2 * ring buffer based function tracer
4 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 Nadia Yvette Chambers
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/nmi.h>
41 #include <linux/sched/rt.h>
44 #include "trace_output.h"
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
50 bool ring_buffer_expanded;
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
59 static bool __read_mostly tracing_selftest_running;
62 * If a tracer is running, we do not want to run SELFTEST.
64 bool __read_mostly tracing_selftest_disabled;
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt[] = {
71 static struct tracer_flags dummy_tracer_flags = {
73 .opts = dummy_tracer_opt
76 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
82 * To prevent the comm cache from being overwritten when no
83 * tracing is active, only save the comm when a trace event
86 static DEFINE_PER_CPU(bool, trace_cmdline_save);
89 * Kill all tracing for good (never come back).
90 * It is initialized to 1 but will turn to zero if the initialization
91 * of the tracer is successful. But that is the only place that sets
94 static int tracing_disabled = 1;
96 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
98 cpumask_var_t __read_mostly tracing_buffer_mask;
101 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
103 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
104 * is set, then ftrace_dump is called. This will output the contents
105 * of the ftrace buffers to the console. This is very useful for
106 * capturing traces that lead to crashes and outputing it to a
109 * It is default off, but you can enable it with either specifying
110 * "ftrace_dump_on_oops" in the kernel command line, or setting
111 * /proc/sys/kernel/ftrace_dump_on_oops
112 * Set 1 if you want to dump buffers of all CPUs
113 * Set 2 if you want to dump the buffer of the CPU that triggered oops
116 enum ftrace_dump_mode ftrace_dump_on_oops;
118 static int tracing_set_tracer(const char *buf);
120 #define MAX_TRACER_SIZE 100
121 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
122 static char *default_bootup_tracer;
124 static bool allocate_snapshot;
126 static int __init set_cmdline_ftrace(char *str)
128 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
129 default_bootup_tracer = bootup_tracer_buf;
130 /* We are using ftrace early, expand it */
131 ring_buffer_expanded = true;
134 __setup("ftrace=", set_cmdline_ftrace);
136 static int __init set_ftrace_dump_on_oops(char *str)
138 if (*str++ != '=' || !*str) {
139 ftrace_dump_on_oops = DUMP_ALL;
143 if (!strcmp("orig_cpu", str)) {
144 ftrace_dump_on_oops = DUMP_ORIG;
150 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
152 static int __init boot_alloc_snapshot(char *str)
154 allocate_snapshot = true;
155 /* We also need the main ring buffer expanded */
156 ring_buffer_expanded = true;
159 __setup("alloc_snapshot", boot_alloc_snapshot);
162 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
163 static char *trace_boot_options __initdata;
165 static int __init set_trace_boot_options(char *str)
167 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
168 trace_boot_options = trace_boot_options_buf;
171 __setup("trace_options=", set_trace_boot_options);
173 unsigned long long ns2usecs(cycle_t nsec)
181 * The global_trace is the descriptor that holds the tracing
182 * buffers for the live tracing. For each CPU, it contains
183 * a link list of pages that will store trace entries. The
184 * page descriptor of the pages in the memory is used to hold
185 * the link list by linking the lru item in the page descriptor
186 * to each of the pages in the buffer per CPU.
188 * For each active CPU there is a data field that holds the
189 * pages for the buffer for that CPU. Each CPU has the same number
190 * of pages allocated for its buffer.
192 static struct trace_array global_trace;
194 LIST_HEAD(ftrace_trace_arrays);
196 int 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 | TRACE_ITER_FUNCTION;
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[] = {
645 int in_ns; /* is this clock in nanoseconds? */
647 { trace_clock_local, "local", 1 },
648 { trace_clock_global, "global", 1 },
649 { trace_clock_counter, "counter", 0 },
650 { trace_clock_jiffies, "uptime", 1 },
651 { trace_clock, "perf", 1 },
656 * trace_parser_get_init - gets the buffer for trace parser
658 int trace_parser_get_init(struct trace_parser *parser, int size)
660 memset(parser, 0, sizeof(*parser));
662 parser->buffer = kmalloc(size, GFP_KERNEL);
671 * trace_parser_put - frees the buffer for trace parser
673 void trace_parser_put(struct trace_parser *parser)
675 kfree(parser->buffer);
679 * trace_get_user - reads the user input string separated by space
680 * (matched by isspace(ch))
682 * For each string found the 'struct trace_parser' is updated,
683 * and the function returns.
685 * Returns number of bytes read.
687 * See kernel/trace/trace.h for 'struct trace_parser' details.
689 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
690 size_t cnt, loff_t *ppos)
697 trace_parser_clear(parser);
699 ret = get_user(ch, ubuf++);
707 * The parser is not finished with the last write,
708 * continue reading the user input without skipping spaces.
711 /* skip white space */
712 while (cnt && isspace(ch)) {
713 ret = get_user(ch, ubuf++);
720 /* only spaces were written */
730 /* read the non-space input */
731 while (cnt && !isspace(ch)) {
732 if (parser->idx < parser->size - 1)
733 parser->buffer[parser->idx++] = ch;
738 ret = get_user(ch, ubuf++);
745 /* We either got finished input or we have to wait for another call. */
747 parser->buffer[parser->idx] = 0;
748 parser->cont = false;
751 parser->buffer[parser->idx++] = ch;
761 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
769 if (s->len <= s->readpos)
772 len = s->len - s->readpos;
775 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
785 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
789 if (s->len <= s->readpos)
792 len = s->len - s->readpos;
795 memcpy(buf, s->buffer + s->readpos, cnt);
802 * ftrace_max_lock is used to protect the swapping of buffers
803 * when taking a max snapshot. The buffers themselves are
804 * protected by per_cpu spinlocks. But the action of the swap
805 * needs its own lock.
807 * This is defined as a arch_spinlock_t in order to help
808 * with performance when lockdep debugging is enabled.
810 * It is also used in other places outside the update_max_tr
811 * so it needs to be defined outside of the
812 * CONFIG_TRACER_MAX_TRACE.
814 static arch_spinlock_t ftrace_max_lock =
815 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
817 unsigned long __read_mostly tracing_thresh;
819 #ifdef CONFIG_TRACER_MAX_TRACE
820 unsigned long __read_mostly tracing_max_latency;
823 * Copy the new maximum trace into the separate maximum-trace
824 * structure. (this way the maximum trace is permanently saved,
825 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
828 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
830 struct trace_buffer *trace_buf = &tr->trace_buffer;
831 struct trace_buffer *max_buf = &tr->max_buffer;
832 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
833 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
836 max_buf->time_start = data->preempt_timestamp;
838 max_data->saved_latency = tracing_max_latency;
839 max_data->critical_start = data->critical_start;
840 max_data->critical_end = data->critical_end;
842 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
843 max_data->pid = tsk->pid;
845 * If tsk == current, then use current_uid(), as that does not use
846 * RCU. The irq tracer can be called out of RCU scope.
849 max_data->uid = current_uid();
851 max_data->uid = task_uid(tsk);
853 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
854 max_data->policy = tsk->policy;
855 max_data->rt_priority = tsk->rt_priority;
857 /* record this tasks comm */
858 tracing_record_cmdline(tsk);
862 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
864 * @tsk: the task with the latency
865 * @cpu: The cpu that initiated the trace.
867 * Flip the buffers between the @tr and the max_tr and record information
868 * about which task was the cause of this latency.
871 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
873 struct ring_buffer *buf;
878 WARN_ON_ONCE(!irqs_disabled());
880 if (!tr->allocated_snapshot) {
881 /* Only the nop tracer should hit this when disabling */
882 WARN_ON_ONCE(tr->current_trace != &nop_trace);
886 arch_spin_lock(&ftrace_max_lock);
888 buf = tr->trace_buffer.buffer;
889 tr->trace_buffer.buffer = tr->max_buffer.buffer;
890 tr->max_buffer.buffer = buf;
892 __update_max_tr(tr, tsk, cpu);
893 arch_spin_unlock(&ftrace_max_lock);
897 * update_max_tr_single - only copy one trace over, and reset the rest
899 * @tsk - task with the latency
900 * @cpu - the cpu of the buffer to copy.
902 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
905 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
912 WARN_ON_ONCE(!irqs_disabled());
913 if (!tr->allocated_snapshot) {
914 /* Only the nop tracer should hit this when disabling */
915 WARN_ON_ONCE(tr->current_trace != &nop_trace);
919 arch_spin_lock(&ftrace_max_lock);
921 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
925 * We failed to swap the buffer due to a commit taking
926 * place on this CPU. We fail to record, but we reset
927 * the max trace buffer (no one writes directly to it)
928 * and flag that it failed.
930 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
931 "Failed to swap buffers due to commit in progress\n");
934 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
936 __update_max_tr(tr, tsk, cpu);
937 arch_spin_unlock(&ftrace_max_lock);
939 #endif /* CONFIG_TRACER_MAX_TRACE */
941 static void default_wait_pipe(struct trace_iterator *iter)
943 /* Iterators are static, they should be filled or empty */
944 if (trace_buffer_iter(iter, iter->cpu_file))
947 ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file);
950 #ifdef CONFIG_FTRACE_STARTUP_TEST
951 static int run_tracer_selftest(struct tracer *type)
953 struct trace_array *tr = &global_trace;
954 struct tracer *saved_tracer = tr->current_trace;
957 if (!type->selftest || tracing_selftest_disabled)
961 * Run a selftest on this tracer.
962 * Here we reset the trace buffer, and set the current
963 * tracer to be this tracer. The tracer can then run some
964 * internal tracing to verify that everything is in order.
965 * If we fail, we do not register this tracer.
967 tracing_reset_online_cpus(&tr->trace_buffer);
969 tr->current_trace = type;
971 #ifdef CONFIG_TRACER_MAX_TRACE
972 if (type->use_max_tr) {
973 /* If we expanded the buffers, make sure the max is expanded too */
974 if (ring_buffer_expanded)
975 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
976 RING_BUFFER_ALL_CPUS);
977 tr->allocated_snapshot = true;
981 /* the test is responsible for initializing and enabling */
982 pr_info("Testing tracer %s: ", type->name);
983 ret = type->selftest(type, tr);
984 /* the test is responsible for resetting too */
985 tr->current_trace = saved_tracer;
987 printk(KERN_CONT "FAILED!\n");
988 /* Add the warning after printing 'FAILED' */
992 /* Only reset on passing, to avoid touching corrupted buffers */
993 tracing_reset_online_cpus(&tr->trace_buffer);
995 #ifdef CONFIG_TRACER_MAX_TRACE
996 if (type->use_max_tr) {
997 tr->allocated_snapshot = false;
999 /* Shrink the max buffer again */
1000 if (ring_buffer_expanded)
1001 ring_buffer_resize(tr->max_buffer.buffer, 1,
1002 RING_BUFFER_ALL_CPUS);
1006 printk(KERN_CONT "PASSED\n");
1010 static inline int run_tracer_selftest(struct tracer *type)
1014 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1017 * register_tracer - register a tracer with the ftrace system.
1018 * @type - the plugin for the tracer
1020 * Register a new plugin tracer.
1022 int register_tracer(struct tracer *type)
1028 pr_info("Tracer must have a name\n");
1032 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1033 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1037 mutex_lock(&trace_types_lock);
1039 tracing_selftest_running = true;
1041 for (t = trace_types; t; t = t->next) {
1042 if (strcmp(type->name, t->name) == 0) {
1044 pr_info("Tracer %s already registered\n",
1051 if (!type->set_flag)
1052 type->set_flag = &dummy_set_flag;
1054 type->flags = &dummy_tracer_flags;
1056 if (!type->flags->opts)
1057 type->flags->opts = dummy_tracer_opt;
1058 if (!type->wait_pipe)
1059 type->wait_pipe = default_wait_pipe;
1061 ret = run_tracer_selftest(type);
1065 type->next = trace_types;
1069 tracing_selftest_running = false;
1070 mutex_unlock(&trace_types_lock);
1072 if (ret || !default_bootup_tracer)
1075 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1078 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1079 /* Do we want this tracer to start on bootup? */
1080 tracing_set_tracer(type->name);
1081 default_bootup_tracer = NULL;
1082 /* disable other selftests, since this will break it. */
1083 tracing_selftest_disabled = true;
1084 #ifdef CONFIG_FTRACE_STARTUP_TEST
1085 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1093 void tracing_reset(struct trace_buffer *buf, int cpu)
1095 struct ring_buffer *buffer = buf->buffer;
1100 ring_buffer_record_disable(buffer);
1102 /* Make sure all commits have finished */
1103 synchronize_sched();
1104 ring_buffer_reset_cpu(buffer, cpu);
1106 ring_buffer_record_enable(buffer);
1109 void tracing_reset_online_cpus(struct trace_buffer *buf)
1111 struct ring_buffer *buffer = buf->buffer;
1117 ring_buffer_record_disable(buffer);
1119 /* Make sure all commits have finished */
1120 synchronize_sched();
1122 buf->time_start = ftrace_now(buf->cpu);
1124 for_each_online_cpu(cpu)
1125 ring_buffer_reset_cpu(buffer, cpu);
1127 ring_buffer_record_enable(buffer);
1130 void tracing_reset_current(int cpu)
1132 tracing_reset(&global_trace.trace_buffer, cpu);
1135 void tracing_reset_all_online_cpus(void)
1137 struct trace_array *tr;
1139 mutex_lock(&trace_types_lock);
1140 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1141 tracing_reset_online_cpus(&tr->trace_buffer);
1142 #ifdef CONFIG_TRACER_MAX_TRACE
1143 tracing_reset_online_cpus(&tr->max_buffer);
1146 mutex_unlock(&trace_types_lock);
1149 #define SAVED_CMDLINES 128
1150 #define NO_CMDLINE_MAP UINT_MAX
1151 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1152 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
1153 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
1154 static int cmdline_idx;
1155 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1157 /* temporary disable recording */
1158 static atomic_t trace_record_cmdline_disabled __read_mostly;
1160 static void trace_init_cmdlines(void)
1162 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
1163 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
1167 int is_tracing_stopped(void)
1169 return global_trace.stop_count;
1173 * ftrace_off_permanent - disable all ftrace code permanently
1175 * This should only be called when a serious anomally has
1176 * been detected. This will turn off the function tracing,
1177 * ring buffers, and other tracing utilites. It takes no
1178 * locks and can be called from any context.
1180 void ftrace_off_permanent(void)
1182 tracing_disabled = 1;
1184 tracing_off_permanent();
1188 * tracing_start - quick start of the tracer
1190 * If tracing is enabled but was stopped by tracing_stop,
1191 * this will start the tracer back up.
1193 void tracing_start(void)
1195 struct ring_buffer *buffer;
1196 unsigned long flags;
1198 if (tracing_disabled)
1201 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1202 if (--global_trace.stop_count) {
1203 if (global_trace.stop_count < 0) {
1204 /* Someone screwed up their debugging */
1206 global_trace.stop_count = 0;
1211 /* Prevent the buffers from switching */
1212 arch_spin_lock(&ftrace_max_lock);
1214 buffer = global_trace.trace_buffer.buffer;
1216 ring_buffer_record_enable(buffer);
1218 #ifdef CONFIG_TRACER_MAX_TRACE
1219 buffer = global_trace.max_buffer.buffer;
1221 ring_buffer_record_enable(buffer);
1224 arch_spin_unlock(&ftrace_max_lock);
1228 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1231 static void tracing_start_tr(struct trace_array *tr)
1233 struct ring_buffer *buffer;
1234 unsigned long flags;
1236 if (tracing_disabled)
1239 /* If global, we need to also start the max tracer */
1240 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1241 return tracing_start();
1243 raw_spin_lock_irqsave(&tr->start_lock, flags);
1245 if (--tr->stop_count) {
1246 if (tr->stop_count < 0) {
1247 /* Someone screwed up their debugging */
1254 buffer = tr->trace_buffer.buffer;
1256 ring_buffer_record_enable(buffer);
1259 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1263 * tracing_stop - quick stop of the tracer
1265 * Light weight way to stop tracing. Use in conjunction with
1268 void tracing_stop(void)
1270 struct ring_buffer *buffer;
1271 unsigned long flags;
1274 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1275 if (global_trace.stop_count++)
1278 /* Prevent the buffers from switching */
1279 arch_spin_lock(&ftrace_max_lock);
1281 buffer = global_trace.trace_buffer.buffer;
1283 ring_buffer_record_disable(buffer);
1285 #ifdef CONFIG_TRACER_MAX_TRACE
1286 buffer = global_trace.max_buffer.buffer;
1288 ring_buffer_record_disable(buffer);
1291 arch_spin_unlock(&ftrace_max_lock);
1294 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1297 static void tracing_stop_tr(struct trace_array *tr)
1299 struct ring_buffer *buffer;
1300 unsigned long flags;
1302 /* If global, we need to also stop the max tracer */
1303 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1304 return tracing_stop();
1306 raw_spin_lock_irqsave(&tr->start_lock, flags);
1307 if (tr->stop_count++)
1310 buffer = tr->trace_buffer.buffer;
1312 ring_buffer_record_disable(buffer);
1315 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1318 void trace_stop_cmdline_recording(void);
1320 static void trace_save_cmdline(struct task_struct *tsk)
1324 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1328 * It's not the end of the world if we don't get
1329 * the lock, but we also don't want to spin
1330 * nor do we want to disable interrupts,
1331 * so if we miss here, then better luck next time.
1333 if (!arch_spin_trylock(&trace_cmdline_lock))
1336 idx = map_pid_to_cmdline[tsk->pid];
1337 if (idx == NO_CMDLINE_MAP) {
1338 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1341 * Check whether the cmdline buffer at idx has a pid
1342 * mapped. We are going to overwrite that entry so we
1343 * need to clear the map_pid_to_cmdline. Otherwise we
1344 * would read the new comm for the old pid.
1346 pid = map_cmdline_to_pid[idx];
1347 if (pid != NO_CMDLINE_MAP)
1348 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1350 map_cmdline_to_pid[idx] = tsk->pid;
1351 map_pid_to_cmdline[tsk->pid] = idx;
1356 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1358 arch_spin_unlock(&trace_cmdline_lock);
1361 void trace_find_cmdline(int pid, char comm[])
1366 strcpy(comm, "<idle>");
1370 if (WARN_ON_ONCE(pid < 0)) {
1371 strcpy(comm, "<XXX>");
1375 if (pid > PID_MAX_DEFAULT) {
1376 strcpy(comm, "<...>");
1381 arch_spin_lock(&trace_cmdline_lock);
1382 map = map_pid_to_cmdline[pid];
1383 if (map != NO_CMDLINE_MAP)
1384 strcpy(comm, saved_cmdlines[map]);
1386 strcpy(comm, "<...>");
1388 arch_spin_unlock(&trace_cmdline_lock);
1392 void tracing_record_cmdline(struct task_struct *tsk)
1394 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1397 if (!__this_cpu_read(trace_cmdline_save))
1400 __this_cpu_write(trace_cmdline_save, false);
1402 trace_save_cmdline(tsk);
1406 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1409 struct task_struct *tsk = current;
1411 entry->preempt_count = pc & 0xff;
1412 entry->pid = (tsk) ? tsk->pid : 0;
1414 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1415 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1417 TRACE_FLAG_IRQS_NOSUPPORT |
1419 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1420 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1421 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1423 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1425 struct ring_buffer_event *
1426 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1429 unsigned long flags, int pc)
1431 struct ring_buffer_event *event;
1433 event = ring_buffer_lock_reserve(buffer, len);
1434 if (event != NULL) {
1435 struct trace_entry *ent = ring_buffer_event_data(event);
1437 tracing_generic_entry_update(ent, flags, pc);
1445 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1447 __this_cpu_write(trace_cmdline_save, true);
1448 ring_buffer_unlock_commit(buffer, event);
1452 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1453 struct ring_buffer_event *event,
1454 unsigned long flags, int pc)
1456 __buffer_unlock_commit(buffer, event);
1458 ftrace_trace_stack(buffer, flags, 6, pc);
1459 ftrace_trace_userstack(buffer, flags, pc);
1462 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1463 struct ring_buffer_event *event,
1464 unsigned long flags, int pc)
1466 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1468 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1470 struct ring_buffer_event *
1471 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1472 struct ftrace_event_file *ftrace_file,
1473 int type, unsigned long len,
1474 unsigned long flags, int pc)
1476 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1477 return trace_buffer_lock_reserve(*current_rb,
1478 type, len, flags, pc);
1480 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1482 struct ring_buffer_event *
1483 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1484 int type, unsigned long len,
1485 unsigned long flags, int pc)
1487 *current_rb = global_trace.trace_buffer.buffer;
1488 return trace_buffer_lock_reserve(*current_rb,
1489 type, len, flags, pc);
1491 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1493 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1494 struct ring_buffer_event *event,
1495 unsigned long flags, int pc)
1497 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1499 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1501 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1502 struct ring_buffer_event *event,
1503 unsigned long flags, int pc,
1504 struct pt_regs *regs)
1506 __buffer_unlock_commit(buffer, event);
1508 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1509 ftrace_trace_userstack(buffer, flags, pc);
1511 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1513 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1514 struct ring_buffer_event *event)
1516 ring_buffer_discard_commit(buffer, event);
1518 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1521 trace_function(struct trace_array *tr,
1522 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1525 struct ftrace_event_call *call = &event_function;
1526 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1527 struct ring_buffer_event *event;
1528 struct ftrace_entry *entry;
1530 /* If we are reading the ring buffer, don't trace */
1531 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1534 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1538 entry = ring_buffer_event_data(event);
1540 entry->parent_ip = parent_ip;
1542 if (!filter_check_discard(call, entry, buffer, event))
1543 __buffer_unlock_commit(buffer, event);
1547 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1548 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1551 if (likely(!atomic_read(&data->disabled)))
1552 trace_function(tr, ip, parent_ip, flags, pc);
1555 #ifdef CONFIG_STACKTRACE
1557 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1558 struct ftrace_stack {
1559 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1562 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1563 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1565 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1566 unsigned long flags,
1567 int skip, int pc, struct pt_regs *regs)
1569 struct ftrace_event_call *call = &event_kernel_stack;
1570 struct ring_buffer_event *event;
1571 struct stack_entry *entry;
1572 struct stack_trace trace;
1574 int size = FTRACE_STACK_ENTRIES;
1576 trace.nr_entries = 0;
1580 * Since events can happen in NMIs there's no safe way to
1581 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1582 * or NMI comes in, it will just have to use the default
1583 * FTRACE_STACK_SIZE.
1585 preempt_disable_notrace();
1587 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1589 * We don't need any atomic variables, just a barrier.
1590 * If an interrupt comes in, we don't care, because it would
1591 * have exited and put the counter back to what we want.
1592 * We just need a barrier to keep gcc from moving things
1596 if (use_stack == 1) {
1597 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1598 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1601 save_stack_trace_regs(regs, &trace);
1603 save_stack_trace(&trace);
1605 if (trace.nr_entries > size)
1606 size = trace.nr_entries;
1608 /* From now on, use_stack is a boolean */
1611 size *= sizeof(unsigned long);
1613 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1614 sizeof(*entry) + size, flags, pc);
1617 entry = ring_buffer_event_data(event);
1619 memset(&entry->caller, 0, size);
1622 memcpy(&entry->caller, trace.entries,
1623 trace.nr_entries * sizeof(unsigned long));
1625 trace.max_entries = FTRACE_STACK_ENTRIES;
1626 trace.entries = entry->caller;
1628 save_stack_trace_regs(regs, &trace);
1630 save_stack_trace(&trace);
1633 entry->size = trace.nr_entries;
1635 if (!filter_check_discard(call, entry, buffer, event))
1636 __buffer_unlock_commit(buffer, event);
1639 /* Again, don't let gcc optimize things here */
1641 __this_cpu_dec(ftrace_stack_reserve);
1642 preempt_enable_notrace();
1646 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1647 int skip, int pc, struct pt_regs *regs)
1649 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1652 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1655 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1658 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1661 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1664 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1667 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1671 * trace_dump_stack - record a stack back trace in the trace buffer
1672 * @skip: Number of functions to skip (helper handlers)
1674 void trace_dump_stack(int skip)
1676 unsigned long flags;
1678 if (tracing_disabled || tracing_selftest_running)
1681 local_save_flags(flags);
1684 * Skip 3 more, seems to get us at the caller of
1688 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
1689 flags, skip, preempt_count(), NULL);
1692 static DEFINE_PER_CPU(int, user_stack_count);
1695 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1697 struct ftrace_event_call *call = &event_user_stack;
1698 struct ring_buffer_event *event;
1699 struct userstack_entry *entry;
1700 struct stack_trace trace;
1702 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1706 * NMIs can not handle page faults, even with fix ups.
1707 * The save user stack can (and often does) fault.
1709 if (unlikely(in_nmi()))
1713 * prevent recursion, since the user stack tracing may
1714 * trigger other kernel events.
1717 if (__this_cpu_read(user_stack_count))
1720 __this_cpu_inc(user_stack_count);
1722 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1723 sizeof(*entry), flags, pc);
1725 goto out_drop_count;
1726 entry = ring_buffer_event_data(event);
1728 entry->tgid = current->tgid;
1729 memset(&entry->caller, 0, sizeof(entry->caller));
1731 trace.nr_entries = 0;
1732 trace.max_entries = FTRACE_STACK_ENTRIES;
1734 trace.entries = entry->caller;
1736 save_stack_trace_user(&trace);
1737 if (!filter_check_discard(call, entry, buffer, event))
1738 __buffer_unlock_commit(buffer, event);
1741 __this_cpu_dec(user_stack_count);
1747 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1749 ftrace_trace_userstack(tr, flags, preempt_count());
1753 #endif /* CONFIG_STACKTRACE */
1755 /* created for use with alloc_percpu */
1756 struct trace_buffer_struct {
1757 char buffer[TRACE_BUF_SIZE];
1760 static struct trace_buffer_struct *trace_percpu_buffer;
1761 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1762 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1763 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1766 * The buffer used is dependent on the context. There is a per cpu
1767 * buffer for normal context, softirq contex, hard irq context and
1768 * for NMI context. Thise allows for lockless recording.
1770 * Note, if the buffers failed to be allocated, then this returns NULL
1772 static char *get_trace_buf(void)
1774 struct trace_buffer_struct *percpu_buffer;
1777 * If we have allocated per cpu buffers, then we do not
1778 * need to do any locking.
1781 percpu_buffer = trace_percpu_nmi_buffer;
1783 percpu_buffer = trace_percpu_irq_buffer;
1784 else if (in_softirq())
1785 percpu_buffer = trace_percpu_sirq_buffer;
1787 percpu_buffer = trace_percpu_buffer;
1792 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1795 static int alloc_percpu_trace_buffer(void)
1797 struct trace_buffer_struct *buffers;
1798 struct trace_buffer_struct *sirq_buffers;
1799 struct trace_buffer_struct *irq_buffers;
1800 struct trace_buffer_struct *nmi_buffers;
1802 buffers = alloc_percpu(struct trace_buffer_struct);
1806 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1810 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1814 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1818 trace_percpu_buffer = buffers;
1819 trace_percpu_sirq_buffer = sirq_buffers;
1820 trace_percpu_irq_buffer = irq_buffers;
1821 trace_percpu_nmi_buffer = nmi_buffers;
1826 free_percpu(irq_buffers);
1828 free_percpu(sirq_buffers);
1830 free_percpu(buffers);
1832 WARN(1, "Could not allocate percpu trace_printk buffer");
1836 static int buffers_allocated;
1838 void trace_printk_init_buffers(void)
1840 if (buffers_allocated)
1843 if (alloc_percpu_trace_buffer())
1846 pr_info("ftrace: Allocated trace_printk buffers\n");
1848 /* Expand the buffers to set size */
1849 tracing_update_buffers();
1851 buffers_allocated = 1;
1854 * trace_printk_init_buffers() can be called by modules.
1855 * If that happens, then we need to start cmdline recording
1856 * directly here. If the global_trace.buffer is already
1857 * allocated here, then this was called by module code.
1859 if (global_trace.trace_buffer.buffer)
1860 tracing_start_cmdline_record();
1863 void trace_printk_start_comm(void)
1865 /* Start tracing comms if trace printk is set */
1866 if (!buffers_allocated)
1868 tracing_start_cmdline_record();
1871 static void trace_printk_start_stop_comm(int enabled)
1873 if (!buffers_allocated)
1877 tracing_start_cmdline_record();
1879 tracing_stop_cmdline_record();
1883 * trace_vbprintk - write binary msg to tracing buffer
1886 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1888 struct ftrace_event_call *call = &event_bprint;
1889 struct ring_buffer_event *event;
1890 struct ring_buffer *buffer;
1891 struct trace_array *tr = &global_trace;
1892 struct bprint_entry *entry;
1893 unsigned long flags;
1895 int len = 0, size, pc;
1897 if (unlikely(tracing_selftest_running || tracing_disabled))
1900 /* Don't pollute graph traces with trace_vprintk internals */
1901 pause_graph_tracing();
1903 pc = preempt_count();
1904 preempt_disable_notrace();
1906 tbuffer = get_trace_buf();
1912 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1914 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
1917 local_save_flags(flags);
1918 size = sizeof(*entry) + sizeof(u32) * len;
1919 buffer = tr->trace_buffer.buffer;
1920 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1924 entry = ring_buffer_event_data(event);
1928 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
1929 if (!filter_check_discard(call, entry, buffer, event)) {
1930 __buffer_unlock_commit(buffer, event);
1931 ftrace_trace_stack(buffer, flags, 6, pc);
1935 preempt_enable_notrace();
1936 unpause_graph_tracing();
1940 EXPORT_SYMBOL_GPL(trace_vbprintk);
1943 __trace_array_vprintk(struct ring_buffer *buffer,
1944 unsigned long ip, const char *fmt, va_list args)
1946 struct ftrace_event_call *call = &event_print;
1947 struct ring_buffer_event *event;
1948 int len = 0, size, pc;
1949 struct print_entry *entry;
1950 unsigned long flags;
1953 if (tracing_disabled || tracing_selftest_running)
1956 /* Don't pollute graph traces with trace_vprintk internals */
1957 pause_graph_tracing();
1959 pc = preempt_count();
1960 preempt_disable_notrace();
1963 tbuffer = get_trace_buf();
1969 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
1970 if (len > TRACE_BUF_SIZE)
1973 local_save_flags(flags);
1974 size = sizeof(*entry) + len + 1;
1975 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1979 entry = ring_buffer_event_data(event);
1982 memcpy(&entry->buf, tbuffer, len);
1983 entry->buf[len] = '\0';
1984 if (!filter_check_discard(call, entry, buffer, event)) {
1985 __buffer_unlock_commit(buffer, event);
1986 ftrace_trace_stack(buffer, flags, 6, pc);
1989 preempt_enable_notrace();
1990 unpause_graph_tracing();
1995 int trace_array_vprintk(struct trace_array *tr,
1996 unsigned long ip, const char *fmt, va_list args)
1998 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
2001 int trace_array_printk(struct trace_array *tr,
2002 unsigned long ip, const char *fmt, ...)
2007 if (!(trace_flags & TRACE_ITER_PRINTK))
2011 ret = trace_array_vprintk(tr, ip, fmt, ap);
2016 int trace_array_printk_buf(struct ring_buffer *buffer,
2017 unsigned long ip, const char *fmt, ...)
2022 if (!(trace_flags & TRACE_ITER_PRINTK))
2026 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2031 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2033 return trace_array_vprintk(&global_trace, ip, fmt, args);
2035 EXPORT_SYMBOL_GPL(trace_vprintk);
2037 static void trace_iterator_increment(struct trace_iterator *iter)
2039 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2043 ring_buffer_read(buf_iter, NULL);
2046 static struct trace_entry *
2047 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2048 unsigned long *lost_events)
2050 struct ring_buffer_event *event;
2051 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2054 event = ring_buffer_iter_peek(buf_iter, ts);
2056 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2060 iter->ent_size = ring_buffer_event_length(event);
2061 return ring_buffer_event_data(event);
2067 static struct trace_entry *
2068 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2069 unsigned long *missing_events, u64 *ent_ts)
2071 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2072 struct trace_entry *ent, *next = NULL;
2073 unsigned long lost_events = 0, next_lost = 0;
2074 int cpu_file = iter->cpu_file;
2075 u64 next_ts = 0, ts;
2081 * If we are in a per_cpu trace file, don't bother by iterating over
2082 * all cpu and peek directly.
2084 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2085 if (ring_buffer_empty_cpu(buffer, cpu_file))
2087 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2089 *ent_cpu = cpu_file;
2094 for_each_tracing_cpu(cpu) {
2096 if (ring_buffer_empty_cpu(buffer, cpu))
2099 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2102 * Pick the entry with the smallest timestamp:
2104 if (ent && (!next || ts < next_ts)) {
2108 next_lost = lost_events;
2109 next_size = iter->ent_size;
2113 iter->ent_size = next_size;
2116 *ent_cpu = next_cpu;
2122 *missing_events = next_lost;
2127 /* Find the next real entry, without updating the iterator itself */
2128 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2129 int *ent_cpu, u64 *ent_ts)
2131 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2134 /* Find the next real entry, and increment the iterator to the next entry */
2135 void *trace_find_next_entry_inc(struct trace_iterator *iter)
2137 iter->ent = __find_next_entry(iter, &iter->cpu,
2138 &iter->lost_events, &iter->ts);
2141 trace_iterator_increment(iter);
2143 return iter->ent ? iter : NULL;
2146 static void trace_consume(struct trace_iterator *iter)
2148 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2149 &iter->lost_events);
2152 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2154 struct trace_iterator *iter = m->private;
2158 WARN_ON_ONCE(iter->leftover);
2162 /* can't go backwards */
2167 ent = trace_find_next_entry_inc(iter);
2171 while (ent && iter->idx < i)
2172 ent = trace_find_next_entry_inc(iter);
2179 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2181 struct ring_buffer_event *event;
2182 struct ring_buffer_iter *buf_iter;
2183 unsigned long entries = 0;
2186 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2188 buf_iter = trace_buffer_iter(iter, cpu);
2192 ring_buffer_iter_reset(buf_iter);
2195 * We could have the case with the max latency tracers
2196 * that a reset never took place on a cpu. This is evident
2197 * by the timestamp being before the start of the buffer.
2199 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2200 if (ts >= iter->trace_buffer->time_start)
2203 ring_buffer_read(buf_iter, NULL);
2206 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2210 * The current tracer is copied to avoid a global locking
2213 static void *s_start(struct seq_file *m, loff_t *pos)
2215 struct trace_iterator *iter = m->private;
2216 struct trace_array *tr = iter->tr;
2217 int cpu_file = iter->cpu_file;
2223 * copy the tracer to avoid using a global lock all around.
2224 * iter->trace is a copy of current_trace, the pointer to the
2225 * name may be used instead of a strcmp(), as iter->trace->name
2226 * will point to the same string as current_trace->name.
2228 mutex_lock(&trace_types_lock);
2229 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2230 *iter->trace = *tr->current_trace;
2231 mutex_unlock(&trace_types_lock);
2233 #ifdef CONFIG_TRACER_MAX_TRACE
2234 if (iter->snapshot && iter->trace->use_max_tr)
2235 return ERR_PTR(-EBUSY);
2238 if (!iter->snapshot)
2239 atomic_inc(&trace_record_cmdline_disabled);
2241 if (*pos != iter->pos) {
2246 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2247 for_each_tracing_cpu(cpu)
2248 tracing_iter_reset(iter, cpu);
2250 tracing_iter_reset(iter, cpu_file);
2253 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2258 * If we overflowed the seq_file before, then we want
2259 * to just reuse the trace_seq buffer again.
2265 p = s_next(m, p, &l);
2269 trace_event_read_lock();
2270 trace_access_lock(cpu_file);
2274 static void s_stop(struct seq_file *m, void *p)
2276 struct trace_iterator *iter = m->private;
2278 #ifdef CONFIG_TRACER_MAX_TRACE
2279 if (iter->snapshot && iter->trace->use_max_tr)
2283 if (!iter->snapshot)
2284 atomic_dec(&trace_record_cmdline_disabled);
2286 trace_access_unlock(iter->cpu_file);
2287 trace_event_read_unlock();
2291 get_total_entries(struct trace_buffer *buf,
2292 unsigned long *total, unsigned long *entries)
2294 unsigned long count;
2300 for_each_tracing_cpu(cpu) {
2301 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2303 * If this buffer has skipped entries, then we hold all
2304 * entries for the trace and we need to ignore the
2305 * ones before the time stamp.
2307 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2308 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2309 /* total is the same as the entries */
2313 ring_buffer_overrun_cpu(buf->buffer, cpu);
2318 static void print_lat_help_header(struct seq_file *m)
2320 seq_puts(m, "# _------=> CPU# \n");
2321 seq_puts(m, "# / _-----=> irqs-off \n");
2322 seq_puts(m, "# | / _----=> need-resched \n");
2323 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2324 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2325 seq_puts(m, "# |||| / delay \n");
2326 seq_puts(m, "# cmd pid ||||| time | caller \n");
2327 seq_puts(m, "# \\ / ||||| \\ | / \n");
2330 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2332 unsigned long total;
2333 unsigned long entries;
2335 get_total_entries(buf, &total, &entries);
2336 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2337 entries, total, num_online_cpus());
2341 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2343 print_event_info(buf, m);
2344 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2345 seq_puts(m, "# | | | | |\n");
2348 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2350 print_event_info(buf, m);
2351 seq_puts(m, "# _-----=> irqs-off\n");
2352 seq_puts(m, "# / _----=> need-resched\n");
2353 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2354 seq_puts(m, "# || / _--=> preempt-depth\n");
2355 seq_puts(m, "# ||| / delay\n");
2356 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2357 seq_puts(m, "# | | | |||| | |\n");
2361 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2363 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2364 struct trace_buffer *buf = iter->trace_buffer;
2365 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2366 struct tracer *type = iter->trace;
2367 unsigned long entries;
2368 unsigned long total;
2369 const char *name = "preemption";
2373 get_total_entries(buf, &total, &entries);
2375 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2377 seq_puts(m, "# -----------------------------------"
2378 "---------------------------------\n");
2379 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2380 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2381 nsecs_to_usecs(data->saved_latency),
2385 #if defined(CONFIG_PREEMPT_NONE)
2387 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2389 #elif defined(CONFIG_PREEMPT)
2394 /* These are reserved for later use */
2397 seq_printf(m, " #P:%d)\n", num_online_cpus());
2401 seq_puts(m, "# -----------------\n");
2402 seq_printf(m, "# | task: %.16s-%d "
2403 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2404 data->comm, data->pid,
2405 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2406 data->policy, data->rt_priority);
2407 seq_puts(m, "# -----------------\n");
2409 if (data->critical_start) {
2410 seq_puts(m, "# => started at: ");
2411 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2412 trace_print_seq(m, &iter->seq);
2413 seq_puts(m, "\n# => ended at: ");
2414 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2415 trace_print_seq(m, &iter->seq);
2416 seq_puts(m, "\n#\n");
2422 static void test_cpu_buff_start(struct trace_iterator *iter)
2424 struct trace_seq *s = &iter->seq;
2426 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2429 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2432 if (cpumask_test_cpu(iter->cpu, iter->started))
2435 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2438 cpumask_set_cpu(iter->cpu, iter->started);
2440 /* Don't print started cpu buffer for the first entry of the trace */
2442 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2446 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2448 struct trace_seq *s = &iter->seq;
2449 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2450 struct trace_entry *entry;
2451 struct trace_event *event;
2455 test_cpu_buff_start(iter);
2457 event = ftrace_find_event(entry->type);
2459 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2460 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2461 if (!trace_print_lat_context(iter))
2464 if (!trace_print_context(iter))
2470 return event->funcs->trace(iter, sym_flags, event);
2472 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2475 return TRACE_TYPE_HANDLED;
2477 return TRACE_TYPE_PARTIAL_LINE;
2480 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2482 struct trace_seq *s = &iter->seq;
2483 struct trace_entry *entry;
2484 struct trace_event *event;
2488 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2489 if (!trace_seq_printf(s, "%d %d %llu ",
2490 entry->pid, iter->cpu, iter->ts))
2494 event = ftrace_find_event(entry->type);
2496 return event->funcs->raw(iter, 0, event);
2498 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2501 return TRACE_TYPE_HANDLED;
2503 return TRACE_TYPE_PARTIAL_LINE;
2506 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2508 struct trace_seq *s = &iter->seq;
2509 unsigned char newline = '\n';
2510 struct trace_entry *entry;
2511 struct trace_event *event;
2515 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2516 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2517 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2518 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2521 event = ftrace_find_event(entry->type);
2523 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2524 if (ret != TRACE_TYPE_HANDLED)
2528 SEQ_PUT_FIELD_RET(s, newline);
2530 return TRACE_TYPE_HANDLED;
2533 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2535 struct trace_seq *s = &iter->seq;
2536 struct trace_entry *entry;
2537 struct trace_event *event;
2541 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2542 SEQ_PUT_FIELD_RET(s, entry->pid);
2543 SEQ_PUT_FIELD_RET(s, iter->cpu);
2544 SEQ_PUT_FIELD_RET(s, iter->ts);
2547 event = ftrace_find_event(entry->type);
2548 return event ? event->funcs->binary(iter, 0, event) :
2552 int trace_empty(struct trace_iterator *iter)
2554 struct ring_buffer_iter *buf_iter;
2557 /* If we are looking at one CPU buffer, only check that one */
2558 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2559 cpu = iter->cpu_file;
2560 buf_iter = trace_buffer_iter(iter, cpu);
2562 if (!ring_buffer_iter_empty(buf_iter))
2565 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2571 for_each_tracing_cpu(cpu) {
2572 buf_iter = trace_buffer_iter(iter, cpu);
2574 if (!ring_buffer_iter_empty(buf_iter))
2577 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2585 /* Called with trace_event_read_lock() held. */
2586 enum print_line_t print_trace_line(struct trace_iterator *iter)
2588 enum print_line_t ret;
2590 if (iter->lost_events &&
2591 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2592 iter->cpu, iter->lost_events))
2593 return TRACE_TYPE_PARTIAL_LINE;
2595 if (iter->trace && iter->trace->print_line) {
2596 ret = iter->trace->print_line(iter);
2597 if (ret != TRACE_TYPE_UNHANDLED)
2601 if (iter->ent->type == TRACE_BPUTS &&
2602 trace_flags & TRACE_ITER_PRINTK &&
2603 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2604 return trace_print_bputs_msg_only(iter);
2606 if (iter->ent->type == TRACE_BPRINT &&
2607 trace_flags & TRACE_ITER_PRINTK &&
2608 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2609 return trace_print_bprintk_msg_only(iter);
2611 if (iter->ent->type == TRACE_PRINT &&
2612 trace_flags & TRACE_ITER_PRINTK &&
2613 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2614 return trace_print_printk_msg_only(iter);
2616 if (trace_flags & TRACE_ITER_BIN)
2617 return print_bin_fmt(iter);
2619 if (trace_flags & TRACE_ITER_HEX)
2620 return print_hex_fmt(iter);
2622 if (trace_flags & TRACE_ITER_RAW)
2623 return print_raw_fmt(iter);
2625 return print_trace_fmt(iter);
2628 void trace_latency_header(struct seq_file *m)
2630 struct trace_iterator *iter = m->private;
2632 /* print nothing if the buffers are empty */
2633 if (trace_empty(iter))
2636 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2637 print_trace_header(m, iter);
2639 if (!(trace_flags & TRACE_ITER_VERBOSE))
2640 print_lat_help_header(m);
2643 void trace_default_header(struct seq_file *m)
2645 struct trace_iterator *iter = m->private;
2647 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2650 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2651 /* print nothing if the buffers are empty */
2652 if (trace_empty(iter))
2654 print_trace_header(m, iter);
2655 if (!(trace_flags & TRACE_ITER_VERBOSE))
2656 print_lat_help_header(m);
2658 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2659 if (trace_flags & TRACE_ITER_IRQ_INFO)
2660 print_func_help_header_irq(iter->trace_buffer, m);
2662 print_func_help_header(iter->trace_buffer, m);
2667 static void test_ftrace_alive(struct seq_file *m)
2669 if (!ftrace_is_dead())
2671 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2672 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2675 #ifdef CONFIG_TRACER_MAX_TRACE
2676 static void show_snapshot_main_help(struct seq_file *m)
2678 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2679 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2680 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2681 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate)\n");
2682 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2683 seq_printf(m, "# is not a '0' or '1')\n");
2686 static void show_snapshot_percpu_help(struct seq_file *m)
2688 seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2689 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2690 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2691 seq_printf(m, "# Takes a snapshot of the main buffer for this cpu.\n");
2693 seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n");
2694 seq_printf(m, "# Must use main snapshot file to allocate.\n");
2696 seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2697 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2698 seq_printf(m, "# is not a '0' or '1')\n");
2701 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2703 if (iter->tr->allocated_snapshot)
2704 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2706 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2708 seq_printf(m, "# Snapshot commands:\n");
2709 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2710 show_snapshot_main_help(m);
2712 show_snapshot_percpu_help(m);
2715 /* Should never be called */
2716 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2719 static int s_show(struct seq_file *m, void *v)
2721 struct trace_iterator *iter = v;
2724 if (iter->ent == NULL) {
2726 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2728 test_ftrace_alive(m);
2730 if (iter->snapshot && trace_empty(iter))
2731 print_snapshot_help(m, iter);
2732 else if (iter->trace && iter->trace->print_header)
2733 iter->trace->print_header(m);
2735 trace_default_header(m);
2737 } else if (iter->leftover) {
2739 * If we filled the seq_file buffer earlier, we
2740 * want to just show it now.
2742 ret = trace_print_seq(m, &iter->seq);
2744 /* ret should this time be zero, but you never know */
2745 iter->leftover = ret;
2748 print_trace_line(iter);
2749 ret = trace_print_seq(m, &iter->seq);
2751 * If we overflow the seq_file buffer, then it will
2752 * ask us for this data again at start up.
2754 * ret is 0 if seq_file write succeeded.
2757 iter->leftover = ret;
2763 static const struct seq_operations tracer_seq_ops = {
2770 static struct trace_iterator *
2771 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
2773 struct trace_cpu *tc = inode->i_private;
2774 struct trace_array *tr = tc->tr;
2775 struct trace_iterator *iter;
2778 if (tracing_disabled)
2779 return ERR_PTR(-ENODEV);
2781 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2783 return ERR_PTR(-ENOMEM);
2785 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2787 if (!iter->buffer_iter)
2791 * We make a copy of the current tracer to avoid concurrent
2792 * changes on it while we are reading.
2794 mutex_lock(&trace_types_lock);
2795 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2799 *iter->trace = *tr->current_trace;
2801 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2806 #ifdef CONFIG_TRACER_MAX_TRACE
2807 /* Currently only the top directory has a snapshot */
2808 if (tr->current_trace->print_max || snapshot)
2809 iter->trace_buffer = &tr->max_buffer;
2812 iter->trace_buffer = &tr->trace_buffer;
2813 iter->snapshot = snapshot;
2815 mutex_init(&iter->mutex);
2816 iter->cpu_file = tc->cpu;
2818 /* Notify the tracer early; before we stop tracing. */
2819 if (iter->trace && iter->trace->open)
2820 iter->trace->open(iter);
2822 /* Annotate start of buffers if we had overruns */
2823 if (ring_buffer_overruns(iter->trace_buffer->buffer))
2824 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2826 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2827 if (trace_clocks[tr->clock_id].in_ns)
2828 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2830 /* stop the trace while dumping if we are not opening "snapshot" */
2831 if (!iter->snapshot)
2832 tracing_stop_tr(tr);
2834 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
2835 for_each_tracing_cpu(cpu) {
2836 iter->buffer_iter[cpu] =
2837 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2839 ring_buffer_read_prepare_sync();
2840 for_each_tracing_cpu(cpu) {
2841 ring_buffer_read_start(iter->buffer_iter[cpu]);
2842 tracing_iter_reset(iter, cpu);
2845 cpu = iter->cpu_file;
2846 iter->buffer_iter[cpu] =
2847 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2848 ring_buffer_read_prepare_sync();
2849 ring_buffer_read_start(iter->buffer_iter[cpu]);
2850 tracing_iter_reset(iter, cpu);
2855 mutex_unlock(&trace_types_lock);
2860 mutex_unlock(&trace_types_lock);
2862 kfree(iter->buffer_iter);
2864 seq_release_private(inode, file);
2865 return ERR_PTR(-ENOMEM);
2868 int tracing_open_generic(struct inode *inode, struct file *filp)
2870 if (tracing_disabled)
2873 filp->private_data = inode->i_private;
2877 static int tracing_release(struct inode *inode, struct file *file)
2879 struct seq_file *m = file->private_data;
2880 struct trace_iterator *iter;
2881 struct trace_array *tr;
2884 if (!(file->f_mode & FMODE_READ))
2890 mutex_lock(&trace_types_lock);
2895 for_each_tracing_cpu(cpu) {
2896 if (iter->buffer_iter[cpu])
2897 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2900 if (iter->trace && iter->trace->close)
2901 iter->trace->close(iter);
2903 if (!iter->snapshot)
2904 /* reenable tracing if it was previously enabled */
2905 tracing_start_tr(tr);
2906 mutex_unlock(&trace_types_lock);
2908 mutex_destroy(&iter->mutex);
2909 free_cpumask_var(iter->started);
2911 kfree(iter->buffer_iter);
2912 seq_release_private(inode, file);
2916 static int tracing_open(struct inode *inode, struct file *file)
2918 struct trace_iterator *iter;
2921 /* If this file was open for write, then erase contents */
2922 if ((file->f_mode & FMODE_WRITE) &&
2923 (file->f_flags & O_TRUNC)) {
2924 struct trace_cpu *tc = inode->i_private;
2925 struct trace_array *tr = tc->tr;
2927 if (tc->cpu == RING_BUFFER_ALL_CPUS)
2928 tracing_reset_online_cpus(&tr->trace_buffer);
2930 tracing_reset(&tr->trace_buffer, tc->cpu);
2933 if (file->f_mode & FMODE_READ) {
2934 iter = __tracing_open(inode, file, false);
2936 ret = PTR_ERR(iter);
2937 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2938 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2944 t_next(struct seq_file *m, void *v, loff_t *pos)
2946 struct tracer *t = v;
2956 static void *t_start(struct seq_file *m, loff_t *pos)
2961 mutex_lock(&trace_types_lock);
2962 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2968 static void t_stop(struct seq_file *m, void *p)
2970 mutex_unlock(&trace_types_lock);
2973 static int t_show(struct seq_file *m, void *v)
2975 struct tracer *t = v;
2980 seq_printf(m, "%s", t->name);
2989 static const struct seq_operations show_traces_seq_ops = {
2996 static int show_traces_open(struct inode *inode, struct file *file)
2998 if (tracing_disabled)
3001 return seq_open(file, &show_traces_seq_ops);
3005 tracing_write_stub(struct file *filp, const char __user *ubuf,
3006 size_t count, loff_t *ppos)
3011 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
3013 if (file->f_mode & FMODE_READ)
3014 return seq_lseek(file, offset, origin);
3019 static const struct file_operations tracing_fops = {
3020 .open = tracing_open,
3022 .write = tracing_write_stub,
3023 .llseek = tracing_seek,
3024 .release = tracing_release,
3027 static const struct file_operations show_traces_fops = {
3028 .open = show_traces_open,
3030 .release = seq_release,
3031 .llseek = seq_lseek,
3035 * Only trace on a CPU if the bitmask is set:
3037 static cpumask_var_t tracing_cpumask;
3040 * The tracer itself will not take this lock, but still we want
3041 * to provide a consistent cpumask to user-space:
3043 static DEFINE_MUTEX(tracing_cpumask_update_lock);
3046 * Temporary storage for the character representation of the
3047 * CPU bitmask (and one more byte for the newline):
3049 static char mask_str[NR_CPUS + 1];
3052 tracing_cpumask_read(struct file *filp, char __user *ubuf,
3053 size_t count, loff_t *ppos)
3057 mutex_lock(&tracing_cpumask_update_lock);
3059 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
3060 if (count - len < 2) {
3064 len += sprintf(mask_str + len, "\n");
3065 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3068 mutex_unlock(&tracing_cpumask_update_lock);
3074 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3075 size_t count, loff_t *ppos)
3077 struct trace_array *tr = filp->private_data;
3078 cpumask_var_t tracing_cpumask_new;
3081 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3084 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3088 mutex_lock(&tracing_cpumask_update_lock);
3090 local_irq_disable();
3091 arch_spin_lock(&ftrace_max_lock);
3092 for_each_tracing_cpu(cpu) {
3094 * Increase/decrease the disabled counter if we are
3095 * about to flip a bit in the cpumask:
3097 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
3098 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3099 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3100 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3102 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
3103 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3104 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3105 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3108 arch_spin_unlock(&ftrace_max_lock);
3111 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
3113 mutex_unlock(&tracing_cpumask_update_lock);
3114 free_cpumask_var(tracing_cpumask_new);
3119 free_cpumask_var(tracing_cpumask_new);
3124 static const struct file_operations tracing_cpumask_fops = {
3125 .open = tracing_open_generic,
3126 .read = tracing_cpumask_read,
3127 .write = tracing_cpumask_write,
3128 .llseek = generic_file_llseek,
3131 static int tracing_trace_options_show(struct seq_file *m, void *v)
3133 struct tracer_opt *trace_opts;
3134 struct trace_array *tr = m->private;
3138 mutex_lock(&trace_types_lock);
3139 tracer_flags = tr->current_trace->flags->val;
3140 trace_opts = tr->current_trace->flags->opts;
3142 for (i = 0; trace_options[i]; i++) {
3143 if (trace_flags & (1 << i))
3144 seq_printf(m, "%s\n", trace_options[i]);
3146 seq_printf(m, "no%s\n", trace_options[i]);
3149 for (i = 0; trace_opts[i].name; i++) {
3150 if (tracer_flags & trace_opts[i].bit)
3151 seq_printf(m, "%s\n", trace_opts[i].name);
3153 seq_printf(m, "no%s\n", trace_opts[i].name);
3155 mutex_unlock(&trace_types_lock);
3160 static int __set_tracer_option(struct tracer *trace,
3161 struct tracer_flags *tracer_flags,
3162 struct tracer_opt *opts, int neg)
3166 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
3171 tracer_flags->val &= ~opts->bit;
3173 tracer_flags->val |= opts->bit;
3177 /* Try to assign a tracer specific option */
3178 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
3180 struct tracer_flags *tracer_flags = trace->flags;
3181 struct tracer_opt *opts = NULL;
3184 for (i = 0; tracer_flags->opts[i].name; i++) {
3185 opts = &tracer_flags->opts[i];
3187 if (strcmp(cmp, opts->name) == 0)
3188 return __set_tracer_option(trace, trace->flags,
3195 /* Some tracers require overwrite to stay enabled */
3196 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3198 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3204 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3206 /* do nothing if flag is already set */
3207 if (!!(trace_flags & mask) == !!enabled)
3210 /* Give the tracer a chance to approve the change */
3211 if (tr->current_trace->flag_changed)
3212 if (tr->current_trace->flag_changed(tr->current_trace, mask, !!enabled))
3216 trace_flags |= mask;
3218 trace_flags &= ~mask;
3220 if (mask == TRACE_ITER_RECORD_CMD)
3221 trace_event_enable_cmd_record(enabled);
3223 if (mask == TRACE_ITER_OVERWRITE) {
3224 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3225 #ifdef CONFIG_TRACER_MAX_TRACE
3226 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3230 if (mask == TRACE_ITER_PRINTK)
3231 trace_printk_start_stop_comm(enabled);
3236 static int trace_set_options(struct trace_array *tr, char *option)
3243 cmp = strstrip(option);
3245 if (strncmp(cmp, "no", 2) == 0) {
3250 mutex_lock(&trace_types_lock);
3252 for (i = 0; trace_options[i]; i++) {
3253 if (strcmp(cmp, trace_options[i]) == 0) {
3254 ret = set_tracer_flag(tr, 1 << i, !neg);
3259 /* If no option could be set, test the specific tracer options */
3260 if (!trace_options[i])
3261 ret = set_tracer_option(tr->current_trace, cmp, neg);
3263 mutex_unlock(&trace_types_lock);
3269 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3270 size_t cnt, loff_t *ppos)
3272 struct seq_file *m = filp->private_data;
3273 struct trace_array *tr = m->private;
3277 if (cnt >= sizeof(buf))
3280 if (copy_from_user(&buf, ubuf, cnt))
3285 ret = trace_set_options(tr, buf);
3294 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3296 if (tracing_disabled)
3299 return single_open(file, tracing_trace_options_show, inode->i_private);
3302 static const struct file_operations tracing_iter_fops = {
3303 .open = tracing_trace_options_open,
3305 .llseek = seq_lseek,
3306 .release = single_release,
3307 .write = tracing_trace_options_write,
3310 static const char readme_msg[] =
3311 "tracing mini-HOWTO:\n\n"
3312 "# echo 0 > tracing_on : quick way to disable tracing\n"
3313 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3314 " Important files:\n"
3315 " trace\t\t\t- The static contents of the buffer\n"
3316 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3317 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3318 " current_tracer\t- function and latency tracers\n"
3319 " available_tracers\t- list of configured tracers for current_tracer\n"
3320 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3321 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3322 " trace_clock\t\t-change the clock used to order events\n"
3323 " local: Per cpu clock but may not be synced across CPUs\n"
3324 " global: Synced across CPUs but slows tracing down.\n"
3325 " counter: Not a clock, but just an increment\n"
3326 " uptime: Jiffy counter from time of boot\n"
3327 " perf: Same clock that perf events use\n"
3328 #ifdef CONFIG_X86_64
3329 " x86-tsc: TSC cycle counter\n"
3331 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3332 " tracing_cpumask\t- Limit which CPUs to trace\n"
3333 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3334 "\t\t\t Remove sub-buffer with rmdir\n"
3335 " trace_options\t\t- Set format or modify how tracing happens\n"
3336 "\t\t\t Disable an option by adding a suffix 'no' to the option name\n"
3337 #ifdef CONFIG_DYNAMIC_FTRACE
3338 "\n available_filter_functions - list of functions that can be filtered on\n"
3339 " set_ftrace_filter\t- echo function name in here to only trace these functions\n"
3340 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3341 " modules: Can select a group via module\n"
3342 " Format: :mod:<module-name>\n"
3343 " example: echo :mod:ext3 > set_ftrace_filter\n"
3344 " triggers: a command to perform when function is hit\n"
3345 " Format: <function>:<trigger>[:count]\n"
3346 " trigger: traceon, traceoff\n"
3347 " enable_event:<system>:<event>\n"
3348 " disable_event:<system>:<event>\n"
3349 #ifdef CONFIG_STACKTRACE
3352 #ifdef CONFIG_TRACER_SNAPSHOT
3355 " example: echo do_fault:traceoff > set_ftrace_filter\n"
3356 " echo do_trap:traceoff:3 > set_ftrace_filter\n"
3357 " The first one will disable tracing every time do_fault is hit\n"
3358 " The second will disable tracing at most 3 times when do_trap is hit\n"
3359 " The first time do trap is hit and it disables tracing, the counter\n"
3360 " will decrement to 2. If tracing is already disabled, the counter\n"
3361 " will not decrement. It only decrements when the trigger did work\n"
3362 " To remove trigger without count:\n"
3363 " echo '!<function>:<trigger> > set_ftrace_filter\n"
3364 " To remove trigger with a count:\n"
3365 " echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3366 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3367 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3368 " modules: Can select a group via module command :mod:\n"
3369 " Does not accept triggers\n"
3370 #endif /* CONFIG_DYNAMIC_FTRACE */
3371 #ifdef CONFIG_FUNCTION_TRACER
3372 " set_ftrace_pid\t- Write pid(s) to only function trace those pids (function)\n"
3374 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3375 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3376 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3378 #ifdef CONFIG_TRACER_SNAPSHOT
3379 "\n snapshot\t\t- Like 'trace' but shows the content of the static snapshot buffer\n"
3380 "\t\t\t Read the contents for more information\n"
3382 #ifdef CONFIG_STACKTRACE
3383 " stack_trace\t\t- Shows the max stack trace when active\n"
3384 " stack_max_size\t- Shows current max stack size that was traced\n"
3385 "\t\t\t Write into this file to reset the max size (trigger a new trace)\n"
3386 #ifdef CONFIG_DYNAMIC_FTRACE
3387 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace traces\n"
3389 #endif /* CONFIG_STACKTRACE */
3393 tracing_readme_read(struct file *filp, char __user *ubuf,
3394 size_t cnt, loff_t *ppos)
3396 return simple_read_from_buffer(ubuf, cnt, ppos,
3397 readme_msg, strlen(readme_msg));
3400 static const struct file_operations tracing_readme_fops = {
3401 .open = tracing_open_generic,
3402 .read = tracing_readme_read,
3403 .llseek = generic_file_llseek,
3407 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3408 size_t cnt, loff_t *ppos)
3417 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3421 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3429 for (i = 0; i < SAVED_CMDLINES; i++) {
3432 pid = map_cmdline_to_pid[i];
3433 if (pid == -1 || pid == NO_CMDLINE_MAP)
3436 trace_find_cmdline(pid, buf_comm);
3437 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3442 len = simple_read_from_buffer(ubuf, cnt, ppos,
3451 static const struct file_operations tracing_saved_cmdlines_fops = {
3452 .open = tracing_open_generic,
3453 .read = tracing_saved_cmdlines_read,
3454 .llseek = generic_file_llseek,
3458 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3459 size_t cnt, loff_t *ppos)
3461 struct trace_array *tr = filp->private_data;
3462 char buf[MAX_TRACER_SIZE+2];
3465 mutex_lock(&trace_types_lock);
3466 r = sprintf(buf, "%s\n", tr->current_trace->name);
3467 mutex_unlock(&trace_types_lock);
3469 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3472 int tracer_init(struct tracer *t, struct trace_array *tr)
3474 tracing_reset_online_cpus(&tr->trace_buffer);
3478 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3482 for_each_tracing_cpu(cpu)
3483 per_cpu_ptr(buf->data, cpu)->entries = val;
3486 #ifdef CONFIG_TRACER_MAX_TRACE
3487 /* resize @tr's buffer to the size of @size_tr's entries */
3488 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3489 struct trace_buffer *size_buf, int cpu_id)
3493 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3494 for_each_tracing_cpu(cpu) {
3495 ret = ring_buffer_resize(trace_buf->buffer,
3496 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3499 per_cpu_ptr(trace_buf->data, cpu)->entries =
3500 per_cpu_ptr(size_buf->data, cpu)->entries;
3503 ret = ring_buffer_resize(trace_buf->buffer,
3504 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3506 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3507 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3512 #endif /* CONFIG_TRACER_MAX_TRACE */
3514 static int __tracing_resize_ring_buffer(struct trace_array *tr,
3515 unsigned long size, int cpu)
3520 * If kernel or user changes the size of the ring buffer
3521 * we use the size that was given, and we can forget about
3522 * expanding it later.
3524 ring_buffer_expanded = true;
3526 /* May be called before buffers are initialized */
3527 if (!tr->trace_buffer.buffer)
3530 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3534 #ifdef CONFIG_TRACER_MAX_TRACE
3535 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3536 !tr->current_trace->use_max_tr)
3539 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3541 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3542 &tr->trace_buffer, cpu);
3545 * AARGH! We are left with different
3546 * size max buffer!!!!
3547 * The max buffer is our "snapshot" buffer.
3548 * When a tracer needs a snapshot (one of the
3549 * latency tracers), it swaps the max buffer
3550 * with the saved snap shot. We succeeded to
3551 * update the size of the main buffer, but failed to
3552 * update the size of the max buffer. But when we tried
3553 * to reset the main buffer to the original size, we
3554 * failed there too. This is very unlikely to
3555 * happen, but if it does, warn and kill all
3559 tracing_disabled = 1;
3564 if (cpu == RING_BUFFER_ALL_CPUS)
3565 set_buffer_entries(&tr->max_buffer, size);
3567 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3570 #endif /* CONFIG_TRACER_MAX_TRACE */
3572 if (cpu == RING_BUFFER_ALL_CPUS)
3573 set_buffer_entries(&tr->trace_buffer, size);
3575 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3580 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3581 unsigned long size, int cpu_id)
3585 mutex_lock(&trace_types_lock);
3587 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3588 /* make sure, this cpu is enabled in the mask */
3589 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3595 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3600 mutex_unlock(&trace_types_lock);
3607 * tracing_update_buffers - used by tracing facility to expand ring buffers
3609 * To save on memory when the tracing is never used on a system with it
3610 * configured in. The ring buffers are set to a minimum size. But once
3611 * a user starts to use the tracing facility, then they need to grow
3612 * to their default size.
3614 * This function is to be called when a tracer is about to be used.
3616 int tracing_update_buffers(void)
3620 mutex_lock(&trace_types_lock);
3621 if (!ring_buffer_expanded)
3622 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3623 RING_BUFFER_ALL_CPUS);
3624 mutex_unlock(&trace_types_lock);
3629 struct trace_option_dentry;
3631 static struct trace_option_dentry *
3632 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3635 destroy_trace_option_files(struct trace_option_dentry *topts);
3637 static int tracing_set_tracer(const char *buf)
3639 static struct trace_option_dentry *topts;
3640 struct trace_array *tr = &global_trace;
3642 #ifdef CONFIG_TRACER_MAX_TRACE
3647 mutex_lock(&trace_types_lock);
3649 if (!ring_buffer_expanded) {
3650 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3651 RING_BUFFER_ALL_CPUS);
3657 for (t = trace_types; t; t = t->next) {
3658 if (strcmp(t->name, buf) == 0)
3665 if (t == tr->current_trace)
3668 trace_branch_disable();
3670 tr->current_trace->enabled = false;
3672 if (tr->current_trace->reset)
3673 tr->current_trace->reset(tr);
3675 /* Current trace needs to be nop_trace before synchronize_sched */
3676 tr->current_trace = &nop_trace;
3678 #ifdef CONFIG_TRACER_MAX_TRACE
3679 had_max_tr = tr->allocated_snapshot;
3681 if (had_max_tr && !t->use_max_tr) {
3683 * We need to make sure that the update_max_tr sees that
3684 * current_trace changed to nop_trace to keep it from
3685 * swapping the buffers after we resize it.
3686 * The update_max_tr is called from interrupts disabled
3687 * so a synchronized_sched() is sufficient.
3689 synchronize_sched();
3693 destroy_trace_option_files(topts);
3695 topts = create_trace_option_files(tr, t);
3697 #ifdef CONFIG_TRACER_MAX_TRACE
3698 if (t->use_max_tr && !had_max_tr) {
3699 ret = alloc_snapshot(tr);
3706 ret = tracer_init(t, tr);
3711 tr->current_trace = t;
3712 tr->current_trace->enabled = true;
3713 trace_branch_enable(tr);
3715 mutex_unlock(&trace_types_lock);
3721 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3722 size_t cnt, loff_t *ppos)
3724 char buf[MAX_TRACER_SIZE+1];
3731 if (cnt > MAX_TRACER_SIZE)
3732 cnt = MAX_TRACER_SIZE;
3734 if (copy_from_user(&buf, ubuf, cnt))
3739 /* strip ending whitespace. */
3740 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3743 err = tracing_set_tracer(buf);
3753 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3754 size_t cnt, loff_t *ppos)
3756 unsigned long *ptr = filp->private_data;
3760 r = snprintf(buf, sizeof(buf), "%ld\n",
3761 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3762 if (r > sizeof(buf))
3764 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3768 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3769 size_t cnt, loff_t *ppos)
3771 unsigned long *ptr = filp->private_data;
3775 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3784 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3786 struct trace_cpu *tc = inode->i_private;
3787 struct trace_array *tr = tc->tr;
3788 struct trace_iterator *iter;
3791 if (tracing_disabled)
3794 mutex_lock(&trace_types_lock);
3796 /* create a buffer to store the information to pass to userspace */
3797 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3804 * We make a copy of the current tracer to avoid concurrent
3805 * changes on it while we are reading.
3807 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3812 *iter->trace = *tr->current_trace;
3814 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3819 /* trace pipe does not show start of buffer */
3820 cpumask_setall(iter->started);
3822 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3823 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3825 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3826 if (trace_clocks[tr->clock_id].in_ns)
3827 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3829 iter->cpu_file = tc->cpu;
3831 iter->trace_buffer = &tc->tr->trace_buffer;
3832 mutex_init(&iter->mutex);
3833 filp->private_data = iter;
3835 if (iter->trace->pipe_open)
3836 iter->trace->pipe_open(iter);
3838 nonseekable_open(inode, filp);
3840 mutex_unlock(&trace_types_lock);
3846 mutex_unlock(&trace_types_lock);
3850 static int tracing_release_pipe(struct inode *inode, struct file *file)
3852 struct trace_iterator *iter = file->private_data;
3854 mutex_lock(&trace_types_lock);
3856 if (iter->trace->pipe_close)
3857 iter->trace->pipe_close(iter);
3859 mutex_unlock(&trace_types_lock);
3861 free_cpumask_var(iter->started);
3862 mutex_destroy(&iter->mutex);
3870 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
3872 /* Iterators are static, they should be filled or empty */
3873 if (trace_buffer_iter(iter, iter->cpu_file))
3874 return POLLIN | POLLRDNORM;
3876 if (trace_flags & TRACE_ITER_BLOCK)
3878 * Always select as readable when in blocking mode
3880 return POLLIN | POLLRDNORM;
3882 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
3887 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3889 struct trace_iterator *iter = filp->private_data;
3891 return trace_poll(iter, filp, poll_table);
3895 * This is a make-shift waitqueue.
3896 * A tracer might use this callback on some rare cases:
3898 * 1) the current tracer might hold the runqueue lock when it wakes up
3899 * a reader, hence a deadlock (sched, function, and function graph tracers)
3900 * 2) the function tracers, trace all functions, we don't want
3901 * the overhead of calling wake_up and friends
3902 * (and tracing them too)
3904 * Anyway, this is really very primitive wakeup.
3906 void poll_wait_pipe(struct trace_iterator *iter)
3908 set_current_state(TASK_INTERRUPTIBLE);
3909 /* sleep for 100 msecs, and try again. */
3910 schedule_timeout(HZ / 10);
3913 /* Must be called with trace_types_lock mutex held. */
3914 static int tracing_wait_pipe(struct file *filp)
3916 struct trace_iterator *iter = filp->private_data;
3918 while (trace_empty(iter)) {
3920 if ((filp->f_flags & O_NONBLOCK)) {
3924 mutex_unlock(&iter->mutex);
3926 iter->trace->wait_pipe(iter);
3928 mutex_lock(&iter->mutex);
3930 if (signal_pending(current))
3934 * We block until we read something and tracing is disabled.
3935 * We still block if tracing is disabled, but we have never
3936 * read anything. This allows a user to cat this file, and
3937 * then enable tracing. But after we have read something,
3938 * we give an EOF when tracing is again disabled.
3940 * iter->pos will be 0 if we haven't read anything.
3942 if (!tracing_is_enabled() && iter->pos)
3953 tracing_read_pipe(struct file *filp, char __user *ubuf,
3954 size_t cnt, loff_t *ppos)
3956 struct trace_iterator *iter = filp->private_data;
3957 struct trace_array *tr = iter->tr;
3960 /* return any leftover data */
3961 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3965 trace_seq_init(&iter->seq);
3967 /* copy the tracer to avoid using a global lock all around */
3968 mutex_lock(&trace_types_lock);
3969 if (unlikely(iter->trace->name != tr->current_trace->name))
3970 *iter->trace = *tr->current_trace;
3971 mutex_unlock(&trace_types_lock);
3974 * Avoid more than one consumer on a single file descriptor
3975 * This is just a matter of traces coherency, the ring buffer itself
3978 mutex_lock(&iter->mutex);
3979 if (iter->trace->read) {
3980 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3986 sret = tracing_wait_pipe(filp);
3990 /* stop when tracing is finished */
3991 if (trace_empty(iter)) {
3996 if (cnt >= PAGE_SIZE)
3997 cnt = PAGE_SIZE - 1;
3999 /* reset all but tr, trace, and overruns */
4000 memset(&iter->seq, 0,
4001 sizeof(struct trace_iterator) -
4002 offsetof(struct trace_iterator, seq));
4005 trace_event_read_lock();
4006 trace_access_lock(iter->cpu_file);
4007 while (trace_find_next_entry_inc(iter) != NULL) {
4008 enum print_line_t ret;
4009 int len = iter->seq.len;
4011 ret = print_trace_line(iter);
4012 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4013 /* don't print partial lines */
4014 iter->seq.len = len;
4017 if (ret != TRACE_TYPE_NO_CONSUME)
4018 trace_consume(iter);
4020 if (iter->seq.len >= cnt)
4024 * Setting the full flag means we reached the trace_seq buffer
4025 * size and we should leave by partial output condition above.
4026 * One of the trace_seq_* functions is not used properly.
4028 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4031 trace_access_unlock(iter->cpu_file);
4032 trace_event_read_unlock();
4034 /* Now copy what we have to the user */
4035 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4036 if (iter->seq.readpos >= iter->seq.len)
4037 trace_seq_init(&iter->seq);
4040 * If there was nothing to send to user, in spite of consuming trace
4041 * entries, go back to wait for more entries.
4047 mutex_unlock(&iter->mutex);
4052 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
4053 struct pipe_buffer *buf)
4055 __free_page(buf->page);
4058 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4061 __free_page(spd->pages[idx]);
4064 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4066 .map = generic_pipe_buf_map,
4067 .unmap = generic_pipe_buf_unmap,
4068 .confirm = generic_pipe_buf_confirm,
4069 .release = tracing_pipe_buf_release,
4070 .steal = generic_pipe_buf_steal,
4071 .get = generic_pipe_buf_get,
4075 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4080 /* Seq buffer is page-sized, exactly what we need. */
4082 count = iter->seq.len;
4083 ret = print_trace_line(iter);
4084 count = iter->seq.len - count;
4087 iter->seq.len -= count;
4090 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4091 iter->seq.len -= count;
4095 if (ret != TRACE_TYPE_NO_CONSUME)
4096 trace_consume(iter);
4098 if (!trace_find_next_entry_inc(iter)) {
4108 static ssize_t tracing_splice_read_pipe(struct file *filp,
4110 struct pipe_inode_info *pipe,
4114 struct page *pages_def[PIPE_DEF_BUFFERS];
4115 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4116 struct trace_iterator *iter = filp->private_data;
4117 struct splice_pipe_desc spd = {
4119 .partial = partial_def,
4120 .nr_pages = 0, /* This gets updated below. */
4121 .nr_pages_max = PIPE_DEF_BUFFERS,
4123 .ops = &tracing_pipe_buf_ops,
4124 .spd_release = tracing_spd_release_pipe,
4126 struct trace_array *tr = iter->tr;
4131 if (splice_grow_spd(pipe, &spd))
4134 /* copy the tracer to avoid using a global lock all around */
4135 mutex_lock(&trace_types_lock);
4136 if (unlikely(iter->trace->name != tr->current_trace->name))
4137 *iter->trace = *tr->current_trace;
4138 mutex_unlock(&trace_types_lock);
4140 mutex_lock(&iter->mutex);
4142 if (iter->trace->splice_read) {
4143 ret = iter->trace->splice_read(iter, filp,
4144 ppos, pipe, len, flags);
4149 ret = tracing_wait_pipe(filp);
4153 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4158 trace_event_read_lock();
4159 trace_access_lock(iter->cpu_file);
4161 /* Fill as many pages as possible. */
4162 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
4163 spd.pages[i] = alloc_page(GFP_KERNEL);
4167 rem = tracing_fill_pipe_page(rem, iter);
4169 /* Copy the data into the page, so we can start over. */
4170 ret = trace_seq_to_buffer(&iter->seq,
4171 page_address(spd.pages[i]),
4174 __free_page(spd.pages[i]);
4177 spd.partial[i].offset = 0;
4178 spd.partial[i].len = iter->seq.len;
4180 trace_seq_init(&iter->seq);
4183 trace_access_unlock(iter->cpu_file);
4184 trace_event_read_unlock();
4185 mutex_unlock(&iter->mutex);
4189 ret = splice_to_pipe(pipe, &spd);
4191 splice_shrink_spd(&spd);
4195 mutex_unlock(&iter->mutex);
4200 tracing_entries_read(struct file *filp, char __user *ubuf,
4201 size_t cnt, loff_t *ppos)
4203 struct trace_cpu *tc = filp->private_data;
4204 struct trace_array *tr = tc->tr;
4209 mutex_lock(&trace_types_lock);
4211 if (tc->cpu == RING_BUFFER_ALL_CPUS) {
4212 int cpu, buf_size_same;
4217 /* check if all cpu sizes are same */
4218 for_each_tracing_cpu(cpu) {
4219 /* fill in the size from first enabled cpu */
4221 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4222 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4228 if (buf_size_same) {
4229 if (!ring_buffer_expanded)
4230 r = sprintf(buf, "%lu (expanded: %lu)\n",
4232 trace_buf_size >> 10);
4234 r = sprintf(buf, "%lu\n", size >> 10);
4236 r = sprintf(buf, "X\n");
4238 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, tc->cpu)->entries >> 10);
4240 mutex_unlock(&trace_types_lock);
4242 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4247 tracing_entries_write(struct file *filp, const char __user *ubuf,
4248 size_t cnt, loff_t *ppos)
4250 struct trace_cpu *tc = filp->private_data;
4254 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4258 /* must have at least 1 entry */
4262 /* value is in KB */
4265 ret = tracing_resize_ring_buffer(tc->tr, val, tc->cpu);
4275 tracing_total_entries_read(struct file *filp, char __user *ubuf,
4276 size_t cnt, loff_t *ppos)
4278 struct trace_array *tr = filp->private_data;
4281 unsigned long size = 0, expanded_size = 0;
4283 mutex_lock(&trace_types_lock);
4284 for_each_tracing_cpu(cpu) {
4285 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
4286 if (!ring_buffer_expanded)
4287 expanded_size += trace_buf_size >> 10;
4289 if (ring_buffer_expanded)
4290 r = sprintf(buf, "%lu\n", size);
4292 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
4293 mutex_unlock(&trace_types_lock);
4295 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4299 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4300 size_t cnt, loff_t *ppos)
4303 * There is no need to read what the user has written, this function
4304 * is just to make sure that there is no error when "echo" is used
4313 tracing_free_buffer_release(struct inode *inode, struct file *filp)
4315 struct trace_array *tr = inode->i_private;
4317 /* disable tracing ? */
4318 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4320 /* resize the ring buffer to 0 */
4321 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4327 tracing_mark_write(struct file *filp, const char __user *ubuf,
4328 size_t cnt, loff_t *fpos)
4330 unsigned long addr = (unsigned long)ubuf;
4331 struct ring_buffer_event *event;
4332 struct ring_buffer *buffer;
4333 struct print_entry *entry;
4334 unsigned long irq_flags;
4335 struct page *pages[2];
4345 if (tracing_disabled)
4348 if (!(trace_flags & TRACE_ITER_MARKERS))
4351 if (cnt > TRACE_BUF_SIZE)
4352 cnt = TRACE_BUF_SIZE;
4355 * Userspace is injecting traces into the kernel trace buffer.
4356 * We want to be as non intrusive as possible.
4357 * To do so, we do not want to allocate any special buffers
4358 * or take any locks, but instead write the userspace data
4359 * straight into the ring buffer.
4361 * First we need to pin the userspace buffer into memory,
4362 * which, most likely it is, because it just referenced it.
4363 * But there's no guarantee that it is. By using get_user_pages_fast()
4364 * and kmap_atomic/kunmap_atomic() we can get access to the
4365 * pages directly. We then write the data directly into the
4368 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4370 /* check if we cross pages */
4371 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4374 offset = addr & (PAGE_SIZE - 1);
4377 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4378 if (ret < nr_pages) {
4380 put_page(pages[ret]);
4385 for (i = 0; i < nr_pages; i++)
4386 map_page[i] = kmap_atomic(pages[i]);
4388 local_save_flags(irq_flags);
4389 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4390 buffer = global_trace.trace_buffer.buffer;
4391 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4392 irq_flags, preempt_count());
4394 /* Ring buffer disabled, return as if not open for write */
4399 entry = ring_buffer_event_data(event);
4400 entry->ip = _THIS_IP_;
4402 if (nr_pages == 2) {
4403 len = PAGE_SIZE - offset;
4404 memcpy(&entry->buf, map_page[0] + offset, len);
4405 memcpy(&entry->buf[len], map_page[1], cnt - len);
4407 memcpy(&entry->buf, map_page[0] + offset, cnt);
4409 if (entry->buf[cnt - 1] != '\n') {
4410 entry->buf[cnt] = '\n';
4411 entry->buf[cnt + 1] = '\0';
4413 entry->buf[cnt] = '\0';
4415 __buffer_unlock_commit(buffer, event);
4422 for (i = 0; i < nr_pages; i++){
4423 kunmap_atomic(map_page[i]);
4430 static int tracing_clock_show(struct seq_file *m, void *v)
4432 struct trace_array *tr = m->private;
4435 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4437 "%s%s%s%s", i ? " " : "",
4438 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4439 i == tr->clock_id ? "]" : "");
4445 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4446 size_t cnt, loff_t *fpos)
4448 struct seq_file *m = filp->private_data;
4449 struct trace_array *tr = m->private;
4451 const char *clockstr;
4454 if (cnt >= sizeof(buf))
4457 if (copy_from_user(&buf, ubuf, cnt))
4462 clockstr = strstrip(buf);
4464 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4465 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4468 if (i == ARRAY_SIZE(trace_clocks))
4471 mutex_lock(&trace_types_lock);
4475 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4478 * New clock may not be consistent with the previous clock.
4479 * Reset the buffer so that it doesn't have incomparable timestamps.
4481 tracing_reset_online_cpus(&global_trace.trace_buffer);
4483 #ifdef CONFIG_TRACER_MAX_TRACE
4484 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4485 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4486 tracing_reset_online_cpus(&global_trace.max_buffer);
4489 mutex_unlock(&trace_types_lock);
4496 static int tracing_clock_open(struct inode *inode, struct file *file)
4498 if (tracing_disabled)
4501 return single_open(file, tracing_clock_show, inode->i_private);
4504 struct ftrace_buffer_info {
4505 struct trace_iterator iter;
4510 #ifdef CONFIG_TRACER_SNAPSHOT
4511 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4513 struct trace_cpu *tc = inode->i_private;
4514 struct trace_iterator *iter;
4518 if (file->f_mode & FMODE_READ) {
4519 iter = __tracing_open(inode, file, true);
4521 ret = PTR_ERR(iter);
4523 /* Writes still need the seq_file to hold the private data */
4524 m = kzalloc(sizeof(*m), GFP_KERNEL);
4527 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4533 iter->trace_buffer = &tc->tr->max_buffer;
4534 iter->cpu_file = tc->cpu;
4536 file->private_data = m;
4543 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4546 struct seq_file *m = filp->private_data;
4547 struct trace_iterator *iter = m->private;
4548 struct trace_array *tr = iter->tr;
4552 ret = tracing_update_buffers();
4556 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4560 mutex_lock(&trace_types_lock);
4562 if (tr->current_trace->use_max_tr) {
4569 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4573 if (tr->allocated_snapshot)
4577 /* Only allow per-cpu swap if the ring buffer supports it */
4578 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4579 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4584 if (!tr->allocated_snapshot) {
4585 ret = alloc_snapshot(tr);
4589 local_irq_disable();
4590 /* Now, we're going to swap */
4591 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4592 update_max_tr(tr, current, smp_processor_id());
4594 update_max_tr_single(tr, current, iter->cpu_file);
4598 if (tr->allocated_snapshot) {
4599 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4600 tracing_reset_online_cpus(&tr->max_buffer);
4602 tracing_reset(&tr->max_buffer, iter->cpu_file);
4612 mutex_unlock(&trace_types_lock);
4616 static int tracing_snapshot_release(struct inode *inode, struct file *file)
4618 struct seq_file *m = file->private_data;
4620 if (file->f_mode & FMODE_READ)
4621 return tracing_release(inode, file);
4623 /* If write only, the seq_file is just a stub */
4631 static int tracing_buffers_open(struct inode *inode, struct file *filp);
4632 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4633 size_t count, loff_t *ppos);
4634 static int tracing_buffers_release(struct inode *inode, struct file *file);
4635 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4636 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
4638 static int snapshot_raw_open(struct inode *inode, struct file *filp)
4640 struct ftrace_buffer_info *info;
4643 ret = tracing_buffers_open(inode, filp);
4647 info = filp->private_data;
4649 if (info->iter.trace->use_max_tr) {
4650 tracing_buffers_release(inode, filp);
4654 info->iter.snapshot = true;
4655 info->iter.trace_buffer = &info->iter.tr->max_buffer;
4660 #endif /* CONFIG_TRACER_SNAPSHOT */
4663 static const struct file_operations tracing_max_lat_fops = {
4664 .open = tracing_open_generic,
4665 .read = tracing_max_lat_read,
4666 .write = tracing_max_lat_write,
4667 .llseek = generic_file_llseek,
4670 static const struct file_operations set_tracer_fops = {
4671 .open = tracing_open_generic,
4672 .read = tracing_set_trace_read,
4673 .write = tracing_set_trace_write,
4674 .llseek = generic_file_llseek,
4677 static const struct file_operations tracing_pipe_fops = {
4678 .open = tracing_open_pipe,
4679 .poll = tracing_poll_pipe,
4680 .read = tracing_read_pipe,
4681 .splice_read = tracing_splice_read_pipe,
4682 .release = tracing_release_pipe,
4683 .llseek = no_llseek,
4686 static const struct file_operations tracing_entries_fops = {
4687 .open = tracing_open_generic,
4688 .read = tracing_entries_read,
4689 .write = tracing_entries_write,
4690 .llseek = generic_file_llseek,
4693 static const struct file_operations tracing_total_entries_fops = {
4694 .open = tracing_open_generic,
4695 .read = tracing_total_entries_read,
4696 .llseek = generic_file_llseek,
4699 static const struct file_operations tracing_free_buffer_fops = {
4700 .write = tracing_free_buffer_write,
4701 .release = tracing_free_buffer_release,
4704 static const struct file_operations tracing_mark_fops = {
4705 .open = tracing_open_generic,
4706 .write = tracing_mark_write,
4707 .llseek = generic_file_llseek,
4710 static const struct file_operations trace_clock_fops = {
4711 .open = tracing_clock_open,
4713 .llseek = seq_lseek,
4714 .release = single_release,
4715 .write = tracing_clock_write,
4718 #ifdef CONFIG_TRACER_SNAPSHOT
4719 static const struct file_operations snapshot_fops = {
4720 .open = tracing_snapshot_open,
4722 .write = tracing_snapshot_write,
4723 .llseek = tracing_seek,
4724 .release = tracing_snapshot_release,
4727 static const struct file_operations snapshot_raw_fops = {
4728 .open = snapshot_raw_open,
4729 .read = tracing_buffers_read,
4730 .release = tracing_buffers_release,
4731 .splice_read = tracing_buffers_splice_read,
4732 .llseek = no_llseek,
4735 #endif /* CONFIG_TRACER_SNAPSHOT */
4737 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4739 struct trace_cpu *tc = inode->i_private;
4740 struct trace_array *tr = tc->tr;
4741 struct ftrace_buffer_info *info;
4743 if (tracing_disabled)
4746 info = kzalloc(sizeof(*info), GFP_KERNEL);
4750 mutex_lock(&trace_types_lock);
4755 info->iter.cpu_file = tc->cpu;
4756 info->iter.trace = tr->current_trace;
4757 info->iter.trace_buffer = &tr->trace_buffer;
4759 /* Force reading ring buffer for first read */
4760 info->read = (unsigned int)-1;
4762 filp->private_data = info;
4764 mutex_unlock(&trace_types_lock);
4766 return nonseekable_open(inode, filp);
4770 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
4772 struct ftrace_buffer_info *info = filp->private_data;
4773 struct trace_iterator *iter = &info->iter;
4775 return trace_poll(iter, filp, poll_table);
4779 tracing_buffers_read(struct file *filp, char __user *ubuf,
4780 size_t count, loff_t *ppos)
4782 struct ftrace_buffer_info *info = filp->private_data;
4783 struct trace_iterator *iter = &info->iter;
4790 mutex_lock(&trace_types_lock);
4792 #ifdef CONFIG_TRACER_MAX_TRACE
4793 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
4800 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
4806 /* Do we have previous read data to read? */
4807 if (info->read < PAGE_SIZE)
4811 trace_access_lock(iter->cpu_file);
4812 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
4816 trace_access_unlock(iter->cpu_file);
4819 if (trace_empty(iter)) {
4820 if ((filp->f_flags & O_NONBLOCK)) {
4824 mutex_unlock(&trace_types_lock);
4825 iter->trace->wait_pipe(iter);
4826 mutex_lock(&trace_types_lock);
4827 if (signal_pending(current)) {
4839 size = PAGE_SIZE - info->read;
4843 ret = copy_to_user(ubuf, info->spare + info->read, size);
4854 mutex_unlock(&trace_types_lock);
4859 static int tracing_buffers_release(struct inode *inode, struct file *file)
4861 struct ftrace_buffer_info *info = file->private_data;
4862 struct trace_iterator *iter = &info->iter;
4864 mutex_lock(&trace_types_lock);
4866 WARN_ON(!iter->tr->ref);
4870 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
4873 mutex_unlock(&trace_types_lock);
4879 struct ring_buffer *buffer;
4884 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4885 struct pipe_buffer *buf)
4887 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4892 ring_buffer_free_read_page(ref->buffer, ref->page);
4897 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4898 struct pipe_buffer *buf)
4900 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4905 /* Pipe buffer operations for a buffer. */
4906 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4908 .map = generic_pipe_buf_map,
4909 .unmap = generic_pipe_buf_unmap,
4910 .confirm = generic_pipe_buf_confirm,
4911 .release = buffer_pipe_buf_release,
4912 .steal = generic_pipe_buf_steal,
4913 .get = buffer_pipe_buf_get,
4917 * Callback from splice_to_pipe(), if we need to release some pages
4918 * at the end of the spd in case we error'ed out in filling the pipe.
4920 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4922 struct buffer_ref *ref =
4923 (struct buffer_ref *)spd->partial[i].private;
4928 ring_buffer_free_read_page(ref->buffer, ref->page);
4930 spd->partial[i].private = 0;
4934 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4935 struct pipe_inode_info *pipe, size_t len,
4938 struct ftrace_buffer_info *info = file->private_data;
4939 struct trace_iterator *iter = &info->iter;
4940 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4941 struct page *pages_def[PIPE_DEF_BUFFERS];
4942 struct splice_pipe_desc spd = {
4944 .partial = partial_def,
4945 .nr_pages_max = PIPE_DEF_BUFFERS,
4947 .ops = &buffer_pipe_buf_ops,
4948 .spd_release = buffer_spd_release,
4950 struct buffer_ref *ref;
4951 int entries, size, i;
4954 mutex_lock(&trace_types_lock);
4956 #ifdef CONFIG_TRACER_MAX_TRACE
4957 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
4963 if (splice_grow_spd(pipe, &spd)) {
4968 if (*ppos & (PAGE_SIZE - 1)) {
4973 if (len & (PAGE_SIZE - 1)) {
4974 if (len < PAGE_SIZE) {
4982 trace_access_lock(iter->cpu_file);
4983 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
4985 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4989 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4994 ref->buffer = iter->trace_buffer->buffer;
4995 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5001 r = ring_buffer_read_page(ref->buffer, &ref->page,
5002 len, iter->cpu_file, 1);
5004 ring_buffer_free_read_page(ref->buffer, ref->page);
5010 * zero out any left over data, this is going to
5013 size = ring_buffer_page_len(ref->page);
5014 if (size < PAGE_SIZE)
5015 memset(ref->page + size, 0, PAGE_SIZE - size);
5017 page = virt_to_page(ref->page);
5019 spd.pages[i] = page;
5020 spd.partial[i].len = PAGE_SIZE;
5021 spd.partial[i].offset = 0;
5022 spd.partial[i].private = (unsigned long)ref;
5026 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5029 trace_access_unlock(iter->cpu_file);
5032 /* did we read anything? */
5033 if (!spd.nr_pages) {
5034 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
5038 mutex_unlock(&trace_types_lock);
5039 iter->trace->wait_pipe(iter);
5040 mutex_lock(&trace_types_lock);
5041 if (signal_pending(current)) {
5048 ret = splice_to_pipe(pipe, &spd);
5049 splice_shrink_spd(&spd);
5051 mutex_unlock(&trace_types_lock);
5056 static const struct file_operations tracing_buffers_fops = {
5057 .open = tracing_buffers_open,
5058 .read = tracing_buffers_read,
5059 .poll = tracing_buffers_poll,
5060 .release = tracing_buffers_release,
5061 .splice_read = tracing_buffers_splice_read,
5062 .llseek = no_llseek,
5066 tracing_stats_read(struct file *filp, char __user *ubuf,
5067 size_t count, loff_t *ppos)
5069 struct trace_cpu *tc = filp->private_data;
5070 struct trace_array *tr = tc->tr;
5071 struct trace_buffer *trace_buf = &tr->trace_buffer;
5072 struct trace_seq *s;
5074 unsigned long long t;
5075 unsigned long usec_rem;
5078 s = kmalloc(sizeof(*s), GFP_KERNEL);
5084 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5085 trace_seq_printf(s, "entries: %ld\n", cnt);
5087 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5088 trace_seq_printf(s, "overrun: %ld\n", cnt);
5090 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5091 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5093 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5094 trace_seq_printf(s, "bytes: %ld\n", cnt);
5096 if (trace_clocks[tr->clock_id].in_ns) {
5097 /* local or global for trace_clock */
5098 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5099 usec_rem = do_div(t, USEC_PER_SEC);
5100 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5103 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5104 usec_rem = do_div(t, USEC_PER_SEC);
5105 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5107 /* counter or tsc mode for trace_clock */
5108 trace_seq_printf(s, "oldest event ts: %llu\n",
5109 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5111 trace_seq_printf(s, "now ts: %llu\n",
5112 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5115 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5116 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5118 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5119 trace_seq_printf(s, "read events: %ld\n", cnt);
5121 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
5128 static const struct file_operations tracing_stats_fops = {
5129 .open = tracing_open_generic,
5130 .read = tracing_stats_read,
5131 .llseek = generic_file_llseek,
5134 #ifdef CONFIG_DYNAMIC_FTRACE
5136 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5142 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5143 size_t cnt, loff_t *ppos)
5145 static char ftrace_dyn_info_buffer[1024];
5146 static DEFINE_MUTEX(dyn_info_mutex);
5147 unsigned long *p = filp->private_data;
5148 char *buf = ftrace_dyn_info_buffer;
5149 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5152 mutex_lock(&dyn_info_mutex);
5153 r = sprintf(buf, "%ld ", *p);
5155 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5158 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5160 mutex_unlock(&dyn_info_mutex);
5165 static const struct file_operations tracing_dyn_info_fops = {
5166 .open = tracing_open_generic,
5167 .read = tracing_read_dyn_info,
5168 .llseek = generic_file_llseek,
5170 #endif /* CONFIG_DYNAMIC_FTRACE */
5172 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5174 ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5180 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5182 unsigned long *count = (long *)data;
5194 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5195 struct ftrace_probe_ops *ops, void *data)
5197 long count = (long)data;
5199 seq_printf(m, "%ps:", (void *)ip);
5201 seq_printf(m, "snapshot");
5204 seq_printf(m, ":unlimited\n");
5206 seq_printf(m, ":count=%ld\n", count);
5211 static struct ftrace_probe_ops snapshot_probe_ops = {
5212 .func = ftrace_snapshot,
5213 .print = ftrace_snapshot_print,
5216 static struct ftrace_probe_ops snapshot_count_probe_ops = {
5217 .func = ftrace_count_snapshot,
5218 .print = ftrace_snapshot_print,
5222 ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
5223 char *glob, char *cmd, char *param, int enable)
5225 struct ftrace_probe_ops *ops;
5226 void *count = (void *)-1;
5230 /* hash funcs only work with set_ftrace_filter */
5234 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
5236 if (glob[0] == '!') {
5237 unregister_ftrace_function_probe_func(glob+1, ops);
5244 number = strsep(¶m, ":");
5246 if (!strlen(number))
5250 * We use the callback data field (which is a pointer)
5253 ret = kstrtoul(number, 0, (unsigned long *)&count);
5258 ret = register_ftrace_function_probe(glob, ops, count);
5261 alloc_snapshot(&global_trace);
5263 return ret < 0 ? ret : 0;
5266 static struct ftrace_func_command ftrace_snapshot_cmd = {
5268 .func = ftrace_trace_snapshot_callback,
5271 static int register_snapshot_cmd(void)
5273 return register_ftrace_command(&ftrace_snapshot_cmd);
5276 static inline int register_snapshot_cmd(void) { return 0; }
5277 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5279 struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
5284 if (!debugfs_initialized())
5287 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
5288 tr->dir = debugfs_create_dir("tracing", NULL);
5291 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5296 struct dentry *tracing_init_dentry(void)
5298 return tracing_init_dentry_tr(&global_trace);
5301 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
5303 struct dentry *d_tracer;
5306 return tr->percpu_dir;
5308 d_tracer = tracing_init_dentry_tr(tr);
5312 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
5314 WARN_ONCE(!tr->percpu_dir,
5315 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
5317 return tr->percpu_dir;
5321 tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
5323 struct trace_array_cpu *data = per_cpu_ptr(tr->trace_buffer.data, cpu);
5324 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
5325 struct dentry *d_cpu;
5326 char cpu_dir[30]; /* 30 characters should be more than enough */
5331 snprintf(cpu_dir, 30, "cpu%ld", cpu);
5332 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
5334 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
5338 /* per cpu trace_pipe */
5339 trace_create_file("trace_pipe", 0444, d_cpu,
5340 (void *)&data->trace_cpu, &tracing_pipe_fops);
5343 trace_create_file("trace", 0644, d_cpu,
5344 (void *)&data->trace_cpu, &tracing_fops);
5346 trace_create_file("trace_pipe_raw", 0444, d_cpu,
5347 (void *)&data->trace_cpu, &tracing_buffers_fops);
5349 trace_create_file("stats", 0444, d_cpu,
5350 (void *)&data->trace_cpu, &tracing_stats_fops);
5352 trace_create_file("buffer_size_kb", 0444, d_cpu,
5353 (void *)&data->trace_cpu, &tracing_entries_fops);
5355 #ifdef CONFIG_TRACER_SNAPSHOT
5356 trace_create_file("snapshot", 0644, d_cpu,
5357 (void *)&data->trace_cpu, &snapshot_fops);
5359 trace_create_file("snapshot_raw", 0444, d_cpu,
5360 (void *)&data->trace_cpu, &snapshot_raw_fops);
5364 #ifdef CONFIG_FTRACE_SELFTEST
5365 /* Let selftest have access to static functions in this file */
5366 #include "trace_selftest.c"
5369 struct trace_option_dentry {
5370 struct tracer_opt *opt;
5371 struct tracer_flags *flags;
5372 struct trace_array *tr;
5373 struct dentry *entry;
5377 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
5380 struct trace_option_dentry *topt = filp->private_data;
5383 if (topt->flags->val & topt->opt->bit)
5388 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5392 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
5395 struct trace_option_dentry *topt = filp->private_data;
5399 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5403 if (val != 0 && val != 1)
5406 if (!!(topt->flags->val & topt->opt->bit) != val) {
5407 mutex_lock(&trace_types_lock);
5408 ret = __set_tracer_option(topt->tr->current_trace, topt->flags,
5410 mutex_unlock(&trace_types_lock);
5421 static const struct file_operations trace_options_fops = {
5422 .open = tracing_open_generic,
5423 .read = trace_options_read,
5424 .write = trace_options_write,
5425 .llseek = generic_file_llseek,
5429 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5432 long index = (long)filp->private_data;
5435 if (trace_flags & (1 << index))
5440 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5444 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5447 struct trace_array *tr = &global_trace;
5448 long index = (long)filp->private_data;
5452 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5456 if (val != 0 && val != 1)
5459 mutex_lock(&trace_types_lock);
5460 ret = set_tracer_flag(tr, 1 << index, val);
5461 mutex_unlock(&trace_types_lock);
5471 static const struct file_operations trace_options_core_fops = {
5472 .open = tracing_open_generic,
5473 .read = trace_options_core_read,
5474 .write = trace_options_core_write,
5475 .llseek = generic_file_llseek,
5478 struct dentry *trace_create_file(const char *name,
5480 struct dentry *parent,
5482 const struct file_operations *fops)
5486 ret = debugfs_create_file(name, mode, parent, data, fops);
5488 pr_warning("Could not create debugfs '%s' entry\n", name);
5494 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5496 struct dentry *d_tracer;
5501 d_tracer = tracing_init_dentry_tr(tr);
5505 tr->options = debugfs_create_dir("options", d_tracer);
5507 pr_warning("Could not create debugfs directory 'options'\n");
5515 create_trace_option_file(struct trace_array *tr,
5516 struct trace_option_dentry *topt,
5517 struct tracer_flags *flags,
5518 struct tracer_opt *opt)
5520 struct dentry *t_options;
5522 t_options = trace_options_init_dentry(tr);
5526 topt->flags = flags;
5530 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5531 &trace_options_fops);
5535 static struct trace_option_dentry *
5536 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5538 struct trace_option_dentry *topts;
5539 struct tracer_flags *flags;
5540 struct tracer_opt *opts;
5546 flags = tracer->flags;
5548 if (!flags || !flags->opts)
5553 for (cnt = 0; opts[cnt].name; cnt++)
5556 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5560 for (cnt = 0; opts[cnt].name; cnt++)
5561 create_trace_option_file(tr, &topts[cnt], flags,
5568 destroy_trace_option_files(struct trace_option_dentry *topts)
5575 for (cnt = 0; topts[cnt].opt; cnt++) {
5576 if (topts[cnt].entry)
5577 debugfs_remove(topts[cnt].entry);
5583 static struct dentry *
5584 create_trace_option_core_file(struct trace_array *tr,
5585 const char *option, long index)
5587 struct dentry *t_options;
5589 t_options = trace_options_init_dentry(tr);
5593 return trace_create_file(option, 0644, t_options, (void *)index,
5594 &trace_options_core_fops);
5597 static __init void create_trace_options_dir(struct trace_array *tr)
5599 struct dentry *t_options;
5602 t_options = trace_options_init_dentry(tr);
5606 for (i = 0; trace_options[i]; i++)
5607 create_trace_option_core_file(tr, trace_options[i], i);
5611 rb_simple_read(struct file *filp, char __user *ubuf,
5612 size_t cnt, loff_t *ppos)
5614 struct trace_array *tr = filp->private_data;
5615 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5620 r = ring_buffer_record_is_on(buffer);
5624 r = sprintf(buf, "%d\n", r);
5626 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5630 rb_simple_write(struct file *filp, const char __user *ubuf,
5631 size_t cnt, loff_t *ppos)
5633 struct trace_array *tr = filp->private_data;
5634 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5638 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5643 mutex_lock(&trace_types_lock);
5645 ring_buffer_record_on(buffer);
5646 if (tr->current_trace->start)
5647 tr->current_trace->start(tr);
5649 ring_buffer_record_off(buffer);
5650 if (tr->current_trace->stop)
5651 tr->current_trace->stop(tr);
5653 mutex_unlock(&trace_types_lock);
5661 static const struct file_operations rb_simple_fops = {
5662 .open = tracing_open_generic,
5663 .read = rb_simple_read,
5664 .write = rb_simple_write,
5665 .llseek = default_llseek,
5668 struct dentry *trace_instance_dir;
5671 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
5673 static void init_trace_buffers(struct trace_array *tr, struct trace_buffer *buf)
5677 for_each_tracing_cpu(cpu) {
5678 memset(per_cpu_ptr(buf->data, cpu), 0, sizeof(struct trace_array_cpu));
5679 per_cpu_ptr(buf->data, cpu)->trace_cpu.cpu = cpu;
5680 per_cpu_ptr(buf->data, cpu)->trace_cpu.tr = tr;
5685 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
5687 enum ring_buffer_flags rb_flags;
5689 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5691 buf->buffer = ring_buffer_alloc(size, rb_flags);
5695 buf->data = alloc_percpu(struct trace_array_cpu);
5697 ring_buffer_free(buf->buffer);
5701 init_trace_buffers(tr, buf);
5703 /* Allocate the first page for all buffers */
5704 set_buffer_entries(&tr->trace_buffer,
5705 ring_buffer_size(tr->trace_buffer.buffer, 0));
5710 static int allocate_trace_buffers(struct trace_array *tr, int size)
5714 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
5718 #ifdef CONFIG_TRACER_MAX_TRACE
5719 ret = allocate_trace_buffer(tr, &tr->max_buffer,
5720 allocate_snapshot ? size : 1);
5722 ring_buffer_free(tr->trace_buffer.buffer);
5723 free_percpu(tr->trace_buffer.data);
5726 tr->allocated_snapshot = allocate_snapshot;
5729 * Only the top level trace array gets its snapshot allocated
5730 * from the kernel command line.
5732 allocate_snapshot = false;
5737 static int new_instance_create(const char *name)
5739 struct trace_array *tr;
5742 mutex_lock(&trace_types_lock);
5745 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5746 if (tr->name && strcmp(tr->name, name) == 0)
5751 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
5755 tr->name = kstrdup(name, GFP_KERNEL);
5759 raw_spin_lock_init(&tr->start_lock);
5761 tr->current_trace = &nop_trace;
5763 INIT_LIST_HEAD(&tr->systems);
5764 INIT_LIST_HEAD(&tr->events);
5766 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
5769 /* Holder for file callbacks */
5770 tr->trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
5771 tr->trace_cpu.tr = tr;
5773 tr->dir = debugfs_create_dir(name, trace_instance_dir);
5777 ret = event_trace_add_tracer(tr->dir, tr);
5781 init_tracer_debugfs(tr, tr->dir);
5783 list_add(&tr->list, &ftrace_trace_arrays);
5785 mutex_unlock(&trace_types_lock);
5790 if (tr->trace_buffer.buffer)
5791 ring_buffer_free(tr->trace_buffer.buffer);
5796 mutex_unlock(&trace_types_lock);
5802 static int instance_delete(const char *name)
5804 struct trace_array *tr;
5808 mutex_lock(&trace_types_lock);
5811 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5812 if (tr->name && strcmp(tr->name, name) == 0) {
5824 list_del(&tr->list);
5826 event_trace_del_tracer(tr);
5827 debugfs_remove_recursive(tr->dir);
5828 free_percpu(tr->trace_buffer.data);
5829 ring_buffer_free(tr->trace_buffer.buffer);
5837 mutex_unlock(&trace_types_lock);
5842 static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
5844 struct dentry *parent;
5847 /* Paranoid: Make sure the parent is the "instances" directory */
5848 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
5849 if (WARN_ON_ONCE(parent != trace_instance_dir))
5853 * The inode mutex is locked, but debugfs_create_dir() will also
5854 * take the mutex. As the instances directory can not be destroyed
5855 * or changed in any other way, it is safe to unlock it, and
5856 * let the dentry try. If two users try to make the same dir at
5857 * the same time, then the new_instance_create() will determine the
5860 mutex_unlock(&inode->i_mutex);
5862 ret = new_instance_create(dentry->d_iname);
5864 mutex_lock(&inode->i_mutex);
5869 static int instance_rmdir(struct inode *inode, struct dentry *dentry)
5871 struct dentry *parent;
5874 /* Paranoid: Make sure the parent is the "instances" directory */
5875 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
5876 if (WARN_ON_ONCE(parent != trace_instance_dir))
5879 /* The caller did a dget() on dentry */
5880 mutex_unlock(&dentry->d_inode->i_mutex);
5883 * The inode mutex is locked, but debugfs_create_dir() will also
5884 * take the mutex. As the instances directory can not be destroyed
5885 * or changed in any other way, it is safe to unlock it, and
5886 * let the dentry try. If two users try to make the same dir at
5887 * the same time, then the instance_delete() will determine the
5890 mutex_unlock(&inode->i_mutex);
5892 ret = instance_delete(dentry->d_iname);
5894 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
5895 mutex_lock(&dentry->d_inode->i_mutex);
5900 static const struct inode_operations instance_dir_inode_operations = {
5901 .lookup = simple_lookup,
5902 .mkdir = instance_mkdir,
5903 .rmdir = instance_rmdir,
5906 static __init void create_trace_instances(struct dentry *d_tracer)
5908 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
5909 if (WARN_ON(!trace_instance_dir))
5912 /* Hijack the dir inode operations, to allow mkdir */
5913 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
5917 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
5921 trace_create_file("trace_options", 0644, d_tracer,
5922 tr, &tracing_iter_fops);
5924 trace_create_file("trace", 0644, d_tracer,
5925 (void *)&tr->trace_cpu, &tracing_fops);
5927 trace_create_file("trace_pipe", 0444, d_tracer,
5928 (void *)&tr->trace_cpu, &tracing_pipe_fops);
5930 trace_create_file("buffer_size_kb", 0644, d_tracer,
5931 (void *)&tr->trace_cpu, &tracing_entries_fops);
5933 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
5934 tr, &tracing_total_entries_fops);
5936 trace_create_file("free_buffer", 0644, d_tracer,
5937 tr, &tracing_free_buffer_fops);
5939 trace_create_file("trace_marker", 0220, d_tracer,
5940 tr, &tracing_mark_fops);
5942 trace_create_file("trace_clock", 0644, d_tracer, tr,
5945 trace_create_file("tracing_on", 0644, d_tracer,
5946 tr, &rb_simple_fops);
5948 #ifdef CONFIG_TRACER_SNAPSHOT
5949 trace_create_file("snapshot", 0644, d_tracer,
5950 (void *)&tr->trace_cpu, &snapshot_fops);
5953 for_each_tracing_cpu(cpu)
5954 tracing_init_debugfs_percpu(tr, cpu);
5958 static __init int tracer_init_debugfs(void)
5960 struct dentry *d_tracer;
5962 trace_access_lock_init();
5964 d_tracer = tracing_init_dentry();
5968 init_tracer_debugfs(&global_trace, d_tracer);
5970 trace_create_file("tracing_cpumask", 0644, d_tracer,
5971 &global_trace, &tracing_cpumask_fops);
5973 trace_create_file("available_tracers", 0444, d_tracer,
5974 &global_trace, &show_traces_fops);
5976 trace_create_file("current_tracer", 0644, d_tracer,
5977 &global_trace, &set_tracer_fops);
5979 #ifdef CONFIG_TRACER_MAX_TRACE
5980 trace_create_file("tracing_max_latency", 0644, d_tracer,
5981 &tracing_max_latency, &tracing_max_lat_fops);
5984 trace_create_file("tracing_thresh", 0644, d_tracer,
5985 &tracing_thresh, &tracing_max_lat_fops);
5987 trace_create_file("README", 0444, d_tracer,
5988 NULL, &tracing_readme_fops);
5990 trace_create_file("saved_cmdlines", 0444, d_tracer,
5991 NULL, &tracing_saved_cmdlines_fops);
5993 #ifdef CONFIG_DYNAMIC_FTRACE
5994 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
5995 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
5998 create_trace_instances(d_tracer);
6000 create_trace_options_dir(&global_trace);
6005 static int trace_panic_handler(struct notifier_block *this,
6006 unsigned long event, void *unused)
6008 if (ftrace_dump_on_oops)
6009 ftrace_dump(ftrace_dump_on_oops);
6013 static struct notifier_block trace_panic_notifier = {
6014 .notifier_call = trace_panic_handler,
6016 .priority = 150 /* priority: INT_MAX >= x >= 0 */
6019 static int trace_die_handler(struct notifier_block *self,
6025 if (ftrace_dump_on_oops)
6026 ftrace_dump(ftrace_dump_on_oops);
6034 static struct notifier_block trace_die_notifier = {
6035 .notifier_call = trace_die_handler,
6040 * printk is set to max of 1024, we really don't need it that big.
6041 * Nothing should be printing 1000 characters anyway.
6043 #define TRACE_MAX_PRINT 1000
6046 * Define here KERN_TRACE so that we have one place to modify
6047 * it if we decide to change what log level the ftrace dump
6050 #define KERN_TRACE KERN_EMERG
6053 trace_printk_seq(struct trace_seq *s)
6055 /* Probably should print a warning here. */
6056 if (s->len >= TRACE_MAX_PRINT)
6057 s->len = TRACE_MAX_PRINT;
6059 /* should be zero ended, but we are paranoid. */
6060 s->buffer[s->len] = 0;
6062 printk(KERN_TRACE "%s", s->buffer);
6067 void trace_init_global_iter(struct trace_iterator *iter)
6069 iter->tr = &global_trace;
6070 iter->trace = iter->tr->current_trace;
6071 iter->cpu_file = RING_BUFFER_ALL_CPUS;
6072 iter->trace_buffer = &global_trace.trace_buffer;
6075 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
6077 /* use static because iter can be a bit big for the stack */
6078 static struct trace_iterator iter;
6079 static atomic_t dump_running;
6080 unsigned int old_userobj;
6081 unsigned long flags;
6084 /* Only allow one dump user at a time. */
6085 if (atomic_inc_return(&dump_running) != 1) {
6086 atomic_dec(&dump_running);
6091 * Always turn off tracing when we dump.
6092 * We don't need to show trace output of what happens
6093 * between multiple crashes.
6095 * If the user does a sysrq-z, then they can re-enable
6096 * tracing with echo 1 > tracing_on.
6100 local_irq_save(flags);
6102 /* Simulate the iterator */
6103 trace_init_global_iter(&iter);
6105 for_each_tracing_cpu(cpu) {
6106 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
6109 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
6111 /* don't look at user memory in panic mode */
6112 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
6114 switch (oops_dump_mode) {
6116 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6119 iter.cpu_file = raw_smp_processor_id();
6124 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
6125 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6128 printk(KERN_TRACE "Dumping ftrace buffer:\n");
6130 /* Did function tracer already get disabled? */
6131 if (ftrace_is_dead()) {
6132 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6133 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6137 * We need to stop all tracing on all CPUS to read the
6138 * the next buffer. This is a bit expensive, but is
6139 * not done often. We fill all what we can read,
6140 * and then release the locks again.
6143 while (!trace_empty(&iter)) {
6146 printk(KERN_TRACE "---------------------------------\n");
6150 /* reset all but tr, trace, and overruns */
6151 memset(&iter.seq, 0,
6152 sizeof(struct trace_iterator) -
6153 offsetof(struct trace_iterator, seq));
6154 iter.iter_flags |= TRACE_FILE_LAT_FMT;
6157 if (trace_find_next_entry_inc(&iter) != NULL) {
6160 ret = print_trace_line(&iter);
6161 if (ret != TRACE_TYPE_NO_CONSUME)
6162 trace_consume(&iter);
6164 touch_nmi_watchdog();
6166 trace_printk_seq(&iter.seq);
6170 printk(KERN_TRACE " (ftrace buffer empty)\n");
6172 printk(KERN_TRACE "---------------------------------\n");
6175 trace_flags |= old_userobj;
6177 for_each_tracing_cpu(cpu) {
6178 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
6180 atomic_dec(&dump_running);
6181 local_irq_restore(flags);
6183 EXPORT_SYMBOL_GPL(ftrace_dump);
6185 __init static int tracer_alloc_buffers(void)
6191 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
6194 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
6195 goto out_free_buffer_mask;
6197 /* Only allocate trace_printk buffers if a trace_printk exists */
6198 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
6199 /* Must be called before global_trace.buffer is allocated */
6200 trace_printk_init_buffers();
6202 /* To save memory, keep the ring buffer size to its minimum */
6203 if (ring_buffer_expanded)
6204 ring_buf_size = trace_buf_size;
6208 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
6209 cpumask_copy(tracing_cpumask, cpu_all_mask);
6211 raw_spin_lock_init(&global_trace.start_lock);
6213 /* TODO: make the number of buffers hot pluggable with CPUS */
6214 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
6215 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
6217 goto out_free_cpumask;
6220 if (global_trace.buffer_disabled)
6223 trace_init_cmdlines();
6226 * register_tracer() might reference current_trace, so it
6227 * needs to be set before we register anything. This is
6228 * just a bootstrap of current_trace anyway.
6230 global_trace.current_trace = &nop_trace;
6232 register_tracer(&nop_trace);
6234 /* All seems OK, enable tracing */
6235 tracing_disabled = 0;
6237 atomic_notifier_chain_register(&panic_notifier_list,
6238 &trace_panic_notifier);
6240 register_die_notifier(&trace_die_notifier);
6242 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
6244 /* Holder for file callbacks */
6245 global_trace.trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
6246 global_trace.trace_cpu.tr = &global_trace;
6248 INIT_LIST_HEAD(&global_trace.systems);
6249 INIT_LIST_HEAD(&global_trace.events);
6250 list_add(&global_trace.list, &ftrace_trace_arrays);
6252 while (trace_boot_options) {
6255 option = strsep(&trace_boot_options, ",");
6256 trace_set_options(&global_trace, option);
6259 register_snapshot_cmd();
6264 free_percpu(global_trace.trace_buffer.data);
6265 #ifdef CONFIG_TRACER_MAX_TRACE
6266 free_percpu(global_trace.max_buffer.data);
6268 free_cpumask_var(tracing_cpumask);
6269 out_free_buffer_mask:
6270 free_cpumask_var(tracing_buffer_mask);
6275 __init static int clear_boot_tracer(void)
6278 * The default tracer at boot buffer is an init section.
6279 * This function is called in lateinit. If we did not
6280 * find the boot tracer, then clear it out, to prevent
6281 * later registration from accessing the buffer that is
6282 * about to be freed.
6284 if (!default_bootup_tracer)
6287 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
6288 default_bootup_tracer);
6289 default_bootup_tracer = NULL;
6294 early_initcall(tracer_alloc_buffers);
6295 fs_initcall(tracer_init_debugfs);
6296 late_initcall(clear_boot_tracer);