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 int 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 return 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);
1310 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1313 static void tracing_start_tr(struct trace_array *tr)
1315 struct ring_buffer *buffer;
1316 unsigned long flags;
1318 if (tracing_disabled)
1321 /* If global, we need to also start the max tracer */
1322 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1323 return tracing_start();
1325 raw_spin_lock_irqsave(&tr->start_lock, flags);
1327 if (--tr->stop_count) {
1328 if (tr->stop_count < 0) {
1329 /* Someone screwed up their debugging */
1336 buffer = tr->trace_buffer.buffer;
1338 ring_buffer_record_enable(buffer);
1341 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1345 * tracing_stop - quick stop of the tracer
1347 * Light weight way to stop tracing. Use in conjunction with
1350 void tracing_stop(void)
1352 struct ring_buffer *buffer;
1353 unsigned long flags;
1355 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1356 if (global_trace.stop_count++)
1359 /* Prevent the buffers from switching */
1360 arch_spin_lock(&ftrace_max_lock);
1362 buffer = global_trace.trace_buffer.buffer;
1364 ring_buffer_record_disable(buffer);
1366 #ifdef CONFIG_TRACER_MAX_TRACE
1367 buffer = global_trace.max_buffer.buffer;
1369 ring_buffer_record_disable(buffer);
1372 arch_spin_unlock(&ftrace_max_lock);
1375 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1378 static void tracing_stop_tr(struct trace_array *tr)
1380 struct ring_buffer *buffer;
1381 unsigned long flags;
1383 /* If global, we need to also stop the max tracer */
1384 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1385 return tracing_stop();
1387 raw_spin_lock_irqsave(&tr->start_lock, flags);
1388 if (tr->stop_count++)
1391 buffer = tr->trace_buffer.buffer;
1393 ring_buffer_record_disable(buffer);
1396 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1399 void trace_stop_cmdline_recording(void);
1401 static int trace_save_cmdline(struct task_struct *tsk)
1405 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1409 * It's not the end of the world if we don't get
1410 * the lock, but we also don't want to spin
1411 * nor do we want to disable interrupts,
1412 * so if we miss here, then better luck next time.
1414 if (!arch_spin_trylock(&trace_cmdline_lock))
1417 idx = map_pid_to_cmdline[tsk->pid];
1418 if (idx == NO_CMDLINE_MAP) {
1419 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1422 * Check whether the cmdline buffer at idx has a pid
1423 * mapped. We are going to overwrite that entry so we
1424 * need to clear the map_pid_to_cmdline. Otherwise we
1425 * would read the new comm for the old pid.
1427 pid = map_cmdline_to_pid[idx];
1428 if (pid != NO_CMDLINE_MAP)
1429 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1431 map_cmdline_to_pid[idx] = tsk->pid;
1432 map_pid_to_cmdline[tsk->pid] = idx;
1437 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1439 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 if (trace_save_cmdline(tsk))
1484 __this_cpu_write(trace_cmdline_save, false);
1488 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1491 struct task_struct *tsk = current;
1493 entry->preempt_count = pc & 0xff;
1494 entry->pid = (tsk) ? tsk->pid : 0;
1496 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1497 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1499 TRACE_FLAG_IRQS_NOSUPPORT |
1501 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1502 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1503 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1505 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1507 struct ring_buffer_event *
1508 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1511 unsigned long flags, int pc)
1513 struct ring_buffer_event *event;
1515 event = ring_buffer_lock_reserve(buffer, len);
1516 if (event != NULL) {
1517 struct trace_entry *ent = ring_buffer_event_data(event);
1519 tracing_generic_entry_update(ent, flags, pc);
1527 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1529 __this_cpu_write(trace_cmdline_save, true);
1530 ring_buffer_unlock_commit(buffer, event);
1534 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1535 struct ring_buffer_event *event,
1536 unsigned long flags, int pc)
1538 __buffer_unlock_commit(buffer, event);
1540 ftrace_trace_stack(buffer, flags, 6, pc);
1541 ftrace_trace_userstack(buffer, flags, pc);
1544 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1545 struct ring_buffer_event *event,
1546 unsigned long flags, int pc)
1548 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1550 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1552 struct ring_buffer_event *
1553 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1554 struct ftrace_event_file *ftrace_file,
1555 int type, unsigned long len,
1556 unsigned long flags, int pc)
1558 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1559 return trace_buffer_lock_reserve(*current_rb,
1560 type, len, flags, pc);
1562 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1564 struct ring_buffer_event *
1565 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1566 int type, unsigned long len,
1567 unsigned long flags, int pc)
1569 *current_rb = global_trace.trace_buffer.buffer;
1570 return trace_buffer_lock_reserve(*current_rb,
1571 type, len, flags, pc);
1573 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1575 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1576 struct ring_buffer_event *event,
1577 unsigned long flags, int pc)
1579 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1581 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1583 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1584 struct ring_buffer_event *event,
1585 unsigned long flags, int pc,
1586 struct pt_regs *regs)
1588 __buffer_unlock_commit(buffer, event);
1590 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1591 ftrace_trace_userstack(buffer, flags, pc);
1593 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1595 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1596 struct ring_buffer_event *event)
1598 ring_buffer_discard_commit(buffer, event);
1600 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1603 trace_function(struct trace_array *tr,
1604 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1607 struct ftrace_event_call *call = &event_function;
1608 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1609 struct ring_buffer_event *event;
1610 struct ftrace_entry *entry;
1612 /* If we are reading the ring buffer, don't trace */
1613 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1616 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1620 entry = ring_buffer_event_data(event);
1622 entry->parent_ip = parent_ip;
1624 if (!filter_check_discard(call, entry, buffer, event))
1625 __buffer_unlock_commit(buffer, event);
1629 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1630 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1633 if (likely(!atomic_read(&data->disabled)))
1634 trace_function(tr, ip, parent_ip, flags, pc);
1637 #ifdef CONFIG_STACKTRACE
1639 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1640 struct ftrace_stack {
1641 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1644 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1645 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1647 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1648 unsigned long flags,
1649 int skip, int pc, struct pt_regs *regs)
1651 struct ftrace_event_call *call = &event_kernel_stack;
1652 struct ring_buffer_event *event;
1653 struct stack_entry *entry;
1654 struct stack_trace trace;
1656 int size = FTRACE_STACK_ENTRIES;
1658 trace.nr_entries = 0;
1662 * Since events can happen in NMIs there's no safe way to
1663 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1664 * or NMI comes in, it will just have to use the default
1665 * FTRACE_STACK_SIZE.
1667 preempt_disable_notrace();
1669 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1671 * We don't need any atomic variables, just a barrier.
1672 * If an interrupt comes in, we don't care, because it would
1673 * have exited and put the counter back to what we want.
1674 * We just need a barrier to keep gcc from moving things
1678 if (use_stack == 1) {
1679 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1680 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1683 save_stack_trace_regs(regs, &trace);
1685 save_stack_trace(&trace);
1687 if (trace.nr_entries > size)
1688 size = trace.nr_entries;
1690 /* From now on, use_stack is a boolean */
1693 size *= sizeof(unsigned long);
1695 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1696 sizeof(*entry) + size, flags, pc);
1699 entry = ring_buffer_event_data(event);
1701 memset(&entry->caller, 0, size);
1704 memcpy(&entry->caller, trace.entries,
1705 trace.nr_entries * sizeof(unsigned long));
1707 trace.max_entries = FTRACE_STACK_ENTRIES;
1708 trace.entries = entry->caller;
1710 save_stack_trace_regs(regs, &trace);
1712 save_stack_trace(&trace);
1715 entry->size = trace.nr_entries;
1717 if (!filter_check_discard(call, entry, buffer, event))
1718 __buffer_unlock_commit(buffer, event);
1721 /* Again, don't let gcc optimize things here */
1723 __this_cpu_dec(ftrace_stack_reserve);
1724 preempt_enable_notrace();
1728 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1729 int skip, int pc, struct pt_regs *regs)
1731 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1734 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1737 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1740 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1743 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1746 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1749 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1753 * trace_dump_stack - record a stack back trace in the trace buffer
1754 * @skip: Number of functions to skip (helper handlers)
1756 void trace_dump_stack(int skip)
1758 unsigned long flags;
1760 if (tracing_disabled || tracing_selftest_running)
1763 local_save_flags(flags);
1766 * Skip 3 more, seems to get us at the caller of
1770 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
1771 flags, skip, preempt_count(), NULL);
1774 static DEFINE_PER_CPU(int, user_stack_count);
1777 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1779 struct ftrace_event_call *call = &event_user_stack;
1780 struct ring_buffer_event *event;
1781 struct userstack_entry *entry;
1782 struct stack_trace trace;
1784 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1788 * NMIs can not handle page faults, even with fix ups.
1789 * The save user stack can (and often does) fault.
1791 if (unlikely(in_nmi()))
1795 * prevent recursion, since the user stack tracing may
1796 * trigger other kernel events.
1799 if (__this_cpu_read(user_stack_count))
1802 __this_cpu_inc(user_stack_count);
1804 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1805 sizeof(*entry), flags, pc);
1807 goto out_drop_count;
1808 entry = ring_buffer_event_data(event);
1810 entry->tgid = current->tgid;
1811 memset(&entry->caller, 0, sizeof(entry->caller));
1813 trace.nr_entries = 0;
1814 trace.max_entries = FTRACE_STACK_ENTRIES;
1816 trace.entries = entry->caller;
1818 save_stack_trace_user(&trace);
1819 if (!filter_check_discard(call, entry, buffer, event))
1820 __buffer_unlock_commit(buffer, event);
1823 __this_cpu_dec(user_stack_count);
1829 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1831 ftrace_trace_userstack(tr, flags, preempt_count());
1835 #endif /* CONFIG_STACKTRACE */
1837 /* created for use with alloc_percpu */
1838 struct trace_buffer_struct {
1839 char buffer[TRACE_BUF_SIZE];
1842 static struct trace_buffer_struct *trace_percpu_buffer;
1843 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1844 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1845 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1848 * The buffer used is dependent on the context. There is a per cpu
1849 * buffer for normal context, softirq contex, hard irq context and
1850 * for NMI context. Thise allows for lockless recording.
1852 * Note, if the buffers failed to be allocated, then this returns NULL
1854 static char *get_trace_buf(void)
1856 struct trace_buffer_struct *percpu_buffer;
1859 * If we have allocated per cpu buffers, then we do not
1860 * need to do any locking.
1863 percpu_buffer = trace_percpu_nmi_buffer;
1865 percpu_buffer = trace_percpu_irq_buffer;
1866 else if (in_softirq())
1867 percpu_buffer = trace_percpu_sirq_buffer;
1869 percpu_buffer = trace_percpu_buffer;
1874 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1877 static int alloc_percpu_trace_buffer(void)
1879 struct trace_buffer_struct *buffers;
1880 struct trace_buffer_struct *sirq_buffers;
1881 struct trace_buffer_struct *irq_buffers;
1882 struct trace_buffer_struct *nmi_buffers;
1884 buffers = alloc_percpu(struct trace_buffer_struct);
1888 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1892 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1896 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1900 trace_percpu_buffer = buffers;
1901 trace_percpu_sirq_buffer = sirq_buffers;
1902 trace_percpu_irq_buffer = irq_buffers;
1903 trace_percpu_nmi_buffer = nmi_buffers;
1908 free_percpu(irq_buffers);
1910 free_percpu(sirq_buffers);
1912 free_percpu(buffers);
1914 WARN(1, "Could not allocate percpu trace_printk buffer");
1918 static int buffers_allocated;
1920 void trace_printk_init_buffers(void)
1922 if (buffers_allocated)
1925 if (alloc_percpu_trace_buffer())
1928 pr_info("ftrace: Allocated trace_printk buffers\n");
1930 /* Expand the buffers to set size */
1931 tracing_update_buffers();
1933 buffers_allocated = 1;
1936 * trace_printk_init_buffers() can be called by modules.
1937 * If that happens, then we need to start cmdline recording
1938 * directly here. If the global_trace.buffer is already
1939 * allocated here, then this was called by module code.
1941 if (global_trace.trace_buffer.buffer)
1942 tracing_start_cmdline_record();
1945 void trace_printk_start_comm(void)
1947 /* Start tracing comms if trace printk is set */
1948 if (!buffers_allocated)
1950 tracing_start_cmdline_record();
1953 static void trace_printk_start_stop_comm(int enabled)
1955 if (!buffers_allocated)
1959 tracing_start_cmdline_record();
1961 tracing_stop_cmdline_record();
1965 * trace_vbprintk - write binary msg to tracing buffer
1968 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1970 struct ftrace_event_call *call = &event_bprint;
1971 struct ring_buffer_event *event;
1972 struct ring_buffer *buffer;
1973 struct trace_array *tr = &global_trace;
1974 struct bprint_entry *entry;
1975 unsigned long flags;
1977 int len = 0, size, pc;
1979 if (unlikely(tracing_selftest_running || tracing_disabled))
1982 /* Don't pollute graph traces with trace_vprintk internals */
1983 pause_graph_tracing();
1985 pc = preempt_count();
1986 preempt_disable_notrace();
1988 tbuffer = get_trace_buf();
1994 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1996 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
1999 local_save_flags(flags);
2000 size = sizeof(*entry) + sizeof(u32) * len;
2001 buffer = tr->trace_buffer.buffer;
2002 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
2006 entry = ring_buffer_event_data(event);
2010 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
2011 if (!filter_check_discard(call, entry, buffer, event)) {
2012 __buffer_unlock_commit(buffer, event);
2013 ftrace_trace_stack(buffer, flags, 6, pc);
2017 preempt_enable_notrace();
2018 unpause_graph_tracing();
2022 EXPORT_SYMBOL_GPL(trace_vbprintk);
2025 __trace_array_vprintk(struct ring_buffer *buffer,
2026 unsigned long ip, const char *fmt, va_list args)
2028 struct ftrace_event_call *call = &event_print;
2029 struct ring_buffer_event *event;
2030 int len = 0, size, pc;
2031 struct print_entry *entry;
2032 unsigned long flags;
2035 if (tracing_disabled || tracing_selftest_running)
2038 /* Don't pollute graph traces with trace_vprintk internals */
2039 pause_graph_tracing();
2041 pc = preempt_count();
2042 preempt_disable_notrace();
2045 tbuffer = get_trace_buf();
2051 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
2052 if (len > TRACE_BUF_SIZE)
2055 local_save_flags(flags);
2056 size = sizeof(*entry) + len + 1;
2057 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
2061 entry = ring_buffer_event_data(event);
2064 memcpy(&entry->buf, tbuffer, len);
2065 entry->buf[len] = '\0';
2066 if (!filter_check_discard(call, entry, buffer, event)) {
2067 __buffer_unlock_commit(buffer, event);
2068 ftrace_trace_stack(buffer, flags, 6, pc);
2071 preempt_enable_notrace();
2072 unpause_graph_tracing();
2077 int trace_array_vprintk(struct trace_array *tr,
2078 unsigned long ip, const char *fmt, va_list args)
2080 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
2083 int trace_array_printk(struct trace_array *tr,
2084 unsigned long ip, const char *fmt, ...)
2089 if (!(trace_flags & TRACE_ITER_PRINTK))
2093 ret = trace_array_vprintk(tr, ip, fmt, ap);
2098 int trace_array_printk_buf(struct ring_buffer *buffer,
2099 unsigned long ip, const char *fmt, ...)
2104 if (!(trace_flags & TRACE_ITER_PRINTK))
2108 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2113 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2115 return trace_array_vprintk(&global_trace, ip, fmt, args);
2117 EXPORT_SYMBOL_GPL(trace_vprintk);
2119 static void trace_iterator_increment(struct trace_iterator *iter)
2121 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2125 ring_buffer_read(buf_iter, NULL);
2128 static struct trace_entry *
2129 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2130 unsigned long *lost_events)
2132 struct ring_buffer_event *event;
2133 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2136 event = ring_buffer_iter_peek(buf_iter, ts);
2138 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2142 iter->ent_size = ring_buffer_event_length(event);
2143 return ring_buffer_event_data(event);
2149 static struct trace_entry *
2150 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2151 unsigned long *missing_events, u64 *ent_ts)
2153 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2154 struct trace_entry *ent, *next = NULL;
2155 unsigned long lost_events = 0, next_lost = 0;
2156 int cpu_file = iter->cpu_file;
2157 u64 next_ts = 0, ts;
2163 * If we are in a per_cpu trace file, don't bother by iterating over
2164 * all cpu and peek directly.
2166 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2167 if (ring_buffer_empty_cpu(buffer, cpu_file))
2169 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2171 *ent_cpu = cpu_file;
2176 for_each_tracing_cpu(cpu) {
2178 if (ring_buffer_empty_cpu(buffer, cpu))
2181 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2184 * Pick the entry with the smallest timestamp:
2186 if (ent && (!next || ts < next_ts)) {
2190 next_lost = lost_events;
2191 next_size = iter->ent_size;
2195 iter->ent_size = next_size;
2198 *ent_cpu = next_cpu;
2204 *missing_events = next_lost;
2209 /* Find the next real entry, without updating the iterator itself */
2210 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2211 int *ent_cpu, u64 *ent_ts)
2213 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2216 /* Find the next real entry, and increment the iterator to the next entry */
2217 void *trace_find_next_entry_inc(struct trace_iterator *iter)
2219 iter->ent = __find_next_entry(iter, &iter->cpu,
2220 &iter->lost_events, &iter->ts);
2223 trace_iterator_increment(iter);
2225 return iter->ent ? iter : NULL;
2228 static void trace_consume(struct trace_iterator *iter)
2230 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2231 &iter->lost_events);
2234 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2236 struct trace_iterator *iter = m->private;
2240 WARN_ON_ONCE(iter->leftover);
2244 /* can't go backwards */
2249 ent = trace_find_next_entry_inc(iter);
2253 while (ent && iter->idx < i)
2254 ent = trace_find_next_entry_inc(iter);
2261 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2263 struct ring_buffer_event *event;
2264 struct ring_buffer_iter *buf_iter;
2265 unsigned long entries = 0;
2268 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2270 buf_iter = trace_buffer_iter(iter, cpu);
2274 ring_buffer_iter_reset(buf_iter);
2277 * We could have the case with the max latency tracers
2278 * that a reset never took place on a cpu. This is evident
2279 * by the timestamp being before the start of the buffer.
2281 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2282 if (ts >= iter->trace_buffer->time_start)
2285 ring_buffer_read(buf_iter, NULL);
2288 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2292 * The current tracer is copied to avoid a global locking
2295 static void *s_start(struct seq_file *m, loff_t *pos)
2297 struct trace_iterator *iter = m->private;
2298 struct trace_array *tr = iter->tr;
2299 int cpu_file = iter->cpu_file;
2305 * copy the tracer to avoid using a global lock all around.
2306 * iter->trace is a copy of current_trace, the pointer to the
2307 * name may be used instead of a strcmp(), as iter->trace->name
2308 * will point to the same string as current_trace->name.
2310 mutex_lock(&trace_types_lock);
2311 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2312 *iter->trace = *tr->current_trace;
2313 mutex_unlock(&trace_types_lock);
2315 #ifdef CONFIG_TRACER_MAX_TRACE
2316 if (iter->snapshot && iter->trace->use_max_tr)
2317 return ERR_PTR(-EBUSY);
2320 if (!iter->snapshot)
2321 atomic_inc(&trace_record_cmdline_disabled);
2323 if (*pos != iter->pos) {
2328 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2329 for_each_tracing_cpu(cpu)
2330 tracing_iter_reset(iter, cpu);
2332 tracing_iter_reset(iter, cpu_file);
2335 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2340 * If we overflowed the seq_file before, then we want
2341 * to just reuse the trace_seq buffer again.
2347 p = s_next(m, p, &l);
2351 trace_event_read_lock();
2352 trace_access_lock(cpu_file);
2356 static void s_stop(struct seq_file *m, void *p)
2358 struct trace_iterator *iter = m->private;
2360 #ifdef CONFIG_TRACER_MAX_TRACE
2361 if (iter->snapshot && iter->trace->use_max_tr)
2365 if (!iter->snapshot)
2366 atomic_dec(&trace_record_cmdline_disabled);
2368 trace_access_unlock(iter->cpu_file);
2369 trace_event_read_unlock();
2373 get_total_entries(struct trace_buffer *buf,
2374 unsigned long *total, unsigned long *entries)
2376 unsigned long count;
2382 for_each_tracing_cpu(cpu) {
2383 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2385 * If this buffer has skipped entries, then we hold all
2386 * entries for the trace and we need to ignore the
2387 * ones before the time stamp.
2389 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2390 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2391 /* total is the same as the entries */
2395 ring_buffer_overrun_cpu(buf->buffer, cpu);
2400 static void print_lat_help_header(struct seq_file *m)
2402 seq_puts(m, "# _------=> CPU# \n");
2403 seq_puts(m, "# / _-----=> irqs-off \n");
2404 seq_puts(m, "# | / _----=> need-resched \n");
2405 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2406 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2407 seq_puts(m, "# |||| / delay \n");
2408 seq_puts(m, "# cmd pid ||||| time | caller \n");
2409 seq_puts(m, "# \\ / ||||| \\ | / \n");
2412 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2414 unsigned long total;
2415 unsigned long entries;
2417 get_total_entries(buf, &total, &entries);
2418 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2419 entries, total, num_online_cpus());
2423 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2425 print_event_info(buf, m);
2426 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2427 seq_puts(m, "# | | | | |\n");
2430 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2432 print_event_info(buf, m);
2433 seq_puts(m, "# _-----=> irqs-off\n");
2434 seq_puts(m, "# / _----=> need-resched\n");
2435 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2436 seq_puts(m, "# || / _--=> preempt-depth\n");
2437 seq_puts(m, "# ||| / delay\n");
2438 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2439 seq_puts(m, "# | | | |||| | |\n");
2443 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2445 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2446 struct trace_buffer *buf = iter->trace_buffer;
2447 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2448 struct tracer *type = iter->trace;
2449 unsigned long entries;
2450 unsigned long total;
2451 const char *name = "preemption";
2455 get_total_entries(buf, &total, &entries);
2457 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2459 seq_puts(m, "# -----------------------------------"
2460 "---------------------------------\n");
2461 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2462 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2463 nsecs_to_usecs(data->saved_latency),
2467 #if defined(CONFIG_PREEMPT_NONE)
2469 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2471 #elif defined(CONFIG_PREEMPT)
2476 /* These are reserved for later use */
2479 seq_printf(m, " #P:%d)\n", num_online_cpus());
2483 seq_puts(m, "# -----------------\n");
2484 seq_printf(m, "# | task: %.16s-%d "
2485 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2486 data->comm, data->pid,
2487 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2488 data->policy, data->rt_priority);
2489 seq_puts(m, "# -----------------\n");
2491 if (data->critical_start) {
2492 seq_puts(m, "# => started at: ");
2493 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2494 trace_print_seq(m, &iter->seq);
2495 seq_puts(m, "\n# => ended at: ");
2496 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2497 trace_print_seq(m, &iter->seq);
2498 seq_puts(m, "\n#\n");
2504 static void test_cpu_buff_start(struct trace_iterator *iter)
2506 struct trace_seq *s = &iter->seq;
2508 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2511 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2514 if (cpumask_test_cpu(iter->cpu, iter->started))
2517 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2520 cpumask_set_cpu(iter->cpu, iter->started);
2522 /* Don't print started cpu buffer for the first entry of the trace */
2524 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2528 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2530 struct trace_seq *s = &iter->seq;
2531 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2532 struct trace_entry *entry;
2533 struct trace_event *event;
2537 test_cpu_buff_start(iter);
2539 event = ftrace_find_event(entry->type);
2541 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2542 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2543 if (!trace_print_lat_context(iter))
2546 if (!trace_print_context(iter))
2552 return event->funcs->trace(iter, sym_flags, event);
2554 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2557 return TRACE_TYPE_HANDLED;
2559 return TRACE_TYPE_PARTIAL_LINE;
2562 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2564 struct trace_seq *s = &iter->seq;
2565 struct trace_entry *entry;
2566 struct trace_event *event;
2570 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2571 if (!trace_seq_printf(s, "%d %d %llu ",
2572 entry->pid, iter->cpu, iter->ts))
2576 event = ftrace_find_event(entry->type);
2578 return event->funcs->raw(iter, 0, event);
2580 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2583 return TRACE_TYPE_HANDLED;
2585 return TRACE_TYPE_PARTIAL_LINE;
2588 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2590 struct trace_seq *s = &iter->seq;
2591 unsigned char newline = '\n';
2592 struct trace_entry *entry;
2593 struct trace_event *event;
2597 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2598 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2599 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2600 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2603 event = ftrace_find_event(entry->type);
2605 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2606 if (ret != TRACE_TYPE_HANDLED)
2610 SEQ_PUT_FIELD_RET(s, newline);
2612 return TRACE_TYPE_HANDLED;
2615 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2617 struct trace_seq *s = &iter->seq;
2618 struct trace_entry *entry;
2619 struct trace_event *event;
2623 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2624 SEQ_PUT_FIELD_RET(s, entry->pid);
2625 SEQ_PUT_FIELD_RET(s, iter->cpu);
2626 SEQ_PUT_FIELD_RET(s, iter->ts);
2629 event = ftrace_find_event(entry->type);
2630 return event ? event->funcs->binary(iter, 0, event) :
2634 int trace_empty(struct trace_iterator *iter)
2636 struct ring_buffer_iter *buf_iter;
2639 /* If we are looking at one CPU buffer, only check that one */
2640 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2641 cpu = iter->cpu_file;
2642 buf_iter = trace_buffer_iter(iter, cpu);
2644 if (!ring_buffer_iter_empty(buf_iter))
2647 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2653 for_each_tracing_cpu(cpu) {
2654 buf_iter = trace_buffer_iter(iter, cpu);
2656 if (!ring_buffer_iter_empty(buf_iter))
2659 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2667 /* Called with trace_event_read_lock() held. */
2668 enum print_line_t print_trace_line(struct trace_iterator *iter)
2670 enum print_line_t ret;
2672 if (iter->lost_events &&
2673 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2674 iter->cpu, iter->lost_events))
2675 return TRACE_TYPE_PARTIAL_LINE;
2677 if (iter->trace && iter->trace->print_line) {
2678 ret = iter->trace->print_line(iter);
2679 if (ret != TRACE_TYPE_UNHANDLED)
2683 if (iter->ent->type == TRACE_BPUTS &&
2684 trace_flags & TRACE_ITER_PRINTK &&
2685 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2686 return trace_print_bputs_msg_only(iter);
2688 if (iter->ent->type == TRACE_BPRINT &&
2689 trace_flags & TRACE_ITER_PRINTK &&
2690 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2691 return trace_print_bprintk_msg_only(iter);
2693 if (iter->ent->type == TRACE_PRINT &&
2694 trace_flags & TRACE_ITER_PRINTK &&
2695 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2696 return trace_print_printk_msg_only(iter);
2698 if (trace_flags & TRACE_ITER_BIN)
2699 return print_bin_fmt(iter);
2701 if (trace_flags & TRACE_ITER_HEX)
2702 return print_hex_fmt(iter);
2704 if (trace_flags & TRACE_ITER_RAW)
2705 return print_raw_fmt(iter);
2707 return print_trace_fmt(iter);
2710 void trace_latency_header(struct seq_file *m)
2712 struct trace_iterator *iter = m->private;
2714 /* print nothing if the buffers are empty */
2715 if (trace_empty(iter))
2718 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2719 print_trace_header(m, iter);
2721 if (!(trace_flags & TRACE_ITER_VERBOSE))
2722 print_lat_help_header(m);
2725 void trace_default_header(struct seq_file *m)
2727 struct trace_iterator *iter = m->private;
2729 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2732 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2733 /* print nothing if the buffers are empty */
2734 if (trace_empty(iter))
2736 print_trace_header(m, iter);
2737 if (!(trace_flags & TRACE_ITER_VERBOSE))
2738 print_lat_help_header(m);
2740 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2741 if (trace_flags & TRACE_ITER_IRQ_INFO)
2742 print_func_help_header_irq(iter->trace_buffer, m);
2744 print_func_help_header(iter->trace_buffer, m);
2749 static void test_ftrace_alive(struct seq_file *m)
2751 if (!ftrace_is_dead())
2753 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2754 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2757 #ifdef CONFIG_TRACER_MAX_TRACE
2758 static void show_snapshot_main_help(struct seq_file *m)
2760 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2761 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2762 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2763 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate)\n");
2764 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2765 seq_printf(m, "# is not a '0' or '1')\n");
2768 static void show_snapshot_percpu_help(struct seq_file *m)
2770 seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2771 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2772 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2773 seq_printf(m, "# Takes a snapshot of the main buffer for this cpu.\n");
2775 seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n");
2776 seq_printf(m, "# Must use main snapshot file to allocate.\n");
2778 seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2779 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2780 seq_printf(m, "# is not a '0' or '1')\n");
2783 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2785 if (iter->tr->allocated_snapshot)
2786 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2788 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2790 seq_printf(m, "# Snapshot commands:\n");
2791 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2792 show_snapshot_main_help(m);
2794 show_snapshot_percpu_help(m);
2797 /* Should never be called */
2798 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2801 static int s_show(struct seq_file *m, void *v)
2803 struct trace_iterator *iter = v;
2806 if (iter->ent == NULL) {
2808 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2810 test_ftrace_alive(m);
2812 if (iter->snapshot && trace_empty(iter))
2813 print_snapshot_help(m, iter);
2814 else if (iter->trace && iter->trace->print_header)
2815 iter->trace->print_header(m);
2817 trace_default_header(m);
2819 } else if (iter->leftover) {
2821 * If we filled the seq_file buffer earlier, we
2822 * want to just show it now.
2824 ret = trace_print_seq(m, &iter->seq);
2826 /* ret should this time be zero, but you never know */
2827 iter->leftover = ret;
2830 print_trace_line(iter);
2831 ret = trace_print_seq(m, &iter->seq);
2833 * If we overflow the seq_file buffer, then it will
2834 * ask us for this data again at start up.
2836 * ret is 0 if seq_file write succeeded.
2839 iter->leftover = ret;
2846 * Should be used after trace_array_get(), trace_types_lock
2847 * ensures that i_cdev was already initialized.
2849 static inline int tracing_get_cpu(struct inode *inode)
2851 if (inode->i_cdev) /* See trace_create_cpu_file() */
2852 return (long)inode->i_cdev - 1;
2853 return RING_BUFFER_ALL_CPUS;
2856 static const struct seq_operations tracer_seq_ops = {
2863 static struct trace_iterator *
2864 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
2866 struct trace_array *tr = inode->i_private;
2867 struct trace_iterator *iter;
2870 if (tracing_disabled)
2871 return ERR_PTR(-ENODEV);
2873 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2875 return ERR_PTR(-ENOMEM);
2877 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2879 if (!iter->buffer_iter)
2883 * We make a copy of the current tracer to avoid concurrent
2884 * changes on it while we are reading.
2886 mutex_lock(&trace_types_lock);
2887 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2891 *iter->trace = *tr->current_trace;
2893 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2898 #ifdef CONFIG_TRACER_MAX_TRACE
2899 /* Currently only the top directory has a snapshot */
2900 if (tr->current_trace->print_max || snapshot)
2901 iter->trace_buffer = &tr->max_buffer;
2904 iter->trace_buffer = &tr->trace_buffer;
2905 iter->snapshot = snapshot;
2907 iter->cpu_file = tracing_get_cpu(inode);
2908 mutex_init(&iter->mutex);
2910 /* Notify the tracer early; before we stop tracing. */
2911 if (iter->trace && iter->trace->open)
2912 iter->trace->open(iter);
2914 /* Annotate start of buffers if we had overruns */
2915 if (ring_buffer_overruns(iter->trace_buffer->buffer))
2916 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2918 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2919 if (trace_clocks[tr->clock_id].in_ns)
2920 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2922 /* stop the trace while dumping if we are not opening "snapshot" */
2923 if (!iter->snapshot)
2924 tracing_stop_tr(tr);
2926 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
2927 for_each_tracing_cpu(cpu) {
2928 iter->buffer_iter[cpu] =
2929 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2931 ring_buffer_read_prepare_sync();
2932 for_each_tracing_cpu(cpu) {
2933 ring_buffer_read_start(iter->buffer_iter[cpu]);
2934 tracing_iter_reset(iter, cpu);
2937 cpu = iter->cpu_file;
2938 iter->buffer_iter[cpu] =
2939 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2940 ring_buffer_read_prepare_sync();
2941 ring_buffer_read_start(iter->buffer_iter[cpu]);
2942 tracing_iter_reset(iter, cpu);
2945 mutex_unlock(&trace_types_lock);
2950 mutex_unlock(&trace_types_lock);
2952 kfree(iter->buffer_iter);
2954 seq_release_private(inode, file);
2955 return ERR_PTR(-ENOMEM);
2958 int tracing_open_generic(struct inode *inode, struct file *filp)
2960 if (tracing_disabled)
2963 filp->private_data = inode->i_private;
2968 * Open and update trace_array ref count.
2969 * Must have the current trace_array passed to it.
2971 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
2973 struct trace_array *tr = inode->i_private;
2975 if (tracing_disabled)
2978 if (trace_array_get(tr) < 0)
2981 filp->private_data = inode->i_private;
2986 static int tracing_release(struct inode *inode, struct file *file)
2988 struct trace_array *tr = inode->i_private;
2989 struct seq_file *m = file->private_data;
2990 struct trace_iterator *iter;
2993 if (!(file->f_mode & FMODE_READ)) {
2994 trace_array_put(tr);
2998 /* Writes do not use seq_file */
3000 mutex_lock(&trace_types_lock);
3002 for_each_tracing_cpu(cpu) {
3003 if (iter->buffer_iter[cpu])
3004 ring_buffer_read_finish(iter->buffer_iter[cpu]);
3007 if (iter->trace && iter->trace->close)
3008 iter->trace->close(iter);
3010 if (!iter->snapshot)
3011 /* reenable tracing if it was previously enabled */
3012 tracing_start_tr(tr);
3014 __trace_array_put(tr);
3016 mutex_unlock(&trace_types_lock);
3018 mutex_destroy(&iter->mutex);
3019 free_cpumask_var(iter->started);
3021 kfree(iter->buffer_iter);
3022 seq_release_private(inode, file);
3027 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
3029 struct trace_array *tr = inode->i_private;
3031 trace_array_put(tr);
3035 static int tracing_single_release_tr(struct inode *inode, struct file *file)
3037 struct trace_array *tr = inode->i_private;
3039 trace_array_put(tr);
3041 return single_release(inode, file);
3044 static int tracing_open(struct inode *inode, struct file *file)
3046 struct trace_array *tr = inode->i_private;
3047 struct trace_iterator *iter;
3050 if (trace_array_get(tr) < 0)
3053 /* If this file was open for write, then erase contents */
3054 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
3055 int cpu = tracing_get_cpu(inode);
3057 if (cpu == RING_BUFFER_ALL_CPUS)
3058 tracing_reset_online_cpus(&tr->trace_buffer);
3060 tracing_reset(&tr->trace_buffer, cpu);
3063 if (file->f_mode & FMODE_READ) {
3064 iter = __tracing_open(inode, file, false);
3066 ret = PTR_ERR(iter);
3067 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
3068 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3072 trace_array_put(tr);
3078 t_next(struct seq_file *m, void *v, loff_t *pos)
3080 struct tracer *t = v;
3090 static void *t_start(struct seq_file *m, loff_t *pos)
3095 mutex_lock(&trace_types_lock);
3096 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
3102 static void t_stop(struct seq_file *m, void *p)
3104 mutex_unlock(&trace_types_lock);
3107 static int t_show(struct seq_file *m, void *v)
3109 struct tracer *t = v;
3114 seq_printf(m, "%s", t->name);
3123 static const struct seq_operations show_traces_seq_ops = {
3130 static int show_traces_open(struct inode *inode, struct file *file)
3132 if (tracing_disabled)
3135 return seq_open(file, &show_traces_seq_ops);
3139 tracing_write_stub(struct file *filp, const char __user *ubuf,
3140 size_t count, loff_t *ppos)
3145 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
3147 if (file->f_mode & FMODE_READ)
3148 return seq_lseek(file, offset, origin);
3153 static const struct file_operations tracing_fops = {
3154 .open = tracing_open,
3156 .write = tracing_write_stub,
3157 .llseek = tracing_seek,
3158 .release = tracing_release,
3161 static const struct file_operations show_traces_fops = {
3162 .open = show_traces_open,
3164 .release = seq_release,
3165 .llseek = seq_lseek,
3169 * Only trace on a CPU if the bitmask is set:
3171 static cpumask_var_t tracing_cpumask;
3174 * The tracer itself will not take this lock, but still we want
3175 * to provide a consistent cpumask to user-space:
3177 static DEFINE_MUTEX(tracing_cpumask_update_lock);
3180 * Temporary storage for the character representation of the
3181 * CPU bitmask (and one more byte for the newline):
3183 static char mask_str[NR_CPUS + 1];
3186 tracing_cpumask_read(struct file *filp, char __user *ubuf,
3187 size_t count, loff_t *ppos)
3191 mutex_lock(&tracing_cpumask_update_lock);
3193 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
3194 if (count - len < 2) {
3198 len += sprintf(mask_str + len, "\n");
3199 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3202 mutex_unlock(&tracing_cpumask_update_lock);
3208 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3209 size_t count, loff_t *ppos)
3211 struct trace_array *tr = filp->private_data;
3212 cpumask_var_t tracing_cpumask_new;
3215 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3218 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3222 mutex_lock(&tracing_cpumask_update_lock);
3224 local_irq_disable();
3225 arch_spin_lock(&ftrace_max_lock);
3226 for_each_tracing_cpu(cpu) {
3228 * Increase/decrease the disabled counter if we are
3229 * about to flip a bit in the cpumask:
3231 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
3232 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3233 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3234 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3236 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
3237 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3238 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3239 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3242 arch_spin_unlock(&ftrace_max_lock);
3245 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
3247 mutex_unlock(&tracing_cpumask_update_lock);
3248 free_cpumask_var(tracing_cpumask_new);
3253 free_cpumask_var(tracing_cpumask_new);
3258 static const struct file_operations tracing_cpumask_fops = {
3259 .open = tracing_open_generic,
3260 .read = tracing_cpumask_read,
3261 .write = tracing_cpumask_write,
3262 .llseek = generic_file_llseek,
3265 static int tracing_trace_options_show(struct seq_file *m, void *v)
3267 struct tracer_opt *trace_opts;
3268 struct trace_array *tr = m->private;
3272 mutex_lock(&trace_types_lock);
3273 tracer_flags = tr->current_trace->flags->val;
3274 trace_opts = tr->current_trace->flags->opts;
3276 for (i = 0; trace_options[i]; i++) {
3277 if (trace_flags & (1 << i))
3278 seq_printf(m, "%s\n", trace_options[i]);
3280 seq_printf(m, "no%s\n", trace_options[i]);
3283 for (i = 0; trace_opts[i].name; i++) {
3284 if (tracer_flags & trace_opts[i].bit)
3285 seq_printf(m, "%s\n", trace_opts[i].name);
3287 seq_printf(m, "no%s\n", trace_opts[i].name);
3289 mutex_unlock(&trace_types_lock);
3294 static int __set_tracer_option(struct tracer *trace,
3295 struct tracer_flags *tracer_flags,
3296 struct tracer_opt *opts, int neg)
3300 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
3305 tracer_flags->val &= ~opts->bit;
3307 tracer_flags->val |= opts->bit;
3311 /* Try to assign a tracer specific option */
3312 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
3314 struct tracer_flags *tracer_flags = trace->flags;
3315 struct tracer_opt *opts = NULL;
3318 for (i = 0; tracer_flags->opts[i].name; i++) {
3319 opts = &tracer_flags->opts[i];
3321 if (strcmp(cmp, opts->name) == 0)
3322 return __set_tracer_option(trace, trace->flags,
3329 /* Some tracers require overwrite to stay enabled */
3330 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3332 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3338 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3340 /* do nothing if flag is already set */
3341 if (!!(trace_flags & mask) == !!enabled)
3344 /* Give the tracer a chance to approve the change */
3345 if (tr->current_trace->flag_changed)
3346 if (tr->current_trace->flag_changed(tr->current_trace, mask, !!enabled))
3350 trace_flags |= mask;
3352 trace_flags &= ~mask;
3354 if (mask == TRACE_ITER_RECORD_CMD)
3355 trace_event_enable_cmd_record(enabled);
3357 if (mask == TRACE_ITER_OVERWRITE) {
3358 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3359 #ifdef CONFIG_TRACER_MAX_TRACE
3360 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3364 if (mask == TRACE_ITER_PRINTK)
3365 trace_printk_start_stop_comm(enabled);
3370 static int trace_set_options(struct trace_array *tr, char *option)
3377 cmp = strstrip(option);
3379 if (strncmp(cmp, "no", 2) == 0) {
3384 mutex_lock(&trace_types_lock);
3386 for (i = 0; trace_options[i]; i++) {
3387 if (strcmp(cmp, trace_options[i]) == 0) {
3388 ret = set_tracer_flag(tr, 1 << i, !neg);
3393 /* If no option could be set, test the specific tracer options */
3394 if (!trace_options[i])
3395 ret = set_tracer_option(tr->current_trace, cmp, neg);
3397 mutex_unlock(&trace_types_lock);
3403 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3404 size_t cnt, loff_t *ppos)
3406 struct seq_file *m = filp->private_data;
3407 struct trace_array *tr = m->private;
3411 if (cnt >= sizeof(buf))
3414 if (copy_from_user(&buf, ubuf, cnt))
3419 ret = trace_set_options(tr, buf);
3428 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3430 struct trace_array *tr = inode->i_private;
3433 if (tracing_disabled)
3436 if (trace_array_get(tr) < 0)
3439 ret = single_open(file, tracing_trace_options_show, inode->i_private);
3441 trace_array_put(tr);
3446 static const struct file_operations tracing_iter_fops = {
3447 .open = tracing_trace_options_open,
3449 .llseek = seq_lseek,
3450 .release = tracing_single_release_tr,
3451 .write = tracing_trace_options_write,
3454 static const char readme_msg[] =
3455 "tracing mini-HOWTO:\n\n"
3456 "# echo 0 > tracing_on : quick way to disable tracing\n"
3457 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3458 " Important files:\n"
3459 " trace\t\t\t- The static contents of the buffer\n"
3460 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3461 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3462 " current_tracer\t- function and latency tracers\n"
3463 " available_tracers\t- list of configured tracers for current_tracer\n"
3464 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3465 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3466 " trace_clock\t\t-change the clock used to order events\n"
3467 " local: Per cpu clock but may not be synced across CPUs\n"
3468 " global: Synced across CPUs but slows tracing down.\n"
3469 " counter: Not a clock, but just an increment\n"
3470 " uptime: Jiffy counter from time of boot\n"
3471 " perf: Same clock that perf events use\n"
3472 #ifdef CONFIG_X86_64
3473 " x86-tsc: TSC cycle counter\n"
3475 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3476 " tracing_cpumask\t- Limit which CPUs to trace\n"
3477 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3478 "\t\t\t Remove sub-buffer with rmdir\n"
3479 " trace_options\t\t- Set format or modify how tracing happens\n"
3480 "\t\t\t Disable an option by adding a suffix 'no' to the option name\n"
3481 #ifdef CONFIG_DYNAMIC_FTRACE
3482 "\n available_filter_functions - list of functions that can be filtered on\n"
3483 " set_ftrace_filter\t- echo function name in here to only trace these functions\n"
3484 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3485 " modules: Can select a group via module\n"
3486 " Format: :mod:<module-name>\n"
3487 " example: echo :mod:ext3 > set_ftrace_filter\n"
3488 " triggers: a command to perform when function is hit\n"
3489 " Format: <function>:<trigger>[:count]\n"
3490 " trigger: traceon, traceoff\n"
3491 " enable_event:<system>:<event>\n"
3492 " disable_event:<system>:<event>\n"
3493 #ifdef CONFIG_STACKTRACE
3496 #ifdef CONFIG_TRACER_SNAPSHOT
3499 " example: echo do_fault:traceoff > set_ftrace_filter\n"
3500 " echo do_trap:traceoff:3 > set_ftrace_filter\n"
3501 " The first one will disable tracing every time do_fault is hit\n"
3502 " The second will disable tracing at most 3 times when do_trap is hit\n"
3503 " The first time do trap is hit and it disables tracing, the counter\n"
3504 " will decrement to 2. If tracing is already disabled, the counter\n"
3505 " will not decrement. It only decrements when the trigger did work\n"
3506 " To remove trigger without count:\n"
3507 " echo '!<function>:<trigger> > set_ftrace_filter\n"
3508 " To remove trigger with a count:\n"
3509 " echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3510 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3511 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3512 " modules: Can select a group via module command :mod:\n"
3513 " Does not accept triggers\n"
3514 #endif /* CONFIG_DYNAMIC_FTRACE */
3515 #ifdef CONFIG_FUNCTION_TRACER
3516 " set_ftrace_pid\t- Write pid(s) to only function trace those pids (function)\n"
3518 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3519 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3520 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3522 #ifdef CONFIG_TRACER_SNAPSHOT
3523 "\n snapshot\t\t- Like 'trace' but shows the content of the static snapshot buffer\n"
3524 "\t\t\t Read the contents for more information\n"
3526 #ifdef CONFIG_STACKTRACE
3527 " stack_trace\t\t- Shows the max stack trace when active\n"
3528 " stack_max_size\t- Shows current max stack size that was traced\n"
3529 "\t\t\t Write into this file to reset the max size (trigger a new trace)\n"
3530 #ifdef CONFIG_DYNAMIC_FTRACE
3531 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace traces\n"
3533 #endif /* CONFIG_STACKTRACE */
3537 tracing_readme_read(struct file *filp, char __user *ubuf,
3538 size_t cnt, loff_t *ppos)
3540 return simple_read_from_buffer(ubuf, cnt, ppos,
3541 readme_msg, strlen(readme_msg));
3544 static const struct file_operations tracing_readme_fops = {
3545 .open = tracing_open_generic,
3546 .read = tracing_readme_read,
3547 .llseek = generic_file_llseek,
3551 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3552 size_t cnt, loff_t *ppos)
3561 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3565 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3573 for (i = 0; i < SAVED_CMDLINES; i++) {
3576 pid = map_cmdline_to_pid[i];
3577 if (pid == -1 || pid == NO_CMDLINE_MAP)
3580 trace_find_cmdline(pid, buf_comm);
3581 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3586 len = simple_read_from_buffer(ubuf, cnt, ppos,
3595 static const struct file_operations tracing_saved_cmdlines_fops = {
3596 .open = tracing_open_generic,
3597 .read = tracing_saved_cmdlines_read,
3598 .llseek = generic_file_llseek,
3602 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3603 size_t cnt, loff_t *ppos)
3605 struct trace_array *tr = filp->private_data;
3606 char buf[MAX_TRACER_SIZE+2];
3609 mutex_lock(&trace_types_lock);
3610 r = sprintf(buf, "%s\n", tr->current_trace->name);
3611 mutex_unlock(&trace_types_lock);
3613 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3616 int tracer_init(struct tracer *t, struct trace_array *tr)
3618 tracing_reset_online_cpus(&tr->trace_buffer);
3622 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3626 for_each_tracing_cpu(cpu)
3627 per_cpu_ptr(buf->data, cpu)->entries = val;
3630 #ifdef CONFIG_TRACER_MAX_TRACE
3631 /* resize @tr's buffer to the size of @size_tr's entries */
3632 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3633 struct trace_buffer *size_buf, int cpu_id)
3637 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3638 for_each_tracing_cpu(cpu) {
3639 ret = ring_buffer_resize(trace_buf->buffer,
3640 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3643 per_cpu_ptr(trace_buf->data, cpu)->entries =
3644 per_cpu_ptr(size_buf->data, cpu)->entries;
3647 ret = ring_buffer_resize(trace_buf->buffer,
3648 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3650 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3651 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3656 #endif /* CONFIG_TRACER_MAX_TRACE */
3658 static int __tracing_resize_ring_buffer(struct trace_array *tr,
3659 unsigned long size, int cpu)
3664 * If kernel or user changes the size of the ring buffer
3665 * we use the size that was given, and we can forget about
3666 * expanding it later.
3668 ring_buffer_expanded = true;
3670 /* May be called before buffers are initialized */
3671 if (!tr->trace_buffer.buffer)
3674 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3678 #ifdef CONFIG_TRACER_MAX_TRACE
3679 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3680 !tr->current_trace->use_max_tr)
3683 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3685 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3686 &tr->trace_buffer, cpu);
3689 * AARGH! We are left with different
3690 * size max buffer!!!!
3691 * The max buffer is our "snapshot" buffer.
3692 * When a tracer needs a snapshot (one of the
3693 * latency tracers), it swaps the max buffer
3694 * with the saved snap shot. We succeeded to
3695 * update the size of the main buffer, but failed to
3696 * update the size of the max buffer. But when we tried
3697 * to reset the main buffer to the original size, we
3698 * failed there too. This is very unlikely to
3699 * happen, but if it does, warn and kill all
3703 tracing_disabled = 1;
3708 if (cpu == RING_BUFFER_ALL_CPUS)
3709 set_buffer_entries(&tr->max_buffer, size);
3711 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3714 #endif /* CONFIG_TRACER_MAX_TRACE */
3716 if (cpu == RING_BUFFER_ALL_CPUS)
3717 set_buffer_entries(&tr->trace_buffer, size);
3719 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3724 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3725 unsigned long size, int cpu_id)
3729 mutex_lock(&trace_types_lock);
3731 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3732 /* make sure, this cpu is enabled in the mask */
3733 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3739 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3744 mutex_unlock(&trace_types_lock);
3751 * tracing_update_buffers - used by tracing facility to expand ring buffers
3753 * To save on memory when the tracing is never used on a system with it
3754 * configured in. The ring buffers are set to a minimum size. But once
3755 * a user starts to use the tracing facility, then they need to grow
3756 * to their default size.
3758 * This function is to be called when a tracer is about to be used.
3760 int tracing_update_buffers(void)
3764 mutex_lock(&trace_types_lock);
3765 if (!ring_buffer_expanded)
3766 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3767 RING_BUFFER_ALL_CPUS);
3768 mutex_unlock(&trace_types_lock);
3773 struct trace_option_dentry;
3775 static struct trace_option_dentry *
3776 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3779 destroy_trace_option_files(struct trace_option_dentry *topts);
3781 static int tracing_set_tracer(const char *buf)
3783 static struct trace_option_dentry *topts;
3784 struct trace_array *tr = &global_trace;
3786 #ifdef CONFIG_TRACER_MAX_TRACE
3791 mutex_lock(&trace_types_lock);
3793 if (!ring_buffer_expanded) {
3794 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3795 RING_BUFFER_ALL_CPUS);
3801 for (t = trace_types; t; t = t->next) {
3802 if (strcmp(t->name, buf) == 0)
3809 if (t == tr->current_trace)
3812 trace_branch_disable();
3814 tr->current_trace->enabled = false;
3816 if (tr->current_trace->reset)
3817 tr->current_trace->reset(tr);
3819 /* Current trace needs to be nop_trace before synchronize_sched */
3820 tr->current_trace = &nop_trace;
3822 #ifdef CONFIG_TRACER_MAX_TRACE
3823 had_max_tr = tr->allocated_snapshot;
3825 if (had_max_tr && !t->use_max_tr) {
3827 * We need to make sure that the update_max_tr sees that
3828 * current_trace changed to nop_trace to keep it from
3829 * swapping the buffers after we resize it.
3830 * The update_max_tr is called from interrupts disabled
3831 * so a synchronized_sched() is sufficient.
3833 synchronize_sched();
3837 destroy_trace_option_files(topts);
3839 topts = create_trace_option_files(tr, t);
3841 #ifdef CONFIG_TRACER_MAX_TRACE
3842 if (t->use_max_tr && !had_max_tr) {
3843 ret = alloc_snapshot(tr);
3850 ret = tracer_init(t, tr);
3855 tr->current_trace = t;
3856 tr->current_trace->enabled = true;
3857 trace_branch_enable(tr);
3859 mutex_unlock(&trace_types_lock);
3865 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3866 size_t cnt, loff_t *ppos)
3868 char buf[MAX_TRACER_SIZE+1];
3875 if (cnt > MAX_TRACER_SIZE)
3876 cnt = MAX_TRACER_SIZE;
3878 if (copy_from_user(&buf, ubuf, cnt))
3883 /* strip ending whitespace. */
3884 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3887 err = tracing_set_tracer(buf);
3897 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3898 size_t cnt, loff_t *ppos)
3900 unsigned long *ptr = filp->private_data;
3904 r = snprintf(buf, sizeof(buf), "%ld\n",
3905 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3906 if (r > sizeof(buf))
3908 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3912 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3913 size_t cnt, loff_t *ppos)
3915 unsigned long *ptr = filp->private_data;
3919 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3928 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3930 struct trace_array *tr = inode->i_private;
3931 struct trace_iterator *iter;
3934 if (tracing_disabled)
3937 if (trace_array_get(tr) < 0)
3940 mutex_lock(&trace_types_lock);
3942 /* create a buffer to store the information to pass to userspace */
3943 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3946 __trace_array_put(tr);
3951 * We make a copy of the current tracer to avoid concurrent
3952 * changes on it while we are reading.
3954 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3959 *iter->trace = *tr->current_trace;
3961 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3966 /* trace pipe does not show start of buffer */
3967 cpumask_setall(iter->started);
3969 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3970 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3972 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3973 if (trace_clocks[tr->clock_id].in_ns)
3974 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3977 iter->trace_buffer = &tr->trace_buffer;
3978 iter->cpu_file = tracing_get_cpu(inode);
3979 mutex_init(&iter->mutex);
3980 filp->private_data = iter;
3982 if (iter->trace->pipe_open)
3983 iter->trace->pipe_open(iter);
3985 nonseekable_open(inode, filp);
3987 mutex_unlock(&trace_types_lock);
3993 __trace_array_put(tr);
3994 mutex_unlock(&trace_types_lock);
3998 static int tracing_release_pipe(struct inode *inode, struct file *file)
4000 struct trace_iterator *iter = file->private_data;
4001 struct trace_array *tr = inode->i_private;
4003 mutex_lock(&trace_types_lock);
4005 if (iter->trace->pipe_close)
4006 iter->trace->pipe_close(iter);
4008 mutex_unlock(&trace_types_lock);
4010 free_cpumask_var(iter->started);
4011 mutex_destroy(&iter->mutex);
4015 trace_array_put(tr);
4021 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
4023 /* Iterators are static, they should be filled or empty */
4024 if (trace_buffer_iter(iter, iter->cpu_file))
4025 return POLLIN | POLLRDNORM;
4027 if (trace_flags & TRACE_ITER_BLOCK)
4029 * Always select as readable when in blocking mode
4031 return POLLIN | POLLRDNORM;
4033 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
4038 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
4040 struct trace_iterator *iter = filp->private_data;
4042 return trace_poll(iter, filp, poll_table);
4046 * This is a make-shift waitqueue.
4047 * A tracer might use this callback on some rare cases:
4049 * 1) the current tracer might hold the runqueue lock when it wakes up
4050 * a reader, hence a deadlock (sched, function, and function graph tracers)
4051 * 2) the function tracers, trace all functions, we don't want
4052 * the overhead of calling wake_up and friends
4053 * (and tracing them too)
4055 * Anyway, this is really very primitive wakeup.
4057 int poll_wait_pipe(struct trace_iterator *iter)
4059 set_current_state(TASK_INTERRUPTIBLE);
4060 /* sleep for 100 msecs, and try again. */
4061 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;
4071 while (trace_empty(iter)) {
4073 if ((filp->f_flags & O_NONBLOCK)) {
4077 mutex_unlock(&iter->mutex);
4079 ret = iter->trace->wait_pipe(iter);
4081 mutex_lock(&iter->mutex);
4086 if (signal_pending(current))
4090 * We block until we read something and tracing is disabled.
4091 * We still block if tracing is disabled, but we have never
4092 * read anything. This allows a user to cat this file, and
4093 * then enable tracing. But after we have read something,
4094 * we give an EOF when tracing is again disabled.
4096 * iter->pos will be 0 if we haven't read anything.
4098 if (!tracing_is_on() && iter->pos)
4109 tracing_read_pipe(struct file *filp, char __user *ubuf,
4110 size_t cnt, loff_t *ppos)
4112 struct trace_iterator *iter = filp->private_data;
4113 struct trace_array *tr = iter->tr;
4116 /* return any leftover data */
4117 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4121 trace_seq_init(&iter->seq);
4123 /* copy the tracer to avoid using a global lock all around */
4124 mutex_lock(&trace_types_lock);
4125 if (unlikely(iter->trace->name != tr->current_trace->name))
4126 *iter->trace = *tr->current_trace;
4127 mutex_unlock(&trace_types_lock);
4130 * Avoid more than one consumer on a single file descriptor
4131 * This is just a matter of traces coherency, the ring buffer itself
4134 mutex_lock(&iter->mutex);
4135 if (iter->trace->read) {
4136 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
4142 sret = tracing_wait_pipe(filp);
4146 /* stop when tracing is finished */
4147 if (trace_empty(iter)) {
4152 if (cnt >= PAGE_SIZE)
4153 cnt = PAGE_SIZE - 1;
4155 /* reset all but tr, trace, and overruns */
4156 memset(&iter->seq, 0,
4157 sizeof(struct trace_iterator) -
4158 offsetof(struct trace_iterator, seq));
4159 cpumask_clear(iter->started);
4162 trace_event_read_lock();
4163 trace_access_lock(iter->cpu_file);
4164 while (trace_find_next_entry_inc(iter) != NULL) {
4165 enum print_line_t ret;
4166 int len = iter->seq.len;
4168 ret = print_trace_line(iter);
4169 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4170 /* don't print partial lines */
4171 iter->seq.len = len;
4174 if (ret != TRACE_TYPE_NO_CONSUME)
4175 trace_consume(iter);
4177 if (iter->seq.len >= cnt)
4181 * Setting the full flag means we reached the trace_seq buffer
4182 * size and we should leave by partial output condition above.
4183 * One of the trace_seq_* functions is not used properly.
4185 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4188 trace_access_unlock(iter->cpu_file);
4189 trace_event_read_unlock();
4191 /* Now copy what we have to the user */
4192 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4193 if (iter->seq.readpos >= iter->seq.len)
4194 trace_seq_init(&iter->seq);
4197 * If there was nothing to send to user, in spite of consuming trace
4198 * entries, go back to wait for more entries.
4204 mutex_unlock(&iter->mutex);
4209 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
4210 struct pipe_buffer *buf)
4212 __free_page(buf->page);
4215 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4218 __free_page(spd->pages[idx]);
4221 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4223 .map = generic_pipe_buf_map,
4224 .unmap = generic_pipe_buf_unmap,
4225 .confirm = generic_pipe_buf_confirm,
4226 .release = tracing_pipe_buf_release,
4227 .steal = generic_pipe_buf_steal,
4228 .get = generic_pipe_buf_get,
4232 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4237 /* Seq buffer is page-sized, exactly what we need. */
4239 count = iter->seq.len;
4240 ret = print_trace_line(iter);
4241 count = iter->seq.len - count;
4244 iter->seq.len -= count;
4247 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4248 iter->seq.len -= count;
4252 if (ret != TRACE_TYPE_NO_CONSUME)
4253 trace_consume(iter);
4255 if (!trace_find_next_entry_inc(iter)) {
4265 static ssize_t tracing_splice_read_pipe(struct file *filp,
4267 struct pipe_inode_info *pipe,
4271 struct page *pages_def[PIPE_DEF_BUFFERS];
4272 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4273 struct trace_iterator *iter = filp->private_data;
4274 struct splice_pipe_desc spd = {
4276 .partial = partial_def,
4277 .nr_pages = 0, /* This gets updated below. */
4278 .nr_pages_max = PIPE_DEF_BUFFERS,
4280 .ops = &tracing_pipe_buf_ops,
4281 .spd_release = tracing_spd_release_pipe,
4283 struct trace_array *tr = iter->tr;
4288 if (splice_grow_spd(pipe, &spd))
4291 /* copy the tracer to avoid using a global lock all around */
4292 mutex_lock(&trace_types_lock);
4293 if (unlikely(iter->trace->name != tr->current_trace->name))
4294 *iter->trace = *tr->current_trace;
4295 mutex_unlock(&trace_types_lock);
4297 mutex_lock(&iter->mutex);
4299 if (iter->trace->splice_read) {
4300 ret = iter->trace->splice_read(iter, filp,
4301 ppos, pipe, len, flags);
4306 ret = tracing_wait_pipe(filp);
4310 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4315 trace_event_read_lock();
4316 trace_access_lock(iter->cpu_file);
4318 /* Fill as many pages as possible. */
4319 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
4320 spd.pages[i] = alloc_page(GFP_KERNEL);
4324 rem = tracing_fill_pipe_page(rem, iter);
4326 /* Copy the data into the page, so we can start over. */
4327 ret = trace_seq_to_buffer(&iter->seq,
4328 page_address(spd.pages[i]),
4331 __free_page(spd.pages[i]);
4334 spd.partial[i].offset = 0;
4335 spd.partial[i].len = iter->seq.len;
4337 trace_seq_init(&iter->seq);
4340 trace_access_unlock(iter->cpu_file);
4341 trace_event_read_unlock();
4342 mutex_unlock(&iter->mutex);
4346 ret = splice_to_pipe(pipe, &spd);
4348 splice_shrink_spd(&spd);
4352 mutex_unlock(&iter->mutex);
4357 tracing_entries_read(struct file *filp, char __user *ubuf,
4358 size_t cnt, loff_t *ppos)
4360 struct inode *inode = file_inode(filp);
4361 struct trace_array *tr = inode->i_private;
4362 int cpu = tracing_get_cpu(inode);
4367 mutex_lock(&trace_types_lock);
4369 if (cpu == RING_BUFFER_ALL_CPUS) {
4370 int cpu, buf_size_same;
4375 /* check if all cpu sizes are same */
4376 for_each_tracing_cpu(cpu) {
4377 /* fill in the size from first enabled cpu */
4379 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4380 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4386 if (buf_size_same) {
4387 if (!ring_buffer_expanded)
4388 r = sprintf(buf, "%lu (expanded: %lu)\n",
4390 trace_buf_size >> 10);
4392 r = sprintf(buf, "%lu\n", size >> 10);
4394 r = sprintf(buf, "X\n");
4396 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10);
4398 mutex_unlock(&trace_types_lock);
4400 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4405 tracing_entries_write(struct file *filp, const char __user *ubuf,
4406 size_t cnt, loff_t *ppos)
4408 struct inode *inode = file_inode(filp);
4409 struct trace_array *tr = inode->i_private;
4413 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4417 /* must have at least 1 entry */
4421 /* value is in KB */
4423 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
4433 tracing_total_entries_read(struct file *filp, char __user *ubuf,
4434 size_t cnt, loff_t *ppos)
4436 struct trace_array *tr = filp->private_data;
4439 unsigned long size = 0, expanded_size = 0;
4441 mutex_lock(&trace_types_lock);
4442 for_each_tracing_cpu(cpu) {
4443 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
4444 if (!ring_buffer_expanded)
4445 expanded_size += trace_buf_size >> 10;
4447 if (ring_buffer_expanded)
4448 r = sprintf(buf, "%lu\n", size);
4450 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
4451 mutex_unlock(&trace_types_lock);
4453 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4457 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4458 size_t cnt, loff_t *ppos)
4461 * There is no need to read what the user has written, this function
4462 * is just to make sure that there is no error when "echo" is used
4471 tracing_free_buffer_release(struct inode *inode, struct file *filp)
4473 struct trace_array *tr = inode->i_private;
4475 /* disable tracing ? */
4476 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4477 tracer_tracing_off(tr);
4478 /* resize the ring buffer to 0 */
4479 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4481 trace_array_put(tr);
4487 tracing_mark_write(struct file *filp, const char __user *ubuf,
4488 size_t cnt, loff_t *fpos)
4490 unsigned long addr = (unsigned long)ubuf;
4491 struct trace_array *tr = filp->private_data;
4492 struct ring_buffer_event *event;
4493 struct ring_buffer *buffer;
4494 struct print_entry *entry;
4495 unsigned long irq_flags;
4496 struct page *pages[2];
4506 if (tracing_disabled)
4509 if (!(trace_flags & TRACE_ITER_MARKERS))
4512 if (cnt > TRACE_BUF_SIZE)
4513 cnt = TRACE_BUF_SIZE;
4516 * Userspace is injecting traces into the kernel trace buffer.
4517 * We want to be as non intrusive as possible.
4518 * To do so, we do not want to allocate any special buffers
4519 * or take any locks, but instead write the userspace data
4520 * straight into the ring buffer.
4522 * First we need to pin the userspace buffer into memory,
4523 * which, most likely it is, because it just referenced it.
4524 * But there's no guarantee that it is. By using get_user_pages_fast()
4525 * and kmap_atomic/kunmap_atomic() we can get access to the
4526 * pages directly. We then write the data directly into the
4529 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4531 /* check if we cross pages */
4532 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4535 offset = addr & (PAGE_SIZE - 1);
4538 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4539 if (ret < nr_pages) {
4541 put_page(pages[ret]);
4546 for (i = 0; i < nr_pages; i++)
4547 map_page[i] = kmap_atomic(pages[i]);
4549 local_save_flags(irq_flags);
4550 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4551 buffer = tr->trace_buffer.buffer;
4552 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4553 irq_flags, preempt_count());
4555 /* Ring buffer disabled, return as if not open for write */
4560 entry = ring_buffer_event_data(event);
4561 entry->ip = _THIS_IP_;
4563 if (nr_pages == 2) {
4564 len = PAGE_SIZE - offset;
4565 memcpy(&entry->buf, map_page[0] + offset, len);
4566 memcpy(&entry->buf[len], map_page[1], cnt - len);
4568 memcpy(&entry->buf, map_page[0] + offset, cnt);
4570 if (entry->buf[cnt - 1] != '\n') {
4571 entry->buf[cnt] = '\n';
4572 entry->buf[cnt + 1] = '\0';
4574 entry->buf[cnt] = '\0';
4576 __buffer_unlock_commit(buffer, event);
4583 for (i = 0; i < nr_pages; i++){
4584 kunmap_atomic(map_page[i]);
4591 static int tracing_clock_show(struct seq_file *m, void *v)
4593 struct trace_array *tr = m->private;
4596 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4598 "%s%s%s%s", i ? " " : "",
4599 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4600 i == tr->clock_id ? "]" : "");
4606 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4607 size_t cnt, loff_t *fpos)
4609 struct seq_file *m = filp->private_data;
4610 struct trace_array *tr = m->private;
4612 const char *clockstr;
4615 if (cnt >= sizeof(buf))
4618 if (copy_from_user(&buf, ubuf, cnt))
4623 clockstr = strstrip(buf);
4625 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4626 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4629 if (i == ARRAY_SIZE(trace_clocks))
4632 mutex_lock(&trace_types_lock);
4636 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4639 * New clock may not be consistent with the previous clock.
4640 * Reset the buffer so that it doesn't have incomparable timestamps.
4642 tracing_reset_online_cpus(&tr->trace_buffer);
4644 #ifdef CONFIG_TRACER_MAX_TRACE
4645 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4646 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4647 tracing_reset_online_cpus(&tr->max_buffer);
4650 mutex_unlock(&trace_types_lock);
4657 static int tracing_clock_open(struct inode *inode, struct file *file)
4659 struct trace_array *tr = inode->i_private;
4662 if (tracing_disabled)
4665 if (trace_array_get(tr))
4668 ret = single_open(file, tracing_clock_show, inode->i_private);
4670 trace_array_put(tr);
4675 struct ftrace_buffer_info {
4676 struct trace_iterator iter;
4681 #ifdef CONFIG_TRACER_SNAPSHOT
4682 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4684 struct trace_array *tr = inode->i_private;
4685 struct trace_iterator *iter;
4689 if (trace_array_get(tr) < 0)
4692 if (file->f_mode & FMODE_READ) {
4693 iter = __tracing_open(inode, file, true);
4695 ret = PTR_ERR(iter);
4697 /* Writes still need the seq_file to hold the private data */
4699 m = kzalloc(sizeof(*m), GFP_KERNEL);
4702 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4710 iter->trace_buffer = &tr->max_buffer;
4711 iter->cpu_file = tracing_get_cpu(inode);
4713 file->private_data = m;
4717 trace_array_put(tr);
4723 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4726 struct seq_file *m = filp->private_data;
4727 struct trace_iterator *iter = m->private;
4728 struct trace_array *tr = iter->tr;
4732 ret = tracing_update_buffers();
4736 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4740 mutex_lock(&trace_types_lock);
4742 if (tr->current_trace->use_max_tr) {
4749 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4753 if (tr->allocated_snapshot)
4757 /* Only allow per-cpu swap if the ring buffer supports it */
4758 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4759 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4764 if (!tr->allocated_snapshot) {
4765 ret = alloc_snapshot(tr);
4769 local_irq_disable();
4770 /* Now, we're going to swap */
4771 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4772 update_max_tr(tr, current, smp_processor_id());
4774 update_max_tr_single(tr, current, iter->cpu_file);
4778 if (tr->allocated_snapshot) {
4779 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4780 tracing_reset_online_cpus(&tr->max_buffer);
4782 tracing_reset(&tr->max_buffer, iter->cpu_file);
4792 mutex_unlock(&trace_types_lock);
4796 static int tracing_snapshot_release(struct inode *inode, struct file *file)
4798 struct seq_file *m = file->private_data;
4801 ret = tracing_release(inode, file);
4803 if (file->f_mode & FMODE_READ)
4806 /* If write only, the seq_file is just a stub */
4814 static int tracing_buffers_open(struct inode *inode, struct file *filp);
4815 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4816 size_t count, loff_t *ppos);
4817 static int tracing_buffers_release(struct inode *inode, struct file *file);
4818 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4819 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
4821 static int snapshot_raw_open(struct inode *inode, struct file *filp)
4823 struct ftrace_buffer_info *info;
4826 ret = tracing_buffers_open(inode, filp);
4830 info = filp->private_data;
4832 if (info->iter.trace->use_max_tr) {
4833 tracing_buffers_release(inode, filp);
4837 info->iter.snapshot = true;
4838 info->iter.trace_buffer = &info->iter.tr->max_buffer;
4843 #endif /* CONFIG_TRACER_SNAPSHOT */
4846 static const struct file_operations tracing_max_lat_fops = {
4847 .open = tracing_open_generic,
4848 .read = tracing_max_lat_read,
4849 .write = tracing_max_lat_write,
4850 .llseek = generic_file_llseek,
4853 static const struct file_operations set_tracer_fops = {
4854 .open = tracing_open_generic,
4855 .read = tracing_set_trace_read,
4856 .write = tracing_set_trace_write,
4857 .llseek = generic_file_llseek,
4860 static const struct file_operations tracing_pipe_fops = {
4861 .open = tracing_open_pipe,
4862 .poll = tracing_poll_pipe,
4863 .read = tracing_read_pipe,
4864 .splice_read = tracing_splice_read_pipe,
4865 .release = tracing_release_pipe,
4866 .llseek = no_llseek,
4869 static const struct file_operations tracing_entries_fops = {
4870 .open = tracing_open_generic_tr,
4871 .read = tracing_entries_read,
4872 .write = tracing_entries_write,
4873 .llseek = generic_file_llseek,
4874 .release = tracing_release_generic_tr,
4877 static const struct file_operations tracing_total_entries_fops = {
4878 .open = tracing_open_generic_tr,
4879 .read = tracing_total_entries_read,
4880 .llseek = generic_file_llseek,
4881 .release = tracing_release_generic_tr,
4884 static const struct file_operations tracing_free_buffer_fops = {
4885 .open = tracing_open_generic_tr,
4886 .write = tracing_free_buffer_write,
4887 .release = tracing_free_buffer_release,
4890 static const struct file_operations tracing_mark_fops = {
4891 .open = tracing_open_generic_tr,
4892 .write = tracing_mark_write,
4893 .llseek = generic_file_llseek,
4894 .release = tracing_release_generic_tr,
4897 static const struct file_operations trace_clock_fops = {
4898 .open = tracing_clock_open,
4900 .llseek = seq_lseek,
4901 .release = tracing_single_release_tr,
4902 .write = tracing_clock_write,
4905 #ifdef CONFIG_TRACER_SNAPSHOT
4906 static const struct file_operations snapshot_fops = {
4907 .open = tracing_snapshot_open,
4909 .write = tracing_snapshot_write,
4910 .llseek = tracing_seek,
4911 .release = tracing_snapshot_release,
4914 static const struct file_operations snapshot_raw_fops = {
4915 .open = snapshot_raw_open,
4916 .read = tracing_buffers_read,
4917 .release = tracing_buffers_release,
4918 .splice_read = tracing_buffers_splice_read,
4919 .llseek = no_llseek,
4922 #endif /* CONFIG_TRACER_SNAPSHOT */
4924 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4926 struct trace_array *tr = inode->i_private;
4927 struct ftrace_buffer_info *info;
4930 if (tracing_disabled)
4933 if (trace_array_get(tr) < 0)
4936 info = kzalloc(sizeof(*info), GFP_KERNEL);
4938 trace_array_put(tr);
4942 mutex_lock(&trace_types_lock);
4945 info->iter.cpu_file = tracing_get_cpu(inode);
4946 info->iter.trace = tr->current_trace;
4947 info->iter.trace_buffer = &tr->trace_buffer;
4949 /* Force reading ring buffer for first read */
4950 info->read = (unsigned int)-1;
4952 filp->private_data = info;
4954 mutex_unlock(&trace_types_lock);
4956 ret = nonseekable_open(inode, filp);
4958 trace_array_put(tr);
4964 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
4966 struct ftrace_buffer_info *info = filp->private_data;
4967 struct trace_iterator *iter = &info->iter;
4969 return trace_poll(iter, filp, poll_table);
4973 tracing_buffers_read(struct file *filp, char __user *ubuf,
4974 size_t count, loff_t *ppos)
4976 struct ftrace_buffer_info *info = filp->private_data;
4977 struct trace_iterator *iter = &info->iter;
4984 mutex_lock(&trace_types_lock);
4986 #ifdef CONFIG_TRACER_MAX_TRACE
4987 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
4994 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
5000 /* Do we have previous read data to read? */
5001 if (info->read < PAGE_SIZE)
5005 trace_access_lock(iter->cpu_file);
5006 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
5010 trace_access_unlock(iter->cpu_file);
5013 if (trace_empty(iter)) {
5014 if ((filp->f_flags & O_NONBLOCK)) {
5018 mutex_unlock(&trace_types_lock);
5019 ret = iter->trace->wait_pipe(iter);
5020 mutex_lock(&trace_types_lock);
5025 if (signal_pending(current)) {
5037 size = PAGE_SIZE - info->read;
5041 ret = copy_to_user(ubuf, info->spare + info->read, size);
5052 mutex_unlock(&trace_types_lock);
5057 static int tracing_buffers_release(struct inode *inode, struct file *file)
5059 struct ftrace_buffer_info *info = file->private_data;
5060 struct trace_iterator *iter = &info->iter;
5062 mutex_lock(&trace_types_lock);
5064 __trace_array_put(iter->tr);
5067 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
5070 mutex_unlock(&trace_types_lock);
5076 struct ring_buffer *buffer;
5081 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
5082 struct pipe_buffer *buf)
5084 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5089 ring_buffer_free_read_page(ref->buffer, ref->page);
5094 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
5095 struct pipe_buffer *buf)
5097 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5102 /* Pipe buffer operations for a buffer. */
5103 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
5105 .map = generic_pipe_buf_map,
5106 .unmap = generic_pipe_buf_unmap,
5107 .confirm = generic_pipe_buf_confirm,
5108 .release = buffer_pipe_buf_release,
5109 .steal = generic_pipe_buf_steal,
5110 .get = buffer_pipe_buf_get,
5114 * Callback from splice_to_pipe(), if we need to release some pages
5115 * at the end of the spd in case we error'ed out in filling the pipe.
5117 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
5119 struct buffer_ref *ref =
5120 (struct buffer_ref *)spd->partial[i].private;
5125 ring_buffer_free_read_page(ref->buffer, ref->page);
5127 spd->partial[i].private = 0;
5131 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5132 struct pipe_inode_info *pipe, size_t len,
5135 struct ftrace_buffer_info *info = file->private_data;
5136 struct trace_iterator *iter = &info->iter;
5137 struct partial_page partial_def[PIPE_DEF_BUFFERS];
5138 struct page *pages_def[PIPE_DEF_BUFFERS];
5139 struct splice_pipe_desc spd = {
5141 .partial = partial_def,
5142 .nr_pages_max = PIPE_DEF_BUFFERS,
5144 .ops = &buffer_pipe_buf_ops,
5145 .spd_release = buffer_spd_release,
5147 struct buffer_ref *ref;
5148 int entries, size, i;
5151 mutex_lock(&trace_types_lock);
5153 #ifdef CONFIG_TRACER_MAX_TRACE
5154 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5160 if (splice_grow_spd(pipe, &spd)) {
5165 if (*ppos & (PAGE_SIZE - 1)) {
5170 if (len & (PAGE_SIZE - 1)) {
5171 if (len < PAGE_SIZE) {
5179 trace_access_lock(iter->cpu_file);
5180 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5182 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
5186 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
5191 ref->buffer = iter->trace_buffer->buffer;
5192 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5198 r = ring_buffer_read_page(ref->buffer, &ref->page,
5199 len, iter->cpu_file, 1);
5201 ring_buffer_free_read_page(ref->buffer, ref->page);
5207 * zero out any left over data, this is going to
5210 size = ring_buffer_page_len(ref->page);
5211 if (size < PAGE_SIZE)
5212 memset(ref->page + size, 0, PAGE_SIZE - size);
5214 page = virt_to_page(ref->page);
5216 spd.pages[i] = page;
5217 spd.partial[i].len = PAGE_SIZE;
5218 spd.partial[i].offset = 0;
5219 spd.partial[i].private = (unsigned long)ref;
5223 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5226 trace_access_unlock(iter->cpu_file);
5229 /* did we read anything? */
5230 if (!spd.nr_pages) {
5231 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
5235 mutex_unlock(&trace_types_lock);
5236 ret = iter->trace->wait_pipe(iter);
5237 mutex_lock(&trace_types_lock);
5240 if (signal_pending(current)) {
5247 ret = splice_to_pipe(pipe, &spd);
5248 splice_shrink_spd(&spd);
5250 mutex_unlock(&trace_types_lock);
5255 static const struct file_operations tracing_buffers_fops = {
5256 .open = tracing_buffers_open,
5257 .read = tracing_buffers_read,
5258 .poll = tracing_buffers_poll,
5259 .release = tracing_buffers_release,
5260 .splice_read = tracing_buffers_splice_read,
5261 .llseek = no_llseek,
5265 tracing_stats_read(struct file *filp, char __user *ubuf,
5266 size_t count, loff_t *ppos)
5268 struct inode *inode = file_inode(filp);
5269 struct trace_array *tr = inode->i_private;
5270 struct trace_buffer *trace_buf = &tr->trace_buffer;
5271 int cpu = tracing_get_cpu(inode);
5272 struct trace_seq *s;
5274 unsigned long long t;
5275 unsigned long usec_rem;
5277 s = kmalloc(sizeof(*s), GFP_KERNEL);
5283 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5284 trace_seq_printf(s, "entries: %ld\n", cnt);
5286 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5287 trace_seq_printf(s, "overrun: %ld\n", cnt);
5289 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5290 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5292 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5293 trace_seq_printf(s, "bytes: %ld\n", cnt);
5295 if (trace_clocks[tr->clock_id].in_ns) {
5296 /* local or global for trace_clock */
5297 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5298 usec_rem = do_div(t, USEC_PER_SEC);
5299 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5302 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5303 usec_rem = do_div(t, USEC_PER_SEC);
5304 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5306 /* counter or tsc mode for trace_clock */
5307 trace_seq_printf(s, "oldest event ts: %llu\n",
5308 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5310 trace_seq_printf(s, "now ts: %llu\n",
5311 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5314 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5315 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5317 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5318 trace_seq_printf(s, "read events: %ld\n", cnt);
5320 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
5327 static const struct file_operations tracing_stats_fops = {
5328 .open = tracing_open_generic_tr,
5329 .read = tracing_stats_read,
5330 .llseek = generic_file_llseek,
5331 .release = tracing_release_generic_tr,
5334 #ifdef CONFIG_DYNAMIC_FTRACE
5336 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5342 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5343 size_t cnt, loff_t *ppos)
5345 static char ftrace_dyn_info_buffer[1024];
5346 static DEFINE_MUTEX(dyn_info_mutex);
5347 unsigned long *p = filp->private_data;
5348 char *buf = ftrace_dyn_info_buffer;
5349 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5352 mutex_lock(&dyn_info_mutex);
5353 r = sprintf(buf, "%ld ", *p);
5355 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5358 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5360 mutex_unlock(&dyn_info_mutex);
5365 static const struct file_operations tracing_dyn_info_fops = {
5366 .open = tracing_open_generic,
5367 .read = tracing_read_dyn_info,
5368 .llseek = generic_file_llseek,
5370 #endif /* CONFIG_DYNAMIC_FTRACE */
5372 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5374 ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5380 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5382 unsigned long *count = (long *)data;
5394 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5395 struct ftrace_probe_ops *ops, void *data)
5397 long count = (long)data;
5399 seq_printf(m, "%ps:", (void *)ip);
5401 seq_printf(m, "snapshot");
5404 seq_printf(m, ":unlimited\n");
5406 seq_printf(m, ":count=%ld\n", count);
5411 static struct ftrace_probe_ops snapshot_probe_ops = {
5412 .func = ftrace_snapshot,
5413 .print = ftrace_snapshot_print,
5416 static struct ftrace_probe_ops snapshot_count_probe_ops = {
5417 .func = ftrace_count_snapshot,
5418 .print = ftrace_snapshot_print,
5422 ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
5423 char *glob, char *cmd, char *param, int enable)
5425 struct ftrace_probe_ops *ops;
5426 void *count = (void *)-1;
5430 /* hash funcs only work with set_ftrace_filter */
5434 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
5436 if (glob[0] == '!') {
5437 unregister_ftrace_function_probe_func(glob+1, ops);
5444 number = strsep(¶m, ":");
5446 if (!strlen(number))
5450 * We use the callback data field (which is a pointer)
5453 ret = kstrtoul(number, 0, (unsigned long *)&count);
5458 ret = register_ftrace_function_probe(glob, ops, count);
5461 alloc_snapshot(&global_trace);
5463 return ret < 0 ? ret : 0;
5466 static struct ftrace_func_command ftrace_snapshot_cmd = {
5468 .func = ftrace_trace_snapshot_callback,
5471 static int register_snapshot_cmd(void)
5473 return register_ftrace_command(&ftrace_snapshot_cmd);
5476 static inline int register_snapshot_cmd(void) { return 0; }
5477 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5479 struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
5484 if (!debugfs_initialized())
5487 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
5488 tr->dir = debugfs_create_dir("tracing", NULL);
5491 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5496 struct dentry *tracing_init_dentry(void)
5498 return tracing_init_dentry_tr(&global_trace);
5501 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
5503 struct dentry *d_tracer;
5506 return tr->percpu_dir;
5508 d_tracer = tracing_init_dentry_tr(tr);
5512 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
5514 WARN_ONCE(!tr->percpu_dir,
5515 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
5517 return tr->percpu_dir;
5520 static struct dentry *
5521 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
5522 void *data, long cpu, const struct file_operations *fops)
5524 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
5526 if (ret) /* See tracing_get_cpu() */
5527 ret->d_inode->i_cdev = (void *)(cpu + 1);
5532 tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
5534 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
5535 struct dentry *d_cpu;
5536 char cpu_dir[30]; /* 30 characters should be more than enough */
5541 snprintf(cpu_dir, 30, "cpu%ld", cpu);
5542 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
5544 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
5548 /* per cpu trace_pipe */
5549 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
5550 tr, cpu, &tracing_pipe_fops);
5553 trace_create_cpu_file("trace", 0644, d_cpu,
5554 tr, cpu, &tracing_fops);
5556 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
5557 tr, cpu, &tracing_buffers_fops);
5559 trace_create_cpu_file("stats", 0444, d_cpu,
5560 tr, cpu, &tracing_stats_fops);
5562 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
5563 tr, cpu, &tracing_entries_fops);
5565 #ifdef CONFIG_TRACER_SNAPSHOT
5566 trace_create_cpu_file("snapshot", 0644, d_cpu,
5567 tr, cpu, &snapshot_fops);
5569 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
5570 tr, cpu, &snapshot_raw_fops);
5574 #ifdef CONFIG_FTRACE_SELFTEST
5575 /* Let selftest have access to static functions in this file */
5576 #include "trace_selftest.c"
5579 struct trace_option_dentry {
5580 struct tracer_opt *opt;
5581 struct tracer_flags *flags;
5582 struct trace_array *tr;
5583 struct dentry *entry;
5587 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
5590 struct trace_option_dentry *topt = filp->private_data;
5593 if (topt->flags->val & topt->opt->bit)
5598 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5602 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
5605 struct trace_option_dentry *topt = filp->private_data;
5609 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5613 if (val != 0 && val != 1)
5616 if (!!(topt->flags->val & topt->opt->bit) != val) {
5617 mutex_lock(&trace_types_lock);
5618 ret = __set_tracer_option(topt->tr->current_trace, topt->flags,
5620 mutex_unlock(&trace_types_lock);
5631 static const struct file_operations trace_options_fops = {
5632 .open = tracing_open_generic,
5633 .read = trace_options_read,
5634 .write = trace_options_write,
5635 .llseek = generic_file_llseek,
5639 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5642 long index = (long)filp->private_data;
5645 if (trace_flags & (1 << index))
5650 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5654 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5657 struct trace_array *tr = &global_trace;
5658 long index = (long)filp->private_data;
5662 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5666 if (val != 0 && val != 1)
5669 mutex_lock(&trace_types_lock);
5670 ret = set_tracer_flag(tr, 1 << index, val);
5671 mutex_unlock(&trace_types_lock);
5681 static const struct file_operations trace_options_core_fops = {
5682 .open = tracing_open_generic,
5683 .read = trace_options_core_read,
5684 .write = trace_options_core_write,
5685 .llseek = generic_file_llseek,
5688 struct dentry *trace_create_file(const char *name,
5690 struct dentry *parent,
5692 const struct file_operations *fops)
5696 ret = debugfs_create_file(name, mode, parent, data, fops);
5698 pr_warning("Could not create debugfs '%s' entry\n", name);
5704 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5706 struct dentry *d_tracer;
5711 d_tracer = tracing_init_dentry_tr(tr);
5715 tr->options = debugfs_create_dir("options", d_tracer);
5717 pr_warning("Could not create debugfs directory 'options'\n");
5725 create_trace_option_file(struct trace_array *tr,
5726 struct trace_option_dentry *topt,
5727 struct tracer_flags *flags,
5728 struct tracer_opt *opt)
5730 struct dentry *t_options;
5732 t_options = trace_options_init_dentry(tr);
5736 topt->flags = flags;
5740 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5741 &trace_options_fops);
5745 static struct trace_option_dentry *
5746 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5748 struct trace_option_dentry *topts;
5749 struct tracer_flags *flags;
5750 struct tracer_opt *opts;
5756 flags = tracer->flags;
5758 if (!flags || !flags->opts)
5763 for (cnt = 0; opts[cnt].name; cnt++)
5766 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5770 for (cnt = 0; opts[cnt].name; cnt++)
5771 create_trace_option_file(tr, &topts[cnt], flags,
5778 destroy_trace_option_files(struct trace_option_dentry *topts)
5785 for (cnt = 0; topts[cnt].opt; cnt++) {
5786 if (topts[cnt].entry)
5787 debugfs_remove(topts[cnt].entry);
5793 static struct dentry *
5794 create_trace_option_core_file(struct trace_array *tr,
5795 const char *option, long index)
5797 struct dentry *t_options;
5799 t_options = trace_options_init_dentry(tr);
5803 return trace_create_file(option, 0644, t_options, (void *)index,
5804 &trace_options_core_fops);
5807 static __init void create_trace_options_dir(struct trace_array *tr)
5809 struct dentry *t_options;
5812 t_options = trace_options_init_dentry(tr);
5816 for (i = 0; trace_options[i]; i++)
5817 create_trace_option_core_file(tr, trace_options[i], i);
5821 rb_simple_read(struct file *filp, char __user *ubuf,
5822 size_t cnt, loff_t *ppos)
5824 struct trace_array *tr = filp->private_data;
5828 r = tracer_tracing_is_on(tr);
5829 r = sprintf(buf, "%d\n", r);
5831 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5835 rb_simple_write(struct file *filp, const char __user *ubuf,
5836 size_t cnt, loff_t *ppos)
5838 struct trace_array *tr = filp->private_data;
5839 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5843 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5848 mutex_lock(&trace_types_lock);
5850 tracer_tracing_on(tr);
5851 if (tr->current_trace->start)
5852 tr->current_trace->start(tr);
5854 tracer_tracing_off(tr);
5855 if (tr->current_trace->stop)
5856 tr->current_trace->stop(tr);
5858 mutex_unlock(&trace_types_lock);
5866 static const struct file_operations rb_simple_fops = {
5867 .open = tracing_open_generic_tr,
5868 .read = rb_simple_read,
5869 .write = rb_simple_write,
5870 .release = tracing_release_generic_tr,
5871 .llseek = default_llseek,
5874 struct dentry *trace_instance_dir;
5877 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
5879 static void init_trace_buffers(struct trace_array *tr, struct trace_buffer *buf)
5883 for_each_tracing_cpu(cpu) {
5884 memset(per_cpu_ptr(buf->data, cpu), 0, sizeof(struct trace_array_cpu));
5885 per_cpu_ptr(buf->data, cpu)->trace_cpu.cpu = cpu;
5886 per_cpu_ptr(buf->data, cpu)->trace_cpu.tr = tr;
5891 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
5893 enum ring_buffer_flags rb_flags;
5895 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5899 buf->buffer = ring_buffer_alloc(size, rb_flags);
5903 buf->data = alloc_percpu(struct trace_array_cpu);
5905 ring_buffer_free(buf->buffer);
5909 init_trace_buffers(tr, buf);
5911 /* Allocate the first page for all buffers */
5912 set_buffer_entries(&tr->trace_buffer,
5913 ring_buffer_size(tr->trace_buffer.buffer, 0));
5918 static int allocate_trace_buffers(struct trace_array *tr, int size)
5922 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
5926 #ifdef CONFIG_TRACER_MAX_TRACE
5927 ret = allocate_trace_buffer(tr, &tr->max_buffer,
5928 allocate_snapshot ? size : 1);
5930 ring_buffer_free(tr->trace_buffer.buffer);
5931 free_percpu(tr->trace_buffer.data);
5934 tr->allocated_snapshot = allocate_snapshot;
5937 * Only the top level trace array gets its snapshot allocated
5938 * from the kernel command line.
5940 allocate_snapshot = false;
5945 static int new_instance_create(const char *name)
5947 struct trace_array *tr;
5950 mutex_lock(&trace_types_lock);
5953 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5954 if (tr->name && strcmp(tr->name, name) == 0)
5959 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
5963 tr->name = kstrdup(name, GFP_KERNEL);
5967 raw_spin_lock_init(&tr->start_lock);
5969 tr->current_trace = &nop_trace;
5971 INIT_LIST_HEAD(&tr->systems);
5972 INIT_LIST_HEAD(&tr->events);
5974 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
5977 /* Holder for file callbacks */
5978 tr->trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
5979 tr->trace_cpu.tr = tr;
5981 tr->dir = debugfs_create_dir(name, trace_instance_dir);
5985 ret = event_trace_add_tracer(tr->dir, tr);
5987 debugfs_remove_recursive(tr->dir);
5991 init_tracer_debugfs(tr, tr->dir);
5993 list_add(&tr->list, &ftrace_trace_arrays);
5995 mutex_unlock(&trace_types_lock);
6000 if (tr->trace_buffer.buffer)
6001 ring_buffer_free(tr->trace_buffer.buffer);
6006 mutex_unlock(&trace_types_lock);
6012 static int instance_delete(const char *name)
6014 struct trace_array *tr;
6018 mutex_lock(&trace_types_lock);
6021 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6022 if (tr->name && strcmp(tr->name, name) == 0) {
6034 list_del(&tr->list);
6036 event_trace_del_tracer(tr);
6037 debugfs_remove_recursive(tr->dir);
6038 free_percpu(tr->trace_buffer.data);
6039 ring_buffer_free(tr->trace_buffer.buffer);
6047 mutex_unlock(&trace_types_lock);
6052 static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
6054 struct dentry *parent;
6057 /* Paranoid: Make sure the parent is the "instances" directory */
6058 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6059 if (WARN_ON_ONCE(parent != trace_instance_dir))
6063 * The inode mutex is locked, but debugfs_create_dir() will also
6064 * take the mutex. As the instances directory can not be destroyed
6065 * or changed in any other way, it is safe to unlock it, and
6066 * let the dentry try. If two users try to make the same dir at
6067 * the same time, then the new_instance_create() will determine the
6070 mutex_unlock(&inode->i_mutex);
6072 ret = new_instance_create(dentry->d_iname);
6074 mutex_lock(&inode->i_mutex);
6079 static int instance_rmdir(struct inode *inode, struct dentry *dentry)
6081 struct dentry *parent;
6084 /* Paranoid: Make sure the parent is the "instances" directory */
6085 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6086 if (WARN_ON_ONCE(parent != trace_instance_dir))
6089 /* The caller did a dget() on dentry */
6090 mutex_unlock(&dentry->d_inode->i_mutex);
6093 * The inode mutex is locked, but debugfs_create_dir() will also
6094 * take the mutex. As the instances directory can not be destroyed
6095 * or changed in any other way, it is safe to unlock it, and
6096 * let the dentry try. If two users try to make the same dir at
6097 * the same time, then the instance_delete() will determine the
6100 mutex_unlock(&inode->i_mutex);
6102 ret = instance_delete(dentry->d_iname);
6104 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
6105 mutex_lock(&dentry->d_inode->i_mutex);
6110 static const struct inode_operations instance_dir_inode_operations = {
6111 .lookup = simple_lookup,
6112 .mkdir = instance_mkdir,
6113 .rmdir = instance_rmdir,
6116 static __init void create_trace_instances(struct dentry *d_tracer)
6118 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
6119 if (WARN_ON(!trace_instance_dir))
6122 /* Hijack the dir inode operations, to allow mkdir */
6123 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
6127 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
6131 trace_create_file("trace_options", 0644, d_tracer,
6132 tr, &tracing_iter_fops);
6134 trace_create_file("trace", 0644, d_tracer,
6137 trace_create_file("trace_pipe", 0444, d_tracer,
6138 tr, &tracing_pipe_fops);
6140 trace_create_file("buffer_size_kb", 0644, d_tracer,
6141 tr, &tracing_entries_fops);
6143 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
6144 tr, &tracing_total_entries_fops);
6146 trace_create_file("free_buffer", 0644, d_tracer,
6147 tr, &tracing_free_buffer_fops);
6149 trace_create_file("trace_marker", 0220, d_tracer,
6150 tr, &tracing_mark_fops);
6152 trace_create_file("trace_clock", 0644, d_tracer, tr,
6155 trace_create_file("tracing_on", 0644, d_tracer,
6156 tr, &rb_simple_fops);
6158 #ifdef CONFIG_TRACER_SNAPSHOT
6159 trace_create_file("snapshot", 0644, d_tracer,
6160 tr, &snapshot_fops);
6163 for_each_tracing_cpu(cpu)
6164 tracing_init_debugfs_percpu(tr, cpu);
6168 static __init int tracer_init_debugfs(void)
6170 struct dentry *d_tracer;
6172 trace_access_lock_init();
6174 d_tracer = tracing_init_dentry();
6178 init_tracer_debugfs(&global_trace, d_tracer);
6180 trace_create_file("tracing_cpumask", 0644, d_tracer,
6181 &global_trace, &tracing_cpumask_fops);
6183 trace_create_file("available_tracers", 0444, d_tracer,
6184 &global_trace, &show_traces_fops);
6186 trace_create_file("current_tracer", 0644, d_tracer,
6187 &global_trace, &set_tracer_fops);
6189 #ifdef CONFIG_TRACER_MAX_TRACE
6190 trace_create_file("tracing_max_latency", 0644, d_tracer,
6191 &tracing_max_latency, &tracing_max_lat_fops);
6194 trace_create_file("tracing_thresh", 0644, d_tracer,
6195 &tracing_thresh, &tracing_max_lat_fops);
6197 trace_create_file("README", 0444, d_tracer,
6198 NULL, &tracing_readme_fops);
6200 trace_create_file("saved_cmdlines", 0444, d_tracer,
6201 NULL, &tracing_saved_cmdlines_fops);
6203 #ifdef CONFIG_DYNAMIC_FTRACE
6204 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
6205 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
6208 create_trace_instances(d_tracer);
6210 create_trace_options_dir(&global_trace);
6215 static int trace_panic_handler(struct notifier_block *this,
6216 unsigned long event, void *unused)
6218 if (ftrace_dump_on_oops)
6219 ftrace_dump(ftrace_dump_on_oops);
6223 static struct notifier_block trace_panic_notifier = {
6224 .notifier_call = trace_panic_handler,
6226 .priority = 150 /* priority: INT_MAX >= x >= 0 */
6229 static int trace_die_handler(struct notifier_block *self,
6235 if (ftrace_dump_on_oops)
6236 ftrace_dump(ftrace_dump_on_oops);
6244 static struct notifier_block trace_die_notifier = {
6245 .notifier_call = trace_die_handler,
6250 * printk is set to max of 1024, we really don't need it that big.
6251 * Nothing should be printing 1000 characters anyway.
6253 #define TRACE_MAX_PRINT 1000
6256 * Define here KERN_TRACE so that we have one place to modify
6257 * it if we decide to change what log level the ftrace dump
6260 #define KERN_TRACE KERN_EMERG
6263 trace_printk_seq(struct trace_seq *s)
6265 /* Probably should print a warning here. */
6266 if (s->len >= TRACE_MAX_PRINT)
6267 s->len = TRACE_MAX_PRINT;
6269 /* should be zero ended, but we are paranoid. */
6270 s->buffer[s->len] = 0;
6272 printk(KERN_TRACE "%s", s->buffer);
6277 void trace_init_global_iter(struct trace_iterator *iter)
6279 iter->tr = &global_trace;
6280 iter->trace = iter->tr->current_trace;
6281 iter->cpu_file = RING_BUFFER_ALL_CPUS;
6282 iter->trace_buffer = &global_trace.trace_buffer;
6285 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
6287 /* use static because iter can be a bit big for the stack */
6288 static struct trace_iterator iter;
6289 static atomic_t dump_running;
6290 unsigned int old_userobj;
6291 unsigned long flags;
6294 /* Only allow one dump user at a time. */
6295 if (atomic_inc_return(&dump_running) != 1) {
6296 atomic_dec(&dump_running);
6301 * Always turn off tracing when we dump.
6302 * We don't need to show trace output of what happens
6303 * between multiple crashes.
6305 * If the user does a sysrq-z, then they can re-enable
6306 * tracing with echo 1 > tracing_on.
6310 local_irq_save(flags);
6312 /* Simulate the iterator */
6313 trace_init_global_iter(&iter);
6315 for_each_tracing_cpu(cpu) {
6316 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
6319 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
6321 /* don't look at user memory in panic mode */
6322 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
6324 switch (oops_dump_mode) {
6326 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6329 iter.cpu_file = raw_smp_processor_id();
6334 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
6335 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6338 printk(KERN_TRACE "Dumping ftrace buffer:\n");
6340 /* Did function tracer already get disabled? */
6341 if (ftrace_is_dead()) {
6342 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6343 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6347 * We need to stop all tracing on all CPUS to read the
6348 * the next buffer. This is a bit expensive, but is
6349 * not done often. We fill all what we can read,
6350 * and then release the locks again.
6353 while (!trace_empty(&iter)) {
6356 printk(KERN_TRACE "---------------------------------\n");
6360 /* reset all but tr, trace, and overruns */
6361 memset(&iter.seq, 0,
6362 sizeof(struct trace_iterator) -
6363 offsetof(struct trace_iterator, seq));
6364 iter.iter_flags |= TRACE_FILE_LAT_FMT;
6367 if (trace_find_next_entry_inc(&iter) != NULL) {
6370 ret = print_trace_line(&iter);
6371 if (ret != TRACE_TYPE_NO_CONSUME)
6372 trace_consume(&iter);
6374 touch_nmi_watchdog();
6376 trace_printk_seq(&iter.seq);
6380 printk(KERN_TRACE " (ftrace buffer empty)\n");
6382 printk(KERN_TRACE "---------------------------------\n");
6385 trace_flags |= old_userobj;
6387 for_each_tracing_cpu(cpu) {
6388 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
6390 atomic_dec(&dump_running);
6391 local_irq_restore(flags);
6393 EXPORT_SYMBOL_GPL(ftrace_dump);
6395 __init static int tracer_alloc_buffers(void)
6401 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
6404 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
6405 goto out_free_buffer_mask;
6407 /* Only allocate trace_printk buffers if a trace_printk exists */
6408 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
6409 /* Must be called before global_trace.buffer is allocated */
6410 trace_printk_init_buffers();
6412 /* To save memory, keep the ring buffer size to its minimum */
6413 if (ring_buffer_expanded)
6414 ring_buf_size = trace_buf_size;
6418 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
6419 cpumask_copy(tracing_cpumask, cpu_all_mask);
6421 raw_spin_lock_init(&global_trace.start_lock);
6423 /* TODO: make the number of buffers hot pluggable with CPUS */
6424 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
6425 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
6427 goto out_free_cpumask;
6430 if (global_trace.buffer_disabled)
6433 trace_init_cmdlines();
6436 * register_tracer() might reference current_trace, so it
6437 * needs to be set before we register anything. This is
6438 * just a bootstrap of current_trace anyway.
6440 global_trace.current_trace = &nop_trace;
6442 register_tracer(&nop_trace);
6444 /* All seems OK, enable tracing */
6445 tracing_disabled = 0;
6447 atomic_notifier_chain_register(&panic_notifier_list,
6448 &trace_panic_notifier);
6450 register_die_notifier(&trace_die_notifier);
6452 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
6454 /* Holder for file callbacks */
6455 global_trace.trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
6456 global_trace.trace_cpu.tr = &global_trace;
6458 INIT_LIST_HEAD(&global_trace.systems);
6459 INIT_LIST_HEAD(&global_trace.events);
6460 list_add(&global_trace.list, &ftrace_trace_arrays);
6462 while (trace_boot_options) {
6465 option = strsep(&trace_boot_options, ",");
6466 trace_set_options(&global_trace, option);
6469 register_snapshot_cmd();
6474 free_percpu(global_trace.trace_buffer.data);
6475 #ifdef CONFIG_TRACER_MAX_TRACE
6476 free_percpu(global_trace.max_buffer.data);
6478 free_cpumask_var(tracing_cpumask);
6479 out_free_buffer_mask:
6480 free_cpumask_var(tracing_buffer_mask);
6485 __init static int clear_boot_tracer(void)
6488 * The default tracer at boot buffer is an init section.
6489 * This function is called in lateinit. If we did not
6490 * find the boot tracer, then clear it out, to prevent
6491 * later registration from accessing the buffer that is
6492 * about to be freed.
6494 if (!default_bootup_tracer)
6497 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
6498 default_bootup_tracer);
6499 default_bootup_tracer = NULL;
6504 early_initcall(tracer_alloc_buffers);
6505 fs_initcall(tracer_init_debugfs);
6506 late_initcall(clear_boot_tracer);