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/tracefs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/mount.h>
36 #include <linux/rwsem.h>
37 #include <linux/slab.h>
38 #include <linux/ctype.h>
39 #include <linux/init.h>
40 #include <linux/poll.h>
41 #include <linux/nmi.h>
43 #include <linux/sched/rt.h>
46 #include "trace_output.h"
49 * On boot up, the ring buffer is set to the minimum size, so that
50 * we do not waste memory on systems that are not using tracing.
52 bool ring_buffer_expanded;
55 * We need to change this state when a selftest is running.
56 * A selftest will lurk into the ring-buffer to count the
57 * entries inserted during the selftest although some concurrent
58 * insertions into the ring-buffer such as trace_printk could occurred
59 * at the same time, giving false positive or negative results.
61 static bool __read_mostly tracing_selftest_running;
64 * If a tracer is running, we do not want to run SELFTEST.
66 bool __read_mostly tracing_selftest_disabled;
68 /* Pipe tracepoints to printk */
69 struct trace_iterator *tracepoint_print_iter;
70 int tracepoint_printk;
72 /* For tracers that don't implement custom flags */
73 static struct tracer_opt dummy_tracer_opt[] = {
77 static struct tracer_flags dummy_tracer_flags = {
79 .opts = dummy_tracer_opt
83 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
89 * To prevent the comm cache from being overwritten when no
90 * tracing is active, only save the comm when a trace event
93 static DEFINE_PER_CPU(bool, trace_cmdline_save);
96 * Kill all tracing for good (never come back).
97 * It is initialized to 1 but will turn to zero if the initialization
98 * of the tracer is successful. But that is the only place that sets
101 static int tracing_disabled = 1;
103 cpumask_var_t __read_mostly tracing_buffer_mask;
106 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
108 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
109 * is set, then ftrace_dump is called. This will output the contents
110 * of the ftrace buffers to the console. This is very useful for
111 * capturing traces that lead to crashes and outputing it to a
114 * It is default off, but you can enable it with either specifying
115 * "ftrace_dump_on_oops" in the kernel command line, or setting
116 * /proc/sys/kernel/ftrace_dump_on_oops
117 * Set 1 if you want to dump buffers of all CPUs
118 * Set 2 if you want to dump the buffer of the CPU that triggered oops
121 enum ftrace_dump_mode ftrace_dump_on_oops;
123 /* When set, tracing will stop when a WARN*() is hit */
124 int __disable_trace_on_warning;
126 #ifdef CONFIG_TRACE_ENUM_MAP_FILE
127 /* Map of enums to their values, for "enum_map" file */
128 struct trace_enum_map_head {
130 unsigned long length;
133 union trace_enum_map_item;
135 struct trace_enum_map_tail {
137 * "end" is first and points to NULL as it must be different
138 * than "mod" or "enum_string"
140 union trace_enum_map_item *next;
141 const char *end; /* points to NULL */
144 static DEFINE_MUTEX(trace_enum_mutex);
147 * The trace_enum_maps are saved in an array with two extra elements,
148 * one at the beginning, and one at the end. The beginning item contains
149 * the count of the saved maps (head.length), and the module they
150 * belong to if not built in (head.mod). The ending item contains a
151 * pointer to the next array of saved enum_map items.
153 union trace_enum_map_item {
154 struct trace_enum_map map;
155 struct trace_enum_map_head head;
156 struct trace_enum_map_tail tail;
159 static union trace_enum_map_item *trace_enum_maps;
160 #endif /* CONFIG_TRACE_ENUM_MAP_FILE */
162 static int tracing_set_tracer(struct trace_array *tr, const char *buf);
164 #define MAX_TRACER_SIZE 100
165 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
166 static char *default_bootup_tracer;
168 static bool allocate_snapshot;
170 static int __init set_cmdline_ftrace(char *str)
172 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
173 default_bootup_tracer = bootup_tracer_buf;
174 /* We are using ftrace early, expand it */
175 ring_buffer_expanded = true;
178 __setup("ftrace=", set_cmdline_ftrace);
180 static int __init set_ftrace_dump_on_oops(char *str)
182 if (*str++ != '=' || !*str) {
183 ftrace_dump_on_oops = DUMP_ALL;
187 if (!strcmp("orig_cpu", str)) {
188 ftrace_dump_on_oops = DUMP_ORIG;
194 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
196 static int __init stop_trace_on_warning(char *str)
198 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
199 __disable_trace_on_warning = 1;
202 __setup("traceoff_on_warning", stop_trace_on_warning);
204 static int __init boot_alloc_snapshot(char *str)
206 allocate_snapshot = true;
207 /* We also need the main ring buffer expanded */
208 ring_buffer_expanded = true;
211 __setup("alloc_snapshot", boot_alloc_snapshot);
214 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
216 static int __init set_trace_boot_options(char *str)
218 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
221 __setup("trace_options=", set_trace_boot_options);
223 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
224 static char *trace_boot_clock __initdata;
226 static int __init set_trace_boot_clock(char *str)
228 strlcpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
229 trace_boot_clock = trace_boot_clock_buf;
232 __setup("trace_clock=", set_trace_boot_clock);
234 static int __init set_tracepoint_printk(char *str)
236 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
237 tracepoint_printk = 1;
240 __setup("tp_printk", set_tracepoint_printk);
242 unsigned long long ns2usecs(cycle_t nsec)
249 /* trace_flags holds trace_options default values */
250 #define TRACE_DEFAULT_FLAGS \
251 (FUNCTION_DEFAULT_FLAGS | \
252 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
253 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
254 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
255 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS)
257 /* trace_options that are only supported by global_trace */
258 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
259 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
263 * The global_trace is the descriptor that holds the tracing
264 * buffers for the live tracing. For each CPU, it contains
265 * a link list of pages that will store trace entries. The
266 * page descriptor of the pages in the memory is used to hold
267 * the link list by linking the lru item in the page descriptor
268 * to each of the pages in the buffer per CPU.
270 * For each active CPU there is a data field that holds the
271 * pages for the buffer for that CPU. Each CPU has the same number
272 * of pages allocated for its buffer.
274 static struct trace_array global_trace = {
275 .trace_flags = TRACE_DEFAULT_FLAGS,
278 LIST_HEAD(ftrace_trace_arrays);
280 int trace_array_get(struct trace_array *this_tr)
282 struct trace_array *tr;
285 mutex_lock(&trace_types_lock);
286 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
293 mutex_unlock(&trace_types_lock);
298 static void __trace_array_put(struct trace_array *this_tr)
300 WARN_ON(!this_tr->ref);
304 void trace_array_put(struct trace_array *this_tr)
306 mutex_lock(&trace_types_lock);
307 __trace_array_put(this_tr);
308 mutex_unlock(&trace_types_lock);
311 int filter_check_discard(struct trace_event_file *file, void *rec,
312 struct ring_buffer *buffer,
313 struct ring_buffer_event *event)
315 if (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
316 !filter_match_preds(file->filter, rec)) {
317 ring_buffer_discard_commit(buffer, event);
323 EXPORT_SYMBOL_GPL(filter_check_discard);
325 int call_filter_check_discard(struct trace_event_call *call, void *rec,
326 struct ring_buffer *buffer,
327 struct ring_buffer_event *event)
329 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
330 !filter_match_preds(call->filter, rec)) {
331 ring_buffer_discard_commit(buffer, event);
337 EXPORT_SYMBOL_GPL(call_filter_check_discard);
339 static cycle_t buffer_ftrace_now(struct trace_buffer *buf, int cpu)
343 /* Early boot up does not have a buffer yet */
345 return trace_clock_local();
347 ts = ring_buffer_time_stamp(buf->buffer, cpu);
348 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
353 cycle_t ftrace_now(int cpu)
355 return buffer_ftrace_now(&global_trace.trace_buffer, cpu);
359 * tracing_is_enabled - Show if global_trace has been disabled
361 * Shows if the global trace has been enabled or not. It uses the
362 * mirror flag "buffer_disabled" to be used in fast paths such as for
363 * the irqsoff tracer. But it may be inaccurate due to races. If you
364 * need to know the accurate state, use tracing_is_on() which is a little
365 * slower, but accurate.
367 int tracing_is_enabled(void)
370 * For quick access (irqsoff uses this in fast path), just
371 * return the mirror variable of the state of the ring buffer.
372 * It's a little racy, but we don't really care.
375 return !global_trace.buffer_disabled;
379 * trace_buf_size is the size in bytes that is allocated
380 * for a buffer. Note, the number of bytes is always rounded
383 * This number is purposely set to a low number of 16384.
384 * If the dump on oops happens, it will be much appreciated
385 * to not have to wait for all that output. Anyway this can be
386 * boot time and run time configurable.
388 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
390 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
392 /* trace_types holds a link list of available tracers. */
393 static struct tracer *trace_types __read_mostly;
396 * trace_types_lock is used to protect the trace_types list.
398 DEFINE_MUTEX(trace_types_lock);
401 * serialize the access of the ring buffer
403 * ring buffer serializes readers, but it is low level protection.
404 * The validity of the events (which returns by ring_buffer_peek() ..etc)
405 * are not protected by ring buffer.
407 * The content of events may become garbage if we allow other process consumes
408 * these events concurrently:
409 * A) the page of the consumed events may become a normal page
410 * (not reader page) in ring buffer, and this page will be rewrited
411 * by events producer.
412 * B) The page of the consumed events may become a page for splice_read,
413 * and this page will be returned to system.
415 * These primitives allow multi process access to different cpu ring buffer
418 * These primitives don't distinguish read-only and read-consume access.
419 * Multi read-only access are also serialized.
423 static DECLARE_RWSEM(all_cpu_access_lock);
424 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
426 static inline void trace_access_lock(int cpu)
428 if (cpu == RING_BUFFER_ALL_CPUS) {
429 /* gain it for accessing the whole ring buffer. */
430 down_write(&all_cpu_access_lock);
432 /* gain it for accessing a cpu ring buffer. */
434 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
435 down_read(&all_cpu_access_lock);
437 /* Secondly block other access to this @cpu ring buffer. */
438 mutex_lock(&per_cpu(cpu_access_lock, cpu));
442 static inline void trace_access_unlock(int cpu)
444 if (cpu == RING_BUFFER_ALL_CPUS) {
445 up_write(&all_cpu_access_lock);
447 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
448 up_read(&all_cpu_access_lock);
452 static inline void trace_access_lock_init(void)
456 for_each_possible_cpu(cpu)
457 mutex_init(&per_cpu(cpu_access_lock, cpu));
462 static DEFINE_MUTEX(access_lock);
464 static inline void trace_access_lock(int cpu)
467 mutex_lock(&access_lock);
470 static inline void trace_access_unlock(int cpu)
473 mutex_unlock(&access_lock);
476 static inline void trace_access_lock_init(void)
482 #ifdef CONFIG_STACKTRACE
483 static void __ftrace_trace_stack(struct ring_buffer *buffer,
485 int skip, int pc, struct pt_regs *regs);
486 static inline void ftrace_trace_stack(struct trace_array *tr,
487 struct ring_buffer *buffer,
489 int skip, int pc, struct pt_regs *regs);
492 static inline void __ftrace_trace_stack(struct ring_buffer *buffer,
494 int skip, int pc, struct pt_regs *regs)
497 static inline void ftrace_trace_stack(struct trace_array *tr,
498 struct ring_buffer *buffer,
500 int skip, int pc, struct pt_regs *regs)
506 static void tracer_tracing_on(struct trace_array *tr)
508 if (tr->trace_buffer.buffer)
509 ring_buffer_record_on(tr->trace_buffer.buffer);
511 * This flag is looked at when buffers haven't been allocated
512 * yet, or by some tracers (like irqsoff), that just want to
513 * know if the ring buffer has been disabled, but it can handle
514 * races of where it gets disabled but we still do a record.
515 * As the check is in the fast path of the tracers, it is more
516 * important to be fast than accurate.
518 tr->buffer_disabled = 0;
519 /* Make the flag seen by readers */
524 * tracing_on - enable tracing buffers
526 * This function enables tracing buffers that may have been
527 * disabled with tracing_off.
529 void tracing_on(void)
531 tracer_tracing_on(&global_trace);
533 EXPORT_SYMBOL_GPL(tracing_on);
536 * __trace_puts - write a constant string into the trace buffer.
537 * @ip: The address of the caller
538 * @str: The constant string to write
539 * @size: The size of the string.
541 int __trace_puts(unsigned long ip, const char *str, int size)
543 struct ring_buffer_event *event;
544 struct ring_buffer *buffer;
545 struct print_entry *entry;
546 unsigned long irq_flags;
550 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
553 pc = preempt_count();
555 if (unlikely(tracing_selftest_running || tracing_disabled))
558 alloc = sizeof(*entry) + size + 2; /* possible \n added */
560 local_save_flags(irq_flags);
561 buffer = global_trace.trace_buffer.buffer;
562 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
567 entry = ring_buffer_event_data(event);
570 memcpy(&entry->buf, str, size);
572 /* Add a newline if necessary */
573 if (entry->buf[size - 1] != '\n') {
574 entry->buf[size] = '\n';
575 entry->buf[size + 1] = '\0';
577 entry->buf[size] = '\0';
579 __buffer_unlock_commit(buffer, event);
580 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
584 EXPORT_SYMBOL_GPL(__trace_puts);
587 * __trace_bputs - write the pointer to a constant string into trace buffer
588 * @ip: The address of the caller
589 * @str: The constant string to write to the buffer to
591 int __trace_bputs(unsigned long ip, const char *str)
593 struct ring_buffer_event *event;
594 struct ring_buffer *buffer;
595 struct bputs_entry *entry;
596 unsigned long irq_flags;
597 int size = sizeof(struct bputs_entry);
600 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
603 pc = preempt_count();
605 if (unlikely(tracing_selftest_running || tracing_disabled))
608 local_save_flags(irq_flags);
609 buffer = global_trace.trace_buffer.buffer;
610 event = trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
615 entry = ring_buffer_event_data(event);
619 __buffer_unlock_commit(buffer, event);
620 ftrace_trace_stack(&global_trace, buffer, irq_flags, 4, pc, NULL);
624 EXPORT_SYMBOL_GPL(__trace_bputs);
626 #ifdef CONFIG_TRACER_SNAPSHOT
628 * trace_snapshot - take a snapshot of the current buffer.
630 * This causes a swap between the snapshot buffer and the current live
631 * tracing buffer. You can use this to take snapshots of the live
632 * trace when some condition is triggered, but continue to trace.
634 * Note, make sure to allocate the snapshot with either
635 * a tracing_snapshot_alloc(), or by doing it manually
636 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
638 * If the snapshot buffer is not allocated, it will stop tracing.
639 * Basically making a permanent snapshot.
641 void tracing_snapshot(void)
643 struct trace_array *tr = &global_trace;
644 struct tracer *tracer = tr->current_trace;
648 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
649 internal_trace_puts("*** snapshot is being ignored ***\n");
653 if (!tr->allocated_snapshot) {
654 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
655 internal_trace_puts("*** stopping trace here! ***\n");
660 /* Note, snapshot can not be used when the tracer uses it */
661 if (tracer->use_max_tr) {
662 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
663 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
667 local_irq_save(flags);
668 update_max_tr(tr, current, smp_processor_id());
669 local_irq_restore(flags);
671 EXPORT_SYMBOL_GPL(tracing_snapshot);
673 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
674 struct trace_buffer *size_buf, int cpu_id);
675 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val);
677 static int alloc_snapshot(struct trace_array *tr)
681 if (!tr->allocated_snapshot) {
683 /* allocate spare buffer */
684 ret = resize_buffer_duplicate_size(&tr->max_buffer,
685 &tr->trace_buffer, RING_BUFFER_ALL_CPUS);
689 tr->allocated_snapshot = true;
695 static void free_snapshot(struct trace_array *tr)
698 * We don't free the ring buffer. instead, resize it because
699 * The max_tr ring buffer has some state (e.g. ring->clock) and
700 * we want preserve it.
702 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
703 set_buffer_entries(&tr->max_buffer, 1);
704 tracing_reset_online_cpus(&tr->max_buffer);
705 tr->allocated_snapshot = false;
709 * tracing_alloc_snapshot - allocate snapshot buffer.
711 * This only allocates the snapshot buffer if it isn't already
712 * allocated - it doesn't also take a snapshot.
714 * This is meant to be used in cases where the snapshot buffer needs
715 * to be set up for events that can't sleep but need to be able to
716 * trigger a snapshot.
718 int tracing_alloc_snapshot(void)
720 struct trace_array *tr = &global_trace;
723 ret = alloc_snapshot(tr);
728 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
731 * trace_snapshot_alloc - allocate and take a snapshot of the current buffer.
733 * This is similar to trace_snapshot(), but it will allocate the
734 * snapshot buffer if it isn't already allocated. Use this only
735 * where it is safe to sleep, as the allocation may sleep.
737 * This causes a swap between the snapshot buffer and the current live
738 * tracing buffer. You can use this to take snapshots of the live
739 * trace when some condition is triggered, but continue to trace.
741 void tracing_snapshot_alloc(void)
745 ret = tracing_alloc_snapshot();
751 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
753 void tracing_snapshot(void)
755 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
757 EXPORT_SYMBOL_GPL(tracing_snapshot);
758 int tracing_alloc_snapshot(void)
760 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
763 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
764 void tracing_snapshot_alloc(void)
769 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
770 #endif /* CONFIG_TRACER_SNAPSHOT */
772 static void tracer_tracing_off(struct trace_array *tr)
774 if (tr->trace_buffer.buffer)
775 ring_buffer_record_off(tr->trace_buffer.buffer);
777 * This flag is looked at when buffers haven't been allocated
778 * yet, or by some tracers (like irqsoff), that just want to
779 * know if the ring buffer has been disabled, but it can handle
780 * races of where it gets disabled but we still do a record.
781 * As the check is in the fast path of the tracers, it is more
782 * important to be fast than accurate.
784 tr->buffer_disabled = 1;
785 /* Make the flag seen by readers */
790 * tracing_off - turn off tracing buffers
792 * This function stops the tracing buffers from recording data.
793 * It does not disable any overhead the tracers themselves may
794 * be causing. This function simply causes all recording to
795 * the ring buffers to fail.
797 void tracing_off(void)
799 tracer_tracing_off(&global_trace);
801 EXPORT_SYMBOL_GPL(tracing_off);
803 void disable_trace_on_warning(void)
805 if (__disable_trace_on_warning)
810 * tracer_tracing_is_on - show real state of ring buffer enabled
811 * @tr : the trace array to know if ring buffer is enabled
813 * Shows real state of the ring buffer if it is enabled or not.
815 static int tracer_tracing_is_on(struct trace_array *tr)
817 if (tr->trace_buffer.buffer)
818 return ring_buffer_record_is_on(tr->trace_buffer.buffer);
819 return !tr->buffer_disabled;
823 * tracing_is_on - show state of ring buffers enabled
825 int tracing_is_on(void)
827 return tracer_tracing_is_on(&global_trace);
829 EXPORT_SYMBOL_GPL(tracing_is_on);
831 static int __init set_buf_size(char *str)
833 unsigned long buf_size;
837 buf_size = memparse(str, &str);
838 /* nr_entries can not be zero */
841 trace_buf_size = buf_size;
844 __setup("trace_buf_size=", set_buf_size);
846 static int __init set_tracing_thresh(char *str)
848 unsigned long threshold;
853 ret = kstrtoul(str, 0, &threshold);
856 tracing_thresh = threshold * 1000;
859 __setup("tracing_thresh=", set_tracing_thresh);
861 unsigned long nsecs_to_usecs(unsigned long nsecs)
867 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
868 * It uses C(a, b) where 'a' is the enum name and 'b' is the string that
869 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
870 * of strings in the order that the enums were defined.
875 /* These must match the bit postions in trace_iterator_flags */
876 static const char *trace_options[] = {
884 int in_ns; /* is this clock in nanoseconds? */
886 { trace_clock_local, "local", 1 },
887 { trace_clock_global, "global", 1 },
888 { trace_clock_counter, "counter", 0 },
889 { trace_clock_jiffies, "uptime", 0 },
890 { trace_clock, "perf", 1 },
891 { ktime_get_mono_fast_ns, "mono", 1 },
892 { ktime_get_raw_fast_ns, "mono_raw", 1 },
897 * trace_parser_get_init - gets the buffer for trace parser
899 int trace_parser_get_init(struct trace_parser *parser, int size)
901 memset(parser, 0, sizeof(*parser));
903 parser->buffer = kmalloc(size, GFP_KERNEL);
912 * trace_parser_put - frees the buffer for trace parser
914 void trace_parser_put(struct trace_parser *parser)
916 kfree(parser->buffer);
920 * trace_get_user - reads the user input string separated by space
921 * (matched by isspace(ch))
923 * For each string found the 'struct trace_parser' is updated,
924 * and the function returns.
926 * Returns number of bytes read.
928 * See kernel/trace/trace.h for 'struct trace_parser' details.
930 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
931 size_t cnt, loff_t *ppos)
938 trace_parser_clear(parser);
940 ret = get_user(ch, ubuf++);
948 * The parser is not finished with the last write,
949 * continue reading the user input without skipping spaces.
952 /* skip white space */
953 while (cnt && isspace(ch)) {
954 ret = get_user(ch, ubuf++);
961 /* only spaces were written */
971 /* read the non-space input */
972 while (cnt && !isspace(ch)) {
973 if (parser->idx < parser->size - 1)
974 parser->buffer[parser->idx++] = ch;
979 ret = get_user(ch, ubuf++);
986 /* We either got finished input or we have to wait for another call. */
988 parser->buffer[parser->idx] = 0;
989 parser->cont = false;
990 } else if (parser->idx < parser->size - 1) {
992 parser->buffer[parser->idx++] = ch;
1005 /* TODO add a seq_buf_to_buffer() */
1006 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1010 if (trace_seq_used(s) <= s->seq.readpos)
1013 len = trace_seq_used(s) - s->seq.readpos;
1016 memcpy(buf, s->buffer + s->seq.readpos, cnt);
1018 s->seq.readpos += cnt;
1022 unsigned long __read_mostly tracing_thresh;
1024 #ifdef CONFIG_TRACER_MAX_TRACE
1026 * Copy the new maximum trace into the separate maximum-trace
1027 * structure. (this way the maximum trace is permanently saved,
1028 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
1031 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1033 struct trace_buffer *trace_buf = &tr->trace_buffer;
1034 struct trace_buffer *max_buf = &tr->max_buffer;
1035 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1036 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1039 max_buf->time_start = data->preempt_timestamp;
1041 max_data->saved_latency = tr->max_latency;
1042 max_data->critical_start = data->critical_start;
1043 max_data->critical_end = data->critical_end;
1045 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1046 max_data->pid = tsk->pid;
1048 * If tsk == current, then use current_uid(), as that does not use
1049 * RCU. The irq tracer can be called out of RCU scope.
1052 max_data->uid = current_uid();
1054 max_data->uid = task_uid(tsk);
1056 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1057 max_data->policy = tsk->policy;
1058 max_data->rt_priority = tsk->rt_priority;
1060 /* record this tasks comm */
1061 tracing_record_cmdline(tsk);
1065 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1067 * @tsk: the task with the latency
1068 * @cpu: The cpu that initiated the trace.
1070 * Flip the buffers between the @tr and the max_tr and record information
1071 * about which task was the cause of this latency.
1074 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1076 struct ring_buffer *buf;
1081 WARN_ON_ONCE(!irqs_disabled());
1083 if (!tr->allocated_snapshot) {
1084 /* Only the nop tracer should hit this when disabling */
1085 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1089 arch_spin_lock(&tr->max_lock);
1091 buf = tr->trace_buffer.buffer;
1092 tr->trace_buffer.buffer = tr->max_buffer.buffer;
1093 tr->max_buffer.buffer = buf;
1095 __update_max_tr(tr, tsk, cpu);
1096 arch_spin_unlock(&tr->max_lock);
1100 * update_max_tr_single - only copy one trace over, and reset the rest
1102 * @tsk - task with the latency
1103 * @cpu - the cpu of the buffer to copy.
1105 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1108 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1115 WARN_ON_ONCE(!irqs_disabled());
1116 if (!tr->allocated_snapshot) {
1117 /* Only the nop tracer should hit this when disabling */
1118 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1122 arch_spin_lock(&tr->max_lock);
1124 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
1126 if (ret == -EBUSY) {
1128 * We failed to swap the buffer due to a commit taking
1129 * place on this CPU. We fail to record, but we reset
1130 * the max trace buffer (no one writes directly to it)
1131 * and flag that it failed.
1133 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1134 "Failed to swap buffers due to commit in progress\n");
1137 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1139 __update_max_tr(tr, tsk, cpu);
1140 arch_spin_unlock(&tr->max_lock);
1142 #endif /* CONFIG_TRACER_MAX_TRACE */
1144 static int wait_on_pipe(struct trace_iterator *iter, bool full)
1146 /* Iterators are static, they should be filled or empty */
1147 if (trace_buffer_iter(iter, iter->cpu_file))
1150 return ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file,
1154 #ifdef CONFIG_FTRACE_STARTUP_TEST
1155 static int run_tracer_selftest(struct tracer *type)
1157 struct trace_array *tr = &global_trace;
1158 struct tracer *saved_tracer = tr->current_trace;
1161 if (!type->selftest || tracing_selftest_disabled)
1165 * Run a selftest on this tracer.
1166 * Here we reset the trace buffer, and set the current
1167 * tracer to be this tracer. The tracer can then run some
1168 * internal tracing to verify that everything is in order.
1169 * If we fail, we do not register this tracer.
1171 tracing_reset_online_cpus(&tr->trace_buffer);
1173 tr->current_trace = type;
1175 #ifdef CONFIG_TRACER_MAX_TRACE
1176 if (type->use_max_tr) {
1177 /* If we expanded the buffers, make sure the max is expanded too */
1178 if (ring_buffer_expanded)
1179 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1180 RING_BUFFER_ALL_CPUS);
1181 tr->allocated_snapshot = true;
1185 /* the test is responsible for initializing and enabling */
1186 pr_info("Testing tracer %s: ", type->name);
1187 ret = type->selftest(type, tr);
1188 /* the test is responsible for resetting too */
1189 tr->current_trace = saved_tracer;
1191 printk(KERN_CONT "FAILED!\n");
1192 /* Add the warning after printing 'FAILED' */
1196 /* Only reset on passing, to avoid touching corrupted buffers */
1197 tracing_reset_online_cpus(&tr->trace_buffer);
1199 #ifdef CONFIG_TRACER_MAX_TRACE
1200 if (type->use_max_tr) {
1201 tr->allocated_snapshot = false;
1203 /* Shrink the max buffer again */
1204 if (ring_buffer_expanded)
1205 ring_buffer_resize(tr->max_buffer.buffer, 1,
1206 RING_BUFFER_ALL_CPUS);
1210 printk(KERN_CONT "PASSED\n");
1214 static inline int run_tracer_selftest(struct tracer *type)
1218 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1220 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
1222 static void __init apply_trace_boot_options(void);
1225 * register_tracer - register a tracer with the ftrace system.
1226 * @type - the plugin for the tracer
1228 * Register a new plugin tracer.
1230 int __init register_tracer(struct tracer *type)
1236 pr_info("Tracer must have a name\n");
1240 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1241 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1245 mutex_lock(&trace_types_lock);
1247 tracing_selftest_running = true;
1249 for (t = trace_types; t; t = t->next) {
1250 if (strcmp(type->name, t->name) == 0) {
1252 pr_info("Tracer %s already registered\n",
1259 if (!type->set_flag)
1260 type->set_flag = &dummy_set_flag;
1262 type->flags = &dummy_tracer_flags;
1264 if (!type->flags->opts)
1265 type->flags->opts = dummy_tracer_opt;
1267 ret = run_tracer_selftest(type);
1271 type->next = trace_types;
1273 add_tracer_options(&global_trace, type);
1276 tracing_selftest_running = false;
1277 mutex_unlock(&trace_types_lock);
1279 if (ret || !default_bootup_tracer)
1282 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1285 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1286 /* Do we want this tracer to start on bootup? */
1287 tracing_set_tracer(&global_trace, type->name);
1288 default_bootup_tracer = NULL;
1290 apply_trace_boot_options();
1292 /* disable other selftests, since this will break it. */
1293 tracing_selftest_disabled = true;
1294 #ifdef CONFIG_FTRACE_STARTUP_TEST
1295 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1303 void tracing_reset(struct trace_buffer *buf, int cpu)
1305 struct ring_buffer *buffer = buf->buffer;
1310 ring_buffer_record_disable(buffer);
1312 /* Make sure all commits have finished */
1313 synchronize_sched();
1314 ring_buffer_reset_cpu(buffer, cpu);
1316 ring_buffer_record_enable(buffer);
1319 void tracing_reset_online_cpus(struct trace_buffer *buf)
1321 struct ring_buffer *buffer = buf->buffer;
1327 ring_buffer_record_disable(buffer);
1329 /* Make sure all commits have finished */
1330 synchronize_sched();
1332 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
1334 for_each_online_cpu(cpu)
1335 ring_buffer_reset_cpu(buffer, cpu);
1337 ring_buffer_record_enable(buffer);
1340 /* Must have trace_types_lock held */
1341 void tracing_reset_all_online_cpus(void)
1343 struct trace_array *tr;
1345 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1346 tracing_reset_online_cpus(&tr->trace_buffer);
1347 #ifdef CONFIG_TRACER_MAX_TRACE
1348 tracing_reset_online_cpus(&tr->max_buffer);
1353 #define SAVED_CMDLINES_DEFAULT 128
1354 #define NO_CMDLINE_MAP UINT_MAX
1355 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1356 struct saved_cmdlines_buffer {
1357 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1358 unsigned *map_cmdline_to_pid;
1359 unsigned cmdline_num;
1361 char *saved_cmdlines;
1363 static struct saved_cmdlines_buffer *savedcmd;
1365 /* temporary disable recording */
1366 static atomic_t trace_record_cmdline_disabled __read_mostly;
1368 static inline char *get_saved_cmdlines(int idx)
1370 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
1373 static inline void set_cmdline(int idx, const char *cmdline)
1375 memcpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
1378 static int allocate_cmdlines_buffer(unsigned int val,
1379 struct saved_cmdlines_buffer *s)
1381 s->map_cmdline_to_pid = kmalloc(val * sizeof(*s->map_cmdline_to_pid),
1383 if (!s->map_cmdline_to_pid)
1386 s->saved_cmdlines = kmalloc(val * TASK_COMM_LEN, GFP_KERNEL);
1387 if (!s->saved_cmdlines) {
1388 kfree(s->map_cmdline_to_pid);
1393 s->cmdline_num = val;
1394 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
1395 sizeof(s->map_pid_to_cmdline));
1396 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
1397 val * sizeof(*s->map_cmdline_to_pid));
1402 static int trace_create_savedcmd(void)
1406 savedcmd = kmalloc(sizeof(*savedcmd), GFP_KERNEL);
1410 ret = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT, savedcmd);
1420 int is_tracing_stopped(void)
1422 return global_trace.stop_count;
1426 * tracing_start - quick start of the tracer
1428 * If tracing is enabled but was stopped by tracing_stop,
1429 * this will start the tracer back up.
1431 void tracing_start(void)
1433 struct ring_buffer *buffer;
1434 unsigned long flags;
1436 if (tracing_disabled)
1439 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1440 if (--global_trace.stop_count) {
1441 if (global_trace.stop_count < 0) {
1442 /* Someone screwed up their debugging */
1444 global_trace.stop_count = 0;
1449 /* Prevent the buffers from switching */
1450 arch_spin_lock(&global_trace.max_lock);
1452 buffer = global_trace.trace_buffer.buffer;
1454 ring_buffer_record_enable(buffer);
1456 #ifdef CONFIG_TRACER_MAX_TRACE
1457 buffer = global_trace.max_buffer.buffer;
1459 ring_buffer_record_enable(buffer);
1462 arch_spin_unlock(&global_trace.max_lock);
1465 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1468 static void tracing_start_tr(struct trace_array *tr)
1470 struct ring_buffer *buffer;
1471 unsigned long flags;
1473 if (tracing_disabled)
1476 /* If global, we need to also start the max tracer */
1477 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1478 return tracing_start();
1480 raw_spin_lock_irqsave(&tr->start_lock, flags);
1482 if (--tr->stop_count) {
1483 if (tr->stop_count < 0) {
1484 /* Someone screwed up their debugging */
1491 buffer = tr->trace_buffer.buffer;
1493 ring_buffer_record_enable(buffer);
1496 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1500 * tracing_stop - quick stop of the tracer
1502 * Light weight way to stop tracing. Use in conjunction with
1505 void tracing_stop(void)
1507 struct ring_buffer *buffer;
1508 unsigned long flags;
1510 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1511 if (global_trace.stop_count++)
1514 /* Prevent the buffers from switching */
1515 arch_spin_lock(&global_trace.max_lock);
1517 buffer = global_trace.trace_buffer.buffer;
1519 ring_buffer_record_disable(buffer);
1521 #ifdef CONFIG_TRACER_MAX_TRACE
1522 buffer = global_trace.max_buffer.buffer;
1524 ring_buffer_record_disable(buffer);
1527 arch_spin_unlock(&global_trace.max_lock);
1530 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1533 static void tracing_stop_tr(struct trace_array *tr)
1535 struct ring_buffer *buffer;
1536 unsigned long flags;
1538 /* If global, we need to also stop the max tracer */
1539 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1540 return tracing_stop();
1542 raw_spin_lock_irqsave(&tr->start_lock, flags);
1543 if (tr->stop_count++)
1546 buffer = tr->trace_buffer.buffer;
1548 ring_buffer_record_disable(buffer);
1551 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1554 void trace_stop_cmdline_recording(void);
1556 static int trace_save_cmdline(struct task_struct *tsk)
1560 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1564 * It's not the end of the world if we don't get
1565 * the lock, but we also don't want to spin
1566 * nor do we want to disable interrupts,
1567 * so if we miss here, then better luck next time.
1569 if (!arch_spin_trylock(&trace_cmdline_lock))
1572 idx = savedcmd->map_pid_to_cmdline[tsk->pid];
1573 if (idx == NO_CMDLINE_MAP) {
1574 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
1577 * Check whether the cmdline buffer at idx has a pid
1578 * mapped. We are going to overwrite that entry so we
1579 * need to clear the map_pid_to_cmdline. Otherwise we
1580 * would read the new comm for the old pid.
1582 pid = savedcmd->map_cmdline_to_pid[idx];
1583 if (pid != NO_CMDLINE_MAP)
1584 savedcmd->map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1586 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
1587 savedcmd->map_pid_to_cmdline[tsk->pid] = idx;
1589 savedcmd->cmdline_idx = idx;
1592 set_cmdline(idx, tsk->comm);
1594 arch_spin_unlock(&trace_cmdline_lock);
1599 static void __trace_find_cmdline(int pid, char comm[])
1604 strcpy(comm, "<idle>");
1608 if (WARN_ON_ONCE(pid < 0)) {
1609 strcpy(comm, "<XXX>");
1613 if (pid > PID_MAX_DEFAULT) {
1614 strcpy(comm, "<...>");
1618 map = savedcmd->map_pid_to_cmdline[pid];
1619 if (map != NO_CMDLINE_MAP)
1620 strcpy(comm, get_saved_cmdlines(map));
1622 strcpy(comm, "<...>");
1625 void trace_find_cmdline(int pid, char comm[])
1628 arch_spin_lock(&trace_cmdline_lock);
1630 __trace_find_cmdline(pid, comm);
1632 arch_spin_unlock(&trace_cmdline_lock);
1636 void tracing_record_cmdline(struct task_struct *tsk)
1638 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1641 if (!__this_cpu_read(trace_cmdline_save))
1644 if (trace_save_cmdline(tsk))
1645 __this_cpu_write(trace_cmdline_save, false);
1649 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1652 struct task_struct *tsk = current;
1654 entry->preempt_count = pc & 0xff;
1655 entry->pid = (tsk) ? tsk->pid : 0;
1657 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1658 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1660 TRACE_FLAG_IRQS_NOSUPPORT |
1662 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1663 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1664 (tif_need_resched() ? TRACE_FLAG_NEED_RESCHED : 0) |
1665 (test_preempt_need_resched() ? TRACE_FLAG_PREEMPT_RESCHED : 0);
1667 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1669 struct ring_buffer_event *
1670 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1673 unsigned long flags, int pc)
1675 struct ring_buffer_event *event;
1677 event = ring_buffer_lock_reserve(buffer, len);
1678 if (event != NULL) {
1679 struct trace_entry *ent = ring_buffer_event_data(event);
1681 tracing_generic_entry_update(ent, flags, pc);
1689 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1691 __this_cpu_write(trace_cmdline_save, true);
1692 ring_buffer_unlock_commit(buffer, event);
1695 void trace_buffer_unlock_commit(struct trace_array *tr,
1696 struct ring_buffer *buffer,
1697 struct ring_buffer_event *event,
1698 unsigned long flags, int pc)
1700 __buffer_unlock_commit(buffer, event);
1702 ftrace_trace_stack(tr, buffer, flags, 6, pc, NULL);
1703 ftrace_trace_userstack(buffer, flags, pc);
1705 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1707 static struct ring_buffer *temp_buffer;
1709 struct ring_buffer_event *
1710 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1711 struct trace_event_file *trace_file,
1712 int type, unsigned long len,
1713 unsigned long flags, int pc)
1715 struct ring_buffer_event *entry;
1717 *current_rb = trace_file->tr->trace_buffer.buffer;
1718 entry = trace_buffer_lock_reserve(*current_rb,
1719 type, len, flags, pc);
1721 * If tracing is off, but we have triggers enabled
1722 * we still need to look at the event data. Use the temp_buffer
1723 * to store the trace event for the tigger to use. It's recusive
1724 * safe and will not be recorded anywhere.
1726 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
1727 *current_rb = temp_buffer;
1728 entry = trace_buffer_lock_reserve(*current_rb,
1729 type, len, flags, pc);
1733 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1735 struct ring_buffer_event *
1736 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1737 int type, unsigned long len,
1738 unsigned long flags, int pc)
1740 *current_rb = global_trace.trace_buffer.buffer;
1741 return trace_buffer_lock_reserve(*current_rb,
1742 type, len, flags, pc);
1744 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1746 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
1747 struct ring_buffer *buffer,
1748 struct ring_buffer_event *event,
1749 unsigned long flags, int pc,
1750 struct pt_regs *regs)
1752 __buffer_unlock_commit(buffer, event);
1754 ftrace_trace_stack(tr, buffer, flags, 6, pc, regs);
1755 ftrace_trace_userstack(buffer, flags, pc);
1757 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1759 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1760 struct ring_buffer_event *event)
1762 ring_buffer_discard_commit(buffer, event);
1764 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1767 trace_function(struct trace_array *tr,
1768 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1771 struct trace_event_call *call = &event_function;
1772 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1773 struct ring_buffer_event *event;
1774 struct ftrace_entry *entry;
1776 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1780 entry = ring_buffer_event_data(event);
1782 entry->parent_ip = parent_ip;
1784 if (!call_filter_check_discard(call, entry, buffer, event))
1785 __buffer_unlock_commit(buffer, event);
1788 #ifdef CONFIG_STACKTRACE
1790 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1791 struct ftrace_stack {
1792 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1795 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1796 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1798 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1799 unsigned long flags,
1800 int skip, int pc, struct pt_regs *regs)
1802 struct trace_event_call *call = &event_kernel_stack;
1803 struct ring_buffer_event *event;
1804 struct stack_entry *entry;
1805 struct stack_trace trace;
1807 int size = FTRACE_STACK_ENTRIES;
1809 trace.nr_entries = 0;
1813 * Since events can happen in NMIs there's no safe way to
1814 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1815 * or NMI comes in, it will just have to use the default
1816 * FTRACE_STACK_SIZE.
1818 preempt_disable_notrace();
1820 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1822 * We don't need any atomic variables, just a barrier.
1823 * If an interrupt comes in, we don't care, because it would
1824 * have exited and put the counter back to what we want.
1825 * We just need a barrier to keep gcc from moving things
1829 if (use_stack == 1) {
1830 trace.entries = this_cpu_ptr(ftrace_stack.calls);
1831 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1834 save_stack_trace_regs(regs, &trace);
1836 save_stack_trace(&trace);
1838 if (trace.nr_entries > size)
1839 size = trace.nr_entries;
1841 /* From now on, use_stack is a boolean */
1844 size *= sizeof(unsigned long);
1846 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1847 sizeof(*entry) + size, flags, pc);
1850 entry = ring_buffer_event_data(event);
1852 memset(&entry->caller, 0, size);
1855 memcpy(&entry->caller, trace.entries,
1856 trace.nr_entries * sizeof(unsigned long));
1858 trace.max_entries = FTRACE_STACK_ENTRIES;
1859 trace.entries = entry->caller;
1861 save_stack_trace_regs(regs, &trace);
1863 save_stack_trace(&trace);
1866 entry->size = trace.nr_entries;
1868 if (!call_filter_check_discard(call, entry, buffer, event))
1869 __buffer_unlock_commit(buffer, event);
1872 /* Again, don't let gcc optimize things here */
1874 __this_cpu_dec(ftrace_stack_reserve);
1875 preempt_enable_notrace();
1879 static inline void ftrace_trace_stack(struct trace_array *tr,
1880 struct ring_buffer *buffer,
1881 unsigned long flags,
1882 int skip, int pc, struct pt_regs *regs)
1884 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
1887 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1890 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1893 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1897 * trace_dump_stack - record a stack back trace in the trace buffer
1898 * @skip: Number of functions to skip (helper handlers)
1900 void trace_dump_stack(int skip)
1902 unsigned long flags;
1904 if (tracing_disabled || tracing_selftest_running)
1907 local_save_flags(flags);
1910 * Skip 3 more, seems to get us at the caller of
1914 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
1915 flags, skip, preempt_count(), NULL);
1918 static DEFINE_PER_CPU(int, user_stack_count);
1921 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1923 struct trace_event_call *call = &event_user_stack;
1924 struct ring_buffer_event *event;
1925 struct userstack_entry *entry;
1926 struct stack_trace trace;
1928 if (!(global_trace.trace_flags & TRACE_ITER_USERSTACKTRACE))
1932 * NMIs can not handle page faults, even with fix ups.
1933 * The save user stack can (and often does) fault.
1935 if (unlikely(in_nmi()))
1939 * prevent recursion, since the user stack tracing may
1940 * trigger other kernel events.
1943 if (__this_cpu_read(user_stack_count))
1946 __this_cpu_inc(user_stack_count);
1948 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1949 sizeof(*entry), flags, pc);
1951 goto out_drop_count;
1952 entry = ring_buffer_event_data(event);
1954 entry->tgid = current->tgid;
1955 memset(&entry->caller, 0, sizeof(entry->caller));
1957 trace.nr_entries = 0;
1958 trace.max_entries = FTRACE_STACK_ENTRIES;
1960 trace.entries = entry->caller;
1962 save_stack_trace_user(&trace);
1963 if (!call_filter_check_discard(call, entry, buffer, event))
1964 __buffer_unlock_commit(buffer, event);
1967 __this_cpu_dec(user_stack_count);
1973 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1975 ftrace_trace_userstack(tr, flags, preempt_count());
1979 #endif /* CONFIG_STACKTRACE */
1981 /* created for use with alloc_percpu */
1982 struct trace_buffer_struct {
1983 char buffer[TRACE_BUF_SIZE];
1986 static struct trace_buffer_struct *trace_percpu_buffer;
1987 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1988 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1989 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1992 * The buffer used is dependent on the context. There is a per cpu
1993 * buffer for normal context, softirq contex, hard irq context and
1994 * for NMI context. Thise allows for lockless recording.
1996 * Note, if the buffers failed to be allocated, then this returns NULL
1998 static char *get_trace_buf(void)
2000 struct trace_buffer_struct *percpu_buffer;
2003 * If we have allocated per cpu buffers, then we do not
2004 * need to do any locking.
2007 percpu_buffer = trace_percpu_nmi_buffer;
2009 percpu_buffer = trace_percpu_irq_buffer;
2010 else if (in_softirq())
2011 percpu_buffer = trace_percpu_sirq_buffer;
2013 percpu_buffer = trace_percpu_buffer;
2018 return this_cpu_ptr(&percpu_buffer->buffer[0]);
2021 static int alloc_percpu_trace_buffer(void)
2023 struct trace_buffer_struct *buffers;
2024 struct trace_buffer_struct *sirq_buffers;
2025 struct trace_buffer_struct *irq_buffers;
2026 struct trace_buffer_struct *nmi_buffers;
2028 buffers = alloc_percpu(struct trace_buffer_struct);
2032 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
2036 irq_buffers = alloc_percpu(struct trace_buffer_struct);
2040 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
2044 trace_percpu_buffer = buffers;
2045 trace_percpu_sirq_buffer = sirq_buffers;
2046 trace_percpu_irq_buffer = irq_buffers;
2047 trace_percpu_nmi_buffer = nmi_buffers;
2052 free_percpu(irq_buffers);
2054 free_percpu(sirq_buffers);
2056 free_percpu(buffers);
2058 WARN(1, "Could not allocate percpu trace_printk buffer");
2062 static int buffers_allocated;
2064 void trace_printk_init_buffers(void)
2066 if (buffers_allocated)
2069 if (alloc_percpu_trace_buffer())
2072 /* trace_printk() is for debug use only. Don't use it in production. */
2075 pr_warning("**********************************************************\n");
2076 pr_warning("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
2077 pr_warning("** **\n");
2078 pr_warning("** trace_printk() being used. Allocating extra memory. **\n");
2079 pr_warning("** **\n");
2080 pr_warning("** This means that this is a DEBUG kernel and it is **\n");
2081 pr_warning("** unsafe for production use. **\n");
2082 pr_warning("** **\n");
2083 pr_warning("** If you see this message and you are not debugging **\n");
2084 pr_warning("** the kernel, report this immediately to your vendor! **\n");
2085 pr_warning("** **\n");
2086 pr_warning("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
2087 pr_warning("**********************************************************\n");
2089 /* Expand the buffers to set size */
2090 tracing_update_buffers();
2092 buffers_allocated = 1;
2095 * trace_printk_init_buffers() can be called by modules.
2096 * If that happens, then we need to start cmdline recording
2097 * directly here. If the global_trace.buffer is already
2098 * allocated here, then this was called by module code.
2100 if (global_trace.trace_buffer.buffer)
2101 tracing_start_cmdline_record();
2104 void trace_printk_start_comm(void)
2106 /* Start tracing comms if trace printk is set */
2107 if (!buffers_allocated)
2109 tracing_start_cmdline_record();
2112 static void trace_printk_start_stop_comm(int enabled)
2114 if (!buffers_allocated)
2118 tracing_start_cmdline_record();
2120 tracing_stop_cmdline_record();
2124 * trace_vbprintk - write binary msg to tracing buffer
2127 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
2129 struct trace_event_call *call = &event_bprint;
2130 struct ring_buffer_event *event;
2131 struct ring_buffer *buffer;
2132 struct trace_array *tr = &global_trace;
2133 struct bprint_entry *entry;
2134 unsigned long flags;
2136 int len = 0, size, pc;
2138 if (unlikely(tracing_selftest_running || tracing_disabled))
2141 /* Don't pollute graph traces with trace_vprintk internals */
2142 pause_graph_tracing();
2144 pc = preempt_count();
2145 preempt_disable_notrace();
2147 tbuffer = get_trace_buf();
2153 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
2155 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
2158 local_save_flags(flags);
2159 size = sizeof(*entry) + sizeof(u32) * len;
2160 buffer = tr->trace_buffer.buffer;
2161 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
2165 entry = ring_buffer_event_data(event);
2169 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
2170 if (!call_filter_check_discard(call, entry, buffer, event)) {
2171 __buffer_unlock_commit(buffer, event);
2172 ftrace_trace_stack(tr, buffer, flags, 6, pc, NULL);
2176 preempt_enable_notrace();
2177 unpause_graph_tracing();
2181 EXPORT_SYMBOL_GPL(trace_vbprintk);
2184 __trace_array_vprintk(struct ring_buffer *buffer,
2185 unsigned long ip, const char *fmt, va_list args)
2187 struct trace_event_call *call = &event_print;
2188 struct ring_buffer_event *event;
2189 int len = 0, size, pc;
2190 struct print_entry *entry;
2191 unsigned long flags;
2194 if (tracing_disabled || tracing_selftest_running)
2197 /* Don't pollute graph traces with trace_vprintk internals */
2198 pause_graph_tracing();
2200 pc = preempt_count();
2201 preempt_disable_notrace();
2204 tbuffer = get_trace_buf();
2210 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
2212 local_save_flags(flags);
2213 size = sizeof(*entry) + len + 1;
2214 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
2218 entry = ring_buffer_event_data(event);
2221 memcpy(&entry->buf, tbuffer, len + 1);
2222 if (!call_filter_check_discard(call, entry, buffer, event)) {
2223 __buffer_unlock_commit(buffer, event);
2224 ftrace_trace_stack(&global_trace, buffer, flags, 6, pc, NULL);
2227 preempt_enable_notrace();
2228 unpause_graph_tracing();
2233 int trace_array_vprintk(struct trace_array *tr,
2234 unsigned long ip, const char *fmt, va_list args)
2236 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
2239 int trace_array_printk(struct trace_array *tr,
2240 unsigned long ip, const char *fmt, ...)
2245 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
2249 ret = trace_array_vprintk(tr, ip, fmt, ap);
2254 int trace_array_printk_buf(struct ring_buffer *buffer,
2255 unsigned long ip, const char *fmt, ...)
2260 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
2264 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2269 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2271 return trace_array_vprintk(&global_trace, ip, fmt, args);
2273 EXPORT_SYMBOL_GPL(trace_vprintk);
2275 static void trace_iterator_increment(struct trace_iterator *iter)
2277 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2281 ring_buffer_read(buf_iter, NULL);
2284 static struct trace_entry *
2285 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2286 unsigned long *lost_events)
2288 struct ring_buffer_event *event;
2289 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2292 event = ring_buffer_iter_peek(buf_iter, ts);
2294 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2298 iter->ent_size = ring_buffer_event_length(event);
2299 return ring_buffer_event_data(event);
2305 static struct trace_entry *
2306 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2307 unsigned long *missing_events, u64 *ent_ts)
2309 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2310 struct trace_entry *ent, *next = NULL;
2311 unsigned long lost_events = 0, next_lost = 0;
2312 int cpu_file = iter->cpu_file;
2313 u64 next_ts = 0, ts;
2319 * If we are in a per_cpu trace file, don't bother by iterating over
2320 * all cpu and peek directly.
2322 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2323 if (ring_buffer_empty_cpu(buffer, cpu_file))
2325 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2327 *ent_cpu = cpu_file;
2332 for_each_tracing_cpu(cpu) {
2334 if (ring_buffer_empty_cpu(buffer, cpu))
2337 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2340 * Pick the entry with the smallest timestamp:
2342 if (ent && (!next || ts < next_ts)) {
2346 next_lost = lost_events;
2347 next_size = iter->ent_size;
2351 iter->ent_size = next_size;
2354 *ent_cpu = next_cpu;
2360 *missing_events = next_lost;
2365 /* Find the next real entry, without updating the iterator itself */
2366 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2367 int *ent_cpu, u64 *ent_ts)
2369 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2372 /* Find the next real entry, and increment the iterator to the next entry */
2373 void *trace_find_next_entry_inc(struct trace_iterator *iter)
2375 iter->ent = __find_next_entry(iter, &iter->cpu,
2376 &iter->lost_events, &iter->ts);
2379 trace_iterator_increment(iter);
2381 return iter->ent ? iter : NULL;
2384 static void trace_consume(struct trace_iterator *iter)
2386 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2387 &iter->lost_events);
2390 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2392 struct trace_iterator *iter = m->private;
2396 WARN_ON_ONCE(iter->leftover);
2400 /* can't go backwards */
2405 ent = trace_find_next_entry_inc(iter);
2409 while (ent && iter->idx < i)
2410 ent = trace_find_next_entry_inc(iter);
2417 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2419 struct ring_buffer_event *event;
2420 struct ring_buffer_iter *buf_iter;
2421 unsigned long entries = 0;
2424 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2426 buf_iter = trace_buffer_iter(iter, cpu);
2430 ring_buffer_iter_reset(buf_iter);
2433 * We could have the case with the max latency tracers
2434 * that a reset never took place on a cpu. This is evident
2435 * by the timestamp being before the start of the buffer.
2437 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2438 if (ts >= iter->trace_buffer->time_start)
2441 ring_buffer_read(buf_iter, NULL);
2444 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2448 * The current tracer is copied to avoid a global locking
2451 static void *s_start(struct seq_file *m, loff_t *pos)
2453 struct trace_iterator *iter = m->private;
2454 struct trace_array *tr = iter->tr;
2455 int cpu_file = iter->cpu_file;
2461 * copy the tracer to avoid using a global lock all around.
2462 * iter->trace is a copy of current_trace, the pointer to the
2463 * name may be used instead of a strcmp(), as iter->trace->name
2464 * will point to the same string as current_trace->name.
2466 mutex_lock(&trace_types_lock);
2467 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2468 *iter->trace = *tr->current_trace;
2469 mutex_unlock(&trace_types_lock);
2471 #ifdef CONFIG_TRACER_MAX_TRACE
2472 if (iter->snapshot && iter->trace->use_max_tr)
2473 return ERR_PTR(-EBUSY);
2476 if (!iter->snapshot)
2477 atomic_inc(&trace_record_cmdline_disabled);
2479 if (*pos != iter->pos) {
2484 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2485 for_each_tracing_cpu(cpu)
2486 tracing_iter_reset(iter, cpu);
2488 tracing_iter_reset(iter, cpu_file);
2491 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2496 * If we overflowed the seq_file before, then we want
2497 * to just reuse the trace_seq buffer again.
2503 p = s_next(m, p, &l);
2507 trace_event_read_lock();
2508 trace_access_lock(cpu_file);
2512 static void s_stop(struct seq_file *m, void *p)
2514 struct trace_iterator *iter = m->private;
2516 #ifdef CONFIG_TRACER_MAX_TRACE
2517 if (iter->snapshot && iter->trace->use_max_tr)
2521 if (!iter->snapshot)
2522 atomic_dec(&trace_record_cmdline_disabled);
2524 trace_access_unlock(iter->cpu_file);
2525 trace_event_read_unlock();
2529 get_total_entries(struct trace_buffer *buf,
2530 unsigned long *total, unsigned long *entries)
2532 unsigned long count;
2538 for_each_tracing_cpu(cpu) {
2539 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2541 * If this buffer has skipped entries, then we hold all
2542 * entries for the trace and we need to ignore the
2543 * ones before the time stamp.
2545 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2546 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2547 /* total is the same as the entries */
2551 ring_buffer_overrun_cpu(buf->buffer, cpu);
2556 static void print_lat_help_header(struct seq_file *m)
2558 seq_puts(m, "# _------=> CPU# \n"
2559 "# / _-----=> irqs-off \n"
2560 "# | / _----=> need-resched \n"
2561 "# || / _---=> hardirq/softirq \n"
2562 "# ||| / _--=> preempt-depth \n"
2564 "# cmd pid ||||| time | caller \n"
2565 "# \\ / ||||| \\ | / \n");
2568 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2570 unsigned long total;
2571 unsigned long entries;
2573 get_total_entries(buf, &total, &entries);
2574 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2575 entries, total, num_online_cpus());
2579 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2581 print_event_info(buf, m);
2582 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n"
2586 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2588 print_event_info(buf, m);
2589 seq_puts(m, "# _-----=> irqs-off\n"
2590 "# / _----=> need-resched\n"
2591 "# | / _---=> hardirq/softirq\n"
2592 "# || / _--=> preempt-depth\n"
2594 "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n"
2595 "# | | | |||| | |\n");
2599 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2601 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
2602 struct trace_buffer *buf = iter->trace_buffer;
2603 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2604 struct tracer *type = iter->trace;
2605 unsigned long entries;
2606 unsigned long total;
2607 const char *name = "preemption";
2611 get_total_entries(buf, &total, &entries);
2613 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2615 seq_puts(m, "# -----------------------------------"
2616 "---------------------------------\n");
2617 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2618 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2619 nsecs_to_usecs(data->saved_latency),
2623 #if defined(CONFIG_PREEMPT_NONE)
2625 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2627 #elif defined(CONFIG_PREEMPT)
2632 /* These are reserved for later use */
2635 seq_printf(m, " #P:%d)\n", num_online_cpus());
2639 seq_puts(m, "# -----------------\n");
2640 seq_printf(m, "# | task: %.16s-%d "
2641 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2642 data->comm, data->pid,
2643 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2644 data->policy, data->rt_priority);
2645 seq_puts(m, "# -----------------\n");
2647 if (data->critical_start) {
2648 seq_puts(m, "# => started at: ");
2649 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2650 trace_print_seq(m, &iter->seq);
2651 seq_puts(m, "\n# => ended at: ");
2652 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2653 trace_print_seq(m, &iter->seq);
2654 seq_puts(m, "\n#\n");
2660 static void test_cpu_buff_start(struct trace_iterator *iter)
2662 struct trace_seq *s = &iter->seq;
2663 struct trace_array *tr = iter->tr;
2665 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
2668 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2671 if (iter->started && cpumask_test_cpu(iter->cpu, iter->started))
2674 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2678 cpumask_set_cpu(iter->cpu, iter->started);
2680 /* Don't print started cpu buffer for the first entry of the trace */
2682 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2686 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2688 struct trace_array *tr = iter->tr;
2689 struct trace_seq *s = &iter->seq;
2690 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
2691 struct trace_entry *entry;
2692 struct trace_event *event;
2696 test_cpu_buff_start(iter);
2698 event = ftrace_find_event(entry->type);
2700 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
2701 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2702 trace_print_lat_context(iter);
2704 trace_print_context(iter);
2707 if (trace_seq_has_overflowed(s))
2708 return TRACE_TYPE_PARTIAL_LINE;
2711 return event->funcs->trace(iter, sym_flags, event);
2713 trace_seq_printf(s, "Unknown type %d\n", entry->type);
2715 return trace_handle_return(s);
2718 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2720 struct trace_array *tr = iter->tr;
2721 struct trace_seq *s = &iter->seq;
2722 struct trace_entry *entry;
2723 struct trace_event *event;
2727 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
2728 trace_seq_printf(s, "%d %d %llu ",
2729 entry->pid, iter->cpu, iter->ts);
2731 if (trace_seq_has_overflowed(s))
2732 return TRACE_TYPE_PARTIAL_LINE;
2734 event = ftrace_find_event(entry->type);
2736 return event->funcs->raw(iter, 0, event);
2738 trace_seq_printf(s, "%d ?\n", entry->type);
2740 return trace_handle_return(s);
2743 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2745 struct trace_array *tr = iter->tr;
2746 struct trace_seq *s = &iter->seq;
2747 unsigned char newline = '\n';
2748 struct trace_entry *entry;
2749 struct trace_event *event;
2753 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
2754 SEQ_PUT_HEX_FIELD(s, entry->pid);
2755 SEQ_PUT_HEX_FIELD(s, iter->cpu);
2756 SEQ_PUT_HEX_FIELD(s, iter->ts);
2757 if (trace_seq_has_overflowed(s))
2758 return TRACE_TYPE_PARTIAL_LINE;
2761 event = ftrace_find_event(entry->type);
2763 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2764 if (ret != TRACE_TYPE_HANDLED)
2768 SEQ_PUT_FIELD(s, newline);
2770 return trace_handle_return(s);
2773 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2775 struct trace_array *tr = iter->tr;
2776 struct trace_seq *s = &iter->seq;
2777 struct trace_entry *entry;
2778 struct trace_event *event;
2782 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
2783 SEQ_PUT_FIELD(s, entry->pid);
2784 SEQ_PUT_FIELD(s, iter->cpu);
2785 SEQ_PUT_FIELD(s, iter->ts);
2786 if (trace_seq_has_overflowed(s))
2787 return TRACE_TYPE_PARTIAL_LINE;
2790 event = ftrace_find_event(entry->type);
2791 return event ? event->funcs->binary(iter, 0, event) :
2795 int trace_empty(struct trace_iterator *iter)
2797 struct ring_buffer_iter *buf_iter;
2800 /* If we are looking at one CPU buffer, only check that one */
2801 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2802 cpu = iter->cpu_file;
2803 buf_iter = trace_buffer_iter(iter, cpu);
2805 if (!ring_buffer_iter_empty(buf_iter))
2808 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2814 for_each_tracing_cpu(cpu) {
2815 buf_iter = trace_buffer_iter(iter, cpu);
2817 if (!ring_buffer_iter_empty(buf_iter))
2820 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2828 /* Called with trace_event_read_lock() held. */
2829 enum print_line_t print_trace_line(struct trace_iterator *iter)
2831 struct trace_array *tr = iter->tr;
2832 unsigned long trace_flags = tr->trace_flags;
2833 enum print_line_t ret;
2835 if (iter->lost_events) {
2836 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2837 iter->cpu, iter->lost_events);
2838 if (trace_seq_has_overflowed(&iter->seq))
2839 return TRACE_TYPE_PARTIAL_LINE;
2842 if (iter->trace && iter->trace->print_line) {
2843 ret = iter->trace->print_line(iter);
2844 if (ret != TRACE_TYPE_UNHANDLED)
2848 if (iter->ent->type == TRACE_BPUTS &&
2849 trace_flags & TRACE_ITER_PRINTK &&
2850 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2851 return trace_print_bputs_msg_only(iter);
2853 if (iter->ent->type == TRACE_BPRINT &&
2854 trace_flags & TRACE_ITER_PRINTK &&
2855 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2856 return trace_print_bprintk_msg_only(iter);
2858 if (iter->ent->type == TRACE_PRINT &&
2859 trace_flags & TRACE_ITER_PRINTK &&
2860 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2861 return trace_print_printk_msg_only(iter);
2863 if (trace_flags & TRACE_ITER_BIN)
2864 return print_bin_fmt(iter);
2866 if (trace_flags & TRACE_ITER_HEX)
2867 return print_hex_fmt(iter);
2869 if (trace_flags & TRACE_ITER_RAW)
2870 return print_raw_fmt(iter);
2872 return print_trace_fmt(iter);
2875 void trace_latency_header(struct seq_file *m)
2877 struct trace_iterator *iter = m->private;
2878 struct trace_array *tr = iter->tr;
2880 /* print nothing if the buffers are empty */
2881 if (trace_empty(iter))
2884 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2885 print_trace_header(m, iter);
2887 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
2888 print_lat_help_header(m);
2891 void trace_default_header(struct seq_file *m)
2893 struct trace_iterator *iter = m->private;
2894 struct trace_array *tr = iter->tr;
2895 unsigned long trace_flags = tr->trace_flags;
2897 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2900 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2901 /* print nothing if the buffers are empty */
2902 if (trace_empty(iter))
2904 print_trace_header(m, iter);
2905 if (!(trace_flags & TRACE_ITER_VERBOSE))
2906 print_lat_help_header(m);
2908 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2909 if (trace_flags & TRACE_ITER_IRQ_INFO)
2910 print_func_help_header_irq(iter->trace_buffer, m);
2912 print_func_help_header(iter->trace_buffer, m);
2917 static void test_ftrace_alive(struct seq_file *m)
2919 if (!ftrace_is_dead())
2921 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
2922 "# MAY BE MISSING FUNCTION EVENTS\n");
2925 #ifdef CONFIG_TRACER_MAX_TRACE
2926 static void show_snapshot_main_help(struct seq_file *m)
2928 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
2929 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
2930 "# Takes a snapshot of the main buffer.\n"
2931 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
2932 "# (Doesn't have to be '2' works with any number that\n"
2933 "# is not a '0' or '1')\n");
2936 static void show_snapshot_percpu_help(struct seq_file *m)
2938 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2939 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2940 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
2941 "# Takes a snapshot of the main buffer for this cpu.\n");
2943 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
2944 "# Must use main snapshot file to allocate.\n");
2946 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
2947 "# (Doesn't have to be '2' works with any number that\n"
2948 "# is not a '0' or '1')\n");
2951 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2953 if (iter->tr->allocated_snapshot)
2954 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
2956 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
2958 seq_puts(m, "# Snapshot commands:\n");
2959 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2960 show_snapshot_main_help(m);
2962 show_snapshot_percpu_help(m);
2965 /* Should never be called */
2966 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2969 static int s_show(struct seq_file *m, void *v)
2971 struct trace_iterator *iter = v;
2974 if (iter->ent == NULL) {
2976 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2978 test_ftrace_alive(m);
2980 if (iter->snapshot && trace_empty(iter))
2981 print_snapshot_help(m, iter);
2982 else if (iter->trace && iter->trace->print_header)
2983 iter->trace->print_header(m);
2985 trace_default_header(m);
2987 } else if (iter->leftover) {
2989 * If we filled the seq_file buffer earlier, we
2990 * want to just show it now.
2992 ret = trace_print_seq(m, &iter->seq);
2994 /* ret should this time be zero, but you never know */
2995 iter->leftover = ret;
2998 print_trace_line(iter);
2999 ret = trace_print_seq(m, &iter->seq);
3001 * If we overflow the seq_file buffer, then it will
3002 * ask us for this data again at start up.
3004 * ret is 0 if seq_file write succeeded.
3007 iter->leftover = ret;
3014 * Should be used after trace_array_get(), trace_types_lock
3015 * ensures that i_cdev was already initialized.
3017 static inline int tracing_get_cpu(struct inode *inode)
3019 if (inode->i_cdev) /* See trace_create_cpu_file() */
3020 return (long)inode->i_cdev - 1;
3021 return RING_BUFFER_ALL_CPUS;
3024 static const struct seq_operations tracer_seq_ops = {
3031 static struct trace_iterator *
3032 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
3034 struct trace_array *tr = inode->i_private;
3035 struct trace_iterator *iter;
3038 if (tracing_disabled)
3039 return ERR_PTR(-ENODEV);
3041 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
3043 return ERR_PTR(-ENOMEM);
3045 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
3047 if (!iter->buffer_iter)
3051 * We make a copy of the current tracer to avoid concurrent
3052 * changes on it while we are reading.
3054 mutex_lock(&trace_types_lock);
3055 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
3059 *iter->trace = *tr->current_trace;
3061 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
3066 #ifdef CONFIG_TRACER_MAX_TRACE
3067 /* Currently only the top directory has a snapshot */
3068 if (tr->current_trace->print_max || snapshot)
3069 iter->trace_buffer = &tr->max_buffer;
3072 iter->trace_buffer = &tr->trace_buffer;
3073 iter->snapshot = snapshot;
3075 iter->cpu_file = tracing_get_cpu(inode);
3076 mutex_init(&iter->mutex);
3078 /* Notify the tracer early; before we stop tracing. */
3079 if (iter->trace && iter->trace->open)
3080 iter->trace->open(iter);
3082 /* Annotate start of buffers if we had overruns */
3083 if (ring_buffer_overruns(iter->trace_buffer->buffer))
3084 iter->iter_flags |= TRACE_FILE_ANNOTATE;
3086 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3087 if (trace_clocks[tr->clock_id].in_ns)
3088 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3090 /* stop the trace while dumping if we are not opening "snapshot" */
3091 if (!iter->snapshot)
3092 tracing_stop_tr(tr);
3094 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
3095 for_each_tracing_cpu(cpu) {
3096 iter->buffer_iter[cpu] =
3097 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
3099 ring_buffer_read_prepare_sync();
3100 for_each_tracing_cpu(cpu) {
3101 ring_buffer_read_start(iter->buffer_iter[cpu]);
3102 tracing_iter_reset(iter, cpu);
3105 cpu = iter->cpu_file;
3106 iter->buffer_iter[cpu] =
3107 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
3108 ring_buffer_read_prepare_sync();
3109 ring_buffer_read_start(iter->buffer_iter[cpu]);
3110 tracing_iter_reset(iter, cpu);
3113 mutex_unlock(&trace_types_lock);
3118 mutex_unlock(&trace_types_lock);
3120 kfree(iter->buffer_iter);
3122 seq_release_private(inode, file);
3123 return ERR_PTR(-ENOMEM);
3126 int tracing_open_generic(struct inode *inode, struct file *filp)
3128 if (tracing_disabled)
3131 filp->private_data = inode->i_private;
3135 bool tracing_is_disabled(void)
3137 return (tracing_disabled) ? true: false;
3141 * Open and update trace_array ref count.
3142 * Must have the current trace_array passed to it.
3144 static int tracing_open_generic_tr(struct inode *inode, struct file *filp)
3146 struct trace_array *tr = inode->i_private;
3148 if (tracing_disabled)
3151 if (trace_array_get(tr) < 0)
3154 filp->private_data = inode->i_private;
3159 static int tracing_release(struct inode *inode, struct file *file)
3161 struct trace_array *tr = inode->i_private;
3162 struct seq_file *m = file->private_data;
3163 struct trace_iterator *iter;
3166 if (!(file->f_mode & FMODE_READ)) {
3167 trace_array_put(tr);
3171 /* Writes do not use seq_file */
3173 mutex_lock(&trace_types_lock);
3175 for_each_tracing_cpu(cpu) {
3176 if (iter->buffer_iter[cpu])
3177 ring_buffer_read_finish(iter->buffer_iter[cpu]);
3180 if (iter->trace && iter->trace->close)
3181 iter->trace->close(iter);
3183 if (!iter->snapshot)
3184 /* reenable tracing if it was previously enabled */
3185 tracing_start_tr(tr);
3187 __trace_array_put(tr);
3189 mutex_unlock(&trace_types_lock);
3191 mutex_destroy(&iter->mutex);
3192 free_cpumask_var(iter->started);
3194 kfree(iter->buffer_iter);
3195 seq_release_private(inode, file);
3200 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
3202 struct trace_array *tr = inode->i_private;
3204 trace_array_put(tr);
3208 static int tracing_single_release_tr(struct inode *inode, struct file *file)
3210 struct trace_array *tr = inode->i_private;
3212 trace_array_put(tr);
3214 return single_release(inode, file);
3217 static int tracing_open(struct inode *inode, struct file *file)
3219 struct trace_array *tr = inode->i_private;
3220 struct trace_iterator *iter;
3223 if (trace_array_get(tr) < 0)
3226 /* If this file was open for write, then erase contents */
3227 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
3228 int cpu = tracing_get_cpu(inode);
3230 if (cpu == RING_BUFFER_ALL_CPUS)
3231 tracing_reset_online_cpus(&tr->trace_buffer);
3233 tracing_reset(&tr->trace_buffer, cpu);
3236 if (file->f_mode & FMODE_READ) {
3237 iter = __tracing_open(inode, file, false);
3239 ret = PTR_ERR(iter);
3240 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
3241 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3245 trace_array_put(tr);
3251 * Some tracers are not suitable for instance buffers.
3252 * A tracer is always available for the global array (toplevel)
3253 * or if it explicitly states that it is.
3256 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
3258 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
3261 /* Find the next tracer that this trace array may use */
3262 static struct tracer *
3263 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
3265 while (t && !trace_ok_for_array(t, tr))
3272 t_next(struct seq_file *m, void *v, loff_t *pos)
3274 struct trace_array *tr = m->private;
3275 struct tracer *t = v;
3280 t = get_tracer_for_array(tr, t->next);
3285 static void *t_start(struct seq_file *m, loff_t *pos)
3287 struct trace_array *tr = m->private;
3291 mutex_lock(&trace_types_lock);
3293 t = get_tracer_for_array(tr, trace_types);
3294 for (; t && l < *pos; t = t_next(m, t, &l))
3300 static void t_stop(struct seq_file *m, void *p)
3302 mutex_unlock(&trace_types_lock);
3305 static int t_show(struct seq_file *m, void *v)
3307 struct tracer *t = v;
3312 seq_puts(m, t->name);
3321 static const struct seq_operations show_traces_seq_ops = {
3328 static int show_traces_open(struct inode *inode, struct file *file)
3330 struct trace_array *tr = inode->i_private;
3334 if (tracing_disabled)
3337 ret = seq_open(file, &show_traces_seq_ops);
3341 m = file->private_data;
3348 tracing_write_stub(struct file *filp, const char __user *ubuf,
3349 size_t count, loff_t *ppos)
3354 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
3358 if (file->f_mode & FMODE_READ)
3359 ret = seq_lseek(file, offset, whence);
3361 file->f_pos = ret = 0;
3366 static const struct file_operations tracing_fops = {
3367 .open = tracing_open,
3369 .write = tracing_write_stub,
3370 .llseek = tracing_lseek,
3371 .release = tracing_release,
3374 static const struct file_operations show_traces_fops = {
3375 .open = show_traces_open,
3377 .release = seq_release,
3378 .llseek = seq_lseek,
3382 * The tracer itself will not take this lock, but still we want
3383 * to provide a consistent cpumask to user-space:
3385 static DEFINE_MUTEX(tracing_cpumask_update_lock);
3388 * Temporary storage for the character representation of the
3389 * CPU bitmask (and one more byte for the newline):
3391 static char mask_str[NR_CPUS + 1];
3394 tracing_cpumask_read(struct file *filp, char __user *ubuf,
3395 size_t count, loff_t *ppos)
3397 struct trace_array *tr = file_inode(filp)->i_private;
3400 mutex_lock(&tracing_cpumask_update_lock);
3402 len = snprintf(mask_str, count, "%*pb\n",
3403 cpumask_pr_args(tr->tracing_cpumask));
3408 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3411 mutex_unlock(&tracing_cpumask_update_lock);
3417 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3418 size_t count, loff_t *ppos)
3420 struct trace_array *tr = file_inode(filp)->i_private;
3421 cpumask_var_t tracing_cpumask_new;
3424 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3427 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3431 mutex_lock(&tracing_cpumask_update_lock);
3433 local_irq_disable();
3434 arch_spin_lock(&tr->max_lock);
3435 for_each_tracing_cpu(cpu) {
3437 * Increase/decrease the disabled counter if we are
3438 * about to flip a bit in the cpumask:
3440 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
3441 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3442 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3443 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3445 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
3446 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3447 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3448 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3451 arch_spin_unlock(&tr->max_lock);
3454 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
3456 mutex_unlock(&tracing_cpumask_update_lock);
3457 free_cpumask_var(tracing_cpumask_new);
3462 free_cpumask_var(tracing_cpumask_new);
3467 static const struct file_operations tracing_cpumask_fops = {
3468 .open = tracing_open_generic_tr,
3469 .read = tracing_cpumask_read,
3470 .write = tracing_cpumask_write,
3471 .release = tracing_release_generic_tr,
3472 .llseek = generic_file_llseek,
3475 static int tracing_trace_options_show(struct seq_file *m, void *v)
3477 struct tracer_opt *trace_opts;
3478 struct trace_array *tr = m->private;
3482 mutex_lock(&trace_types_lock);
3483 tracer_flags = tr->current_trace->flags->val;
3484 trace_opts = tr->current_trace->flags->opts;
3486 for (i = 0; trace_options[i]; i++) {
3487 if (tr->trace_flags & (1 << i))
3488 seq_printf(m, "%s\n", trace_options[i]);
3490 seq_printf(m, "no%s\n", trace_options[i]);
3493 for (i = 0; trace_opts[i].name; i++) {
3494 if (tracer_flags & trace_opts[i].bit)
3495 seq_printf(m, "%s\n", trace_opts[i].name);
3497 seq_printf(m, "no%s\n", trace_opts[i].name);
3499 mutex_unlock(&trace_types_lock);
3504 static int __set_tracer_option(struct trace_array *tr,
3505 struct tracer_flags *tracer_flags,
3506 struct tracer_opt *opts, int neg)
3508 struct tracer *trace = tr->current_trace;
3511 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
3516 tracer_flags->val &= ~opts->bit;
3518 tracer_flags->val |= opts->bit;
3522 /* Try to assign a tracer specific option */
3523 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
3525 struct tracer *trace = tr->current_trace;
3526 struct tracer_flags *tracer_flags = trace->flags;
3527 struct tracer_opt *opts = NULL;
3530 for (i = 0; tracer_flags->opts[i].name; i++) {
3531 opts = &tracer_flags->opts[i];
3533 if (strcmp(cmp, opts->name) == 0)
3534 return __set_tracer_option(tr, trace->flags, opts, neg);
3540 /* Some tracers require overwrite to stay enabled */
3541 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3543 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3549 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3551 /* do nothing if flag is already set */
3552 if (!!(tr->trace_flags & mask) == !!enabled)
3555 /* Give the tracer a chance to approve the change */
3556 if (tr->current_trace->flag_changed)
3557 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
3561 tr->trace_flags |= mask;
3563 tr->trace_flags &= ~mask;
3565 if (mask == TRACE_ITER_RECORD_CMD)
3566 trace_event_enable_cmd_record(enabled);
3568 if (mask == TRACE_ITER_OVERWRITE) {
3569 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3570 #ifdef CONFIG_TRACER_MAX_TRACE
3571 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3575 if (mask == TRACE_ITER_PRINTK) {
3576 trace_printk_start_stop_comm(enabled);
3577 trace_printk_control(enabled);
3583 static int trace_set_options(struct trace_array *tr, char *option)
3589 size_t orig_len = strlen(option);
3591 cmp = strstrip(option);
3593 if (strncmp(cmp, "no", 2) == 0) {
3598 mutex_lock(&trace_types_lock);
3600 for (i = 0; trace_options[i]; i++) {
3601 if (strcmp(cmp, trace_options[i]) == 0) {
3602 ret = set_tracer_flag(tr, 1 << i, !neg);
3607 /* If no option could be set, test the specific tracer options */
3608 if (!trace_options[i])
3609 ret = set_tracer_option(tr, cmp, neg);
3611 mutex_unlock(&trace_types_lock);
3614 * If the first trailing whitespace is replaced with '\0' by strstrip,
3615 * turn it back into a space.
3617 if (orig_len > strlen(option))
3618 option[strlen(option)] = ' ';
3623 static void __init apply_trace_boot_options(void)
3625 char *buf = trace_boot_options_buf;
3629 option = strsep(&buf, ",");
3635 trace_set_options(&global_trace, option);
3637 /* Put back the comma to allow this to be called again */
3644 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3645 size_t cnt, loff_t *ppos)
3647 struct seq_file *m = filp->private_data;
3648 struct trace_array *tr = m->private;
3652 if (cnt >= sizeof(buf))
3655 if (copy_from_user(&buf, ubuf, cnt))
3660 ret = trace_set_options(tr, buf);
3669 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3671 struct trace_array *tr = inode->i_private;
3674 if (tracing_disabled)
3677 if (trace_array_get(tr) < 0)
3680 ret = single_open(file, tracing_trace_options_show, inode->i_private);
3682 trace_array_put(tr);
3687 static const struct file_operations tracing_iter_fops = {
3688 .open = tracing_trace_options_open,
3690 .llseek = seq_lseek,
3691 .release = tracing_single_release_tr,
3692 .write = tracing_trace_options_write,
3695 static const char readme_msg[] =
3696 "tracing mini-HOWTO:\n\n"
3697 "# echo 0 > tracing_on : quick way to disable tracing\n"
3698 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3699 " Important files:\n"
3700 " trace\t\t\t- The static contents of the buffer\n"
3701 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3702 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3703 " current_tracer\t- function and latency tracers\n"
3704 " available_tracers\t- list of configured tracers for current_tracer\n"
3705 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3706 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3707 " trace_clock\t\t-change the clock used to order events\n"
3708 " local: Per cpu clock but may not be synced across CPUs\n"
3709 " global: Synced across CPUs but slows tracing down.\n"
3710 " counter: Not a clock, but just an increment\n"
3711 " uptime: Jiffy counter from time of boot\n"
3712 " perf: Same clock that perf events use\n"
3713 #ifdef CONFIG_X86_64
3714 " x86-tsc: TSC cycle counter\n"
3716 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3717 " tracing_cpumask\t- Limit which CPUs to trace\n"
3718 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3719 "\t\t\t Remove sub-buffer with rmdir\n"
3720 " trace_options\t\t- Set format or modify how tracing happens\n"
3721 "\t\t\t Disable an option by adding a suffix 'no' to the\n"
3722 "\t\t\t option name\n"
3723 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
3724 #ifdef CONFIG_DYNAMIC_FTRACE
3725 "\n available_filter_functions - list of functions that can be filtered on\n"
3726 " set_ftrace_filter\t- echo function name in here to only trace these\n"
3727 "\t\t\t functions\n"
3728 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3729 "\t modules: Can select a group via module\n"
3730 "\t Format: :mod:<module-name>\n"
3731 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
3732 "\t triggers: a command to perform when function is hit\n"
3733 "\t Format: <function>:<trigger>[:count]\n"
3734 "\t trigger: traceon, traceoff\n"
3735 "\t\t enable_event:<system>:<event>\n"
3736 "\t\t disable_event:<system>:<event>\n"
3737 #ifdef CONFIG_STACKTRACE
3740 #ifdef CONFIG_TRACER_SNAPSHOT
3745 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
3746 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
3747 "\t The first one will disable tracing every time do_fault is hit\n"
3748 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
3749 "\t The first time do trap is hit and it disables tracing, the\n"
3750 "\t counter will decrement to 2. If tracing is already disabled,\n"
3751 "\t the counter will not decrement. It only decrements when the\n"
3752 "\t trigger did work\n"
3753 "\t To remove trigger without count:\n"
3754 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
3755 "\t To remove trigger with a count:\n"
3756 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3757 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3758 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3759 "\t modules: Can select a group via module command :mod:\n"
3760 "\t Does not accept triggers\n"
3761 #endif /* CONFIG_DYNAMIC_FTRACE */
3762 #ifdef CONFIG_FUNCTION_TRACER
3763 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
3766 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3767 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3768 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
3769 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3771 #ifdef CONFIG_TRACER_SNAPSHOT
3772 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
3773 "\t\t\t snapshot buffer. Read the contents for more\n"
3774 "\t\t\t information\n"
3776 #ifdef CONFIG_STACK_TRACER
3777 " stack_trace\t\t- Shows the max stack trace when active\n"
3778 " stack_max_size\t- Shows current max stack size that was traced\n"
3779 "\t\t\t Write into this file to reset the max size (trigger a\n"
3780 "\t\t\t new trace)\n"
3781 #ifdef CONFIG_DYNAMIC_FTRACE
3782 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
3785 #endif /* CONFIG_STACK_TRACER */
3786 " events/\t\t- Directory containing all trace event subsystems:\n"
3787 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
3788 " events/<system>/\t- Directory containing all trace events for <system>:\n"
3789 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
3791 " filter\t\t- If set, only events passing filter are traced\n"
3792 " events/<system>/<event>/\t- Directory containing control files for\n"
3794 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
3795 " filter\t\t- If set, only events passing filter are traced\n"
3796 " trigger\t\t- If set, a command to perform when event is hit\n"
3797 "\t Format: <trigger>[:count][if <filter>]\n"
3798 "\t trigger: traceon, traceoff\n"
3799 "\t enable_event:<system>:<event>\n"
3800 "\t disable_event:<system>:<event>\n"
3801 #ifdef CONFIG_STACKTRACE
3804 #ifdef CONFIG_TRACER_SNAPSHOT
3807 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
3808 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
3809 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
3810 "\t events/block/block_unplug/trigger\n"
3811 "\t The first disables tracing every time block_unplug is hit.\n"
3812 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
3813 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
3814 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
3815 "\t Like function triggers, the counter is only decremented if it\n"
3816 "\t enabled or disabled tracing.\n"
3817 "\t To remove a trigger without a count:\n"
3818 "\t echo '!<trigger> > <system>/<event>/trigger\n"
3819 "\t To remove a trigger with a count:\n"
3820 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
3821 "\t Filters can be ignored when removing a trigger.\n"
3825 tracing_readme_read(struct file *filp, char __user *ubuf,
3826 size_t cnt, loff_t *ppos)
3828 return simple_read_from_buffer(ubuf, cnt, ppos,
3829 readme_msg, strlen(readme_msg));
3832 static const struct file_operations tracing_readme_fops = {
3833 .open = tracing_open_generic,
3834 .read = tracing_readme_read,
3835 .llseek = generic_file_llseek,
3838 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
3840 unsigned int *ptr = v;
3842 if (*pos || m->count)
3847 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
3849 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
3858 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
3864 arch_spin_lock(&trace_cmdline_lock);
3866 v = &savedcmd->map_cmdline_to_pid[0];
3868 v = saved_cmdlines_next(m, v, &l);
3876 static void saved_cmdlines_stop(struct seq_file *m, void *v)
3878 arch_spin_unlock(&trace_cmdline_lock);
3882 static int saved_cmdlines_show(struct seq_file *m, void *v)
3884 char buf[TASK_COMM_LEN];
3885 unsigned int *pid = v;
3887 __trace_find_cmdline(*pid, buf);
3888 seq_printf(m, "%d %s\n", *pid, buf);
3892 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
3893 .start = saved_cmdlines_start,
3894 .next = saved_cmdlines_next,
3895 .stop = saved_cmdlines_stop,
3896 .show = saved_cmdlines_show,
3899 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
3901 if (tracing_disabled)
3904 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
3907 static const struct file_operations tracing_saved_cmdlines_fops = {
3908 .open = tracing_saved_cmdlines_open,
3910 .llseek = seq_lseek,
3911 .release = seq_release,
3915 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
3916 size_t cnt, loff_t *ppos)
3921 arch_spin_lock(&trace_cmdline_lock);
3922 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
3923 arch_spin_unlock(&trace_cmdline_lock);
3925 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3928 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
3930 kfree(s->saved_cmdlines);
3931 kfree(s->map_cmdline_to_pid);
3935 static int tracing_resize_saved_cmdlines(unsigned int val)
3937 struct saved_cmdlines_buffer *s, *savedcmd_temp;
3939 s = kmalloc(sizeof(*s), GFP_KERNEL);
3943 if (allocate_cmdlines_buffer(val, s) < 0) {
3948 arch_spin_lock(&trace_cmdline_lock);
3949 savedcmd_temp = savedcmd;
3951 arch_spin_unlock(&trace_cmdline_lock);
3952 free_saved_cmdlines_buffer(savedcmd_temp);
3958 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
3959 size_t cnt, loff_t *ppos)
3964 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3968 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
3969 if (!val || val > PID_MAX_DEFAULT)
3972 ret = tracing_resize_saved_cmdlines((unsigned int)val);
3981 static const struct file_operations tracing_saved_cmdlines_size_fops = {
3982 .open = tracing_open_generic,
3983 .read = tracing_saved_cmdlines_size_read,
3984 .write = tracing_saved_cmdlines_size_write,
3987 #ifdef CONFIG_TRACE_ENUM_MAP_FILE
3988 static union trace_enum_map_item *
3989 update_enum_map(union trace_enum_map_item *ptr)
3991 if (!ptr->map.enum_string) {
3992 if (ptr->tail.next) {
3993 ptr = ptr->tail.next;
3994 /* Set ptr to the next real item (skip head) */
4002 static void *enum_map_next(struct seq_file *m, void *v, loff_t *pos)
4004 union trace_enum_map_item *ptr = v;
4007 * Paranoid! If ptr points to end, we don't want to increment past it.
4008 * This really should never happen.
4010 ptr = update_enum_map(ptr);
4011 if (WARN_ON_ONCE(!ptr))
4018 ptr = update_enum_map(ptr);
4023 static void *enum_map_start(struct seq_file *m, loff_t *pos)
4025 union trace_enum_map_item *v;
4028 mutex_lock(&trace_enum_mutex);
4030 v = trace_enum_maps;
4034 while (v && l < *pos) {
4035 v = enum_map_next(m, v, &l);
4041 static void enum_map_stop(struct seq_file *m, void *v)
4043 mutex_unlock(&trace_enum_mutex);
4046 static int enum_map_show(struct seq_file *m, void *v)
4048 union trace_enum_map_item *ptr = v;
4050 seq_printf(m, "%s %ld (%s)\n",
4051 ptr->map.enum_string, ptr->map.enum_value,
4057 static const struct seq_operations tracing_enum_map_seq_ops = {
4058 .start = enum_map_start,
4059 .next = enum_map_next,
4060 .stop = enum_map_stop,
4061 .show = enum_map_show,
4064 static int tracing_enum_map_open(struct inode *inode, struct file *filp)
4066 if (tracing_disabled)
4069 return seq_open(filp, &tracing_enum_map_seq_ops);
4072 static const struct file_operations tracing_enum_map_fops = {
4073 .open = tracing_enum_map_open,
4075 .llseek = seq_lseek,
4076 .release = seq_release,
4079 static inline union trace_enum_map_item *
4080 trace_enum_jmp_to_tail(union trace_enum_map_item *ptr)
4082 /* Return tail of array given the head */
4083 return ptr + ptr->head.length + 1;
4087 trace_insert_enum_map_file(struct module *mod, struct trace_enum_map **start,
4090 struct trace_enum_map **stop;
4091 struct trace_enum_map **map;
4092 union trace_enum_map_item *map_array;
4093 union trace_enum_map_item *ptr;
4098 * The trace_enum_maps contains the map plus a head and tail item,
4099 * where the head holds the module and length of array, and the
4100 * tail holds a pointer to the next list.
4102 map_array = kmalloc(sizeof(*map_array) * (len + 2), GFP_KERNEL);
4104 pr_warning("Unable to allocate trace enum mapping\n");
4108 mutex_lock(&trace_enum_mutex);
4110 if (!trace_enum_maps)
4111 trace_enum_maps = map_array;
4113 ptr = trace_enum_maps;
4115 ptr = trace_enum_jmp_to_tail(ptr);
4116 if (!ptr->tail.next)
4118 ptr = ptr->tail.next;
4121 ptr->tail.next = map_array;
4123 map_array->head.mod = mod;
4124 map_array->head.length = len;
4127 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
4128 map_array->map = **map;
4131 memset(map_array, 0, sizeof(*map_array));
4133 mutex_unlock(&trace_enum_mutex);
4136 static void trace_create_enum_file(struct dentry *d_tracer)
4138 trace_create_file("enum_map", 0444, d_tracer,
4139 NULL, &tracing_enum_map_fops);
4142 #else /* CONFIG_TRACE_ENUM_MAP_FILE */
4143 static inline void trace_create_enum_file(struct dentry *d_tracer) { }
4144 static inline void trace_insert_enum_map_file(struct module *mod,
4145 struct trace_enum_map **start, int len) { }
4146 #endif /* !CONFIG_TRACE_ENUM_MAP_FILE */
4148 static void trace_insert_enum_map(struct module *mod,
4149 struct trace_enum_map **start, int len)
4151 struct trace_enum_map **map;
4158 trace_event_enum_update(map, len);
4160 trace_insert_enum_map_file(mod, start, len);
4164 tracing_set_trace_read(struct file *filp, char __user *ubuf,
4165 size_t cnt, loff_t *ppos)
4167 struct trace_array *tr = filp->private_data;
4168 char buf[MAX_TRACER_SIZE+2];
4171 mutex_lock(&trace_types_lock);
4172 r = sprintf(buf, "%s\n", tr->current_trace->name);
4173 mutex_unlock(&trace_types_lock);
4175 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4178 int tracer_init(struct tracer *t, struct trace_array *tr)
4180 tracing_reset_online_cpus(&tr->trace_buffer);
4184 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
4188 for_each_tracing_cpu(cpu)
4189 per_cpu_ptr(buf->data, cpu)->entries = val;
4192 #ifdef CONFIG_TRACER_MAX_TRACE
4193 /* resize @tr's buffer to the size of @size_tr's entries */
4194 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
4195 struct trace_buffer *size_buf, int cpu_id)
4199 if (cpu_id == RING_BUFFER_ALL_CPUS) {
4200 for_each_tracing_cpu(cpu) {
4201 ret = ring_buffer_resize(trace_buf->buffer,
4202 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
4205 per_cpu_ptr(trace_buf->data, cpu)->entries =
4206 per_cpu_ptr(size_buf->data, cpu)->entries;
4209 ret = ring_buffer_resize(trace_buf->buffer,
4210 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
4212 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
4213 per_cpu_ptr(size_buf->data, cpu_id)->entries;
4218 #endif /* CONFIG_TRACER_MAX_TRACE */
4220 static int __tracing_resize_ring_buffer(struct trace_array *tr,
4221 unsigned long size, int cpu)
4226 * If kernel or user changes the size of the ring buffer
4227 * we use the size that was given, and we can forget about
4228 * expanding it later.
4230 ring_buffer_expanded = true;
4232 /* May be called before buffers are initialized */
4233 if (!tr->trace_buffer.buffer)
4236 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
4240 #ifdef CONFIG_TRACER_MAX_TRACE
4241 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
4242 !tr->current_trace->use_max_tr)
4245 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
4247 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
4248 &tr->trace_buffer, cpu);
4251 * AARGH! We are left with different
4252 * size max buffer!!!!
4253 * The max buffer is our "snapshot" buffer.
4254 * When a tracer needs a snapshot (one of the
4255 * latency tracers), it swaps the max buffer
4256 * with the saved snap shot. We succeeded to
4257 * update the size of the main buffer, but failed to
4258 * update the size of the max buffer. But when we tried
4259 * to reset the main buffer to the original size, we
4260 * failed there too. This is very unlikely to
4261 * happen, but if it does, warn and kill all
4265 tracing_disabled = 1;
4270 if (cpu == RING_BUFFER_ALL_CPUS)
4271 set_buffer_entries(&tr->max_buffer, size);
4273 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
4276 #endif /* CONFIG_TRACER_MAX_TRACE */
4278 if (cpu == RING_BUFFER_ALL_CPUS)
4279 set_buffer_entries(&tr->trace_buffer, size);
4281 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
4286 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
4287 unsigned long size, int cpu_id)
4291 mutex_lock(&trace_types_lock);
4293 if (cpu_id != RING_BUFFER_ALL_CPUS) {
4294 /* make sure, this cpu is enabled in the mask */
4295 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
4301 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
4306 mutex_unlock(&trace_types_lock);
4313 * tracing_update_buffers - used by tracing facility to expand ring buffers
4315 * To save on memory when the tracing is never used on a system with it
4316 * configured in. The ring buffers are set to a minimum size. But once
4317 * a user starts to use the tracing facility, then they need to grow
4318 * to their default size.
4320 * This function is to be called when a tracer is about to be used.
4322 int tracing_update_buffers(void)
4326 mutex_lock(&trace_types_lock);
4327 if (!ring_buffer_expanded)
4328 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
4329 RING_BUFFER_ALL_CPUS);
4330 mutex_unlock(&trace_types_lock);
4335 struct trace_option_dentry;
4338 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
4341 * Used to clear out the tracer before deletion of an instance.
4342 * Must have trace_types_lock held.
4344 static void tracing_set_nop(struct trace_array *tr)
4346 if (tr->current_trace == &nop_trace)
4349 tr->current_trace->enabled--;
4351 if (tr->current_trace->reset)
4352 tr->current_trace->reset(tr);
4354 tr->current_trace = &nop_trace;
4357 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
4359 /* Only enable if the directory has been created already. */
4363 create_trace_option_files(tr, t);
4366 static int tracing_set_tracer(struct trace_array *tr, const char *buf)
4369 #ifdef CONFIG_TRACER_MAX_TRACE
4374 mutex_lock(&trace_types_lock);
4376 if (!ring_buffer_expanded) {
4377 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
4378 RING_BUFFER_ALL_CPUS);
4384 for (t = trace_types; t; t = t->next) {
4385 if (strcmp(t->name, buf) == 0)
4392 if (t == tr->current_trace)
4395 /* Some tracers are only allowed for the top level buffer */
4396 if (!trace_ok_for_array(t, tr)) {
4401 /* If trace pipe files are being read, we can't change the tracer */
4402 if (tr->current_trace->ref) {
4407 trace_branch_disable();
4409 tr->current_trace->enabled--;
4411 if (tr->current_trace->reset)
4412 tr->current_trace->reset(tr);
4414 /* Current trace needs to be nop_trace before synchronize_sched */
4415 tr->current_trace = &nop_trace;
4417 #ifdef CONFIG_TRACER_MAX_TRACE
4418 had_max_tr = tr->allocated_snapshot;
4420 if (had_max_tr && !t->use_max_tr) {
4422 * We need to make sure that the update_max_tr sees that
4423 * current_trace changed to nop_trace to keep it from
4424 * swapping the buffers after we resize it.
4425 * The update_max_tr is called from interrupts disabled
4426 * so a synchronized_sched() is sufficient.
4428 synchronize_sched();
4433 #ifdef CONFIG_TRACER_MAX_TRACE
4434 if (t->use_max_tr && !had_max_tr) {
4435 ret = alloc_snapshot(tr);
4442 ret = tracer_init(t, tr);
4447 tr->current_trace = t;
4448 tr->current_trace->enabled++;
4449 trace_branch_enable(tr);
4451 mutex_unlock(&trace_types_lock);
4457 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
4458 size_t cnt, loff_t *ppos)
4460 struct trace_array *tr = filp->private_data;
4461 char buf[MAX_TRACER_SIZE+1];
4468 if (cnt > MAX_TRACER_SIZE)
4469 cnt = MAX_TRACER_SIZE;
4471 if (copy_from_user(&buf, ubuf, cnt))
4476 /* strip ending whitespace. */
4477 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
4480 err = tracing_set_tracer(tr, buf);
4490 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
4491 size_t cnt, loff_t *ppos)
4496 r = snprintf(buf, sizeof(buf), "%ld\n",
4497 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
4498 if (r > sizeof(buf))
4500 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4504 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
4505 size_t cnt, loff_t *ppos)
4510 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4520 tracing_thresh_read(struct file *filp, char __user *ubuf,
4521 size_t cnt, loff_t *ppos)
4523 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
4527 tracing_thresh_write(struct file *filp, const char __user *ubuf,
4528 size_t cnt, loff_t *ppos)
4530 struct trace_array *tr = filp->private_data;
4533 mutex_lock(&trace_types_lock);
4534 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
4538 if (tr->current_trace->update_thresh) {
4539 ret = tr->current_trace->update_thresh(tr);
4546 mutex_unlock(&trace_types_lock);
4552 tracing_max_lat_read(struct file *filp, char __user *ubuf,
4553 size_t cnt, loff_t *ppos)
4555 return tracing_nsecs_read(filp->private_data, ubuf, cnt, ppos);
4559 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
4560 size_t cnt, loff_t *ppos)
4562 return tracing_nsecs_write(filp->private_data, ubuf, cnt, ppos);
4565 static int tracing_open_pipe(struct inode *inode, struct file *filp)
4567 struct trace_array *tr = inode->i_private;
4568 struct trace_iterator *iter;
4571 if (tracing_disabled)
4574 if (trace_array_get(tr) < 0)
4577 mutex_lock(&trace_types_lock);
4579 /* create a buffer to store the information to pass to userspace */
4580 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4583 __trace_array_put(tr);
4587 trace_seq_init(&iter->seq);
4588 iter->trace = tr->current_trace;
4590 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
4595 /* trace pipe does not show start of buffer */
4596 cpumask_setall(iter->started);
4598 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
4599 iter->iter_flags |= TRACE_FILE_LAT_FMT;
4601 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4602 if (trace_clocks[tr->clock_id].in_ns)
4603 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4606 iter->trace_buffer = &tr->trace_buffer;
4607 iter->cpu_file = tracing_get_cpu(inode);
4608 mutex_init(&iter->mutex);
4609 filp->private_data = iter;
4611 if (iter->trace->pipe_open)
4612 iter->trace->pipe_open(iter);
4614 nonseekable_open(inode, filp);
4616 tr->current_trace->ref++;
4618 mutex_unlock(&trace_types_lock);
4624 __trace_array_put(tr);
4625 mutex_unlock(&trace_types_lock);
4629 static int tracing_release_pipe(struct inode *inode, struct file *file)
4631 struct trace_iterator *iter = file->private_data;
4632 struct trace_array *tr = inode->i_private;
4634 mutex_lock(&trace_types_lock);
4636 tr->current_trace->ref--;
4638 if (iter->trace->pipe_close)
4639 iter->trace->pipe_close(iter);
4641 mutex_unlock(&trace_types_lock);
4643 free_cpumask_var(iter->started);
4644 mutex_destroy(&iter->mutex);
4647 trace_array_put(tr);
4653 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
4655 struct trace_array *tr = iter->tr;
4657 /* Iterators are static, they should be filled or empty */
4658 if (trace_buffer_iter(iter, iter->cpu_file))
4659 return POLLIN | POLLRDNORM;
4661 if (tr->trace_flags & TRACE_ITER_BLOCK)
4663 * Always select as readable when in blocking mode
4665 return POLLIN | POLLRDNORM;
4667 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
4672 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
4674 struct trace_iterator *iter = filp->private_data;
4676 return trace_poll(iter, filp, poll_table);
4679 /* Must be called with iter->mutex held. */
4680 static int tracing_wait_pipe(struct file *filp)
4682 struct trace_iterator *iter = filp->private_data;
4685 while (trace_empty(iter)) {
4687 if ((filp->f_flags & O_NONBLOCK)) {
4692 * We block until we read something and tracing is disabled.
4693 * We still block if tracing is disabled, but we have never
4694 * read anything. This allows a user to cat this file, and
4695 * then enable tracing. But after we have read something,
4696 * we give an EOF when tracing is again disabled.
4698 * iter->pos will be 0 if we haven't read anything.
4700 if (!tracing_is_on() && iter->pos)
4703 mutex_unlock(&iter->mutex);
4705 ret = wait_on_pipe(iter, false);
4707 mutex_lock(&iter->mutex);
4720 tracing_read_pipe(struct file *filp, char __user *ubuf,
4721 size_t cnt, loff_t *ppos)
4723 struct trace_iterator *iter = filp->private_data;
4726 /* return any leftover data */
4727 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4731 trace_seq_init(&iter->seq);
4734 * Avoid more than one consumer on a single file descriptor
4735 * This is just a matter of traces coherency, the ring buffer itself
4738 mutex_lock(&iter->mutex);
4739 if (iter->trace->read) {
4740 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
4746 sret = tracing_wait_pipe(filp);
4750 /* stop when tracing is finished */
4751 if (trace_empty(iter)) {
4756 if (cnt >= PAGE_SIZE)
4757 cnt = PAGE_SIZE - 1;
4759 /* reset all but tr, trace, and overruns */
4760 memset(&iter->seq, 0,
4761 sizeof(struct trace_iterator) -
4762 offsetof(struct trace_iterator, seq));
4763 cpumask_clear(iter->started);
4766 trace_event_read_lock();
4767 trace_access_lock(iter->cpu_file);
4768 while (trace_find_next_entry_inc(iter) != NULL) {
4769 enum print_line_t ret;
4770 int save_len = iter->seq.seq.len;
4772 ret = print_trace_line(iter);
4773 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4774 /* don't print partial lines */
4775 iter->seq.seq.len = save_len;
4778 if (ret != TRACE_TYPE_NO_CONSUME)
4779 trace_consume(iter);
4781 if (trace_seq_used(&iter->seq) >= cnt)
4785 * Setting the full flag means we reached the trace_seq buffer
4786 * size and we should leave by partial output condition above.
4787 * One of the trace_seq_* functions is not used properly.
4789 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4792 trace_access_unlock(iter->cpu_file);
4793 trace_event_read_unlock();
4795 /* Now copy what we have to the user */
4796 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4797 if (iter->seq.seq.readpos >= trace_seq_used(&iter->seq))
4798 trace_seq_init(&iter->seq);
4801 * If there was nothing to send to user, in spite of consuming trace
4802 * entries, go back to wait for more entries.
4808 mutex_unlock(&iter->mutex);
4813 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4816 __free_page(spd->pages[idx]);
4819 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4821 .confirm = generic_pipe_buf_confirm,
4822 .release = generic_pipe_buf_release,
4823 .steal = generic_pipe_buf_steal,
4824 .get = generic_pipe_buf_get,
4828 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4834 /* Seq buffer is page-sized, exactly what we need. */
4836 save_len = iter->seq.seq.len;
4837 ret = print_trace_line(iter);
4839 if (trace_seq_has_overflowed(&iter->seq)) {
4840 iter->seq.seq.len = save_len;
4845 * This should not be hit, because it should only
4846 * be set if the iter->seq overflowed. But check it
4847 * anyway to be safe.
4849 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4850 iter->seq.seq.len = save_len;
4854 count = trace_seq_used(&iter->seq) - save_len;
4857 iter->seq.seq.len = save_len;
4861 if (ret != TRACE_TYPE_NO_CONSUME)
4862 trace_consume(iter);
4864 if (!trace_find_next_entry_inc(iter)) {
4874 static ssize_t tracing_splice_read_pipe(struct file *filp,
4876 struct pipe_inode_info *pipe,
4880 struct page *pages_def[PIPE_DEF_BUFFERS];
4881 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4882 struct trace_iterator *iter = filp->private_data;
4883 struct splice_pipe_desc spd = {
4885 .partial = partial_def,
4886 .nr_pages = 0, /* This gets updated below. */
4887 .nr_pages_max = PIPE_DEF_BUFFERS,
4889 .ops = &tracing_pipe_buf_ops,
4890 .spd_release = tracing_spd_release_pipe,
4896 if (splice_grow_spd(pipe, &spd))
4899 mutex_lock(&iter->mutex);
4901 if (iter->trace->splice_read) {
4902 ret = iter->trace->splice_read(iter, filp,
4903 ppos, pipe, len, flags);
4908 ret = tracing_wait_pipe(filp);
4912 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4917 trace_event_read_lock();
4918 trace_access_lock(iter->cpu_file);
4920 /* Fill as many pages as possible. */
4921 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
4922 spd.pages[i] = alloc_page(GFP_KERNEL);
4926 rem = tracing_fill_pipe_page(rem, iter);
4928 /* Copy the data into the page, so we can start over. */
4929 ret = trace_seq_to_buffer(&iter->seq,
4930 page_address(spd.pages[i]),
4931 trace_seq_used(&iter->seq));
4933 __free_page(spd.pages[i]);
4936 spd.partial[i].offset = 0;
4937 spd.partial[i].len = trace_seq_used(&iter->seq);
4939 trace_seq_init(&iter->seq);
4942 trace_access_unlock(iter->cpu_file);
4943 trace_event_read_unlock();
4944 mutex_unlock(&iter->mutex);
4948 ret = splice_to_pipe(pipe, &spd);
4950 splice_shrink_spd(&spd);
4954 mutex_unlock(&iter->mutex);
4959 tracing_entries_read(struct file *filp, char __user *ubuf,
4960 size_t cnt, loff_t *ppos)
4962 struct inode *inode = file_inode(filp);
4963 struct trace_array *tr = inode->i_private;
4964 int cpu = tracing_get_cpu(inode);
4969 mutex_lock(&trace_types_lock);
4971 if (cpu == RING_BUFFER_ALL_CPUS) {
4972 int cpu, buf_size_same;
4977 /* check if all cpu sizes are same */
4978 for_each_tracing_cpu(cpu) {
4979 /* fill in the size from first enabled cpu */
4981 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4982 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4988 if (buf_size_same) {
4989 if (!ring_buffer_expanded)
4990 r = sprintf(buf, "%lu (expanded: %lu)\n",
4992 trace_buf_size >> 10);
4994 r = sprintf(buf, "%lu\n", size >> 10);
4996 r = sprintf(buf, "X\n");
4998 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10);
5000 mutex_unlock(&trace_types_lock);
5002 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5007 tracing_entries_write(struct file *filp, const char __user *ubuf,
5008 size_t cnt, loff_t *ppos)
5010 struct inode *inode = file_inode(filp);
5011 struct trace_array *tr = inode->i_private;
5015 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5019 /* must have at least 1 entry */
5023 /* value is in KB */
5025 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
5035 tracing_total_entries_read(struct file *filp, char __user *ubuf,
5036 size_t cnt, loff_t *ppos)
5038 struct trace_array *tr = filp->private_data;
5041 unsigned long size = 0, expanded_size = 0;
5043 mutex_lock(&trace_types_lock);
5044 for_each_tracing_cpu(cpu) {
5045 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
5046 if (!ring_buffer_expanded)
5047 expanded_size += trace_buf_size >> 10;
5049 if (ring_buffer_expanded)
5050 r = sprintf(buf, "%lu\n", size);
5052 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
5053 mutex_unlock(&trace_types_lock);
5055 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5059 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
5060 size_t cnt, loff_t *ppos)
5063 * There is no need to read what the user has written, this function
5064 * is just to make sure that there is no error when "echo" is used
5073 tracing_free_buffer_release(struct inode *inode, struct file *filp)
5075 struct trace_array *tr = inode->i_private;
5077 /* disable tracing ? */
5078 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
5079 tracer_tracing_off(tr);
5080 /* resize the ring buffer to 0 */
5081 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
5083 trace_array_put(tr);
5089 tracing_mark_write(struct file *filp, const char __user *ubuf,
5090 size_t cnt, loff_t *fpos)
5092 unsigned long addr = (unsigned long)ubuf;
5093 struct trace_array *tr = filp->private_data;
5094 struct ring_buffer_event *event;
5095 struct ring_buffer *buffer;
5096 struct print_entry *entry;
5097 unsigned long irq_flags;
5098 struct page *pages[2];
5108 if (tracing_disabled)
5111 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
5114 if (cnt > TRACE_BUF_SIZE)
5115 cnt = TRACE_BUF_SIZE;
5118 * Userspace is injecting traces into the kernel trace buffer.
5119 * We want to be as non intrusive as possible.
5120 * To do so, we do not want to allocate any special buffers
5121 * or take any locks, but instead write the userspace data
5122 * straight into the ring buffer.
5124 * First we need to pin the userspace buffer into memory,
5125 * which, most likely it is, because it just referenced it.
5126 * But there's no guarantee that it is. By using get_user_pages_fast()
5127 * and kmap_atomic/kunmap_atomic() we can get access to the
5128 * pages directly. We then write the data directly into the
5131 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
5133 /* check if we cross pages */
5134 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
5137 offset = addr & (PAGE_SIZE - 1);
5140 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
5141 if (ret < nr_pages) {
5143 put_page(pages[ret]);
5148 for (i = 0; i < nr_pages; i++)
5149 map_page[i] = kmap_atomic(pages[i]);
5151 local_save_flags(irq_flags);
5152 size = sizeof(*entry) + cnt + 2; /* possible \n added */
5153 buffer = tr->trace_buffer.buffer;
5154 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
5155 irq_flags, preempt_count());
5157 /* Ring buffer disabled, return as if not open for write */
5162 entry = ring_buffer_event_data(event);
5163 entry->ip = _THIS_IP_;
5165 if (nr_pages == 2) {
5166 len = PAGE_SIZE - offset;
5167 memcpy(&entry->buf, map_page[0] + offset, len);
5168 memcpy(&entry->buf[len], map_page[1], cnt - len);
5170 memcpy(&entry->buf, map_page[0] + offset, cnt);
5172 if (entry->buf[cnt - 1] != '\n') {
5173 entry->buf[cnt] = '\n';
5174 entry->buf[cnt + 1] = '\0';
5176 entry->buf[cnt] = '\0';
5178 __buffer_unlock_commit(buffer, event);
5185 for (i = nr_pages - 1; i >= 0; i--) {
5186 kunmap_atomic(map_page[i]);
5193 static int tracing_clock_show(struct seq_file *m, void *v)
5195 struct trace_array *tr = m->private;
5198 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
5200 "%s%s%s%s", i ? " " : "",
5201 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
5202 i == tr->clock_id ? "]" : "");
5208 static int tracing_set_clock(struct trace_array *tr, const char *clockstr)
5212 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
5213 if (strcmp(trace_clocks[i].name, clockstr) == 0)
5216 if (i == ARRAY_SIZE(trace_clocks))
5219 mutex_lock(&trace_types_lock);
5223 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
5226 * New clock may not be consistent with the previous clock.
5227 * Reset the buffer so that it doesn't have incomparable timestamps.
5229 tracing_reset_online_cpus(&tr->trace_buffer);
5231 #ifdef CONFIG_TRACER_MAX_TRACE
5232 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
5233 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
5234 tracing_reset_online_cpus(&tr->max_buffer);
5237 mutex_unlock(&trace_types_lock);
5242 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
5243 size_t cnt, loff_t *fpos)
5245 struct seq_file *m = filp->private_data;
5246 struct trace_array *tr = m->private;
5248 const char *clockstr;
5251 if (cnt >= sizeof(buf))
5254 if (copy_from_user(&buf, ubuf, cnt))
5259 clockstr = strstrip(buf);
5261 ret = tracing_set_clock(tr, clockstr);
5270 static int tracing_clock_open(struct inode *inode, struct file *file)
5272 struct trace_array *tr = inode->i_private;
5275 if (tracing_disabled)
5278 if (trace_array_get(tr))
5281 ret = single_open(file, tracing_clock_show, inode->i_private);
5283 trace_array_put(tr);
5288 struct ftrace_buffer_info {
5289 struct trace_iterator iter;
5294 #ifdef CONFIG_TRACER_SNAPSHOT
5295 static int tracing_snapshot_open(struct inode *inode, struct file *file)
5297 struct trace_array *tr = inode->i_private;
5298 struct trace_iterator *iter;
5302 if (trace_array_get(tr) < 0)
5305 if (file->f_mode & FMODE_READ) {
5306 iter = __tracing_open(inode, file, true);
5308 ret = PTR_ERR(iter);
5310 /* Writes still need the seq_file to hold the private data */
5312 m = kzalloc(sizeof(*m), GFP_KERNEL);
5315 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
5323 iter->trace_buffer = &tr->max_buffer;
5324 iter->cpu_file = tracing_get_cpu(inode);
5326 file->private_data = m;
5330 trace_array_put(tr);
5336 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
5339 struct seq_file *m = filp->private_data;
5340 struct trace_iterator *iter = m->private;
5341 struct trace_array *tr = iter->tr;
5345 ret = tracing_update_buffers();
5349 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5353 mutex_lock(&trace_types_lock);
5355 if (tr->current_trace->use_max_tr) {
5362 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
5366 if (tr->allocated_snapshot)
5370 /* Only allow per-cpu swap if the ring buffer supports it */
5371 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
5372 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
5377 if (!tr->allocated_snapshot) {
5378 ret = alloc_snapshot(tr);
5382 local_irq_disable();
5383 /* Now, we're going to swap */
5384 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
5385 update_max_tr(tr, current, smp_processor_id());
5387 update_max_tr_single(tr, current, iter->cpu_file);
5391 if (tr->allocated_snapshot) {
5392 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
5393 tracing_reset_online_cpus(&tr->max_buffer);
5395 tracing_reset(&tr->max_buffer, iter->cpu_file);
5405 mutex_unlock(&trace_types_lock);
5409 static int tracing_snapshot_release(struct inode *inode, struct file *file)
5411 struct seq_file *m = file->private_data;
5414 ret = tracing_release(inode, file);
5416 if (file->f_mode & FMODE_READ)
5419 /* If write only, the seq_file is just a stub */
5427 static int tracing_buffers_open(struct inode *inode, struct file *filp);
5428 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
5429 size_t count, loff_t *ppos);
5430 static int tracing_buffers_release(struct inode *inode, struct file *file);
5431 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5432 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
5434 static int snapshot_raw_open(struct inode *inode, struct file *filp)
5436 struct ftrace_buffer_info *info;
5439 ret = tracing_buffers_open(inode, filp);
5443 info = filp->private_data;
5445 if (info->iter.trace->use_max_tr) {
5446 tracing_buffers_release(inode, filp);
5450 info->iter.snapshot = true;
5451 info->iter.trace_buffer = &info->iter.tr->max_buffer;
5456 #endif /* CONFIG_TRACER_SNAPSHOT */
5459 static const struct file_operations tracing_thresh_fops = {
5460 .open = tracing_open_generic,
5461 .read = tracing_thresh_read,
5462 .write = tracing_thresh_write,
5463 .llseek = generic_file_llseek,
5466 static const struct file_operations tracing_max_lat_fops = {
5467 .open = tracing_open_generic,
5468 .read = tracing_max_lat_read,
5469 .write = tracing_max_lat_write,
5470 .llseek = generic_file_llseek,
5473 static const struct file_operations set_tracer_fops = {
5474 .open = tracing_open_generic,
5475 .read = tracing_set_trace_read,
5476 .write = tracing_set_trace_write,
5477 .llseek = generic_file_llseek,
5480 static const struct file_operations tracing_pipe_fops = {
5481 .open = tracing_open_pipe,
5482 .poll = tracing_poll_pipe,
5483 .read = tracing_read_pipe,
5484 .splice_read = tracing_splice_read_pipe,
5485 .release = tracing_release_pipe,
5486 .llseek = no_llseek,
5489 static const struct file_operations tracing_entries_fops = {
5490 .open = tracing_open_generic_tr,
5491 .read = tracing_entries_read,
5492 .write = tracing_entries_write,
5493 .llseek = generic_file_llseek,
5494 .release = tracing_release_generic_tr,
5497 static const struct file_operations tracing_total_entries_fops = {
5498 .open = tracing_open_generic_tr,
5499 .read = tracing_total_entries_read,
5500 .llseek = generic_file_llseek,
5501 .release = tracing_release_generic_tr,
5504 static const struct file_operations tracing_free_buffer_fops = {
5505 .open = tracing_open_generic_tr,
5506 .write = tracing_free_buffer_write,
5507 .release = tracing_free_buffer_release,
5510 static const struct file_operations tracing_mark_fops = {
5511 .open = tracing_open_generic_tr,
5512 .write = tracing_mark_write,
5513 .llseek = generic_file_llseek,
5514 .release = tracing_release_generic_tr,
5517 static const struct file_operations trace_clock_fops = {
5518 .open = tracing_clock_open,
5520 .llseek = seq_lseek,
5521 .release = tracing_single_release_tr,
5522 .write = tracing_clock_write,
5525 #ifdef CONFIG_TRACER_SNAPSHOT
5526 static const struct file_operations snapshot_fops = {
5527 .open = tracing_snapshot_open,
5529 .write = tracing_snapshot_write,
5530 .llseek = tracing_lseek,
5531 .release = tracing_snapshot_release,
5534 static const struct file_operations snapshot_raw_fops = {
5535 .open = snapshot_raw_open,
5536 .read = tracing_buffers_read,
5537 .release = tracing_buffers_release,
5538 .splice_read = tracing_buffers_splice_read,
5539 .llseek = no_llseek,
5542 #endif /* CONFIG_TRACER_SNAPSHOT */
5544 static int tracing_buffers_open(struct inode *inode, struct file *filp)
5546 struct trace_array *tr = inode->i_private;
5547 struct ftrace_buffer_info *info;
5550 if (tracing_disabled)
5553 if (trace_array_get(tr) < 0)
5556 info = kzalloc(sizeof(*info), GFP_KERNEL);
5558 trace_array_put(tr);
5562 mutex_lock(&trace_types_lock);
5565 info->iter.cpu_file = tracing_get_cpu(inode);
5566 info->iter.trace = tr->current_trace;
5567 info->iter.trace_buffer = &tr->trace_buffer;
5569 /* Force reading ring buffer for first read */
5570 info->read = (unsigned int)-1;
5572 filp->private_data = info;
5574 tr->current_trace->ref++;
5576 mutex_unlock(&trace_types_lock);
5578 ret = nonseekable_open(inode, filp);
5580 trace_array_put(tr);
5586 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
5588 struct ftrace_buffer_info *info = filp->private_data;
5589 struct trace_iterator *iter = &info->iter;
5591 return trace_poll(iter, filp, poll_table);
5595 tracing_buffers_read(struct file *filp, char __user *ubuf,
5596 size_t count, loff_t *ppos)
5598 struct ftrace_buffer_info *info = filp->private_data;
5599 struct trace_iterator *iter = &info->iter;
5606 #ifdef CONFIG_TRACER_MAX_TRACE
5607 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
5612 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
5617 /* Do we have previous read data to read? */
5618 if (info->read < PAGE_SIZE)
5622 trace_access_lock(iter->cpu_file);
5623 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
5627 trace_access_unlock(iter->cpu_file);
5630 if (trace_empty(iter)) {
5631 if ((filp->f_flags & O_NONBLOCK))
5634 ret = wait_on_pipe(iter, false);
5645 size = PAGE_SIZE - info->read;
5649 ret = copy_to_user(ubuf, info->spare + info->read, size);
5661 static int tracing_buffers_release(struct inode *inode, struct file *file)
5663 struct ftrace_buffer_info *info = file->private_data;
5664 struct trace_iterator *iter = &info->iter;
5666 mutex_lock(&trace_types_lock);
5668 iter->tr->current_trace->ref--;
5670 __trace_array_put(iter->tr);
5673 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
5676 mutex_unlock(&trace_types_lock);
5682 struct ring_buffer *buffer;
5687 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
5688 struct pipe_buffer *buf)
5690 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5695 ring_buffer_free_read_page(ref->buffer, ref->page);
5700 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
5701 struct pipe_buffer *buf)
5703 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5708 /* Pipe buffer operations for a buffer. */
5709 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
5711 .confirm = generic_pipe_buf_confirm,
5712 .release = buffer_pipe_buf_release,
5713 .steal = generic_pipe_buf_steal,
5714 .get = buffer_pipe_buf_get,
5718 * Callback from splice_to_pipe(), if we need to release some pages
5719 * at the end of the spd in case we error'ed out in filling the pipe.
5721 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
5723 struct buffer_ref *ref =
5724 (struct buffer_ref *)spd->partial[i].private;
5729 ring_buffer_free_read_page(ref->buffer, ref->page);
5731 spd->partial[i].private = 0;
5735 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5736 struct pipe_inode_info *pipe, size_t len,
5739 struct ftrace_buffer_info *info = file->private_data;
5740 struct trace_iterator *iter = &info->iter;
5741 struct partial_page partial_def[PIPE_DEF_BUFFERS];
5742 struct page *pages_def[PIPE_DEF_BUFFERS];
5743 struct splice_pipe_desc spd = {
5745 .partial = partial_def,
5746 .nr_pages_max = PIPE_DEF_BUFFERS,
5748 .ops = &buffer_pipe_buf_ops,
5749 .spd_release = buffer_spd_release,
5751 struct buffer_ref *ref;
5752 int entries, size, i;
5755 #ifdef CONFIG_TRACER_MAX_TRACE
5756 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
5760 if (splice_grow_spd(pipe, &spd))
5763 if (*ppos & (PAGE_SIZE - 1))
5766 if (len & (PAGE_SIZE - 1)) {
5767 if (len < PAGE_SIZE)
5773 trace_access_lock(iter->cpu_file);
5774 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5776 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= PAGE_SIZE) {
5780 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
5787 ref->buffer = iter->trace_buffer->buffer;
5788 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5795 r = ring_buffer_read_page(ref->buffer, &ref->page,
5796 len, iter->cpu_file, 1);
5798 ring_buffer_free_read_page(ref->buffer, ref->page);
5804 * zero out any left over data, this is going to
5807 size = ring_buffer_page_len(ref->page);
5808 if (size < PAGE_SIZE)
5809 memset(ref->page + size, 0, PAGE_SIZE - size);
5811 page = virt_to_page(ref->page);
5813 spd.pages[i] = page;
5814 spd.partial[i].len = PAGE_SIZE;
5815 spd.partial[i].offset = 0;
5816 spd.partial[i].private = (unsigned long)ref;
5820 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5823 trace_access_unlock(iter->cpu_file);
5826 /* did we read anything? */
5827 if (!spd.nr_pages) {
5831 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
5834 ret = wait_on_pipe(iter, true);
5841 ret = splice_to_pipe(pipe, &spd);
5842 splice_shrink_spd(&spd);
5847 static const struct file_operations tracing_buffers_fops = {
5848 .open = tracing_buffers_open,
5849 .read = tracing_buffers_read,
5850 .poll = tracing_buffers_poll,
5851 .release = tracing_buffers_release,
5852 .splice_read = tracing_buffers_splice_read,
5853 .llseek = no_llseek,
5857 tracing_stats_read(struct file *filp, char __user *ubuf,
5858 size_t count, loff_t *ppos)
5860 struct inode *inode = file_inode(filp);
5861 struct trace_array *tr = inode->i_private;
5862 struct trace_buffer *trace_buf = &tr->trace_buffer;
5863 int cpu = tracing_get_cpu(inode);
5864 struct trace_seq *s;
5866 unsigned long long t;
5867 unsigned long usec_rem;
5869 s = kmalloc(sizeof(*s), GFP_KERNEL);
5875 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5876 trace_seq_printf(s, "entries: %ld\n", cnt);
5878 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5879 trace_seq_printf(s, "overrun: %ld\n", cnt);
5881 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5882 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5884 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5885 trace_seq_printf(s, "bytes: %ld\n", cnt);
5887 if (trace_clocks[tr->clock_id].in_ns) {
5888 /* local or global for trace_clock */
5889 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5890 usec_rem = do_div(t, USEC_PER_SEC);
5891 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5894 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5895 usec_rem = do_div(t, USEC_PER_SEC);
5896 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5898 /* counter or tsc mode for trace_clock */
5899 trace_seq_printf(s, "oldest event ts: %llu\n",
5900 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5902 trace_seq_printf(s, "now ts: %llu\n",
5903 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5906 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5907 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5909 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5910 trace_seq_printf(s, "read events: %ld\n", cnt);
5912 count = simple_read_from_buffer(ubuf, count, ppos,
5913 s->buffer, trace_seq_used(s));
5920 static const struct file_operations tracing_stats_fops = {
5921 .open = tracing_open_generic_tr,
5922 .read = tracing_stats_read,
5923 .llseek = generic_file_llseek,
5924 .release = tracing_release_generic_tr,
5927 #ifdef CONFIG_DYNAMIC_FTRACE
5929 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5935 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5936 size_t cnt, loff_t *ppos)
5938 static char ftrace_dyn_info_buffer[1024];
5939 static DEFINE_MUTEX(dyn_info_mutex);
5940 unsigned long *p = filp->private_data;
5941 char *buf = ftrace_dyn_info_buffer;
5942 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5945 mutex_lock(&dyn_info_mutex);
5946 r = sprintf(buf, "%ld ", *p);
5948 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5951 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5953 mutex_unlock(&dyn_info_mutex);
5958 static const struct file_operations tracing_dyn_info_fops = {
5959 .open = tracing_open_generic,
5960 .read = tracing_read_dyn_info,
5961 .llseek = generic_file_llseek,
5963 #endif /* CONFIG_DYNAMIC_FTRACE */
5965 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5967 ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5973 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5975 unsigned long *count = (long *)data;
5987 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5988 struct ftrace_probe_ops *ops, void *data)
5990 long count = (long)data;
5992 seq_printf(m, "%ps:", (void *)ip);
5994 seq_puts(m, "snapshot");
5997 seq_puts(m, ":unlimited\n");
5999 seq_printf(m, ":count=%ld\n", count);
6004 static struct ftrace_probe_ops snapshot_probe_ops = {
6005 .func = ftrace_snapshot,
6006 .print = ftrace_snapshot_print,
6009 static struct ftrace_probe_ops snapshot_count_probe_ops = {
6010 .func = ftrace_count_snapshot,
6011 .print = ftrace_snapshot_print,
6015 ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
6016 char *glob, char *cmd, char *param, int enable)
6018 struct ftrace_probe_ops *ops;
6019 void *count = (void *)-1;
6023 /* hash funcs only work with set_ftrace_filter */
6027 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
6029 if (glob[0] == '!') {
6030 unregister_ftrace_function_probe_func(glob+1, ops);
6037 number = strsep(¶m, ":");
6039 if (!strlen(number))
6043 * We use the callback data field (which is a pointer)
6046 ret = kstrtoul(number, 0, (unsigned long *)&count);
6051 ret = register_ftrace_function_probe(glob, ops, count);
6054 alloc_snapshot(&global_trace);
6056 return ret < 0 ? ret : 0;
6059 static struct ftrace_func_command ftrace_snapshot_cmd = {
6061 .func = ftrace_trace_snapshot_callback,
6064 static __init int register_snapshot_cmd(void)
6066 return register_ftrace_command(&ftrace_snapshot_cmd);
6069 static inline __init int register_snapshot_cmd(void) { return 0; }
6070 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
6072 static struct dentry *tracing_get_dentry(struct trace_array *tr)
6074 if (WARN_ON(!tr->dir))
6075 return ERR_PTR(-ENODEV);
6077 /* Top directory uses NULL as the parent */
6078 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
6081 /* All sub buffers have a descriptor */
6085 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
6087 struct dentry *d_tracer;
6090 return tr->percpu_dir;
6092 d_tracer = tracing_get_dentry(tr);
6093 if (IS_ERR(d_tracer))
6096 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
6098 WARN_ONCE(!tr->percpu_dir,
6099 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
6101 return tr->percpu_dir;
6104 static struct dentry *
6105 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
6106 void *data, long cpu, const struct file_operations *fops)
6108 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
6110 if (ret) /* See tracing_get_cpu() */
6111 d_inode(ret)->i_cdev = (void *)(cpu + 1);
6116 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
6118 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
6119 struct dentry *d_cpu;
6120 char cpu_dir[30]; /* 30 characters should be more than enough */
6125 snprintf(cpu_dir, 30, "cpu%ld", cpu);
6126 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
6128 pr_warning("Could not create tracefs '%s' entry\n", cpu_dir);
6132 /* per cpu trace_pipe */
6133 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
6134 tr, cpu, &tracing_pipe_fops);
6137 trace_create_cpu_file("trace", 0644, d_cpu,
6138 tr, cpu, &tracing_fops);
6140 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
6141 tr, cpu, &tracing_buffers_fops);
6143 trace_create_cpu_file("stats", 0444, d_cpu,
6144 tr, cpu, &tracing_stats_fops);
6146 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
6147 tr, cpu, &tracing_entries_fops);
6149 #ifdef CONFIG_TRACER_SNAPSHOT
6150 trace_create_cpu_file("snapshot", 0644, d_cpu,
6151 tr, cpu, &snapshot_fops);
6153 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
6154 tr, cpu, &snapshot_raw_fops);
6158 #ifdef CONFIG_FTRACE_SELFTEST
6159 /* Let selftest have access to static functions in this file */
6160 #include "trace_selftest.c"
6164 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
6167 struct trace_option_dentry *topt = filp->private_data;
6170 if (topt->flags->val & topt->opt->bit)
6175 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
6179 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
6182 struct trace_option_dentry *topt = filp->private_data;
6186 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6190 if (val != 0 && val != 1)
6193 if (!!(topt->flags->val & topt->opt->bit) != val) {
6194 mutex_lock(&trace_types_lock);
6195 ret = __set_tracer_option(topt->tr, topt->flags,
6197 mutex_unlock(&trace_types_lock);
6208 static const struct file_operations trace_options_fops = {
6209 .open = tracing_open_generic,
6210 .read = trace_options_read,
6211 .write = trace_options_write,
6212 .llseek = generic_file_llseek,
6216 * In order to pass in both the trace_array descriptor as well as the index
6217 * to the flag that the trace option file represents, the trace_array
6218 * has a character array of trace_flags_index[], which holds the index
6219 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
6220 * The address of this character array is passed to the flag option file
6221 * read/write callbacks.
6223 * In order to extract both the index and the trace_array descriptor,
6224 * get_tr_index() uses the following algorithm.
6228 * As the pointer itself contains the address of the index (remember
6231 * Then to get the trace_array descriptor, by subtracting that index
6232 * from the ptr, we get to the start of the index itself.
6234 * ptr - idx == &index[0]
6236 * Then a simple container_of() from that pointer gets us to the
6237 * trace_array descriptor.
6239 static void get_tr_index(void *data, struct trace_array **ptr,
6240 unsigned int *pindex)
6242 *pindex = *(unsigned char *)data;
6244 *ptr = container_of(data - *pindex, struct trace_array,
6249 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
6252 void *tr_index = filp->private_data;
6253 struct trace_array *tr;
6257 get_tr_index(tr_index, &tr, &index);
6259 if (tr->trace_flags & (1 << index))
6264 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
6268 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
6271 void *tr_index = filp->private_data;
6272 struct trace_array *tr;
6277 get_tr_index(tr_index, &tr, &index);
6279 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6283 if (val != 0 && val != 1)
6286 mutex_lock(&trace_types_lock);
6287 ret = set_tracer_flag(tr, 1 << index, val);
6288 mutex_unlock(&trace_types_lock);
6298 static const struct file_operations trace_options_core_fops = {
6299 .open = tracing_open_generic,
6300 .read = trace_options_core_read,
6301 .write = trace_options_core_write,
6302 .llseek = generic_file_llseek,
6305 struct dentry *trace_create_file(const char *name,
6307 struct dentry *parent,
6309 const struct file_operations *fops)
6313 ret = tracefs_create_file(name, mode, parent, data, fops);
6315 pr_warning("Could not create tracefs '%s' entry\n", name);
6321 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
6323 struct dentry *d_tracer;
6328 d_tracer = tracing_get_dentry(tr);
6329 if (IS_ERR(d_tracer))
6332 tr->options = tracefs_create_dir("options", d_tracer);
6334 pr_warning("Could not create tracefs directory 'options'\n");
6342 create_trace_option_file(struct trace_array *tr,
6343 struct trace_option_dentry *topt,
6344 struct tracer_flags *flags,
6345 struct tracer_opt *opt)
6347 struct dentry *t_options;
6349 t_options = trace_options_init_dentry(tr);
6353 topt->flags = flags;
6357 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
6358 &trace_options_fops);
6363 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
6365 struct trace_option_dentry *topts;
6366 struct trace_options *tr_topts;
6367 struct tracer_flags *flags;
6368 struct tracer_opt *opts;
6375 flags = tracer->flags;
6377 if (!flags || !flags->opts)
6381 * If this is an instance, only create flags for tracers
6382 * the instance may have.
6384 if (!trace_ok_for_array(tracer, tr))
6387 for (i = 0; i < tr->nr_topts; i++) {
6389 * Check if these flags have already been added.
6390 * Some tracers share flags.
6392 if (tr->topts[i].tracer->flags == tracer->flags)
6398 for (cnt = 0; opts[cnt].name; cnt++)
6401 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
6405 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
6412 tr->topts = tr_topts;
6413 tr->topts[tr->nr_topts].tracer = tracer;
6414 tr->topts[tr->nr_topts].topts = topts;
6417 for (cnt = 0; opts[cnt].name; cnt++) {
6418 create_trace_option_file(tr, &topts[cnt], flags,
6420 WARN_ONCE(topts[cnt].entry == NULL,
6421 "Failed to create trace option: %s",
6426 static struct dentry *
6427 create_trace_option_core_file(struct trace_array *tr,
6428 const char *option, long index)
6430 struct dentry *t_options;
6432 t_options = trace_options_init_dentry(tr);
6436 return trace_create_file(option, 0644, t_options,
6437 (void *)&tr->trace_flags_index[index],
6438 &trace_options_core_fops);
6441 static void create_trace_options_dir(struct trace_array *tr)
6443 struct dentry *t_options;
6444 bool top_level = tr == &global_trace;
6447 t_options = trace_options_init_dentry(tr);
6451 for (i = 0; trace_options[i]; i++) {
6453 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
6454 create_trace_option_core_file(tr, trace_options[i], i);
6459 rb_simple_read(struct file *filp, char __user *ubuf,
6460 size_t cnt, loff_t *ppos)
6462 struct trace_array *tr = filp->private_data;
6466 r = tracer_tracing_is_on(tr);
6467 r = sprintf(buf, "%d\n", r);
6469 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6473 rb_simple_write(struct file *filp, const char __user *ubuf,
6474 size_t cnt, loff_t *ppos)
6476 struct trace_array *tr = filp->private_data;
6477 struct ring_buffer *buffer = tr->trace_buffer.buffer;
6481 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6486 mutex_lock(&trace_types_lock);
6488 tracer_tracing_on(tr);
6489 if (tr->current_trace->start)
6490 tr->current_trace->start(tr);
6492 tracer_tracing_off(tr);
6493 if (tr->current_trace->stop)
6494 tr->current_trace->stop(tr);
6496 mutex_unlock(&trace_types_lock);
6504 static const struct file_operations rb_simple_fops = {
6505 .open = tracing_open_generic_tr,
6506 .read = rb_simple_read,
6507 .write = rb_simple_write,
6508 .release = tracing_release_generic_tr,
6509 .llseek = default_llseek,
6512 struct dentry *trace_instance_dir;
6515 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
6518 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
6520 enum ring_buffer_flags rb_flags;
6522 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
6526 buf->buffer = ring_buffer_alloc(size, rb_flags);
6530 buf->data = alloc_percpu(struct trace_array_cpu);
6532 ring_buffer_free(buf->buffer);
6536 /* Allocate the first page for all buffers */
6537 set_buffer_entries(&tr->trace_buffer,
6538 ring_buffer_size(tr->trace_buffer.buffer, 0));
6543 static int allocate_trace_buffers(struct trace_array *tr, int size)
6547 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
6551 #ifdef CONFIG_TRACER_MAX_TRACE
6552 ret = allocate_trace_buffer(tr, &tr->max_buffer,
6553 allocate_snapshot ? size : 1);
6555 ring_buffer_free(tr->trace_buffer.buffer);
6556 free_percpu(tr->trace_buffer.data);
6559 tr->allocated_snapshot = allocate_snapshot;
6562 * Only the top level trace array gets its snapshot allocated
6563 * from the kernel command line.
6565 allocate_snapshot = false;
6570 static void free_trace_buffer(struct trace_buffer *buf)
6573 ring_buffer_free(buf->buffer);
6575 free_percpu(buf->data);
6580 static void free_trace_buffers(struct trace_array *tr)
6585 free_trace_buffer(&tr->trace_buffer);
6587 #ifdef CONFIG_TRACER_MAX_TRACE
6588 free_trace_buffer(&tr->max_buffer);
6592 static void init_trace_flags_index(struct trace_array *tr)
6596 /* Used by the trace options files */
6597 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
6598 tr->trace_flags_index[i] = i;
6601 static void __update_tracer_options(struct trace_array *tr)
6605 for (t = trace_types; t; t = t->next)
6606 add_tracer_options(tr, t);
6609 static void update_tracer_options(struct trace_array *tr)
6611 mutex_lock(&trace_types_lock);
6612 __update_tracer_options(tr);
6613 mutex_unlock(&trace_types_lock);
6616 static int instance_mkdir(const char *name)
6618 struct trace_array *tr;
6621 mutex_lock(&trace_types_lock);
6624 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6625 if (tr->name && strcmp(tr->name, name) == 0)
6630 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
6634 tr->name = kstrdup(name, GFP_KERNEL);
6638 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
6641 tr->trace_flags = global_trace.trace_flags;
6643 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
6645 raw_spin_lock_init(&tr->start_lock);
6647 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
6649 tr->current_trace = &nop_trace;
6651 INIT_LIST_HEAD(&tr->systems);
6652 INIT_LIST_HEAD(&tr->events);
6654 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
6657 tr->dir = tracefs_create_dir(name, trace_instance_dir);
6661 ret = event_trace_add_tracer(tr->dir, tr);
6663 tracefs_remove_recursive(tr->dir);
6667 init_tracer_tracefs(tr, tr->dir);
6668 init_trace_flags_index(tr);
6669 __update_tracer_options(tr);
6671 list_add(&tr->list, &ftrace_trace_arrays);
6673 mutex_unlock(&trace_types_lock);
6678 free_trace_buffers(tr);
6679 free_cpumask_var(tr->tracing_cpumask);
6684 mutex_unlock(&trace_types_lock);
6690 static int instance_rmdir(const char *name)
6692 struct trace_array *tr;
6697 mutex_lock(&trace_types_lock);
6700 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6701 if (tr->name && strcmp(tr->name, name) == 0) {
6710 if (tr->ref || (tr->current_trace && tr->current_trace->ref))
6713 list_del(&tr->list);
6715 tracing_set_nop(tr);
6716 event_trace_del_tracer(tr);
6717 ftrace_destroy_function_files(tr);
6718 tracefs_remove_recursive(tr->dir);
6719 free_trace_buffers(tr);
6721 for (i = 0; i < tr->nr_topts; i++) {
6722 kfree(tr->topts[i].topts);
6732 mutex_unlock(&trace_types_lock);
6737 static __init void create_trace_instances(struct dentry *d_tracer)
6739 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
6742 if (WARN_ON(!trace_instance_dir))
6747 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
6751 trace_create_file("available_tracers", 0444, d_tracer,
6752 tr, &show_traces_fops);
6754 trace_create_file("current_tracer", 0644, d_tracer,
6755 tr, &set_tracer_fops);
6757 trace_create_file("tracing_cpumask", 0644, d_tracer,
6758 tr, &tracing_cpumask_fops);
6760 trace_create_file("trace_options", 0644, d_tracer,
6761 tr, &tracing_iter_fops);
6763 trace_create_file("trace", 0644, d_tracer,
6766 trace_create_file("trace_pipe", 0444, d_tracer,
6767 tr, &tracing_pipe_fops);
6769 trace_create_file("buffer_size_kb", 0644, d_tracer,
6770 tr, &tracing_entries_fops);
6772 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
6773 tr, &tracing_total_entries_fops);
6775 trace_create_file("free_buffer", 0200, d_tracer,
6776 tr, &tracing_free_buffer_fops);
6778 trace_create_file("trace_marker", 0220, d_tracer,
6779 tr, &tracing_mark_fops);
6781 trace_create_file("trace_clock", 0644, d_tracer, tr,
6784 trace_create_file("tracing_on", 0644, d_tracer,
6785 tr, &rb_simple_fops);
6787 create_trace_options_dir(tr);
6789 #ifdef CONFIG_TRACER_MAX_TRACE
6790 trace_create_file("tracing_max_latency", 0644, d_tracer,
6791 &tr->max_latency, &tracing_max_lat_fops);
6794 if (ftrace_create_function_files(tr, d_tracer))
6795 WARN(1, "Could not allocate function filter files");
6797 #ifdef CONFIG_TRACER_SNAPSHOT
6798 trace_create_file("snapshot", 0644, d_tracer,
6799 tr, &snapshot_fops);
6802 for_each_tracing_cpu(cpu)
6803 tracing_init_tracefs_percpu(tr, cpu);
6807 static struct vfsmount *trace_automount(void *ingore)
6809 struct vfsmount *mnt;
6810 struct file_system_type *type;
6813 * To maintain backward compatibility for tools that mount
6814 * debugfs to get to the tracing facility, tracefs is automatically
6815 * mounted to the debugfs/tracing directory.
6817 type = get_fs_type("tracefs");
6820 mnt = vfs_kern_mount(type, 0, "tracefs", NULL);
6821 put_filesystem(type);
6830 * tracing_init_dentry - initialize top level trace array
6832 * This is called when creating files or directories in the tracing
6833 * directory. It is called via fs_initcall() by any of the boot up code
6834 * and expects to return the dentry of the top level tracing directory.
6836 struct dentry *tracing_init_dentry(void)
6838 struct trace_array *tr = &global_trace;
6840 /* The top level trace array uses NULL as parent */
6844 if (WARN_ON(!tracefs_initialized()) ||
6845 (IS_ENABLED(CONFIG_DEBUG_FS) &&
6846 WARN_ON(!debugfs_initialized())))
6847 return ERR_PTR(-ENODEV);
6850 * As there may still be users that expect the tracing
6851 * files to exist in debugfs/tracing, we must automount
6852 * the tracefs file system there, so older tools still
6853 * work with the newer kerenl.
6855 tr->dir = debugfs_create_automount("tracing", NULL,
6856 trace_automount, NULL);
6858 pr_warn_once("Could not create debugfs directory 'tracing'\n");
6859 return ERR_PTR(-ENOMEM);
6865 extern struct trace_enum_map *__start_ftrace_enum_maps[];
6866 extern struct trace_enum_map *__stop_ftrace_enum_maps[];
6868 static void __init trace_enum_init(void)
6872 len = __stop_ftrace_enum_maps - __start_ftrace_enum_maps;
6873 trace_insert_enum_map(NULL, __start_ftrace_enum_maps, len);
6876 #ifdef CONFIG_MODULES
6877 static void trace_module_add_enums(struct module *mod)
6879 if (!mod->num_trace_enums)
6883 * Modules with bad taint do not have events created, do
6884 * not bother with enums either.
6886 if (trace_module_has_bad_taint(mod))
6889 trace_insert_enum_map(mod, mod->trace_enums, mod->num_trace_enums);
6892 #ifdef CONFIG_TRACE_ENUM_MAP_FILE
6893 static void trace_module_remove_enums(struct module *mod)
6895 union trace_enum_map_item *map;
6896 union trace_enum_map_item **last = &trace_enum_maps;
6898 if (!mod->num_trace_enums)
6901 mutex_lock(&trace_enum_mutex);
6903 map = trace_enum_maps;
6906 if (map->head.mod == mod)
6908 map = trace_enum_jmp_to_tail(map);
6909 last = &map->tail.next;
6910 map = map->tail.next;
6915 *last = trace_enum_jmp_to_tail(map)->tail.next;
6918 mutex_unlock(&trace_enum_mutex);
6921 static inline void trace_module_remove_enums(struct module *mod) { }
6922 #endif /* CONFIG_TRACE_ENUM_MAP_FILE */
6924 static int trace_module_notify(struct notifier_block *self,
6925 unsigned long val, void *data)
6927 struct module *mod = data;
6930 case MODULE_STATE_COMING:
6931 trace_module_add_enums(mod);
6933 case MODULE_STATE_GOING:
6934 trace_module_remove_enums(mod);
6941 static struct notifier_block trace_module_nb = {
6942 .notifier_call = trace_module_notify,
6945 #endif /* CONFIG_MODULES */
6947 static __init int tracer_init_tracefs(void)
6949 struct dentry *d_tracer;
6951 trace_access_lock_init();
6953 d_tracer = tracing_init_dentry();
6954 if (IS_ERR(d_tracer))
6957 init_tracer_tracefs(&global_trace, d_tracer);
6959 trace_create_file("tracing_thresh", 0644, d_tracer,
6960 &global_trace, &tracing_thresh_fops);
6962 trace_create_file("README", 0444, d_tracer,
6963 NULL, &tracing_readme_fops);
6965 trace_create_file("saved_cmdlines", 0444, d_tracer,
6966 NULL, &tracing_saved_cmdlines_fops);
6968 trace_create_file("saved_cmdlines_size", 0644, d_tracer,
6969 NULL, &tracing_saved_cmdlines_size_fops);
6973 trace_create_enum_file(d_tracer);
6975 #ifdef CONFIG_MODULES
6976 register_module_notifier(&trace_module_nb);
6979 #ifdef CONFIG_DYNAMIC_FTRACE
6980 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
6981 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
6984 create_trace_instances(d_tracer);
6986 update_tracer_options(&global_trace);
6991 static int trace_panic_handler(struct notifier_block *this,
6992 unsigned long event, void *unused)
6994 if (ftrace_dump_on_oops)
6995 ftrace_dump(ftrace_dump_on_oops);
6999 static struct notifier_block trace_panic_notifier = {
7000 .notifier_call = trace_panic_handler,
7002 .priority = 150 /* priority: INT_MAX >= x >= 0 */
7005 static int trace_die_handler(struct notifier_block *self,
7011 if (ftrace_dump_on_oops)
7012 ftrace_dump(ftrace_dump_on_oops);
7020 static struct notifier_block trace_die_notifier = {
7021 .notifier_call = trace_die_handler,
7026 * printk is set to max of 1024, we really don't need it that big.
7027 * Nothing should be printing 1000 characters anyway.
7029 #define TRACE_MAX_PRINT 1000
7032 * Define here KERN_TRACE so that we have one place to modify
7033 * it if we decide to change what log level the ftrace dump
7036 #define KERN_TRACE KERN_EMERG
7039 trace_printk_seq(struct trace_seq *s)
7041 /* Probably should print a warning here. */
7042 if (s->seq.len >= TRACE_MAX_PRINT)
7043 s->seq.len = TRACE_MAX_PRINT;
7046 * More paranoid code. Although the buffer size is set to
7047 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
7048 * an extra layer of protection.
7050 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
7051 s->seq.len = s->seq.size - 1;
7053 /* should be zero ended, but we are paranoid. */
7054 s->buffer[s->seq.len] = 0;
7056 printk(KERN_TRACE "%s", s->buffer);
7061 void trace_init_global_iter(struct trace_iterator *iter)
7063 iter->tr = &global_trace;
7064 iter->trace = iter->tr->current_trace;
7065 iter->cpu_file = RING_BUFFER_ALL_CPUS;
7066 iter->trace_buffer = &global_trace.trace_buffer;
7068 if (iter->trace && iter->trace->open)
7069 iter->trace->open(iter);
7071 /* Annotate start of buffers if we had overruns */
7072 if (ring_buffer_overruns(iter->trace_buffer->buffer))
7073 iter->iter_flags |= TRACE_FILE_ANNOTATE;
7075 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
7076 if (trace_clocks[iter->tr->clock_id].in_ns)
7077 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
7080 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
7082 /* use static because iter can be a bit big for the stack */
7083 static struct trace_iterator iter;
7084 static atomic_t dump_running;
7085 struct trace_array *tr = &global_trace;
7086 unsigned int old_userobj;
7087 unsigned long flags;
7090 /* Only allow one dump user at a time. */
7091 if (atomic_inc_return(&dump_running) != 1) {
7092 atomic_dec(&dump_running);
7097 * Always turn off tracing when we dump.
7098 * We don't need to show trace output of what happens
7099 * between multiple crashes.
7101 * If the user does a sysrq-z, then they can re-enable
7102 * tracing with echo 1 > tracing_on.
7106 local_irq_save(flags);
7108 /* Simulate the iterator */
7109 trace_init_global_iter(&iter);
7111 for_each_tracing_cpu(cpu) {
7112 atomic_inc(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
7115 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
7117 /* don't look at user memory in panic mode */
7118 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
7120 switch (oops_dump_mode) {
7122 iter.cpu_file = RING_BUFFER_ALL_CPUS;
7125 iter.cpu_file = raw_smp_processor_id();
7130 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
7131 iter.cpu_file = RING_BUFFER_ALL_CPUS;
7134 printk(KERN_TRACE "Dumping ftrace buffer:\n");
7136 /* Did function tracer already get disabled? */
7137 if (ftrace_is_dead()) {
7138 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
7139 printk("# MAY BE MISSING FUNCTION EVENTS\n");
7143 * We need to stop all tracing on all CPUS to read the
7144 * the next buffer. This is a bit expensive, but is
7145 * not done often. We fill all what we can read,
7146 * and then release the locks again.
7149 while (!trace_empty(&iter)) {
7152 printk(KERN_TRACE "---------------------------------\n");
7156 /* reset all but tr, trace, and overruns */
7157 memset(&iter.seq, 0,
7158 sizeof(struct trace_iterator) -
7159 offsetof(struct trace_iterator, seq));
7160 iter.iter_flags |= TRACE_FILE_LAT_FMT;
7163 if (trace_find_next_entry_inc(&iter) != NULL) {
7166 ret = print_trace_line(&iter);
7167 if (ret != TRACE_TYPE_NO_CONSUME)
7168 trace_consume(&iter);
7170 touch_nmi_watchdog();
7172 trace_printk_seq(&iter.seq);
7176 printk(KERN_TRACE " (ftrace buffer empty)\n");
7178 printk(KERN_TRACE "---------------------------------\n");
7181 tr->trace_flags |= old_userobj;
7183 for_each_tracing_cpu(cpu) {
7184 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
7186 atomic_dec(&dump_running);
7187 local_irq_restore(flags);
7189 EXPORT_SYMBOL_GPL(ftrace_dump);
7191 __init static int tracer_alloc_buffers(void)
7197 * Make sure we don't accidently add more trace options
7198 * than we have bits for.
7200 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
7202 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
7205 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
7206 goto out_free_buffer_mask;
7208 /* Only allocate trace_printk buffers if a trace_printk exists */
7209 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
7210 /* Must be called before global_trace.buffer is allocated */
7211 trace_printk_init_buffers();
7213 /* To save memory, keep the ring buffer size to its minimum */
7214 if (ring_buffer_expanded)
7215 ring_buf_size = trace_buf_size;
7219 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
7220 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
7222 raw_spin_lock_init(&global_trace.start_lock);
7224 /* Used for event triggers */
7225 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
7227 goto out_free_cpumask;
7229 if (trace_create_savedcmd() < 0)
7230 goto out_free_temp_buffer;
7232 /* TODO: make the number of buffers hot pluggable with CPUS */
7233 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
7234 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
7236 goto out_free_savedcmd;
7239 if (global_trace.buffer_disabled)
7242 if (trace_boot_clock) {
7243 ret = tracing_set_clock(&global_trace, trace_boot_clock);
7245 pr_warning("Trace clock %s not defined, going back to default\n",
7250 * register_tracer() might reference current_trace, so it
7251 * needs to be set before we register anything. This is
7252 * just a bootstrap of current_trace anyway.
7254 global_trace.current_trace = &nop_trace;
7256 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
7258 ftrace_init_global_array_ops(&global_trace);
7260 init_trace_flags_index(&global_trace);
7262 register_tracer(&nop_trace);
7264 /* All seems OK, enable tracing */
7265 tracing_disabled = 0;
7267 atomic_notifier_chain_register(&panic_notifier_list,
7268 &trace_panic_notifier);
7270 register_die_notifier(&trace_die_notifier);
7272 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
7274 INIT_LIST_HEAD(&global_trace.systems);
7275 INIT_LIST_HEAD(&global_trace.events);
7276 list_add(&global_trace.list, &ftrace_trace_arrays);
7278 apply_trace_boot_options();
7280 register_snapshot_cmd();
7285 free_saved_cmdlines_buffer(savedcmd);
7286 out_free_temp_buffer:
7287 ring_buffer_free(temp_buffer);
7289 free_cpumask_var(global_trace.tracing_cpumask);
7290 out_free_buffer_mask:
7291 free_cpumask_var(tracing_buffer_mask);
7296 void __init trace_init(void)
7298 if (tracepoint_printk) {
7299 tracepoint_print_iter =
7300 kmalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
7301 if (WARN_ON(!tracepoint_print_iter))
7302 tracepoint_printk = 0;
7304 tracer_alloc_buffers();
7308 __init static int clear_boot_tracer(void)
7311 * The default tracer at boot buffer is an init section.
7312 * This function is called in lateinit. If we did not
7313 * find the boot tracer, then clear it out, to prevent
7314 * later registration from accessing the buffer that is
7315 * about to be freed.
7317 if (!default_bootup_tracer)
7320 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
7321 default_bootup_tracer);
7322 default_bootup_tracer = NULL;
7327 fs_initcall(tracer_init_tracefs);
7328 late_initcall(clear_boot_tracer);