2 * ring buffer based function tracer
4 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 Nadia Yvette Chambers
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/nmi.h>
41 #include <linux/sched/rt.h>
44 #include "trace_output.h"
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
50 bool ring_buffer_expanded;
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
59 static bool __read_mostly tracing_selftest_running;
62 * If a tracer is running, we do not want to run SELFTEST.
64 bool __read_mostly tracing_selftest_disabled;
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt[] = {
71 static struct tracer_flags dummy_tracer_flags = {
73 .opts = dummy_tracer_opt
76 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
82 * To prevent the comm cache from being overwritten when no
83 * tracing is active, only save the comm when a trace event
86 static DEFINE_PER_CPU(bool, trace_cmdline_save);
89 * Kill all tracing for good (never come back).
90 * It is initialized to 1 but will turn to zero if the initialization
91 * of the tracer is successful. But that is the only place that sets
94 static int tracing_disabled = 1;
96 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
98 cpumask_var_t __read_mostly tracing_buffer_mask;
101 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
103 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
104 * is set, then ftrace_dump is called. This will output the contents
105 * of the ftrace buffers to the console. This is very useful for
106 * capturing traces that lead to crashes and outputing it to a
109 * It is default off, but you can enable it with either specifying
110 * "ftrace_dump_on_oops" in the kernel command line, or setting
111 * /proc/sys/kernel/ftrace_dump_on_oops
112 * Set 1 if you want to dump buffers of all CPUs
113 * Set 2 if you want to dump the buffer of the CPU that triggered oops
116 enum ftrace_dump_mode ftrace_dump_on_oops;
118 /* When set, tracing will stop when a WARN*() is hit */
119 int __disable_trace_on_warning;
121 static int tracing_set_tracer(const char *buf);
123 #define MAX_TRACER_SIZE 100
124 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
125 static char *default_bootup_tracer;
127 static bool allocate_snapshot;
129 static int __init set_cmdline_ftrace(char *str)
131 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
132 default_bootup_tracer = bootup_tracer_buf;
133 /* We are using ftrace early, expand it */
134 ring_buffer_expanded = true;
137 __setup("ftrace=", set_cmdline_ftrace);
139 static int __init set_ftrace_dump_on_oops(char *str)
141 if (*str++ != '=' || !*str) {
142 ftrace_dump_on_oops = DUMP_ALL;
146 if (!strcmp("orig_cpu", str)) {
147 ftrace_dump_on_oops = DUMP_ORIG;
153 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
155 static int __init stop_trace_on_warning(char *str)
157 __disable_trace_on_warning = 1;
160 __setup("traceoff_on_warning=", stop_trace_on_warning);
162 static int __init boot_alloc_snapshot(char *str)
164 allocate_snapshot = true;
165 /* We also need the main ring buffer expanded */
166 ring_buffer_expanded = true;
169 __setup("alloc_snapshot", boot_alloc_snapshot);
172 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
173 static char *trace_boot_options __initdata;
175 static int __init set_trace_boot_options(char *str)
177 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
178 trace_boot_options = trace_boot_options_buf;
181 __setup("trace_options=", set_trace_boot_options);
184 unsigned long long ns2usecs(cycle_t nsec)
192 * The global_trace is the descriptor that holds the tracing
193 * buffers for the live tracing. For each CPU, it contains
194 * a link list of pages that will store trace entries. The
195 * page descriptor of the pages in the memory is used to hold
196 * the link list by linking the lru item in the page descriptor
197 * to each of the pages in the buffer per CPU.
199 * For each active CPU there is a data field that holds the
200 * pages for the buffer for that CPU. Each CPU has the same number
201 * of pages allocated for its buffer.
203 static struct trace_array global_trace;
205 LIST_HEAD(ftrace_trace_arrays);
207 int filter_current_check_discard(struct ring_buffer *buffer,
208 struct ftrace_event_call *call, void *rec,
209 struct ring_buffer_event *event)
211 return filter_check_discard(call, rec, buffer, event);
213 EXPORT_SYMBOL_GPL(filter_current_check_discard);
215 cycle_t ftrace_now(int cpu)
219 /* Early boot up does not have a buffer yet */
220 if (!global_trace.trace_buffer.buffer)
221 return trace_clock_local();
223 ts = ring_buffer_time_stamp(global_trace.trace_buffer.buffer, cpu);
224 ring_buffer_normalize_time_stamp(global_trace.trace_buffer.buffer, cpu, &ts);
229 int tracing_is_enabled(void)
231 return tracing_is_on();
235 * trace_buf_size is the size in bytes that is allocated
236 * for a buffer. Note, the number of bytes is always rounded
239 * This number is purposely set to a low number of 16384.
240 * If the dump on oops happens, it will be much appreciated
241 * to not have to wait for all that output. Anyway this can be
242 * boot time and run time configurable.
244 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
246 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
248 /* trace_types holds a link list of available tracers. */
249 static struct tracer *trace_types __read_mostly;
252 * trace_types_lock is used to protect the trace_types list.
254 static DEFINE_MUTEX(trace_types_lock);
257 * serialize the access of the ring buffer
259 * ring buffer serializes readers, but it is low level protection.
260 * The validity of the events (which returns by ring_buffer_peek() ..etc)
261 * are not protected by ring buffer.
263 * The content of events may become garbage if we allow other process consumes
264 * these events concurrently:
265 * A) the page of the consumed events may become a normal page
266 * (not reader page) in ring buffer, and this page will be rewrited
267 * by events producer.
268 * B) The page of the consumed events may become a page for splice_read,
269 * and this page will be returned to system.
271 * These primitives allow multi process access to different cpu ring buffer
274 * These primitives don't distinguish read-only and read-consume access.
275 * Multi read-only access are also serialized.
279 static DECLARE_RWSEM(all_cpu_access_lock);
280 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
282 static inline void trace_access_lock(int cpu)
284 if (cpu == RING_BUFFER_ALL_CPUS) {
285 /* gain it for accessing the whole ring buffer. */
286 down_write(&all_cpu_access_lock);
288 /* gain it for accessing a cpu ring buffer. */
290 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
291 down_read(&all_cpu_access_lock);
293 /* Secondly block other access to this @cpu ring buffer. */
294 mutex_lock(&per_cpu(cpu_access_lock, cpu));
298 static inline void trace_access_unlock(int cpu)
300 if (cpu == RING_BUFFER_ALL_CPUS) {
301 up_write(&all_cpu_access_lock);
303 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
304 up_read(&all_cpu_access_lock);
308 static inline void trace_access_lock_init(void)
312 for_each_possible_cpu(cpu)
313 mutex_init(&per_cpu(cpu_access_lock, cpu));
318 static DEFINE_MUTEX(access_lock);
320 static inline void trace_access_lock(int cpu)
323 mutex_lock(&access_lock);
326 static inline void trace_access_unlock(int cpu)
329 mutex_unlock(&access_lock);
332 static inline void trace_access_lock_init(void)
338 /* trace_flags holds trace_options default values */
339 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
340 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
341 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
342 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | TRACE_ITER_FUNCTION;
345 * tracing_on - enable tracing buffers
347 * This function enables tracing buffers that may have been
348 * disabled with tracing_off.
350 void tracing_on(void)
352 if (global_trace.trace_buffer.buffer)
353 ring_buffer_record_on(global_trace.trace_buffer.buffer);
355 * This flag is only looked at when buffers haven't been
356 * allocated yet. We don't really care about the race
357 * between setting this flag and actually turning
360 global_trace.buffer_disabled = 0;
362 EXPORT_SYMBOL_GPL(tracing_on);
365 * __trace_puts - write a constant string into the trace buffer.
366 * @ip: The address of the caller
367 * @str: The constant string to write
368 * @size: The size of the string.
370 int __trace_puts(unsigned long ip, const char *str, int size)
372 struct ring_buffer_event *event;
373 struct ring_buffer *buffer;
374 struct print_entry *entry;
375 unsigned long irq_flags;
378 alloc = sizeof(*entry) + size + 2; /* possible \n added */
380 local_save_flags(irq_flags);
381 buffer = global_trace.trace_buffer.buffer;
382 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
383 irq_flags, preempt_count());
387 entry = ring_buffer_event_data(event);
390 memcpy(&entry->buf, str, size);
392 /* Add a newline if necessary */
393 if (entry->buf[size - 1] != '\n') {
394 entry->buf[size] = '\n';
395 entry->buf[size + 1] = '\0';
397 entry->buf[size] = '\0';
399 __buffer_unlock_commit(buffer, event);
403 EXPORT_SYMBOL_GPL(__trace_puts);
406 * __trace_bputs - write the pointer to a constant string into trace buffer
407 * @ip: The address of the caller
408 * @str: The constant string to write to the buffer to
410 int __trace_bputs(unsigned long ip, const char *str)
412 struct ring_buffer_event *event;
413 struct ring_buffer *buffer;
414 struct bputs_entry *entry;
415 unsigned long irq_flags;
416 int size = sizeof(struct bputs_entry);
418 local_save_flags(irq_flags);
419 buffer = global_trace.trace_buffer.buffer;
420 event = trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
421 irq_flags, preempt_count());
425 entry = ring_buffer_event_data(event);
429 __buffer_unlock_commit(buffer, event);
433 EXPORT_SYMBOL_GPL(__trace_bputs);
435 #ifdef CONFIG_TRACER_SNAPSHOT
437 * trace_snapshot - take a snapshot of the current buffer.
439 * This causes a swap between the snapshot buffer and the current live
440 * tracing buffer. You can use this to take snapshots of the live
441 * trace when some condition is triggered, but continue to trace.
443 * Note, make sure to allocate the snapshot with either
444 * a tracing_snapshot_alloc(), or by doing it manually
445 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
447 * If the snapshot buffer is not allocated, it will stop tracing.
448 * Basically making a permanent snapshot.
450 void tracing_snapshot(void)
452 struct trace_array *tr = &global_trace;
453 struct tracer *tracer = tr->current_trace;
457 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
458 internal_trace_puts("*** snapshot is being ignored ***\n");
462 if (!tr->allocated_snapshot) {
463 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
464 internal_trace_puts("*** stopping trace here! ***\n");
469 /* Note, snapshot can not be used when the tracer uses it */
470 if (tracer->use_max_tr) {
471 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
472 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
476 local_irq_save(flags);
477 update_max_tr(tr, current, smp_processor_id());
478 local_irq_restore(flags);
480 EXPORT_SYMBOL_GPL(tracing_snapshot);
482 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
483 struct trace_buffer *size_buf, int cpu_id);
484 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val);
486 static int alloc_snapshot(struct trace_array *tr)
490 if (!tr->allocated_snapshot) {
492 /* allocate spare buffer */
493 ret = resize_buffer_duplicate_size(&tr->max_buffer,
494 &tr->trace_buffer, RING_BUFFER_ALL_CPUS);
498 tr->allocated_snapshot = true;
504 void free_snapshot(struct trace_array *tr)
507 * We don't free the ring buffer. instead, resize it because
508 * The max_tr ring buffer has some state (e.g. ring->clock) and
509 * we want preserve it.
511 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
512 set_buffer_entries(&tr->max_buffer, 1);
513 tracing_reset_online_cpus(&tr->max_buffer);
514 tr->allocated_snapshot = false;
518 * trace_snapshot_alloc - allocate and take a snapshot of the current buffer.
520 * This is similar to trace_snapshot(), but it will allocate the
521 * snapshot buffer if it isn't already allocated. Use this only
522 * where it is safe to sleep, as the allocation may sleep.
524 * This causes a swap between the snapshot buffer and the current live
525 * tracing buffer. You can use this to take snapshots of the live
526 * trace when some condition is triggered, but continue to trace.
528 void tracing_snapshot_alloc(void)
530 struct trace_array *tr = &global_trace;
533 ret = alloc_snapshot(tr);
534 if (WARN_ON(ret < 0))
539 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
541 void tracing_snapshot(void)
543 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
545 EXPORT_SYMBOL_GPL(tracing_snapshot);
546 void tracing_snapshot_alloc(void)
551 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
552 #endif /* CONFIG_TRACER_SNAPSHOT */
555 * tracing_off - turn off tracing buffers
557 * This function stops the tracing buffers from recording data.
558 * It does not disable any overhead the tracers themselves may
559 * be causing. This function simply causes all recording to
560 * the ring buffers to fail.
562 void tracing_off(void)
564 if (global_trace.trace_buffer.buffer)
565 ring_buffer_record_off(global_trace.trace_buffer.buffer);
567 * This flag is only looked at when buffers haven't been
568 * allocated yet. We don't really care about the race
569 * between setting this flag and actually turning
572 global_trace.buffer_disabled = 1;
574 EXPORT_SYMBOL_GPL(tracing_off);
576 void disable_trace_on_warning(void)
578 if (__disable_trace_on_warning)
583 * tracing_is_on - show state of ring buffers enabled
585 int tracing_is_on(void)
587 if (global_trace.trace_buffer.buffer)
588 return ring_buffer_record_is_on(global_trace.trace_buffer.buffer);
589 return !global_trace.buffer_disabled;
591 EXPORT_SYMBOL_GPL(tracing_is_on);
593 static int __init set_buf_size(char *str)
595 unsigned long buf_size;
599 buf_size = memparse(str, &str);
600 /* nr_entries can not be zero */
603 trace_buf_size = buf_size;
606 __setup("trace_buf_size=", set_buf_size);
608 static int __init set_tracing_thresh(char *str)
610 unsigned long threshold;
615 ret = kstrtoul(str, 0, &threshold);
618 tracing_thresh = threshold * 1000;
621 __setup("tracing_thresh=", set_tracing_thresh);
623 unsigned long nsecs_to_usecs(unsigned long nsecs)
628 /* These must match the bit postions in trace_iterator_flags */
629 static const char *trace_options[] = {
662 int in_ns; /* is this clock in nanoseconds? */
664 { trace_clock_local, "local", 1 },
665 { trace_clock_global, "global", 1 },
666 { trace_clock_counter, "counter", 0 },
667 { trace_clock_jiffies, "uptime", 1 },
668 { trace_clock, "perf", 1 },
675 * trace_parser_get_init - gets the buffer for trace parser
677 int trace_parser_get_init(struct trace_parser *parser, int size)
679 memset(parser, 0, sizeof(*parser));
681 parser->buffer = kmalloc(size, GFP_KERNEL);
690 * trace_parser_put - frees the buffer for trace parser
692 void trace_parser_put(struct trace_parser *parser)
694 kfree(parser->buffer);
698 * trace_get_user - reads the user input string separated by space
699 * (matched by isspace(ch))
701 * For each string found the 'struct trace_parser' is updated,
702 * and the function returns.
704 * Returns number of bytes read.
706 * See kernel/trace/trace.h for 'struct trace_parser' details.
708 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
709 size_t cnt, loff_t *ppos)
716 trace_parser_clear(parser);
718 ret = get_user(ch, ubuf++);
726 * The parser is not finished with the last write,
727 * continue reading the user input without skipping spaces.
730 /* skip white space */
731 while (cnt && isspace(ch)) {
732 ret = get_user(ch, ubuf++);
739 /* only spaces were written */
749 /* read the non-space input */
750 while (cnt && !isspace(ch)) {
751 if (parser->idx < parser->size - 1)
752 parser->buffer[parser->idx++] = ch;
757 ret = get_user(ch, ubuf++);
764 /* We either got finished input or we have to wait for another call. */
766 parser->buffer[parser->idx] = 0;
767 parser->cont = false;
770 parser->buffer[parser->idx++] = ch;
780 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
788 if (s->len <= s->readpos)
791 len = s->len - s->readpos;
794 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
804 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
808 if (s->len <= s->readpos)
811 len = s->len - s->readpos;
814 memcpy(buf, s->buffer + s->readpos, cnt);
821 * ftrace_max_lock is used to protect the swapping of buffers
822 * when taking a max snapshot. The buffers themselves are
823 * protected by per_cpu spinlocks. But the action of the swap
824 * needs its own lock.
826 * This is defined as a arch_spinlock_t in order to help
827 * with performance when lockdep debugging is enabled.
829 * It is also used in other places outside the update_max_tr
830 * so it needs to be defined outside of the
831 * CONFIG_TRACER_MAX_TRACE.
833 static arch_spinlock_t ftrace_max_lock =
834 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
836 unsigned long __read_mostly tracing_thresh;
838 #ifdef CONFIG_TRACER_MAX_TRACE
839 unsigned long __read_mostly tracing_max_latency;
842 * Copy the new maximum trace into the separate maximum-trace
843 * structure. (this way the maximum trace is permanently saved,
844 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
847 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
849 struct trace_buffer *trace_buf = &tr->trace_buffer;
850 struct trace_buffer *max_buf = &tr->max_buffer;
851 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
852 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
855 max_buf->time_start = data->preempt_timestamp;
857 max_data->saved_latency = tracing_max_latency;
858 max_data->critical_start = data->critical_start;
859 max_data->critical_end = data->critical_end;
861 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
862 max_data->pid = tsk->pid;
864 * If tsk == current, then use current_uid(), as that does not use
865 * RCU. The irq tracer can be called out of RCU scope.
868 max_data->uid = current_uid();
870 max_data->uid = task_uid(tsk);
872 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
873 max_data->policy = tsk->policy;
874 max_data->rt_priority = tsk->rt_priority;
876 /* record this tasks comm */
877 tracing_record_cmdline(tsk);
881 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
883 * @tsk: the task with the latency
884 * @cpu: The cpu that initiated the trace.
886 * Flip the buffers between the @tr and the max_tr and record information
887 * about which task was the cause of this latency.
890 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
892 struct ring_buffer *buf;
897 WARN_ON_ONCE(!irqs_disabled());
899 if (!tr->allocated_snapshot) {
900 /* Only the nop tracer should hit this when disabling */
901 WARN_ON_ONCE(tr->current_trace != &nop_trace);
905 arch_spin_lock(&ftrace_max_lock);
907 buf = tr->trace_buffer.buffer;
908 tr->trace_buffer.buffer = tr->max_buffer.buffer;
909 tr->max_buffer.buffer = buf;
911 __update_max_tr(tr, tsk, cpu);
912 arch_spin_unlock(&ftrace_max_lock);
916 * update_max_tr_single - only copy one trace over, and reset the rest
918 * @tsk - task with the latency
919 * @cpu - the cpu of the buffer to copy.
921 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
924 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
931 WARN_ON_ONCE(!irqs_disabled());
932 if (!tr->allocated_snapshot) {
933 /* Only the nop tracer should hit this when disabling */
934 WARN_ON_ONCE(tr->current_trace != &nop_trace);
938 arch_spin_lock(&ftrace_max_lock);
940 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
944 * We failed to swap the buffer due to a commit taking
945 * place on this CPU. We fail to record, but we reset
946 * the max trace buffer (no one writes directly to it)
947 * and flag that it failed.
949 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
950 "Failed to swap buffers due to commit in progress\n");
953 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
955 __update_max_tr(tr, tsk, cpu);
956 arch_spin_unlock(&ftrace_max_lock);
958 #endif /* CONFIG_TRACER_MAX_TRACE */
960 static void default_wait_pipe(struct trace_iterator *iter)
962 /* Iterators are static, they should be filled or empty */
963 if (trace_buffer_iter(iter, iter->cpu_file))
966 ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file);
969 #ifdef CONFIG_FTRACE_STARTUP_TEST
970 static int run_tracer_selftest(struct tracer *type)
972 struct trace_array *tr = &global_trace;
973 struct tracer *saved_tracer = tr->current_trace;
976 if (!type->selftest || tracing_selftest_disabled)
980 * Run a selftest on this tracer.
981 * Here we reset the trace buffer, and set the current
982 * tracer to be this tracer. The tracer can then run some
983 * internal tracing to verify that everything is in order.
984 * If we fail, we do not register this tracer.
986 tracing_reset_online_cpus(&tr->trace_buffer);
988 tr->current_trace = type;
990 #ifdef CONFIG_TRACER_MAX_TRACE
991 if (type->use_max_tr) {
992 /* If we expanded the buffers, make sure the max is expanded too */
993 if (ring_buffer_expanded)
994 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
995 RING_BUFFER_ALL_CPUS);
996 tr->allocated_snapshot = true;
1000 /* the test is responsible for initializing and enabling */
1001 pr_info("Testing tracer %s: ", type->name);
1002 ret = type->selftest(type, tr);
1003 /* the test is responsible for resetting too */
1004 tr->current_trace = saved_tracer;
1006 printk(KERN_CONT "FAILED!\n");
1007 /* Add the warning after printing 'FAILED' */
1011 /* Only reset on passing, to avoid touching corrupted buffers */
1012 tracing_reset_online_cpus(&tr->trace_buffer);
1014 #ifdef CONFIG_TRACER_MAX_TRACE
1015 if (type->use_max_tr) {
1016 tr->allocated_snapshot = false;
1018 /* Shrink the max buffer again */
1019 if (ring_buffer_expanded)
1020 ring_buffer_resize(tr->max_buffer.buffer, 1,
1021 RING_BUFFER_ALL_CPUS);
1025 printk(KERN_CONT "PASSED\n");
1029 static inline int run_tracer_selftest(struct tracer *type)
1033 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1036 * register_tracer - register a tracer with the ftrace system.
1037 * @type - the plugin for the tracer
1039 * Register a new plugin tracer.
1041 int register_tracer(struct tracer *type)
1047 pr_info("Tracer must have a name\n");
1051 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1052 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1056 mutex_lock(&trace_types_lock);
1058 tracing_selftest_running = true;
1060 for (t = trace_types; t; t = t->next) {
1061 if (strcmp(type->name, t->name) == 0) {
1063 pr_info("Tracer %s already registered\n",
1070 if (!type->set_flag)
1071 type->set_flag = &dummy_set_flag;
1073 type->flags = &dummy_tracer_flags;
1075 if (!type->flags->opts)
1076 type->flags->opts = dummy_tracer_opt;
1077 if (!type->wait_pipe)
1078 type->wait_pipe = default_wait_pipe;
1080 ret = run_tracer_selftest(type);
1084 type->next = trace_types;
1088 tracing_selftest_running = false;
1089 mutex_unlock(&trace_types_lock);
1091 if (ret || !default_bootup_tracer)
1094 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1097 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1098 /* Do we want this tracer to start on bootup? */
1099 tracing_set_tracer(type->name);
1100 default_bootup_tracer = NULL;
1101 /* disable other selftests, since this will break it. */
1102 tracing_selftest_disabled = true;
1103 #ifdef CONFIG_FTRACE_STARTUP_TEST
1104 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1112 void tracing_reset(struct trace_buffer *buf, int cpu)
1114 struct ring_buffer *buffer = buf->buffer;
1119 ring_buffer_record_disable(buffer);
1121 /* Make sure all commits have finished */
1122 synchronize_sched();
1123 ring_buffer_reset_cpu(buffer, cpu);
1125 ring_buffer_record_enable(buffer);
1128 void tracing_reset_online_cpus(struct trace_buffer *buf)
1130 struct ring_buffer *buffer = buf->buffer;
1136 ring_buffer_record_disable(buffer);
1138 /* Make sure all commits have finished */
1139 synchronize_sched();
1141 buf->time_start = ftrace_now(buf->cpu);
1143 for_each_online_cpu(cpu)
1144 ring_buffer_reset_cpu(buffer, cpu);
1146 ring_buffer_record_enable(buffer);
1149 void tracing_reset_current(int cpu)
1151 tracing_reset(&global_trace.trace_buffer, cpu);
1154 void tracing_reset_all_online_cpus(void)
1156 struct trace_array *tr;
1158 mutex_lock(&trace_types_lock);
1159 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1160 tracing_reset_online_cpus(&tr->trace_buffer);
1161 #ifdef CONFIG_TRACER_MAX_TRACE
1162 tracing_reset_online_cpus(&tr->max_buffer);
1165 mutex_unlock(&trace_types_lock);
1168 #define SAVED_CMDLINES 128
1169 #define NO_CMDLINE_MAP UINT_MAX
1170 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1171 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
1172 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
1173 static int cmdline_idx;
1174 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1176 /* temporary disable recording */
1177 static atomic_t trace_record_cmdline_disabled __read_mostly;
1179 static void trace_init_cmdlines(void)
1181 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
1182 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
1186 int is_tracing_stopped(void)
1188 return global_trace.stop_count;
1192 * ftrace_off_permanent - disable all ftrace code permanently
1194 * This should only be called when a serious anomally has
1195 * been detected. This will turn off the function tracing,
1196 * ring buffers, and other tracing utilites. It takes no
1197 * locks and can be called from any context.
1199 void ftrace_off_permanent(void)
1201 tracing_disabled = 1;
1203 tracing_off_permanent();
1207 * tracing_start - quick start of the tracer
1209 * If tracing is enabled but was stopped by tracing_stop,
1210 * this will start the tracer back up.
1212 void tracing_start(void)
1214 struct ring_buffer *buffer;
1215 unsigned long flags;
1217 if (tracing_disabled)
1220 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1221 if (--global_trace.stop_count) {
1222 if (global_trace.stop_count < 0) {
1223 /* Someone screwed up their debugging */
1225 global_trace.stop_count = 0;
1230 /* Prevent the buffers from switching */
1231 arch_spin_lock(&ftrace_max_lock);
1233 buffer = global_trace.trace_buffer.buffer;
1235 ring_buffer_record_enable(buffer);
1237 #ifdef CONFIG_TRACER_MAX_TRACE
1238 buffer = global_trace.max_buffer.buffer;
1240 ring_buffer_record_enable(buffer);
1243 arch_spin_unlock(&ftrace_max_lock);
1247 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1250 static void tracing_start_tr(struct trace_array *tr)
1252 struct ring_buffer *buffer;
1253 unsigned long flags;
1255 if (tracing_disabled)
1258 /* If global, we need to also start the max tracer */
1259 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1260 return tracing_start();
1262 raw_spin_lock_irqsave(&tr->start_lock, flags);
1264 if (--tr->stop_count) {
1265 if (tr->stop_count < 0) {
1266 /* Someone screwed up their debugging */
1273 buffer = tr->trace_buffer.buffer;
1275 ring_buffer_record_enable(buffer);
1278 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1282 * tracing_stop - quick stop of the tracer
1284 * Light weight way to stop tracing. Use in conjunction with
1287 void tracing_stop(void)
1289 struct ring_buffer *buffer;
1290 unsigned long flags;
1293 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1294 if (global_trace.stop_count++)
1297 /* Prevent the buffers from switching */
1298 arch_spin_lock(&ftrace_max_lock);
1300 buffer = global_trace.trace_buffer.buffer;
1302 ring_buffer_record_disable(buffer);
1304 #ifdef CONFIG_TRACER_MAX_TRACE
1305 buffer = global_trace.max_buffer.buffer;
1307 ring_buffer_record_disable(buffer);
1310 arch_spin_unlock(&ftrace_max_lock);
1313 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1316 static void tracing_stop_tr(struct trace_array *tr)
1318 struct ring_buffer *buffer;
1319 unsigned long flags;
1321 /* If global, we need to also stop the max tracer */
1322 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1323 return tracing_stop();
1325 raw_spin_lock_irqsave(&tr->start_lock, flags);
1326 if (tr->stop_count++)
1329 buffer = tr->trace_buffer.buffer;
1331 ring_buffer_record_disable(buffer);
1334 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1337 void trace_stop_cmdline_recording(void);
1339 static void trace_save_cmdline(struct task_struct *tsk)
1343 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1347 * It's not the end of the world if we don't get
1348 * the lock, but we also don't want to spin
1349 * nor do we want to disable interrupts,
1350 * so if we miss here, then better luck next time.
1352 if (!arch_spin_trylock(&trace_cmdline_lock))
1355 idx = map_pid_to_cmdline[tsk->pid];
1356 if (idx == NO_CMDLINE_MAP) {
1357 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1360 * Check whether the cmdline buffer at idx has a pid
1361 * mapped. We are going to overwrite that entry so we
1362 * need to clear the map_pid_to_cmdline. Otherwise we
1363 * would read the new comm for the old pid.
1365 pid = map_cmdline_to_pid[idx];
1366 if (pid != NO_CMDLINE_MAP)
1367 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1369 map_cmdline_to_pid[idx] = tsk->pid;
1370 map_pid_to_cmdline[tsk->pid] = idx;
1375 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1377 arch_spin_unlock(&trace_cmdline_lock);
1380 void trace_find_cmdline(int pid, char comm[])
1385 strcpy(comm, "<idle>");
1389 if (WARN_ON_ONCE(pid < 0)) {
1390 strcpy(comm, "<XXX>");
1394 if (pid > PID_MAX_DEFAULT) {
1395 strcpy(comm, "<...>");
1400 arch_spin_lock(&trace_cmdline_lock);
1401 map = map_pid_to_cmdline[pid];
1402 if (map != NO_CMDLINE_MAP)
1403 strcpy(comm, saved_cmdlines[map]);
1405 strcpy(comm, "<...>");
1407 arch_spin_unlock(&trace_cmdline_lock);
1411 void tracing_record_cmdline(struct task_struct *tsk)
1413 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1416 if (!__this_cpu_read(trace_cmdline_save))
1419 __this_cpu_write(trace_cmdline_save, false);
1421 trace_save_cmdline(tsk);
1425 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1428 struct task_struct *tsk = current;
1430 entry->preempt_count = pc & 0xff;
1431 entry->pid = (tsk) ? tsk->pid : 0;
1433 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1434 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1436 TRACE_FLAG_IRQS_NOSUPPORT |
1438 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1439 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1440 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1442 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1444 struct ring_buffer_event *
1445 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1448 unsigned long flags, int pc)
1450 struct ring_buffer_event *event;
1452 event = ring_buffer_lock_reserve(buffer, len);
1453 if (event != NULL) {
1454 struct trace_entry *ent = ring_buffer_event_data(event);
1456 tracing_generic_entry_update(ent, flags, pc);
1464 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1466 __this_cpu_write(trace_cmdline_save, true);
1467 ring_buffer_unlock_commit(buffer, event);
1471 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1472 struct ring_buffer_event *event,
1473 unsigned long flags, int pc)
1475 __buffer_unlock_commit(buffer, event);
1477 ftrace_trace_stack(buffer, flags, 6, pc);
1478 ftrace_trace_userstack(buffer, flags, pc);
1481 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1482 struct ring_buffer_event *event,
1483 unsigned long flags, int pc)
1485 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1487 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1489 struct ring_buffer_event *
1490 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1491 struct ftrace_event_file *ftrace_file,
1492 int type, unsigned long len,
1493 unsigned long flags, int pc)
1495 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1496 return trace_buffer_lock_reserve(*current_rb,
1497 type, len, flags, pc);
1499 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1501 struct ring_buffer_event *
1502 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1503 int type, unsigned long len,
1504 unsigned long flags, int pc)
1506 *current_rb = global_trace.trace_buffer.buffer;
1507 return trace_buffer_lock_reserve(*current_rb,
1508 type, len, flags, pc);
1510 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1512 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1513 struct ring_buffer_event *event,
1514 unsigned long flags, int pc)
1516 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1518 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1520 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1521 struct ring_buffer_event *event,
1522 unsigned long flags, int pc,
1523 struct pt_regs *regs)
1525 __buffer_unlock_commit(buffer, event);
1527 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1528 ftrace_trace_userstack(buffer, flags, pc);
1530 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1532 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1533 struct ring_buffer_event *event)
1535 ring_buffer_discard_commit(buffer, event);
1537 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1540 trace_function(struct trace_array *tr,
1541 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1544 struct ftrace_event_call *call = &event_function;
1545 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1546 struct ring_buffer_event *event;
1547 struct ftrace_entry *entry;
1549 /* If we are reading the ring buffer, don't trace */
1550 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1553 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1557 entry = ring_buffer_event_data(event);
1559 entry->parent_ip = parent_ip;
1561 if (!filter_check_discard(call, entry, buffer, event))
1562 __buffer_unlock_commit(buffer, event);
1566 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1567 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1570 if (likely(!atomic_read(&data->disabled)))
1571 trace_function(tr, ip, parent_ip, flags, pc);
1574 #ifdef CONFIG_STACKTRACE
1576 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1577 struct ftrace_stack {
1578 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1581 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1582 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1584 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1585 unsigned long flags,
1586 int skip, int pc, struct pt_regs *regs)
1588 struct ftrace_event_call *call = &event_kernel_stack;
1589 struct ring_buffer_event *event;
1590 struct stack_entry *entry;
1591 struct stack_trace trace;
1593 int size = FTRACE_STACK_ENTRIES;
1595 trace.nr_entries = 0;
1599 * Since events can happen in NMIs there's no safe way to
1600 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1601 * or NMI comes in, it will just have to use the default
1602 * FTRACE_STACK_SIZE.
1604 preempt_disable_notrace();
1606 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1608 * We don't need any atomic variables, just a barrier.
1609 * If an interrupt comes in, we don't care, because it would
1610 * have exited and put the counter back to what we want.
1611 * We just need a barrier to keep gcc from moving things
1615 if (use_stack == 1) {
1616 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1617 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1620 save_stack_trace_regs(regs, &trace);
1622 save_stack_trace(&trace);
1624 if (trace.nr_entries > size)
1625 size = trace.nr_entries;
1627 /* From now on, use_stack is a boolean */
1630 size *= sizeof(unsigned long);
1632 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1633 sizeof(*entry) + size, flags, pc);
1636 entry = ring_buffer_event_data(event);
1638 memset(&entry->caller, 0, size);
1641 memcpy(&entry->caller, trace.entries,
1642 trace.nr_entries * sizeof(unsigned long));
1644 trace.max_entries = FTRACE_STACK_ENTRIES;
1645 trace.entries = entry->caller;
1647 save_stack_trace_regs(regs, &trace);
1649 save_stack_trace(&trace);
1652 entry->size = trace.nr_entries;
1654 if (!filter_check_discard(call, entry, buffer, event))
1655 __buffer_unlock_commit(buffer, event);
1658 /* Again, don't let gcc optimize things here */
1660 __this_cpu_dec(ftrace_stack_reserve);
1661 preempt_enable_notrace();
1665 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1666 int skip, int pc, struct pt_regs *regs)
1668 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1671 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1674 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1677 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1680 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1683 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1686 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1690 * trace_dump_stack - record a stack back trace in the trace buffer
1691 * @skip: Number of functions to skip (helper handlers)
1693 void trace_dump_stack(int skip)
1695 unsigned long flags;
1697 if (tracing_disabled || tracing_selftest_running)
1700 local_save_flags(flags);
1703 * Skip 3 more, seems to get us at the caller of
1707 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
1708 flags, skip, preempt_count(), NULL);
1711 static DEFINE_PER_CPU(int, user_stack_count);
1714 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1716 struct ftrace_event_call *call = &event_user_stack;
1717 struct ring_buffer_event *event;
1718 struct userstack_entry *entry;
1719 struct stack_trace trace;
1721 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1725 * NMIs can not handle page faults, even with fix ups.
1726 * The save user stack can (and often does) fault.
1728 if (unlikely(in_nmi()))
1732 * prevent recursion, since the user stack tracing may
1733 * trigger other kernel events.
1736 if (__this_cpu_read(user_stack_count))
1739 __this_cpu_inc(user_stack_count);
1741 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1742 sizeof(*entry), flags, pc);
1744 goto out_drop_count;
1745 entry = ring_buffer_event_data(event);
1747 entry->tgid = current->tgid;
1748 memset(&entry->caller, 0, sizeof(entry->caller));
1750 trace.nr_entries = 0;
1751 trace.max_entries = FTRACE_STACK_ENTRIES;
1753 trace.entries = entry->caller;
1755 save_stack_trace_user(&trace);
1756 if (!filter_check_discard(call, entry, buffer, event))
1757 __buffer_unlock_commit(buffer, event);
1760 __this_cpu_dec(user_stack_count);
1766 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1768 ftrace_trace_userstack(tr, flags, preempt_count());
1772 #endif /* CONFIG_STACKTRACE */
1774 /* created for use with alloc_percpu */
1775 struct trace_buffer_struct {
1776 char buffer[TRACE_BUF_SIZE];
1779 static struct trace_buffer_struct *trace_percpu_buffer;
1780 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1781 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1782 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1785 * The buffer used is dependent on the context. There is a per cpu
1786 * buffer for normal context, softirq contex, hard irq context and
1787 * for NMI context. Thise allows for lockless recording.
1789 * Note, if the buffers failed to be allocated, then this returns NULL
1791 static char *get_trace_buf(void)
1793 struct trace_buffer_struct *percpu_buffer;
1796 * If we have allocated per cpu buffers, then we do not
1797 * need to do any locking.
1800 percpu_buffer = trace_percpu_nmi_buffer;
1802 percpu_buffer = trace_percpu_irq_buffer;
1803 else if (in_softirq())
1804 percpu_buffer = trace_percpu_sirq_buffer;
1806 percpu_buffer = trace_percpu_buffer;
1811 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1814 static int alloc_percpu_trace_buffer(void)
1816 struct trace_buffer_struct *buffers;
1817 struct trace_buffer_struct *sirq_buffers;
1818 struct trace_buffer_struct *irq_buffers;
1819 struct trace_buffer_struct *nmi_buffers;
1821 buffers = alloc_percpu(struct trace_buffer_struct);
1825 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1829 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1833 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1837 trace_percpu_buffer = buffers;
1838 trace_percpu_sirq_buffer = sirq_buffers;
1839 trace_percpu_irq_buffer = irq_buffers;
1840 trace_percpu_nmi_buffer = nmi_buffers;
1845 free_percpu(irq_buffers);
1847 free_percpu(sirq_buffers);
1849 free_percpu(buffers);
1851 WARN(1, "Could not allocate percpu trace_printk buffer");
1855 static int buffers_allocated;
1857 void trace_printk_init_buffers(void)
1859 if (buffers_allocated)
1862 if (alloc_percpu_trace_buffer())
1865 pr_info("ftrace: Allocated trace_printk buffers\n");
1867 /* Expand the buffers to set size */
1868 tracing_update_buffers();
1870 buffers_allocated = 1;
1873 * trace_printk_init_buffers() can be called by modules.
1874 * If that happens, then we need to start cmdline recording
1875 * directly here. If the global_trace.buffer is already
1876 * allocated here, then this was called by module code.
1878 if (global_trace.trace_buffer.buffer)
1879 tracing_start_cmdline_record();
1882 void trace_printk_start_comm(void)
1884 /* Start tracing comms if trace printk is set */
1885 if (!buffers_allocated)
1887 tracing_start_cmdline_record();
1890 static void trace_printk_start_stop_comm(int enabled)
1892 if (!buffers_allocated)
1896 tracing_start_cmdline_record();
1898 tracing_stop_cmdline_record();
1902 * trace_vbprintk - write binary msg to tracing buffer
1905 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1907 struct ftrace_event_call *call = &event_bprint;
1908 struct ring_buffer_event *event;
1909 struct ring_buffer *buffer;
1910 struct trace_array *tr = &global_trace;
1911 struct bprint_entry *entry;
1912 unsigned long flags;
1914 int len = 0, size, pc;
1916 if (unlikely(tracing_selftest_running || tracing_disabled))
1919 /* Don't pollute graph traces with trace_vprintk internals */
1920 pause_graph_tracing();
1922 pc = preempt_count();
1923 preempt_disable_notrace();
1925 tbuffer = get_trace_buf();
1931 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1933 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
1936 local_save_flags(flags);
1937 size = sizeof(*entry) + sizeof(u32) * len;
1938 buffer = tr->trace_buffer.buffer;
1939 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1943 entry = ring_buffer_event_data(event);
1947 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
1948 if (!filter_check_discard(call, entry, buffer, event)) {
1949 __buffer_unlock_commit(buffer, event);
1950 ftrace_trace_stack(buffer, flags, 6, pc);
1954 preempt_enable_notrace();
1955 unpause_graph_tracing();
1959 EXPORT_SYMBOL_GPL(trace_vbprintk);
1962 __trace_array_vprintk(struct ring_buffer *buffer,
1963 unsigned long ip, const char *fmt, va_list args)
1965 struct ftrace_event_call *call = &event_print;
1966 struct ring_buffer_event *event;
1967 int len = 0, size, pc;
1968 struct print_entry *entry;
1969 unsigned long flags;
1972 if (tracing_disabled || tracing_selftest_running)
1975 /* Don't pollute graph traces with trace_vprintk internals */
1976 pause_graph_tracing();
1978 pc = preempt_count();
1979 preempt_disable_notrace();
1982 tbuffer = get_trace_buf();
1988 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
1989 if (len > TRACE_BUF_SIZE)
1992 local_save_flags(flags);
1993 size = sizeof(*entry) + len + 1;
1994 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1998 entry = ring_buffer_event_data(event);
2001 memcpy(&entry->buf, tbuffer, len);
2002 entry->buf[len] = '\0';
2003 if (!filter_check_discard(call, entry, buffer, event)) {
2004 __buffer_unlock_commit(buffer, event);
2005 ftrace_trace_stack(buffer, flags, 6, pc);
2008 preempt_enable_notrace();
2009 unpause_graph_tracing();
2014 int trace_array_vprintk(struct trace_array *tr,
2015 unsigned long ip, const char *fmt, va_list args)
2017 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
2020 int trace_array_printk(struct trace_array *tr,
2021 unsigned long ip, const char *fmt, ...)
2026 if (!(trace_flags & TRACE_ITER_PRINTK))
2030 ret = trace_array_vprintk(tr, ip, fmt, ap);
2035 int trace_array_printk_buf(struct ring_buffer *buffer,
2036 unsigned long ip, const char *fmt, ...)
2041 if (!(trace_flags & TRACE_ITER_PRINTK))
2045 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2050 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2052 return trace_array_vprintk(&global_trace, ip, fmt, args);
2054 EXPORT_SYMBOL_GPL(trace_vprintk);
2056 static void trace_iterator_increment(struct trace_iterator *iter)
2058 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2062 ring_buffer_read(buf_iter, NULL);
2065 static struct trace_entry *
2066 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2067 unsigned long *lost_events)
2069 struct ring_buffer_event *event;
2070 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2073 event = ring_buffer_iter_peek(buf_iter, ts);
2075 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2079 iter->ent_size = ring_buffer_event_length(event);
2080 return ring_buffer_event_data(event);
2086 static struct trace_entry *
2087 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2088 unsigned long *missing_events, u64 *ent_ts)
2090 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2091 struct trace_entry *ent, *next = NULL;
2092 unsigned long lost_events = 0, next_lost = 0;
2093 int cpu_file = iter->cpu_file;
2094 u64 next_ts = 0, ts;
2100 * If we are in a per_cpu trace file, don't bother by iterating over
2101 * all cpu and peek directly.
2103 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2104 if (ring_buffer_empty_cpu(buffer, cpu_file))
2106 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2108 *ent_cpu = cpu_file;
2113 for_each_tracing_cpu(cpu) {
2115 if (ring_buffer_empty_cpu(buffer, cpu))
2118 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2121 * Pick the entry with the smallest timestamp:
2123 if (ent && (!next || ts < next_ts)) {
2127 next_lost = lost_events;
2128 next_size = iter->ent_size;
2132 iter->ent_size = next_size;
2135 *ent_cpu = next_cpu;
2141 *missing_events = next_lost;
2146 /* Find the next real entry, without updating the iterator itself */
2147 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2148 int *ent_cpu, u64 *ent_ts)
2150 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2153 /* Find the next real entry, and increment the iterator to the next entry */
2154 void *trace_find_next_entry_inc(struct trace_iterator *iter)
2156 iter->ent = __find_next_entry(iter, &iter->cpu,
2157 &iter->lost_events, &iter->ts);
2160 trace_iterator_increment(iter);
2162 return iter->ent ? iter : NULL;
2165 static void trace_consume(struct trace_iterator *iter)
2167 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2168 &iter->lost_events);
2171 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2173 struct trace_iterator *iter = m->private;
2177 WARN_ON_ONCE(iter->leftover);
2181 /* can't go backwards */
2186 ent = trace_find_next_entry_inc(iter);
2190 while (ent && iter->idx < i)
2191 ent = trace_find_next_entry_inc(iter);
2198 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2200 struct ring_buffer_event *event;
2201 struct ring_buffer_iter *buf_iter;
2202 unsigned long entries = 0;
2205 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2207 buf_iter = trace_buffer_iter(iter, cpu);
2211 ring_buffer_iter_reset(buf_iter);
2214 * We could have the case with the max latency tracers
2215 * that a reset never took place on a cpu. This is evident
2216 * by the timestamp being before the start of the buffer.
2218 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2219 if (ts >= iter->trace_buffer->time_start)
2222 ring_buffer_read(buf_iter, NULL);
2225 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2229 * The current tracer is copied to avoid a global locking
2232 static void *s_start(struct seq_file *m, loff_t *pos)
2234 struct trace_iterator *iter = m->private;
2235 struct trace_array *tr = iter->tr;
2236 int cpu_file = iter->cpu_file;
2242 * copy the tracer to avoid using a global lock all around.
2243 * iter->trace is a copy of current_trace, the pointer to the
2244 * name may be used instead of a strcmp(), as iter->trace->name
2245 * will point to the same string as current_trace->name.
2247 mutex_lock(&trace_types_lock);
2248 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2249 *iter->trace = *tr->current_trace;
2250 mutex_unlock(&trace_types_lock);
2252 #ifdef CONFIG_TRACER_MAX_TRACE
2253 if (iter->snapshot && iter->trace->use_max_tr)
2254 return ERR_PTR(-EBUSY);
2257 if (!iter->snapshot)
2258 atomic_inc(&trace_record_cmdline_disabled);
2260 if (*pos != iter->pos) {
2265 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2266 for_each_tracing_cpu(cpu)
2267 tracing_iter_reset(iter, cpu);
2269 tracing_iter_reset(iter, cpu_file);
2272 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2277 * If we overflowed the seq_file before, then we want
2278 * to just reuse the trace_seq buffer again.
2284 p = s_next(m, p, &l);
2288 trace_event_read_lock();
2289 trace_access_lock(cpu_file);
2293 static void s_stop(struct seq_file *m, void *p)
2295 struct trace_iterator *iter = m->private;
2297 #ifdef CONFIG_TRACER_MAX_TRACE
2298 if (iter->snapshot && iter->trace->use_max_tr)
2302 if (!iter->snapshot)
2303 atomic_dec(&trace_record_cmdline_disabled);
2305 trace_access_unlock(iter->cpu_file);
2306 trace_event_read_unlock();
2310 get_total_entries(struct trace_buffer *buf,
2311 unsigned long *total, unsigned long *entries)
2313 unsigned long count;
2319 for_each_tracing_cpu(cpu) {
2320 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2322 * If this buffer has skipped entries, then we hold all
2323 * entries for the trace and we need to ignore the
2324 * ones before the time stamp.
2326 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2327 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2328 /* total is the same as the entries */
2332 ring_buffer_overrun_cpu(buf->buffer, cpu);
2337 static void print_lat_help_header(struct seq_file *m)
2339 seq_puts(m, "# _------=> CPU# \n");
2340 seq_puts(m, "# / _-----=> irqs-off \n");
2341 seq_puts(m, "# | / _----=> need-resched \n");
2342 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2343 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2344 seq_puts(m, "# |||| / delay \n");
2345 seq_puts(m, "# cmd pid ||||| time | caller \n");
2346 seq_puts(m, "# \\ / ||||| \\ | / \n");
2349 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2351 unsigned long total;
2352 unsigned long entries;
2354 get_total_entries(buf, &total, &entries);
2355 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2356 entries, total, num_online_cpus());
2360 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2362 print_event_info(buf, m);
2363 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2364 seq_puts(m, "# | | | | |\n");
2367 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2369 print_event_info(buf, m);
2370 seq_puts(m, "# _-----=> irqs-off\n");
2371 seq_puts(m, "# / _----=> need-resched\n");
2372 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2373 seq_puts(m, "# || / _--=> preempt-depth\n");
2374 seq_puts(m, "# ||| / delay\n");
2375 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2376 seq_puts(m, "# | | | |||| | |\n");
2380 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2382 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2383 struct trace_buffer *buf = iter->trace_buffer;
2384 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2385 struct tracer *type = iter->trace;
2386 unsigned long entries;
2387 unsigned long total;
2388 const char *name = "preemption";
2392 get_total_entries(buf, &total, &entries);
2394 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2396 seq_puts(m, "# -----------------------------------"
2397 "---------------------------------\n");
2398 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2399 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2400 nsecs_to_usecs(data->saved_latency),
2404 #if defined(CONFIG_PREEMPT_NONE)
2406 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2408 #elif defined(CONFIG_PREEMPT)
2413 /* These are reserved for later use */
2416 seq_printf(m, " #P:%d)\n", num_online_cpus());
2420 seq_puts(m, "# -----------------\n");
2421 seq_printf(m, "# | task: %.16s-%d "
2422 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2423 data->comm, data->pid,
2424 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2425 data->policy, data->rt_priority);
2426 seq_puts(m, "# -----------------\n");
2428 if (data->critical_start) {
2429 seq_puts(m, "# => started at: ");
2430 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2431 trace_print_seq(m, &iter->seq);
2432 seq_puts(m, "\n# => ended at: ");
2433 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2434 trace_print_seq(m, &iter->seq);
2435 seq_puts(m, "\n#\n");
2441 static void test_cpu_buff_start(struct trace_iterator *iter)
2443 struct trace_seq *s = &iter->seq;
2445 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2448 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2451 if (cpumask_test_cpu(iter->cpu, iter->started))
2454 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2457 cpumask_set_cpu(iter->cpu, iter->started);
2459 /* Don't print started cpu buffer for the first entry of the trace */
2461 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2465 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2467 struct trace_seq *s = &iter->seq;
2468 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2469 struct trace_entry *entry;
2470 struct trace_event *event;
2474 test_cpu_buff_start(iter);
2476 event = ftrace_find_event(entry->type);
2478 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2479 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2480 if (!trace_print_lat_context(iter))
2483 if (!trace_print_context(iter))
2489 return event->funcs->trace(iter, sym_flags, event);
2491 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2494 return TRACE_TYPE_HANDLED;
2496 return TRACE_TYPE_PARTIAL_LINE;
2499 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2501 struct trace_seq *s = &iter->seq;
2502 struct trace_entry *entry;
2503 struct trace_event *event;
2507 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2508 if (!trace_seq_printf(s, "%d %d %llu ",
2509 entry->pid, iter->cpu, iter->ts))
2513 event = ftrace_find_event(entry->type);
2515 return event->funcs->raw(iter, 0, event);
2517 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2520 return TRACE_TYPE_HANDLED;
2522 return TRACE_TYPE_PARTIAL_LINE;
2525 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2527 struct trace_seq *s = &iter->seq;
2528 unsigned char newline = '\n';
2529 struct trace_entry *entry;
2530 struct trace_event *event;
2534 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2535 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2536 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2537 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2540 event = ftrace_find_event(entry->type);
2542 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2543 if (ret != TRACE_TYPE_HANDLED)
2547 SEQ_PUT_FIELD_RET(s, newline);
2549 return TRACE_TYPE_HANDLED;
2552 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2554 struct trace_seq *s = &iter->seq;
2555 struct trace_entry *entry;
2556 struct trace_event *event;
2560 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2561 SEQ_PUT_FIELD_RET(s, entry->pid);
2562 SEQ_PUT_FIELD_RET(s, iter->cpu);
2563 SEQ_PUT_FIELD_RET(s, iter->ts);
2566 event = ftrace_find_event(entry->type);
2567 return event ? event->funcs->binary(iter, 0, event) :
2571 int trace_empty(struct trace_iterator *iter)
2573 struct ring_buffer_iter *buf_iter;
2576 /* If we are looking at one CPU buffer, only check that one */
2577 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2578 cpu = iter->cpu_file;
2579 buf_iter = trace_buffer_iter(iter, cpu);
2581 if (!ring_buffer_iter_empty(buf_iter))
2584 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2590 for_each_tracing_cpu(cpu) {
2591 buf_iter = trace_buffer_iter(iter, cpu);
2593 if (!ring_buffer_iter_empty(buf_iter))
2596 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2604 /* Called with trace_event_read_lock() held. */
2605 enum print_line_t print_trace_line(struct trace_iterator *iter)
2607 enum print_line_t ret;
2609 if (iter->lost_events &&
2610 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2611 iter->cpu, iter->lost_events))
2612 return TRACE_TYPE_PARTIAL_LINE;
2614 if (iter->trace && iter->trace->print_line) {
2615 ret = iter->trace->print_line(iter);
2616 if (ret != TRACE_TYPE_UNHANDLED)
2620 if (iter->ent->type == TRACE_BPUTS &&
2621 trace_flags & TRACE_ITER_PRINTK &&
2622 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2623 return trace_print_bputs_msg_only(iter);
2625 if (iter->ent->type == TRACE_BPRINT &&
2626 trace_flags & TRACE_ITER_PRINTK &&
2627 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2628 return trace_print_bprintk_msg_only(iter);
2630 if (iter->ent->type == TRACE_PRINT &&
2631 trace_flags & TRACE_ITER_PRINTK &&
2632 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2633 return trace_print_printk_msg_only(iter);
2635 if (trace_flags & TRACE_ITER_BIN)
2636 return print_bin_fmt(iter);
2638 if (trace_flags & TRACE_ITER_HEX)
2639 return print_hex_fmt(iter);
2641 if (trace_flags & TRACE_ITER_RAW)
2642 return print_raw_fmt(iter);
2644 return print_trace_fmt(iter);
2647 void trace_latency_header(struct seq_file *m)
2649 struct trace_iterator *iter = m->private;
2651 /* print nothing if the buffers are empty */
2652 if (trace_empty(iter))
2655 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2656 print_trace_header(m, iter);
2658 if (!(trace_flags & TRACE_ITER_VERBOSE))
2659 print_lat_help_header(m);
2662 void trace_default_header(struct seq_file *m)
2664 struct trace_iterator *iter = m->private;
2666 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2669 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2670 /* print nothing if the buffers are empty */
2671 if (trace_empty(iter))
2673 print_trace_header(m, iter);
2674 if (!(trace_flags & TRACE_ITER_VERBOSE))
2675 print_lat_help_header(m);
2677 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2678 if (trace_flags & TRACE_ITER_IRQ_INFO)
2679 print_func_help_header_irq(iter->trace_buffer, m);
2681 print_func_help_header(iter->trace_buffer, m);
2686 static void test_ftrace_alive(struct seq_file *m)
2688 if (!ftrace_is_dead())
2690 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2691 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2694 #ifdef CONFIG_TRACER_MAX_TRACE
2695 static void show_snapshot_main_help(struct seq_file *m)
2697 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2698 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2699 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2700 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate)\n");
2701 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2702 seq_printf(m, "# is not a '0' or '1')\n");
2705 static void show_snapshot_percpu_help(struct seq_file *m)
2707 seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2708 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2709 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2710 seq_printf(m, "# Takes a snapshot of the main buffer for this cpu.\n");
2712 seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n");
2713 seq_printf(m, "# Must use main snapshot file to allocate.\n");
2715 seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2716 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2717 seq_printf(m, "# is not a '0' or '1')\n");
2720 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2722 if (iter->tr->allocated_snapshot)
2723 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2725 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2727 seq_printf(m, "# Snapshot commands:\n");
2728 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2729 show_snapshot_main_help(m);
2731 show_snapshot_percpu_help(m);
2734 /* Should never be called */
2735 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2738 static int s_show(struct seq_file *m, void *v)
2740 struct trace_iterator *iter = v;
2743 if (iter->ent == NULL) {
2745 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2747 test_ftrace_alive(m);
2749 if (iter->snapshot && trace_empty(iter))
2750 print_snapshot_help(m, iter);
2751 else if (iter->trace && iter->trace->print_header)
2752 iter->trace->print_header(m);
2754 trace_default_header(m);
2756 } else if (iter->leftover) {
2758 * If we filled the seq_file buffer earlier, we
2759 * want to just show it now.
2761 ret = trace_print_seq(m, &iter->seq);
2763 /* ret should this time be zero, but you never know */
2764 iter->leftover = ret;
2767 print_trace_line(iter);
2768 ret = trace_print_seq(m, &iter->seq);
2770 * If we overflow the seq_file buffer, then it will
2771 * ask us for this data again at start up.
2773 * ret is 0 if seq_file write succeeded.
2776 iter->leftover = ret;
2782 static const struct seq_operations tracer_seq_ops = {
2789 static struct trace_iterator *
2790 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
2792 struct trace_cpu *tc = inode->i_private;
2793 struct trace_array *tr = tc->tr;
2794 struct trace_iterator *iter;
2797 if (tracing_disabled)
2798 return ERR_PTR(-ENODEV);
2800 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2802 return ERR_PTR(-ENOMEM);
2804 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2806 if (!iter->buffer_iter)
2810 * We make a copy of the current tracer to avoid concurrent
2811 * changes on it while we are reading.
2813 mutex_lock(&trace_types_lock);
2814 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2818 *iter->trace = *tr->current_trace;
2820 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2825 #ifdef CONFIG_TRACER_MAX_TRACE
2826 /* Currently only the top directory has a snapshot */
2827 if (tr->current_trace->print_max || snapshot)
2828 iter->trace_buffer = &tr->max_buffer;
2831 iter->trace_buffer = &tr->trace_buffer;
2832 iter->snapshot = snapshot;
2834 mutex_init(&iter->mutex);
2835 iter->cpu_file = tc->cpu;
2837 /* Notify the tracer early; before we stop tracing. */
2838 if (iter->trace && iter->trace->open)
2839 iter->trace->open(iter);
2841 /* Annotate start of buffers if we had overruns */
2842 if (ring_buffer_overruns(iter->trace_buffer->buffer))
2843 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2845 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2846 if (trace_clocks[trace_clock_id].in_ns)
2847 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2849 /* stop the trace while dumping if we are not opening "snapshot" */
2850 if (!iter->snapshot)
2851 tracing_stop_tr(tr);
2853 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
2854 for_each_tracing_cpu(cpu) {
2855 iter->buffer_iter[cpu] =
2856 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2858 ring_buffer_read_prepare_sync();
2859 for_each_tracing_cpu(cpu) {
2860 ring_buffer_read_start(iter->buffer_iter[cpu]);
2861 tracing_iter_reset(iter, cpu);
2864 cpu = iter->cpu_file;
2865 iter->buffer_iter[cpu] =
2866 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2867 ring_buffer_read_prepare_sync();
2868 ring_buffer_read_start(iter->buffer_iter[cpu]);
2869 tracing_iter_reset(iter, cpu);
2874 mutex_unlock(&trace_types_lock);
2879 mutex_unlock(&trace_types_lock);
2881 kfree(iter->buffer_iter);
2883 seq_release_private(inode, file);
2884 return ERR_PTR(-ENOMEM);
2887 int tracing_open_generic(struct inode *inode, struct file *filp)
2889 if (tracing_disabled)
2892 filp->private_data = inode->i_private;
2896 static int tracing_release(struct inode *inode, struct file *file)
2898 struct seq_file *m = file->private_data;
2899 struct trace_iterator *iter;
2900 struct trace_array *tr;
2903 if (!(file->f_mode & FMODE_READ))
2909 mutex_lock(&trace_types_lock);
2914 for_each_tracing_cpu(cpu) {
2915 if (iter->buffer_iter[cpu])
2916 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2919 if (iter->trace && iter->trace->close)
2920 iter->trace->close(iter);
2922 if (!iter->snapshot)
2923 /* reenable tracing if it was previously enabled */
2924 tracing_start_tr(tr);
2925 mutex_unlock(&trace_types_lock);
2927 mutex_destroy(&iter->mutex);
2928 free_cpumask_var(iter->started);
2930 kfree(iter->buffer_iter);
2931 seq_release_private(inode, file);
2935 static int tracing_open(struct inode *inode, struct file *file)
2937 struct trace_iterator *iter;
2940 /* If this file was open for write, then erase contents */
2941 if ((file->f_mode & FMODE_WRITE) &&
2942 (file->f_flags & O_TRUNC)) {
2943 struct trace_cpu *tc = inode->i_private;
2944 struct trace_array *tr = tc->tr;
2946 if (tc->cpu == RING_BUFFER_ALL_CPUS)
2947 tracing_reset_online_cpus(&tr->trace_buffer);
2949 tracing_reset(&tr->trace_buffer, tc->cpu);
2952 if (file->f_mode & FMODE_READ) {
2953 iter = __tracing_open(inode, file, false);
2955 ret = PTR_ERR(iter);
2956 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2957 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2963 t_next(struct seq_file *m, void *v, loff_t *pos)
2965 struct tracer *t = v;
2975 static void *t_start(struct seq_file *m, loff_t *pos)
2980 mutex_lock(&trace_types_lock);
2981 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2987 static void t_stop(struct seq_file *m, void *p)
2989 mutex_unlock(&trace_types_lock);
2992 static int t_show(struct seq_file *m, void *v)
2994 struct tracer *t = v;
2999 seq_printf(m, "%s", t->name);
3008 static const struct seq_operations show_traces_seq_ops = {
3015 static int show_traces_open(struct inode *inode, struct file *file)
3017 if (tracing_disabled)
3020 return seq_open(file, &show_traces_seq_ops);
3024 tracing_write_stub(struct file *filp, const char __user *ubuf,
3025 size_t count, loff_t *ppos)
3030 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
3032 if (file->f_mode & FMODE_READ)
3033 return seq_lseek(file, offset, origin);
3038 static const struct file_operations tracing_fops = {
3039 .open = tracing_open,
3041 .write = tracing_write_stub,
3042 .llseek = tracing_seek,
3043 .release = tracing_release,
3046 static const struct file_operations show_traces_fops = {
3047 .open = show_traces_open,
3049 .release = seq_release,
3050 .llseek = seq_lseek,
3054 * Only trace on a CPU if the bitmask is set:
3056 static cpumask_var_t tracing_cpumask;
3059 * The tracer itself will not take this lock, but still we want
3060 * to provide a consistent cpumask to user-space:
3062 static DEFINE_MUTEX(tracing_cpumask_update_lock);
3065 * Temporary storage for the character representation of the
3066 * CPU bitmask (and one more byte for the newline):
3068 static char mask_str[NR_CPUS + 1];
3071 tracing_cpumask_read(struct file *filp, char __user *ubuf,
3072 size_t count, loff_t *ppos)
3076 mutex_lock(&tracing_cpumask_update_lock);
3078 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
3079 if (count - len < 2) {
3083 len += sprintf(mask_str + len, "\n");
3084 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3087 mutex_unlock(&tracing_cpumask_update_lock);
3093 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3094 size_t count, loff_t *ppos)
3096 struct trace_array *tr = filp->private_data;
3097 cpumask_var_t tracing_cpumask_new;
3100 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3103 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3107 mutex_lock(&tracing_cpumask_update_lock);
3109 local_irq_disable();
3110 arch_spin_lock(&ftrace_max_lock);
3111 for_each_tracing_cpu(cpu) {
3113 * Increase/decrease the disabled counter if we are
3114 * about to flip a bit in the cpumask:
3116 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
3117 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3118 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3119 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3121 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
3122 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3123 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3124 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3127 arch_spin_unlock(&ftrace_max_lock);
3130 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
3132 mutex_unlock(&tracing_cpumask_update_lock);
3133 free_cpumask_var(tracing_cpumask_new);
3138 free_cpumask_var(tracing_cpumask_new);
3143 static const struct file_operations tracing_cpumask_fops = {
3144 .open = tracing_open_generic,
3145 .read = tracing_cpumask_read,
3146 .write = tracing_cpumask_write,
3147 .llseek = generic_file_llseek,
3150 static int tracing_trace_options_show(struct seq_file *m, void *v)
3152 struct tracer_opt *trace_opts;
3153 struct trace_array *tr = m->private;
3157 mutex_lock(&trace_types_lock);
3158 tracer_flags = tr->current_trace->flags->val;
3159 trace_opts = tr->current_trace->flags->opts;
3161 for (i = 0; trace_options[i]; i++) {
3162 if (trace_flags & (1 << i))
3163 seq_printf(m, "%s\n", trace_options[i]);
3165 seq_printf(m, "no%s\n", trace_options[i]);
3168 for (i = 0; trace_opts[i].name; i++) {
3169 if (tracer_flags & trace_opts[i].bit)
3170 seq_printf(m, "%s\n", trace_opts[i].name);
3172 seq_printf(m, "no%s\n", trace_opts[i].name);
3174 mutex_unlock(&trace_types_lock);
3179 static int __set_tracer_option(struct tracer *trace,
3180 struct tracer_flags *tracer_flags,
3181 struct tracer_opt *opts, int neg)
3185 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
3190 tracer_flags->val &= ~opts->bit;
3192 tracer_flags->val |= opts->bit;
3196 /* Try to assign a tracer specific option */
3197 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
3199 struct tracer_flags *tracer_flags = trace->flags;
3200 struct tracer_opt *opts = NULL;
3203 for (i = 0; tracer_flags->opts[i].name; i++) {
3204 opts = &tracer_flags->opts[i];
3206 if (strcmp(cmp, opts->name) == 0)
3207 return __set_tracer_option(trace, trace->flags,
3214 /* Some tracers require overwrite to stay enabled */
3215 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3217 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3223 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3225 /* do nothing if flag is already set */
3226 if (!!(trace_flags & mask) == !!enabled)
3229 /* Give the tracer a chance to approve the change */
3230 if (tr->current_trace->flag_changed)
3231 if (tr->current_trace->flag_changed(tr->current_trace, mask, !!enabled))
3235 trace_flags |= mask;
3237 trace_flags &= ~mask;
3239 if (mask == TRACE_ITER_RECORD_CMD)
3240 trace_event_enable_cmd_record(enabled);
3242 if (mask == TRACE_ITER_OVERWRITE) {
3243 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3244 #ifdef CONFIG_TRACER_MAX_TRACE
3245 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3249 if (mask == TRACE_ITER_PRINTK)
3250 trace_printk_start_stop_comm(enabled);
3255 static int trace_set_options(struct trace_array *tr, char *option)
3262 cmp = strstrip(option);
3264 if (strncmp(cmp, "no", 2) == 0) {
3269 mutex_lock(&trace_types_lock);
3271 for (i = 0; trace_options[i]; i++) {
3272 if (strcmp(cmp, trace_options[i]) == 0) {
3273 ret = set_tracer_flag(tr, 1 << i, !neg);
3278 /* If no option could be set, test the specific tracer options */
3279 if (!trace_options[i])
3280 ret = set_tracer_option(tr->current_trace, cmp, neg);
3282 mutex_unlock(&trace_types_lock);
3288 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3289 size_t cnt, loff_t *ppos)
3291 struct seq_file *m = filp->private_data;
3292 struct trace_array *tr = m->private;
3296 if (cnt >= sizeof(buf))
3299 if (copy_from_user(&buf, ubuf, cnt))
3304 ret = trace_set_options(tr, buf);
3313 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3315 if (tracing_disabled)
3318 return single_open(file, tracing_trace_options_show, inode->i_private);
3321 static const struct file_operations tracing_iter_fops = {
3322 .open = tracing_trace_options_open,
3324 .llseek = seq_lseek,
3325 .release = single_release,
3326 .write = tracing_trace_options_write,
3329 static const char readme_msg[] =
3330 "tracing mini-HOWTO:\n\n"
3331 "# echo 0 > tracing_on : quick way to disable tracing\n"
3332 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3333 " Important files:\n"
3334 " trace\t\t\t- The static contents of the buffer\n"
3335 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3336 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3337 " current_tracer\t- function and latency tracers\n"
3338 " available_tracers\t- list of configured tracers for current_tracer\n"
3339 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3340 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3341 " trace_clock\t\t-change the clock used to order events\n"
3342 " local: Per cpu clock but may not be synced across CPUs\n"
3343 " global: Synced across CPUs but slows tracing down.\n"
3344 " counter: Not a clock, but just an increment\n"
3345 " uptime: Jiffy counter from time of boot\n"
3346 " perf: Same clock that perf events use\n"
3347 #ifdef CONFIG_X86_64
3348 " x86-tsc: TSC cycle counter\n"
3350 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3351 " tracing_cpumask\t- Limit which CPUs to trace\n"
3352 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3353 "\t\t\t Remove sub-buffer with rmdir\n"
3354 " trace_options\t\t- Set format or modify how tracing happens\n"
3355 "\t\t\t Disable an option by adding a suffix 'no' to the option name\n"
3356 #ifdef CONFIG_DYNAMIC_FTRACE
3357 "\n available_filter_functions - list of functions that can be filtered on\n"
3358 " set_ftrace_filter\t- echo function name in here to only trace these functions\n"
3359 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3360 " modules: Can select a group via module\n"
3361 " Format: :mod:<module-name>\n"
3362 " example: echo :mod:ext3 > set_ftrace_filter\n"
3363 " triggers: a command to perform when function is hit\n"
3364 " Format: <function>:<trigger>[:count]\n"
3365 " trigger: traceon, traceoff\n"
3366 " enable_event:<system>:<event>\n"
3367 " disable_event:<system>:<event>\n"
3368 #ifdef CONFIG_STACKTRACE
3371 #ifdef CONFIG_TRACER_SNAPSHOT
3374 " example: echo do_fault:traceoff > set_ftrace_filter\n"
3375 " echo do_trap:traceoff:3 > set_ftrace_filter\n"
3376 " The first one will disable tracing every time do_fault is hit\n"
3377 " The second will disable tracing at most 3 times when do_trap is hit\n"
3378 " The first time do trap is hit and it disables tracing, the counter\n"
3379 " will decrement to 2. If tracing is already disabled, the counter\n"
3380 " will not decrement. It only decrements when the trigger did work\n"
3381 " To remove trigger without count:\n"
3382 " echo '!<function>:<trigger> > set_ftrace_filter\n"
3383 " To remove trigger with a count:\n"
3384 " echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3385 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3386 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3387 " modules: Can select a group via module command :mod:\n"
3388 " Does not accept triggers\n"
3389 #endif /* CONFIG_DYNAMIC_FTRACE */
3390 #ifdef CONFIG_FUNCTION_TRACER
3391 " set_ftrace_pid\t- Write pid(s) to only function trace those pids (function)\n"
3393 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3394 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3395 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3397 #ifdef CONFIG_TRACER_SNAPSHOT
3398 "\n snapshot\t\t- Like 'trace' but shows the content of the static snapshot buffer\n"
3399 "\t\t\t Read the contents for more information\n"
3401 #ifdef CONFIG_STACKTRACE
3402 " stack_trace\t\t- Shows the max stack trace when active\n"
3403 " stack_max_size\t- Shows current max stack size that was traced\n"
3404 "\t\t\t Write into this file to reset the max size (trigger a new trace)\n"
3405 #ifdef CONFIG_DYNAMIC_FTRACE
3406 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace traces\n"
3408 #endif /* CONFIG_STACKTRACE */
3412 tracing_readme_read(struct file *filp, char __user *ubuf,
3413 size_t cnt, loff_t *ppos)
3415 return simple_read_from_buffer(ubuf, cnt, ppos,
3416 readme_msg, strlen(readme_msg));
3419 static const struct file_operations tracing_readme_fops = {
3420 .open = tracing_open_generic,
3421 .read = tracing_readme_read,
3422 .llseek = generic_file_llseek,
3426 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3427 size_t cnt, loff_t *ppos)
3436 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3440 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3448 for (i = 0; i < SAVED_CMDLINES; i++) {
3451 pid = map_cmdline_to_pid[i];
3452 if (pid == -1 || pid == NO_CMDLINE_MAP)
3455 trace_find_cmdline(pid, buf_comm);
3456 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3461 len = simple_read_from_buffer(ubuf, cnt, ppos,
3470 static const struct file_operations tracing_saved_cmdlines_fops = {
3471 .open = tracing_open_generic,
3472 .read = tracing_saved_cmdlines_read,
3473 .llseek = generic_file_llseek,
3477 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3478 size_t cnt, loff_t *ppos)
3480 struct trace_array *tr = filp->private_data;
3481 char buf[MAX_TRACER_SIZE+2];
3484 mutex_lock(&trace_types_lock);
3485 r = sprintf(buf, "%s\n", tr->current_trace->name);
3486 mutex_unlock(&trace_types_lock);
3488 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3491 int tracer_init(struct tracer *t, struct trace_array *tr)
3493 tracing_reset_online_cpus(&tr->trace_buffer);
3497 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3501 for_each_tracing_cpu(cpu)
3502 per_cpu_ptr(buf->data, cpu)->entries = val;
3505 #ifdef CONFIG_TRACER_MAX_TRACE
3506 /* resize @tr's buffer to the size of @size_tr's entries */
3507 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3508 struct trace_buffer *size_buf, int cpu_id)
3512 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3513 for_each_tracing_cpu(cpu) {
3514 ret = ring_buffer_resize(trace_buf->buffer,
3515 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3518 per_cpu_ptr(trace_buf->data, cpu)->entries =
3519 per_cpu_ptr(size_buf->data, cpu)->entries;
3522 ret = ring_buffer_resize(trace_buf->buffer,
3523 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3525 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3526 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3531 #endif /* CONFIG_TRACER_MAX_TRACE */
3533 static int __tracing_resize_ring_buffer(struct trace_array *tr,
3534 unsigned long size, int cpu)
3539 * If kernel or user changes the size of the ring buffer
3540 * we use the size that was given, and we can forget about
3541 * expanding it later.
3543 ring_buffer_expanded = true;
3545 /* May be called before buffers are initialized */
3546 if (!tr->trace_buffer.buffer)
3549 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3553 #ifdef CONFIG_TRACER_MAX_TRACE
3554 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3555 !tr->current_trace->use_max_tr)
3558 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3560 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3561 &tr->trace_buffer, cpu);
3564 * AARGH! We are left with different
3565 * size max buffer!!!!
3566 * The max buffer is our "snapshot" buffer.
3567 * When a tracer needs a snapshot (one of the
3568 * latency tracers), it swaps the max buffer
3569 * with the saved snap shot. We succeeded to
3570 * update the size of the main buffer, but failed to
3571 * update the size of the max buffer. But when we tried
3572 * to reset the main buffer to the original size, we
3573 * failed there too. This is very unlikely to
3574 * happen, but if it does, warn and kill all
3578 tracing_disabled = 1;
3583 if (cpu == RING_BUFFER_ALL_CPUS)
3584 set_buffer_entries(&tr->max_buffer, size);
3586 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3589 #endif /* CONFIG_TRACER_MAX_TRACE */
3591 if (cpu == RING_BUFFER_ALL_CPUS)
3592 set_buffer_entries(&tr->trace_buffer, size);
3594 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3599 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3600 unsigned long size, int cpu_id)
3604 mutex_lock(&trace_types_lock);
3606 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3607 /* make sure, this cpu is enabled in the mask */
3608 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3614 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3619 mutex_unlock(&trace_types_lock);
3626 * tracing_update_buffers - used by tracing facility to expand ring buffers
3628 * To save on memory when the tracing is never used on a system with it
3629 * configured in. The ring buffers are set to a minimum size. But once
3630 * a user starts to use the tracing facility, then they need to grow
3631 * to their default size.
3633 * This function is to be called when a tracer is about to be used.
3635 int tracing_update_buffers(void)
3639 mutex_lock(&trace_types_lock);
3640 if (!ring_buffer_expanded)
3641 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3642 RING_BUFFER_ALL_CPUS);
3643 mutex_unlock(&trace_types_lock);
3648 struct trace_option_dentry;
3650 static struct trace_option_dentry *
3651 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3654 destroy_trace_option_files(struct trace_option_dentry *topts);
3656 static int tracing_set_tracer(const char *buf)
3658 static struct trace_option_dentry *topts;
3659 struct trace_array *tr = &global_trace;
3661 #ifdef CONFIG_TRACER_MAX_TRACE
3666 mutex_lock(&trace_types_lock);
3668 if (!ring_buffer_expanded) {
3669 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3670 RING_BUFFER_ALL_CPUS);
3676 for (t = trace_types; t; t = t->next) {
3677 if (strcmp(t->name, buf) == 0)
3684 if (t == tr->current_trace)
3687 trace_branch_disable();
3689 tr->current_trace->enabled = false;
3691 if (tr->current_trace->reset)
3692 tr->current_trace->reset(tr);
3694 /* Current trace needs to be nop_trace before synchronize_sched */
3695 tr->current_trace = &nop_trace;
3697 #ifdef CONFIG_TRACER_MAX_TRACE
3698 had_max_tr = tr->allocated_snapshot;
3700 if (had_max_tr && !t->use_max_tr) {
3702 * We need to make sure that the update_max_tr sees that
3703 * current_trace changed to nop_trace to keep it from
3704 * swapping the buffers after we resize it.
3705 * The update_max_tr is called from interrupts disabled
3706 * so a synchronized_sched() is sufficient.
3708 synchronize_sched();
3712 destroy_trace_option_files(topts);
3714 topts = create_trace_option_files(tr, t);
3716 #ifdef CONFIG_TRACER_MAX_TRACE
3717 if (t->use_max_tr && !had_max_tr) {
3718 ret = alloc_snapshot(tr);
3725 ret = tracer_init(t, tr);
3730 tr->current_trace = t;
3731 tr->current_trace->enabled = true;
3732 trace_branch_enable(tr);
3734 mutex_unlock(&trace_types_lock);
3740 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3741 size_t cnt, loff_t *ppos)
3743 char buf[MAX_TRACER_SIZE+1];
3750 if (cnt > MAX_TRACER_SIZE)
3751 cnt = MAX_TRACER_SIZE;
3753 if (copy_from_user(&buf, ubuf, cnt))
3758 /* strip ending whitespace. */
3759 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3762 err = tracing_set_tracer(buf);
3772 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3773 size_t cnt, loff_t *ppos)
3775 unsigned long *ptr = filp->private_data;
3779 r = snprintf(buf, sizeof(buf), "%ld\n",
3780 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3781 if (r > sizeof(buf))
3783 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3787 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3788 size_t cnt, loff_t *ppos)
3790 unsigned long *ptr = filp->private_data;
3794 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3803 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3805 struct trace_cpu *tc = inode->i_private;
3806 struct trace_array *tr = tc->tr;
3807 struct trace_iterator *iter;
3810 if (tracing_disabled)
3813 mutex_lock(&trace_types_lock);
3815 /* create a buffer to store the information to pass to userspace */
3816 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3823 * We make a copy of the current tracer to avoid concurrent
3824 * changes on it while we are reading.
3826 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3831 *iter->trace = *tr->current_trace;
3833 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3838 /* trace pipe does not show start of buffer */
3839 cpumask_setall(iter->started);
3841 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3842 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3844 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3845 if (trace_clocks[trace_clock_id].in_ns)
3846 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3848 iter->cpu_file = tc->cpu;
3850 iter->trace_buffer = &tc->tr->trace_buffer;
3851 mutex_init(&iter->mutex);
3852 filp->private_data = iter;
3854 if (iter->trace->pipe_open)
3855 iter->trace->pipe_open(iter);
3857 nonseekable_open(inode, filp);
3859 mutex_unlock(&trace_types_lock);
3865 mutex_unlock(&trace_types_lock);
3869 static int tracing_release_pipe(struct inode *inode, struct file *file)
3871 struct trace_iterator *iter = file->private_data;
3873 mutex_lock(&trace_types_lock);
3875 if (iter->trace->pipe_close)
3876 iter->trace->pipe_close(iter);
3878 mutex_unlock(&trace_types_lock);
3880 free_cpumask_var(iter->started);
3881 mutex_destroy(&iter->mutex);
3889 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
3891 /* Iterators are static, they should be filled or empty */
3892 if (trace_buffer_iter(iter, iter->cpu_file))
3893 return POLLIN | POLLRDNORM;
3895 if (trace_flags & TRACE_ITER_BLOCK)
3897 * Always select as readable when in blocking mode
3899 return POLLIN | POLLRDNORM;
3901 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
3906 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3908 struct trace_iterator *iter = filp->private_data;
3910 return trace_poll(iter, filp, poll_table);
3914 * This is a make-shift waitqueue.
3915 * A tracer might use this callback on some rare cases:
3917 * 1) the current tracer might hold the runqueue lock when it wakes up
3918 * a reader, hence a deadlock (sched, function, and function graph tracers)
3919 * 2) the function tracers, trace all functions, we don't want
3920 * the overhead of calling wake_up and friends
3921 * (and tracing them too)
3923 * Anyway, this is really very primitive wakeup.
3925 void poll_wait_pipe(struct trace_iterator *iter)
3927 set_current_state(TASK_INTERRUPTIBLE);
3928 /* sleep for 100 msecs, and try again. */
3929 schedule_timeout(HZ / 10);
3932 /* Must be called with trace_types_lock mutex held. */
3933 static int tracing_wait_pipe(struct file *filp)
3935 struct trace_iterator *iter = filp->private_data;
3937 while (trace_empty(iter)) {
3939 if ((filp->f_flags & O_NONBLOCK)) {
3943 mutex_unlock(&iter->mutex);
3945 iter->trace->wait_pipe(iter);
3947 mutex_lock(&iter->mutex);
3949 if (signal_pending(current))
3953 * We block until we read something and tracing is disabled.
3954 * We still block if tracing is disabled, but we have never
3955 * read anything. This allows a user to cat this file, and
3956 * then enable tracing. But after we have read something,
3957 * we give an EOF when tracing is again disabled.
3959 * iter->pos will be 0 if we haven't read anything.
3961 if (!tracing_is_enabled() && iter->pos)
3972 tracing_read_pipe(struct file *filp, char __user *ubuf,
3973 size_t cnt, loff_t *ppos)
3975 struct trace_iterator *iter = filp->private_data;
3976 struct trace_array *tr = iter->tr;
3979 /* return any leftover data */
3980 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3984 trace_seq_init(&iter->seq);
3986 /* copy the tracer to avoid using a global lock all around */
3987 mutex_lock(&trace_types_lock);
3988 if (unlikely(iter->trace->name != tr->current_trace->name))
3989 *iter->trace = *tr->current_trace;
3990 mutex_unlock(&trace_types_lock);
3993 * Avoid more than one consumer on a single file descriptor
3994 * This is just a matter of traces coherency, the ring buffer itself
3997 mutex_lock(&iter->mutex);
3998 if (iter->trace->read) {
3999 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
4005 sret = tracing_wait_pipe(filp);
4009 /* stop when tracing is finished */
4010 if (trace_empty(iter)) {
4015 if (cnt >= PAGE_SIZE)
4016 cnt = PAGE_SIZE - 1;
4018 /* reset all but tr, trace, and overruns */
4019 memset(&iter->seq, 0,
4020 sizeof(struct trace_iterator) -
4021 offsetof(struct trace_iterator, seq));
4024 trace_event_read_lock();
4025 trace_access_lock(iter->cpu_file);
4026 while (trace_find_next_entry_inc(iter) != NULL) {
4027 enum print_line_t ret;
4028 int len = iter->seq.len;
4030 ret = print_trace_line(iter);
4031 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4032 /* don't print partial lines */
4033 iter->seq.len = len;
4036 if (ret != TRACE_TYPE_NO_CONSUME)
4037 trace_consume(iter);
4039 if (iter->seq.len >= cnt)
4043 * Setting the full flag means we reached the trace_seq buffer
4044 * size and we should leave by partial output condition above.
4045 * One of the trace_seq_* functions is not used properly.
4047 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4050 trace_access_unlock(iter->cpu_file);
4051 trace_event_read_unlock();
4053 /* Now copy what we have to the user */
4054 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4055 if (iter->seq.readpos >= iter->seq.len)
4056 trace_seq_init(&iter->seq);
4059 * If there was nothing to send to user, in spite of consuming trace
4060 * entries, go back to wait for more entries.
4066 mutex_unlock(&iter->mutex);
4071 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
4072 struct pipe_buffer *buf)
4074 __free_page(buf->page);
4077 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4080 __free_page(spd->pages[idx]);
4083 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4085 .map = generic_pipe_buf_map,
4086 .unmap = generic_pipe_buf_unmap,
4087 .confirm = generic_pipe_buf_confirm,
4088 .release = tracing_pipe_buf_release,
4089 .steal = generic_pipe_buf_steal,
4090 .get = generic_pipe_buf_get,
4094 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4099 /* Seq buffer is page-sized, exactly what we need. */
4101 count = iter->seq.len;
4102 ret = print_trace_line(iter);
4103 count = iter->seq.len - count;
4106 iter->seq.len -= count;
4109 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4110 iter->seq.len -= count;
4114 if (ret != TRACE_TYPE_NO_CONSUME)
4115 trace_consume(iter);
4117 if (!trace_find_next_entry_inc(iter)) {
4127 static ssize_t tracing_splice_read_pipe(struct file *filp,
4129 struct pipe_inode_info *pipe,
4133 struct page *pages_def[PIPE_DEF_BUFFERS];
4134 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4135 struct trace_iterator *iter = filp->private_data;
4136 struct splice_pipe_desc spd = {
4138 .partial = partial_def,
4139 .nr_pages = 0, /* This gets updated below. */
4140 .nr_pages_max = PIPE_DEF_BUFFERS,
4142 .ops = &tracing_pipe_buf_ops,
4143 .spd_release = tracing_spd_release_pipe,
4145 struct trace_array *tr = iter->tr;
4150 if (splice_grow_spd(pipe, &spd))
4153 /* copy the tracer to avoid using a global lock all around */
4154 mutex_lock(&trace_types_lock);
4155 if (unlikely(iter->trace->name != tr->current_trace->name))
4156 *iter->trace = *tr->current_trace;
4157 mutex_unlock(&trace_types_lock);
4159 mutex_lock(&iter->mutex);
4161 if (iter->trace->splice_read) {
4162 ret = iter->trace->splice_read(iter, filp,
4163 ppos, pipe, len, flags);
4168 ret = tracing_wait_pipe(filp);
4172 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4177 trace_event_read_lock();
4178 trace_access_lock(iter->cpu_file);
4180 /* Fill as many pages as possible. */
4181 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
4182 spd.pages[i] = alloc_page(GFP_KERNEL);
4186 rem = tracing_fill_pipe_page(rem, iter);
4188 /* Copy the data into the page, so we can start over. */
4189 ret = trace_seq_to_buffer(&iter->seq,
4190 page_address(spd.pages[i]),
4193 __free_page(spd.pages[i]);
4196 spd.partial[i].offset = 0;
4197 spd.partial[i].len = iter->seq.len;
4199 trace_seq_init(&iter->seq);
4202 trace_access_unlock(iter->cpu_file);
4203 trace_event_read_unlock();
4204 mutex_unlock(&iter->mutex);
4208 ret = splice_to_pipe(pipe, &spd);
4210 splice_shrink_spd(&spd);
4214 mutex_unlock(&iter->mutex);
4219 tracing_entries_read(struct file *filp, char __user *ubuf,
4220 size_t cnt, loff_t *ppos)
4222 struct trace_cpu *tc = filp->private_data;
4223 struct trace_array *tr = tc->tr;
4228 mutex_lock(&trace_types_lock);
4230 if (tc->cpu == RING_BUFFER_ALL_CPUS) {
4231 int cpu, buf_size_same;
4236 /* check if all cpu sizes are same */
4237 for_each_tracing_cpu(cpu) {
4238 /* fill in the size from first enabled cpu */
4240 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4241 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4247 if (buf_size_same) {
4248 if (!ring_buffer_expanded)
4249 r = sprintf(buf, "%lu (expanded: %lu)\n",
4251 trace_buf_size >> 10);
4253 r = sprintf(buf, "%lu\n", size >> 10);
4255 r = sprintf(buf, "X\n");
4257 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, tc->cpu)->entries >> 10);
4259 mutex_unlock(&trace_types_lock);
4261 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4266 tracing_entries_write(struct file *filp, const char __user *ubuf,
4267 size_t cnt, loff_t *ppos)
4269 struct trace_cpu *tc = filp->private_data;
4273 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4277 /* must have at least 1 entry */
4281 /* value is in KB */
4284 ret = tracing_resize_ring_buffer(tc->tr, val, tc->cpu);
4294 tracing_total_entries_read(struct file *filp, char __user *ubuf,
4295 size_t cnt, loff_t *ppos)
4297 struct trace_array *tr = filp->private_data;
4300 unsigned long size = 0, expanded_size = 0;
4302 mutex_lock(&trace_types_lock);
4303 for_each_tracing_cpu(cpu) {
4304 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
4305 if (!ring_buffer_expanded)
4306 expanded_size += trace_buf_size >> 10;
4308 if (ring_buffer_expanded)
4309 r = sprintf(buf, "%lu\n", size);
4311 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
4312 mutex_unlock(&trace_types_lock);
4314 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4318 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4319 size_t cnt, loff_t *ppos)
4322 * There is no need to read what the user has written, this function
4323 * is just to make sure that there is no error when "echo" is used
4332 tracing_free_buffer_release(struct inode *inode, struct file *filp)
4334 struct trace_array *tr = inode->i_private;
4336 /* disable tracing ? */
4337 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4339 /* resize the ring buffer to 0 */
4340 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4346 tracing_mark_write(struct file *filp, const char __user *ubuf,
4347 size_t cnt, loff_t *fpos)
4349 unsigned long addr = (unsigned long)ubuf;
4350 struct ring_buffer_event *event;
4351 struct ring_buffer *buffer;
4352 struct print_entry *entry;
4353 unsigned long irq_flags;
4354 struct page *pages[2];
4364 if (tracing_disabled)
4367 if (!(trace_flags & TRACE_ITER_MARKERS))
4370 if (cnt > TRACE_BUF_SIZE)
4371 cnt = TRACE_BUF_SIZE;
4374 * Userspace is injecting traces into the kernel trace buffer.
4375 * We want to be as non intrusive as possible.
4376 * To do so, we do not want to allocate any special buffers
4377 * or take any locks, but instead write the userspace data
4378 * straight into the ring buffer.
4380 * First we need to pin the userspace buffer into memory,
4381 * which, most likely it is, because it just referenced it.
4382 * But there's no guarantee that it is. By using get_user_pages_fast()
4383 * and kmap_atomic/kunmap_atomic() we can get access to the
4384 * pages directly. We then write the data directly into the
4387 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4389 /* check if we cross pages */
4390 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4393 offset = addr & (PAGE_SIZE - 1);
4396 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4397 if (ret < nr_pages) {
4399 put_page(pages[ret]);
4404 for (i = 0; i < nr_pages; i++)
4405 map_page[i] = kmap_atomic(pages[i]);
4407 local_save_flags(irq_flags);
4408 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4409 buffer = global_trace.trace_buffer.buffer;
4410 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4411 irq_flags, preempt_count());
4413 /* Ring buffer disabled, return as if not open for write */
4418 entry = ring_buffer_event_data(event);
4419 entry->ip = _THIS_IP_;
4421 if (nr_pages == 2) {
4422 len = PAGE_SIZE - offset;
4423 memcpy(&entry->buf, map_page[0] + offset, len);
4424 memcpy(&entry->buf[len], map_page[1], cnt - len);
4426 memcpy(&entry->buf, map_page[0] + offset, cnt);
4428 if (entry->buf[cnt - 1] != '\n') {
4429 entry->buf[cnt] = '\n';
4430 entry->buf[cnt + 1] = '\0';
4432 entry->buf[cnt] = '\0';
4434 __buffer_unlock_commit(buffer, event);
4441 for (i = 0; i < nr_pages; i++){
4442 kunmap_atomic(map_page[i]);
4449 static int tracing_clock_show(struct seq_file *m, void *v)
4451 struct trace_array *tr = m->private;
4454 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4456 "%s%s%s%s", i ? " " : "",
4457 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4458 i == tr->clock_id ? "]" : "");
4464 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4465 size_t cnt, loff_t *fpos)
4467 struct seq_file *m = filp->private_data;
4468 struct trace_array *tr = m->private;
4470 const char *clockstr;
4473 if (cnt >= sizeof(buf))
4476 if (copy_from_user(&buf, ubuf, cnt))
4481 clockstr = strstrip(buf);
4483 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4484 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4487 if (i == ARRAY_SIZE(trace_clocks))
4490 mutex_lock(&trace_types_lock);
4494 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4497 * New clock may not be consistent with the previous clock.
4498 * Reset the buffer so that it doesn't have incomparable timestamps.
4500 tracing_reset_online_cpus(&global_trace.trace_buffer);
4502 #ifdef CONFIG_TRACER_MAX_TRACE
4503 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4504 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4505 tracing_reset_online_cpus(&global_trace.max_buffer);
4508 mutex_unlock(&trace_types_lock);
4515 static int tracing_clock_open(struct inode *inode, struct file *file)
4517 if (tracing_disabled)
4520 return single_open(file, tracing_clock_show, inode->i_private);
4523 struct ftrace_buffer_info {
4524 struct trace_iterator iter;
4529 #ifdef CONFIG_TRACER_SNAPSHOT
4530 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4532 struct trace_cpu *tc = inode->i_private;
4533 struct trace_iterator *iter;
4537 if (file->f_mode & FMODE_READ) {
4538 iter = __tracing_open(inode, file, true);
4540 ret = PTR_ERR(iter);
4542 /* Writes still need the seq_file to hold the private data */
4543 m = kzalloc(sizeof(*m), GFP_KERNEL);
4546 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4552 iter->trace_buffer = &tc->tr->max_buffer;
4553 iter->cpu_file = tc->cpu;
4555 file->private_data = m;
4562 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4565 struct seq_file *m = filp->private_data;
4566 struct trace_iterator *iter = m->private;
4567 struct trace_array *tr = iter->tr;
4571 ret = tracing_update_buffers();
4575 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4579 mutex_lock(&trace_types_lock);
4581 if (tr->current_trace->use_max_tr) {
4588 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4592 if (tr->allocated_snapshot)
4596 /* Only allow per-cpu swap if the ring buffer supports it */
4597 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4598 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4603 if (!tr->allocated_snapshot) {
4604 ret = alloc_snapshot(tr);
4608 local_irq_disable();
4609 /* Now, we're going to swap */
4610 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4611 update_max_tr(tr, current, smp_processor_id());
4613 update_max_tr_single(tr, current, iter->cpu_file);
4617 if (tr->allocated_snapshot) {
4618 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4619 tracing_reset_online_cpus(&tr->max_buffer);
4621 tracing_reset(&tr->max_buffer, iter->cpu_file);
4631 mutex_unlock(&trace_types_lock);
4635 static int tracing_snapshot_release(struct inode *inode, struct file *file)
4637 struct seq_file *m = file->private_data;
4639 if (file->f_mode & FMODE_READ)
4640 return tracing_release(inode, file);
4642 /* If write only, the seq_file is just a stub */
4650 static int tracing_buffers_open(struct inode *inode, struct file *filp);
4651 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4652 size_t count, loff_t *ppos);
4653 static int tracing_buffers_release(struct inode *inode, struct file *file);
4654 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4655 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
4657 static int snapshot_raw_open(struct inode *inode, struct file *filp)
4659 struct ftrace_buffer_info *info;
4662 ret = tracing_buffers_open(inode, filp);
4666 info = filp->private_data;
4668 if (info->iter.trace->use_max_tr) {
4669 tracing_buffers_release(inode, filp);
4673 info->iter.snapshot = true;
4674 info->iter.trace_buffer = &info->iter.tr->max_buffer;
4679 #endif /* CONFIG_TRACER_SNAPSHOT */
4682 static const struct file_operations tracing_max_lat_fops = {
4683 .open = tracing_open_generic,
4684 .read = tracing_max_lat_read,
4685 .write = tracing_max_lat_write,
4686 .llseek = generic_file_llseek,
4689 static const struct file_operations set_tracer_fops = {
4690 .open = tracing_open_generic,
4691 .read = tracing_set_trace_read,
4692 .write = tracing_set_trace_write,
4693 .llseek = generic_file_llseek,
4696 static const struct file_operations tracing_pipe_fops = {
4697 .open = tracing_open_pipe,
4698 .poll = tracing_poll_pipe,
4699 .read = tracing_read_pipe,
4700 .splice_read = tracing_splice_read_pipe,
4701 .release = tracing_release_pipe,
4702 .llseek = no_llseek,
4705 static const struct file_operations tracing_entries_fops = {
4706 .open = tracing_open_generic,
4707 .read = tracing_entries_read,
4708 .write = tracing_entries_write,
4709 .llseek = generic_file_llseek,
4712 static const struct file_operations tracing_total_entries_fops = {
4713 .open = tracing_open_generic,
4714 .read = tracing_total_entries_read,
4715 .llseek = generic_file_llseek,
4718 static const struct file_operations tracing_free_buffer_fops = {
4719 .write = tracing_free_buffer_write,
4720 .release = tracing_free_buffer_release,
4723 static const struct file_operations tracing_mark_fops = {
4724 .open = tracing_open_generic,
4725 .write = tracing_mark_write,
4726 .llseek = generic_file_llseek,
4729 static const struct file_operations trace_clock_fops = {
4730 .open = tracing_clock_open,
4732 .llseek = seq_lseek,
4733 .release = single_release,
4734 .write = tracing_clock_write,
4737 #ifdef CONFIG_TRACER_SNAPSHOT
4738 static const struct file_operations snapshot_fops = {
4739 .open = tracing_snapshot_open,
4741 .write = tracing_snapshot_write,
4742 .llseek = tracing_seek,
4743 .release = tracing_snapshot_release,
4746 static const struct file_operations snapshot_raw_fops = {
4747 .open = snapshot_raw_open,
4748 .read = tracing_buffers_read,
4749 .release = tracing_buffers_release,
4750 .splice_read = tracing_buffers_splice_read,
4751 .llseek = no_llseek,
4754 #endif /* CONFIG_TRACER_SNAPSHOT */
4756 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4758 struct trace_cpu *tc = inode->i_private;
4759 struct trace_array *tr = tc->tr;
4760 struct ftrace_buffer_info *info;
4762 if (tracing_disabled)
4765 info = kzalloc(sizeof(*info), GFP_KERNEL);
4769 mutex_lock(&trace_types_lock);
4774 info->iter.cpu_file = tc->cpu;
4775 info->iter.trace = tr->current_trace;
4776 info->iter.trace_buffer = &tr->trace_buffer;
4778 /* Force reading ring buffer for first read */
4779 info->read = (unsigned int)-1;
4781 filp->private_data = info;
4783 mutex_unlock(&trace_types_lock);
4785 return nonseekable_open(inode, filp);
4789 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
4791 struct ftrace_buffer_info *info = filp->private_data;
4792 struct trace_iterator *iter = &info->iter;
4794 return trace_poll(iter, filp, poll_table);
4798 tracing_buffers_read(struct file *filp, char __user *ubuf,
4799 size_t count, loff_t *ppos)
4801 struct ftrace_buffer_info *info = filp->private_data;
4802 struct trace_iterator *iter = &info->iter;
4809 mutex_lock(&trace_types_lock);
4811 #ifdef CONFIG_TRACER_MAX_TRACE
4812 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
4819 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
4825 /* Do we have previous read data to read? */
4826 if (info->read < PAGE_SIZE)
4830 trace_access_lock(iter->cpu_file);
4831 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
4835 trace_access_unlock(iter->cpu_file);
4838 if (trace_empty(iter)) {
4839 if ((filp->f_flags & O_NONBLOCK)) {
4843 mutex_unlock(&trace_types_lock);
4844 iter->trace->wait_pipe(iter);
4845 mutex_lock(&trace_types_lock);
4846 if (signal_pending(current)) {
4858 size = PAGE_SIZE - info->read;
4862 ret = copy_to_user(ubuf, info->spare + info->read, size);
4873 mutex_unlock(&trace_types_lock);
4878 static int tracing_buffers_release(struct inode *inode, struct file *file)
4880 struct ftrace_buffer_info *info = file->private_data;
4881 struct trace_iterator *iter = &info->iter;
4883 mutex_lock(&trace_types_lock);
4885 WARN_ON(!iter->tr->ref);
4889 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
4892 mutex_unlock(&trace_types_lock);
4898 struct ring_buffer *buffer;
4903 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4904 struct pipe_buffer *buf)
4906 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4911 ring_buffer_free_read_page(ref->buffer, ref->page);
4916 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4917 struct pipe_buffer *buf)
4919 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4924 /* Pipe buffer operations for a buffer. */
4925 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4927 .map = generic_pipe_buf_map,
4928 .unmap = generic_pipe_buf_unmap,
4929 .confirm = generic_pipe_buf_confirm,
4930 .release = buffer_pipe_buf_release,
4931 .steal = generic_pipe_buf_steal,
4932 .get = buffer_pipe_buf_get,
4936 * Callback from splice_to_pipe(), if we need to release some pages
4937 * at the end of the spd in case we error'ed out in filling the pipe.
4939 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4941 struct buffer_ref *ref =
4942 (struct buffer_ref *)spd->partial[i].private;
4947 ring_buffer_free_read_page(ref->buffer, ref->page);
4949 spd->partial[i].private = 0;
4953 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4954 struct pipe_inode_info *pipe, size_t len,
4957 struct ftrace_buffer_info *info = file->private_data;
4958 struct trace_iterator *iter = &info->iter;
4959 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4960 struct page *pages_def[PIPE_DEF_BUFFERS];
4961 struct splice_pipe_desc spd = {
4963 .partial = partial_def,
4964 .nr_pages_max = PIPE_DEF_BUFFERS,
4966 .ops = &buffer_pipe_buf_ops,
4967 .spd_release = buffer_spd_release,
4969 struct buffer_ref *ref;
4970 int entries, size, i;
4973 mutex_lock(&trace_types_lock);
4975 #ifdef CONFIG_TRACER_MAX_TRACE
4976 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
4982 if (splice_grow_spd(pipe, &spd)) {
4987 if (*ppos & (PAGE_SIZE - 1)) {
4992 if (len & (PAGE_SIZE - 1)) {
4993 if (len < PAGE_SIZE) {
5001 trace_access_lock(iter->cpu_file);
5002 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5004 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
5008 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
5013 ref->buffer = iter->trace_buffer->buffer;
5014 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5020 r = ring_buffer_read_page(ref->buffer, &ref->page,
5021 len, iter->cpu_file, 1);
5023 ring_buffer_free_read_page(ref->buffer, ref->page);
5029 * zero out any left over data, this is going to
5032 size = ring_buffer_page_len(ref->page);
5033 if (size < PAGE_SIZE)
5034 memset(ref->page + size, 0, PAGE_SIZE - size);
5036 page = virt_to_page(ref->page);
5038 spd.pages[i] = page;
5039 spd.partial[i].len = PAGE_SIZE;
5040 spd.partial[i].offset = 0;
5041 spd.partial[i].private = (unsigned long)ref;
5045 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5048 trace_access_unlock(iter->cpu_file);
5051 /* did we read anything? */
5052 if (!spd.nr_pages) {
5053 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
5057 mutex_unlock(&trace_types_lock);
5058 iter->trace->wait_pipe(iter);
5059 mutex_lock(&trace_types_lock);
5060 if (signal_pending(current)) {
5067 ret = splice_to_pipe(pipe, &spd);
5068 splice_shrink_spd(&spd);
5070 mutex_unlock(&trace_types_lock);
5075 static const struct file_operations tracing_buffers_fops = {
5076 .open = tracing_buffers_open,
5077 .read = tracing_buffers_read,
5078 .poll = tracing_buffers_poll,
5079 .release = tracing_buffers_release,
5080 .splice_read = tracing_buffers_splice_read,
5081 .llseek = no_llseek,
5085 tracing_stats_read(struct file *filp, char __user *ubuf,
5086 size_t count, loff_t *ppos)
5088 struct trace_cpu *tc = filp->private_data;
5089 struct trace_array *tr = tc->tr;
5090 struct trace_buffer *trace_buf = &tr->trace_buffer;
5091 struct trace_seq *s;
5093 unsigned long long t;
5094 unsigned long usec_rem;
5097 s = kmalloc(sizeof(*s), GFP_KERNEL);
5103 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5104 trace_seq_printf(s, "entries: %ld\n", cnt);
5106 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5107 trace_seq_printf(s, "overrun: %ld\n", cnt);
5109 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5110 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5112 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5113 trace_seq_printf(s, "bytes: %ld\n", cnt);
5115 if (trace_clocks[trace_clock_id].in_ns) {
5116 /* local or global for trace_clock */
5117 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5118 usec_rem = do_div(t, USEC_PER_SEC);
5119 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5122 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5123 usec_rem = do_div(t, USEC_PER_SEC);
5124 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5126 /* counter or tsc mode for trace_clock */
5127 trace_seq_printf(s, "oldest event ts: %llu\n",
5128 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5130 trace_seq_printf(s, "now ts: %llu\n",
5131 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5134 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5135 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5137 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5138 trace_seq_printf(s, "read events: %ld\n", cnt);
5140 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
5147 static const struct file_operations tracing_stats_fops = {
5148 .open = tracing_open_generic,
5149 .read = tracing_stats_read,
5150 .llseek = generic_file_llseek,
5153 #ifdef CONFIG_DYNAMIC_FTRACE
5155 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5161 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5162 size_t cnt, loff_t *ppos)
5164 static char ftrace_dyn_info_buffer[1024];
5165 static DEFINE_MUTEX(dyn_info_mutex);
5166 unsigned long *p = filp->private_data;
5167 char *buf = ftrace_dyn_info_buffer;
5168 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5171 mutex_lock(&dyn_info_mutex);
5172 r = sprintf(buf, "%ld ", *p);
5174 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5177 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5179 mutex_unlock(&dyn_info_mutex);
5184 static const struct file_operations tracing_dyn_info_fops = {
5185 .open = tracing_open_generic,
5186 .read = tracing_read_dyn_info,
5187 .llseek = generic_file_llseek,
5189 #endif /* CONFIG_DYNAMIC_FTRACE */
5191 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5193 ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5199 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5201 unsigned long *count = (long *)data;
5213 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5214 struct ftrace_probe_ops *ops, void *data)
5216 long count = (long)data;
5218 seq_printf(m, "%ps:", (void *)ip);
5220 seq_printf(m, "snapshot");
5223 seq_printf(m, ":unlimited\n");
5225 seq_printf(m, ":count=%ld\n", count);
5230 static struct ftrace_probe_ops snapshot_probe_ops = {
5231 .func = ftrace_snapshot,
5232 .print = ftrace_snapshot_print,
5235 static struct ftrace_probe_ops snapshot_count_probe_ops = {
5236 .func = ftrace_count_snapshot,
5237 .print = ftrace_snapshot_print,
5241 ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
5242 char *glob, char *cmd, char *param, int enable)
5244 struct ftrace_probe_ops *ops;
5245 void *count = (void *)-1;
5249 /* hash funcs only work with set_ftrace_filter */
5253 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
5255 if (glob[0] == '!') {
5256 unregister_ftrace_function_probe_func(glob+1, ops);
5263 number = strsep(¶m, ":");
5265 if (!strlen(number))
5269 * We use the callback data field (which is a pointer)
5272 ret = kstrtoul(number, 0, (unsigned long *)&count);
5277 ret = register_ftrace_function_probe(glob, ops, count);
5280 alloc_snapshot(&global_trace);
5282 return ret < 0 ? ret : 0;
5285 static struct ftrace_func_command ftrace_snapshot_cmd = {
5287 .func = ftrace_trace_snapshot_callback,
5290 static int register_snapshot_cmd(void)
5292 return register_ftrace_command(&ftrace_snapshot_cmd);
5295 static inline int register_snapshot_cmd(void) { return 0; }
5296 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5298 struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
5303 if (!debugfs_initialized())
5306 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
5307 tr->dir = debugfs_create_dir("tracing", NULL);
5310 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5315 struct dentry *tracing_init_dentry(void)
5317 return tracing_init_dentry_tr(&global_trace);
5320 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
5322 struct dentry *d_tracer;
5325 return tr->percpu_dir;
5327 d_tracer = tracing_init_dentry_tr(tr);
5331 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
5333 WARN_ONCE(!tr->percpu_dir,
5334 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
5336 return tr->percpu_dir;
5340 tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
5342 struct trace_array_cpu *data = per_cpu_ptr(tr->trace_buffer.data, cpu);
5343 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
5344 struct dentry *d_cpu;
5345 char cpu_dir[30]; /* 30 characters should be more than enough */
5350 snprintf(cpu_dir, 30, "cpu%ld", cpu);
5351 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
5353 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
5357 /* per cpu trace_pipe */
5358 trace_create_file("trace_pipe", 0444, d_cpu,
5359 (void *)&data->trace_cpu, &tracing_pipe_fops);
5362 trace_create_file("trace", 0644, d_cpu,
5363 (void *)&data->trace_cpu, &tracing_fops);
5365 trace_create_file("trace_pipe_raw", 0444, d_cpu,
5366 (void *)&data->trace_cpu, &tracing_buffers_fops);
5368 trace_create_file("stats", 0444, d_cpu,
5369 (void *)&data->trace_cpu, &tracing_stats_fops);
5371 trace_create_file("buffer_size_kb", 0444, d_cpu,
5372 (void *)&data->trace_cpu, &tracing_entries_fops);
5374 #ifdef CONFIG_TRACER_SNAPSHOT
5375 trace_create_file("snapshot", 0644, d_cpu,
5376 (void *)&data->trace_cpu, &snapshot_fops);
5378 trace_create_file("snapshot_raw", 0444, d_cpu,
5379 (void *)&data->trace_cpu, &snapshot_raw_fops);
5383 #ifdef CONFIG_FTRACE_SELFTEST
5384 /* Let selftest have access to static functions in this file */
5385 #include "trace_selftest.c"
5388 struct trace_option_dentry {
5389 struct tracer_opt *opt;
5390 struct tracer_flags *flags;
5391 struct trace_array *tr;
5392 struct dentry *entry;
5396 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
5399 struct trace_option_dentry *topt = filp->private_data;
5402 if (topt->flags->val & topt->opt->bit)
5407 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5411 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
5414 struct trace_option_dentry *topt = filp->private_data;
5418 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5422 if (val != 0 && val != 1)
5425 if (!!(topt->flags->val & topt->opt->bit) != val) {
5426 mutex_lock(&trace_types_lock);
5427 ret = __set_tracer_option(topt->tr->current_trace, topt->flags,
5429 mutex_unlock(&trace_types_lock);
5440 static const struct file_operations trace_options_fops = {
5441 .open = tracing_open_generic,
5442 .read = trace_options_read,
5443 .write = trace_options_write,
5444 .llseek = generic_file_llseek,
5448 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5451 long index = (long)filp->private_data;
5454 if (trace_flags & (1 << index))
5459 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5463 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5466 struct trace_array *tr = &global_trace;
5467 long index = (long)filp->private_data;
5471 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5475 if (val != 0 && val != 1)
5478 mutex_lock(&trace_types_lock);
5479 ret = set_tracer_flag(tr, 1 << index, val);
5480 mutex_unlock(&trace_types_lock);
5490 static const struct file_operations trace_options_core_fops = {
5491 .open = tracing_open_generic,
5492 .read = trace_options_core_read,
5493 .write = trace_options_core_write,
5494 .llseek = generic_file_llseek,
5497 struct dentry *trace_create_file(const char *name,
5499 struct dentry *parent,
5501 const struct file_operations *fops)
5505 ret = debugfs_create_file(name, mode, parent, data, fops);
5507 pr_warning("Could not create debugfs '%s' entry\n", name);
5513 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5515 struct dentry *d_tracer;
5520 d_tracer = tracing_init_dentry_tr(tr);
5524 tr->options = debugfs_create_dir("options", d_tracer);
5526 pr_warning("Could not create debugfs directory 'options'\n");
5534 create_trace_option_file(struct trace_array *tr,
5535 struct trace_option_dentry *topt,
5536 struct tracer_flags *flags,
5537 struct tracer_opt *opt)
5539 struct dentry *t_options;
5541 t_options = trace_options_init_dentry(tr);
5545 topt->flags = flags;
5549 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5550 &trace_options_fops);
5554 static struct trace_option_dentry *
5555 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5557 struct trace_option_dentry *topts;
5558 struct tracer_flags *flags;
5559 struct tracer_opt *opts;
5565 flags = tracer->flags;
5567 if (!flags || !flags->opts)
5572 for (cnt = 0; opts[cnt].name; cnt++)
5575 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5579 for (cnt = 0; opts[cnt].name; cnt++)
5580 create_trace_option_file(tr, &topts[cnt], flags,
5587 destroy_trace_option_files(struct trace_option_dentry *topts)
5594 for (cnt = 0; topts[cnt].opt; cnt++) {
5595 if (topts[cnt].entry)
5596 debugfs_remove(topts[cnt].entry);
5602 static struct dentry *
5603 create_trace_option_core_file(struct trace_array *tr,
5604 const char *option, long index)
5606 struct dentry *t_options;
5608 t_options = trace_options_init_dentry(tr);
5612 return trace_create_file(option, 0644, t_options, (void *)index,
5613 &trace_options_core_fops);
5616 static __init void create_trace_options_dir(struct trace_array *tr)
5618 struct dentry *t_options;
5621 t_options = trace_options_init_dentry(tr);
5625 for (i = 0; trace_options[i]; i++)
5626 create_trace_option_core_file(tr, trace_options[i], i);
5630 rb_simple_read(struct file *filp, char __user *ubuf,
5631 size_t cnt, loff_t *ppos)
5633 struct trace_array *tr = filp->private_data;
5634 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5639 r = ring_buffer_record_is_on(buffer);
5643 r = sprintf(buf, "%d\n", r);
5645 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5649 rb_simple_write(struct file *filp, const char __user *ubuf,
5650 size_t cnt, loff_t *ppos)
5652 struct trace_array *tr = filp->private_data;
5653 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5657 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5662 mutex_lock(&trace_types_lock);
5664 ring_buffer_record_on(buffer);
5665 if (tr->current_trace->start)
5666 tr->current_trace->start(tr);
5668 ring_buffer_record_off(buffer);
5669 if (tr->current_trace->stop)
5670 tr->current_trace->stop(tr);
5672 mutex_unlock(&trace_types_lock);
5680 static const struct file_operations rb_simple_fops = {
5681 .open = tracing_open_generic,
5682 .read = rb_simple_read,
5683 .write = rb_simple_write,
5684 .llseek = default_llseek,
5687 struct dentry *trace_instance_dir;
5690 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
5692 static void init_trace_buffers(struct trace_array *tr, struct trace_buffer *buf)
5696 for_each_tracing_cpu(cpu) {
5697 memset(per_cpu_ptr(buf->data, cpu), 0, sizeof(struct trace_array_cpu));
5698 per_cpu_ptr(buf->data, cpu)->trace_cpu.cpu = cpu;
5699 per_cpu_ptr(buf->data, cpu)->trace_cpu.tr = tr;
5704 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
5706 enum ring_buffer_flags rb_flags;
5708 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5710 buf->buffer = ring_buffer_alloc(size, rb_flags);
5714 buf->data = alloc_percpu(struct trace_array_cpu);
5716 ring_buffer_free(buf->buffer);
5720 init_trace_buffers(tr, buf);
5722 /* Allocate the first page for all buffers */
5723 set_buffer_entries(&tr->trace_buffer,
5724 ring_buffer_size(tr->trace_buffer.buffer, 0));
5729 static int allocate_trace_buffers(struct trace_array *tr, int size)
5733 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
5737 #ifdef CONFIG_TRACER_MAX_TRACE
5738 ret = allocate_trace_buffer(tr, &tr->max_buffer,
5739 allocate_snapshot ? size : 1);
5741 ring_buffer_free(tr->trace_buffer.buffer);
5742 free_percpu(tr->trace_buffer.data);
5745 tr->allocated_snapshot = allocate_snapshot;
5748 * Only the top level trace array gets its snapshot allocated
5749 * from the kernel command line.
5751 allocate_snapshot = false;
5756 static int new_instance_create(const char *name)
5758 struct trace_array *tr;
5761 mutex_lock(&trace_types_lock);
5764 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5765 if (tr->name && strcmp(tr->name, name) == 0)
5770 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
5774 tr->name = kstrdup(name, GFP_KERNEL);
5778 raw_spin_lock_init(&tr->start_lock);
5780 tr->current_trace = &nop_trace;
5782 INIT_LIST_HEAD(&tr->systems);
5783 INIT_LIST_HEAD(&tr->events);
5785 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
5788 /* Holder for file callbacks */
5789 tr->trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
5790 tr->trace_cpu.tr = tr;
5792 tr->dir = debugfs_create_dir(name, trace_instance_dir);
5796 ret = event_trace_add_tracer(tr->dir, tr);
5800 init_tracer_debugfs(tr, tr->dir);
5802 list_add(&tr->list, &ftrace_trace_arrays);
5804 mutex_unlock(&trace_types_lock);
5809 if (tr->trace_buffer.buffer)
5810 ring_buffer_free(tr->trace_buffer.buffer);
5815 mutex_unlock(&trace_types_lock);
5821 static int instance_delete(const char *name)
5823 struct trace_array *tr;
5827 mutex_lock(&trace_types_lock);
5830 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5831 if (tr->name && strcmp(tr->name, name) == 0) {
5843 list_del(&tr->list);
5845 event_trace_del_tracer(tr);
5846 debugfs_remove_recursive(tr->dir);
5847 free_percpu(tr->trace_buffer.data);
5848 ring_buffer_free(tr->trace_buffer.buffer);
5856 mutex_unlock(&trace_types_lock);
5861 static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
5863 struct dentry *parent;
5866 /* Paranoid: Make sure the parent is the "instances" directory */
5867 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
5868 if (WARN_ON_ONCE(parent != trace_instance_dir))
5872 * The inode mutex is locked, but debugfs_create_dir() will also
5873 * take the mutex. As the instances directory can not be destroyed
5874 * or changed in any other way, it is safe to unlock it, and
5875 * let the dentry try. If two users try to make the same dir at
5876 * the same time, then the new_instance_create() will determine the
5879 mutex_unlock(&inode->i_mutex);
5881 ret = new_instance_create(dentry->d_iname);
5883 mutex_lock(&inode->i_mutex);
5888 static int instance_rmdir(struct inode *inode, struct dentry *dentry)
5890 struct dentry *parent;
5893 /* Paranoid: Make sure the parent is the "instances" directory */
5894 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
5895 if (WARN_ON_ONCE(parent != trace_instance_dir))
5898 /* The caller did a dget() on dentry */
5899 mutex_unlock(&dentry->d_inode->i_mutex);
5902 * The inode mutex is locked, but debugfs_create_dir() will also
5903 * take the mutex. As the instances directory can not be destroyed
5904 * or changed in any other way, it is safe to unlock it, and
5905 * let the dentry try. If two users try to make the same dir at
5906 * the same time, then the instance_delete() will determine the
5909 mutex_unlock(&inode->i_mutex);
5911 ret = instance_delete(dentry->d_iname);
5913 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
5914 mutex_lock(&dentry->d_inode->i_mutex);
5919 static const struct inode_operations instance_dir_inode_operations = {
5920 .lookup = simple_lookup,
5921 .mkdir = instance_mkdir,
5922 .rmdir = instance_rmdir,
5925 static __init void create_trace_instances(struct dentry *d_tracer)
5927 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
5928 if (WARN_ON(!trace_instance_dir))
5931 /* Hijack the dir inode operations, to allow mkdir */
5932 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
5936 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
5940 trace_create_file("trace_options", 0644, d_tracer,
5941 tr, &tracing_iter_fops);
5943 trace_create_file("trace", 0644, d_tracer,
5944 (void *)&tr->trace_cpu, &tracing_fops);
5946 trace_create_file("trace_pipe", 0444, d_tracer,
5947 (void *)&tr->trace_cpu, &tracing_pipe_fops);
5949 trace_create_file("buffer_size_kb", 0644, d_tracer,
5950 (void *)&tr->trace_cpu, &tracing_entries_fops);
5952 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
5953 tr, &tracing_total_entries_fops);
5955 trace_create_file("free_buffer", 0200, d_tracer,
5956 tr, &tracing_free_buffer_fops);
5958 trace_create_file("trace_marker", 0220, d_tracer,
5959 tr, &tracing_mark_fops);
5961 trace_create_file("trace_clock", 0644, d_tracer, tr,
5964 trace_create_file("tracing_on", 0644, d_tracer,
5965 tr, &rb_simple_fops);
5967 #ifdef CONFIG_TRACER_SNAPSHOT
5968 trace_create_file("snapshot", 0644, d_tracer,
5969 (void *)&tr->trace_cpu, &snapshot_fops);
5972 for_each_tracing_cpu(cpu)
5973 tracing_init_debugfs_percpu(tr, cpu);
5977 static __init int tracer_init_debugfs(void)
5979 struct dentry *d_tracer;
5981 trace_access_lock_init();
5983 d_tracer = tracing_init_dentry();
5987 init_tracer_debugfs(&global_trace, d_tracer);
5989 trace_create_file("tracing_cpumask", 0644, d_tracer,
5990 &global_trace, &tracing_cpumask_fops);
5992 trace_create_file("available_tracers", 0444, d_tracer,
5993 &global_trace, &show_traces_fops);
5995 trace_create_file("current_tracer", 0644, d_tracer,
5996 &global_trace, &set_tracer_fops);
5998 #ifdef CONFIG_TRACER_MAX_TRACE
5999 trace_create_file("tracing_max_latency", 0644, d_tracer,
6000 &tracing_max_latency, &tracing_max_lat_fops);
6003 trace_create_file("tracing_thresh", 0644, d_tracer,
6004 &tracing_thresh, &tracing_max_lat_fops);
6006 trace_create_file("README", 0444, d_tracer,
6007 NULL, &tracing_readme_fops);
6009 trace_create_file("saved_cmdlines", 0444, d_tracer,
6010 NULL, &tracing_saved_cmdlines_fops);
6012 #ifdef CONFIG_DYNAMIC_FTRACE
6013 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
6014 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
6017 create_trace_instances(d_tracer);
6019 create_trace_options_dir(&global_trace);
6024 static int trace_panic_handler(struct notifier_block *this,
6025 unsigned long event, void *unused)
6027 if (ftrace_dump_on_oops)
6028 ftrace_dump(ftrace_dump_on_oops);
6032 static struct notifier_block trace_panic_notifier = {
6033 .notifier_call = trace_panic_handler,
6035 .priority = 150 /* priority: INT_MAX >= x >= 0 */
6038 static int trace_die_handler(struct notifier_block *self,
6044 if (ftrace_dump_on_oops)
6045 ftrace_dump(ftrace_dump_on_oops);
6053 static struct notifier_block trace_die_notifier = {
6054 .notifier_call = trace_die_handler,
6059 * printk is set to max of 1024, we really don't need it that big.
6060 * Nothing should be printing 1000 characters anyway.
6062 #define TRACE_MAX_PRINT 1000
6065 * Define here KERN_TRACE so that we have one place to modify
6066 * it if we decide to change what log level the ftrace dump
6069 #define KERN_TRACE KERN_EMERG
6072 trace_printk_seq(struct trace_seq *s)
6074 /* Probably should print a warning here. */
6075 if (s->len >= TRACE_MAX_PRINT)
6076 s->len = TRACE_MAX_PRINT;
6078 /* should be zero ended, but we are paranoid. */
6079 s->buffer[s->len] = 0;
6081 printk(KERN_TRACE "%s", s->buffer);
6086 void trace_init_global_iter(struct trace_iterator *iter)
6088 iter->tr = &global_trace;
6089 iter->trace = iter->tr->current_trace;
6090 iter->cpu_file = RING_BUFFER_ALL_CPUS;
6091 iter->trace_buffer = &global_trace.trace_buffer;
6094 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
6096 /* use static because iter can be a bit big for the stack */
6097 static struct trace_iterator iter;
6098 static atomic_t dump_running;
6099 unsigned int old_userobj;
6100 unsigned long flags;
6103 /* Only allow one dump user at a time. */
6104 if (atomic_inc_return(&dump_running) != 1) {
6105 atomic_dec(&dump_running);
6110 * Always turn off tracing when we dump.
6111 * We don't need to show trace output of what happens
6112 * between multiple crashes.
6114 * If the user does a sysrq-z, then they can re-enable
6115 * tracing with echo 1 > tracing_on.
6119 local_irq_save(flags);
6121 /* Simulate the iterator */
6122 trace_init_global_iter(&iter);
6124 for_each_tracing_cpu(cpu) {
6125 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
6128 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
6130 /* don't look at user memory in panic mode */
6131 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
6133 switch (oops_dump_mode) {
6135 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6138 iter.cpu_file = raw_smp_processor_id();
6143 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
6144 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6147 printk(KERN_TRACE "Dumping ftrace buffer:\n");
6149 /* Did function tracer already get disabled? */
6150 if (ftrace_is_dead()) {
6151 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6152 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6156 * We need to stop all tracing on all CPUS to read the
6157 * the next buffer. This is a bit expensive, but is
6158 * not done often. We fill all what we can read,
6159 * and then release the locks again.
6162 while (!trace_empty(&iter)) {
6165 printk(KERN_TRACE "---------------------------------\n");
6169 /* reset all but tr, trace, and overruns */
6170 memset(&iter.seq, 0,
6171 sizeof(struct trace_iterator) -
6172 offsetof(struct trace_iterator, seq));
6173 iter.iter_flags |= TRACE_FILE_LAT_FMT;
6176 if (trace_find_next_entry_inc(&iter) != NULL) {
6179 ret = print_trace_line(&iter);
6180 if (ret != TRACE_TYPE_NO_CONSUME)
6181 trace_consume(&iter);
6183 touch_nmi_watchdog();
6185 trace_printk_seq(&iter.seq);
6189 printk(KERN_TRACE " (ftrace buffer empty)\n");
6191 printk(KERN_TRACE "---------------------------------\n");
6194 trace_flags |= old_userobj;
6196 for_each_tracing_cpu(cpu) {
6197 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
6199 atomic_dec(&dump_running);
6200 local_irq_restore(flags);
6202 EXPORT_SYMBOL_GPL(ftrace_dump);
6204 __init static int tracer_alloc_buffers(void)
6210 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
6213 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
6214 goto out_free_buffer_mask;
6216 /* Only allocate trace_printk buffers if a trace_printk exists */
6217 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
6218 /* Must be called before global_trace.buffer is allocated */
6219 trace_printk_init_buffers();
6221 /* To save memory, keep the ring buffer size to its minimum */
6222 if (ring_buffer_expanded)
6223 ring_buf_size = trace_buf_size;
6227 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
6228 cpumask_copy(tracing_cpumask, cpu_all_mask);
6230 raw_spin_lock_init(&global_trace.start_lock);
6232 /* TODO: make the number of buffers hot pluggable with CPUS */
6233 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
6234 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
6236 goto out_free_cpumask;
6239 if (global_trace.buffer_disabled)
6242 trace_init_cmdlines();
6245 * register_tracer() might reference current_trace, so it
6246 * needs to be set before we register anything. This is
6247 * just a bootstrap of current_trace anyway.
6249 global_trace.current_trace = &nop_trace;
6251 register_tracer(&nop_trace);
6253 /* All seems OK, enable tracing */
6254 tracing_disabled = 0;
6256 atomic_notifier_chain_register(&panic_notifier_list,
6257 &trace_panic_notifier);
6259 register_die_notifier(&trace_die_notifier);
6261 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
6263 /* Holder for file callbacks */
6264 global_trace.trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
6265 global_trace.trace_cpu.tr = &global_trace;
6267 INIT_LIST_HEAD(&global_trace.systems);
6268 INIT_LIST_HEAD(&global_trace.events);
6269 list_add(&global_trace.list, &ftrace_trace_arrays);
6271 while (trace_boot_options) {
6274 option = strsep(&trace_boot_options, ",");
6275 trace_set_options(&global_trace, option);
6278 register_snapshot_cmd();
6283 free_percpu(global_trace.trace_buffer.data);
6284 #ifdef CONFIG_TRACER_MAX_TRACE
6285 free_percpu(global_trace.max_buffer.data);
6287 free_cpumask_var(tracing_cpumask);
6288 out_free_buffer_mask:
6289 free_cpumask_var(tracing_buffer_mask);
6294 __init static int clear_boot_tracer(void)
6297 * The default tracer at boot buffer is an init section.
6298 * This function is called in lateinit. If we did not
6299 * find the boot tracer, then clear it out, to prevent
6300 * later registration from accessing the buffer that is
6301 * about to be freed.
6303 if (!default_bootup_tracer)
6306 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
6307 default_bootup_tracer);
6308 default_bootup_tracer = NULL;
6313 early_initcall(tracer_alloc_buffers);
6314 fs_initcall(tracer_init_debugfs);
6315 late_initcall(clear_boot_tracer);