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 static int tracing_set_tracer(const char *buf);
120 #define MAX_TRACER_SIZE 100
121 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
122 static char *default_bootup_tracer;
124 static bool allocate_snapshot;
126 static int __init set_cmdline_ftrace(char *str)
128 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
129 default_bootup_tracer = bootup_tracer_buf;
130 /* We are using ftrace early, expand it */
131 ring_buffer_expanded = true;
134 __setup("ftrace=", set_cmdline_ftrace);
136 static int __init set_ftrace_dump_on_oops(char *str)
138 if (*str++ != '=' || !*str) {
139 ftrace_dump_on_oops = DUMP_ALL;
143 if (!strcmp("orig_cpu", str)) {
144 ftrace_dump_on_oops = DUMP_ORIG;
150 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
152 static int __init boot_alloc_snapshot(char *str)
154 allocate_snapshot = true;
155 /* We also need the main ring buffer expanded */
156 ring_buffer_expanded = true;
159 __setup("alloc_snapshot", boot_alloc_snapshot);
162 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
163 static char *trace_boot_options __initdata;
165 static int __init set_trace_boot_options(char *str)
167 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
168 trace_boot_options = trace_boot_options_buf;
171 __setup("trace_options=", set_trace_boot_options);
173 unsigned long long ns2usecs(cycle_t nsec)
181 * The global_trace is the descriptor that holds the tracing
182 * buffers for the live tracing. For each CPU, it contains
183 * a link list of pages that will store trace entries. The
184 * page descriptor of the pages in the memory is used to hold
185 * the link list by linking the lru item in the page descriptor
186 * to each of the pages in the buffer per CPU.
188 * For each active CPU there is a data field that holds the
189 * pages for the buffer for that CPU. Each CPU has the same number
190 * of pages allocated for its buffer.
192 static struct trace_array global_trace;
194 LIST_HEAD(ftrace_trace_arrays);
196 int trace_array_get(struct trace_array *this_tr)
198 struct trace_array *tr;
201 mutex_lock(&trace_types_lock);
202 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
209 mutex_unlock(&trace_types_lock);
214 static void __trace_array_put(struct trace_array *this_tr)
216 WARN_ON(!this_tr->ref);
220 void trace_array_put(struct trace_array *this_tr)
222 mutex_lock(&trace_types_lock);
223 __trace_array_put(this_tr);
224 mutex_unlock(&trace_types_lock);
227 int filter_current_check_discard(struct ring_buffer *buffer,
228 struct ftrace_event_call *call, void *rec,
229 struct ring_buffer_event *event)
231 return filter_check_discard(call, rec, buffer, event);
233 EXPORT_SYMBOL_GPL(filter_current_check_discard);
235 cycle_t buffer_ftrace_now(struct trace_buffer *buf, int cpu)
239 /* Early boot up does not have a buffer yet */
241 return trace_clock_local();
243 ts = ring_buffer_time_stamp(buf->buffer, cpu);
244 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
249 cycle_t ftrace_now(int cpu)
251 return buffer_ftrace_now(&global_trace.trace_buffer, cpu);
255 * tracing_is_enabled - Show if global_trace has been disabled
257 * Shows if the global trace has been enabled or not. It uses the
258 * mirror flag "buffer_disabled" to be used in fast paths such as for
259 * the irqsoff tracer. But it may be inaccurate due to races. If you
260 * need to know the accurate state, use tracing_is_on() which is a little
261 * slower, but accurate.
263 int tracing_is_enabled(void)
266 * For quick access (irqsoff uses this in fast path), just
267 * return the mirror variable of the state of the ring buffer.
268 * It's a little racy, but we don't really care.
271 return !global_trace.buffer_disabled;
275 * trace_buf_size is the size in bytes that is allocated
276 * for a buffer. Note, the number of bytes is always rounded
279 * This number is purposely set to a low number of 16384.
280 * If the dump on oops happens, it will be much appreciated
281 * to not have to wait for all that output. Anyway this can be
282 * boot time and run time configurable.
284 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
286 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
288 /* trace_types holds a link list of available tracers. */
289 static struct tracer *trace_types __read_mostly;
292 * trace_types_lock is used to protect the trace_types list.
294 DEFINE_MUTEX(trace_types_lock);
297 * serialize the access of the ring buffer
299 * ring buffer serializes readers, but it is low level protection.
300 * The validity of the events (which returns by ring_buffer_peek() ..etc)
301 * are not protected by ring buffer.
303 * The content of events may become garbage if we allow other process consumes
304 * these events concurrently:
305 * A) the page of the consumed events may become a normal page
306 * (not reader page) in ring buffer, and this page will be rewrited
307 * by events producer.
308 * B) The page of the consumed events may become a page for splice_read,
309 * and this page will be returned to system.
311 * These primitives allow multi process access to different cpu ring buffer
314 * These primitives don't distinguish read-only and read-consume access.
315 * Multi read-only access are also serialized.
319 static DECLARE_RWSEM(all_cpu_access_lock);
320 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
322 static inline void trace_access_lock(int cpu)
324 if (cpu == RING_BUFFER_ALL_CPUS) {
325 /* gain it for accessing the whole ring buffer. */
326 down_write(&all_cpu_access_lock);
328 /* gain it for accessing a cpu ring buffer. */
330 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
331 down_read(&all_cpu_access_lock);
333 /* Secondly block other access to this @cpu ring buffer. */
334 mutex_lock(&per_cpu(cpu_access_lock, cpu));
338 static inline void trace_access_unlock(int cpu)
340 if (cpu == RING_BUFFER_ALL_CPUS) {
341 up_write(&all_cpu_access_lock);
343 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
344 up_read(&all_cpu_access_lock);
348 static inline void trace_access_lock_init(void)
352 for_each_possible_cpu(cpu)
353 mutex_init(&per_cpu(cpu_access_lock, cpu));
358 static DEFINE_MUTEX(access_lock);
360 static inline void trace_access_lock(int cpu)
363 mutex_lock(&access_lock);
366 static inline void trace_access_unlock(int cpu)
369 mutex_unlock(&access_lock);
372 static inline void trace_access_lock_init(void)
378 /* trace_flags holds trace_options default values */
379 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
380 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
381 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
382 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | TRACE_ITER_FUNCTION;
384 void tracer_tracing_on(struct trace_array *tr)
386 if (tr->trace_buffer.buffer)
387 ring_buffer_record_on(tr->trace_buffer.buffer);
389 * This flag is looked at when buffers haven't been allocated
390 * yet, or by some tracers (like irqsoff), that just want to
391 * know if the ring buffer has been disabled, but it can handle
392 * races of where it gets disabled but we still do a record.
393 * As the check is in the fast path of the tracers, it is more
394 * important to be fast than accurate.
396 tr->buffer_disabled = 0;
397 /* Make the flag seen by readers */
402 * tracing_on - enable tracing buffers
404 * This function enables tracing buffers that may have been
405 * disabled with tracing_off.
407 void tracing_on(void)
409 tracer_tracing_on(&global_trace);
411 EXPORT_SYMBOL_GPL(tracing_on);
414 * __trace_puts - write a constant string into the trace buffer.
415 * @ip: The address of the caller
416 * @str: The constant string to write
417 * @size: The size of the string.
419 int __trace_puts(unsigned long ip, const char *str, int size)
421 struct ring_buffer_event *event;
422 struct ring_buffer *buffer;
423 struct print_entry *entry;
424 unsigned long irq_flags;
427 if (unlikely(tracing_selftest_running || tracing_disabled))
430 alloc = sizeof(*entry) + size + 2; /* possible \n added */
432 local_save_flags(irq_flags);
433 buffer = global_trace.trace_buffer.buffer;
434 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
435 irq_flags, preempt_count());
439 entry = ring_buffer_event_data(event);
442 memcpy(&entry->buf, str, size);
444 /* Add a newline if necessary */
445 if (entry->buf[size - 1] != '\n') {
446 entry->buf[size] = '\n';
447 entry->buf[size + 1] = '\0';
449 entry->buf[size] = '\0';
451 __buffer_unlock_commit(buffer, event);
455 EXPORT_SYMBOL_GPL(__trace_puts);
458 * __trace_bputs - write the pointer to a constant string into trace buffer
459 * @ip: The address of the caller
460 * @str: The constant string to write to the buffer to
462 int __trace_bputs(unsigned long ip, const char *str)
464 struct ring_buffer_event *event;
465 struct ring_buffer *buffer;
466 struct bputs_entry *entry;
467 unsigned long irq_flags;
468 int size = sizeof(struct bputs_entry);
470 if (unlikely(tracing_selftest_running || tracing_disabled))
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 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);
641 * tracer_tracing_is_on - show real state of ring buffer enabled
642 * @tr : the trace array to know if ring buffer is enabled
644 * Shows real state of the ring buffer if it is enabled or not.
646 int tracer_tracing_is_on(struct trace_array *tr)
648 if (tr->trace_buffer.buffer)
649 return ring_buffer_record_is_on(tr->trace_buffer.buffer);
650 return !tr->buffer_disabled;
654 * tracing_is_on - show state of ring buffers enabled
656 int tracing_is_on(void)
658 return tracer_tracing_is_on(&global_trace);
660 EXPORT_SYMBOL_GPL(tracing_is_on);
662 static int __init set_buf_size(char *str)
664 unsigned long buf_size;
668 buf_size = memparse(str, &str);
669 /* nr_entries can not be zero */
672 trace_buf_size = buf_size;
675 __setup("trace_buf_size=", set_buf_size);
677 static int __init set_tracing_thresh(char *str)
679 unsigned long threshold;
684 ret = kstrtoul(str, 0, &threshold);
687 tracing_thresh = threshold * 1000;
690 __setup("tracing_thresh=", set_tracing_thresh);
692 unsigned long nsecs_to_usecs(unsigned long nsecs)
697 /* These must match the bit postions in trace_iterator_flags */
698 static const char *trace_options[] = {
731 int in_ns; /* is this clock in nanoseconds? */
733 { trace_clock_local, "local", 1 },
734 { trace_clock_global, "global", 1 },
735 { trace_clock_counter, "counter", 0 },
736 { trace_clock_jiffies, "uptime", 1 },
737 { trace_clock, "perf", 1 },
742 * trace_parser_get_init - gets the buffer for trace parser
744 int trace_parser_get_init(struct trace_parser *parser, int size)
746 memset(parser, 0, sizeof(*parser));
748 parser->buffer = kmalloc(size, GFP_KERNEL);
757 * trace_parser_put - frees the buffer for trace parser
759 void trace_parser_put(struct trace_parser *parser)
761 kfree(parser->buffer);
765 * trace_get_user - reads the user input string separated by space
766 * (matched by isspace(ch))
768 * For each string found the 'struct trace_parser' is updated,
769 * and the function returns.
771 * Returns number of bytes read.
773 * See kernel/trace/trace.h for 'struct trace_parser' details.
775 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
776 size_t cnt, loff_t *ppos)
783 trace_parser_clear(parser);
785 ret = get_user(ch, ubuf++);
793 * The parser is not finished with the last write,
794 * continue reading the user input without skipping spaces.
797 /* skip white space */
798 while (cnt && isspace(ch)) {
799 ret = get_user(ch, ubuf++);
806 /* only spaces were written */
816 /* read the non-space input */
817 while (cnt && !isspace(ch)) {
818 if (parser->idx < parser->size - 1)
819 parser->buffer[parser->idx++] = ch;
824 ret = get_user(ch, ubuf++);
831 /* We either got finished input or we have to wait for another call. */
833 parser->buffer[parser->idx] = 0;
834 parser->cont = false;
835 } else if (parser->idx < parser->size - 1) {
837 parser->buffer[parser->idx++] = ch;
850 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
858 if (s->len <= s->readpos)
861 len = s->len - s->readpos;
864 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
874 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
878 if (s->len <= s->readpos)
881 len = s->len - s->readpos;
884 memcpy(buf, s->buffer + s->readpos, cnt);
891 * ftrace_max_lock is used to protect the swapping of buffers
892 * when taking a max snapshot. The buffers themselves are
893 * protected by per_cpu spinlocks. But the action of the swap
894 * needs its own lock.
896 * This is defined as a arch_spinlock_t in order to help
897 * with performance when lockdep debugging is enabled.
899 * It is also used in other places outside the update_max_tr
900 * so it needs to be defined outside of the
901 * CONFIG_TRACER_MAX_TRACE.
903 static arch_spinlock_t ftrace_max_lock =
904 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
906 unsigned long __read_mostly tracing_thresh;
908 #ifdef CONFIG_TRACER_MAX_TRACE
909 unsigned long __read_mostly tracing_max_latency;
912 * Copy the new maximum trace into the separate maximum-trace
913 * structure. (this way the maximum trace is permanently saved,
914 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
917 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
919 struct trace_buffer *trace_buf = &tr->trace_buffer;
920 struct trace_buffer *max_buf = &tr->max_buffer;
921 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
922 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
925 max_buf->time_start = data->preempt_timestamp;
927 max_data->saved_latency = tracing_max_latency;
928 max_data->critical_start = data->critical_start;
929 max_data->critical_end = data->critical_end;
931 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
932 max_data->pid = tsk->pid;
934 * If tsk == current, then use current_uid(), as that does not use
935 * RCU. The irq tracer can be called out of RCU scope.
938 max_data->uid = current_uid();
940 max_data->uid = task_uid(tsk);
942 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
943 max_data->policy = tsk->policy;
944 max_data->rt_priority = tsk->rt_priority;
946 /* record this tasks comm */
947 tracing_record_cmdline(tsk);
951 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
953 * @tsk: the task with the latency
954 * @cpu: The cpu that initiated the trace.
956 * Flip the buffers between the @tr and the max_tr and record information
957 * about which task was the cause of this latency.
960 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
962 struct ring_buffer *buf;
967 WARN_ON_ONCE(!irqs_disabled());
969 if (!tr->allocated_snapshot) {
970 /* Only the nop tracer should hit this when disabling */
971 WARN_ON_ONCE(tr->current_trace != &nop_trace);
975 arch_spin_lock(&ftrace_max_lock);
977 buf = tr->trace_buffer.buffer;
978 tr->trace_buffer.buffer = tr->max_buffer.buffer;
979 tr->max_buffer.buffer = buf;
981 __update_max_tr(tr, tsk, cpu);
982 arch_spin_unlock(&ftrace_max_lock);
986 * update_max_tr_single - only copy one trace over, and reset the rest
988 * @tsk - task with the latency
989 * @cpu - the cpu of the buffer to copy.
991 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
994 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1001 WARN_ON_ONCE(!irqs_disabled());
1002 if (!tr->allocated_snapshot) {
1003 /* Only the nop tracer should hit this when disabling */
1004 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1008 arch_spin_lock(&ftrace_max_lock);
1010 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
1012 if (ret == -EBUSY) {
1014 * We failed to swap the buffer due to a commit taking
1015 * place on this CPU. We fail to record, but we reset
1016 * the max trace buffer (no one writes directly to it)
1017 * and flag that it failed.
1019 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1020 "Failed to swap buffers due to commit in progress\n");
1023 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1025 __update_max_tr(tr, tsk, cpu);
1026 arch_spin_unlock(&ftrace_max_lock);
1028 #endif /* CONFIG_TRACER_MAX_TRACE */
1030 static void default_wait_pipe(struct trace_iterator *iter)
1032 /* Iterators are static, they should be filled or empty */
1033 if (trace_buffer_iter(iter, iter->cpu_file))
1036 ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file);
1039 #ifdef CONFIG_FTRACE_STARTUP_TEST
1040 static int run_tracer_selftest(struct tracer *type)
1042 struct trace_array *tr = &global_trace;
1043 struct tracer *saved_tracer = tr->current_trace;
1046 if (!type->selftest || tracing_selftest_disabled)
1050 * Run a selftest on this tracer.
1051 * Here we reset the trace buffer, and set the current
1052 * tracer to be this tracer. The tracer can then run some
1053 * internal tracing to verify that everything is in order.
1054 * If we fail, we do not register this tracer.
1056 tracing_reset_online_cpus(&tr->trace_buffer);
1058 tr->current_trace = type;
1060 #ifdef CONFIG_TRACER_MAX_TRACE
1061 if (type->use_max_tr) {
1062 /* If we expanded the buffers, make sure the max is expanded too */
1063 if (ring_buffer_expanded)
1064 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1065 RING_BUFFER_ALL_CPUS);
1066 tr->allocated_snapshot = true;
1070 /* the test is responsible for initializing and enabling */
1071 pr_info("Testing tracer %s: ", type->name);
1072 ret = type->selftest(type, tr);
1073 /* the test is responsible for resetting too */
1074 tr->current_trace = saved_tracer;
1076 printk(KERN_CONT "FAILED!\n");
1077 /* Add the warning after printing 'FAILED' */
1081 /* Only reset on passing, to avoid touching corrupted buffers */
1082 tracing_reset_online_cpus(&tr->trace_buffer);
1084 #ifdef CONFIG_TRACER_MAX_TRACE
1085 if (type->use_max_tr) {
1086 tr->allocated_snapshot = false;
1088 /* Shrink the max buffer again */
1089 if (ring_buffer_expanded)
1090 ring_buffer_resize(tr->max_buffer.buffer, 1,
1091 RING_BUFFER_ALL_CPUS);
1095 printk(KERN_CONT "PASSED\n");
1099 static inline int run_tracer_selftest(struct tracer *type)
1103 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1106 * register_tracer - register a tracer with the ftrace system.
1107 * @type - the plugin for the tracer
1109 * Register a new plugin tracer.
1111 int register_tracer(struct tracer *type)
1117 pr_info("Tracer must have a name\n");
1121 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1122 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1126 mutex_lock(&trace_types_lock);
1128 tracing_selftest_running = true;
1130 for (t = trace_types; t; t = t->next) {
1131 if (strcmp(type->name, t->name) == 0) {
1133 pr_info("Tracer %s already registered\n",
1140 if (!type->set_flag)
1141 type->set_flag = &dummy_set_flag;
1143 type->flags = &dummy_tracer_flags;
1145 if (!type->flags->opts)
1146 type->flags->opts = dummy_tracer_opt;
1147 if (!type->wait_pipe)
1148 type->wait_pipe = default_wait_pipe;
1150 ret = run_tracer_selftest(type);
1154 type->next = trace_types;
1158 tracing_selftest_running = false;
1159 mutex_unlock(&trace_types_lock);
1161 if (ret || !default_bootup_tracer)
1164 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1167 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1168 /* Do we want this tracer to start on bootup? */
1169 tracing_set_tracer(type->name);
1170 default_bootup_tracer = NULL;
1171 /* disable other selftests, since this will break it. */
1172 tracing_selftest_disabled = true;
1173 #ifdef CONFIG_FTRACE_STARTUP_TEST
1174 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1182 void tracing_reset(struct trace_buffer *buf, int cpu)
1184 struct ring_buffer *buffer = buf->buffer;
1189 ring_buffer_record_disable(buffer);
1191 /* Make sure all commits have finished */
1192 synchronize_sched();
1193 ring_buffer_reset_cpu(buffer, cpu);
1195 ring_buffer_record_enable(buffer);
1198 void tracing_reset_online_cpus(struct trace_buffer *buf)
1200 struct ring_buffer *buffer = buf->buffer;
1206 ring_buffer_record_disable(buffer);
1208 /* Make sure all commits have finished */
1209 synchronize_sched();
1211 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
1213 for_each_online_cpu(cpu)
1214 ring_buffer_reset_cpu(buffer, cpu);
1216 ring_buffer_record_enable(buffer);
1219 /* Must have trace_types_lock held */
1220 void tracing_reset_all_online_cpus(void)
1222 struct trace_array *tr;
1224 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1225 tracing_reset_online_cpus(&tr->trace_buffer);
1226 #ifdef CONFIG_TRACER_MAX_TRACE
1227 tracing_reset_online_cpus(&tr->max_buffer);
1232 #define SAVED_CMDLINES 128
1233 #define NO_CMDLINE_MAP UINT_MAX
1234 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1235 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
1236 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
1237 static int cmdline_idx;
1238 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1240 /* temporary disable recording */
1241 static atomic_t trace_record_cmdline_disabled __read_mostly;
1243 static void trace_init_cmdlines(void)
1245 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
1246 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
1250 int is_tracing_stopped(void)
1252 return global_trace.stop_count;
1256 * ftrace_off_permanent - disable all ftrace code permanently
1258 * This should only be called when a serious anomally has
1259 * been detected. This will turn off the function tracing,
1260 * ring buffers, and other tracing utilites. It takes no
1261 * locks and can be called from any context.
1263 void ftrace_off_permanent(void)
1265 tracing_disabled = 1;
1267 tracing_off_permanent();
1271 * tracing_start - quick start of the tracer
1273 * If tracing is enabled but was stopped by tracing_stop,
1274 * this will start the tracer back up.
1276 void tracing_start(void)
1278 struct ring_buffer *buffer;
1279 unsigned long flags;
1281 if (tracing_disabled)
1284 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1285 if (--global_trace.stop_count) {
1286 if (global_trace.stop_count < 0) {
1287 /* Someone screwed up their debugging */
1289 global_trace.stop_count = 0;
1294 /* Prevent the buffers from switching */
1295 arch_spin_lock(&ftrace_max_lock);
1297 buffer = global_trace.trace_buffer.buffer;
1299 ring_buffer_record_enable(buffer);
1301 #ifdef CONFIG_TRACER_MAX_TRACE
1302 buffer = global_trace.max_buffer.buffer;
1304 ring_buffer_record_enable(buffer);
1307 arch_spin_unlock(&ftrace_max_lock);
1311 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1314 static void tracing_start_tr(struct trace_array *tr)
1316 struct ring_buffer *buffer;
1317 unsigned long flags;
1319 if (tracing_disabled)
1322 /* If global, we need to also start the max tracer */
1323 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1324 return tracing_start();
1326 raw_spin_lock_irqsave(&tr->start_lock, flags);
1328 if (--tr->stop_count) {
1329 if (tr->stop_count < 0) {
1330 /* Someone screwed up their debugging */
1337 buffer = tr->trace_buffer.buffer;
1339 ring_buffer_record_enable(buffer);
1342 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1346 * tracing_stop - quick stop of the tracer
1348 * Light weight way to stop tracing. Use in conjunction with
1351 void tracing_stop(void)
1353 struct ring_buffer *buffer;
1354 unsigned long flags;
1357 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1358 if (global_trace.stop_count++)
1361 /* Prevent the buffers from switching */
1362 arch_spin_lock(&ftrace_max_lock);
1364 buffer = global_trace.trace_buffer.buffer;
1366 ring_buffer_record_disable(buffer);
1368 #ifdef CONFIG_TRACER_MAX_TRACE
1369 buffer = global_trace.max_buffer.buffer;
1371 ring_buffer_record_disable(buffer);
1374 arch_spin_unlock(&ftrace_max_lock);
1377 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1380 static void tracing_stop_tr(struct trace_array *tr)
1382 struct ring_buffer *buffer;
1383 unsigned long flags;
1385 /* If global, we need to also stop the max tracer */
1386 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1387 return tracing_stop();
1389 raw_spin_lock_irqsave(&tr->start_lock, flags);
1390 if (tr->stop_count++)
1393 buffer = tr->trace_buffer.buffer;
1395 ring_buffer_record_disable(buffer);
1398 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1401 void trace_stop_cmdline_recording(void);
1403 static void trace_save_cmdline(struct task_struct *tsk)
1407 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1411 * It's not the end of the world if we don't get
1412 * the lock, but we also don't want to spin
1413 * nor do we want to disable interrupts,
1414 * so if we miss here, then better luck next time.
1416 if (!arch_spin_trylock(&trace_cmdline_lock))
1419 idx = map_pid_to_cmdline[tsk->pid];
1420 if (idx == NO_CMDLINE_MAP) {
1421 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1424 * Check whether the cmdline buffer at idx has a pid
1425 * mapped. We are going to overwrite that entry so we
1426 * need to clear the map_pid_to_cmdline. Otherwise we
1427 * would read the new comm for the old pid.
1429 pid = map_cmdline_to_pid[idx];
1430 if (pid != NO_CMDLINE_MAP)
1431 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1433 map_cmdline_to_pid[idx] = tsk->pid;
1434 map_pid_to_cmdline[tsk->pid] = idx;
1439 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1441 arch_spin_unlock(&trace_cmdline_lock);
1444 void trace_find_cmdline(int pid, char comm[])
1449 strcpy(comm, "<idle>");
1453 if (WARN_ON_ONCE(pid < 0)) {
1454 strcpy(comm, "<XXX>");
1458 if (pid > PID_MAX_DEFAULT) {
1459 strcpy(comm, "<...>");
1464 arch_spin_lock(&trace_cmdline_lock);
1465 map = map_pid_to_cmdline[pid];
1466 if (map != NO_CMDLINE_MAP)
1467 strcpy(comm, saved_cmdlines[map]);
1469 strcpy(comm, "<...>");
1471 arch_spin_unlock(&trace_cmdline_lock);
1475 void tracing_record_cmdline(struct task_struct *tsk)
1477 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1480 if (!__this_cpu_read(trace_cmdline_save))
1483 __this_cpu_write(trace_cmdline_save, false);
1485 trace_save_cmdline(tsk);
1489 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1492 struct task_struct *tsk = current;
1494 entry->preempt_count = pc & 0xff;
1495 entry->pid = (tsk) ? tsk->pid : 0;
1497 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1498 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1500 TRACE_FLAG_IRQS_NOSUPPORT |
1502 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1503 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1504 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1506 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1508 struct ring_buffer_event *
1509 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1512 unsigned long flags, int pc)
1514 struct ring_buffer_event *event;
1516 event = ring_buffer_lock_reserve(buffer, len);
1517 if (event != NULL) {
1518 struct trace_entry *ent = ring_buffer_event_data(event);
1520 tracing_generic_entry_update(ent, flags, pc);
1528 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1530 __this_cpu_write(trace_cmdline_save, true);
1531 ring_buffer_unlock_commit(buffer, event);
1535 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1536 struct ring_buffer_event *event,
1537 unsigned long flags, int pc)
1539 __buffer_unlock_commit(buffer, event);
1541 ftrace_trace_stack(buffer, flags, 6, pc);
1542 ftrace_trace_userstack(buffer, flags, pc);
1545 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1546 struct ring_buffer_event *event,
1547 unsigned long flags, int pc)
1549 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1551 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1553 struct ring_buffer_event *
1554 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1555 struct ftrace_event_file *ftrace_file,
1556 int type, unsigned long len,
1557 unsigned long flags, int pc)
1559 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1560 return trace_buffer_lock_reserve(*current_rb,
1561 type, len, flags, pc);
1563 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1565 struct ring_buffer_event *
1566 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1567 int type, unsigned long len,
1568 unsigned long flags, int pc)
1570 *current_rb = global_trace.trace_buffer.buffer;
1571 return trace_buffer_lock_reserve(*current_rb,
1572 type, len, flags, pc);
1574 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1576 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1577 struct ring_buffer_event *event,
1578 unsigned long flags, int pc)
1580 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1582 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1584 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1585 struct ring_buffer_event *event,
1586 unsigned long flags, int pc,
1587 struct pt_regs *regs)
1589 __buffer_unlock_commit(buffer, event);
1591 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1592 ftrace_trace_userstack(buffer, flags, pc);
1594 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1596 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1597 struct ring_buffer_event *event)
1599 ring_buffer_discard_commit(buffer, event);
1601 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1604 trace_function(struct trace_array *tr,
1605 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1608 struct ftrace_event_call *call = &event_function;
1609 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1610 struct ring_buffer_event *event;
1611 struct ftrace_entry *entry;
1613 /* If we are reading the ring buffer, don't trace */
1614 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1617 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1621 entry = ring_buffer_event_data(event);
1623 entry->parent_ip = parent_ip;
1625 if (!filter_check_discard(call, entry, buffer, event))
1626 __buffer_unlock_commit(buffer, event);
1630 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1631 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1634 if (likely(!atomic_read(&data->disabled)))
1635 trace_function(tr, ip, parent_ip, flags, pc);
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 inode *inode, struct file *file, bool snapshot)
2867 struct trace_array *tr = inode->i_private;
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 iter->cpu_file = tracing_get_cpu(inode);
2909 mutex_init(&iter->mutex);
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 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 trace_array *tr = inode->i_private;
2990 struct seq_file *m = file->private_data;
2991 struct trace_iterator *iter;
2994 if (!(file->f_mode & FMODE_READ)) {
2995 trace_array_put(tr);
2999 /* Writes do not use seq_file */
3001 mutex_lock(&trace_types_lock);
3003 for_each_tracing_cpu(cpu) {
3004 if (iter->buffer_iter[cpu])
3005 ring_buffer_read_finish(iter->buffer_iter[cpu]);
3008 if (iter->trace && iter->trace->close)
3009 iter->trace->close(iter);
3011 if (!iter->snapshot)
3012 /* reenable tracing if it was previously enabled */
3013 tracing_start_tr(tr);
3015 __trace_array_put(tr);
3017 mutex_unlock(&trace_types_lock);
3019 mutex_destroy(&iter->mutex);
3020 free_cpumask_var(iter->started);
3022 kfree(iter->buffer_iter);
3023 seq_release_private(inode, file);
3028 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
3030 struct trace_array *tr = inode->i_private;
3032 trace_array_put(tr);
3036 static int tracing_single_release_tr(struct inode *inode, struct file *file)
3038 struct trace_array *tr = inode->i_private;
3040 trace_array_put(tr);
3042 return single_release(inode, file);
3045 static int tracing_open(struct inode *inode, struct file *file)
3047 struct trace_array *tr = inode->i_private;
3048 struct trace_iterator *iter;
3051 if (trace_array_get(tr) < 0)
3054 /* If this file was open for write, then erase contents */
3055 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
3056 int cpu = tracing_get_cpu(inode);
3058 if (cpu == RING_BUFFER_ALL_CPUS)
3059 tracing_reset_online_cpus(&tr->trace_buffer);
3061 tracing_reset(&tr->trace_buffer, cpu);
3064 if (file->f_mode & FMODE_READ) {
3065 iter = __tracing_open(inode, file, false);
3067 ret = PTR_ERR(iter);
3068 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
3069 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3073 trace_array_put(tr);
3079 t_next(struct seq_file *m, void *v, loff_t *pos)
3081 struct tracer *t = v;
3091 static void *t_start(struct seq_file *m, loff_t *pos)
3096 mutex_lock(&trace_types_lock);
3097 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
3103 static void t_stop(struct seq_file *m, void *p)
3105 mutex_unlock(&trace_types_lock);
3108 static int t_show(struct seq_file *m, void *v)
3110 struct tracer *t = v;
3115 seq_printf(m, "%s", t->name);
3124 static const struct seq_operations show_traces_seq_ops = {
3131 static int show_traces_open(struct inode *inode, struct file *file)
3133 if (tracing_disabled)
3136 return seq_open(file, &show_traces_seq_ops);
3140 tracing_write_stub(struct file *filp, const char __user *ubuf,
3141 size_t count, loff_t *ppos)
3146 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
3148 if (file->f_mode & FMODE_READ)
3149 return seq_lseek(file, offset, origin);
3154 static const struct file_operations tracing_fops = {
3155 .open = tracing_open,
3157 .write = tracing_write_stub,
3158 .llseek = tracing_seek,
3159 .release = tracing_release,
3162 static const struct file_operations show_traces_fops = {
3163 .open = show_traces_open,
3165 .release = seq_release,
3166 .llseek = seq_lseek,
3170 * Only trace on a CPU if the bitmask is set:
3172 static cpumask_var_t tracing_cpumask;
3175 * The tracer itself will not take this lock, but still we want
3176 * to provide a consistent cpumask to user-space:
3178 static DEFINE_MUTEX(tracing_cpumask_update_lock);
3181 * Temporary storage for the character representation of the
3182 * CPU bitmask (and one more byte for the newline):
3184 static char mask_str[NR_CPUS + 1];
3187 tracing_cpumask_read(struct file *filp, char __user *ubuf,
3188 size_t count, loff_t *ppos)
3192 mutex_lock(&tracing_cpumask_update_lock);
3194 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
3195 if (count - len < 2) {
3199 len += sprintf(mask_str + len, "\n");
3200 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3203 mutex_unlock(&tracing_cpumask_update_lock);
3209 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3210 size_t count, loff_t *ppos)
3212 struct trace_array *tr = filp->private_data;
3213 cpumask_var_t tracing_cpumask_new;
3216 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3219 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3223 mutex_lock(&tracing_cpumask_update_lock);
3225 local_irq_disable();
3226 arch_spin_lock(&ftrace_max_lock);
3227 for_each_tracing_cpu(cpu) {
3229 * Increase/decrease the disabled counter if we are
3230 * about to flip a bit in the cpumask:
3232 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
3233 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3234 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3235 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3237 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
3238 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3239 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3240 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3243 arch_spin_unlock(&ftrace_max_lock);
3246 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
3248 mutex_unlock(&tracing_cpumask_update_lock);
3249 free_cpumask_var(tracing_cpumask_new);
3254 free_cpumask_var(tracing_cpumask_new);
3259 static const struct file_operations tracing_cpumask_fops = {
3260 .open = tracing_open_generic,
3261 .read = tracing_cpumask_read,
3262 .write = tracing_cpumask_write,
3263 .llseek = generic_file_llseek,
3266 static int tracing_trace_options_show(struct seq_file *m, void *v)
3268 struct tracer_opt *trace_opts;
3269 struct trace_array *tr = m->private;
3273 mutex_lock(&trace_types_lock);
3274 tracer_flags = tr->current_trace->flags->val;
3275 trace_opts = tr->current_trace->flags->opts;
3277 for (i = 0; trace_options[i]; i++) {
3278 if (trace_flags & (1 << i))
3279 seq_printf(m, "%s\n", trace_options[i]);
3281 seq_printf(m, "no%s\n", trace_options[i]);
3284 for (i = 0; trace_opts[i].name; i++) {
3285 if (tracer_flags & trace_opts[i].bit)
3286 seq_printf(m, "%s\n", trace_opts[i].name);
3288 seq_printf(m, "no%s\n", trace_opts[i].name);
3290 mutex_unlock(&trace_types_lock);
3295 static int __set_tracer_option(struct tracer *trace,
3296 struct tracer_flags *tracer_flags,
3297 struct tracer_opt *opts, int neg)
3301 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
3306 tracer_flags->val &= ~opts->bit;
3308 tracer_flags->val |= opts->bit;
3312 /* Try to assign a tracer specific option */
3313 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
3315 struct tracer_flags *tracer_flags = trace->flags;
3316 struct tracer_opt *opts = NULL;
3319 for (i = 0; tracer_flags->opts[i].name; i++) {
3320 opts = &tracer_flags->opts[i];
3322 if (strcmp(cmp, opts->name) == 0)
3323 return __set_tracer_option(trace, trace->flags,
3330 /* Some tracers require overwrite to stay enabled */
3331 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3333 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3339 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3341 /* do nothing if flag is already set */
3342 if (!!(trace_flags & mask) == !!enabled)
3345 /* Give the tracer a chance to approve the change */
3346 if (tr->current_trace->flag_changed)
3347 if (tr->current_trace->flag_changed(tr->current_trace, mask, !!enabled))
3351 trace_flags |= mask;
3353 trace_flags &= ~mask;
3355 if (mask == TRACE_ITER_RECORD_CMD)
3356 trace_event_enable_cmd_record(enabled);
3358 if (mask == TRACE_ITER_OVERWRITE) {
3359 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3360 #ifdef CONFIG_TRACER_MAX_TRACE
3361 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3365 if (mask == TRACE_ITER_PRINTK)
3366 trace_printk_start_stop_comm(enabled);
3371 static int trace_set_options(struct trace_array *tr, char *option)
3378 cmp = strstrip(option);
3380 if (strncmp(cmp, "no", 2) == 0) {
3385 mutex_lock(&trace_types_lock);
3387 for (i = 0; trace_options[i]; i++) {
3388 if (strcmp(cmp, trace_options[i]) == 0) {
3389 ret = set_tracer_flag(tr, 1 << i, !neg);
3394 /* If no option could be set, test the specific tracer options */
3395 if (!trace_options[i])
3396 ret = set_tracer_option(tr->current_trace, cmp, neg);
3398 mutex_unlock(&trace_types_lock);
3404 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3405 size_t cnt, loff_t *ppos)
3407 struct seq_file *m = filp->private_data;
3408 struct trace_array *tr = m->private;
3412 if (cnt >= sizeof(buf))
3415 if (copy_from_user(&buf, ubuf, cnt))
3420 ret = trace_set_options(tr, buf);
3429 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3431 struct trace_array *tr = inode->i_private;
3434 if (tracing_disabled)
3437 if (trace_array_get(tr) < 0)
3440 ret = single_open(file, tracing_trace_options_show, inode->i_private);
3442 trace_array_put(tr);
3447 static const struct file_operations tracing_iter_fops = {
3448 .open = tracing_trace_options_open,
3450 .llseek = seq_lseek,
3451 .release = tracing_single_release_tr,
3452 .write = tracing_trace_options_write,
3455 static const char readme_msg[] =
3456 "tracing mini-HOWTO:\n\n"
3457 "# echo 0 > tracing_on : quick way to disable tracing\n"
3458 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3459 " Important files:\n"
3460 " trace\t\t\t- The static contents of the buffer\n"
3461 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3462 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3463 " current_tracer\t- function and latency tracers\n"
3464 " available_tracers\t- list of configured tracers for current_tracer\n"
3465 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3466 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3467 " trace_clock\t\t-change the clock used to order events\n"
3468 " local: Per cpu clock but may not be synced across CPUs\n"
3469 " global: Synced across CPUs but slows tracing down.\n"
3470 " counter: Not a clock, but just an increment\n"
3471 " uptime: Jiffy counter from time of boot\n"
3472 " perf: Same clock that perf events use\n"
3473 #ifdef CONFIG_X86_64
3474 " x86-tsc: TSC cycle counter\n"
3476 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3477 " tracing_cpumask\t- Limit which CPUs to trace\n"
3478 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3479 "\t\t\t Remove sub-buffer with rmdir\n"
3480 " trace_options\t\t- Set format or modify how tracing happens\n"
3481 "\t\t\t Disable an option by adding a suffix 'no' to the option name\n"
3482 #ifdef CONFIG_DYNAMIC_FTRACE
3483 "\n available_filter_functions - list of functions that can be filtered on\n"
3484 " set_ftrace_filter\t- echo function name in here to only trace these functions\n"
3485 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3486 " modules: Can select a group via module\n"
3487 " Format: :mod:<module-name>\n"
3488 " example: echo :mod:ext3 > set_ftrace_filter\n"
3489 " triggers: a command to perform when function is hit\n"
3490 " Format: <function>:<trigger>[:count]\n"
3491 " trigger: traceon, traceoff\n"
3492 " enable_event:<system>:<event>\n"
3493 " disable_event:<system>:<event>\n"
3494 #ifdef CONFIG_STACKTRACE
3497 #ifdef CONFIG_TRACER_SNAPSHOT
3500 " example: echo do_fault:traceoff > set_ftrace_filter\n"
3501 " echo do_trap:traceoff:3 > set_ftrace_filter\n"
3502 " The first one will disable tracing every time do_fault is hit\n"
3503 " The second will disable tracing at most 3 times when do_trap is hit\n"
3504 " The first time do trap is hit and it disables tracing, the counter\n"
3505 " will decrement to 2. If tracing is already disabled, the counter\n"
3506 " will not decrement. It only decrements when the trigger did work\n"
3507 " To remove trigger without count:\n"
3508 " echo '!<function>:<trigger> > set_ftrace_filter\n"
3509 " To remove trigger with a count:\n"
3510 " echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3511 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3512 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3513 " modules: Can select a group via module command :mod:\n"
3514 " Does not accept triggers\n"
3515 #endif /* CONFIG_DYNAMIC_FTRACE */
3516 #ifdef CONFIG_FUNCTION_TRACER
3517 " set_ftrace_pid\t- Write pid(s) to only function trace those pids (function)\n"
3519 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3520 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3521 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3523 #ifdef CONFIG_TRACER_SNAPSHOT
3524 "\n snapshot\t\t- Like 'trace' but shows the content of the static snapshot buffer\n"
3525 "\t\t\t Read the contents for more information\n"
3527 #ifdef CONFIG_STACKTRACE
3528 " stack_trace\t\t- Shows the max stack trace when active\n"
3529 " stack_max_size\t- Shows current max stack size that was traced\n"
3530 "\t\t\t Write into this file to reset the max size (trigger a new trace)\n"
3531 #ifdef CONFIG_DYNAMIC_FTRACE
3532 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace traces\n"
3534 #endif /* CONFIG_STACKTRACE */
3538 tracing_readme_read(struct file *filp, char __user *ubuf,
3539 size_t cnt, loff_t *ppos)
3541 return simple_read_from_buffer(ubuf, cnt, ppos,
3542 readme_msg, strlen(readme_msg));
3545 static const struct file_operations tracing_readme_fops = {
3546 .open = tracing_open_generic,
3547 .read = tracing_readme_read,
3548 .llseek = generic_file_llseek,
3552 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3553 size_t cnt, loff_t *ppos)
3562 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3566 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3574 for (i = 0; i < SAVED_CMDLINES; i++) {
3577 pid = map_cmdline_to_pid[i];
3578 if (pid == -1 || pid == NO_CMDLINE_MAP)
3581 trace_find_cmdline(pid, buf_comm);
3582 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3587 len = simple_read_from_buffer(ubuf, cnt, ppos,
3596 static const struct file_operations tracing_saved_cmdlines_fops = {
3597 .open = tracing_open_generic,
3598 .read = tracing_saved_cmdlines_read,
3599 .llseek = generic_file_llseek,
3603 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3604 size_t cnt, loff_t *ppos)
3606 struct trace_array *tr = filp->private_data;
3607 char buf[MAX_TRACER_SIZE+2];
3610 mutex_lock(&trace_types_lock);
3611 r = sprintf(buf, "%s\n", tr->current_trace->name);
3612 mutex_unlock(&trace_types_lock);
3614 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3617 int tracer_init(struct tracer *t, struct trace_array *tr)
3619 tracing_reset_online_cpus(&tr->trace_buffer);
3623 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3627 for_each_tracing_cpu(cpu)
3628 per_cpu_ptr(buf->data, cpu)->entries = val;
3631 #ifdef CONFIG_TRACER_MAX_TRACE
3632 /* resize @tr's buffer to the size of @size_tr's entries */
3633 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3634 struct trace_buffer *size_buf, int cpu_id)
3638 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3639 for_each_tracing_cpu(cpu) {
3640 ret = ring_buffer_resize(trace_buf->buffer,
3641 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3644 per_cpu_ptr(trace_buf->data, cpu)->entries =
3645 per_cpu_ptr(size_buf->data, cpu)->entries;
3648 ret = ring_buffer_resize(trace_buf->buffer,
3649 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3651 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3652 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3657 #endif /* CONFIG_TRACER_MAX_TRACE */
3659 static int __tracing_resize_ring_buffer(struct trace_array *tr,
3660 unsigned long size, int cpu)
3665 * If kernel or user changes the size of the ring buffer
3666 * we use the size that was given, and we can forget about
3667 * expanding it later.
3669 ring_buffer_expanded = true;
3671 /* May be called before buffers are initialized */
3672 if (!tr->trace_buffer.buffer)
3675 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3679 #ifdef CONFIG_TRACER_MAX_TRACE
3680 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3681 !tr->current_trace->use_max_tr)
3684 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3686 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3687 &tr->trace_buffer, cpu);
3690 * AARGH! We are left with different
3691 * size max buffer!!!!
3692 * The max buffer is our "snapshot" buffer.
3693 * When a tracer needs a snapshot (one of the
3694 * latency tracers), it swaps the max buffer
3695 * with the saved snap shot. We succeeded to
3696 * update the size of the main buffer, but failed to
3697 * update the size of the max buffer. But when we tried
3698 * to reset the main buffer to the original size, we
3699 * failed there too. This is very unlikely to
3700 * happen, but if it does, warn and kill all
3704 tracing_disabled = 1;
3709 if (cpu == RING_BUFFER_ALL_CPUS)
3710 set_buffer_entries(&tr->max_buffer, size);
3712 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3715 #endif /* CONFIG_TRACER_MAX_TRACE */
3717 if (cpu == RING_BUFFER_ALL_CPUS)
3718 set_buffer_entries(&tr->trace_buffer, size);
3720 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3725 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3726 unsigned long size, int cpu_id)
3730 mutex_lock(&trace_types_lock);
3732 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3733 /* make sure, this cpu is enabled in the mask */
3734 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3740 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3745 mutex_unlock(&trace_types_lock);
3752 * tracing_update_buffers - used by tracing facility to expand ring buffers
3754 * To save on memory when the tracing is never used on a system with it
3755 * configured in. The ring buffers are set to a minimum size. But once
3756 * a user starts to use the tracing facility, then they need to grow
3757 * to their default size.
3759 * This function is to be called when a tracer is about to be used.
3761 int tracing_update_buffers(void)
3765 mutex_lock(&trace_types_lock);
3766 if (!ring_buffer_expanded)
3767 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3768 RING_BUFFER_ALL_CPUS);
3769 mutex_unlock(&trace_types_lock);
3774 struct trace_option_dentry;
3776 static struct trace_option_dentry *
3777 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3780 destroy_trace_option_files(struct trace_option_dentry *topts);
3782 static int tracing_set_tracer(const char *buf)
3784 static struct trace_option_dentry *topts;
3785 struct trace_array *tr = &global_trace;
3787 #ifdef CONFIG_TRACER_MAX_TRACE
3792 mutex_lock(&trace_types_lock);
3794 if (!ring_buffer_expanded) {
3795 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3796 RING_BUFFER_ALL_CPUS);
3802 for (t = trace_types; t; t = t->next) {
3803 if (strcmp(t->name, buf) == 0)
3810 if (t == tr->current_trace)
3813 trace_branch_disable();
3815 tr->current_trace->enabled = false;
3817 if (tr->current_trace->reset)
3818 tr->current_trace->reset(tr);
3820 /* Current trace needs to be nop_trace before synchronize_sched */
3821 tr->current_trace = &nop_trace;
3823 #ifdef CONFIG_TRACER_MAX_TRACE
3824 had_max_tr = tr->allocated_snapshot;
3826 if (had_max_tr && !t->use_max_tr) {
3828 * We need to make sure that the update_max_tr sees that
3829 * current_trace changed to nop_trace to keep it from
3830 * swapping the buffers after we resize it.
3831 * The update_max_tr is called from interrupts disabled
3832 * so a synchronized_sched() is sufficient.
3834 synchronize_sched();
3838 destroy_trace_option_files(topts);
3840 topts = create_trace_option_files(tr, t);
3842 #ifdef CONFIG_TRACER_MAX_TRACE
3843 if (t->use_max_tr && !had_max_tr) {
3844 ret = alloc_snapshot(tr);
3851 ret = tracer_init(t, tr);
3856 tr->current_trace = t;
3857 tr->current_trace->enabled = true;
3858 trace_branch_enable(tr);
3860 mutex_unlock(&trace_types_lock);
3866 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3867 size_t cnt, loff_t *ppos)
3869 char buf[MAX_TRACER_SIZE+1];
3876 if (cnt > MAX_TRACER_SIZE)
3877 cnt = MAX_TRACER_SIZE;
3879 if (copy_from_user(&buf, ubuf, cnt))
3884 /* strip ending whitespace. */
3885 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3888 err = tracing_set_tracer(buf);
3898 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3899 size_t cnt, loff_t *ppos)
3901 unsigned long *ptr = filp->private_data;
3905 r = snprintf(buf, sizeof(buf), "%ld\n",
3906 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3907 if (r > sizeof(buf))
3909 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3913 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3914 size_t cnt, loff_t *ppos)
3916 unsigned long *ptr = filp->private_data;
3920 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3929 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3931 struct trace_array *tr = inode->i_private;
3932 struct trace_iterator *iter;
3935 if (tracing_disabled)
3938 if (trace_array_get(tr) < 0)
3941 mutex_lock(&trace_types_lock);
3943 /* create a buffer to store the information to pass to userspace */
3944 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3947 __trace_array_put(tr);
3952 * We make a copy of the current tracer to avoid concurrent
3953 * changes on it while we are reading.
3955 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3960 *iter->trace = *tr->current_trace;
3962 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3967 /* trace pipe does not show start of buffer */
3968 cpumask_setall(iter->started);
3970 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3971 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3973 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3974 if (trace_clocks[tr->clock_id].in_ns)
3975 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3978 iter->trace_buffer = &tr->trace_buffer;
3979 iter->cpu_file = tracing_get_cpu(inode);
3980 mutex_init(&iter->mutex);
3981 filp->private_data = iter;
3983 if (iter->trace->pipe_open)
3984 iter->trace->pipe_open(iter);
3986 nonseekable_open(inode, filp);
3988 mutex_unlock(&trace_types_lock);
3994 __trace_array_put(tr);
3995 mutex_unlock(&trace_types_lock);
3999 static int tracing_release_pipe(struct inode *inode, struct file *file)
4001 struct trace_iterator *iter = file->private_data;
4002 struct trace_array *tr = inode->i_private;
4004 mutex_lock(&trace_types_lock);
4006 if (iter->trace->pipe_close)
4007 iter->trace->pipe_close(iter);
4009 mutex_unlock(&trace_types_lock);
4011 free_cpumask_var(iter->started);
4012 mutex_destroy(&iter->mutex);
4016 trace_array_put(tr);
4022 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
4024 /* Iterators are static, they should be filled or empty */
4025 if (trace_buffer_iter(iter, iter->cpu_file))
4026 return POLLIN | POLLRDNORM;
4028 if (trace_flags & TRACE_ITER_BLOCK)
4030 * Always select as readable when in blocking mode
4032 return POLLIN | POLLRDNORM;
4034 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
4039 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
4041 struct trace_iterator *iter = filp->private_data;
4043 return trace_poll(iter, filp, poll_table);
4047 * This is a make-shift waitqueue.
4048 * A tracer might use this callback on some rare cases:
4050 * 1) the current tracer might hold the runqueue lock when it wakes up
4051 * a reader, hence a deadlock (sched, function, and function graph tracers)
4052 * 2) the function tracers, trace all functions, we don't want
4053 * the overhead of calling wake_up and friends
4054 * (and tracing them too)
4056 * Anyway, this is really very primitive wakeup.
4058 void poll_wait_pipe(struct trace_iterator *iter)
4060 set_current_state(TASK_INTERRUPTIBLE);
4061 /* sleep for 100 msecs, and try again. */
4062 schedule_timeout(HZ / 10);
4065 /* Must be called with trace_types_lock mutex held. */
4066 static int tracing_wait_pipe(struct file *filp)
4068 struct trace_iterator *iter = filp->private_data;
4070 while (trace_empty(iter)) {
4072 if ((filp->f_flags & O_NONBLOCK)) {
4076 mutex_unlock(&iter->mutex);
4078 iter->trace->wait_pipe(iter);
4080 mutex_lock(&iter->mutex);
4082 if (signal_pending(current))
4086 * We block until we read something and tracing is disabled.
4087 * We still block if tracing is disabled, but we have never
4088 * read anything. This allows a user to cat this file, and
4089 * then enable tracing. But after we have read something,
4090 * we give an EOF when tracing is again disabled.
4092 * iter->pos will be 0 if we haven't read anything.
4094 if (!tracing_is_on() && iter->pos)
4105 tracing_read_pipe(struct file *filp, char __user *ubuf,
4106 size_t cnt, loff_t *ppos)
4108 struct trace_iterator *iter = filp->private_data;
4109 struct trace_array *tr = iter->tr;
4112 /* return any leftover data */
4113 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4117 trace_seq_init(&iter->seq);
4119 /* copy the tracer to avoid using a global lock all around */
4120 mutex_lock(&trace_types_lock);
4121 if (unlikely(iter->trace->name != tr->current_trace->name))
4122 *iter->trace = *tr->current_trace;
4123 mutex_unlock(&trace_types_lock);
4126 * Avoid more than one consumer on a single file descriptor
4127 * This is just a matter of traces coherency, the ring buffer itself
4130 mutex_lock(&iter->mutex);
4131 if (iter->trace->read) {
4132 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
4138 sret = tracing_wait_pipe(filp);
4142 /* stop when tracing is finished */
4143 if (trace_empty(iter)) {
4148 if (cnt >= PAGE_SIZE)
4149 cnt = PAGE_SIZE - 1;
4151 /* reset all but tr, trace, and overruns */
4152 memset(&iter->seq, 0,
4153 sizeof(struct trace_iterator) -
4154 offsetof(struct trace_iterator, seq));
4155 cpumask_clear(iter->started);
4158 trace_event_read_lock();
4159 trace_access_lock(iter->cpu_file);
4160 while (trace_find_next_entry_inc(iter) != NULL) {
4161 enum print_line_t ret;
4162 int len = iter->seq.len;
4164 ret = print_trace_line(iter);
4165 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4166 /* don't print partial lines */
4167 iter->seq.len = len;
4170 if (ret != TRACE_TYPE_NO_CONSUME)
4171 trace_consume(iter);
4173 if (iter->seq.len >= cnt)
4177 * Setting the full flag means we reached the trace_seq buffer
4178 * size and we should leave by partial output condition above.
4179 * One of the trace_seq_* functions is not used properly.
4181 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4184 trace_access_unlock(iter->cpu_file);
4185 trace_event_read_unlock();
4187 /* Now copy what we have to the user */
4188 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4189 if (iter->seq.readpos >= iter->seq.len)
4190 trace_seq_init(&iter->seq);
4193 * If there was nothing to send to user, in spite of consuming trace
4194 * entries, go back to wait for more entries.
4200 mutex_unlock(&iter->mutex);
4205 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
4206 struct pipe_buffer *buf)
4208 __free_page(buf->page);
4211 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4214 __free_page(spd->pages[idx]);
4217 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4219 .map = generic_pipe_buf_map,
4220 .unmap = generic_pipe_buf_unmap,
4221 .confirm = generic_pipe_buf_confirm,
4222 .release = tracing_pipe_buf_release,
4223 .steal = generic_pipe_buf_steal,
4224 .get = generic_pipe_buf_get,
4228 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4233 /* Seq buffer is page-sized, exactly what we need. */
4235 count = iter->seq.len;
4236 ret = print_trace_line(iter);
4237 count = iter->seq.len - count;
4240 iter->seq.len -= count;
4243 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4244 iter->seq.len -= count;
4248 if (ret != TRACE_TYPE_NO_CONSUME)
4249 trace_consume(iter);
4251 if (!trace_find_next_entry_inc(iter)) {
4261 static ssize_t tracing_splice_read_pipe(struct file *filp,
4263 struct pipe_inode_info *pipe,
4267 struct page *pages_def[PIPE_DEF_BUFFERS];
4268 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4269 struct trace_iterator *iter = filp->private_data;
4270 struct splice_pipe_desc spd = {
4272 .partial = partial_def,
4273 .nr_pages = 0, /* This gets updated below. */
4274 .nr_pages_max = PIPE_DEF_BUFFERS,
4276 .ops = &tracing_pipe_buf_ops,
4277 .spd_release = tracing_spd_release_pipe,
4279 struct trace_array *tr = iter->tr;
4284 if (splice_grow_spd(pipe, &spd))
4287 /* copy the tracer to avoid using a global lock all around */
4288 mutex_lock(&trace_types_lock);
4289 if (unlikely(iter->trace->name != tr->current_trace->name))
4290 *iter->trace = *tr->current_trace;
4291 mutex_unlock(&trace_types_lock);
4293 mutex_lock(&iter->mutex);
4295 if (iter->trace->splice_read) {
4296 ret = iter->trace->splice_read(iter, filp,
4297 ppos, pipe, len, flags);
4302 ret = tracing_wait_pipe(filp);
4306 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4311 trace_event_read_lock();
4312 trace_access_lock(iter->cpu_file);
4314 /* Fill as many pages as possible. */
4315 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
4316 spd.pages[i] = alloc_page(GFP_KERNEL);
4320 rem = tracing_fill_pipe_page(rem, iter);
4322 /* Copy the data into the page, so we can start over. */
4323 ret = trace_seq_to_buffer(&iter->seq,
4324 page_address(spd.pages[i]),
4327 __free_page(spd.pages[i]);
4330 spd.partial[i].offset = 0;
4331 spd.partial[i].len = iter->seq.len;
4333 trace_seq_init(&iter->seq);
4336 trace_access_unlock(iter->cpu_file);
4337 trace_event_read_unlock();
4338 mutex_unlock(&iter->mutex);
4342 ret = splice_to_pipe(pipe, &spd);
4344 splice_shrink_spd(&spd);
4348 mutex_unlock(&iter->mutex);
4353 tracing_entries_read(struct file *filp, char __user *ubuf,
4354 size_t cnt, loff_t *ppos)
4356 struct inode *inode = file_inode(filp);
4357 struct trace_array *tr = inode->i_private;
4358 int cpu = tracing_get_cpu(inode);
4363 mutex_lock(&trace_types_lock);
4365 if (cpu == RING_BUFFER_ALL_CPUS) {
4366 int cpu, buf_size_same;
4371 /* check if all cpu sizes are same */
4372 for_each_tracing_cpu(cpu) {
4373 /* fill in the size from first enabled cpu */
4375 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4376 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4382 if (buf_size_same) {
4383 if (!ring_buffer_expanded)
4384 r = sprintf(buf, "%lu (expanded: %lu)\n",
4386 trace_buf_size >> 10);
4388 r = sprintf(buf, "%lu\n", size >> 10);
4390 r = sprintf(buf, "X\n");
4392 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10);
4394 mutex_unlock(&trace_types_lock);
4396 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4401 tracing_entries_write(struct file *filp, const char __user *ubuf,
4402 size_t cnt, loff_t *ppos)
4404 struct inode *inode = file_inode(filp);
4405 struct trace_array *tr = inode->i_private;
4409 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4413 /* must have at least 1 entry */
4417 /* value is in KB */
4419 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
4429 tracing_total_entries_read(struct file *filp, char __user *ubuf,
4430 size_t cnt, loff_t *ppos)
4432 struct trace_array *tr = filp->private_data;
4435 unsigned long size = 0, expanded_size = 0;
4437 mutex_lock(&trace_types_lock);
4438 for_each_tracing_cpu(cpu) {
4439 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
4440 if (!ring_buffer_expanded)
4441 expanded_size += trace_buf_size >> 10;
4443 if (ring_buffer_expanded)
4444 r = sprintf(buf, "%lu\n", size);
4446 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
4447 mutex_unlock(&trace_types_lock);
4449 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4453 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4454 size_t cnt, loff_t *ppos)
4457 * There is no need to read what the user has written, this function
4458 * is just to make sure that there is no error when "echo" is used
4467 tracing_free_buffer_release(struct inode *inode, struct file *filp)
4469 struct trace_array *tr = inode->i_private;
4471 /* disable tracing ? */
4472 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4473 tracer_tracing_off(tr);
4474 /* resize the ring buffer to 0 */
4475 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4477 trace_array_put(tr);
4483 tracing_mark_write(struct file *filp, const char __user *ubuf,
4484 size_t cnt, loff_t *fpos)
4486 unsigned long addr = (unsigned long)ubuf;
4487 struct trace_array *tr = filp->private_data;
4488 struct ring_buffer_event *event;
4489 struct ring_buffer *buffer;
4490 struct print_entry *entry;
4491 unsigned long irq_flags;
4492 struct page *pages[2];
4502 if (tracing_disabled)
4505 if (!(trace_flags & TRACE_ITER_MARKERS))
4508 if (cnt > TRACE_BUF_SIZE)
4509 cnt = TRACE_BUF_SIZE;
4512 * Userspace is injecting traces into the kernel trace buffer.
4513 * We want to be as non intrusive as possible.
4514 * To do so, we do not want to allocate any special buffers
4515 * or take any locks, but instead write the userspace data
4516 * straight into the ring buffer.
4518 * First we need to pin the userspace buffer into memory,
4519 * which, most likely it is, because it just referenced it.
4520 * But there's no guarantee that it is. By using get_user_pages_fast()
4521 * and kmap_atomic/kunmap_atomic() we can get access to the
4522 * pages directly. We then write the data directly into the
4525 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4527 /* check if we cross pages */
4528 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4531 offset = addr & (PAGE_SIZE - 1);
4534 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4535 if (ret < nr_pages) {
4537 put_page(pages[ret]);
4542 for (i = 0; i < nr_pages; i++)
4543 map_page[i] = kmap_atomic(pages[i]);
4545 local_save_flags(irq_flags);
4546 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4547 buffer = tr->trace_buffer.buffer;
4548 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4549 irq_flags, preempt_count());
4551 /* Ring buffer disabled, return as if not open for write */
4556 entry = ring_buffer_event_data(event);
4557 entry->ip = _THIS_IP_;
4559 if (nr_pages == 2) {
4560 len = PAGE_SIZE - offset;
4561 memcpy(&entry->buf, map_page[0] + offset, len);
4562 memcpy(&entry->buf[len], map_page[1], cnt - len);
4564 memcpy(&entry->buf, map_page[0] + offset, cnt);
4566 if (entry->buf[cnt - 1] != '\n') {
4567 entry->buf[cnt] = '\n';
4568 entry->buf[cnt + 1] = '\0';
4570 entry->buf[cnt] = '\0';
4572 __buffer_unlock_commit(buffer, event);
4579 for (i = 0; i < nr_pages; i++){
4580 kunmap_atomic(map_page[i]);
4587 static int tracing_clock_show(struct seq_file *m, void *v)
4589 struct trace_array *tr = m->private;
4592 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4594 "%s%s%s%s", i ? " " : "",
4595 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4596 i == tr->clock_id ? "]" : "");
4602 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4603 size_t cnt, loff_t *fpos)
4605 struct seq_file *m = filp->private_data;
4606 struct trace_array *tr = m->private;
4608 const char *clockstr;
4611 if (cnt >= sizeof(buf))
4614 if (copy_from_user(&buf, ubuf, cnt))
4619 clockstr = strstrip(buf);
4621 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4622 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4625 if (i == ARRAY_SIZE(trace_clocks))
4628 mutex_lock(&trace_types_lock);
4632 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4635 * New clock may not be consistent with the previous clock.
4636 * Reset the buffer so that it doesn't have incomparable timestamps.
4638 tracing_reset_online_cpus(&tr->trace_buffer);
4640 #ifdef CONFIG_TRACER_MAX_TRACE
4641 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4642 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4643 tracing_reset_online_cpus(&tr->max_buffer);
4646 mutex_unlock(&trace_types_lock);
4653 static int tracing_clock_open(struct inode *inode, struct file *file)
4655 struct trace_array *tr = inode->i_private;
4658 if (tracing_disabled)
4661 if (trace_array_get(tr))
4664 ret = single_open(file, tracing_clock_show, inode->i_private);
4666 trace_array_put(tr);
4671 struct ftrace_buffer_info {
4672 struct trace_iterator iter;
4677 #ifdef CONFIG_TRACER_SNAPSHOT
4678 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4680 struct trace_array *tr = inode->i_private;
4681 struct trace_iterator *iter;
4685 if (trace_array_get(tr) < 0)
4688 if (file->f_mode & FMODE_READ) {
4689 iter = __tracing_open(inode, file, true);
4691 ret = PTR_ERR(iter);
4693 /* Writes still need the seq_file to hold the private data */
4695 m = kzalloc(sizeof(*m), GFP_KERNEL);
4698 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4706 iter->trace_buffer = &tr->max_buffer;
4707 iter->cpu_file = tracing_get_cpu(inode);
4709 file->private_data = m;
4713 trace_array_put(tr);
4719 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4722 struct seq_file *m = filp->private_data;
4723 struct trace_iterator *iter = m->private;
4724 struct trace_array *tr = iter->tr;
4728 ret = tracing_update_buffers();
4732 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4736 mutex_lock(&trace_types_lock);
4738 if (tr->current_trace->use_max_tr) {
4745 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4749 if (tr->allocated_snapshot)
4753 /* Only allow per-cpu swap if the ring buffer supports it */
4754 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4755 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4760 if (!tr->allocated_snapshot) {
4761 ret = alloc_snapshot(tr);
4765 local_irq_disable();
4766 /* Now, we're going to swap */
4767 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4768 update_max_tr(tr, current, smp_processor_id());
4770 update_max_tr_single(tr, current, iter->cpu_file);
4774 if (tr->allocated_snapshot) {
4775 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4776 tracing_reset_online_cpus(&tr->max_buffer);
4778 tracing_reset(&tr->max_buffer, iter->cpu_file);
4788 mutex_unlock(&trace_types_lock);
4792 static int tracing_snapshot_release(struct inode *inode, struct file *file)
4794 struct seq_file *m = file->private_data;
4797 ret = tracing_release(inode, file);
4799 if (file->f_mode & FMODE_READ)
4802 /* If write only, the seq_file is just a stub */
4810 static int tracing_buffers_open(struct inode *inode, struct file *filp);
4811 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4812 size_t count, loff_t *ppos);
4813 static int tracing_buffers_release(struct inode *inode, struct file *file);
4814 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4815 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
4817 static int snapshot_raw_open(struct inode *inode, struct file *filp)
4819 struct ftrace_buffer_info *info;
4822 ret = tracing_buffers_open(inode, filp);
4826 info = filp->private_data;
4828 if (info->iter.trace->use_max_tr) {
4829 tracing_buffers_release(inode, filp);
4833 info->iter.snapshot = true;
4834 info->iter.trace_buffer = &info->iter.tr->max_buffer;
4839 #endif /* CONFIG_TRACER_SNAPSHOT */
4842 static const struct file_operations tracing_max_lat_fops = {
4843 .open = tracing_open_generic,
4844 .read = tracing_max_lat_read,
4845 .write = tracing_max_lat_write,
4846 .llseek = generic_file_llseek,
4849 static const struct file_operations set_tracer_fops = {
4850 .open = tracing_open_generic,
4851 .read = tracing_set_trace_read,
4852 .write = tracing_set_trace_write,
4853 .llseek = generic_file_llseek,
4856 static const struct file_operations tracing_pipe_fops = {
4857 .open = tracing_open_pipe,
4858 .poll = tracing_poll_pipe,
4859 .read = tracing_read_pipe,
4860 .splice_read = tracing_splice_read_pipe,
4861 .release = tracing_release_pipe,
4862 .llseek = no_llseek,
4865 static const struct file_operations tracing_entries_fops = {
4866 .open = tracing_open_generic_tr,
4867 .read = tracing_entries_read,
4868 .write = tracing_entries_write,
4869 .llseek = generic_file_llseek,
4870 .release = tracing_release_generic_tr,
4873 static const struct file_operations tracing_total_entries_fops = {
4874 .open = tracing_open_generic_tr,
4875 .read = tracing_total_entries_read,
4876 .llseek = generic_file_llseek,
4877 .release = tracing_release_generic_tr,
4880 static const struct file_operations tracing_free_buffer_fops = {
4881 .open = tracing_open_generic_tr,
4882 .write = tracing_free_buffer_write,
4883 .release = tracing_free_buffer_release,
4886 static const struct file_operations tracing_mark_fops = {
4887 .open = tracing_open_generic_tr,
4888 .write = tracing_mark_write,
4889 .llseek = generic_file_llseek,
4890 .release = tracing_release_generic_tr,
4893 static const struct file_operations trace_clock_fops = {
4894 .open = tracing_clock_open,
4896 .llseek = seq_lseek,
4897 .release = tracing_single_release_tr,
4898 .write = tracing_clock_write,
4901 #ifdef CONFIG_TRACER_SNAPSHOT
4902 static const struct file_operations snapshot_fops = {
4903 .open = tracing_snapshot_open,
4905 .write = tracing_snapshot_write,
4906 .llseek = tracing_seek,
4907 .release = tracing_snapshot_release,
4910 static const struct file_operations snapshot_raw_fops = {
4911 .open = snapshot_raw_open,
4912 .read = tracing_buffers_read,
4913 .release = tracing_buffers_release,
4914 .splice_read = tracing_buffers_splice_read,
4915 .llseek = no_llseek,
4918 #endif /* CONFIG_TRACER_SNAPSHOT */
4920 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4922 struct trace_array *tr = inode->i_private;
4923 struct ftrace_buffer_info *info;
4926 if (tracing_disabled)
4929 if (trace_array_get(tr) < 0)
4932 info = kzalloc(sizeof(*info), GFP_KERNEL);
4934 trace_array_put(tr);
4938 mutex_lock(&trace_types_lock);
4941 info->iter.cpu_file = tracing_get_cpu(inode);
4942 info->iter.trace = tr->current_trace;
4943 info->iter.trace_buffer = &tr->trace_buffer;
4945 /* Force reading ring buffer for first read */
4946 info->read = (unsigned int)-1;
4948 filp->private_data = info;
4950 mutex_unlock(&trace_types_lock);
4952 ret = nonseekable_open(inode, filp);
4954 trace_array_put(tr);
4960 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
4962 struct ftrace_buffer_info *info = filp->private_data;
4963 struct trace_iterator *iter = &info->iter;
4965 return trace_poll(iter, filp, poll_table);
4969 tracing_buffers_read(struct file *filp, char __user *ubuf,
4970 size_t count, loff_t *ppos)
4972 struct ftrace_buffer_info *info = filp->private_data;
4973 struct trace_iterator *iter = &info->iter;
4980 mutex_lock(&trace_types_lock);
4982 #ifdef CONFIG_TRACER_MAX_TRACE
4983 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
4990 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
4996 /* Do we have previous read data to read? */
4997 if (info->read < PAGE_SIZE)
5001 trace_access_lock(iter->cpu_file);
5002 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
5006 trace_access_unlock(iter->cpu_file);
5009 if (trace_empty(iter)) {
5010 if ((filp->f_flags & O_NONBLOCK)) {
5014 mutex_unlock(&trace_types_lock);
5015 iter->trace->wait_pipe(iter);
5016 mutex_lock(&trace_types_lock);
5017 if (signal_pending(current)) {
5029 size = PAGE_SIZE - info->read;
5033 ret = copy_to_user(ubuf, info->spare + info->read, size);
5044 mutex_unlock(&trace_types_lock);
5049 static int tracing_buffers_release(struct inode *inode, struct file *file)
5051 struct ftrace_buffer_info *info = file->private_data;
5052 struct trace_iterator *iter = &info->iter;
5054 mutex_lock(&trace_types_lock);
5056 __trace_array_put(iter->tr);
5059 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
5062 mutex_unlock(&trace_types_lock);
5068 struct ring_buffer *buffer;
5073 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
5074 struct pipe_buffer *buf)
5076 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5081 ring_buffer_free_read_page(ref->buffer, ref->page);
5086 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
5087 struct pipe_buffer *buf)
5089 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5094 /* Pipe buffer operations for a buffer. */
5095 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
5097 .map = generic_pipe_buf_map,
5098 .unmap = generic_pipe_buf_unmap,
5099 .confirm = generic_pipe_buf_confirm,
5100 .release = buffer_pipe_buf_release,
5101 .steal = generic_pipe_buf_steal,
5102 .get = buffer_pipe_buf_get,
5106 * Callback from splice_to_pipe(), if we need to release some pages
5107 * at the end of the spd in case we error'ed out in filling the pipe.
5109 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
5111 struct buffer_ref *ref =
5112 (struct buffer_ref *)spd->partial[i].private;
5117 ring_buffer_free_read_page(ref->buffer, ref->page);
5119 spd->partial[i].private = 0;
5123 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5124 struct pipe_inode_info *pipe, size_t len,
5127 struct ftrace_buffer_info *info = file->private_data;
5128 struct trace_iterator *iter = &info->iter;
5129 struct partial_page partial_def[PIPE_DEF_BUFFERS];
5130 struct page *pages_def[PIPE_DEF_BUFFERS];
5131 struct splice_pipe_desc spd = {
5133 .partial = partial_def,
5134 .nr_pages_max = PIPE_DEF_BUFFERS,
5136 .ops = &buffer_pipe_buf_ops,
5137 .spd_release = buffer_spd_release,
5139 struct buffer_ref *ref;
5140 int entries, size, i;
5143 mutex_lock(&trace_types_lock);
5145 #ifdef CONFIG_TRACER_MAX_TRACE
5146 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5152 if (splice_grow_spd(pipe, &spd)) {
5157 if (*ppos & (PAGE_SIZE - 1)) {
5162 if (len & (PAGE_SIZE - 1)) {
5163 if (len < PAGE_SIZE) {
5171 trace_access_lock(iter->cpu_file);
5172 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5174 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
5178 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
5183 ref->buffer = iter->trace_buffer->buffer;
5184 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5190 r = ring_buffer_read_page(ref->buffer, &ref->page,
5191 len, iter->cpu_file, 1);
5193 ring_buffer_free_read_page(ref->buffer, ref->page);
5199 * zero out any left over data, this is going to
5202 size = ring_buffer_page_len(ref->page);
5203 if (size < PAGE_SIZE)
5204 memset(ref->page + size, 0, PAGE_SIZE - size);
5206 page = virt_to_page(ref->page);
5208 spd.pages[i] = page;
5209 spd.partial[i].len = PAGE_SIZE;
5210 spd.partial[i].offset = 0;
5211 spd.partial[i].private = (unsigned long)ref;
5215 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5218 trace_access_unlock(iter->cpu_file);
5221 /* did we read anything? */
5222 if (!spd.nr_pages) {
5223 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
5227 mutex_unlock(&trace_types_lock);
5228 iter->trace->wait_pipe(iter);
5229 mutex_lock(&trace_types_lock);
5230 if (signal_pending(current)) {
5237 ret = splice_to_pipe(pipe, &spd);
5238 splice_shrink_spd(&spd);
5240 mutex_unlock(&trace_types_lock);
5245 static const struct file_operations tracing_buffers_fops = {
5246 .open = tracing_buffers_open,
5247 .read = tracing_buffers_read,
5248 .poll = tracing_buffers_poll,
5249 .release = tracing_buffers_release,
5250 .splice_read = tracing_buffers_splice_read,
5251 .llseek = no_llseek,
5255 tracing_stats_read(struct file *filp, char __user *ubuf,
5256 size_t count, loff_t *ppos)
5258 struct inode *inode = file_inode(filp);
5259 struct trace_array *tr = inode->i_private;
5260 struct trace_buffer *trace_buf = &tr->trace_buffer;
5261 int cpu = tracing_get_cpu(inode);
5262 struct trace_seq *s;
5264 unsigned long long t;
5265 unsigned long usec_rem;
5267 s = kmalloc(sizeof(*s), GFP_KERNEL);
5273 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5274 trace_seq_printf(s, "entries: %ld\n", cnt);
5276 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5277 trace_seq_printf(s, "overrun: %ld\n", cnt);
5279 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5280 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5282 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5283 trace_seq_printf(s, "bytes: %ld\n", cnt);
5285 if (trace_clocks[tr->clock_id].in_ns) {
5286 /* local or global for trace_clock */
5287 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5288 usec_rem = do_div(t, USEC_PER_SEC);
5289 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5292 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5293 usec_rem = do_div(t, USEC_PER_SEC);
5294 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5296 /* counter or tsc mode for trace_clock */
5297 trace_seq_printf(s, "oldest event ts: %llu\n",
5298 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5300 trace_seq_printf(s, "now ts: %llu\n",
5301 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5304 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5305 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5307 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5308 trace_seq_printf(s, "read events: %ld\n", cnt);
5310 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
5317 static const struct file_operations tracing_stats_fops = {
5318 .open = tracing_open_generic_tr,
5319 .read = tracing_stats_read,
5320 .llseek = generic_file_llseek,
5321 .release = tracing_release_generic_tr,
5324 #ifdef CONFIG_DYNAMIC_FTRACE
5326 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5332 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5333 size_t cnt, loff_t *ppos)
5335 static char ftrace_dyn_info_buffer[1024];
5336 static DEFINE_MUTEX(dyn_info_mutex);
5337 unsigned long *p = filp->private_data;
5338 char *buf = ftrace_dyn_info_buffer;
5339 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5342 mutex_lock(&dyn_info_mutex);
5343 r = sprintf(buf, "%ld ", *p);
5345 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5348 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5350 mutex_unlock(&dyn_info_mutex);
5355 static const struct file_operations tracing_dyn_info_fops = {
5356 .open = tracing_open_generic,
5357 .read = tracing_read_dyn_info,
5358 .llseek = generic_file_llseek,
5360 #endif /* CONFIG_DYNAMIC_FTRACE */
5362 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5364 ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5370 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5372 unsigned long *count = (long *)data;
5384 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5385 struct ftrace_probe_ops *ops, void *data)
5387 long count = (long)data;
5389 seq_printf(m, "%ps:", (void *)ip);
5391 seq_printf(m, "snapshot");
5394 seq_printf(m, ":unlimited\n");
5396 seq_printf(m, ":count=%ld\n", count);
5401 static struct ftrace_probe_ops snapshot_probe_ops = {
5402 .func = ftrace_snapshot,
5403 .print = ftrace_snapshot_print,
5406 static struct ftrace_probe_ops snapshot_count_probe_ops = {
5407 .func = ftrace_count_snapshot,
5408 .print = ftrace_snapshot_print,
5412 ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
5413 char *glob, char *cmd, char *param, int enable)
5415 struct ftrace_probe_ops *ops;
5416 void *count = (void *)-1;
5420 /* hash funcs only work with set_ftrace_filter */
5424 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
5426 if (glob[0] == '!') {
5427 unregister_ftrace_function_probe_func(glob+1, ops);
5434 number = strsep(¶m, ":");
5436 if (!strlen(number))
5440 * We use the callback data field (which is a pointer)
5443 ret = kstrtoul(number, 0, (unsigned long *)&count);
5448 ret = register_ftrace_function_probe(glob, ops, count);
5451 alloc_snapshot(&global_trace);
5453 return ret < 0 ? ret : 0;
5456 static struct ftrace_func_command ftrace_snapshot_cmd = {
5458 .func = ftrace_trace_snapshot_callback,
5461 static int register_snapshot_cmd(void)
5463 return register_ftrace_command(&ftrace_snapshot_cmd);
5466 static inline int register_snapshot_cmd(void) { return 0; }
5467 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5469 struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
5474 if (!debugfs_initialized())
5477 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
5478 tr->dir = debugfs_create_dir("tracing", NULL);
5481 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5486 struct dentry *tracing_init_dentry(void)
5488 return tracing_init_dentry_tr(&global_trace);
5491 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
5493 struct dentry *d_tracer;
5496 return tr->percpu_dir;
5498 d_tracer = tracing_init_dentry_tr(tr);
5502 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
5504 WARN_ONCE(!tr->percpu_dir,
5505 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
5507 return tr->percpu_dir;
5510 static struct dentry *
5511 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
5512 void *data, long cpu, const struct file_operations *fops)
5514 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
5516 if (ret) /* See tracing_get_cpu() */
5517 ret->d_inode->i_cdev = (void *)(cpu + 1);
5522 tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
5524 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
5525 struct dentry *d_cpu;
5526 char cpu_dir[30]; /* 30 characters should be more than enough */
5531 snprintf(cpu_dir, 30, "cpu%ld", cpu);
5532 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
5534 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
5538 /* per cpu trace_pipe */
5539 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
5540 tr, cpu, &tracing_pipe_fops);
5543 trace_create_cpu_file("trace", 0644, d_cpu,
5544 tr, cpu, &tracing_fops);
5546 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
5547 tr, cpu, &tracing_buffers_fops);
5549 trace_create_cpu_file("stats", 0444, d_cpu,
5550 tr, cpu, &tracing_stats_fops);
5552 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
5553 tr, cpu, &tracing_entries_fops);
5555 #ifdef CONFIG_TRACER_SNAPSHOT
5556 trace_create_cpu_file("snapshot", 0644, d_cpu,
5557 tr, cpu, &snapshot_fops);
5559 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
5560 tr, cpu, &snapshot_raw_fops);
5564 #ifdef CONFIG_FTRACE_SELFTEST
5565 /* Let selftest have access to static functions in this file */
5566 #include "trace_selftest.c"
5569 struct trace_option_dentry {
5570 struct tracer_opt *opt;
5571 struct tracer_flags *flags;
5572 struct trace_array *tr;
5573 struct dentry *entry;
5577 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
5580 struct trace_option_dentry *topt = filp->private_data;
5583 if (topt->flags->val & topt->opt->bit)
5588 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5592 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
5595 struct trace_option_dentry *topt = filp->private_data;
5599 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5603 if (val != 0 && val != 1)
5606 if (!!(topt->flags->val & topt->opt->bit) != val) {
5607 mutex_lock(&trace_types_lock);
5608 ret = __set_tracer_option(topt->tr->current_trace, topt->flags,
5610 mutex_unlock(&trace_types_lock);
5621 static const struct file_operations trace_options_fops = {
5622 .open = tracing_open_generic,
5623 .read = trace_options_read,
5624 .write = trace_options_write,
5625 .llseek = generic_file_llseek,
5629 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5632 long index = (long)filp->private_data;
5635 if (trace_flags & (1 << index))
5640 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5644 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5647 struct trace_array *tr = &global_trace;
5648 long index = (long)filp->private_data;
5652 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5656 if (val != 0 && val != 1)
5659 mutex_lock(&trace_types_lock);
5660 ret = set_tracer_flag(tr, 1 << index, val);
5661 mutex_unlock(&trace_types_lock);
5671 static const struct file_operations trace_options_core_fops = {
5672 .open = tracing_open_generic,
5673 .read = trace_options_core_read,
5674 .write = trace_options_core_write,
5675 .llseek = generic_file_llseek,
5678 struct dentry *trace_create_file(const char *name,
5680 struct dentry *parent,
5682 const struct file_operations *fops)
5686 ret = debugfs_create_file(name, mode, parent, data, fops);
5688 pr_warning("Could not create debugfs '%s' entry\n", name);
5694 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5696 struct dentry *d_tracer;
5701 d_tracer = tracing_init_dentry_tr(tr);
5705 tr->options = debugfs_create_dir("options", d_tracer);
5707 pr_warning("Could not create debugfs directory 'options'\n");
5715 create_trace_option_file(struct trace_array *tr,
5716 struct trace_option_dentry *topt,
5717 struct tracer_flags *flags,
5718 struct tracer_opt *opt)
5720 struct dentry *t_options;
5722 t_options = trace_options_init_dentry(tr);
5726 topt->flags = flags;
5730 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5731 &trace_options_fops);
5735 static struct trace_option_dentry *
5736 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5738 struct trace_option_dentry *topts;
5739 struct tracer_flags *flags;
5740 struct tracer_opt *opts;
5746 flags = tracer->flags;
5748 if (!flags || !flags->opts)
5753 for (cnt = 0; opts[cnt].name; cnt++)
5756 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5760 for (cnt = 0; opts[cnt].name; cnt++)
5761 create_trace_option_file(tr, &topts[cnt], flags,
5768 destroy_trace_option_files(struct trace_option_dentry *topts)
5775 for (cnt = 0; topts[cnt].opt; cnt++) {
5776 if (topts[cnt].entry)
5777 debugfs_remove(topts[cnt].entry);
5783 static struct dentry *
5784 create_trace_option_core_file(struct trace_array *tr,
5785 const char *option, long index)
5787 struct dentry *t_options;
5789 t_options = trace_options_init_dentry(tr);
5793 return trace_create_file(option, 0644, t_options, (void *)index,
5794 &trace_options_core_fops);
5797 static __init void create_trace_options_dir(struct trace_array *tr)
5799 struct dentry *t_options;
5802 t_options = trace_options_init_dentry(tr);
5806 for (i = 0; trace_options[i]; i++)
5807 create_trace_option_core_file(tr, trace_options[i], i);
5811 rb_simple_read(struct file *filp, char __user *ubuf,
5812 size_t cnt, loff_t *ppos)
5814 struct trace_array *tr = filp->private_data;
5818 r = tracer_tracing_is_on(tr);
5819 r = sprintf(buf, "%d\n", r);
5821 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5825 rb_simple_write(struct file *filp, const char __user *ubuf,
5826 size_t cnt, loff_t *ppos)
5828 struct trace_array *tr = filp->private_data;
5829 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5833 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5838 mutex_lock(&trace_types_lock);
5840 tracer_tracing_on(tr);
5841 if (tr->current_trace->start)
5842 tr->current_trace->start(tr);
5844 tracer_tracing_off(tr);
5845 if (tr->current_trace->stop)
5846 tr->current_trace->stop(tr);
5848 mutex_unlock(&trace_types_lock);
5856 static const struct file_operations rb_simple_fops = {
5857 .open = tracing_open_generic_tr,
5858 .read = rb_simple_read,
5859 .write = rb_simple_write,
5860 .release = tracing_release_generic_tr,
5861 .llseek = default_llseek,
5864 struct dentry *trace_instance_dir;
5867 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
5869 static void init_trace_buffers(struct trace_array *tr, struct trace_buffer *buf)
5873 for_each_tracing_cpu(cpu) {
5874 memset(per_cpu_ptr(buf->data, cpu), 0, sizeof(struct trace_array_cpu));
5875 per_cpu_ptr(buf->data, cpu)->trace_cpu.cpu = cpu;
5876 per_cpu_ptr(buf->data, cpu)->trace_cpu.tr = tr;
5881 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
5883 enum ring_buffer_flags rb_flags;
5885 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5889 buf->buffer = ring_buffer_alloc(size, rb_flags);
5893 buf->data = alloc_percpu(struct trace_array_cpu);
5895 ring_buffer_free(buf->buffer);
5899 init_trace_buffers(tr, buf);
5901 /* Allocate the first page for all buffers */
5902 set_buffer_entries(&tr->trace_buffer,
5903 ring_buffer_size(tr->trace_buffer.buffer, 0));
5908 static int allocate_trace_buffers(struct trace_array *tr, int size)
5912 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
5916 #ifdef CONFIG_TRACER_MAX_TRACE
5917 ret = allocate_trace_buffer(tr, &tr->max_buffer,
5918 allocate_snapshot ? size : 1);
5920 ring_buffer_free(tr->trace_buffer.buffer);
5921 free_percpu(tr->trace_buffer.data);
5924 tr->allocated_snapshot = allocate_snapshot;
5927 * Only the top level trace array gets its snapshot allocated
5928 * from the kernel command line.
5930 allocate_snapshot = false;
5935 static int new_instance_create(const char *name)
5937 struct trace_array *tr;
5940 mutex_lock(&trace_types_lock);
5943 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5944 if (tr->name && strcmp(tr->name, name) == 0)
5949 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
5953 tr->name = kstrdup(name, GFP_KERNEL);
5957 raw_spin_lock_init(&tr->start_lock);
5959 tr->current_trace = &nop_trace;
5961 INIT_LIST_HEAD(&tr->systems);
5962 INIT_LIST_HEAD(&tr->events);
5964 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
5967 /* Holder for file callbacks */
5968 tr->trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
5969 tr->trace_cpu.tr = tr;
5971 tr->dir = debugfs_create_dir(name, trace_instance_dir);
5975 ret = event_trace_add_tracer(tr->dir, tr);
5977 debugfs_remove_recursive(tr->dir);
5981 init_tracer_debugfs(tr, tr->dir);
5983 list_add(&tr->list, &ftrace_trace_arrays);
5985 mutex_unlock(&trace_types_lock);
5990 if (tr->trace_buffer.buffer)
5991 ring_buffer_free(tr->trace_buffer.buffer);
5996 mutex_unlock(&trace_types_lock);
6002 static int instance_delete(const char *name)
6004 struct trace_array *tr;
6008 mutex_lock(&trace_types_lock);
6011 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6012 if (tr->name && strcmp(tr->name, name) == 0) {
6024 list_del(&tr->list);
6026 event_trace_del_tracer(tr);
6027 debugfs_remove_recursive(tr->dir);
6028 free_percpu(tr->trace_buffer.data);
6029 ring_buffer_free(tr->trace_buffer.buffer);
6037 mutex_unlock(&trace_types_lock);
6042 static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
6044 struct dentry *parent;
6047 /* Paranoid: Make sure the parent is the "instances" directory */
6048 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6049 if (WARN_ON_ONCE(parent != trace_instance_dir))
6053 * The inode mutex is locked, but debugfs_create_dir() will also
6054 * take the mutex. As the instances directory can not be destroyed
6055 * or changed in any other way, it is safe to unlock it, and
6056 * let the dentry try. If two users try to make the same dir at
6057 * the same time, then the new_instance_create() will determine the
6060 mutex_unlock(&inode->i_mutex);
6062 ret = new_instance_create(dentry->d_iname);
6064 mutex_lock(&inode->i_mutex);
6069 static int instance_rmdir(struct inode *inode, struct dentry *dentry)
6071 struct dentry *parent;
6074 /* Paranoid: Make sure the parent is the "instances" directory */
6075 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6076 if (WARN_ON_ONCE(parent != trace_instance_dir))
6079 /* The caller did a dget() on dentry */
6080 mutex_unlock(&dentry->d_inode->i_mutex);
6083 * The inode mutex is locked, but debugfs_create_dir() will also
6084 * take the mutex. As the instances directory can not be destroyed
6085 * or changed in any other way, it is safe to unlock it, and
6086 * let the dentry try. If two users try to make the same dir at
6087 * the same time, then the instance_delete() will determine the
6090 mutex_unlock(&inode->i_mutex);
6092 ret = instance_delete(dentry->d_iname);
6094 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
6095 mutex_lock(&dentry->d_inode->i_mutex);
6100 static const struct inode_operations instance_dir_inode_operations = {
6101 .lookup = simple_lookup,
6102 .mkdir = instance_mkdir,
6103 .rmdir = instance_rmdir,
6106 static __init void create_trace_instances(struct dentry *d_tracer)
6108 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
6109 if (WARN_ON(!trace_instance_dir))
6112 /* Hijack the dir inode operations, to allow mkdir */
6113 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
6117 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
6121 trace_create_file("trace_options", 0644, d_tracer,
6122 tr, &tracing_iter_fops);
6124 trace_create_file("trace", 0644, d_tracer,
6127 trace_create_file("trace_pipe", 0444, d_tracer,
6128 tr, &tracing_pipe_fops);
6130 trace_create_file("buffer_size_kb", 0644, d_tracer,
6131 tr, &tracing_entries_fops);
6133 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
6134 tr, &tracing_total_entries_fops);
6136 trace_create_file("free_buffer", 0644, d_tracer,
6137 tr, &tracing_free_buffer_fops);
6139 trace_create_file("trace_marker", 0220, d_tracer,
6140 tr, &tracing_mark_fops);
6142 trace_create_file("trace_clock", 0644, d_tracer, tr,
6145 trace_create_file("tracing_on", 0644, d_tracer,
6146 tr, &rb_simple_fops);
6148 #ifdef CONFIG_TRACER_SNAPSHOT
6149 trace_create_file("snapshot", 0644, d_tracer,
6150 tr, &snapshot_fops);
6153 for_each_tracing_cpu(cpu)
6154 tracing_init_debugfs_percpu(tr, cpu);
6158 static __init int tracer_init_debugfs(void)
6160 struct dentry *d_tracer;
6162 trace_access_lock_init();
6164 d_tracer = tracing_init_dentry();
6168 init_tracer_debugfs(&global_trace, d_tracer);
6170 trace_create_file("tracing_cpumask", 0644, d_tracer,
6171 &global_trace, &tracing_cpumask_fops);
6173 trace_create_file("available_tracers", 0444, d_tracer,
6174 &global_trace, &show_traces_fops);
6176 trace_create_file("current_tracer", 0644, d_tracer,
6177 &global_trace, &set_tracer_fops);
6179 #ifdef CONFIG_TRACER_MAX_TRACE
6180 trace_create_file("tracing_max_latency", 0644, d_tracer,
6181 &tracing_max_latency, &tracing_max_lat_fops);
6184 trace_create_file("tracing_thresh", 0644, d_tracer,
6185 &tracing_thresh, &tracing_max_lat_fops);
6187 trace_create_file("README", 0444, d_tracer,
6188 NULL, &tracing_readme_fops);
6190 trace_create_file("saved_cmdlines", 0444, d_tracer,
6191 NULL, &tracing_saved_cmdlines_fops);
6193 #ifdef CONFIG_DYNAMIC_FTRACE
6194 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
6195 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
6198 create_trace_instances(d_tracer);
6200 create_trace_options_dir(&global_trace);
6205 static int trace_panic_handler(struct notifier_block *this,
6206 unsigned long event, void *unused)
6208 if (ftrace_dump_on_oops)
6209 ftrace_dump(ftrace_dump_on_oops);
6213 static struct notifier_block trace_panic_notifier = {
6214 .notifier_call = trace_panic_handler,
6216 .priority = 150 /* priority: INT_MAX >= x >= 0 */
6219 static int trace_die_handler(struct notifier_block *self,
6225 if (ftrace_dump_on_oops)
6226 ftrace_dump(ftrace_dump_on_oops);
6234 static struct notifier_block trace_die_notifier = {
6235 .notifier_call = trace_die_handler,
6240 * printk is set to max of 1024, we really don't need it that big.
6241 * Nothing should be printing 1000 characters anyway.
6243 #define TRACE_MAX_PRINT 1000
6246 * Define here KERN_TRACE so that we have one place to modify
6247 * it if we decide to change what log level the ftrace dump
6250 #define KERN_TRACE KERN_EMERG
6253 trace_printk_seq(struct trace_seq *s)
6255 /* Probably should print a warning here. */
6256 if (s->len >= TRACE_MAX_PRINT)
6257 s->len = TRACE_MAX_PRINT;
6259 /* should be zero ended, but we are paranoid. */
6260 s->buffer[s->len] = 0;
6262 printk(KERN_TRACE "%s", s->buffer);
6267 void trace_init_global_iter(struct trace_iterator *iter)
6269 iter->tr = &global_trace;
6270 iter->trace = iter->tr->current_trace;
6271 iter->cpu_file = RING_BUFFER_ALL_CPUS;
6272 iter->trace_buffer = &global_trace.trace_buffer;
6275 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
6277 /* use static because iter can be a bit big for the stack */
6278 static struct trace_iterator iter;
6279 static atomic_t dump_running;
6280 unsigned int old_userobj;
6281 unsigned long flags;
6284 /* Only allow one dump user at a time. */
6285 if (atomic_inc_return(&dump_running) != 1) {
6286 atomic_dec(&dump_running);
6291 * Always turn off tracing when we dump.
6292 * We don't need to show trace output of what happens
6293 * between multiple crashes.
6295 * If the user does a sysrq-z, then they can re-enable
6296 * tracing with echo 1 > tracing_on.
6300 local_irq_save(flags);
6302 /* Simulate the iterator */
6303 trace_init_global_iter(&iter);
6305 for_each_tracing_cpu(cpu) {
6306 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
6309 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
6311 /* don't look at user memory in panic mode */
6312 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
6314 switch (oops_dump_mode) {
6316 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6319 iter.cpu_file = raw_smp_processor_id();
6324 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
6325 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6328 printk(KERN_TRACE "Dumping ftrace buffer:\n");
6330 /* Did function tracer already get disabled? */
6331 if (ftrace_is_dead()) {
6332 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6333 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6337 * We need to stop all tracing on all CPUS to read the
6338 * the next buffer. This is a bit expensive, but is
6339 * not done often. We fill all what we can read,
6340 * and then release the locks again.
6343 while (!trace_empty(&iter)) {
6346 printk(KERN_TRACE "---------------------------------\n");
6350 /* reset all but tr, trace, and overruns */
6351 memset(&iter.seq, 0,
6352 sizeof(struct trace_iterator) -
6353 offsetof(struct trace_iterator, seq));
6354 iter.iter_flags |= TRACE_FILE_LAT_FMT;
6357 if (trace_find_next_entry_inc(&iter) != NULL) {
6360 ret = print_trace_line(&iter);
6361 if (ret != TRACE_TYPE_NO_CONSUME)
6362 trace_consume(&iter);
6364 touch_nmi_watchdog();
6366 trace_printk_seq(&iter.seq);
6370 printk(KERN_TRACE " (ftrace buffer empty)\n");
6372 printk(KERN_TRACE "---------------------------------\n");
6375 trace_flags |= old_userobj;
6377 for_each_tracing_cpu(cpu) {
6378 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
6380 atomic_dec(&dump_running);
6381 local_irq_restore(flags);
6383 EXPORT_SYMBOL_GPL(ftrace_dump);
6385 __init static int tracer_alloc_buffers(void)
6391 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
6394 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
6395 goto out_free_buffer_mask;
6397 /* Only allocate trace_printk buffers if a trace_printk exists */
6398 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
6399 /* Must be called before global_trace.buffer is allocated */
6400 trace_printk_init_buffers();
6402 /* To save memory, keep the ring buffer size to its minimum */
6403 if (ring_buffer_expanded)
6404 ring_buf_size = trace_buf_size;
6408 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
6409 cpumask_copy(tracing_cpumask, cpu_all_mask);
6411 raw_spin_lock_init(&global_trace.start_lock);
6413 /* TODO: make the number of buffers hot pluggable with CPUS */
6414 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
6415 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
6417 goto out_free_cpumask;
6420 if (global_trace.buffer_disabled)
6423 trace_init_cmdlines();
6426 * register_tracer() might reference current_trace, so it
6427 * needs to be set before we register anything. This is
6428 * just a bootstrap of current_trace anyway.
6430 global_trace.current_trace = &nop_trace;
6432 register_tracer(&nop_trace);
6434 /* All seems OK, enable tracing */
6435 tracing_disabled = 0;
6437 atomic_notifier_chain_register(&panic_notifier_list,
6438 &trace_panic_notifier);
6440 register_die_notifier(&trace_die_notifier);
6442 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
6444 /* Holder for file callbacks */
6445 global_trace.trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
6446 global_trace.trace_cpu.tr = &global_trace;
6448 INIT_LIST_HEAD(&global_trace.systems);
6449 INIT_LIST_HEAD(&global_trace.events);
6450 list_add(&global_trace.list, &ftrace_trace_arrays);
6452 while (trace_boot_options) {
6455 option = strsep(&trace_boot_options, ",");
6456 trace_set_options(&global_trace, option);
6459 register_snapshot_cmd();
6464 free_percpu(global_trace.trace_buffer.data);
6465 #ifdef CONFIG_TRACER_MAX_TRACE
6466 free_percpu(global_trace.max_buffer.data);
6468 free_cpumask_var(tracing_cpumask);
6469 out_free_buffer_mask:
6470 free_cpumask_var(tracing_buffer_mask);
6475 __init static int clear_boot_tracer(void)
6478 * The default tracer at boot buffer is an init section.
6479 * This function is called in lateinit. If we did not
6480 * find the boot tracer, then clear it out, to prevent
6481 * later registration from accessing the buffer that is
6482 * about to be freed.
6484 if (!default_bootup_tracer)
6487 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
6488 default_bootup_tracer);
6489 default_bootup_tracer = NULL;
6494 early_initcall(tracer_alloc_buffers);
6495 fs_initcall(tracer_init_debugfs);
6496 late_initcall(clear_boot_tracer);