2 * ring buffer based function tracer
4 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 Nadia Yvette Chambers
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/nmi.h>
41 #include <linux/sched/rt.h>
44 #include "trace_output.h"
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
50 bool ring_buffer_expanded;
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
59 static bool __read_mostly tracing_selftest_running;
62 * If a tracer is running, we do not want to run SELFTEST.
64 bool __read_mostly tracing_selftest_disabled;
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt[] = {
71 static struct tracer_flags dummy_tracer_flags = {
73 .opts = dummy_tracer_opt
76 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
82 * To prevent the comm cache from being overwritten when no
83 * tracing is active, only save the comm when a trace event
86 static DEFINE_PER_CPU(bool, trace_cmdline_save);
89 * Kill all tracing for good (never come back).
90 * It is initialized to 1 but will turn to zero if the initialization
91 * of the tracer is successful. But that is the only place that sets
94 static int tracing_disabled = 1;
96 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
98 cpumask_var_t __read_mostly tracing_buffer_mask;
101 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
103 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
104 * is set, then ftrace_dump is called. This will output the contents
105 * of the ftrace buffers to the console. This is very useful for
106 * capturing traces that lead to crashes and outputing it to a
109 * It is default off, but you can enable it with either specifying
110 * "ftrace_dump_on_oops" in the kernel command line, or setting
111 * /proc/sys/kernel/ftrace_dump_on_oops
112 * Set 1 if you want to dump buffers of all CPUs
113 * Set 2 if you want to dump the buffer of the CPU that triggered oops
116 enum ftrace_dump_mode ftrace_dump_on_oops;
118 static int tracing_set_tracer(const char *buf);
120 #define MAX_TRACER_SIZE 100
121 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
122 static char *default_bootup_tracer;
124 static bool allocate_snapshot;
126 static int __init set_cmdline_ftrace(char *str)
128 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
129 default_bootup_tracer = bootup_tracer_buf;
130 /* We are using ftrace early, expand it */
131 ring_buffer_expanded = true;
134 __setup("ftrace=", set_cmdline_ftrace);
136 static int __init set_ftrace_dump_on_oops(char *str)
138 if (*str++ != '=' || !*str) {
139 ftrace_dump_on_oops = DUMP_ALL;
143 if (!strcmp("orig_cpu", str)) {
144 ftrace_dump_on_oops = DUMP_ORIG;
150 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
152 static int __init boot_alloc_snapshot(char *str)
154 allocate_snapshot = true;
155 /* We also need the main ring buffer expanded */
156 ring_buffer_expanded = true;
159 __setup("alloc_snapshot", boot_alloc_snapshot);
162 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
163 static char *trace_boot_options __initdata;
165 static int __init set_trace_boot_options(char *str)
167 strncpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
168 trace_boot_options = trace_boot_options_buf;
171 __setup("trace_options=", set_trace_boot_options);
173 unsigned long long ns2usecs(cycle_t nsec)
181 * The global_trace is the descriptor that holds the tracing
182 * buffers for the live tracing. For each CPU, it contains
183 * a link list of pages that will store trace entries. The
184 * page descriptor of the pages in the memory is used to hold
185 * the link list by linking the lru item in the page descriptor
186 * to each of the pages in the buffer per CPU.
188 * For each active CPU there is a data field that holds the
189 * pages for the buffer for that CPU. Each CPU has the same number
190 * of pages allocated for its buffer.
192 static struct trace_array global_trace;
194 LIST_HEAD(ftrace_trace_arrays);
196 int filter_current_check_discard(struct ring_buffer *buffer,
197 struct ftrace_event_call *call, void *rec,
198 struct ring_buffer_event *event)
200 return filter_check_discard(call, rec, buffer, event);
202 EXPORT_SYMBOL_GPL(filter_current_check_discard);
204 cycle_t ftrace_now(int cpu)
208 /* Early boot up does not have a buffer yet */
209 if (!global_trace.trace_buffer.buffer)
210 return trace_clock_local();
212 ts = ring_buffer_time_stamp(global_trace.trace_buffer.buffer, cpu);
213 ring_buffer_normalize_time_stamp(global_trace.trace_buffer.buffer, cpu, &ts);
218 int tracing_is_enabled(void)
220 return tracing_is_on();
224 * trace_buf_size is the size in bytes that is allocated
225 * for a buffer. Note, the number of bytes is always rounded
228 * This number is purposely set to a low number of 16384.
229 * If the dump on oops happens, it will be much appreciated
230 * to not have to wait for all that output. Anyway this can be
231 * boot time and run time configurable.
233 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
235 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
237 /* trace_types holds a link list of available tracers. */
238 static struct tracer *trace_types __read_mostly;
241 * trace_types_lock is used to protect the trace_types list.
243 static DEFINE_MUTEX(trace_types_lock);
246 * serialize the access of the ring buffer
248 * ring buffer serializes readers, but it is low level protection.
249 * The validity of the events (which returns by ring_buffer_peek() ..etc)
250 * are not protected by ring buffer.
252 * The content of events may become garbage if we allow other process consumes
253 * these events concurrently:
254 * A) the page of the consumed events may become a normal page
255 * (not reader page) in ring buffer, and this page will be rewrited
256 * by events producer.
257 * B) The page of the consumed events may become a page for splice_read,
258 * and this page will be returned to system.
260 * These primitives allow multi process access to different cpu ring buffer
263 * These primitives don't distinguish read-only and read-consume access.
264 * Multi read-only access are also serialized.
268 static DECLARE_RWSEM(all_cpu_access_lock);
269 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
271 static inline void trace_access_lock(int cpu)
273 if (cpu == RING_BUFFER_ALL_CPUS) {
274 /* gain it for accessing the whole ring buffer. */
275 down_write(&all_cpu_access_lock);
277 /* gain it for accessing a cpu ring buffer. */
279 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
280 down_read(&all_cpu_access_lock);
282 /* Secondly block other access to this @cpu ring buffer. */
283 mutex_lock(&per_cpu(cpu_access_lock, cpu));
287 static inline void trace_access_unlock(int cpu)
289 if (cpu == RING_BUFFER_ALL_CPUS) {
290 up_write(&all_cpu_access_lock);
292 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
293 up_read(&all_cpu_access_lock);
297 static inline void trace_access_lock_init(void)
301 for_each_possible_cpu(cpu)
302 mutex_init(&per_cpu(cpu_access_lock, cpu));
307 static DEFINE_MUTEX(access_lock);
309 static inline void trace_access_lock(int cpu)
312 mutex_lock(&access_lock);
315 static inline void trace_access_unlock(int cpu)
318 mutex_unlock(&access_lock);
321 static inline void trace_access_lock_init(void)
327 /* trace_flags holds trace_options default values */
328 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
329 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
330 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
331 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS;
334 * tracing_on - enable tracing buffers
336 * This function enables tracing buffers that may have been
337 * disabled with tracing_off.
339 void tracing_on(void)
341 if (global_trace.trace_buffer.buffer)
342 ring_buffer_record_on(global_trace.trace_buffer.buffer);
344 * This flag is only looked at when buffers haven't been
345 * allocated yet. We don't really care about the race
346 * between setting this flag and actually turning
349 global_trace.buffer_disabled = 0;
351 EXPORT_SYMBOL_GPL(tracing_on);
354 * __trace_puts - write a constant string into the trace buffer.
355 * @ip: The address of the caller
356 * @str: The constant string to write
357 * @size: The size of the string.
359 int __trace_puts(unsigned long ip, const char *str, int size)
361 struct ring_buffer_event *event;
362 struct ring_buffer *buffer;
363 struct print_entry *entry;
364 unsigned long irq_flags;
367 alloc = sizeof(*entry) + size + 2; /* possible \n added */
369 local_save_flags(irq_flags);
370 buffer = global_trace.trace_buffer.buffer;
371 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
372 irq_flags, preempt_count());
376 entry = ring_buffer_event_data(event);
379 memcpy(&entry->buf, str, size);
381 /* Add a newline if necessary */
382 if (entry->buf[size - 1] != '\n') {
383 entry->buf[size] = '\n';
384 entry->buf[size + 1] = '\0';
386 entry->buf[size] = '\0';
388 __buffer_unlock_commit(buffer, event);
392 EXPORT_SYMBOL_GPL(__trace_puts);
395 * __trace_bputs - write the pointer to a constant string into trace buffer
396 * @ip: The address of the caller
397 * @str: The constant string to write to the buffer to
399 int __trace_bputs(unsigned long ip, const char *str)
401 struct ring_buffer_event *event;
402 struct ring_buffer *buffer;
403 struct bputs_entry *entry;
404 unsigned long irq_flags;
405 int size = sizeof(struct bputs_entry);
407 local_save_flags(irq_flags);
408 buffer = global_trace.trace_buffer.buffer;
409 event = trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
410 irq_flags, preempt_count());
414 entry = ring_buffer_event_data(event);
418 __buffer_unlock_commit(buffer, event);
422 EXPORT_SYMBOL_GPL(__trace_bputs);
424 #ifdef CONFIG_TRACER_SNAPSHOT
426 * trace_snapshot - take a snapshot of the current buffer.
428 * This causes a swap between the snapshot buffer and the current live
429 * tracing buffer. You can use this to take snapshots of the live
430 * trace when some condition is triggered, but continue to trace.
432 * Note, make sure to allocate the snapshot with either
433 * a tracing_snapshot_alloc(), or by doing it manually
434 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
436 * If the snapshot buffer is not allocated, it will stop tracing.
437 * Basically making a permanent snapshot.
439 void tracing_snapshot(void)
441 struct trace_array *tr = &global_trace;
442 struct tracer *tracer = tr->current_trace;
446 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
447 internal_trace_puts("*** snapshot is being ignored ***\n");
451 if (!tr->allocated_snapshot) {
452 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
453 internal_trace_puts("*** stopping trace here! ***\n");
458 /* Note, snapshot can not be used when the tracer uses it */
459 if (tracer->use_max_tr) {
460 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
461 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
465 local_irq_save(flags);
466 update_max_tr(tr, current, smp_processor_id());
467 local_irq_restore(flags);
469 EXPORT_SYMBOL_GPL(tracing_snapshot);
471 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
472 struct trace_buffer *size_buf, int cpu_id);
473 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val);
475 static int alloc_snapshot(struct trace_array *tr)
479 if (!tr->allocated_snapshot) {
481 /* allocate spare buffer */
482 ret = resize_buffer_duplicate_size(&tr->max_buffer,
483 &tr->trace_buffer, RING_BUFFER_ALL_CPUS);
487 tr->allocated_snapshot = true;
493 void free_snapshot(struct trace_array *tr)
496 * We don't free the ring buffer. instead, resize it because
497 * The max_tr ring buffer has some state (e.g. ring->clock) and
498 * we want preserve it.
500 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
501 set_buffer_entries(&tr->max_buffer, 1);
502 tracing_reset_online_cpus(&tr->max_buffer);
503 tr->allocated_snapshot = false;
507 * trace_snapshot_alloc - allocate and take a snapshot of the current buffer.
509 * This is similar to trace_snapshot(), but it will allocate the
510 * snapshot buffer if it isn't already allocated. Use this only
511 * where it is safe to sleep, as the allocation may sleep.
513 * This causes a swap between the snapshot buffer and the current live
514 * tracing buffer. You can use this to take snapshots of the live
515 * trace when some condition is triggered, but continue to trace.
517 void tracing_snapshot_alloc(void)
519 struct trace_array *tr = &global_trace;
522 ret = alloc_snapshot(tr);
523 if (WARN_ON(ret < 0))
528 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
530 void tracing_snapshot(void)
532 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
534 EXPORT_SYMBOL_GPL(tracing_snapshot);
535 void tracing_snapshot_alloc(void)
540 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
541 #endif /* CONFIG_TRACER_SNAPSHOT */
544 * tracing_off - turn off tracing buffers
546 * This function stops the tracing buffers from recording data.
547 * It does not disable any overhead the tracers themselves may
548 * be causing. This function simply causes all recording to
549 * the ring buffers to fail.
551 void tracing_off(void)
553 if (global_trace.trace_buffer.buffer)
554 ring_buffer_record_off(global_trace.trace_buffer.buffer);
556 * This flag is only looked at when buffers haven't been
557 * allocated yet. We don't really care about the race
558 * between setting this flag and actually turning
561 global_trace.buffer_disabled = 1;
563 EXPORT_SYMBOL_GPL(tracing_off);
566 * tracing_is_on - show state of ring buffers enabled
568 int tracing_is_on(void)
570 if (global_trace.trace_buffer.buffer)
571 return ring_buffer_record_is_on(global_trace.trace_buffer.buffer);
572 return !global_trace.buffer_disabled;
574 EXPORT_SYMBOL_GPL(tracing_is_on);
576 static int __init set_buf_size(char *str)
578 unsigned long buf_size;
582 buf_size = memparse(str, &str);
583 /* nr_entries can not be zero */
586 trace_buf_size = buf_size;
589 __setup("trace_buf_size=", set_buf_size);
591 static int __init set_tracing_thresh(char *str)
593 unsigned long threshold;
598 ret = kstrtoul(str, 0, &threshold);
601 tracing_thresh = threshold * 1000;
604 __setup("tracing_thresh=", set_tracing_thresh);
606 unsigned long nsecs_to_usecs(unsigned long nsecs)
611 /* These must match the bit postions in trace_iterator_flags */
612 static const char *trace_options[] = {
644 int in_ns; /* is this clock in nanoseconds? */
646 { trace_clock_local, "local", 1 },
647 { trace_clock_global, "global", 1 },
648 { trace_clock_counter, "counter", 0 },
655 * trace_parser_get_init - gets the buffer for trace parser
657 int trace_parser_get_init(struct trace_parser *parser, int size)
659 memset(parser, 0, sizeof(*parser));
661 parser->buffer = kmalloc(size, GFP_KERNEL);
670 * trace_parser_put - frees the buffer for trace parser
672 void trace_parser_put(struct trace_parser *parser)
674 kfree(parser->buffer);
678 * trace_get_user - reads the user input string separated by space
679 * (matched by isspace(ch))
681 * For each string found the 'struct trace_parser' is updated,
682 * and the function returns.
684 * Returns number of bytes read.
686 * See kernel/trace/trace.h for 'struct trace_parser' details.
688 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
689 size_t cnt, loff_t *ppos)
696 trace_parser_clear(parser);
698 ret = get_user(ch, ubuf++);
706 * The parser is not finished with the last write,
707 * continue reading the user input without skipping spaces.
710 /* skip white space */
711 while (cnt && isspace(ch)) {
712 ret = get_user(ch, ubuf++);
719 /* only spaces were written */
729 /* read the non-space input */
730 while (cnt && !isspace(ch)) {
731 if (parser->idx < parser->size - 1)
732 parser->buffer[parser->idx++] = ch;
737 ret = get_user(ch, ubuf++);
744 /* We either got finished input or we have to wait for another call. */
746 parser->buffer[parser->idx] = 0;
747 parser->cont = false;
750 parser->buffer[parser->idx++] = ch;
760 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
768 if (s->len <= s->readpos)
771 len = s->len - s->readpos;
774 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
784 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
788 if (s->len <= s->readpos)
791 len = s->len - s->readpos;
794 memcpy(buf, s->buffer + s->readpos, cnt);
801 * ftrace_max_lock is used to protect the swapping of buffers
802 * when taking a max snapshot. The buffers themselves are
803 * protected by per_cpu spinlocks. But the action of the swap
804 * needs its own lock.
806 * This is defined as a arch_spinlock_t in order to help
807 * with performance when lockdep debugging is enabled.
809 * It is also used in other places outside the update_max_tr
810 * so it needs to be defined outside of the
811 * CONFIG_TRACER_MAX_TRACE.
813 static arch_spinlock_t ftrace_max_lock =
814 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
816 unsigned long __read_mostly tracing_thresh;
818 #ifdef CONFIG_TRACER_MAX_TRACE
819 unsigned long __read_mostly tracing_max_latency;
822 * Copy the new maximum trace into the separate maximum-trace
823 * structure. (this way the maximum trace is permanently saved,
824 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
827 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
829 struct trace_buffer *trace_buf = &tr->trace_buffer;
830 struct trace_buffer *max_buf = &tr->max_buffer;
831 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
832 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
835 max_buf->time_start = data->preempt_timestamp;
837 max_data->saved_latency = tracing_max_latency;
838 max_data->critical_start = data->critical_start;
839 max_data->critical_end = data->critical_end;
841 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
842 max_data->pid = tsk->pid;
843 max_data->uid = task_uid(tsk);
844 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
845 max_data->policy = tsk->policy;
846 max_data->rt_priority = tsk->rt_priority;
848 /* record this tasks comm */
849 tracing_record_cmdline(tsk);
853 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
855 * @tsk: the task with the latency
856 * @cpu: The cpu that initiated the trace.
858 * Flip the buffers between the @tr and the max_tr and record information
859 * about which task was the cause of this latency.
862 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
864 struct ring_buffer *buf;
869 WARN_ON_ONCE(!irqs_disabled());
871 if (!tr->allocated_snapshot) {
872 /* Only the nop tracer should hit this when disabling */
873 WARN_ON_ONCE(tr->current_trace != &nop_trace);
877 arch_spin_lock(&ftrace_max_lock);
879 buf = tr->trace_buffer.buffer;
880 tr->trace_buffer.buffer = tr->max_buffer.buffer;
881 tr->max_buffer.buffer = buf;
883 __update_max_tr(tr, tsk, cpu);
884 arch_spin_unlock(&ftrace_max_lock);
888 * update_max_tr_single - only copy one trace over, and reset the rest
890 * @tsk - task with the latency
891 * @cpu - the cpu of the buffer to copy.
893 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
896 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
903 WARN_ON_ONCE(!irqs_disabled());
904 if (WARN_ON_ONCE(!tr->allocated_snapshot))
907 arch_spin_lock(&ftrace_max_lock);
909 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
913 * We failed to swap the buffer due to a commit taking
914 * place on this CPU. We fail to record, but we reset
915 * the max trace buffer (no one writes directly to it)
916 * and flag that it failed.
918 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
919 "Failed to swap buffers due to commit in progress\n");
922 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
924 __update_max_tr(tr, tsk, cpu);
925 arch_spin_unlock(&ftrace_max_lock);
927 #endif /* CONFIG_TRACER_MAX_TRACE */
929 static void default_wait_pipe(struct trace_iterator *iter)
931 /* Iterators are static, they should be filled or empty */
932 if (trace_buffer_iter(iter, iter->cpu_file))
935 ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file);
938 #ifdef CONFIG_FTRACE_STARTUP_TEST
939 static int run_tracer_selftest(struct tracer *type)
941 struct trace_array *tr = &global_trace;
942 struct tracer *saved_tracer = tr->current_trace;
945 if (!type->selftest || tracing_selftest_disabled)
949 * Run a selftest on this tracer.
950 * Here we reset the trace buffer, and set the current
951 * tracer to be this tracer. The tracer can then run some
952 * internal tracing to verify that everything is in order.
953 * If we fail, we do not register this tracer.
955 tracing_reset_online_cpus(&tr->trace_buffer);
957 tr->current_trace = type;
959 #ifdef CONFIG_TRACER_MAX_TRACE
960 if (type->use_max_tr) {
961 /* If we expanded the buffers, make sure the max is expanded too */
962 if (ring_buffer_expanded)
963 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
964 RING_BUFFER_ALL_CPUS);
965 tr->allocated_snapshot = true;
969 /* the test is responsible for initializing and enabling */
970 pr_info("Testing tracer %s: ", type->name);
971 ret = type->selftest(type, tr);
972 /* the test is responsible for resetting too */
973 tr->current_trace = saved_tracer;
975 printk(KERN_CONT "FAILED!\n");
976 /* Add the warning after printing 'FAILED' */
980 /* Only reset on passing, to avoid touching corrupted buffers */
981 tracing_reset_online_cpus(&tr->trace_buffer);
983 #ifdef CONFIG_TRACER_MAX_TRACE
984 if (type->use_max_tr) {
985 tr->allocated_snapshot = false;
987 /* Shrink the max buffer again */
988 if (ring_buffer_expanded)
989 ring_buffer_resize(tr->max_buffer.buffer, 1,
990 RING_BUFFER_ALL_CPUS);
994 printk(KERN_CONT "PASSED\n");
998 static inline int run_tracer_selftest(struct tracer *type)
1002 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1005 * register_tracer - register a tracer with the ftrace system.
1006 * @type - the plugin for the tracer
1008 * Register a new plugin tracer.
1010 int register_tracer(struct tracer *type)
1016 pr_info("Tracer must have a name\n");
1020 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1021 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1025 mutex_lock(&trace_types_lock);
1027 tracing_selftest_running = true;
1029 for (t = trace_types; t; t = t->next) {
1030 if (strcmp(type->name, t->name) == 0) {
1032 pr_info("Tracer %s already registered\n",
1039 if (!type->set_flag)
1040 type->set_flag = &dummy_set_flag;
1042 type->flags = &dummy_tracer_flags;
1044 if (!type->flags->opts)
1045 type->flags->opts = dummy_tracer_opt;
1046 if (!type->wait_pipe)
1047 type->wait_pipe = default_wait_pipe;
1049 ret = run_tracer_selftest(type);
1053 type->next = trace_types;
1057 tracing_selftest_running = false;
1058 mutex_unlock(&trace_types_lock);
1060 if (ret || !default_bootup_tracer)
1063 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1066 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1067 /* Do we want this tracer to start on bootup? */
1068 tracing_set_tracer(type->name);
1069 default_bootup_tracer = NULL;
1070 /* disable other selftests, since this will break it. */
1071 tracing_selftest_disabled = true;
1072 #ifdef CONFIG_FTRACE_STARTUP_TEST
1073 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1081 void tracing_reset(struct trace_buffer *buf, int cpu)
1083 struct ring_buffer *buffer = buf->buffer;
1088 ring_buffer_record_disable(buffer);
1090 /* Make sure all commits have finished */
1091 synchronize_sched();
1092 ring_buffer_reset_cpu(buffer, cpu);
1094 ring_buffer_record_enable(buffer);
1097 void tracing_reset_online_cpus(struct trace_buffer *buf)
1099 struct ring_buffer *buffer = buf->buffer;
1105 ring_buffer_record_disable(buffer);
1107 /* Make sure all commits have finished */
1108 synchronize_sched();
1110 buf->time_start = ftrace_now(buf->cpu);
1112 for_each_online_cpu(cpu)
1113 ring_buffer_reset_cpu(buffer, cpu);
1115 ring_buffer_record_enable(buffer);
1118 void tracing_reset_current(int cpu)
1120 tracing_reset(&global_trace.trace_buffer, cpu);
1123 void tracing_reset_all_online_cpus(void)
1125 struct trace_array *tr;
1127 mutex_lock(&trace_types_lock);
1128 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1129 tracing_reset_online_cpus(&tr->trace_buffer);
1130 #ifdef CONFIG_TRACER_MAX_TRACE
1131 tracing_reset_online_cpus(&tr->max_buffer);
1134 mutex_unlock(&trace_types_lock);
1137 #define SAVED_CMDLINES 128
1138 #define NO_CMDLINE_MAP UINT_MAX
1139 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1140 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
1141 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
1142 static int cmdline_idx;
1143 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1145 /* temporary disable recording */
1146 static atomic_t trace_record_cmdline_disabled __read_mostly;
1148 static void trace_init_cmdlines(void)
1150 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
1151 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
1155 int is_tracing_stopped(void)
1157 return global_trace.stop_count;
1161 * ftrace_off_permanent - disable all ftrace code permanently
1163 * This should only be called when a serious anomally has
1164 * been detected. This will turn off the function tracing,
1165 * ring buffers, and other tracing utilites. It takes no
1166 * locks and can be called from any context.
1168 void ftrace_off_permanent(void)
1170 tracing_disabled = 1;
1172 tracing_off_permanent();
1176 * tracing_start - quick start of the tracer
1178 * If tracing is enabled but was stopped by tracing_stop,
1179 * this will start the tracer back up.
1181 void tracing_start(void)
1183 struct ring_buffer *buffer;
1184 unsigned long flags;
1186 if (tracing_disabled)
1189 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1190 if (--global_trace.stop_count) {
1191 if (global_trace.stop_count < 0) {
1192 /* Someone screwed up their debugging */
1194 global_trace.stop_count = 0;
1199 /* Prevent the buffers from switching */
1200 arch_spin_lock(&ftrace_max_lock);
1202 buffer = global_trace.trace_buffer.buffer;
1204 ring_buffer_record_enable(buffer);
1206 #ifdef CONFIG_TRACER_MAX_TRACE
1207 buffer = global_trace.max_buffer.buffer;
1209 ring_buffer_record_enable(buffer);
1212 arch_spin_unlock(&ftrace_max_lock);
1216 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1219 static void tracing_start_tr(struct trace_array *tr)
1221 struct ring_buffer *buffer;
1222 unsigned long flags;
1224 if (tracing_disabled)
1227 /* If global, we need to also start the max tracer */
1228 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1229 return tracing_start();
1231 raw_spin_lock_irqsave(&tr->start_lock, flags);
1233 if (--tr->stop_count) {
1234 if (tr->stop_count < 0) {
1235 /* Someone screwed up their debugging */
1242 buffer = tr->trace_buffer.buffer;
1244 ring_buffer_record_enable(buffer);
1247 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1251 * tracing_stop - quick stop of the tracer
1253 * Light weight way to stop tracing. Use in conjunction with
1256 void tracing_stop(void)
1258 struct ring_buffer *buffer;
1259 unsigned long flags;
1262 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1263 if (global_trace.stop_count++)
1266 /* Prevent the buffers from switching */
1267 arch_spin_lock(&ftrace_max_lock);
1269 buffer = global_trace.trace_buffer.buffer;
1271 ring_buffer_record_disable(buffer);
1273 #ifdef CONFIG_TRACER_MAX_TRACE
1274 buffer = global_trace.max_buffer.buffer;
1276 ring_buffer_record_disable(buffer);
1279 arch_spin_unlock(&ftrace_max_lock);
1282 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1285 static void tracing_stop_tr(struct trace_array *tr)
1287 struct ring_buffer *buffer;
1288 unsigned long flags;
1290 /* If global, we need to also stop the max tracer */
1291 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1292 return tracing_stop();
1294 raw_spin_lock_irqsave(&tr->start_lock, flags);
1295 if (tr->stop_count++)
1298 buffer = tr->trace_buffer.buffer;
1300 ring_buffer_record_disable(buffer);
1303 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1306 void trace_stop_cmdline_recording(void);
1308 static void trace_save_cmdline(struct task_struct *tsk)
1312 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1316 * It's not the end of the world if we don't get
1317 * the lock, but we also don't want to spin
1318 * nor do we want to disable interrupts,
1319 * so if we miss here, then better luck next time.
1321 if (!arch_spin_trylock(&trace_cmdline_lock))
1324 idx = map_pid_to_cmdline[tsk->pid];
1325 if (idx == NO_CMDLINE_MAP) {
1326 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1329 * Check whether the cmdline buffer at idx has a pid
1330 * mapped. We are going to overwrite that entry so we
1331 * need to clear the map_pid_to_cmdline. Otherwise we
1332 * would read the new comm for the old pid.
1334 pid = map_cmdline_to_pid[idx];
1335 if (pid != NO_CMDLINE_MAP)
1336 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1338 map_cmdline_to_pid[idx] = tsk->pid;
1339 map_pid_to_cmdline[tsk->pid] = idx;
1344 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1346 arch_spin_unlock(&trace_cmdline_lock);
1349 void trace_find_cmdline(int pid, char comm[])
1354 strcpy(comm, "<idle>");
1358 if (WARN_ON_ONCE(pid < 0)) {
1359 strcpy(comm, "<XXX>");
1363 if (pid > PID_MAX_DEFAULT) {
1364 strcpy(comm, "<...>");
1369 arch_spin_lock(&trace_cmdline_lock);
1370 map = map_pid_to_cmdline[pid];
1371 if (map != NO_CMDLINE_MAP)
1372 strcpy(comm, saved_cmdlines[map]);
1374 strcpy(comm, "<...>");
1376 arch_spin_unlock(&trace_cmdline_lock);
1380 void tracing_record_cmdline(struct task_struct *tsk)
1382 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1385 if (!__this_cpu_read(trace_cmdline_save))
1388 __this_cpu_write(trace_cmdline_save, false);
1390 trace_save_cmdline(tsk);
1394 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1397 struct task_struct *tsk = current;
1399 entry->preempt_count = pc & 0xff;
1400 entry->pid = (tsk) ? tsk->pid : 0;
1402 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1403 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1405 TRACE_FLAG_IRQS_NOSUPPORT |
1407 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1408 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1409 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1411 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1413 struct ring_buffer_event *
1414 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1417 unsigned long flags, int pc)
1419 struct ring_buffer_event *event;
1421 event = ring_buffer_lock_reserve(buffer, len);
1422 if (event != NULL) {
1423 struct trace_entry *ent = ring_buffer_event_data(event);
1425 tracing_generic_entry_update(ent, flags, pc);
1433 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1435 __this_cpu_write(trace_cmdline_save, true);
1436 ring_buffer_unlock_commit(buffer, event);
1440 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1441 struct ring_buffer_event *event,
1442 unsigned long flags, int pc)
1444 __buffer_unlock_commit(buffer, event);
1446 ftrace_trace_stack(buffer, flags, 6, pc);
1447 ftrace_trace_userstack(buffer, flags, pc);
1450 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1451 struct ring_buffer_event *event,
1452 unsigned long flags, int pc)
1454 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1456 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1458 struct ring_buffer_event *
1459 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1460 struct ftrace_event_file *ftrace_file,
1461 int type, unsigned long len,
1462 unsigned long flags, int pc)
1464 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1465 return trace_buffer_lock_reserve(*current_rb,
1466 type, len, flags, pc);
1468 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1470 struct ring_buffer_event *
1471 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1472 int type, unsigned long len,
1473 unsigned long flags, int pc)
1475 *current_rb = global_trace.trace_buffer.buffer;
1476 return trace_buffer_lock_reserve(*current_rb,
1477 type, len, flags, pc);
1479 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1481 void trace_current_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_current_buffer_unlock_commit);
1489 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1490 struct ring_buffer_event *event,
1491 unsigned long flags, int pc,
1492 struct pt_regs *regs)
1494 __buffer_unlock_commit(buffer, event);
1496 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1497 ftrace_trace_userstack(buffer, flags, pc);
1499 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1501 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1502 struct ring_buffer_event *event)
1504 ring_buffer_discard_commit(buffer, event);
1506 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1509 trace_function(struct trace_array *tr,
1510 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1513 struct ftrace_event_call *call = &event_function;
1514 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1515 struct ring_buffer_event *event;
1516 struct ftrace_entry *entry;
1518 /* If we are reading the ring buffer, don't trace */
1519 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1522 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1526 entry = ring_buffer_event_data(event);
1528 entry->parent_ip = parent_ip;
1530 if (!filter_check_discard(call, entry, buffer, event))
1531 __buffer_unlock_commit(buffer, event);
1535 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1536 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1539 if (likely(!atomic_read(&data->disabled)))
1540 trace_function(tr, ip, parent_ip, flags, pc);
1543 #ifdef CONFIG_STACKTRACE
1545 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1546 struct ftrace_stack {
1547 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1550 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1551 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1553 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1554 unsigned long flags,
1555 int skip, int pc, struct pt_regs *regs)
1557 struct ftrace_event_call *call = &event_kernel_stack;
1558 struct ring_buffer_event *event;
1559 struct stack_entry *entry;
1560 struct stack_trace trace;
1562 int size = FTRACE_STACK_ENTRIES;
1564 trace.nr_entries = 0;
1568 * Since events can happen in NMIs there's no safe way to
1569 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1570 * or NMI comes in, it will just have to use the default
1571 * FTRACE_STACK_SIZE.
1573 preempt_disable_notrace();
1575 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1577 * We don't need any atomic variables, just a barrier.
1578 * If an interrupt comes in, we don't care, because it would
1579 * have exited and put the counter back to what we want.
1580 * We just need a barrier to keep gcc from moving things
1584 if (use_stack == 1) {
1585 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1586 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1589 save_stack_trace_regs(regs, &trace);
1591 save_stack_trace(&trace);
1593 if (trace.nr_entries > size)
1594 size = trace.nr_entries;
1596 /* From now on, use_stack is a boolean */
1599 size *= sizeof(unsigned long);
1601 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1602 sizeof(*entry) + size, flags, pc);
1605 entry = ring_buffer_event_data(event);
1607 memset(&entry->caller, 0, size);
1610 memcpy(&entry->caller, trace.entries,
1611 trace.nr_entries * sizeof(unsigned long));
1613 trace.max_entries = FTRACE_STACK_ENTRIES;
1614 trace.entries = entry->caller;
1616 save_stack_trace_regs(regs, &trace);
1618 save_stack_trace(&trace);
1621 entry->size = trace.nr_entries;
1623 if (!filter_check_discard(call, entry, buffer, event))
1624 __buffer_unlock_commit(buffer, event);
1627 /* Again, don't let gcc optimize things here */
1629 __this_cpu_dec(ftrace_stack_reserve);
1630 preempt_enable_notrace();
1634 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1635 int skip, int pc, struct pt_regs *regs)
1637 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1640 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1643 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1646 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1649 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1652 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1655 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1659 * trace_dump_stack - record a stack back trace in the trace buffer
1661 void trace_dump_stack(void)
1663 unsigned long flags;
1665 if (tracing_disabled || tracing_selftest_running)
1668 local_save_flags(flags);
1670 /* skipping 3 traces, seems to get us at the caller of this function */
1671 __ftrace_trace_stack(global_trace.trace_buffer.buffer, flags, 3,
1672 preempt_count(), NULL);
1675 static DEFINE_PER_CPU(int, user_stack_count);
1678 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1680 struct ftrace_event_call *call = &event_user_stack;
1681 struct ring_buffer_event *event;
1682 struct userstack_entry *entry;
1683 struct stack_trace trace;
1685 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1689 * NMIs can not handle page faults, even with fix ups.
1690 * The save user stack can (and often does) fault.
1692 if (unlikely(in_nmi()))
1696 * prevent recursion, since the user stack tracing may
1697 * trigger other kernel events.
1700 if (__this_cpu_read(user_stack_count))
1703 __this_cpu_inc(user_stack_count);
1705 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1706 sizeof(*entry), flags, pc);
1708 goto out_drop_count;
1709 entry = ring_buffer_event_data(event);
1711 entry->tgid = current->tgid;
1712 memset(&entry->caller, 0, sizeof(entry->caller));
1714 trace.nr_entries = 0;
1715 trace.max_entries = FTRACE_STACK_ENTRIES;
1717 trace.entries = entry->caller;
1719 save_stack_trace_user(&trace);
1720 if (!filter_check_discard(call, entry, buffer, event))
1721 __buffer_unlock_commit(buffer, event);
1724 __this_cpu_dec(user_stack_count);
1730 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1732 ftrace_trace_userstack(tr, flags, preempt_count());
1736 #endif /* CONFIG_STACKTRACE */
1738 /* created for use with alloc_percpu */
1739 struct trace_buffer_struct {
1740 char buffer[TRACE_BUF_SIZE];
1743 static struct trace_buffer_struct *trace_percpu_buffer;
1744 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1745 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1746 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1749 * The buffer used is dependent on the context. There is a per cpu
1750 * buffer for normal context, softirq contex, hard irq context and
1751 * for NMI context. Thise allows for lockless recording.
1753 * Note, if the buffers failed to be allocated, then this returns NULL
1755 static char *get_trace_buf(void)
1757 struct trace_buffer_struct *percpu_buffer;
1760 * If we have allocated per cpu buffers, then we do not
1761 * need to do any locking.
1764 percpu_buffer = trace_percpu_nmi_buffer;
1766 percpu_buffer = trace_percpu_irq_buffer;
1767 else if (in_softirq())
1768 percpu_buffer = trace_percpu_sirq_buffer;
1770 percpu_buffer = trace_percpu_buffer;
1775 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1778 static int alloc_percpu_trace_buffer(void)
1780 struct trace_buffer_struct *buffers;
1781 struct trace_buffer_struct *sirq_buffers;
1782 struct trace_buffer_struct *irq_buffers;
1783 struct trace_buffer_struct *nmi_buffers;
1785 buffers = alloc_percpu(struct trace_buffer_struct);
1789 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1793 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1797 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1801 trace_percpu_buffer = buffers;
1802 trace_percpu_sirq_buffer = sirq_buffers;
1803 trace_percpu_irq_buffer = irq_buffers;
1804 trace_percpu_nmi_buffer = nmi_buffers;
1809 free_percpu(irq_buffers);
1811 free_percpu(sirq_buffers);
1813 free_percpu(buffers);
1815 WARN(1, "Could not allocate percpu trace_printk buffer");
1819 static int buffers_allocated;
1821 void trace_printk_init_buffers(void)
1823 if (buffers_allocated)
1826 if (alloc_percpu_trace_buffer())
1829 pr_info("ftrace: Allocated trace_printk buffers\n");
1831 /* Expand the buffers to set size */
1832 tracing_update_buffers();
1834 buffers_allocated = 1;
1837 * trace_printk_init_buffers() can be called by modules.
1838 * If that happens, then we need to start cmdline recording
1839 * directly here. If the global_trace.buffer is already
1840 * allocated here, then this was called by module code.
1842 if (global_trace.trace_buffer.buffer)
1843 tracing_start_cmdline_record();
1846 void trace_printk_start_comm(void)
1848 /* Start tracing comms if trace printk is set */
1849 if (!buffers_allocated)
1851 tracing_start_cmdline_record();
1854 static void trace_printk_start_stop_comm(int enabled)
1856 if (!buffers_allocated)
1860 tracing_start_cmdline_record();
1862 tracing_stop_cmdline_record();
1866 * trace_vbprintk - write binary msg to tracing buffer
1869 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1871 struct ftrace_event_call *call = &event_bprint;
1872 struct ring_buffer_event *event;
1873 struct ring_buffer *buffer;
1874 struct trace_array *tr = &global_trace;
1875 struct bprint_entry *entry;
1876 unsigned long flags;
1878 int len = 0, size, pc;
1880 if (unlikely(tracing_selftest_running || tracing_disabled))
1883 /* Don't pollute graph traces with trace_vprintk internals */
1884 pause_graph_tracing();
1886 pc = preempt_count();
1887 preempt_disable_notrace();
1889 tbuffer = get_trace_buf();
1895 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1897 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
1900 local_save_flags(flags);
1901 size = sizeof(*entry) + sizeof(u32) * len;
1902 buffer = tr->trace_buffer.buffer;
1903 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1907 entry = ring_buffer_event_data(event);
1911 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
1912 if (!filter_check_discard(call, entry, buffer, event)) {
1913 __buffer_unlock_commit(buffer, event);
1914 ftrace_trace_stack(buffer, flags, 6, pc);
1918 preempt_enable_notrace();
1919 unpause_graph_tracing();
1923 EXPORT_SYMBOL_GPL(trace_vbprintk);
1926 __trace_array_vprintk(struct ring_buffer *buffer,
1927 unsigned long ip, const char *fmt, va_list args)
1929 struct ftrace_event_call *call = &event_print;
1930 struct ring_buffer_event *event;
1931 int len = 0, size, pc;
1932 struct print_entry *entry;
1933 unsigned long flags;
1936 if (tracing_disabled || tracing_selftest_running)
1939 /* Don't pollute graph traces with trace_vprintk internals */
1940 pause_graph_tracing();
1942 pc = preempt_count();
1943 preempt_disable_notrace();
1946 tbuffer = get_trace_buf();
1952 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
1953 if (len > TRACE_BUF_SIZE)
1956 local_save_flags(flags);
1957 size = sizeof(*entry) + len + 1;
1958 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1962 entry = ring_buffer_event_data(event);
1965 memcpy(&entry->buf, tbuffer, len);
1966 entry->buf[len] = '\0';
1967 if (!filter_check_discard(call, entry, buffer, event)) {
1968 __buffer_unlock_commit(buffer, event);
1969 ftrace_trace_stack(buffer, flags, 6, pc);
1972 preempt_enable_notrace();
1973 unpause_graph_tracing();
1978 int trace_array_vprintk(struct trace_array *tr,
1979 unsigned long ip, const char *fmt, va_list args)
1981 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
1984 int trace_array_printk(struct trace_array *tr,
1985 unsigned long ip, const char *fmt, ...)
1990 if (!(trace_flags & TRACE_ITER_PRINTK))
1994 ret = trace_array_vprintk(tr, ip, fmt, ap);
1999 int trace_array_printk_buf(struct ring_buffer *buffer,
2000 unsigned long ip, const char *fmt, ...)
2005 if (!(trace_flags & TRACE_ITER_PRINTK))
2009 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2014 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2016 return trace_array_vprintk(&global_trace, ip, fmt, args);
2018 EXPORT_SYMBOL_GPL(trace_vprintk);
2020 static void trace_iterator_increment(struct trace_iterator *iter)
2022 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2026 ring_buffer_read(buf_iter, NULL);
2029 static struct trace_entry *
2030 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2031 unsigned long *lost_events)
2033 struct ring_buffer_event *event;
2034 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2037 event = ring_buffer_iter_peek(buf_iter, ts);
2039 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2043 iter->ent_size = ring_buffer_event_length(event);
2044 return ring_buffer_event_data(event);
2050 static struct trace_entry *
2051 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2052 unsigned long *missing_events, u64 *ent_ts)
2054 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2055 struct trace_entry *ent, *next = NULL;
2056 unsigned long lost_events = 0, next_lost = 0;
2057 int cpu_file = iter->cpu_file;
2058 u64 next_ts = 0, ts;
2064 * If we are in a per_cpu trace file, don't bother by iterating over
2065 * all cpu and peek directly.
2067 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2068 if (ring_buffer_empty_cpu(buffer, cpu_file))
2070 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2072 *ent_cpu = cpu_file;
2077 for_each_tracing_cpu(cpu) {
2079 if (ring_buffer_empty_cpu(buffer, cpu))
2082 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2085 * Pick the entry with the smallest timestamp:
2087 if (ent && (!next || ts < next_ts)) {
2091 next_lost = lost_events;
2092 next_size = iter->ent_size;
2096 iter->ent_size = next_size;
2099 *ent_cpu = next_cpu;
2105 *missing_events = next_lost;
2110 /* Find the next real entry, without updating the iterator itself */
2111 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2112 int *ent_cpu, u64 *ent_ts)
2114 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2117 /* Find the next real entry, and increment the iterator to the next entry */
2118 void *trace_find_next_entry_inc(struct trace_iterator *iter)
2120 iter->ent = __find_next_entry(iter, &iter->cpu,
2121 &iter->lost_events, &iter->ts);
2124 trace_iterator_increment(iter);
2126 return iter->ent ? iter : NULL;
2129 static void trace_consume(struct trace_iterator *iter)
2131 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2132 &iter->lost_events);
2135 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2137 struct trace_iterator *iter = m->private;
2141 WARN_ON_ONCE(iter->leftover);
2145 /* can't go backwards */
2150 ent = trace_find_next_entry_inc(iter);
2154 while (ent && iter->idx < i)
2155 ent = trace_find_next_entry_inc(iter);
2162 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2164 struct ring_buffer_event *event;
2165 struct ring_buffer_iter *buf_iter;
2166 unsigned long entries = 0;
2169 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2171 buf_iter = trace_buffer_iter(iter, cpu);
2175 ring_buffer_iter_reset(buf_iter);
2178 * We could have the case with the max latency tracers
2179 * that a reset never took place on a cpu. This is evident
2180 * by the timestamp being before the start of the buffer.
2182 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2183 if (ts >= iter->trace_buffer->time_start)
2186 ring_buffer_read(buf_iter, NULL);
2189 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2193 * The current tracer is copied to avoid a global locking
2196 static void *s_start(struct seq_file *m, loff_t *pos)
2198 struct trace_iterator *iter = m->private;
2199 struct trace_array *tr = iter->tr;
2200 int cpu_file = iter->cpu_file;
2206 * copy the tracer to avoid using a global lock all around.
2207 * iter->trace is a copy of current_trace, the pointer to the
2208 * name may be used instead of a strcmp(), as iter->trace->name
2209 * will point to the same string as current_trace->name.
2211 mutex_lock(&trace_types_lock);
2212 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2213 *iter->trace = *tr->current_trace;
2214 mutex_unlock(&trace_types_lock);
2216 #ifdef CONFIG_TRACER_MAX_TRACE
2217 if (iter->snapshot && iter->trace->use_max_tr)
2218 return ERR_PTR(-EBUSY);
2221 if (!iter->snapshot)
2222 atomic_inc(&trace_record_cmdline_disabled);
2224 if (*pos != iter->pos) {
2229 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2230 for_each_tracing_cpu(cpu)
2231 tracing_iter_reset(iter, cpu);
2233 tracing_iter_reset(iter, cpu_file);
2236 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2241 * If we overflowed the seq_file before, then we want
2242 * to just reuse the trace_seq buffer again.
2248 p = s_next(m, p, &l);
2252 trace_event_read_lock();
2253 trace_access_lock(cpu_file);
2257 static void s_stop(struct seq_file *m, void *p)
2259 struct trace_iterator *iter = m->private;
2261 #ifdef CONFIG_TRACER_MAX_TRACE
2262 if (iter->snapshot && iter->trace->use_max_tr)
2266 if (!iter->snapshot)
2267 atomic_dec(&trace_record_cmdline_disabled);
2269 trace_access_unlock(iter->cpu_file);
2270 trace_event_read_unlock();
2274 get_total_entries(struct trace_buffer *buf,
2275 unsigned long *total, unsigned long *entries)
2277 unsigned long count;
2283 for_each_tracing_cpu(cpu) {
2284 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2286 * If this buffer has skipped entries, then we hold all
2287 * entries for the trace and we need to ignore the
2288 * ones before the time stamp.
2290 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2291 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2292 /* total is the same as the entries */
2296 ring_buffer_overrun_cpu(buf->buffer, cpu);
2301 static void print_lat_help_header(struct seq_file *m)
2303 seq_puts(m, "# _------=> CPU# \n");
2304 seq_puts(m, "# / _-----=> irqs-off \n");
2305 seq_puts(m, "# | / _----=> need-resched \n");
2306 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2307 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2308 seq_puts(m, "# |||| / delay \n");
2309 seq_puts(m, "# cmd pid ||||| time | caller \n");
2310 seq_puts(m, "# \\ / ||||| \\ | / \n");
2313 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2315 unsigned long total;
2316 unsigned long entries;
2318 get_total_entries(buf, &total, &entries);
2319 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2320 entries, total, num_online_cpus());
2324 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2326 print_event_info(buf, m);
2327 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2328 seq_puts(m, "# | | | | |\n");
2331 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2333 print_event_info(buf, m);
2334 seq_puts(m, "# _-----=> irqs-off\n");
2335 seq_puts(m, "# / _----=> need-resched\n");
2336 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2337 seq_puts(m, "# || / _--=> preempt-depth\n");
2338 seq_puts(m, "# ||| / delay\n");
2339 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2340 seq_puts(m, "# | | | |||| | |\n");
2344 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2346 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2347 struct trace_buffer *buf = iter->trace_buffer;
2348 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2349 struct tracer *type = iter->trace;
2350 unsigned long entries;
2351 unsigned long total;
2352 const char *name = "preemption";
2356 get_total_entries(buf, &total, &entries);
2358 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2360 seq_puts(m, "# -----------------------------------"
2361 "---------------------------------\n");
2362 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2363 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2364 nsecs_to_usecs(data->saved_latency),
2368 #if defined(CONFIG_PREEMPT_NONE)
2370 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2372 #elif defined(CONFIG_PREEMPT)
2377 /* These are reserved for later use */
2380 seq_printf(m, " #P:%d)\n", num_online_cpus());
2384 seq_puts(m, "# -----------------\n");
2385 seq_printf(m, "# | task: %.16s-%d "
2386 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2387 data->comm, data->pid,
2388 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2389 data->policy, data->rt_priority);
2390 seq_puts(m, "# -----------------\n");
2392 if (data->critical_start) {
2393 seq_puts(m, "# => started at: ");
2394 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2395 trace_print_seq(m, &iter->seq);
2396 seq_puts(m, "\n# => ended at: ");
2397 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2398 trace_print_seq(m, &iter->seq);
2399 seq_puts(m, "\n#\n");
2405 static void test_cpu_buff_start(struct trace_iterator *iter)
2407 struct trace_seq *s = &iter->seq;
2409 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2412 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2415 if (cpumask_test_cpu(iter->cpu, iter->started))
2418 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2421 cpumask_set_cpu(iter->cpu, iter->started);
2423 /* Don't print started cpu buffer for the first entry of the trace */
2425 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2429 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2431 struct trace_seq *s = &iter->seq;
2432 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2433 struct trace_entry *entry;
2434 struct trace_event *event;
2438 test_cpu_buff_start(iter);
2440 event = ftrace_find_event(entry->type);
2442 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2443 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2444 if (!trace_print_lat_context(iter))
2447 if (!trace_print_context(iter))
2453 return event->funcs->trace(iter, sym_flags, event);
2455 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2458 return TRACE_TYPE_HANDLED;
2460 return TRACE_TYPE_PARTIAL_LINE;
2463 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2465 struct trace_seq *s = &iter->seq;
2466 struct trace_entry *entry;
2467 struct trace_event *event;
2471 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2472 if (!trace_seq_printf(s, "%d %d %llu ",
2473 entry->pid, iter->cpu, iter->ts))
2477 event = ftrace_find_event(entry->type);
2479 return event->funcs->raw(iter, 0, event);
2481 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2484 return TRACE_TYPE_HANDLED;
2486 return TRACE_TYPE_PARTIAL_LINE;
2489 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2491 struct trace_seq *s = &iter->seq;
2492 unsigned char newline = '\n';
2493 struct trace_entry *entry;
2494 struct trace_event *event;
2498 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2499 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2500 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2501 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2504 event = ftrace_find_event(entry->type);
2506 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2507 if (ret != TRACE_TYPE_HANDLED)
2511 SEQ_PUT_FIELD_RET(s, newline);
2513 return TRACE_TYPE_HANDLED;
2516 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2518 struct trace_seq *s = &iter->seq;
2519 struct trace_entry *entry;
2520 struct trace_event *event;
2524 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2525 SEQ_PUT_FIELD_RET(s, entry->pid);
2526 SEQ_PUT_FIELD_RET(s, iter->cpu);
2527 SEQ_PUT_FIELD_RET(s, iter->ts);
2530 event = ftrace_find_event(entry->type);
2531 return event ? event->funcs->binary(iter, 0, event) :
2535 int trace_empty(struct trace_iterator *iter)
2537 struct ring_buffer_iter *buf_iter;
2540 /* If we are looking at one CPU buffer, only check that one */
2541 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2542 cpu = iter->cpu_file;
2543 buf_iter = trace_buffer_iter(iter, cpu);
2545 if (!ring_buffer_iter_empty(buf_iter))
2548 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2554 for_each_tracing_cpu(cpu) {
2555 buf_iter = trace_buffer_iter(iter, cpu);
2557 if (!ring_buffer_iter_empty(buf_iter))
2560 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2568 /* Called with trace_event_read_lock() held. */
2569 enum print_line_t print_trace_line(struct trace_iterator *iter)
2571 enum print_line_t ret;
2573 if (iter->lost_events &&
2574 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2575 iter->cpu, iter->lost_events))
2576 return TRACE_TYPE_PARTIAL_LINE;
2578 if (iter->trace && iter->trace->print_line) {
2579 ret = iter->trace->print_line(iter);
2580 if (ret != TRACE_TYPE_UNHANDLED)
2584 if (iter->ent->type == TRACE_BPUTS &&
2585 trace_flags & TRACE_ITER_PRINTK &&
2586 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2587 return trace_print_bputs_msg_only(iter);
2589 if (iter->ent->type == TRACE_BPRINT &&
2590 trace_flags & TRACE_ITER_PRINTK &&
2591 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2592 return trace_print_bprintk_msg_only(iter);
2594 if (iter->ent->type == TRACE_PRINT &&
2595 trace_flags & TRACE_ITER_PRINTK &&
2596 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2597 return trace_print_printk_msg_only(iter);
2599 if (trace_flags & TRACE_ITER_BIN)
2600 return print_bin_fmt(iter);
2602 if (trace_flags & TRACE_ITER_HEX)
2603 return print_hex_fmt(iter);
2605 if (trace_flags & TRACE_ITER_RAW)
2606 return print_raw_fmt(iter);
2608 return print_trace_fmt(iter);
2611 void trace_latency_header(struct seq_file *m)
2613 struct trace_iterator *iter = m->private;
2615 /* print nothing if the buffers are empty */
2616 if (trace_empty(iter))
2619 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2620 print_trace_header(m, iter);
2622 if (!(trace_flags & TRACE_ITER_VERBOSE))
2623 print_lat_help_header(m);
2626 void trace_default_header(struct seq_file *m)
2628 struct trace_iterator *iter = m->private;
2630 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2633 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2634 /* print nothing if the buffers are empty */
2635 if (trace_empty(iter))
2637 print_trace_header(m, iter);
2638 if (!(trace_flags & TRACE_ITER_VERBOSE))
2639 print_lat_help_header(m);
2641 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2642 if (trace_flags & TRACE_ITER_IRQ_INFO)
2643 print_func_help_header_irq(iter->trace_buffer, m);
2645 print_func_help_header(iter->trace_buffer, m);
2650 static void test_ftrace_alive(struct seq_file *m)
2652 if (!ftrace_is_dead())
2654 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2655 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2658 #ifdef CONFIG_TRACER_MAX_TRACE
2659 static void show_snapshot_main_help(struct seq_file *m)
2661 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2662 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2663 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2664 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate)\n");
2665 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2666 seq_printf(m, "# is not a '0' or '1')\n");
2669 static void show_snapshot_percpu_help(struct seq_file *m)
2671 seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2672 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2673 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2674 seq_printf(m, "# Takes a snapshot of the main buffer for this cpu.\n");
2676 seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n");
2677 seq_printf(m, "# Must use main snapshot file to allocate.\n");
2679 seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2680 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2681 seq_printf(m, "# is not a '0' or '1')\n");
2684 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2686 if (iter->tr->allocated_snapshot)
2687 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2689 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2691 seq_printf(m, "# Snapshot commands:\n");
2692 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2693 show_snapshot_main_help(m);
2695 show_snapshot_percpu_help(m);
2698 /* Should never be called */
2699 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2702 static int s_show(struct seq_file *m, void *v)
2704 struct trace_iterator *iter = v;
2707 if (iter->ent == NULL) {
2709 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2711 test_ftrace_alive(m);
2713 if (iter->snapshot && trace_empty(iter))
2714 print_snapshot_help(m, iter);
2715 else if (iter->trace && iter->trace->print_header)
2716 iter->trace->print_header(m);
2718 trace_default_header(m);
2720 } else if (iter->leftover) {
2722 * If we filled the seq_file buffer earlier, we
2723 * want to just show it now.
2725 ret = trace_print_seq(m, &iter->seq);
2727 /* ret should this time be zero, but you never know */
2728 iter->leftover = ret;
2731 print_trace_line(iter);
2732 ret = trace_print_seq(m, &iter->seq);
2734 * If we overflow the seq_file buffer, then it will
2735 * ask us for this data again at start up.
2737 * ret is 0 if seq_file write succeeded.
2740 iter->leftover = ret;
2746 static const struct seq_operations tracer_seq_ops = {
2753 static struct trace_iterator *
2754 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
2756 struct trace_cpu *tc = inode->i_private;
2757 struct trace_array *tr = tc->tr;
2758 struct trace_iterator *iter;
2761 if (tracing_disabled)
2762 return ERR_PTR(-ENODEV);
2764 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2766 return ERR_PTR(-ENOMEM);
2768 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2770 if (!iter->buffer_iter)
2774 * We make a copy of the current tracer to avoid concurrent
2775 * changes on it while we are reading.
2777 mutex_lock(&trace_types_lock);
2778 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2782 *iter->trace = *tr->current_trace;
2784 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2789 #ifdef CONFIG_TRACER_MAX_TRACE
2790 /* Currently only the top directory has a snapshot */
2791 if (tr->current_trace->print_max || snapshot)
2792 iter->trace_buffer = &tr->max_buffer;
2795 iter->trace_buffer = &tr->trace_buffer;
2796 iter->snapshot = snapshot;
2798 mutex_init(&iter->mutex);
2799 iter->cpu_file = tc->cpu;
2801 /* Notify the tracer early; before we stop tracing. */
2802 if (iter->trace && iter->trace->open)
2803 iter->trace->open(iter);
2805 /* Annotate start of buffers if we had overruns */
2806 if (ring_buffer_overruns(iter->trace_buffer->buffer))
2807 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2809 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2810 if (trace_clocks[trace_clock_id].in_ns)
2811 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2813 /* stop the trace while dumping if we are not opening "snapshot" */
2814 if (!iter->snapshot)
2815 tracing_stop_tr(tr);
2817 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
2818 for_each_tracing_cpu(cpu) {
2819 iter->buffer_iter[cpu] =
2820 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2822 ring_buffer_read_prepare_sync();
2823 for_each_tracing_cpu(cpu) {
2824 ring_buffer_read_start(iter->buffer_iter[cpu]);
2825 tracing_iter_reset(iter, cpu);
2828 cpu = iter->cpu_file;
2829 iter->buffer_iter[cpu] =
2830 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2831 ring_buffer_read_prepare_sync();
2832 ring_buffer_read_start(iter->buffer_iter[cpu]);
2833 tracing_iter_reset(iter, cpu);
2838 mutex_unlock(&trace_types_lock);
2843 mutex_unlock(&trace_types_lock);
2845 kfree(iter->buffer_iter);
2847 seq_release_private(inode, file);
2848 return ERR_PTR(-ENOMEM);
2851 int tracing_open_generic(struct inode *inode, struct file *filp)
2853 if (tracing_disabled)
2856 filp->private_data = inode->i_private;
2860 static int tracing_release(struct inode *inode, struct file *file)
2862 struct seq_file *m = file->private_data;
2863 struct trace_iterator *iter;
2864 struct trace_array *tr;
2867 if (!(file->f_mode & FMODE_READ))
2873 mutex_lock(&trace_types_lock);
2878 for_each_tracing_cpu(cpu) {
2879 if (iter->buffer_iter[cpu])
2880 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2883 if (iter->trace && iter->trace->close)
2884 iter->trace->close(iter);
2886 if (!iter->snapshot)
2887 /* reenable tracing if it was previously enabled */
2888 tracing_start_tr(tr);
2889 mutex_unlock(&trace_types_lock);
2891 mutex_destroy(&iter->mutex);
2892 free_cpumask_var(iter->started);
2894 kfree(iter->buffer_iter);
2895 seq_release_private(inode, file);
2899 static int tracing_open(struct inode *inode, struct file *file)
2901 struct trace_iterator *iter;
2904 /* If this file was open for write, then erase contents */
2905 if ((file->f_mode & FMODE_WRITE) &&
2906 (file->f_flags & O_TRUNC)) {
2907 struct trace_cpu *tc = inode->i_private;
2908 struct trace_array *tr = tc->tr;
2910 if (tc->cpu == RING_BUFFER_ALL_CPUS)
2911 tracing_reset_online_cpus(&tr->trace_buffer);
2913 tracing_reset(&tr->trace_buffer, tc->cpu);
2916 if (file->f_mode & FMODE_READ) {
2917 iter = __tracing_open(inode, file, false);
2919 ret = PTR_ERR(iter);
2920 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2921 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2927 t_next(struct seq_file *m, void *v, loff_t *pos)
2929 struct tracer *t = v;
2939 static void *t_start(struct seq_file *m, loff_t *pos)
2944 mutex_lock(&trace_types_lock);
2945 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2951 static void t_stop(struct seq_file *m, void *p)
2953 mutex_unlock(&trace_types_lock);
2956 static int t_show(struct seq_file *m, void *v)
2958 struct tracer *t = v;
2963 seq_printf(m, "%s", t->name);
2972 static const struct seq_operations show_traces_seq_ops = {
2979 static int show_traces_open(struct inode *inode, struct file *file)
2981 if (tracing_disabled)
2984 return seq_open(file, &show_traces_seq_ops);
2988 tracing_write_stub(struct file *filp, const char __user *ubuf,
2989 size_t count, loff_t *ppos)
2994 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2996 if (file->f_mode & FMODE_READ)
2997 return seq_lseek(file, offset, origin);
3002 static const struct file_operations tracing_fops = {
3003 .open = tracing_open,
3005 .write = tracing_write_stub,
3006 .llseek = tracing_seek,
3007 .release = tracing_release,
3010 static const struct file_operations show_traces_fops = {
3011 .open = show_traces_open,
3013 .release = seq_release,
3014 .llseek = seq_lseek,
3018 * Only trace on a CPU if the bitmask is set:
3020 static cpumask_var_t tracing_cpumask;
3023 * The tracer itself will not take this lock, but still we want
3024 * to provide a consistent cpumask to user-space:
3026 static DEFINE_MUTEX(tracing_cpumask_update_lock);
3029 * Temporary storage for the character representation of the
3030 * CPU bitmask (and one more byte for the newline):
3032 static char mask_str[NR_CPUS + 1];
3035 tracing_cpumask_read(struct file *filp, char __user *ubuf,
3036 size_t count, loff_t *ppos)
3040 mutex_lock(&tracing_cpumask_update_lock);
3042 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
3043 if (count - len < 2) {
3047 len += sprintf(mask_str + len, "\n");
3048 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3051 mutex_unlock(&tracing_cpumask_update_lock);
3057 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3058 size_t count, loff_t *ppos)
3060 struct trace_array *tr = filp->private_data;
3061 cpumask_var_t tracing_cpumask_new;
3064 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3067 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3071 mutex_lock(&tracing_cpumask_update_lock);
3073 local_irq_disable();
3074 arch_spin_lock(&ftrace_max_lock);
3075 for_each_tracing_cpu(cpu) {
3077 * Increase/decrease the disabled counter if we are
3078 * about to flip a bit in the cpumask:
3080 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
3081 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3082 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3083 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3085 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
3086 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3087 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3088 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3091 arch_spin_unlock(&ftrace_max_lock);
3094 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
3096 mutex_unlock(&tracing_cpumask_update_lock);
3097 free_cpumask_var(tracing_cpumask_new);
3102 free_cpumask_var(tracing_cpumask_new);
3107 static const struct file_operations tracing_cpumask_fops = {
3108 .open = tracing_open_generic,
3109 .read = tracing_cpumask_read,
3110 .write = tracing_cpumask_write,
3111 .llseek = generic_file_llseek,
3114 static int tracing_trace_options_show(struct seq_file *m, void *v)
3116 struct tracer_opt *trace_opts;
3117 struct trace_array *tr = m->private;
3121 mutex_lock(&trace_types_lock);
3122 tracer_flags = tr->current_trace->flags->val;
3123 trace_opts = tr->current_trace->flags->opts;
3125 for (i = 0; trace_options[i]; i++) {
3126 if (trace_flags & (1 << i))
3127 seq_printf(m, "%s\n", trace_options[i]);
3129 seq_printf(m, "no%s\n", trace_options[i]);
3132 for (i = 0; trace_opts[i].name; i++) {
3133 if (tracer_flags & trace_opts[i].bit)
3134 seq_printf(m, "%s\n", trace_opts[i].name);
3136 seq_printf(m, "no%s\n", trace_opts[i].name);
3138 mutex_unlock(&trace_types_lock);
3143 static int __set_tracer_option(struct tracer *trace,
3144 struct tracer_flags *tracer_flags,
3145 struct tracer_opt *opts, int neg)
3149 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
3154 tracer_flags->val &= ~opts->bit;
3156 tracer_flags->val |= opts->bit;
3160 /* Try to assign a tracer specific option */
3161 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
3163 struct tracer_flags *tracer_flags = trace->flags;
3164 struct tracer_opt *opts = NULL;
3167 for (i = 0; tracer_flags->opts[i].name; i++) {
3168 opts = &tracer_flags->opts[i];
3170 if (strcmp(cmp, opts->name) == 0)
3171 return __set_tracer_option(trace, trace->flags,
3178 /* Some tracers require overwrite to stay enabled */
3179 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3181 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3187 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3189 /* do nothing if flag is already set */
3190 if (!!(trace_flags & mask) == !!enabled)
3193 /* Give the tracer a chance to approve the change */
3194 if (tr->current_trace->flag_changed)
3195 if (tr->current_trace->flag_changed(tr->current_trace, mask, !!enabled))
3199 trace_flags |= mask;
3201 trace_flags &= ~mask;
3203 if (mask == TRACE_ITER_RECORD_CMD)
3204 trace_event_enable_cmd_record(enabled);
3206 if (mask == TRACE_ITER_OVERWRITE) {
3207 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3208 #ifdef CONFIG_TRACER_MAX_TRACE
3209 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3213 if (mask == TRACE_ITER_PRINTK)
3214 trace_printk_start_stop_comm(enabled);
3219 static int trace_set_options(struct trace_array *tr, char *option)
3226 cmp = strstrip(option);
3228 if (strncmp(cmp, "no", 2) == 0) {
3233 mutex_lock(&trace_types_lock);
3235 for (i = 0; trace_options[i]; i++) {
3236 if (strcmp(cmp, trace_options[i]) == 0) {
3237 ret = set_tracer_flag(tr, 1 << i, !neg);
3242 /* If no option could be set, test the specific tracer options */
3243 if (!trace_options[i])
3244 ret = set_tracer_option(tr->current_trace, cmp, neg);
3246 mutex_unlock(&trace_types_lock);
3252 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3253 size_t cnt, loff_t *ppos)
3255 struct seq_file *m = filp->private_data;
3256 struct trace_array *tr = m->private;
3260 if (cnt >= sizeof(buf))
3263 if (copy_from_user(&buf, ubuf, cnt))
3268 ret = trace_set_options(tr, buf);
3277 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3279 if (tracing_disabled)
3282 return single_open(file, tracing_trace_options_show, inode->i_private);
3285 static const struct file_operations tracing_iter_fops = {
3286 .open = tracing_trace_options_open,
3288 .llseek = seq_lseek,
3289 .release = single_release,
3290 .write = tracing_trace_options_write,
3293 static const char readme_msg[] =
3294 "tracing mini-HOWTO:\n\n"
3295 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
3296 "# cat /sys/kernel/debug/tracing/available_tracers\n"
3297 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
3298 "# cat /sys/kernel/debug/tracing/current_tracer\n"
3300 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
3301 "# cat /sys/kernel/debug/tracing/current_tracer\n"
3303 "# cat /sys/kernel/debug/tracing/trace_options\n"
3304 "noprint-parent nosym-offset nosym-addr noverbose\n"
3305 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
3306 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
3307 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
3308 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
3312 tracing_readme_read(struct file *filp, char __user *ubuf,
3313 size_t cnt, loff_t *ppos)
3315 return simple_read_from_buffer(ubuf, cnt, ppos,
3316 readme_msg, strlen(readme_msg));
3319 static const struct file_operations tracing_readme_fops = {
3320 .open = tracing_open_generic,
3321 .read = tracing_readme_read,
3322 .llseek = generic_file_llseek,
3326 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3327 size_t cnt, loff_t *ppos)
3336 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3340 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3348 for (i = 0; i < SAVED_CMDLINES; i++) {
3351 pid = map_cmdline_to_pid[i];
3352 if (pid == -1 || pid == NO_CMDLINE_MAP)
3355 trace_find_cmdline(pid, buf_comm);
3356 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3361 len = simple_read_from_buffer(ubuf, cnt, ppos,
3370 static const struct file_operations tracing_saved_cmdlines_fops = {
3371 .open = tracing_open_generic,
3372 .read = tracing_saved_cmdlines_read,
3373 .llseek = generic_file_llseek,
3377 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3378 size_t cnt, loff_t *ppos)
3380 struct trace_array *tr = filp->private_data;
3381 char buf[MAX_TRACER_SIZE+2];
3384 mutex_lock(&trace_types_lock);
3385 r = sprintf(buf, "%s\n", tr->current_trace->name);
3386 mutex_unlock(&trace_types_lock);
3388 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3391 int tracer_init(struct tracer *t, struct trace_array *tr)
3393 tracing_reset_online_cpus(&tr->trace_buffer);
3397 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3401 for_each_tracing_cpu(cpu)
3402 per_cpu_ptr(buf->data, cpu)->entries = val;
3405 #ifdef CONFIG_TRACER_MAX_TRACE
3406 /* resize @tr's buffer to the size of @size_tr's entries */
3407 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3408 struct trace_buffer *size_buf, int cpu_id)
3412 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3413 for_each_tracing_cpu(cpu) {
3414 ret = ring_buffer_resize(trace_buf->buffer,
3415 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3418 per_cpu_ptr(trace_buf->data, cpu)->entries =
3419 per_cpu_ptr(size_buf->data, cpu)->entries;
3422 ret = ring_buffer_resize(trace_buf->buffer,
3423 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3425 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3426 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3431 #endif /* CONFIG_TRACER_MAX_TRACE */
3433 static int __tracing_resize_ring_buffer(struct trace_array *tr,
3434 unsigned long size, int cpu)
3439 * If kernel or user changes the size of the ring buffer
3440 * we use the size that was given, and we can forget about
3441 * expanding it later.
3443 ring_buffer_expanded = true;
3445 /* May be called before buffers are initialized */
3446 if (!tr->trace_buffer.buffer)
3449 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3453 #ifdef CONFIG_TRACER_MAX_TRACE
3454 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3455 !tr->current_trace->use_max_tr)
3458 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3460 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3461 &tr->trace_buffer, cpu);
3464 * AARGH! We are left with different
3465 * size max buffer!!!!
3466 * The max buffer is our "snapshot" buffer.
3467 * When a tracer needs a snapshot (one of the
3468 * latency tracers), it swaps the max buffer
3469 * with the saved snap shot. We succeeded to
3470 * update the size of the main buffer, but failed to
3471 * update the size of the max buffer. But when we tried
3472 * to reset the main buffer to the original size, we
3473 * failed there too. This is very unlikely to
3474 * happen, but if it does, warn and kill all
3478 tracing_disabled = 1;
3483 if (cpu == RING_BUFFER_ALL_CPUS)
3484 set_buffer_entries(&tr->max_buffer, size);
3486 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3489 #endif /* CONFIG_TRACER_MAX_TRACE */
3491 if (cpu == RING_BUFFER_ALL_CPUS)
3492 set_buffer_entries(&tr->trace_buffer, size);
3494 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3499 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3500 unsigned long size, int cpu_id)
3504 mutex_lock(&trace_types_lock);
3506 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3507 /* make sure, this cpu is enabled in the mask */
3508 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3514 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3519 mutex_unlock(&trace_types_lock);
3526 * tracing_update_buffers - used by tracing facility to expand ring buffers
3528 * To save on memory when the tracing is never used on a system with it
3529 * configured in. The ring buffers are set to a minimum size. But once
3530 * a user starts to use the tracing facility, then they need to grow
3531 * to their default size.
3533 * This function is to be called when a tracer is about to be used.
3535 int tracing_update_buffers(void)
3539 mutex_lock(&trace_types_lock);
3540 if (!ring_buffer_expanded)
3541 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3542 RING_BUFFER_ALL_CPUS);
3543 mutex_unlock(&trace_types_lock);
3548 struct trace_option_dentry;
3550 static struct trace_option_dentry *
3551 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3554 destroy_trace_option_files(struct trace_option_dentry *topts);
3556 static int tracing_set_tracer(const char *buf)
3558 static struct trace_option_dentry *topts;
3559 struct trace_array *tr = &global_trace;
3561 #ifdef CONFIG_TRACER_MAX_TRACE
3566 mutex_lock(&trace_types_lock);
3568 if (!ring_buffer_expanded) {
3569 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3570 RING_BUFFER_ALL_CPUS);
3576 for (t = trace_types; t; t = t->next) {
3577 if (strcmp(t->name, buf) == 0)
3584 if (t == tr->current_trace)
3587 trace_branch_disable();
3589 tr->current_trace->enabled = false;
3591 if (tr->current_trace->reset)
3592 tr->current_trace->reset(tr);
3594 /* Current trace needs to be nop_trace before synchronize_sched */
3595 tr->current_trace = &nop_trace;
3597 #ifdef CONFIG_TRACER_MAX_TRACE
3598 had_max_tr = tr->allocated_snapshot;
3600 if (had_max_tr && !t->use_max_tr) {
3602 * We need to make sure that the update_max_tr sees that
3603 * current_trace changed to nop_trace to keep it from
3604 * swapping the buffers after we resize it.
3605 * The update_max_tr is called from interrupts disabled
3606 * so a synchronized_sched() is sufficient.
3608 synchronize_sched();
3612 destroy_trace_option_files(topts);
3614 topts = create_trace_option_files(tr, t);
3616 #ifdef CONFIG_TRACER_MAX_TRACE
3617 if (t->use_max_tr && !had_max_tr) {
3618 ret = alloc_snapshot(tr);
3625 ret = tracer_init(t, tr);
3630 tr->current_trace = t;
3631 tr->current_trace->enabled = true;
3632 trace_branch_enable(tr);
3634 mutex_unlock(&trace_types_lock);
3640 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3641 size_t cnt, loff_t *ppos)
3643 char buf[MAX_TRACER_SIZE+1];
3650 if (cnt > MAX_TRACER_SIZE)
3651 cnt = MAX_TRACER_SIZE;
3653 if (copy_from_user(&buf, ubuf, cnt))
3658 /* strip ending whitespace. */
3659 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3662 err = tracing_set_tracer(buf);
3672 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3673 size_t cnt, loff_t *ppos)
3675 unsigned long *ptr = filp->private_data;
3679 r = snprintf(buf, sizeof(buf), "%ld\n",
3680 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3681 if (r > sizeof(buf))
3683 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3687 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3688 size_t cnt, loff_t *ppos)
3690 unsigned long *ptr = filp->private_data;
3694 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3703 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3705 struct trace_cpu *tc = inode->i_private;
3706 struct trace_array *tr = tc->tr;
3707 struct trace_iterator *iter;
3710 if (tracing_disabled)
3713 mutex_lock(&trace_types_lock);
3715 /* create a buffer to store the information to pass to userspace */
3716 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3723 * We make a copy of the current tracer to avoid concurrent
3724 * changes on it while we are reading.
3726 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3731 *iter->trace = *tr->current_trace;
3733 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3738 /* trace pipe does not show start of buffer */
3739 cpumask_setall(iter->started);
3741 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3742 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3744 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3745 if (trace_clocks[trace_clock_id].in_ns)
3746 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3748 iter->cpu_file = tc->cpu;
3750 iter->trace_buffer = &tc->tr->trace_buffer;
3751 mutex_init(&iter->mutex);
3752 filp->private_data = iter;
3754 if (iter->trace->pipe_open)
3755 iter->trace->pipe_open(iter);
3757 nonseekable_open(inode, filp);
3759 mutex_unlock(&trace_types_lock);
3765 mutex_unlock(&trace_types_lock);
3769 static int tracing_release_pipe(struct inode *inode, struct file *file)
3771 struct trace_iterator *iter = file->private_data;
3773 mutex_lock(&trace_types_lock);
3775 if (iter->trace->pipe_close)
3776 iter->trace->pipe_close(iter);
3778 mutex_unlock(&trace_types_lock);
3780 free_cpumask_var(iter->started);
3781 mutex_destroy(&iter->mutex);
3789 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
3791 /* Iterators are static, they should be filled or empty */
3792 if (trace_buffer_iter(iter, iter->cpu_file))
3793 return POLLIN | POLLRDNORM;
3795 if (trace_flags & TRACE_ITER_BLOCK)
3797 * Always select as readable when in blocking mode
3799 return POLLIN | POLLRDNORM;
3801 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
3806 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3808 struct trace_iterator *iter = filp->private_data;
3810 return trace_poll(iter, filp, poll_table);
3814 * This is a make-shift waitqueue.
3815 * A tracer might use this callback on some rare cases:
3817 * 1) the current tracer might hold the runqueue lock when it wakes up
3818 * a reader, hence a deadlock (sched, function, and function graph tracers)
3819 * 2) the function tracers, trace all functions, we don't want
3820 * the overhead of calling wake_up and friends
3821 * (and tracing them too)
3823 * Anyway, this is really very primitive wakeup.
3825 void poll_wait_pipe(struct trace_iterator *iter)
3827 set_current_state(TASK_INTERRUPTIBLE);
3828 /* sleep for 100 msecs, and try again. */
3829 schedule_timeout(HZ / 10);
3832 /* Must be called with trace_types_lock mutex held. */
3833 static int tracing_wait_pipe(struct file *filp)
3835 struct trace_iterator *iter = filp->private_data;
3837 while (trace_empty(iter)) {
3839 if ((filp->f_flags & O_NONBLOCK)) {
3843 mutex_unlock(&iter->mutex);
3845 iter->trace->wait_pipe(iter);
3847 mutex_lock(&iter->mutex);
3849 if (signal_pending(current))
3853 * We block until we read something and tracing is disabled.
3854 * We still block if tracing is disabled, but we have never
3855 * read anything. This allows a user to cat this file, and
3856 * then enable tracing. But after we have read something,
3857 * we give an EOF when tracing is again disabled.
3859 * iter->pos will be 0 if we haven't read anything.
3861 if (!tracing_is_enabled() && iter->pos)
3872 tracing_read_pipe(struct file *filp, char __user *ubuf,
3873 size_t cnt, loff_t *ppos)
3875 struct trace_iterator *iter = filp->private_data;
3876 struct trace_array *tr = iter->tr;
3879 /* return any leftover data */
3880 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3884 trace_seq_init(&iter->seq);
3886 /* copy the tracer to avoid using a global lock all around */
3887 mutex_lock(&trace_types_lock);
3888 if (unlikely(iter->trace->name != tr->current_trace->name))
3889 *iter->trace = *tr->current_trace;
3890 mutex_unlock(&trace_types_lock);
3893 * Avoid more than one consumer on a single file descriptor
3894 * This is just a matter of traces coherency, the ring buffer itself
3897 mutex_lock(&iter->mutex);
3898 if (iter->trace->read) {
3899 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3905 sret = tracing_wait_pipe(filp);
3909 /* stop when tracing is finished */
3910 if (trace_empty(iter)) {
3915 if (cnt >= PAGE_SIZE)
3916 cnt = PAGE_SIZE - 1;
3918 /* reset all but tr, trace, and overruns */
3919 memset(&iter->seq, 0,
3920 sizeof(struct trace_iterator) -
3921 offsetof(struct trace_iterator, seq));
3924 trace_event_read_lock();
3925 trace_access_lock(iter->cpu_file);
3926 while (trace_find_next_entry_inc(iter) != NULL) {
3927 enum print_line_t ret;
3928 int len = iter->seq.len;
3930 ret = print_trace_line(iter);
3931 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3932 /* don't print partial lines */
3933 iter->seq.len = len;
3936 if (ret != TRACE_TYPE_NO_CONSUME)
3937 trace_consume(iter);
3939 if (iter->seq.len >= cnt)
3943 * Setting the full flag means we reached the trace_seq buffer
3944 * size and we should leave by partial output condition above.
3945 * One of the trace_seq_* functions is not used properly.
3947 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3950 trace_access_unlock(iter->cpu_file);
3951 trace_event_read_unlock();
3953 /* Now copy what we have to the user */
3954 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3955 if (iter->seq.readpos >= iter->seq.len)
3956 trace_seq_init(&iter->seq);
3959 * If there was nothing to send to user, in spite of consuming trace
3960 * entries, go back to wait for more entries.
3966 mutex_unlock(&iter->mutex);
3971 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3972 struct pipe_buffer *buf)
3974 __free_page(buf->page);
3977 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3980 __free_page(spd->pages[idx]);
3983 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3985 .map = generic_pipe_buf_map,
3986 .unmap = generic_pipe_buf_unmap,
3987 .confirm = generic_pipe_buf_confirm,
3988 .release = tracing_pipe_buf_release,
3989 .steal = generic_pipe_buf_steal,
3990 .get = generic_pipe_buf_get,
3994 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3999 /* Seq buffer is page-sized, exactly what we need. */
4001 count = iter->seq.len;
4002 ret = print_trace_line(iter);
4003 count = iter->seq.len - count;
4006 iter->seq.len -= count;
4009 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4010 iter->seq.len -= count;
4014 if (ret != TRACE_TYPE_NO_CONSUME)
4015 trace_consume(iter);
4017 if (!trace_find_next_entry_inc(iter)) {
4027 static ssize_t tracing_splice_read_pipe(struct file *filp,
4029 struct pipe_inode_info *pipe,
4033 struct page *pages_def[PIPE_DEF_BUFFERS];
4034 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4035 struct trace_iterator *iter = filp->private_data;
4036 struct splice_pipe_desc spd = {
4038 .partial = partial_def,
4039 .nr_pages = 0, /* This gets updated below. */
4040 .nr_pages_max = PIPE_DEF_BUFFERS,
4042 .ops = &tracing_pipe_buf_ops,
4043 .spd_release = tracing_spd_release_pipe,
4045 struct trace_array *tr = iter->tr;
4050 if (splice_grow_spd(pipe, &spd))
4053 /* copy the tracer to avoid using a global lock all around */
4054 mutex_lock(&trace_types_lock);
4055 if (unlikely(iter->trace->name != tr->current_trace->name))
4056 *iter->trace = *tr->current_trace;
4057 mutex_unlock(&trace_types_lock);
4059 mutex_lock(&iter->mutex);
4061 if (iter->trace->splice_read) {
4062 ret = iter->trace->splice_read(iter, filp,
4063 ppos, pipe, len, flags);
4068 ret = tracing_wait_pipe(filp);
4072 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4077 trace_event_read_lock();
4078 trace_access_lock(iter->cpu_file);
4080 /* Fill as many pages as possible. */
4081 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
4082 spd.pages[i] = alloc_page(GFP_KERNEL);
4086 rem = tracing_fill_pipe_page(rem, iter);
4088 /* Copy the data into the page, so we can start over. */
4089 ret = trace_seq_to_buffer(&iter->seq,
4090 page_address(spd.pages[i]),
4093 __free_page(spd.pages[i]);
4096 spd.partial[i].offset = 0;
4097 spd.partial[i].len = iter->seq.len;
4099 trace_seq_init(&iter->seq);
4102 trace_access_unlock(iter->cpu_file);
4103 trace_event_read_unlock();
4104 mutex_unlock(&iter->mutex);
4108 ret = splice_to_pipe(pipe, &spd);
4110 splice_shrink_spd(&spd);
4114 mutex_unlock(&iter->mutex);
4119 tracing_entries_read(struct file *filp, char __user *ubuf,
4120 size_t cnt, loff_t *ppos)
4122 struct trace_cpu *tc = filp->private_data;
4123 struct trace_array *tr = tc->tr;
4128 mutex_lock(&trace_types_lock);
4130 if (tc->cpu == RING_BUFFER_ALL_CPUS) {
4131 int cpu, buf_size_same;
4136 /* check if all cpu sizes are same */
4137 for_each_tracing_cpu(cpu) {
4138 /* fill in the size from first enabled cpu */
4140 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4141 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4147 if (buf_size_same) {
4148 if (!ring_buffer_expanded)
4149 r = sprintf(buf, "%lu (expanded: %lu)\n",
4151 trace_buf_size >> 10);
4153 r = sprintf(buf, "%lu\n", size >> 10);
4155 r = sprintf(buf, "X\n");
4157 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, tc->cpu)->entries >> 10);
4159 mutex_unlock(&trace_types_lock);
4161 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4166 tracing_entries_write(struct file *filp, const char __user *ubuf,
4167 size_t cnt, loff_t *ppos)
4169 struct trace_cpu *tc = filp->private_data;
4173 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4177 /* must have at least 1 entry */
4181 /* value is in KB */
4184 ret = tracing_resize_ring_buffer(tc->tr, val, tc->cpu);
4194 tracing_total_entries_read(struct file *filp, char __user *ubuf,
4195 size_t cnt, loff_t *ppos)
4197 struct trace_array *tr = filp->private_data;
4200 unsigned long size = 0, expanded_size = 0;
4202 mutex_lock(&trace_types_lock);
4203 for_each_tracing_cpu(cpu) {
4204 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
4205 if (!ring_buffer_expanded)
4206 expanded_size += trace_buf_size >> 10;
4208 if (ring_buffer_expanded)
4209 r = sprintf(buf, "%lu\n", size);
4211 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
4212 mutex_unlock(&trace_types_lock);
4214 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4218 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4219 size_t cnt, loff_t *ppos)
4222 * There is no need to read what the user has written, this function
4223 * is just to make sure that there is no error when "echo" is used
4232 tracing_free_buffer_release(struct inode *inode, struct file *filp)
4234 struct trace_array *tr = inode->i_private;
4236 /* disable tracing ? */
4237 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4239 /* resize the ring buffer to 0 */
4240 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4246 tracing_mark_write(struct file *filp, const char __user *ubuf,
4247 size_t cnt, loff_t *fpos)
4249 unsigned long addr = (unsigned long)ubuf;
4250 struct ring_buffer_event *event;
4251 struct ring_buffer *buffer;
4252 struct print_entry *entry;
4253 unsigned long irq_flags;
4254 struct page *pages[2];
4264 if (tracing_disabled)
4267 if (!(trace_flags & TRACE_ITER_MARKERS))
4270 if (cnt > TRACE_BUF_SIZE)
4271 cnt = TRACE_BUF_SIZE;
4274 * Userspace is injecting traces into the kernel trace buffer.
4275 * We want to be as non intrusive as possible.
4276 * To do so, we do not want to allocate any special buffers
4277 * or take any locks, but instead write the userspace data
4278 * straight into the ring buffer.
4280 * First we need to pin the userspace buffer into memory,
4281 * which, most likely it is, because it just referenced it.
4282 * But there's no guarantee that it is. By using get_user_pages_fast()
4283 * and kmap_atomic/kunmap_atomic() we can get access to the
4284 * pages directly. We then write the data directly into the
4287 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4289 /* check if we cross pages */
4290 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4293 offset = addr & (PAGE_SIZE - 1);
4296 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4297 if (ret < nr_pages) {
4299 put_page(pages[ret]);
4304 for (i = 0; i < nr_pages; i++)
4305 map_page[i] = kmap_atomic(pages[i]);
4307 local_save_flags(irq_flags);
4308 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4309 buffer = global_trace.trace_buffer.buffer;
4310 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4311 irq_flags, preempt_count());
4313 /* Ring buffer disabled, return as if not open for write */
4318 entry = ring_buffer_event_data(event);
4319 entry->ip = _THIS_IP_;
4321 if (nr_pages == 2) {
4322 len = PAGE_SIZE - offset;
4323 memcpy(&entry->buf, map_page[0] + offset, len);
4324 memcpy(&entry->buf[len], map_page[1], cnt - len);
4326 memcpy(&entry->buf, map_page[0] + offset, cnt);
4328 if (entry->buf[cnt - 1] != '\n') {
4329 entry->buf[cnt] = '\n';
4330 entry->buf[cnt + 1] = '\0';
4332 entry->buf[cnt] = '\0';
4334 __buffer_unlock_commit(buffer, event);
4341 for (i = 0; i < nr_pages; i++){
4342 kunmap_atomic(map_page[i]);
4349 static int tracing_clock_show(struct seq_file *m, void *v)
4351 struct trace_array *tr = m->private;
4354 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4356 "%s%s%s%s", i ? " " : "",
4357 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4358 i == tr->clock_id ? "]" : "");
4364 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4365 size_t cnt, loff_t *fpos)
4367 struct seq_file *m = filp->private_data;
4368 struct trace_array *tr = m->private;
4370 const char *clockstr;
4373 if (cnt >= sizeof(buf))
4376 if (copy_from_user(&buf, ubuf, cnt))
4381 clockstr = strstrip(buf);
4383 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4384 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4387 if (i == ARRAY_SIZE(trace_clocks))
4390 mutex_lock(&trace_types_lock);
4394 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4397 * New clock may not be consistent with the previous clock.
4398 * Reset the buffer so that it doesn't have incomparable timestamps.
4400 tracing_reset_online_cpus(&global_trace.trace_buffer);
4402 #ifdef CONFIG_TRACER_MAX_TRACE
4403 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4404 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4405 tracing_reset_online_cpus(&global_trace.max_buffer);
4408 mutex_unlock(&trace_types_lock);
4415 static int tracing_clock_open(struct inode *inode, struct file *file)
4417 if (tracing_disabled)
4420 return single_open(file, tracing_clock_show, inode->i_private);
4423 struct ftrace_buffer_info {
4424 struct trace_iterator iter;
4429 #ifdef CONFIG_TRACER_SNAPSHOT
4430 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4432 struct trace_cpu *tc = inode->i_private;
4433 struct trace_iterator *iter;
4437 if (file->f_mode & FMODE_READ) {
4438 iter = __tracing_open(inode, file, true);
4440 ret = PTR_ERR(iter);
4442 /* Writes still need the seq_file to hold the private data */
4443 m = kzalloc(sizeof(*m), GFP_KERNEL);
4446 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4452 iter->trace_buffer = &tc->tr->max_buffer;
4453 iter->cpu_file = tc->cpu;
4455 file->private_data = m;
4462 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4465 struct seq_file *m = filp->private_data;
4466 struct trace_iterator *iter = m->private;
4467 struct trace_array *tr = iter->tr;
4471 ret = tracing_update_buffers();
4475 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4479 mutex_lock(&trace_types_lock);
4481 if (tr->current_trace->use_max_tr) {
4488 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4492 if (tr->allocated_snapshot)
4496 /* Only allow per-cpu swap if the ring buffer supports it */
4497 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4498 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4503 if (!tr->allocated_snapshot) {
4504 ret = alloc_snapshot(tr);
4508 local_irq_disable();
4509 /* Now, we're going to swap */
4510 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4511 update_max_tr(tr, current, smp_processor_id());
4513 update_max_tr_single(tr, current, iter->cpu_file);
4517 if (tr->allocated_snapshot) {
4518 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4519 tracing_reset_online_cpus(&tr->max_buffer);
4521 tracing_reset(&tr->max_buffer, iter->cpu_file);
4531 mutex_unlock(&trace_types_lock);
4535 static int tracing_snapshot_release(struct inode *inode, struct file *file)
4537 struct seq_file *m = file->private_data;
4539 if (file->f_mode & FMODE_READ)
4540 return tracing_release(inode, file);
4542 /* If write only, the seq_file is just a stub */
4550 static int tracing_buffers_open(struct inode *inode, struct file *filp);
4551 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4552 size_t count, loff_t *ppos);
4553 static int tracing_buffers_release(struct inode *inode, struct file *file);
4554 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4555 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
4557 static int snapshot_raw_open(struct inode *inode, struct file *filp)
4559 struct ftrace_buffer_info *info;
4562 ret = tracing_buffers_open(inode, filp);
4566 info = filp->private_data;
4568 if (info->iter.trace->use_max_tr) {
4569 tracing_buffers_release(inode, filp);
4573 info->iter.snapshot = true;
4574 info->iter.trace_buffer = &info->iter.tr->max_buffer;
4579 #endif /* CONFIG_TRACER_SNAPSHOT */
4582 static const struct file_operations tracing_max_lat_fops = {
4583 .open = tracing_open_generic,
4584 .read = tracing_max_lat_read,
4585 .write = tracing_max_lat_write,
4586 .llseek = generic_file_llseek,
4589 static const struct file_operations set_tracer_fops = {
4590 .open = tracing_open_generic,
4591 .read = tracing_set_trace_read,
4592 .write = tracing_set_trace_write,
4593 .llseek = generic_file_llseek,
4596 static const struct file_operations tracing_pipe_fops = {
4597 .open = tracing_open_pipe,
4598 .poll = tracing_poll_pipe,
4599 .read = tracing_read_pipe,
4600 .splice_read = tracing_splice_read_pipe,
4601 .release = tracing_release_pipe,
4602 .llseek = no_llseek,
4605 static const struct file_operations tracing_entries_fops = {
4606 .open = tracing_open_generic,
4607 .read = tracing_entries_read,
4608 .write = tracing_entries_write,
4609 .llseek = generic_file_llseek,
4612 static const struct file_operations tracing_total_entries_fops = {
4613 .open = tracing_open_generic,
4614 .read = tracing_total_entries_read,
4615 .llseek = generic_file_llseek,
4618 static const struct file_operations tracing_free_buffer_fops = {
4619 .write = tracing_free_buffer_write,
4620 .release = tracing_free_buffer_release,
4623 static const struct file_operations tracing_mark_fops = {
4624 .open = tracing_open_generic,
4625 .write = tracing_mark_write,
4626 .llseek = generic_file_llseek,
4629 static const struct file_operations trace_clock_fops = {
4630 .open = tracing_clock_open,
4632 .llseek = seq_lseek,
4633 .release = single_release,
4634 .write = tracing_clock_write,
4637 #ifdef CONFIG_TRACER_SNAPSHOT
4638 static const struct file_operations snapshot_fops = {
4639 .open = tracing_snapshot_open,
4641 .write = tracing_snapshot_write,
4642 .llseek = tracing_seek,
4643 .release = tracing_snapshot_release,
4646 static const struct file_operations snapshot_raw_fops = {
4647 .open = snapshot_raw_open,
4648 .read = tracing_buffers_read,
4649 .release = tracing_buffers_release,
4650 .splice_read = tracing_buffers_splice_read,
4651 .llseek = no_llseek,
4654 #endif /* CONFIG_TRACER_SNAPSHOT */
4656 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4658 struct trace_cpu *tc = inode->i_private;
4659 struct trace_array *tr = tc->tr;
4660 struct ftrace_buffer_info *info;
4662 if (tracing_disabled)
4665 info = kzalloc(sizeof(*info), GFP_KERNEL);
4669 mutex_lock(&trace_types_lock);
4674 info->iter.cpu_file = tc->cpu;
4675 info->iter.trace = tr->current_trace;
4676 info->iter.trace_buffer = &tr->trace_buffer;
4678 /* Force reading ring buffer for first read */
4679 info->read = (unsigned int)-1;
4681 filp->private_data = info;
4683 mutex_unlock(&trace_types_lock);
4685 return nonseekable_open(inode, filp);
4689 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
4691 struct ftrace_buffer_info *info = filp->private_data;
4692 struct trace_iterator *iter = &info->iter;
4694 return trace_poll(iter, filp, poll_table);
4698 tracing_buffers_read(struct file *filp, char __user *ubuf,
4699 size_t count, loff_t *ppos)
4701 struct ftrace_buffer_info *info = filp->private_data;
4702 struct trace_iterator *iter = &info->iter;
4709 mutex_lock(&trace_types_lock);
4711 #ifdef CONFIG_TRACER_MAX_TRACE
4712 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
4719 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
4725 /* Do we have previous read data to read? */
4726 if (info->read < PAGE_SIZE)
4730 trace_access_lock(iter->cpu_file);
4731 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
4735 trace_access_unlock(iter->cpu_file);
4738 if (trace_empty(iter)) {
4739 if ((filp->f_flags & O_NONBLOCK)) {
4743 mutex_unlock(&trace_types_lock);
4744 iter->trace->wait_pipe(iter);
4745 mutex_lock(&trace_types_lock);
4746 if (signal_pending(current)) {
4758 size = PAGE_SIZE - info->read;
4762 ret = copy_to_user(ubuf, info->spare + info->read, size);
4773 mutex_unlock(&trace_types_lock);
4778 static int tracing_buffers_release(struct inode *inode, struct file *file)
4780 struct ftrace_buffer_info *info = file->private_data;
4781 struct trace_iterator *iter = &info->iter;
4783 mutex_lock(&trace_types_lock);
4785 WARN_ON(!iter->tr->ref);
4789 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
4792 mutex_unlock(&trace_types_lock);
4798 struct ring_buffer *buffer;
4803 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4804 struct pipe_buffer *buf)
4806 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4811 ring_buffer_free_read_page(ref->buffer, ref->page);
4816 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4817 struct pipe_buffer *buf)
4819 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4824 /* Pipe buffer operations for a buffer. */
4825 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4827 .map = generic_pipe_buf_map,
4828 .unmap = generic_pipe_buf_unmap,
4829 .confirm = generic_pipe_buf_confirm,
4830 .release = buffer_pipe_buf_release,
4831 .steal = generic_pipe_buf_steal,
4832 .get = buffer_pipe_buf_get,
4836 * Callback from splice_to_pipe(), if we need to release some pages
4837 * at the end of the spd in case we error'ed out in filling the pipe.
4839 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4841 struct buffer_ref *ref =
4842 (struct buffer_ref *)spd->partial[i].private;
4847 ring_buffer_free_read_page(ref->buffer, ref->page);
4849 spd->partial[i].private = 0;
4853 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4854 struct pipe_inode_info *pipe, size_t len,
4857 struct ftrace_buffer_info *info = file->private_data;
4858 struct trace_iterator *iter = &info->iter;
4859 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4860 struct page *pages_def[PIPE_DEF_BUFFERS];
4861 struct splice_pipe_desc spd = {
4863 .partial = partial_def,
4864 .nr_pages_max = PIPE_DEF_BUFFERS,
4866 .ops = &buffer_pipe_buf_ops,
4867 .spd_release = buffer_spd_release,
4869 struct buffer_ref *ref;
4870 int entries, size, i;
4873 mutex_lock(&trace_types_lock);
4875 #ifdef CONFIG_TRACER_MAX_TRACE
4876 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
4882 if (splice_grow_spd(pipe, &spd)) {
4887 if (*ppos & (PAGE_SIZE - 1)) {
4892 if (len & (PAGE_SIZE - 1)) {
4893 if (len < PAGE_SIZE) {
4901 trace_access_lock(iter->cpu_file);
4902 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
4904 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4908 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4913 ref->buffer = iter->trace_buffer->buffer;
4914 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
4920 r = ring_buffer_read_page(ref->buffer, &ref->page,
4921 len, iter->cpu_file, 1);
4923 ring_buffer_free_read_page(ref->buffer, ref->page);
4929 * zero out any left over data, this is going to
4932 size = ring_buffer_page_len(ref->page);
4933 if (size < PAGE_SIZE)
4934 memset(ref->page + size, 0, PAGE_SIZE - size);
4936 page = virt_to_page(ref->page);
4938 spd.pages[i] = page;
4939 spd.partial[i].len = PAGE_SIZE;
4940 spd.partial[i].offset = 0;
4941 spd.partial[i].private = (unsigned long)ref;
4945 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
4948 trace_access_unlock(iter->cpu_file);
4951 /* did we read anything? */
4952 if (!spd.nr_pages) {
4953 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
4957 mutex_unlock(&trace_types_lock);
4958 iter->trace->wait_pipe(iter);
4959 mutex_lock(&trace_types_lock);
4960 if (signal_pending(current)) {
4967 ret = splice_to_pipe(pipe, &spd);
4968 splice_shrink_spd(&spd);
4970 mutex_unlock(&trace_types_lock);
4975 static const struct file_operations tracing_buffers_fops = {
4976 .open = tracing_buffers_open,
4977 .read = tracing_buffers_read,
4978 .poll = tracing_buffers_poll,
4979 .release = tracing_buffers_release,
4980 .splice_read = tracing_buffers_splice_read,
4981 .llseek = no_llseek,
4985 tracing_stats_read(struct file *filp, char __user *ubuf,
4986 size_t count, loff_t *ppos)
4988 struct trace_cpu *tc = filp->private_data;
4989 struct trace_array *tr = tc->tr;
4990 struct trace_buffer *trace_buf = &tr->trace_buffer;
4991 struct trace_seq *s;
4993 unsigned long long t;
4994 unsigned long usec_rem;
4997 s = kmalloc(sizeof(*s), GFP_KERNEL);
5003 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5004 trace_seq_printf(s, "entries: %ld\n", cnt);
5006 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5007 trace_seq_printf(s, "overrun: %ld\n", cnt);
5009 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5010 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5012 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5013 trace_seq_printf(s, "bytes: %ld\n", cnt);
5015 if (trace_clocks[trace_clock_id].in_ns) {
5016 /* local or global for trace_clock */
5017 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5018 usec_rem = do_div(t, USEC_PER_SEC);
5019 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5022 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5023 usec_rem = do_div(t, USEC_PER_SEC);
5024 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5026 /* counter or tsc mode for trace_clock */
5027 trace_seq_printf(s, "oldest event ts: %llu\n",
5028 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5030 trace_seq_printf(s, "now ts: %llu\n",
5031 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5034 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5035 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5037 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5038 trace_seq_printf(s, "read events: %ld\n", cnt);
5040 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
5047 static const struct file_operations tracing_stats_fops = {
5048 .open = tracing_open_generic,
5049 .read = tracing_stats_read,
5050 .llseek = generic_file_llseek,
5053 #ifdef CONFIG_DYNAMIC_FTRACE
5055 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5061 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5062 size_t cnt, loff_t *ppos)
5064 static char ftrace_dyn_info_buffer[1024];
5065 static DEFINE_MUTEX(dyn_info_mutex);
5066 unsigned long *p = filp->private_data;
5067 char *buf = ftrace_dyn_info_buffer;
5068 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5071 mutex_lock(&dyn_info_mutex);
5072 r = sprintf(buf, "%ld ", *p);
5074 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5077 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5079 mutex_unlock(&dyn_info_mutex);
5084 static const struct file_operations tracing_dyn_info_fops = {
5085 .open = tracing_open_generic,
5086 .read = tracing_read_dyn_info,
5087 .llseek = generic_file_llseek,
5089 #endif /* CONFIG_DYNAMIC_FTRACE */
5091 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5093 ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5099 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5101 unsigned long *count = (long *)data;
5113 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5114 struct ftrace_probe_ops *ops, void *data)
5116 long count = (long)data;
5118 seq_printf(m, "%ps:", (void *)ip);
5120 seq_printf(m, "snapshot");
5123 seq_printf(m, ":unlimited\n");
5125 seq_printf(m, ":count=%ld\n", count);
5130 static struct ftrace_probe_ops snapshot_probe_ops = {
5131 .func = ftrace_snapshot,
5132 .print = ftrace_snapshot_print,
5135 static struct ftrace_probe_ops snapshot_count_probe_ops = {
5136 .func = ftrace_count_snapshot,
5137 .print = ftrace_snapshot_print,
5141 ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
5142 char *glob, char *cmd, char *param, int enable)
5144 struct ftrace_probe_ops *ops;
5145 void *count = (void *)-1;
5149 /* hash funcs only work with set_ftrace_filter */
5153 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
5155 if (glob[0] == '!') {
5156 unregister_ftrace_function_probe_func(glob+1, ops);
5163 number = strsep(¶m, ":");
5165 if (!strlen(number))
5169 * We use the callback data field (which is a pointer)
5172 ret = kstrtoul(number, 0, (unsigned long *)&count);
5177 ret = register_ftrace_function_probe(glob, ops, count);
5180 alloc_snapshot(&global_trace);
5182 return ret < 0 ? ret : 0;
5185 static struct ftrace_func_command ftrace_snapshot_cmd = {
5187 .func = ftrace_trace_snapshot_callback,
5190 static int register_snapshot_cmd(void)
5192 return register_ftrace_command(&ftrace_snapshot_cmd);
5195 static inline int register_snapshot_cmd(void) { return 0; }
5196 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5198 struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
5205 if (!debugfs_initialized())
5208 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
5209 tr->dir = debugfs_create_dir("tracing", NULL);
5211 if (!tr->dir && !once) {
5213 pr_warning("Could not create debugfs directory 'tracing'\n");
5220 struct dentry *tracing_init_dentry(void)
5222 return tracing_init_dentry_tr(&global_trace);
5225 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
5227 struct dentry *d_tracer;
5230 return tr->percpu_dir;
5232 d_tracer = tracing_init_dentry_tr(tr);
5236 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
5238 WARN_ONCE(!tr->percpu_dir,
5239 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
5241 return tr->percpu_dir;
5245 tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
5247 struct trace_array_cpu *data = per_cpu_ptr(tr->trace_buffer.data, cpu);
5248 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
5249 struct dentry *d_cpu;
5250 char cpu_dir[30]; /* 30 characters should be more than enough */
5255 snprintf(cpu_dir, 30, "cpu%ld", cpu);
5256 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
5258 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
5262 /* per cpu trace_pipe */
5263 trace_create_file("trace_pipe", 0444, d_cpu,
5264 (void *)&data->trace_cpu, &tracing_pipe_fops);
5267 trace_create_file("trace", 0644, d_cpu,
5268 (void *)&data->trace_cpu, &tracing_fops);
5270 trace_create_file("trace_pipe_raw", 0444, d_cpu,
5271 (void *)&data->trace_cpu, &tracing_buffers_fops);
5273 trace_create_file("stats", 0444, d_cpu,
5274 (void *)&data->trace_cpu, &tracing_stats_fops);
5276 trace_create_file("buffer_size_kb", 0444, d_cpu,
5277 (void *)&data->trace_cpu, &tracing_entries_fops);
5279 #ifdef CONFIG_TRACER_SNAPSHOT
5280 trace_create_file("snapshot", 0644, d_cpu,
5281 (void *)&data->trace_cpu, &snapshot_fops);
5283 trace_create_file("snapshot_raw", 0444, d_cpu,
5284 (void *)&data->trace_cpu, &snapshot_raw_fops);
5288 #ifdef CONFIG_FTRACE_SELFTEST
5289 /* Let selftest have access to static functions in this file */
5290 #include "trace_selftest.c"
5293 struct trace_option_dentry {
5294 struct tracer_opt *opt;
5295 struct tracer_flags *flags;
5296 struct trace_array *tr;
5297 struct dentry *entry;
5301 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
5304 struct trace_option_dentry *topt = filp->private_data;
5307 if (topt->flags->val & topt->opt->bit)
5312 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5316 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
5319 struct trace_option_dentry *topt = filp->private_data;
5323 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5327 if (val != 0 && val != 1)
5330 if (!!(topt->flags->val & topt->opt->bit) != val) {
5331 mutex_lock(&trace_types_lock);
5332 ret = __set_tracer_option(topt->tr->current_trace, topt->flags,
5334 mutex_unlock(&trace_types_lock);
5345 static const struct file_operations trace_options_fops = {
5346 .open = tracing_open_generic,
5347 .read = trace_options_read,
5348 .write = trace_options_write,
5349 .llseek = generic_file_llseek,
5353 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5356 long index = (long)filp->private_data;
5359 if (trace_flags & (1 << index))
5364 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5368 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5371 struct trace_array *tr = &global_trace;
5372 long index = (long)filp->private_data;
5376 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5380 if (val != 0 && val != 1)
5383 mutex_lock(&trace_types_lock);
5384 ret = set_tracer_flag(tr, 1 << index, val);
5385 mutex_unlock(&trace_types_lock);
5395 static const struct file_operations trace_options_core_fops = {
5396 .open = tracing_open_generic,
5397 .read = trace_options_core_read,
5398 .write = trace_options_core_write,
5399 .llseek = generic_file_llseek,
5402 struct dentry *trace_create_file(const char *name,
5404 struct dentry *parent,
5406 const struct file_operations *fops)
5410 ret = debugfs_create_file(name, mode, parent, data, fops);
5412 pr_warning("Could not create debugfs '%s' entry\n", name);
5418 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5420 struct dentry *d_tracer;
5425 d_tracer = tracing_init_dentry_tr(tr);
5429 tr->options = debugfs_create_dir("options", d_tracer);
5431 pr_warning("Could not create debugfs directory 'options'\n");
5439 create_trace_option_file(struct trace_array *tr,
5440 struct trace_option_dentry *topt,
5441 struct tracer_flags *flags,
5442 struct tracer_opt *opt)
5444 struct dentry *t_options;
5446 t_options = trace_options_init_dentry(tr);
5450 topt->flags = flags;
5454 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5455 &trace_options_fops);
5459 static struct trace_option_dentry *
5460 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5462 struct trace_option_dentry *topts;
5463 struct tracer_flags *flags;
5464 struct tracer_opt *opts;
5470 flags = tracer->flags;
5472 if (!flags || !flags->opts)
5477 for (cnt = 0; opts[cnt].name; cnt++)
5480 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5484 for (cnt = 0; opts[cnt].name; cnt++)
5485 create_trace_option_file(tr, &topts[cnt], flags,
5492 destroy_trace_option_files(struct trace_option_dentry *topts)
5499 for (cnt = 0; topts[cnt].opt; cnt++) {
5500 if (topts[cnt].entry)
5501 debugfs_remove(topts[cnt].entry);
5507 static struct dentry *
5508 create_trace_option_core_file(struct trace_array *tr,
5509 const char *option, long index)
5511 struct dentry *t_options;
5513 t_options = trace_options_init_dentry(tr);
5517 return trace_create_file(option, 0644, t_options, (void *)index,
5518 &trace_options_core_fops);
5521 static __init void create_trace_options_dir(struct trace_array *tr)
5523 struct dentry *t_options;
5526 t_options = trace_options_init_dentry(tr);
5530 for (i = 0; trace_options[i]; i++)
5531 create_trace_option_core_file(tr, trace_options[i], i);
5535 rb_simple_read(struct file *filp, char __user *ubuf,
5536 size_t cnt, loff_t *ppos)
5538 struct trace_array *tr = filp->private_data;
5539 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5544 r = ring_buffer_record_is_on(buffer);
5548 r = sprintf(buf, "%d\n", r);
5550 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5554 rb_simple_write(struct file *filp, const char __user *ubuf,
5555 size_t cnt, loff_t *ppos)
5557 struct trace_array *tr = filp->private_data;
5558 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5562 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5567 mutex_lock(&trace_types_lock);
5569 ring_buffer_record_on(buffer);
5570 if (tr->current_trace->start)
5571 tr->current_trace->start(tr);
5573 ring_buffer_record_off(buffer);
5574 if (tr->current_trace->stop)
5575 tr->current_trace->stop(tr);
5577 mutex_unlock(&trace_types_lock);
5585 static const struct file_operations rb_simple_fops = {
5586 .open = tracing_open_generic,
5587 .read = rb_simple_read,
5588 .write = rb_simple_write,
5589 .llseek = default_llseek,
5592 struct dentry *trace_instance_dir;
5595 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
5597 static void init_trace_buffers(struct trace_array *tr, struct trace_buffer *buf)
5601 for_each_tracing_cpu(cpu) {
5602 memset(per_cpu_ptr(buf->data, cpu), 0, sizeof(struct trace_array_cpu));
5603 per_cpu_ptr(buf->data, cpu)->trace_cpu.cpu = cpu;
5604 per_cpu_ptr(buf->data, cpu)->trace_cpu.tr = tr;
5609 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
5611 enum ring_buffer_flags rb_flags;
5613 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5615 buf->buffer = ring_buffer_alloc(size, rb_flags);
5619 buf->data = alloc_percpu(struct trace_array_cpu);
5621 ring_buffer_free(buf->buffer);
5625 init_trace_buffers(tr, buf);
5627 /* Allocate the first page for all buffers */
5628 set_buffer_entries(&tr->trace_buffer,
5629 ring_buffer_size(tr->trace_buffer.buffer, 0));
5634 static int allocate_trace_buffers(struct trace_array *tr, int size)
5638 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
5642 #ifdef CONFIG_TRACER_MAX_TRACE
5643 ret = allocate_trace_buffer(tr, &tr->max_buffer,
5644 allocate_snapshot ? size : 1);
5646 ring_buffer_free(tr->trace_buffer.buffer);
5647 free_percpu(tr->trace_buffer.data);
5650 tr->allocated_snapshot = allocate_snapshot;
5653 * Only the top level trace array gets its snapshot allocated
5654 * from the kernel command line.
5656 allocate_snapshot = false;
5661 static int new_instance_create(const char *name)
5663 struct trace_array *tr;
5666 mutex_lock(&trace_types_lock);
5669 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5670 if (tr->name && strcmp(tr->name, name) == 0)
5675 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
5679 tr->name = kstrdup(name, GFP_KERNEL);
5683 raw_spin_lock_init(&tr->start_lock);
5685 tr->current_trace = &nop_trace;
5687 INIT_LIST_HEAD(&tr->systems);
5688 INIT_LIST_HEAD(&tr->events);
5690 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
5693 /* Holder for file callbacks */
5694 tr->trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
5695 tr->trace_cpu.tr = tr;
5697 tr->dir = debugfs_create_dir(name, trace_instance_dir);
5701 ret = event_trace_add_tracer(tr->dir, tr);
5705 init_tracer_debugfs(tr, tr->dir);
5707 list_add(&tr->list, &ftrace_trace_arrays);
5709 mutex_unlock(&trace_types_lock);
5714 if (tr->trace_buffer.buffer)
5715 ring_buffer_free(tr->trace_buffer.buffer);
5720 mutex_unlock(&trace_types_lock);
5726 static int instance_delete(const char *name)
5728 struct trace_array *tr;
5732 mutex_lock(&trace_types_lock);
5735 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5736 if (tr->name && strcmp(tr->name, name) == 0) {
5748 list_del(&tr->list);
5750 event_trace_del_tracer(tr);
5751 debugfs_remove_recursive(tr->dir);
5752 free_percpu(tr->trace_buffer.data);
5753 ring_buffer_free(tr->trace_buffer.buffer);
5761 mutex_unlock(&trace_types_lock);
5766 static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
5768 struct dentry *parent;
5771 /* Paranoid: Make sure the parent is the "instances" directory */
5772 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
5773 if (WARN_ON_ONCE(parent != trace_instance_dir))
5777 * The inode mutex is locked, but debugfs_create_dir() will also
5778 * take the mutex. As the instances directory can not be destroyed
5779 * or changed in any other way, it is safe to unlock it, and
5780 * let the dentry try. If two users try to make the same dir at
5781 * the same time, then the new_instance_create() will determine the
5784 mutex_unlock(&inode->i_mutex);
5786 ret = new_instance_create(dentry->d_iname);
5788 mutex_lock(&inode->i_mutex);
5793 static int instance_rmdir(struct inode *inode, struct dentry *dentry)
5795 struct dentry *parent;
5798 /* Paranoid: Make sure the parent is the "instances" directory */
5799 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
5800 if (WARN_ON_ONCE(parent != trace_instance_dir))
5803 /* The caller did a dget() on dentry */
5804 mutex_unlock(&dentry->d_inode->i_mutex);
5807 * The inode mutex is locked, but debugfs_create_dir() will also
5808 * take the mutex. As the instances directory can not be destroyed
5809 * or changed in any other way, it is safe to unlock it, and
5810 * let the dentry try. If two users try to make the same dir at
5811 * the same time, then the instance_delete() will determine the
5814 mutex_unlock(&inode->i_mutex);
5816 ret = instance_delete(dentry->d_iname);
5818 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
5819 mutex_lock(&dentry->d_inode->i_mutex);
5824 static const struct inode_operations instance_dir_inode_operations = {
5825 .lookup = simple_lookup,
5826 .mkdir = instance_mkdir,
5827 .rmdir = instance_rmdir,
5830 static __init void create_trace_instances(struct dentry *d_tracer)
5832 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
5833 if (WARN_ON(!trace_instance_dir))
5836 /* Hijack the dir inode operations, to allow mkdir */
5837 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
5841 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
5845 trace_create_file("trace_options", 0644, d_tracer,
5846 tr, &tracing_iter_fops);
5848 trace_create_file("trace", 0644, d_tracer,
5849 (void *)&tr->trace_cpu, &tracing_fops);
5851 trace_create_file("trace_pipe", 0444, d_tracer,
5852 (void *)&tr->trace_cpu, &tracing_pipe_fops);
5854 trace_create_file("buffer_size_kb", 0644, d_tracer,
5855 (void *)&tr->trace_cpu, &tracing_entries_fops);
5857 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
5858 tr, &tracing_total_entries_fops);
5860 trace_create_file("free_buffer", 0644, d_tracer,
5861 tr, &tracing_free_buffer_fops);
5863 trace_create_file("trace_marker", 0220, d_tracer,
5864 tr, &tracing_mark_fops);
5866 trace_create_file("trace_clock", 0644, d_tracer, tr,
5869 trace_create_file("tracing_on", 0644, d_tracer,
5870 tr, &rb_simple_fops);
5872 #ifdef CONFIG_TRACER_SNAPSHOT
5873 trace_create_file("snapshot", 0644, d_tracer,
5874 (void *)&tr->trace_cpu, &snapshot_fops);
5877 for_each_tracing_cpu(cpu)
5878 tracing_init_debugfs_percpu(tr, cpu);
5882 static __init int tracer_init_debugfs(void)
5884 struct dentry *d_tracer;
5886 trace_access_lock_init();
5888 d_tracer = tracing_init_dentry();
5890 init_tracer_debugfs(&global_trace, d_tracer);
5892 trace_create_file("tracing_cpumask", 0644, d_tracer,
5893 &global_trace, &tracing_cpumask_fops);
5895 trace_create_file("available_tracers", 0444, d_tracer,
5896 &global_trace, &show_traces_fops);
5898 trace_create_file("current_tracer", 0644, d_tracer,
5899 &global_trace, &set_tracer_fops);
5901 #ifdef CONFIG_TRACER_MAX_TRACE
5902 trace_create_file("tracing_max_latency", 0644, d_tracer,
5903 &tracing_max_latency, &tracing_max_lat_fops);
5906 trace_create_file("tracing_thresh", 0644, d_tracer,
5907 &tracing_thresh, &tracing_max_lat_fops);
5909 trace_create_file("README", 0444, d_tracer,
5910 NULL, &tracing_readme_fops);
5912 trace_create_file("saved_cmdlines", 0444, d_tracer,
5913 NULL, &tracing_saved_cmdlines_fops);
5915 #ifdef CONFIG_DYNAMIC_FTRACE
5916 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
5917 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
5920 create_trace_instances(d_tracer);
5922 create_trace_options_dir(&global_trace);
5927 static int trace_panic_handler(struct notifier_block *this,
5928 unsigned long event, void *unused)
5930 if (ftrace_dump_on_oops)
5931 ftrace_dump(ftrace_dump_on_oops);
5935 static struct notifier_block trace_panic_notifier = {
5936 .notifier_call = trace_panic_handler,
5938 .priority = 150 /* priority: INT_MAX >= x >= 0 */
5941 static int trace_die_handler(struct notifier_block *self,
5947 if (ftrace_dump_on_oops)
5948 ftrace_dump(ftrace_dump_on_oops);
5956 static struct notifier_block trace_die_notifier = {
5957 .notifier_call = trace_die_handler,
5962 * printk is set to max of 1024, we really don't need it that big.
5963 * Nothing should be printing 1000 characters anyway.
5965 #define TRACE_MAX_PRINT 1000
5968 * Define here KERN_TRACE so that we have one place to modify
5969 * it if we decide to change what log level the ftrace dump
5972 #define KERN_TRACE KERN_EMERG
5975 trace_printk_seq(struct trace_seq *s)
5977 /* Probably should print a warning here. */
5981 /* should be zero ended, but we are paranoid. */
5982 s->buffer[s->len] = 0;
5984 printk(KERN_TRACE "%s", s->buffer);
5989 void trace_init_global_iter(struct trace_iterator *iter)
5991 iter->tr = &global_trace;
5992 iter->trace = iter->tr->current_trace;
5993 iter->cpu_file = RING_BUFFER_ALL_CPUS;
5994 iter->trace_buffer = &global_trace.trace_buffer;
5998 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
6000 static arch_spinlock_t ftrace_dump_lock =
6001 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
6002 /* use static because iter can be a bit big for the stack */
6003 static struct trace_iterator iter;
6004 unsigned int old_userobj;
6005 static int dump_ran;
6006 unsigned long flags;
6010 local_irq_save(flags);
6011 arch_spin_lock(&ftrace_dump_lock);
6019 /* Did function tracer already get disabled? */
6020 if (ftrace_is_dead()) {
6021 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6022 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6025 if (disable_tracing)
6028 /* Simulate the iterator */
6029 trace_init_global_iter(&iter);
6031 for_each_tracing_cpu(cpu) {
6032 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
6035 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
6037 /* don't look at user memory in panic mode */
6038 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
6040 switch (oops_dump_mode) {
6042 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6045 iter.cpu_file = raw_smp_processor_id();
6050 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
6051 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6054 printk(KERN_TRACE "Dumping ftrace buffer:\n");
6057 * We need to stop all tracing on all CPUS to read the
6058 * the next buffer. This is a bit expensive, but is
6059 * not done often. We fill all what we can read,
6060 * and then release the locks again.
6063 while (!trace_empty(&iter)) {
6066 printk(KERN_TRACE "---------------------------------\n");
6070 /* reset all but tr, trace, and overruns */
6071 memset(&iter.seq, 0,
6072 sizeof(struct trace_iterator) -
6073 offsetof(struct trace_iterator, seq));
6074 iter.iter_flags |= TRACE_FILE_LAT_FMT;
6077 if (trace_find_next_entry_inc(&iter) != NULL) {
6080 ret = print_trace_line(&iter);
6081 if (ret != TRACE_TYPE_NO_CONSUME)
6082 trace_consume(&iter);
6084 touch_nmi_watchdog();
6086 trace_printk_seq(&iter.seq);
6090 printk(KERN_TRACE " (ftrace buffer empty)\n");
6092 printk(KERN_TRACE "---------------------------------\n");
6095 /* Re-enable tracing if requested */
6096 if (!disable_tracing) {
6097 trace_flags |= old_userobj;
6099 for_each_tracing_cpu(cpu) {
6100 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
6106 arch_spin_unlock(&ftrace_dump_lock);
6107 local_irq_restore(flags);
6110 /* By default: disable tracing after the dump */
6111 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
6113 __ftrace_dump(true, oops_dump_mode);
6115 EXPORT_SYMBOL_GPL(ftrace_dump);
6117 __init static int tracer_alloc_buffers(void)
6123 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
6126 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
6127 goto out_free_buffer_mask;
6129 /* Only allocate trace_printk buffers if a trace_printk exists */
6130 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
6131 /* Must be called before global_trace.buffer is allocated */
6132 trace_printk_init_buffers();
6134 /* To save memory, keep the ring buffer size to its minimum */
6135 if (ring_buffer_expanded)
6136 ring_buf_size = trace_buf_size;
6140 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
6141 cpumask_copy(tracing_cpumask, cpu_all_mask);
6143 raw_spin_lock_init(&global_trace.start_lock);
6145 /* TODO: make the number of buffers hot pluggable with CPUS */
6146 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
6147 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
6149 goto out_free_cpumask;
6152 if (global_trace.buffer_disabled)
6155 trace_init_cmdlines();
6157 register_tracer(&nop_trace);
6159 global_trace.current_trace = &nop_trace;
6161 /* All seems OK, enable tracing */
6162 tracing_disabled = 0;
6164 atomic_notifier_chain_register(&panic_notifier_list,
6165 &trace_panic_notifier);
6167 register_die_notifier(&trace_die_notifier);
6169 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
6171 /* Holder for file callbacks */
6172 global_trace.trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
6173 global_trace.trace_cpu.tr = &global_trace;
6175 INIT_LIST_HEAD(&global_trace.systems);
6176 INIT_LIST_HEAD(&global_trace.events);
6177 list_add(&global_trace.list, &ftrace_trace_arrays);
6179 while (trace_boot_options) {
6182 option = strsep(&trace_boot_options, ",");
6183 trace_set_options(&global_trace, option);
6186 register_snapshot_cmd();
6191 free_percpu(global_trace.trace_buffer.data);
6192 #ifdef CONFIG_TRACER_MAX_TRACE
6193 free_percpu(global_trace.max_buffer.data);
6195 free_cpumask_var(tracing_cpumask);
6196 out_free_buffer_mask:
6197 free_cpumask_var(tracing_buffer_mask);
6202 __init static int clear_boot_tracer(void)
6205 * The default tracer at boot buffer is an init section.
6206 * This function is called in lateinit. If we did not
6207 * find the boot tracer, then clear it out, to prevent
6208 * later registration from accessing the buffer that is
6209 * about to be freed.
6211 if (!default_bootup_tracer)
6214 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
6215 default_bootup_tracer);
6216 default_bootup_tracer = NULL;
6221 early_initcall(tracer_alloc_buffers);
6222 fs_initcall(tracer_init_debugfs);
6223 late_initcall(clear_boot_tracer);