutex_unlock(&trace_types_lock);
4796 static int tracing_snapshot_release(struct inode *inode, struct file *file)
4798 struct seq_file *m = file->private_data;
4801 ret = tracing_release(inode, file);
4803 if (file->f_mode & FMODE_READ)
4806 /* If write only, the seq_file is just a stub */
4814 static int tracing_buffers_open(struct inode *inode, struct file *filp);
4815 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4816 size_t count, loff_t *ppos);
4817 static int tracing_buffers_release(struct inode *inode, struct file *file);
4818 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4819 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
4821 static int snapshot_raw_open(struct inode *inode, struct file *filp)
4823 struct ftrace_buffer_info *info;
4826 ret = tracing_buffers_open(inode, filp);
4830 info = filp->private_data;
4832 if (info->iter.trace->use_max_tr) {
4833 tracing_buffers_release(inode, filp);
4837 info->iter.snapshot = true;
4838 info->iter.trace_buffer = &info->iter.tr->max_buffer;
4843 #endif /* CONFIG_TRACER_SNAPSHOT */
4846 static const struct file_operations tracing_max_lat_fops = {
4847 .open = tracing_open_generic,
4848 .read = tracing_max_lat_read,
4849 .write = tracing_max_lat_write,
4850 .llseek = generic_file_llseek,
4853 static const struct file_operations set_tracer_fops = {
4854 .open = tracing_open_generic,
4855 .read = tracing_set_trace_read,
4856 .write = tracing_set_trace_write,
4857 .llseek = generic_file_llseek,
4860 static const struct file_operations tracing_pipe_fops = {
4861 .open = tracing_open_pipe,
4862 .poll = tracing_poll_pipe,
4863 .read = tracing_read_pipe,
4864 .splice_read = tracing_splice_read_pipe,
4865 .release = tracing_release_pipe,
4866 .llseek = no_llseek,
4869 static const struct file_operations tracing_entries_fops = {
4870 .open = tracing_open_generic_tr,
4871 .read = tracing_entries_read,
4872 .write = tracing_entries_write,
4873 .llseek = generic_file_llseek,
4874 .release = tracing_release_generic_tr,
4877 static const struct file_operations tracing_total_entries_fops = {
4878 .open = tracing_open_generic_tr,
4879 .read = tracing_total_entries_read,
4880 .llseek = generic_file_llseek,
4881 .release = tracing_release_generic_tr,
4884 static const struct file_operations tracing_free_buffer_fops = {
4885 .open = tracing_open_generic_tr,
4886 .write = tracing_free_buffer_write,
4887 .release = tracing_free_buffer_release,
4890 static const struct file_operations tracing_mark_fops = {
4891 .open = tracing_open_generic_tr,
4892 .write = tracing_mark_write,
4893 .llseek = generic_file_llseek,
4894 .release = tracing_release_generic_tr,
4897 static const struct file_operations trace_clock_fops = {
4898 .open = tracing_clock_open,
4900 .llseek = seq_lseek,
4901 .release = tracing_single_release_tr,
4902 .write = tracing_clock_write,
4905 #ifdef CONFIG_TRACER_SNAPSHOT
4906 static const struct file_operations snapshot_fops = {
4907 .open = tracing_snapshot_open,
4909 .write = tracing_snapshot_write,
4910 .llseek = tracing_seek,
4911 .release = tracing_snapshot_release,
4914 static const struct file_operations snapshot_raw_fops = {
4915 .open = snapshot_raw_open,
4916 .read = tracing_buffers_read,
4917 .release = tracing_buffers_release,
4918 .splice_read = tracing_buffers_splice_read,
4919 .llseek = no_llseek,
4922 #endif /* CONFIG_TRACER_SNAPSHOT */
4924 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4926 struct trace_array *tr = inode->i_private;
4927 struct ftrace_buffer_info *info;
4930 if (tracing_disabled)
4933 if (trace_array_get(tr) < 0)
4936 info = kzalloc(sizeof(*info), GFP_KERNEL);
4938 trace_array_put(tr);
4942 mutex_lock(&trace_types_lock);
4945 info->iter.cpu_file = tracing_get_cpu(inode);
4946 info->iter.trace = tr->current_trace;
4947 info->iter.trace_buffer = &tr->trace_buffer;
4949 /* Force reading ring buffer for first read */
4950 info->read = (unsigned int)-1;
4952 filp->private_data = info;
4954 mutex_unlock(&trace_types_lock);
4956 ret = nonseekable_open(inode, filp);
4958 trace_array_put(tr);
4964 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
4966 struct ftrace_buffer_info *info = filp->private_data;
4967 struct trace_iterator *iter = &info->iter;
4969 return trace_poll(iter, filp, poll_table);
4973 tracing_buffers_read(struct file *filp, char __user *ubuf,
4974 size_t count, loff_t *ppos)
4976 struct ftrace_buffer_info *info = filp->private_data;
4977 struct trace_iterator *iter = &info->iter;
4984 mutex_lock(&trace_types_lock);
4986 #ifdef CONFIG_TRACER_MAX_TRACE
4987 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
4994 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
5000 /* Do we have previous read data to read? */
5001 if (info->read < PAGE_SIZE)
5005 trace_access_lock(iter->cpu_file);
5006 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
5010 trace_access_unlock(iter->cpu_file);
5013 if (trace_empty(iter)) {
5014 if ((filp->f_flags & O_NONBLOCK)) {
5018 mutex_unlock(&trace_types_lock);
5019 iter->trace->wait_pipe(iter);
5020 mutex_lock(&trace_types_lock);
5021 if (signal_pending(current)) {
5033 size = PAGE_SIZE - info->read;
5037 ret = copy_to_user(ubuf, info->spare + info->read, size);
5048 mutex_unlock(&trace_types_lock);
5053 static int tracing_buffers_release(struct inode *inode, struct file *file)
5055 struct ftrace_buffer_info *info = file->private_data;
5056 struct trace_iterator *iter = &info->iter;
5058 mutex_lock(&trace_types_lock);
5060 __trace_array_put(iter->tr);
5063 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
5066 mutex_unlock(&trace_types_lock);
5072 struct ring_buffer *buffer;
5077 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
5078 struct pipe_buffer *buf)
5080 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5085 ring_buffer_free_read_page(ref->buffer, ref->page);
5090 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
5091 struct pipe_buffer *buf)
5093 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5098 /* Pipe buffer operations for a buffer. */
5099 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
5101 .map = generic_pipe_buf_map,
5102 .unmap = generic_pipe_buf_unmap,
5103 .confirm = generic_pipe_buf_confirm,
5104 .release = buffer_pipe_buf_release,
5105 .steal = generic_pipe_buf_steal,
5106 .get = buffer_pipe_buf_get,
5110 * Callback from splice_to_pipe(), if we need to release some pages
5111 * at the end of the spd in case we error'ed out in filling the pipe.
5113 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
5115 struct buffer_ref *ref =
5116 (struct buffer_ref *)spd->partial[i].private;
5121 ring_buffer_free_read_page(ref->buffer, ref->page);
5123 spd->partial[i].private = 0;
5127 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5128 struct pipe_inode_info *pipe, size_t len,
5131 struct ftrace_buffer_info *info = file->private_data;
5132 struct trace_iterator *iter = &info->iter;
5133 struct partial_page partial_def[PIPE_DEF_BUFFERS];
5134 struct page *pages_def[PIPE_DEF_BUFFERS];
5135 struct splice_pipe_desc spd = {
5137 .partial = partial_def,
5138 .nr_pages_max = PIPE_DEF_BUFFERS,
5140 .ops = &buffer_pipe_buf_ops,
5141 .spd_release = buffer_spd_release,
5143 struct buffer_ref *ref;
5144 int entries, size, i;
5147 mutex_lock(&trace_types_lock);
5149 #ifdef CONFIG_TRACER_MAX_TRACE
5150 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5156 if (splice_grow_spd(pipe, &spd)) {
5161 if (*ppos & (PAGE_SIZE - 1)) {
5166 if (len & (PAGE_SIZE - 1)) {
5167 if (len < PAGE_SIZE) {
5175 trace_access_lock(iter->cpu_file);
5176 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5178 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
5182 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
5187 ref->buffer = iter->trace_buffer->buffer;
5188 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5194 r = ring_buffer_read_page(ref->buffer, &ref->page,
5195 len, iter->cpu_file, 1);
5197 ring_buffer_free_read_page(ref->buffer, ref->page);
5203 * zero out any left over data, this is going to
5206 size = ring_buffer_page_len(ref->page);
5207 if (size < PAGE_SIZE)
5208 memset(ref->page + size, 0, PAGE_SIZE - size);
5210 page = virt_to_page(ref->page);
5212 spd.pages[i] = page;
5213 spd.partial[i].len = PAGE_SIZE;
5214 spd.partial[i].offset = 0;
5215 spd.partial[i].private = (unsigned long)ref;
5219 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5222 trace_access_unlock(iter->cpu_file);
5225 /* did we read anything? */
5226 if (!spd.nr_pages) {
5227 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
5231 mutex_unlock(&trace_types_lock);
5232 iter->trace->wait_pipe(iter);
5233 mutex_lock(&trace_types_lock);
5234 if (signal_pending(current)) {
5241 ret = splice_to_pipe(pipe, &spd);
5242 splice_shrink_spd(&spd);
5244 mutex_unlock(&trace_types_lock);
5249 static const struct file_operations tracing_buffers_fops = {
5250 .open = tracing_buffers_open,
5251 .read = tracing_buffers_read,
5252 .poll = tracing_buffers_poll,
5253 .release = tracing_buffers_release,
5254 .splice_read = tracing_buffers_splice_read,
5255 .llseek = no_llseek,
5259 tracing_stats_read(struct file *filp, char __user *ubuf,
5260 size_t count, loff_t *ppos)
5262 struct inode *inode = file_inode(filp);
5263 struct trace_array *tr = inode->i_private;
5264 struct trace_buffer *trace_buf = &tr->trace_buffer;
5265 int cpu = tracing_get_cpu(inode);
5266 struct trace_seq *s;
5268 unsigned long long t;
5269 unsigned long usec_rem;
5271 s = kmalloc(sizeof(*s), GFP_KERNEL);
5277 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5278 trace_seq_printf(s, "entries: %ld\n", cnt);
5280 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5281 trace_seq_printf(s, "overrun: %ld\n", cnt);
5283 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5284 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5286 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5287 trace_seq_printf(s, "bytes: %ld\n", cnt);
5289 if (trace_clocks[tr->clock_id].in_ns) {
5290 /* local or global for trace_clock */
5291 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5292 usec_rem = do_div(t, USEC_PER_SEC);
5293 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5296 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5297 usec_rem = do_div(t, USEC_PER_SEC);
5298 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5300 /* counter or tsc mode for trace_clock */
5301 trace_seq_printf(s, "oldest event ts: %llu\n",
5302 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5304 trace_seq_printf(s, "now ts: %llu\n",
5305 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5308 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5309 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5311 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5312 trace_seq_printf(s, "read events: %ld\n", cnt);
5314 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
5321 static const struct file_operations tracing_stats_fops = {
5322 .open = tracing_open_generic_tr,
5323 .read = tracing_stats_read,
5324 .llseek = generic_file_llseek,
5325 .release = tracing_release_generic_tr,
5328 #ifdef CONFIG_DYNAMIC_FTRACE
5330 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5336 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5337 size_t cnt, loff_t *ppos)
5339 static char ftrace_dyn_info_buffer[1024];
5340 static DEFINE_MUTEX(dyn_info_mutex);
5341 unsigned long *p = filp->private_data;
5342 char *buf = ftrace_dyn_info_buffer;
5343 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5346 mutex_lock(&dyn_info_mutex);
5347 r = sprintf(buf, "%ld ", *p);
5349 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5352 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5354 mutex_unlock(&dyn_info_mutex);
5359 static const struct file_operations tracing_dyn_info_fops = {
5360 .open = tracing_open_generic,
5361 .read = tracing_read_dyn_info,
5362 .llseek = generic_file_llseek,
5364 #endif /* CONFIG_DYNAMIC_FTRACE */
5366 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5368 ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5374 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5376 unsigned long *count = (long *)data;
5388 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5389 struct ftrace_probe_ops *ops, void *data)
5391 long count = (long)data;
5393 seq_printf(m, "%ps:", (void *)ip);
5395 seq_printf(m, "snapshot");
5398 seq_printf(m, ":unlimited\n");
5400 seq_printf(m, ":count=%ld\n", count);
5405 static struct ftrace_probe_ops snapshot_probe_ops = {
5406 .func = ftrace_snapshot,
5407 .print = ftrace_snapshot_print,
5410 static struct ftrace_probe_ops snapshot_count_probe_ops = {
5411 .func = ftrace_count_snapshot,
5412 .print = ftrace_snapshot_print,
5416 ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
5417 char *glob, char *cmd, char *param, int enable)
5419 struct ftrace_probe_ops *ops;
5420 void *count = (void *)-1;
5424 /* hash funcs only work with set_ftrace_filter */
5428 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
5430 if (glob[0] == '!') {
5431 unregister_ftrace_function_probe_func(glob+1, ops);
5438 number = strsep(¶m, ":");
5440 if (!strlen(number))
5444 * We use the callback data field (which is a pointer)
5447 ret = kstrtoul(number, 0, (unsigned long *)&count);
5452 ret = register_ftrace_function_probe(glob, ops, count);
5455 alloc_snapshot(&global_trace);
5457 return ret < 0 ? ret : 0;
5460 static struct ftrace_func_command ftrace_snapshot_cmd = {
5462 .func = ftrace_trace_snapshot_callback,
5465 static int register_snapshot_cmd(void)
5467 return register_ftrace_command(&ftrace_snapshot_cmd);
5470 static inline int register_snapshot_cmd(void) { return 0; }
5471 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5473 struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
5478 if (!debugfs_initialized())
5481 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
5482 tr->dir = debugfs_create_dir("tracing", NULL);
5485 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5490 struct dentry *tracing_init_dentry(void)
5492 return tracing_init_dentry_tr(&global_trace);
5495 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
5497 struct dentry *d_tracer;
5500 return tr->percpu_dir;
5502 d_tracer = tracing_init_dentry_tr(tr);
5506 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
5508 WARN_ONCE(!tr->percpu_dir,
5509 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
5511 return tr->percpu_dir;
5514 static struct dentry *
5515 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
5516 void *data, long cpu, const struct file_operations *fops)
5518 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
5520 if (ret) /* See tracing_get_cpu() */
5521 ret->d_inode->i_cdev = (void *)(cpu + 1);
5526 tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
5528 struct trace_array_cpu *data = per_cpu_ptr(tr->trace_buffer.data, cpu);
5529 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
5530 struct dentry *d_cpu;
5531 char cpu_dir[30]; /* 30 characters should be more than enough */
5536 snprintf(cpu_dir, 30, "cpu%ld", cpu);
5537 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
5539 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
5543 /* per cpu trace_pipe */
5544 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
5545 tr, cpu, &tracing_pipe_fops);
5548 trace_create_cpu_file("trace", 0644, d_cpu,
5549 &data->trace_cpu, cpu, &tracing_fops);
5551 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
5552 tr, cpu, &tracing_buffers_fops);
5554 trace_create_cpu_file("stats", 0444, d_cpu,
5555 tr, cpu, &tracing_stats_fops);
5557 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
5558 tr, cpu, &tracing_entries_fops);
5560 #ifdef CONFIG_TRACER_SNAPSHOT
5561 trace_create_cpu_file("snapshot", 0644, d_cpu,
5562 &data->trace_cpu, cpu, &snapshot_fops);
5564 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
5565 tr, cpu, &snapshot_raw_fops);
5569 #ifdef CONFIG_FTRACE_SELFTEST
5570 /* Let selftest have access to static functions in this file */
5571 #include "trace_selftest.c"
5574 struct trace_option_dentry {
5575 struct tracer_opt *opt;
5576 struct tracer_flags *flags;
5577 struct trace_array *tr;
5578 struct dentry *entry;
5582 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
5585 struct trace_option_dentry *topt = filp->private_data;
5588 if (topt->flags->val & topt->opt->bit)
5593 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5597 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
5600 struct trace_option_dentry *topt = filp->private_data;
5604 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5608 if (val != 0 && val != 1)
5611 if (!!(topt->flags->val & topt->opt->bit) != val) {
5612 mutex_lock(&trace_types_lock);
5613 ret = __set_tracer_option(topt->tr->current_trace, topt->flags,
5615 mutex_unlock(&trace_types_lock);
5626 static const struct file_operations trace_options_fops = {
5627 .open = tracing_open_generic,
5628 .read = trace_options_read,
5629 .write = trace_options_write,
5630 .llseek = generic_file_llseek,
5634 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5637 long index = (long)filp->private_data;
5640 if (trace_flags & (1 << index))
5645 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5649 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5652 struct trace_array *tr = &global_trace;
5653 long index = (long)filp->private_data;
5657 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5661 if (val != 0 && val != 1)
5664 mutex_lock(&trace_types_lock);
5665 ret = set_tracer_flag(tr, 1 << index, val);
5666 mutex_unlock(&trace_types_lock);
5676 static const struct file_operations trace_options_core_fops = {
5677 .open = tracing_open_generic,
5678 .read = trace_options_core_read,
5679 .write = trace_options_core_write,
5680 .llseek = generic_file_llseek,
5683 struct dentry *trace_create_file(const char *name,
5685 struct dentry *parent,
5687 const struct file_operations *fops)
5691 ret = debugfs_create_file(name, mode, parent, data, fops);
5693 pr_warning("Could not create debugfs '%s' entry\n", name);
5699 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5701 struct dentry *d_tracer;
5706 d_tracer = tracing_init_dentry_tr(tr);
5710 tr->options = debugfs_create_dir("options", d_tracer);
5712 pr_warning("Could not create debugfs directory 'options'\n");
5720 create_trace_option_file(struct trace_array *tr,
5721 struct trace_option_dentry *topt,
5722 struct tracer_flags *flags,
5723 struct tracer_opt *opt)
5725 struct dentry *t_options;
5727 t_options = trace_options_init_dentry(tr);
5731 topt->flags = flags;
5735 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5736 &trace_options_fops);
5740 static struct trace_option_dentry *
5741 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5743 struct trace_option_dentry *topts;
5744 struct tracer_flags *flags;
5745 struct tracer_opt *opts;
5751 flags = tracer->flags;
5753 if (!flags || !flags->opts)
5758 for (cnt = 0; opts[cnt].name; cnt++)
5761 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5765 for (cnt = 0; opts[cnt].name; cnt++)
5766 create_trace_option_file(tr, &topts[cnt], flags,
5773 destroy_trace_option_files(struct trace_option_dentry *topts)
5780 for (cnt = 0; topts[cnt].opt; cnt++) {
5781 if (topts[cnt].entry)
5782 debugfs_remove(topts[cnt].entry);
5788 static struct dentry *
5789 create_trace_option_core_file(struct trace_array *tr,
5790 const char *option, long index)
5792 struct dentry *t_options;
5794 t_options = trace_options_init_dentry(tr);
5798 return trace_create_file(option, 0644, t_options, (void *)index,
5799 &trace_options_core_fops);
5802 static __init void create_trace_options_dir(struct trace_array *tr)
5804 struct dentry *t_options;
5807 t_options = trace_options_init_dentry(tr);
5811 for (i = 0; trace_options[i]; i++)
5812 create_trace_option_core_file(tr, trace_options[i], i);
5816 rb_simple_read(struct file *filp, char __user *ubuf,
5817 size_t cnt, loff_t *ppos)
5819 struct trace_array *tr = filp->private_data;
5823 r = tracer_tracing_is_on(tr);
5824 r = sprintf(buf, "%d\n", r);
5826 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5830 rb_simple_write(struct file *filp, const char __user *ubuf,
5831 size_t cnt, loff_t *ppos)
5833 struct trace_array *tr = filp->private_data;
5834 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5838 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5843 mutex_lock(&trace_types_lock);
5845 tracer_tracing_on(tr);
5846 if (tr->current_trace->start)
5847 tr->current_trace->start(tr);
5849 tracer_tracing_off(tr);
5850 if (tr->current_trace->stop)
5851 tr->current_trace->stop(tr);
5853 mutex_unlock(&trace_types_lock);
5861 static const struct file_operations rb_simple_fops = {
5862 .open = tracing_open_generic_tr,
5863 .read = rb_simple_read,
5864 .write = rb_simple_write,
5865 .release = tracing_release_generic_tr,
5866 .llseek = default_llseek,
5869 struct dentry *trace_instance_dir;
5872 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
5874 static void init_trace_buffers(struct trace_array *tr, struct trace_buffer *buf)
5878 for_each_tracing_cpu(cpu) {
5879 memset(per_cpu_ptr(buf->data, cpu), 0, sizeof(struct trace_array_cpu));
5880 per_cpu_ptr(buf->data, cpu)->trace_cpu.cpu = cpu;
5881 per_cpu_ptr(buf->data, cpu)->trace_cpu.tr = tr;
5886 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
5888 enum ring_buffer_flags rb_flags;
5890 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5892 buf->buffer = ring_buffer_alloc(size, rb_flags);
5896 buf->data = alloc_percpu(struct trace_array_cpu);
5898 ring_buffer_free(buf->buffer);
5902 init_trace_buffers(tr, buf);
5904 /* Allocate the first page for all buffers */
5905 set_buffer_entries(&tr->trace_buffer,
5906 ring_buffer_size(tr->trace_buffer.buffer, 0));
5911 static int allocate_trace_buffers(struct trace_array *tr, int size)
5915 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
5919 #ifdef CONFIG_TRACER_MAX_TRACE
5920 ret = allocate_trace_buffer(tr, &tr->max_buffer,
5921 allocate_snapshot ? size : 1);
5923 ring_buffer_free(tr->trace_buffer.buffer);
5924 free_percpu(tr->trace_buffer.data);
5927 tr->allocated_snapshot = allocate_snapshot;
5930 * Only the top level trace array gets its snapshot allocated
5931 * from the kernel command line.
5933 allocate_snapshot = false;
5938 static int new_instance_create(const char *name)
5940 struct trace_array *tr;
5943 mutex_lock(&trace_types_lock);
5946 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5947 if (tr->name && strcmp(tr->name, name) == 0)
5952 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
5956 tr->name = kstrdup(name, GFP_KERNEL);
5960 raw_spin_lock_init(&tr->start_lock);
5962 tr->current_trace = &nop_trace;
5964 INIT_LIST_HEAD(&tr->systems);
5965 INIT_LIST_HEAD(&tr->events);
5967 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
5970 /* Holder for file callbacks */
5971 tr->trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
5972 tr->trace_cpu.tr = tr;
5974 tr->dir = debugfs_create_dir(name, trace_instance_dir);
5978 ret = event_trace_add_tracer(tr->dir, tr);
5980 debugfs_remove_recursive(tr->dir);
5984 init_tracer_debugfs(tr, tr->dir);
5986 list_add(&tr->list, &ftrace_trace_arrays);
5988 mutex_unlock(&trace_types_lock);
5993 if (tr->trace_buffer.buffer)
5994 ring_buffer_free(tr->trace_buffer.buffer);
5999 mutex_unlock(&trace_types_lock);
6005 static int instance_delete(const char *name)
6007 struct trace_array *tr;
6011 mutex_lock(&trace_types_lock);
6014 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6015 if (tr->name && strcmp(tr->name, name) == 0) {
6027 list_del(&tr->list);
6029 event_trace_del_tracer(tr);
6030 debugfs_remove_recursive(tr->dir);
6031 free_percpu(tr->trace_buffer.data);
6032 ring_buffer_free(tr->trace_buffer.buffer);
6040 mutex_unlock(&trace_types_lock);
6045 static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
6047 struct dentry *parent;
6050 /* Paranoid: Make sure the parent is the "instances" directory */
6051 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6052 if (WARN_ON_ONCE(parent != trace_instance_dir))
6056 * The inode mutex is locked, but debugfs_create_dir() will also
6057 * take the mutex. As the instances directory can not be destroyed
6058 * or changed in any other way, it is safe to unlock it, and
6059 * let the dentry try. If two users try to make the same dir at
6060 * the same time, then the new_instance_create() will determine the
6063 mutex_unlock(&inode->i_mutex);
6065 ret = new_instance_create(dentry->d_iname);
6067 mutex_lock(&inode->i_mutex);
6072 static int instance_rmdir(struct inode *inode, struct dentry *dentry)
6074 struct dentry *parent;
6077 /* Paranoid: Make sure the parent is the "instances" directory */
6078 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6079 if (WARN_ON_ONCE(parent != trace_instance_dir))
6082 /* The caller did a dget() on dentry */
6083 mutex_unlock(&dentry->d_inode->i_mutex);
6086 * The inode mutex is locked, but debugfs_create_dir() will also
6087 * take the mutex. As the instances directory can not be destroyed
6088 * or changed in any other way, it is safe to unlock it, and
6089 * let the dentry try. If two users try to make the same dir at
6090 * the same time, then the instance_delete() will determine the
6093 mutex_unlock(&inode->i_mutex);
6095 ret = instance_delete(dentry->d_iname);
6097 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
6098 mutex_lock(&dentry->d_inode->i_mutex);
6103 static const struct inode_operations instance_dir_inode_operations = {
6104 .lookup = simple_lookup,
6105 .mkdir = instance_mkdir,
6106 .rmdir = instance_rmdir,
6109 static __init void create_trace_instances(struct dentry *d_tracer)
6111 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
6112 if (WARN_ON(!trace_instance_dir))
6115 /* Hijack the dir inode operations, to allow mkdir */
6116 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
6120 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
6124 trace_create_file("trace_options", 0644, d_tracer,
6125 tr, &tracing_iter_fops);
6127 trace_create_file("trace", 0644, d_tracer,
6128 (void *)&tr->trace_cpu, &tracing_fops);
6130 trace_create_file("trace_pipe", 0444, d_tracer,
6131 tr, &tracing_pipe_fops);
6133 trace_create_file("buffer_size_kb", 0644, d_tracer,
6134 tr, &tracing_entries_fops);
6136 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
6137 tr, &tracing_total_entries_fops);
6139 trace_create_file("free_buffer", 0200, d_tracer,
6140 tr, &tracing_free_buffer_fops);
6142 trace_create_file("trace_marker", 0220, d_tracer,
6143 tr, &tracing_mark_fops);
6145 trace_create_file("trace_clock", 0644, d_tracer, tr,
6148 trace_create_file("tracing_on", 0644, d_tracer,
6149 tr, &rb_simple_fops);
6151 #ifdef CONFIG_TRACER_SNAPSHOT
6152 trace_create_file("snapshot", 0644, d_tracer,
6153 (void *)&tr->trace_cpu, &snapshot_fops);
6156 for_each_tracing_cpu(cpu)
6157 tracing_init_debugfs_percpu(tr, cpu);
6161 static __init int tracer_init_debugfs(void)
6163 struct dentry *d_tracer;
6165 trace_access_lock_init();
6167 d_tracer = tracing_init_dentry();
6171 init_tracer_debugfs(&global_trace, d_tracer);
6173 trace_create_file("tracing_cpumask", 0644, d_tracer,
6174 &global_trace, &tracing_cpumask_fops);
6176 trace_create_file("available_tracers", 0444, d_tracer,
6177 &global_trace, &show_traces_fops);
6179 trace_create_file("current_tracer", 0644, d_tracer,
6180 &global_trace, &set_tracer_fops);
6182 #ifdef CONFIG_TRACER_MAX_TRACE
6183 trace_create_file("tracing_max_latency", 0644, d_tracer,
6184 &tracing_max_latency, &tracing_max_lat_fops);
6187 trace_create_file("tracing_thresh", 0644, d_tracer,
6188 &tracing_thresh, &tracing_max_lat_fops);
6190 trace_create_file("README", 0444, d_tracer,
6191 NULL, &tracing_readme_fops);
6193 trace_create_file("saved_cmdlines", 0444, d_tracer,
6194 NULL, &tracing_saved_cmdlines_fops);
6196 #ifdef CONFIG_DYNAMIC_FTRACE
6197 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
6198 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
6201 create_trace_instances(d_tracer);
6203 create_trace_options_dir(&global_trace);
6208 static int trace_panic_handler(struct notifier_block *this,
6209 unsigned long event, void *unused)
6211 if (ftrace_dump_on_oops)
6212 ftrace_dump(ftrace_dump_on_oops);
6216 static struct notifier_block trace_panic_notifier = {
6217 .notifier_call = trace_panic_handler,
6219 .priority = 150 /* priority: INT_MAX >= x >= 0 */
6222 static int trace_die_handler(struct notifier_block *self,
6228 if (ftrace_dump_on_oops)
6229 ftrace_dump(ftrace_dump_on_oops);
6237 static struct notifier_block trace_die_notifier = {
6238 .notifier_call = trace_die_handler,
6243 * printk is set to max of 1024, we really don't need it that big.
6244 * Nothing should be printing 1000 characters anyway.
6246 #define TRACE_MAX_PRINT 1000
6249 * Define here KERN_TRACE so that we have one place to modify
6250 * it if we decide to change what log level the ftrace dump
6253 #define KERN_TRACE KERN_EMERG
6256 trace_printk_seq(struct trace_seq *s)
6258 /* Probably should print a warning here. */
6259 if (s->len >= TRACE_MAX_PRINT)
6260 s->len = TRACE_MAX_PRINT;
6262 /* should be zero ended, but we are paranoid. */
6263 s->buffer[s->len] = 0;
6265 printk(KERN_TRACE "%s", s->buffer);
6270 void trace_init_global_iter(struct trace_iterator *iter)
6272 iter->tr = &global_trace;
6273 iter->trace = iter->tr->current_trace;
6274 iter->cpu_file = RING_BUFFER_ALL_CPUS;
6275 iter->trace_buffer = &global_trace.trace_buffer;
6278 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
6280 /* use static because iter can be a bit big for the stack */
6281 static struct trace_iterator iter;
6282 static atomic_t dump_running;
6283 unsigned int old_userobj;
6284 unsigned long flags;
6287 /* Only allow one dump user at a time. */
6288 if (atomic_inc_return(&dump_running) != 1) {
6289 atomic_dec(&dump_running);
6294 * Always turn off tracing when we dump.
6295 * We don't need to show trace output of what happens
6296 * between multiple crashes.
6298 * If the user does a sysrq-z, then they can re-enable
6299 * tracing with echo 1 > tracing_on.
6303 local_irq_save(flags);
6305 /* Simulate the iterator */
6306 trace_init_global_iter(&iter);
6308 for_each_tracing_cpu(cpu) {
6309 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
6312 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
6314 /* don't look at user memory in panic mode */
6315 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
6317 switch (oops_dump_mode) {
6319 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6322 iter.cpu_file = raw_smp_processor_id();
6327 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
6328 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6331 printk(KERN_TRACE "Dumping ftrace buffer:\n");
6333 /* Did function tracer already get disabled? */
6334 if (ftrace_is_dead()) {
6335 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6336 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6340 * We need to stop all tracing on all CPUS to read the
6341 * the next buffer. This is a bit expensive, but is
6342 * not done often. We fill all what we can read,
6343 * and then release the locks again.
6346 while (!trace_empty(&iter)) {
6349 printk(KERN_TRACE "---------------------------------\n");
6353 /* reset all but tr, trace, and overruns */
6354 memset(&iter.seq, 0,
6355 sizeof(struct trace_iterator) -
6356 offsetof(struct trace_iterator, seq));
6357 iter.iter_flags |= TRACE_FILE_LAT_FMT;
6360 if (trace_find_next_entry_inc(&iter) != NULL) {
6363 ret = print_trace_line(&iter);
6364 if (ret != TRACE_TYPE_NO_CONSUME)
6365 trace_consume(&iter);
6367 touch_nmi_watchdog();
6369 trace_printk_seq(&iter.seq);
6373 printk(KERN_TRACE " (ftrace buffer empty)\n");
6375 printk(KERN_TRACE "---------------------------------\n");
6378 trace_flags |= old_userobj;
6380 for_each_tracing_cpu(cpu) {
6381 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
6383 atomic_dec(&dump_running);
6384 local_irq_restore(flags);
6386 EXPORT_SYMBOL_GPL(ftrace_dump);
6388 __init static int tracer_alloc_buffers(void)
6394 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
6397 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
6398 goto out_free_buffer_mask;
6400 /* Only allocate trace_printk buffers if a trace_printk exists */
6401 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
6402 /* Must be called before global_trace.buffer is allocated */
6403 trace_printk_init_buffers();
6405 /* To save memory, keep the ring buffer size to its minimum */
6406 if (ring_buffer_expanded)
6407 ring_buf_size = trace_buf_size;
6411 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
6412 cpumask_copy(tracing_cpumask, cpu_all_mask);
6414 raw_spin_lock_init(&global_trace.start_lock);
6416 /* TODO: make the number of buffers hot pluggable with CPUS */
6417 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
6418 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
6420 goto out_free_cpumask;
6423 if (global_trace.buffer_disabled)
6426 trace_init_cmdlines();
6429 * register_tracer() might reference current_trace, so it
6430 * needs to be set before we register anything. This is
6431 * just a bootstrap of current_trace anyway.
6433 global_trace.current_trace = &nop_trace;
6435 register_tracer(&nop_trace);
6437 /* All seems OK, enable tracing */
6438 tracing_disabled = 0;
6440 atomic_notifier_chain_register(&panic_notifier_list,
6441 &trace_panic_notifier);
6443 register_die_notifier(&trace_die_notifier);
6445 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
6447 /* Holder for file callbacks */
6448 global_trace.trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
6449 global_trace.trace_cpu.tr = &global_trace;
6451 INIT_LIST_HEAD(&global_trace.systems);
6452 INIT_LIST_HEAD(&global_trace.events);
6453 list_add(&global_trace.list, &ftrace_trace_arrays);
6455 while (trace_boot_options) {
6458 option = strsep(&trace_boot_options, ",");
6459 trace_set_options(&global_trace, option);
6462 register_snapshot_cmd();
6467 free_percpu(global_trace.trace_buffer.data);
6468 #ifdef CONFIG_TRACER_MAX_TRACE
6469 free_percpu(global_trace.max_buffer.data);
6471 free_cpumask_var(tracing_cpumask);
6472 out_free_buffer_mask:
6473 free_cpumask_var(tracing_buffer_mask);
6478 __init static int clear_boot_tracer(void)
6481 * The default tracer at boot buffer is an init section.
6482 * This function is called in lateinit. If we did not
6483 * find the boot tracer, then clear it out, to prevent
6484 * later registration from accessing the buffer that is
6485 * about to be freed.
6487 if (!default_bootup_tracer)
6490 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
6491 default_bootup_tracer);
6492 default_bootup_tracer = NULL;
6497 early_initcall(tracer_alloc_buffers);
6498 fs_initcall(tracer_init_debugfs);
6499 late_initcall(clear_boot_tracer);