2 * ring buffer based function tracer
4 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 Nadia Yvette Chambers
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/nmi.h>
41 #include <linux/sched/rt.h>
44 #include "trace_output.h"
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
50 bool ring_buffer_expanded;
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
59 static bool __read_mostly tracing_selftest_running;
62 * If a tracer is running, we do not want to run SELFTEST.
64 bool __read_mostly tracing_selftest_disabled;
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt[] = {
71 static struct tracer_flags dummy_tracer_flags = {
73 .opts = dummy_tracer_opt
76 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
82 * To prevent the comm cache from being overwritten when no
83 * tracing is active, only save the comm when a trace event
86 static DEFINE_PER_CPU(bool, trace_cmdline_save);
89 * Kill all tracing for good (never come back).
90 * It is initialized to 1 but will turn to zero if the initialization
91 * of the tracer is successful. But that is the only place that sets
94 static int tracing_disabled = 1;
96 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
98 cpumask_var_t __read_mostly tracing_buffer_mask;
101 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
103 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
104 * is set, then ftrace_dump is called. This will output the contents
105 * of the ftrace buffers to the console. This is very useful for
106 * capturing traces that lead to crashes and outputing it to a
109 * It is default off, but you can enable it with either specifying
110 * "ftrace_dump_on_oops" in the kernel command line, or setting
111 * /proc/sys/kernel/ftrace_dump_on_oops
112 * Set 1 if you want to dump buffers of all CPUs
113 * Set 2 if you want to dump the buffer of the CPU that triggered oops
116 enum ftrace_dump_mode ftrace_dump_on_oops;
118 static int tracing_set_tracer(const char *buf);
120 #define MAX_TRACER_SIZE 100
121 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
122 static char *default_bootup_tracer;
124 static bool allocate_snapshot;
126 static int __init set_cmdline_ftrace(char *str)
128 strlcpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
129 default_bootup_tracer = bootup_tracer_buf;
130 /* We are using ftrace early, expand it */
131 ring_buffer_expanded = true;
134 __setup("ftrace=", set_cmdline_ftrace);
136 static int __init set_ftrace_dump_on_oops(char *str)
138 if (*str++ != '=' || !*str) {
139 ftrace_dump_on_oops = DUMP_ALL;
143 if (!strcmp("orig_cpu", str)) {
144 ftrace_dump_on_oops = DUMP_ORIG;
150 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
152 static int __init boot_alloc_snapshot(char *str)
154 allocate_snapshot = true;
155 /* We also need the main ring buffer expanded */
156 ring_buffer_expanded = true;
159 __setup("alloc_snapshot", boot_alloc_snapshot);
162 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
163 static char *trace_boot_options __initdata;
165 static int __init set_trace_boot_options(char *str)
167 strlcpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
168 trace_boot_options = trace_boot_options_buf;
171 __setup("trace_options=", set_trace_boot_options);
173 unsigned long long ns2usecs(cycle_t nsec)
181 * The global_trace is the descriptor that holds the tracing
182 * buffers for the live tracing. For each CPU, it contains
183 * a link list of pages that will store trace entries. The
184 * page descriptor of the pages in the memory is used to hold
185 * the link list by linking the lru item in the page descriptor
186 * to each of the pages in the buffer per CPU.
188 * For each active CPU there is a data field that holds the
189 * pages for the buffer for that CPU. Each CPU has the same number
190 * of pages allocated for its buffer.
192 static struct trace_array global_trace;
194 LIST_HEAD(ftrace_trace_arrays);
196 int trace_array_get(struct trace_array *this_tr)
198 struct trace_array *tr;
201 mutex_lock(&trace_types_lock);
202 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
209 mutex_unlock(&trace_types_lock);
214 static void __trace_array_put(struct trace_array *this_tr)
216 WARN_ON(!this_tr->ref);
220 void trace_array_put(struct trace_array *this_tr)
222 mutex_lock(&trace_types_lock);
223 __trace_array_put(this_tr);
224 mutex_unlock(&trace_types_lock);
227 int filter_current_check_discard(struct ring_buffer *buffer,
228 struct ftrace_event_call *call, void *rec,
229 struct ring_buffer_event *event)
231 return filter_check_discard(call, rec, buffer, event);
233 EXPORT_SYMBOL_GPL(filter_current_check_discard);
235 cycle_t buffer_ftrace_now(struct trace_buffer *buf, int cpu)
239 /* Early boot up does not have a buffer yet */
241 return trace_clock_local();
243 ts = ring_buffer_time_stamp(buf->buffer, cpu);
244 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
249 cycle_t ftrace_now(int cpu)
251 return buffer_ftrace_now(&global_trace.trace_buffer, cpu);
255 * tracing_is_enabled - Show if global_trace has been disabled
257 * Shows if the global trace has been enabled or not. It uses the
258 * mirror flag "buffer_disabled" to be used in fast paths such as for
259 * the irqsoff tracer. But it may be inaccurate due to races. If you
260 * need to know the accurate state, use tracing_is_on() which is a little
261 * slower, but accurate.
263 int tracing_is_enabled(void)
266 * For quick access (irqsoff uses this in fast path), just
267 * return the mirror variable of the state of the ring buffer.
268 * It's a little racy, but we don't really care.
271 return !global_trace.buffer_disabled;
275 * trace_buf_size is the size in bytes that is allocated
276 * for a buffer. Note, the number of bytes is always rounded
279 * This number is purposely set to a low number of 16384.
280 * If the dump on oops happens, it will be much appreciated
281 * to not have to wait for all that output. Anyway this can be
282 * boot time and run time configurable.
284 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
286 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
288 /* trace_types holds a link list of available tracers. */
289 static struct tracer *trace_types __read_mostly;
292 * trace_types_lock is used to protect the trace_types list.
294 DEFINE_MUTEX(trace_types_lock);
297 * serialize the access of the ring buffer
299 * ring buffer serializes readers, but it is low level protection.
300 * The validity of the events (which returns by ring_buffer_peek() ..etc)
301 * are not protected by ring buffer.
303 * The content of events may become garbage if we allow other process consumes
304 * these events concurrently:
305 * A) the page of the consumed events may become a normal page
306 * (not reader page) in ring buffer, and this page will be rewrited
307 * by events producer.
308 * B) The page of the consumed events may become a page for splice_read,
309 * and this page will be returned to system.
311 * These primitives allow multi process access to different cpu ring buffer
314 * These primitives don't distinguish read-only and read-consume access.
315 * Multi read-only access are also serialized.
319 static DECLARE_RWSEM(all_cpu_access_lock);
320 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
322 static inline void trace_access_lock(int cpu)
324 if (cpu == RING_BUFFER_ALL_CPUS) {
325 /* gain it for accessing the whole ring buffer. */
326 down_write(&all_cpu_access_lock);
328 /* gain it for accessing a cpu ring buffer. */
330 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
331 down_read(&all_cpu_access_lock);
333 /* Secondly block other access to this @cpu ring buffer. */
334 mutex_lock(&per_cpu(cpu_access_lock, cpu));
338 static inline void trace_access_unlock(int cpu)
340 if (cpu == RING_BUFFER_ALL_CPUS) {
341 up_write(&all_cpu_access_lock);
343 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
344 up_read(&all_cpu_access_lock);
348 static inline void trace_access_lock_init(void)
352 for_each_possible_cpu(cpu)
353 mutex_init(&per_cpu(cpu_access_lock, cpu));
358 static DEFINE_MUTEX(access_lock);
360 static inline void trace_access_lock(int cpu)
363 mutex_lock(&access_lock);
366 static inline void trace_access_unlock(int cpu)
369 mutex_unlock(&access_lock);
372 static inline void trace_access_lock_init(void)
378 /* trace_flags holds trace_options default values */
379 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
380 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
381 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
382 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | TRACE_ITER_FUNCTION;
384 void tracer_tracing_on(struct trace_array *tr)
386 if (tr->trace_buffer.buffer)
387 ring_buffer_record_on(tr->trace_buffer.buffer);
389 * This flag is looked at when buffers haven't been allocated
390 * yet, or by some tracers (like irqsoff), that just want to
391 * know if the ring buffer has been disabled, but it can handle
392 * races of where it gets disabled but we still do a record.
393 * As the check is in the fast path of the tracers, it is more
394 * important to be fast than accurate.
396 tr->buffer_disabled = 0;
397 /* Make the flag seen by readers */
402 * tracing_on - enable tracing buffers
404 * This function enables tracing buffers that may have been
405 * disabled with tracing_off.
407 void tracing_on(void)
409 tracer_tracing_on(&global_trace);
411 EXPORT_SYMBOL_GPL(tracing_on);
414 * __trace_puts - write a constant string into the trace buffer.
415 * @ip: The address of the caller
416 * @str: The constant string to write
417 * @size: The size of the string.
419 int __trace_puts(unsigned long ip, const char *str, int size)
421 struct ring_buffer_event *event;
422 struct ring_buffer *buffer;
423 struct print_entry *entry;
424 unsigned long irq_flags;
428 pc = preempt_count();
430 if (unlikely(tracing_selftest_running || tracing_disabled))
433 alloc = sizeof(*entry) + size + 2; /* possible \n added */
435 local_save_flags(irq_flags);
436 buffer = global_trace.trace_buffer.buffer;
437 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
442 entry = ring_buffer_event_data(event);
445 memcpy(&entry->buf, str, size);
447 /* Add a newline if necessary */
448 if (entry->buf[size - 1] != '\n') {
449 entry->buf[size] = '\n';
450 entry->buf[size + 1] = '\0';
452 entry->buf[size] = '\0';
454 __buffer_unlock_commit(buffer, event);
455 ftrace_trace_stack(buffer, irq_flags, 4, pc);
459 EXPORT_SYMBOL_GPL(__trace_puts);
462 * __trace_bputs - write the pointer to a constant string into trace buffer
463 * @ip: The address of the caller
464 * @str: The constant string to write to the buffer to
466 int __trace_bputs(unsigned long ip, const char *str)
468 struct ring_buffer_event *event;
469 struct ring_buffer *buffer;
470 struct bputs_entry *entry;
471 unsigned long irq_flags;
472 int size = sizeof(struct bputs_entry);
475 pc = preempt_count();
477 if (unlikely(tracing_selftest_running || tracing_disabled))
480 local_save_flags(irq_flags);
481 buffer = global_trace.trace_buffer.buffer;
482 event = trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
487 entry = ring_buffer_event_data(event);
491 __buffer_unlock_commit(buffer, event);
492 ftrace_trace_stack(buffer, irq_flags, 4, pc);
496 EXPORT_SYMBOL_GPL(__trace_bputs);
498 #ifdef CONFIG_TRACER_SNAPSHOT
500 * trace_snapshot - take a snapshot of the current buffer.
502 * This causes a swap between the snapshot buffer and the current live
503 * tracing buffer. You can use this to take snapshots of the live
504 * trace when some condition is triggered, but continue to trace.
506 * Note, make sure to allocate the snapshot with either
507 * a tracing_snapshot_alloc(), or by doing it manually
508 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
510 * If the snapshot buffer is not allocated, it will stop tracing.
511 * Basically making a permanent snapshot.
513 void tracing_snapshot(void)
515 struct trace_array *tr = &global_trace;
516 struct tracer *tracer = tr->current_trace;
520 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
521 internal_trace_puts("*** snapshot is being ignored ***\n");
525 if (!tr->allocated_snapshot) {
526 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
527 internal_trace_puts("*** stopping trace here! ***\n");
532 /* Note, snapshot can not be used when the tracer uses it */
533 if (tracer->use_max_tr) {
534 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
535 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
539 local_irq_save(flags);
540 update_max_tr(tr, current, smp_processor_id());
541 local_irq_restore(flags);
543 EXPORT_SYMBOL_GPL(tracing_snapshot);
545 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
546 struct trace_buffer *size_buf, int cpu_id);
547 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val);
549 static int alloc_snapshot(struct trace_array *tr)
553 if (!tr->allocated_snapshot) {
555 /* allocate spare buffer */
556 ret = resize_buffer_duplicate_size(&tr->max_buffer,
557 &tr->trace_buffer, RING_BUFFER_ALL_CPUS);
561 tr->allocated_snapshot = true;
567 void free_snapshot(struct trace_array *tr)
570 * We don't free the ring buffer. instead, resize it because
571 * The max_tr ring buffer has some state (e.g. ring->clock) and
572 * we want preserve it.
574 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
575 set_buffer_entries(&tr->max_buffer, 1);
576 tracing_reset_online_cpus(&tr->max_buffer);
577 tr->allocated_snapshot = false;
581 * trace_snapshot_alloc - allocate and take a snapshot of the current buffer.
583 * This is similar to trace_snapshot(), but it will allocate the
584 * snapshot buffer if it isn't already allocated. Use this only
585 * where it is safe to sleep, as the allocation may sleep.
587 * This causes a swap between the snapshot buffer and the current live
588 * tracing buffer. You can use this to take snapshots of the live
589 * trace when some condition is triggered, but continue to trace.
591 void tracing_snapshot_alloc(void)
593 struct trace_array *tr = &global_trace;
596 ret = alloc_snapshot(tr);
597 if (WARN_ON(ret < 0))
602 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
604 void tracing_snapshot(void)
606 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
608 EXPORT_SYMBOL_GPL(tracing_snapshot);
609 void tracing_snapshot_alloc(void)
614 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
615 #endif /* CONFIG_TRACER_SNAPSHOT */
617 void tracer_tracing_off(struct trace_array *tr)
619 if (tr->trace_buffer.buffer)
620 ring_buffer_record_off(tr->trace_buffer.buffer);
622 * This flag is looked at when buffers haven't been allocated
623 * yet, or by some tracers (like irqsoff), that just want to
624 * know if the ring buffer has been disabled, but it can handle
625 * races of where it gets disabled but we still do a record.
626 * As the check is in the fast path of the tracers, it is more
627 * important to be fast than accurate.
629 tr->buffer_disabled = 1;
630 /* Make the flag seen by readers */
635 * tracing_off - turn off tracing buffers
637 * This function stops the tracing buffers from recording data.
638 * It does not disable any overhead the tracers themselves may
639 * be causing. This function simply causes all recording to
640 * the ring buffers to fail.
642 void tracing_off(void)
644 tracer_tracing_off(&global_trace);
646 EXPORT_SYMBOL_GPL(tracing_off);
649 * tracer_tracing_is_on - show real state of ring buffer enabled
650 * @tr : the trace array to know if ring buffer is enabled
652 * Shows real state of the ring buffer if it is enabled or not.
654 int tracer_tracing_is_on(struct trace_array *tr)
656 if (tr->trace_buffer.buffer)
657 return ring_buffer_record_is_on(tr->trace_buffer.buffer);
658 return !tr->buffer_disabled;
662 * tracing_is_on - show state of ring buffers enabled
664 int tracing_is_on(void)
666 return tracer_tracing_is_on(&global_trace);
668 EXPORT_SYMBOL_GPL(tracing_is_on);
670 static int __init set_buf_size(char *str)
672 unsigned long buf_size;
676 buf_size = memparse(str, &str);
677 /* nr_entries can not be zero */
680 trace_buf_size = buf_size;
683 __setup("trace_buf_size=", set_buf_size);
685 static int __init set_tracing_thresh(char *str)
687 unsigned long threshold;
692 ret = kstrtoul(str, 0, &threshold);
695 tracing_thresh = threshold * 1000;
698 __setup("tracing_thresh=", set_tracing_thresh);
700 unsigned long nsecs_to_usecs(unsigned long nsecs)
705 /* These must match the bit postions in trace_iterator_flags */
706 static const char *trace_options[] = {
739 int in_ns; /* is this clock in nanoseconds? */
741 { trace_clock_local, "local", 1 },
742 { trace_clock_global, "global", 1 },
743 { trace_clock_counter, "counter", 0 },
744 { trace_clock_jiffies, "uptime", 1 },
745 { trace_clock, "perf", 1 },
750 * trace_parser_get_init - gets the buffer for trace parser
752 int trace_parser_get_init(struct trace_parser *parser, int size)
754 memset(parser, 0, sizeof(*parser));
756 parser->buffer = kmalloc(size, GFP_KERNEL);
765 * trace_parser_put - frees the buffer for trace parser
767 void trace_parser_put(struct trace_parser *parser)
769 kfree(parser->buffer);
773 * trace_get_user - reads the user input string separated by space
774 * (matched by isspace(ch))
776 * For each string found the 'struct trace_parser' is updated,
777 * and the function returns.
779 * Returns number of bytes read.
781 * See kernel/trace/trace.h for 'struct trace_parser' details.
783 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
784 size_t cnt, loff_t *ppos)
791 trace_parser_clear(parser);
793 ret = get_user(ch, ubuf++);
801 * The parser is not finished with the last write,
802 * continue reading the user input without skipping spaces.
805 /* skip white space */
806 while (cnt && isspace(ch)) {
807 ret = get_user(ch, ubuf++);
814 /* only spaces were written */
824 /* read the non-space input */
825 while (cnt && !isspace(ch)) {
826 if (parser->idx < parser->size - 1)
827 parser->buffer[parser->idx++] = ch;
832 ret = get_user(ch, ubuf++);
839 /* We either got finished input or we have to wait for another call. */
841 parser->buffer[parser->idx] = 0;
842 parser->cont = false;
843 } else if (parser->idx < parser->size - 1) {
845 parser->buffer[parser->idx++] = ch;
858 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
866 if (s->len <= s->readpos)
869 len = s->len - s->readpos;
872 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
882 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
886 if (s->len <= s->readpos)
889 len = s->len - s->readpos;
892 memcpy(buf, s->buffer + s->readpos, cnt);
899 * ftrace_max_lock is used to protect the swapping of buffers
900 * when taking a max snapshot. The buffers themselves are
901 * protected by per_cpu spinlocks. But the action of the swap
902 * needs its own lock.
904 * This is defined as a arch_spinlock_t in order to help
905 * with performance when lockdep debugging is enabled.
907 * It is also used in other places outside the update_max_tr
908 * so it needs to be defined outside of the
909 * CONFIG_TRACER_MAX_TRACE.
911 static arch_spinlock_t ftrace_max_lock =
912 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
914 unsigned long __read_mostly tracing_thresh;
916 #ifdef CONFIG_TRACER_MAX_TRACE
917 unsigned long __read_mostly tracing_max_latency;
920 * Copy the new maximum trace into the separate maximum-trace
921 * structure. (this way the maximum trace is permanently saved,
922 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
925 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
927 struct trace_buffer *trace_buf = &tr->trace_buffer;
928 struct trace_buffer *max_buf = &tr->max_buffer;
929 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
930 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
933 max_buf->time_start = data->preempt_timestamp;
935 max_data->saved_latency = tracing_max_latency;
936 max_data->critical_start = data->critical_start;
937 max_data->critical_end = data->critical_end;
939 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
940 max_data->pid = tsk->pid;
942 * If tsk == current, then use current_uid(), as that does not use
943 * RCU. The irq tracer can be called out of RCU scope.
946 max_data->uid = current_uid();
948 max_data->uid = task_uid(tsk);
950 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
951 max_data->policy = tsk->policy;
952 max_data->rt_priority = tsk->rt_priority;
954 /* record this tasks comm */
955 tracing_record_cmdline(tsk);
959 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
961 * @tsk: the task with the latency
962 * @cpu: The cpu that initiated the trace.
964 * Flip the buffers between the @tr and the max_tr and record information
965 * about which task was the cause of this latency.
968 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
970 struct ring_buffer *buf;
975 WARN_ON_ONCE(!irqs_disabled());
977 if (!tr->allocated_snapshot) {
978 /* Only the nop tracer should hit this when disabling */
979 WARN_ON_ONCE(tr->current_trace != &nop_trace);
983 arch_spin_lock(&ftrace_max_lock);
985 buf = tr->trace_buffer.buffer;
986 tr->trace_buffer.buffer = tr->max_buffer.buffer;
987 tr->max_buffer.buffer = buf;
989 __update_max_tr(tr, tsk, cpu);
990 arch_spin_unlock(&ftrace_max_lock);
994 * update_max_tr_single - only copy one trace over, and reset the rest
996 * @tsk - task with the latency
997 * @cpu - the cpu of the buffer to copy.
999 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1002 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1009 WARN_ON_ONCE(!irqs_disabled());
1010 if (!tr->allocated_snapshot) {
1011 /* Only the nop tracer should hit this when disabling */
1012 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1016 arch_spin_lock(&ftrace_max_lock);
1018 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->trace_buffer.buffer, cpu);
1020 if (ret == -EBUSY) {
1022 * We failed to swap the buffer due to a commit taking
1023 * place on this CPU. We fail to record, but we reset
1024 * the max trace buffer (no one writes directly to it)
1025 * and flag that it failed.
1027 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1028 "Failed to swap buffers due to commit in progress\n");
1031 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1033 __update_max_tr(tr, tsk, cpu);
1034 arch_spin_unlock(&ftrace_max_lock);
1036 #endif /* CONFIG_TRACER_MAX_TRACE */
1038 static int default_wait_pipe(struct trace_iterator *iter)
1040 /* Iterators are static, they should be filled or empty */
1041 if (trace_buffer_iter(iter, iter->cpu_file))
1044 return ring_buffer_wait(iter->trace_buffer->buffer, iter->cpu_file);
1047 #ifdef CONFIG_FTRACE_STARTUP_TEST
1048 static int run_tracer_selftest(struct tracer *type)
1050 struct trace_array *tr = &global_trace;
1051 struct tracer *saved_tracer = tr->current_trace;
1054 if (!type->selftest || tracing_selftest_disabled)
1058 * Run a selftest on this tracer.
1059 * Here we reset the trace buffer, and set the current
1060 * tracer to be this tracer. The tracer can then run some
1061 * internal tracing to verify that everything is in order.
1062 * If we fail, we do not register this tracer.
1064 tracing_reset_online_cpus(&tr->trace_buffer);
1066 tr->current_trace = type;
1068 #ifdef CONFIG_TRACER_MAX_TRACE
1069 if (type->use_max_tr) {
1070 /* If we expanded the buffers, make sure the max is expanded too */
1071 if (ring_buffer_expanded)
1072 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
1073 RING_BUFFER_ALL_CPUS);
1074 tr->allocated_snapshot = true;
1078 /* the test is responsible for initializing and enabling */
1079 pr_info("Testing tracer %s: ", type->name);
1080 ret = type->selftest(type, tr);
1081 /* the test is responsible for resetting too */
1082 tr->current_trace = saved_tracer;
1084 printk(KERN_CONT "FAILED!\n");
1085 /* Add the warning after printing 'FAILED' */
1089 /* Only reset on passing, to avoid touching corrupted buffers */
1090 tracing_reset_online_cpus(&tr->trace_buffer);
1092 #ifdef CONFIG_TRACER_MAX_TRACE
1093 if (type->use_max_tr) {
1094 tr->allocated_snapshot = false;
1096 /* Shrink the max buffer again */
1097 if (ring_buffer_expanded)
1098 ring_buffer_resize(tr->max_buffer.buffer, 1,
1099 RING_BUFFER_ALL_CPUS);
1103 printk(KERN_CONT "PASSED\n");
1107 static inline int run_tracer_selftest(struct tracer *type)
1111 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1114 * register_tracer - register a tracer with the ftrace system.
1115 * @type - the plugin for the tracer
1117 * Register a new plugin tracer.
1119 int register_tracer(struct tracer *type)
1125 pr_info("Tracer must have a name\n");
1129 if (strlen(type->name) >= MAX_TRACER_SIZE) {
1130 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
1134 mutex_lock(&trace_types_lock);
1136 tracing_selftest_running = true;
1138 for (t = trace_types; t; t = t->next) {
1139 if (strcmp(type->name, t->name) == 0) {
1141 pr_info("Tracer %s already registered\n",
1148 if (!type->set_flag)
1149 type->set_flag = &dummy_set_flag;
1151 type->flags = &dummy_tracer_flags;
1153 if (!type->flags->opts)
1154 type->flags->opts = dummy_tracer_opt;
1155 if (!type->wait_pipe)
1156 type->wait_pipe = default_wait_pipe;
1158 ret = run_tracer_selftest(type);
1162 type->next = trace_types;
1166 tracing_selftest_running = false;
1167 mutex_unlock(&trace_types_lock);
1169 if (ret || !default_bootup_tracer)
1172 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
1175 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
1176 /* Do we want this tracer to start on bootup? */
1177 tracing_set_tracer(type->name);
1178 default_bootup_tracer = NULL;
1179 /* disable other selftests, since this will break it. */
1180 tracing_selftest_disabled = true;
1181 #ifdef CONFIG_FTRACE_STARTUP_TEST
1182 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
1190 void tracing_reset(struct trace_buffer *buf, int cpu)
1192 struct ring_buffer *buffer = buf->buffer;
1197 ring_buffer_record_disable(buffer);
1199 /* Make sure all commits have finished */
1200 synchronize_sched();
1201 ring_buffer_reset_cpu(buffer, cpu);
1203 ring_buffer_record_enable(buffer);
1206 void tracing_reset_online_cpus(struct trace_buffer *buf)
1208 struct ring_buffer *buffer = buf->buffer;
1214 ring_buffer_record_disable(buffer);
1216 /* Make sure all commits have finished */
1217 synchronize_sched();
1219 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
1221 for_each_online_cpu(cpu)
1222 ring_buffer_reset_cpu(buffer, cpu);
1224 ring_buffer_record_enable(buffer);
1227 /* Must have trace_types_lock held */
1228 void tracing_reset_all_online_cpus(void)
1230 struct trace_array *tr;
1232 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
1233 tracing_reset_online_cpus(&tr->trace_buffer);
1234 #ifdef CONFIG_TRACER_MAX_TRACE
1235 tracing_reset_online_cpus(&tr->max_buffer);
1240 #define SAVED_CMDLINES 128
1241 #define NO_CMDLINE_MAP UINT_MAX
1242 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
1243 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
1244 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
1245 static int cmdline_idx;
1246 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1248 /* temporary disable recording */
1249 static atomic_t trace_record_cmdline_disabled __read_mostly;
1251 static void trace_init_cmdlines(void)
1253 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
1254 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
1258 int is_tracing_stopped(void)
1260 return global_trace.stop_count;
1264 * ftrace_off_permanent - disable all ftrace code permanently
1266 * This should only be called when a serious anomally has
1267 * been detected. This will turn off the function tracing,
1268 * ring buffers, and other tracing utilites. It takes no
1269 * locks and can be called from any context.
1271 void ftrace_off_permanent(void)
1273 tracing_disabled = 1;
1275 tracing_off_permanent();
1279 * tracing_start - quick start of the tracer
1281 * If tracing is enabled but was stopped by tracing_stop,
1282 * this will start the tracer back up.
1284 void tracing_start(void)
1286 struct ring_buffer *buffer;
1287 unsigned long flags;
1289 if (tracing_disabled)
1292 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1293 if (--global_trace.stop_count) {
1294 if (global_trace.stop_count < 0) {
1295 /* Someone screwed up their debugging */
1297 global_trace.stop_count = 0;
1302 /* Prevent the buffers from switching */
1303 arch_spin_lock(&ftrace_max_lock);
1305 buffer = global_trace.trace_buffer.buffer;
1307 ring_buffer_record_enable(buffer);
1309 #ifdef CONFIG_TRACER_MAX_TRACE
1310 buffer = global_trace.max_buffer.buffer;
1312 ring_buffer_record_enable(buffer);
1315 arch_spin_unlock(&ftrace_max_lock);
1318 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1321 static void tracing_start_tr(struct trace_array *tr)
1323 struct ring_buffer *buffer;
1324 unsigned long flags;
1326 if (tracing_disabled)
1329 /* If global, we need to also start the max tracer */
1330 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1331 return tracing_start();
1333 raw_spin_lock_irqsave(&tr->start_lock, flags);
1335 if (--tr->stop_count) {
1336 if (tr->stop_count < 0) {
1337 /* Someone screwed up their debugging */
1344 buffer = tr->trace_buffer.buffer;
1346 ring_buffer_record_enable(buffer);
1349 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1353 * tracing_stop - quick stop of the tracer
1355 * Light weight way to stop tracing. Use in conjunction with
1358 void tracing_stop(void)
1360 struct ring_buffer *buffer;
1361 unsigned long flags;
1363 raw_spin_lock_irqsave(&global_trace.start_lock, flags);
1364 if (global_trace.stop_count++)
1367 /* Prevent the buffers from switching */
1368 arch_spin_lock(&ftrace_max_lock);
1370 buffer = global_trace.trace_buffer.buffer;
1372 ring_buffer_record_disable(buffer);
1374 #ifdef CONFIG_TRACER_MAX_TRACE
1375 buffer = global_trace.max_buffer.buffer;
1377 ring_buffer_record_disable(buffer);
1380 arch_spin_unlock(&ftrace_max_lock);
1383 raw_spin_unlock_irqrestore(&global_trace.start_lock, flags);
1386 static void tracing_stop_tr(struct trace_array *tr)
1388 struct ring_buffer *buffer;
1389 unsigned long flags;
1391 /* If global, we need to also stop the max tracer */
1392 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
1393 return tracing_stop();
1395 raw_spin_lock_irqsave(&tr->start_lock, flags);
1396 if (tr->stop_count++)
1399 buffer = tr->trace_buffer.buffer;
1401 ring_buffer_record_disable(buffer);
1404 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
1407 void trace_stop_cmdline_recording(void);
1409 static int trace_save_cmdline(struct task_struct *tsk)
1413 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1417 * It's not the end of the world if we don't get
1418 * the lock, but we also don't want to spin
1419 * nor do we want to disable interrupts,
1420 * so if we miss here, then better luck next time.
1422 if (!arch_spin_trylock(&trace_cmdline_lock))
1425 idx = map_pid_to_cmdline[tsk->pid];
1426 if (idx == NO_CMDLINE_MAP) {
1427 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1430 * Check whether the cmdline buffer at idx has a pid
1431 * mapped. We are going to overwrite that entry so we
1432 * need to clear the map_pid_to_cmdline. Otherwise we
1433 * would read the new comm for the old pid.
1435 pid = map_cmdline_to_pid[idx];
1436 if (pid != NO_CMDLINE_MAP)
1437 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1439 map_cmdline_to_pid[idx] = tsk->pid;
1440 map_pid_to_cmdline[tsk->pid] = idx;
1445 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1447 arch_spin_unlock(&trace_cmdline_lock);
1452 void trace_find_cmdline(int pid, char comm[])
1457 strcpy(comm, "<idle>");
1461 if (WARN_ON_ONCE(pid < 0)) {
1462 strcpy(comm, "<XXX>");
1466 if (pid > PID_MAX_DEFAULT) {
1467 strcpy(comm, "<...>");
1472 arch_spin_lock(&trace_cmdline_lock);
1473 map = map_pid_to_cmdline[pid];
1474 if (map != NO_CMDLINE_MAP)
1475 strcpy(comm, saved_cmdlines[map]);
1477 strcpy(comm, "<...>");
1479 arch_spin_unlock(&trace_cmdline_lock);
1483 void tracing_record_cmdline(struct task_struct *tsk)
1485 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1488 if (!__this_cpu_read(trace_cmdline_save))
1491 if (trace_save_cmdline(tsk))
1492 __this_cpu_write(trace_cmdline_save, false);
1496 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1499 struct task_struct *tsk = current;
1501 entry->preempt_count = pc & 0xff;
1502 entry->pid = (tsk) ? tsk->pid : 0;
1504 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1505 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1507 TRACE_FLAG_IRQS_NOSUPPORT |
1509 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1510 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1511 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1513 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1515 struct ring_buffer_event *
1516 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1519 unsigned long flags, int pc)
1521 struct ring_buffer_event *event;
1523 event = ring_buffer_lock_reserve(buffer, len);
1524 if (event != NULL) {
1525 struct trace_entry *ent = ring_buffer_event_data(event);
1527 tracing_generic_entry_update(ent, flags, pc);
1535 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1537 __this_cpu_write(trace_cmdline_save, true);
1538 ring_buffer_unlock_commit(buffer, event);
1542 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1543 struct ring_buffer_event *event,
1544 unsigned long flags, int pc)
1546 __buffer_unlock_commit(buffer, event);
1548 ftrace_trace_stack(buffer, flags, 6, pc);
1549 ftrace_trace_userstack(buffer, flags, pc);
1552 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1553 struct ring_buffer_event *event,
1554 unsigned long flags, int pc)
1556 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1558 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1560 struct ring_buffer_event *
1561 trace_event_buffer_lock_reserve(struct ring_buffer **current_rb,
1562 struct ftrace_event_file *ftrace_file,
1563 int type, unsigned long len,
1564 unsigned long flags, int pc)
1566 *current_rb = ftrace_file->tr->trace_buffer.buffer;
1567 return trace_buffer_lock_reserve(*current_rb,
1568 type, len, flags, pc);
1570 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
1572 struct ring_buffer_event *
1573 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1574 int type, unsigned long len,
1575 unsigned long flags, int pc)
1577 *current_rb = global_trace.trace_buffer.buffer;
1578 return trace_buffer_lock_reserve(*current_rb,
1579 type, len, flags, pc);
1581 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1583 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1584 struct ring_buffer_event *event,
1585 unsigned long flags, int pc)
1587 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1589 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1591 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1592 struct ring_buffer_event *event,
1593 unsigned long flags, int pc,
1594 struct pt_regs *regs)
1596 __buffer_unlock_commit(buffer, event);
1598 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1599 ftrace_trace_userstack(buffer, flags, pc);
1601 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1603 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1604 struct ring_buffer_event *event)
1606 ring_buffer_discard_commit(buffer, event);
1608 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1611 trace_function(struct trace_array *tr,
1612 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1615 struct ftrace_event_call *call = &event_function;
1616 struct ring_buffer *buffer = tr->trace_buffer.buffer;
1617 struct ring_buffer_event *event;
1618 struct ftrace_entry *entry;
1620 /* If we are reading the ring buffer, don't trace */
1621 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1624 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1628 entry = ring_buffer_event_data(event);
1630 entry->parent_ip = parent_ip;
1632 if (!filter_check_discard(call, entry, buffer, event))
1633 __buffer_unlock_commit(buffer, event);
1637 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1638 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1641 if (likely(!atomic_read(&data->disabled)))
1642 trace_function(tr, ip, parent_ip, flags, pc);
1645 #ifdef CONFIG_STACKTRACE
1647 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1648 struct ftrace_stack {
1649 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1652 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1653 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1655 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1656 unsigned long flags,
1657 int skip, int pc, struct pt_regs *regs)
1659 struct ftrace_event_call *call = &event_kernel_stack;
1660 struct ring_buffer_event *event;
1661 struct stack_entry *entry;
1662 struct stack_trace trace;
1664 int size = FTRACE_STACK_ENTRIES;
1666 trace.nr_entries = 0;
1670 * Since events can happen in NMIs there's no safe way to
1671 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1672 * or NMI comes in, it will just have to use the default
1673 * FTRACE_STACK_SIZE.
1675 preempt_disable_notrace();
1677 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1679 * We don't need any atomic variables, just a barrier.
1680 * If an interrupt comes in, we don't care, because it would
1681 * have exited and put the counter back to what we want.
1682 * We just need a barrier to keep gcc from moving things
1686 if (use_stack == 1) {
1687 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1688 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1691 save_stack_trace_regs(regs, &trace);
1693 save_stack_trace(&trace);
1695 if (trace.nr_entries > size)
1696 size = trace.nr_entries;
1698 /* From now on, use_stack is a boolean */
1701 size *= sizeof(unsigned long);
1703 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1704 sizeof(*entry) + size, flags, pc);
1707 entry = ring_buffer_event_data(event);
1709 memset(&entry->caller, 0, size);
1712 memcpy(&entry->caller, trace.entries,
1713 trace.nr_entries * sizeof(unsigned long));
1715 trace.max_entries = FTRACE_STACK_ENTRIES;
1716 trace.entries = entry->caller;
1718 save_stack_trace_regs(regs, &trace);
1720 save_stack_trace(&trace);
1723 entry->size = trace.nr_entries;
1725 if (!filter_check_discard(call, entry, buffer, event))
1726 __buffer_unlock_commit(buffer, event);
1729 /* Again, don't let gcc optimize things here */
1731 __this_cpu_dec(ftrace_stack_reserve);
1732 preempt_enable_notrace();
1736 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1737 int skip, int pc, struct pt_regs *regs)
1739 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1742 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1745 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1748 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1751 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1754 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1757 __ftrace_trace_stack(tr->trace_buffer.buffer, flags, skip, pc, NULL);
1761 * trace_dump_stack - record a stack back trace in the trace buffer
1762 * @skip: Number of functions to skip (helper handlers)
1764 void trace_dump_stack(int skip)
1766 unsigned long flags;
1768 if (tracing_disabled || tracing_selftest_running)
1771 local_save_flags(flags);
1774 * Skip 3 more, seems to get us at the caller of
1778 __ftrace_trace_stack(global_trace.trace_buffer.buffer,
1779 flags, skip, preempt_count(), NULL);
1782 static DEFINE_PER_CPU(int, user_stack_count);
1785 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1787 struct ftrace_event_call *call = &event_user_stack;
1788 struct ring_buffer_event *event;
1789 struct userstack_entry *entry;
1790 struct stack_trace trace;
1792 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1796 * NMIs can not handle page faults, even with fix ups.
1797 * The save user stack can (and often does) fault.
1799 if (unlikely(in_nmi()))
1803 * prevent recursion, since the user stack tracing may
1804 * trigger other kernel events.
1807 if (__this_cpu_read(user_stack_count))
1810 __this_cpu_inc(user_stack_count);
1812 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1813 sizeof(*entry), flags, pc);
1815 goto out_drop_count;
1816 entry = ring_buffer_event_data(event);
1818 entry->tgid = current->tgid;
1819 memset(&entry->caller, 0, sizeof(entry->caller));
1821 trace.nr_entries = 0;
1822 trace.max_entries = FTRACE_STACK_ENTRIES;
1824 trace.entries = entry->caller;
1826 save_stack_trace_user(&trace);
1827 if (!filter_check_discard(call, entry, buffer, event))
1828 __buffer_unlock_commit(buffer, event);
1831 __this_cpu_dec(user_stack_count);
1837 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1839 ftrace_trace_userstack(tr, flags, preempt_count());
1843 #endif /* CONFIG_STACKTRACE */
1845 /* created for use with alloc_percpu */
1846 struct trace_buffer_struct {
1847 char buffer[TRACE_BUF_SIZE];
1850 static struct trace_buffer_struct *trace_percpu_buffer;
1851 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1852 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1853 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1856 * The buffer used is dependent on the context. There is a per cpu
1857 * buffer for normal context, softirq contex, hard irq context and
1858 * for NMI context. Thise allows for lockless recording.
1860 * Note, if the buffers failed to be allocated, then this returns NULL
1862 static char *get_trace_buf(void)
1864 struct trace_buffer_struct *percpu_buffer;
1867 * If we have allocated per cpu buffers, then we do not
1868 * need to do any locking.
1871 percpu_buffer = trace_percpu_nmi_buffer;
1873 percpu_buffer = trace_percpu_irq_buffer;
1874 else if (in_softirq())
1875 percpu_buffer = trace_percpu_sirq_buffer;
1877 percpu_buffer = trace_percpu_buffer;
1882 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1885 static int alloc_percpu_trace_buffer(void)
1887 struct trace_buffer_struct *buffers;
1888 struct trace_buffer_struct *sirq_buffers;
1889 struct trace_buffer_struct *irq_buffers;
1890 struct trace_buffer_struct *nmi_buffers;
1892 buffers = alloc_percpu(struct trace_buffer_struct);
1896 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1900 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1904 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1908 trace_percpu_buffer = buffers;
1909 trace_percpu_sirq_buffer = sirq_buffers;
1910 trace_percpu_irq_buffer = irq_buffers;
1911 trace_percpu_nmi_buffer = nmi_buffers;
1916 free_percpu(irq_buffers);
1918 free_percpu(sirq_buffers);
1920 free_percpu(buffers);
1922 WARN(1, "Could not allocate percpu trace_printk buffer");
1926 static int buffers_allocated;
1928 void trace_printk_init_buffers(void)
1930 if (buffers_allocated)
1933 if (alloc_percpu_trace_buffer())
1936 pr_info("ftrace: Allocated trace_printk buffers\n");
1938 /* Expand the buffers to set size */
1939 tracing_update_buffers();
1941 buffers_allocated = 1;
1944 * trace_printk_init_buffers() can be called by modules.
1945 * If that happens, then we need to start cmdline recording
1946 * directly here. If the global_trace.buffer is already
1947 * allocated here, then this was called by module code.
1949 if (global_trace.trace_buffer.buffer)
1950 tracing_start_cmdline_record();
1953 void trace_printk_start_comm(void)
1955 /* Start tracing comms if trace printk is set */
1956 if (!buffers_allocated)
1958 tracing_start_cmdline_record();
1961 static void trace_printk_start_stop_comm(int enabled)
1963 if (!buffers_allocated)
1967 tracing_start_cmdline_record();
1969 tracing_stop_cmdline_record();
1973 * trace_vbprintk - write binary msg to tracing buffer
1976 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1978 struct ftrace_event_call *call = &event_bprint;
1979 struct ring_buffer_event *event;
1980 struct ring_buffer *buffer;
1981 struct trace_array *tr = &global_trace;
1982 struct bprint_entry *entry;
1983 unsigned long flags;
1985 int len = 0, size, pc;
1987 if (unlikely(tracing_selftest_running || tracing_disabled))
1990 /* Don't pollute graph traces with trace_vprintk internals */
1991 pause_graph_tracing();
1993 pc = preempt_count();
1994 preempt_disable_notrace();
1996 tbuffer = get_trace_buf();
2002 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
2004 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
2007 local_save_flags(flags);
2008 size = sizeof(*entry) + sizeof(u32) * len;
2009 buffer = tr->trace_buffer.buffer;
2010 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
2014 entry = ring_buffer_event_data(event);
2018 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
2019 if (!filter_check_discard(call, entry, buffer, event)) {
2020 __buffer_unlock_commit(buffer, event);
2021 ftrace_trace_stack(buffer, flags, 6, pc);
2025 preempt_enable_notrace();
2026 unpause_graph_tracing();
2030 EXPORT_SYMBOL_GPL(trace_vbprintk);
2033 __trace_array_vprintk(struct ring_buffer *buffer,
2034 unsigned long ip, const char *fmt, va_list args)
2036 struct ftrace_event_call *call = &event_print;
2037 struct ring_buffer_event *event;
2038 int len = 0, size, pc;
2039 struct print_entry *entry;
2040 unsigned long flags;
2043 if (tracing_disabled || tracing_selftest_running)
2046 /* Don't pollute graph traces with trace_vprintk internals */
2047 pause_graph_tracing();
2049 pc = preempt_count();
2050 preempt_disable_notrace();
2053 tbuffer = get_trace_buf();
2059 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
2060 if (len > TRACE_BUF_SIZE)
2063 local_save_flags(flags);
2064 size = sizeof(*entry) + len + 1;
2065 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
2069 entry = ring_buffer_event_data(event);
2072 memcpy(&entry->buf, tbuffer, len);
2073 entry->buf[len] = '\0';
2074 if (!filter_check_discard(call, entry, buffer, event)) {
2075 __buffer_unlock_commit(buffer, event);
2076 ftrace_trace_stack(buffer, flags, 6, pc);
2079 preempt_enable_notrace();
2080 unpause_graph_tracing();
2085 int trace_array_vprintk(struct trace_array *tr,
2086 unsigned long ip, const char *fmt, va_list args)
2088 return __trace_array_vprintk(tr->trace_buffer.buffer, ip, fmt, args);
2091 int trace_array_printk(struct trace_array *tr,
2092 unsigned long ip, const char *fmt, ...)
2097 if (!(trace_flags & TRACE_ITER_PRINTK))
2101 ret = trace_array_vprintk(tr, ip, fmt, ap);
2106 int trace_array_printk_buf(struct ring_buffer *buffer,
2107 unsigned long ip, const char *fmt, ...)
2112 if (!(trace_flags & TRACE_ITER_PRINTK))
2116 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
2121 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
2123 return trace_array_vprintk(&global_trace, ip, fmt, args);
2125 EXPORT_SYMBOL_GPL(trace_vprintk);
2127 static void trace_iterator_increment(struct trace_iterator *iter)
2129 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
2133 ring_buffer_read(buf_iter, NULL);
2136 static struct trace_entry *
2137 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
2138 unsigned long *lost_events)
2140 struct ring_buffer_event *event;
2141 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
2144 event = ring_buffer_iter_peek(buf_iter, ts);
2146 event = ring_buffer_peek(iter->trace_buffer->buffer, cpu, ts,
2150 iter->ent_size = ring_buffer_event_length(event);
2151 return ring_buffer_event_data(event);
2157 static struct trace_entry *
2158 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
2159 unsigned long *missing_events, u64 *ent_ts)
2161 struct ring_buffer *buffer = iter->trace_buffer->buffer;
2162 struct trace_entry *ent, *next = NULL;
2163 unsigned long lost_events = 0, next_lost = 0;
2164 int cpu_file = iter->cpu_file;
2165 u64 next_ts = 0, ts;
2171 * If we are in a per_cpu trace file, don't bother by iterating over
2172 * all cpu and peek directly.
2174 if (cpu_file > RING_BUFFER_ALL_CPUS) {
2175 if (ring_buffer_empty_cpu(buffer, cpu_file))
2177 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
2179 *ent_cpu = cpu_file;
2184 for_each_tracing_cpu(cpu) {
2186 if (ring_buffer_empty_cpu(buffer, cpu))
2189 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
2192 * Pick the entry with the smallest timestamp:
2194 if (ent && (!next || ts < next_ts)) {
2198 next_lost = lost_events;
2199 next_size = iter->ent_size;
2203 iter->ent_size = next_size;
2206 *ent_cpu = next_cpu;
2212 *missing_events = next_lost;
2217 /* Find the next real entry, without updating the iterator itself */
2218 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
2219 int *ent_cpu, u64 *ent_ts)
2221 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
2224 /* Find the next real entry, and increment the iterator to the next entry */
2225 void *trace_find_next_entry_inc(struct trace_iterator *iter)
2227 iter->ent = __find_next_entry(iter, &iter->cpu,
2228 &iter->lost_events, &iter->ts);
2231 trace_iterator_increment(iter);
2233 return iter->ent ? iter : NULL;
2236 static void trace_consume(struct trace_iterator *iter)
2238 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu, &iter->ts,
2239 &iter->lost_events);
2242 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
2244 struct trace_iterator *iter = m->private;
2248 WARN_ON_ONCE(iter->leftover);
2252 /* can't go backwards */
2257 ent = trace_find_next_entry_inc(iter);
2261 while (ent && iter->idx < i)
2262 ent = trace_find_next_entry_inc(iter);
2269 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
2271 struct ring_buffer_event *event;
2272 struct ring_buffer_iter *buf_iter;
2273 unsigned long entries = 0;
2276 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = 0;
2278 buf_iter = trace_buffer_iter(iter, cpu);
2282 ring_buffer_iter_reset(buf_iter);
2285 * We could have the case with the max latency tracers
2286 * that a reset never took place on a cpu. This is evident
2287 * by the timestamp being before the start of the buffer.
2289 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
2290 if (ts >= iter->trace_buffer->time_start)
2293 ring_buffer_read(buf_iter, NULL);
2296 per_cpu_ptr(iter->trace_buffer->data, cpu)->skipped_entries = entries;
2300 * The current tracer is copied to avoid a global locking
2303 static void *s_start(struct seq_file *m, loff_t *pos)
2305 struct trace_iterator *iter = m->private;
2306 struct trace_array *tr = iter->tr;
2307 int cpu_file = iter->cpu_file;
2313 * copy the tracer to avoid using a global lock all around.
2314 * iter->trace is a copy of current_trace, the pointer to the
2315 * name may be used instead of a strcmp(), as iter->trace->name
2316 * will point to the same string as current_trace->name.
2318 mutex_lock(&trace_types_lock);
2319 if (unlikely(tr->current_trace && iter->trace->name != tr->current_trace->name))
2320 *iter->trace = *tr->current_trace;
2321 mutex_unlock(&trace_types_lock);
2323 #ifdef CONFIG_TRACER_MAX_TRACE
2324 if (iter->snapshot && iter->trace->use_max_tr)
2325 return ERR_PTR(-EBUSY);
2328 if (!iter->snapshot)
2329 atomic_inc(&trace_record_cmdline_disabled);
2331 if (*pos != iter->pos) {
2336 if (cpu_file == RING_BUFFER_ALL_CPUS) {
2337 for_each_tracing_cpu(cpu)
2338 tracing_iter_reset(iter, cpu);
2340 tracing_iter_reset(iter, cpu_file);
2343 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
2348 * If we overflowed the seq_file before, then we want
2349 * to just reuse the trace_seq buffer again.
2355 p = s_next(m, p, &l);
2359 trace_event_read_lock();
2360 trace_access_lock(cpu_file);
2364 static void s_stop(struct seq_file *m, void *p)
2366 struct trace_iterator *iter = m->private;
2368 #ifdef CONFIG_TRACER_MAX_TRACE
2369 if (iter->snapshot && iter->trace->use_max_tr)
2373 if (!iter->snapshot)
2374 atomic_dec(&trace_record_cmdline_disabled);
2376 trace_access_unlock(iter->cpu_file);
2377 trace_event_read_unlock();
2381 get_total_entries(struct trace_buffer *buf,
2382 unsigned long *total, unsigned long *entries)
2384 unsigned long count;
2390 for_each_tracing_cpu(cpu) {
2391 count = ring_buffer_entries_cpu(buf->buffer, cpu);
2393 * If this buffer has skipped entries, then we hold all
2394 * entries for the trace and we need to ignore the
2395 * ones before the time stamp.
2397 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
2398 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
2399 /* total is the same as the entries */
2403 ring_buffer_overrun_cpu(buf->buffer, cpu);
2408 static void print_lat_help_header(struct seq_file *m)
2410 seq_puts(m, "# _------=> CPU# \n");
2411 seq_puts(m, "# / _-----=> irqs-off \n");
2412 seq_puts(m, "# | / _----=> need-resched \n");
2413 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2414 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2415 seq_puts(m, "# |||| / delay \n");
2416 seq_puts(m, "# cmd pid ||||| time | caller \n");
2417 seq_puts(m, "# \\ / ||||| \\ | / \n");
2420 static void print_event_info(struct trace_buffer *buf, struct seq_file *m)
2422 unsigned long total;
2423 unsigned long entries;
2425 get_total_entries(buf, &total, &entries);
2426 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2427 entries, total, num_online_cpus());
2431 static void print_func_help_header(struct trace_buffer *buf, struct seq_file *m)
2433 print_event_info(buf, m);
2434 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2435 seq_puts(m, "# | | | | |\n");
2438 static void print_func_help_header_irq(struct trace_buffer *buf, struct seq_file *m)
2440 print_event_info(buf, m);
2441 seq_puts(m, "# _-----=> irqs-off\n");
2442 seq_puts(m, "# / _----=> need-resched\n");
2443 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2444 seq_puts(m, "# || / _--=> preempt-depth\n");
2445 seq_puts(m, "# ||| / delay\n");
2446 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2447 seq_puts(m, "# | | | |||| | |\n");
2451 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2453 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2454 struct trace_buffer *buf = iter->trace_buffer;
2455 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
2456 struct tracer *type = iter->trace;
2457 unsigned long entries;
2458 unsigned long total;
2459 const char *name = "preemption";
2463 get_total_entries(buf, &total, &entries);
2465 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2467 seq_puts(m, "# -----------------------------------"
2468 "---------------------------------\n");
2469 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2470 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2471 nsecs_to_usecs(data->saved_latency),
2475 #if defined(CONFIG_PREEMPT_NONE)
2477 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2479 #elif defined(CONFIG_PREEMPT)
2484 /* These are reserved for later use */
2487 seq_printf(m, " #P:%d)\n", num_online_cpus());
2491 seq_puts(m, "# -----------------\n");
2492 seq_printf(m, "# | task: %.16s-%d "
2493 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2494 data->comm, data->pid,
2495 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2496 data->policy, data->rt_priority);
2497 seq_puts(m, "# -----------------\n");
2499 if (data->critical_start) {
2500 seq_puts(m, "# => started at: ");
2501 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2502 trace_print_seq(m, &iter->seq);
2503 seq_puts(m, "\n# => ended at: ");
2504 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2505 trace_print_seq(m, &iter->seq);
2506 seq_puts(m, "\n#\n");
2512 static void test_cpu_buff_start(struct trace_iterator *iter)
2514 struct trace_seq *s = &iter->seq;
2516 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2519 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2522 if (cpumask_test_cpu(iter->cpu, iter->started))
2525 if (per_cpu_ptr(iter->trace_buffer->data, iter->cpu)->skipped_entries)
2528 cpumask_set_cpu(iter->cpu, iter->started);
2530 /* Don't print started cpu buffer for the first entry of the trace */
2532 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2536 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2538 struct trace_seq *s = &iter->seq;
2539 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2540 struct trace_entry *entry;
2541 struct trace_event *event;
2545 test_cpu_buff_start(iter);
2547 event = ftrace_find_event(entry->type);
2549 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2550 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2551 if (!trace_print_lat_context(iter))
2554 if (!trace_print_context(iter))
2560 return event->funcs->trace(iter, sym_flags, event);
2562 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2565 return TRACE_TYPE_HANDLED;
2567 return TRACE_TYPE_PARTIAL_LINE;
2570 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2572 struct trace_seq *s = &iter->seq;
2573 struct trace_entry *entry;
2574 struct trace_event *event;
2578 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2579 if (!trace_seq_printf(s, "%d %d %llu ",
2580 entry->pid, iter->cpu, iter->ts))
2584 event = ftrace_find_event(entry->type);
2586 return event->funcs->raw(iter, 0, event);
2588 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2591 return TRACE_TYPE_HANDLED;
2593 return TRACE_TYPE_PARTIAL_LINE;
2596 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2598 struct trace_seq *s = &iter->seq;
2599 unsigned char newline = '\n';
2600 struct trace_entry *entry;
2601 struct trace_event *event;
2605 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2606 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2607 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2608 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2611 event = ftrace_find_event(entry->type);
2613 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2614 if (ret != TRACE_TYPE_HANDLED)
2618 SEQ_PUT_FIELD_RET(s, newline);
2620 return TRACE_TYPE_HANDLED;
2623 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2625 struct trace_seq *s = &iter->seq;
2626 struct trace_entry *entry;
2627 struct trace_event *event;
2631 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2632 SEQ_PUT_FIELD_RET(s, entry->pid);
2633 SEQ_PUT_FIELD_RET(s, iter->cpu);
2634 SEQ_PUT_FIELD_RET(s, iter->ts);
2637 event = ftrace_find_event(entry->type);
2638 return event ? event->funcs->binary(iter, 0, event) :
2642 int trace_empty(struct trace_iterator *iter)
2644 struct ring_buffer_iter *buf_iter;
2647 /* If we are looking at one CPU buffer, only check that one */
2648 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
2649 cpu = iter->cpu_file;
2650 buf_iter = trace_buffer_iter(iter, cpu);
2652 if (!ring_buffer_iter_empty(buf_iter))
2655 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2661 for_each_tracing_cpu(cpu) {
2662 buf_iter = trace_buffer_iter(iter, cpu);
2664 if (!ring_buffer_iter_empty(buf_iter))
2667 if (!ring_buffer_empty_cpu(iter->trace_buffer->buffer, cpu))
2675 /* Called with trace_event_read_lock() held. */
2676 enum print_line_t print_trace_line(struct trace_iterator *iter)
2678 enum print_line_t ret;
2680 if (iter->lost_events &&
2681 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2682 iter->cpu, iter->lost_events))
2683 return TRACE_TYPE_PARTIAL_LINE;
2685 if (iter->trace && iter->trace->print_line) {
2686 ret = iter->trace->print_line(iter);
2687 if (ret != TRACE_TYPE_UNHANDLED)
2691 if (iter->ent->type == TRACE_BPUTS &&
2692 trace_flags & TRACE_ITER_PRINTK &&
2693 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2694 return trace_print_bputs_msg_only(iter);
2696 if (iter->ent->type == TRACE_BPRINT &&
2697 trace_flags & TRACE_ITER_PRINTK &&
2698 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2699 return trace_print_bprintk_msg_only(iter);
2701 if (iter->ent->type == TRACE_PRINT &&
2702 trace_flags & TRACE_ITER_PRINTK &&
2703 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2704 return trace_print_printk_msg_only(iter);
2706 if (trace_flags & TRACE_ITER_BIN)
2707 return print_bin_fmt(iter);
2709 if (trace_flags & TRACE_ITER_HEX)
2710 return print_hex_fmt(iter);
2712 if (trace_flags & TRACE_ITER_RAW)
2713 return print_raw_fmt(iter);
2715 return print_trace_fmt(iter);
2718 void trace_latency_header(struct seq_file *m)
2720 struct trace_iterator *iter = m->private;
2722 /* print nothing if the buffers are empty */
2723 if (trace_empty(iter))
2726 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2727 print_trace_header(m, iter);
2729 if (!(trace_flags & TRACE_ITER_VERBOSE))
2730 print_lat_help_header(m);
2733 void trace_default_header(struct seq_file *m)
2735 struct trace_iterator *iter = m->private;
2737 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2740 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2741 /* print nothing if the buffers are empty */
2742 if (trace_empty(iter))
2744 print_trace_header(m, iter);
2745 if (!(trace_flags & TRACE_ITER_VERBOSE))
2746 print_lat_help_header(m);
2748 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2749 if (trace_flags & TRACE_ITER_IRQ_INFO)
2750 print_func_help_header_irq(iter->trace_buffer, m);
2752 print_func_help_header(iter->trace_buffer, m);
2757 static void test_ftrace_alive(struct seq_file *m)
2759 if (!ftrace_is_dead())
2761 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2762 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2765 #ifdef CONFIG_TRACER_MAX_TRACE
2766 static void show_snapshot_main_help(struct seq_file *m)
2768 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2769 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2770 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2771 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate)\n");
2772 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2773 seq_printf(m, "# is not a '0' or '1')\n");
2776 static void show_snapshot_percpu_help(struct seq_file *m)
2778 seq_printf(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2779 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2780 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2781 seq_printf(m, "# Takes a snapshot of the main buffer for this cpu.\n");
2783 seq_printf(m, "# echo 1 > snapshot : Not supported with this kernel.\n");
2784 seq_printf(m, "# Must use main snapshot file to allocate.\n");
2786 seq_printf(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2787 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2788 seq_printf(m, "# is not a '0' or '1')\n");
2791 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2793 if (iter->tr->allocated_snapshot)
2794 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2796 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2798 seq_printf(m, "# Snapshot commands:\n");
2799 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
2800 show_snapshot_main_help(m);
2802 show_snapshot_percpu_help(m);
2805 /* Should never be called */
2806 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2809 static int s_show(struct seq_file *m, void *v)
2811 struct trace_iterator *iter = v;
2814 if (iter->ent == NULL) {
2816 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2818 test_ftrace_alive(m);
2820 if (iter->snapshot && trace_empty(iter))
2821 print_snapshot_help(m, iter);
2822 else if (iter->trace && iter->trace->print_header)
2823 iter->trace->print_header(m);
2825 trace_default_header(m);
2827 } else if (iter->leftover) {
2829 * If we filled the seq_file buffer earlier, we
2830 * want to just show it now.
2832 ret = trace_print_seq(m, &iter->seq);
2834 /* ret should this time be zero, but you never know */
2835 iter->leftover = ret;
2838 print_trace_line(iter);
2839 ret = trace_print_seq(m, &iter->seq);
2841 * If we overflow the seq_file buffer, then it will
2842 * ask us for this data again at start up.
2844 * ret is 0 if seq_file write succeeded.
2847 iter->leftover = ret;
2854 * Should be used after trace_array_get(), trace_types_lock
2855 * ensures that i_cdev was already initialized.
2857 static inline int tracing_get_cpu(struct inode *inode)
2859 if (inode->i_cdev) /* See trace_create_cpu_file() */
2860 return (long)inode->i_cdev - 1;
2861 return RING_BUFFER_ALL_CPUS;
2864 static const struct seq_operations tracer_seq_ops = {
2871 static struct trace_iterator *
2872 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
2874 struct trace_array *tr = inode->i_private;
2875 struct trace_iterator *iter;
2878 if (tracing_disabled)
2879 return ERR_PTR(-ENODEV);
2881 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2883 return ERR_PTR(-ENOMEM);
2885 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2887 if (!iter->buffer_iter)
2891 * We make a copy of the current tracer to avoid concurrent
2892 * changes on it while we are reading.
2894 mutex_lock(&trace_types_lock);
2895 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2899 *iter->trace = *tr->current_trace;
2901 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2906 #ifdef CONFIG_TRACER_MAX_TRACE
2907 /* Currently only the top directory has a snapshot */
2908 if (tr->current_trace->print_max || snapshot)
2909 iter->trace_buffer = &tr->max_buffer;
2912 iter->trace_buffer = &tr->trace_buffer;
2913 iter->snapshot = snapshot;
2915 iter->cpu_file = tracing_get_cpu(inode);
2916 mutex_init(&iter->mutex);
2918 /* Notify the tracer early; before we stop tracing. */
2919 if (iter->trace && iter->trace->open)
2920 iter->trace->open(iter);
2922 /* Annotate start of buffers if we had overruns */
2923 if (ring_buffer_overruns(iter->trace_buffer->buffer))
2924 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2926 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2927 if (trace_clocks[tr->clock_id].in_ns)
2928 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2930 /* stop the trace while dumping if we are not opening "snapshot" */
2931 if (!iter->snapshot)
2932 tracing_stop_tr(tr);
2934 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
2935 for_each_tracing_cpu(cpu) {
2936 iter->buffer_iter[cpu] =
2937 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2939 ring_buffer_read_prepare_sync();
2940 for_each_tracing_cpu(cpu) {
2941 ring_buffer_read_start(iter->buffer_iter[cpu]);
2942 tracing_iter_reset(iter, cpu);
2945 cpu = iter->cpu_file;
2946 iter->buffer_iter[cpu] =
2947 ring_buffer_read_prepare(iter->trace_buffer->buffer, cpu);
2948 ring_buffer_read_prepare_sync();
2949 ring_buffer_read_start(iter->buffer_iter[cpu]);
2950 tracing_iter_reset(iter, cpu);
2953 mutex_unlock(&trace_types_lock);
2958 mutex_unlock(&trace_types_lock);
2960 kfree(iter->buffer_iter);
2962 seq_release_private(inode, file);
2963 return ERR_PTR(-ENOMEM);
2966 int tracing_open_generic(struct inode *inode, struct file *filp)
2968 if (tracing_disabled)
2971 filp->private_data = inode->i_private;
2976 * Open and update trace_array ref count.
2977 * Must have the current trace_array passed to it.
2979 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
2981 struct trace_array *tr = inode->i_private;
2983 if (tracing_disabled)
2986 if (trace_array_get(tr) < 0)
2989 filp->private_data = inode->i_private;
2994 static int tracing_release(struct inode *inode, struct file *file)
2996 struct trace_array *tr = inode->i_private;
2997 struct seq_file *m = file->private_data;
2998 struct trace_iterator *iter;
3001 if (!(file->f_mode & FMODE_READ)) {
3002 trace_array_put(tr);
3006 /* Writes do not use seq_file */
3008 mutex_lock(&trace_types_lock);
3010 for_each_tracing_cpu(cpu) {
3011 if (iter->buffer_iter[cpu])
3012 ring_buffer_read_finish(iter->buffer_iter[cpu]);
3015 if (iter->trace && iter->trace->close)
3016 iter->trace->close(iter);
3018 if (!iter->snapshot)
3019 /* reenable tracing if it was previously enabled */
3020 tracing_start_tr(tr);
3022 __trace_array_put(tr);
3024 mutex_unlock(&trace_types_lock);
3026 mutex_destroy(&iter->mutex);
3027 free_cpumask_var(iter->started);
3029 kfree(iter->buffer_iter);
3030 seq_release_private(inode, file);
3035 static int tracing_release_generic_tr(struct inode *inode, struct file *file)
3037 struct trace_array *tr = inode->i_private;
3039 trace_array_put(tr);
3043 static int tracing_single_release_tr(struct inode *inode, struct file *file)
3045 struct trace_array *tr = inode->i_private;
3047 trace_array_put(tr);
3049 return single_release(inode, file);
3052 static int tracing_open(struct inode *inode, struct file *file)
3054 struct trace_array *tr = inode->i_private;
3055 struct trace_iterator *iter;
3058 if (trace_array_get(tr) < 0)
3061 /* If this file was open for write, then erase contents */
3062 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
3063 int cpu = tracing_get_cpu(inode);
3065 if (cpu == RING_BUFFER_ALL_CPUS)
3066 tracing_reset_online_cpus(&tr->trace_buffer);
3068 tracing_reset(&tr->trace_buffer, cpu);
3071 if (file->f_mode & FMODE_READ) {
3072 iter = __tracing_open(inode, file, false);
3074 ret = PTR_ERR(iter);
3075 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
3076 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3080 trace_array_put(tr);
3086 t_next(struct seq_file *m, void *v, loff_t *pos)
3088 struct tracer *t = v;
3098 static void *t_start(struct seq_file *m, loff_t *pos)
3103 mutex_lock(&trace_types_lock);
3104 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
3110 static void t_stop(struct seq_file *m, void *p)
3112 mutex_unlock(&trace_types_lock);
3115 static int t_show(struct seq_file *m, void *v)
3117 struct tracer *t = v;
3122 seq_printf(m, "%s", t->name);
3131 static const struct seq_operations show_traces_seq_ops = {
3138 static int show_traces_open(struct inode *inode, struct file *file)
3140 if (tracing_disabled)
3143 return seq_open(file, &show_traces_seq_ops);
3147 tracing_write_stub(struct file *filp, const char __user *ubuf,
3148 size_t count, loff_t *ppos)
3153 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
3155 if (file->f_mode & FMODE_READ)
3156 return seq_lseek(file, offset, origin);
3161 static const struct file_operations tracing_fops = {
3162 .open = tracing_open,
3164 .write = tracing_write_stub,
3165 .llseek = tracing_seek,
3166 .release = tracing_release,
3169 static const struct file_operations show_traces_fops = {
3170 .open = show_traces_open,
3172 .release = seq_release,
3173 .llseek = seq_lseek,
3177 * Only trace on a CPU if the bitmask is set:
3179 static cpumask_var_t tracing_cpumask;
3182 * The tracer itself will not take this lock, but still we want
3183 * to provide a consistent cpumask to user-space:
3185 static DEFINE_MUTEX(tracing_cpumask_update_lock);
3188 * Temporary storage for the character representation of the
3189 * CPU bitmask (and one more byte for the newline):
3191 static char mask_str[NR_CPUS + 1];
3194 tracing_cpumask_read(struct file *filp, char __user *ubuf,
3195 size_t count, loff_t *ppos)
3199 mutex_lock(&tracing_cpumask_update_lock);
3201 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
3202 if (count - len < 2) {
3206 len += sprintf(mask_str + len, "\n");
3207 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
3210 mutex_unlock(&tracing_cpumask_update_lock);
3216 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
3217 size_t count, loff_t *ppos)
3219 struct trace_array *tr = filp->private_data;
3220 cpumask_var_t tracing_cpumask_new;
3223 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
3226 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
3230 mutex_lock(&tracing_cpumask_update_lock);
3232 local_irq_disable();
3233 arch_spin_lock(&ftrace_max_lock);
3234 for_each_tracing_cpu(cpu) {
3236 * Increase/decrease the disabled counter if we are
3237 * about to flip a bit in the cpumask:
3239 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
3240 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3241 atomic_inc(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3242 ring_buffer_record_disable_cpu(tr->trace_buffer.buffer, cpu);
3244 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
3245 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
3246 atomic_dec(&per_cpu_ptr(tr->trace_buffer.data, cpu)->disabled);
3247 ring_buffer_record_enable_cpu(tr->trace_buffer.buffer, cpu);
3250 arch_spin_unlock(&ftrace_max_lock);
3253 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
3255 mutex_unlock(&tracing_cpumask_update_lock);
3256 free_cpumask_var(tracing_cpumask_new);
3261 free_cpumask_var(tracing_cpumask_new);
3266 static const struct file_operations tracing_cpumask_fops = {
3267 .open = tracing_open_generic,
3268 .read = tracing_cpumask_read,
3269 .write = tracing_cpumask_write,
3270 .llseek = generic_file_llseek,
3273 static int tracing_trace_options_show(struct seq_file *m, void *v)
3275 struct tracer_opt *trace_opts;
3276 struct trace_array *tr = m->private;
3280 mutex_lock(&trace_types_lock);
3281 tracer_flags = tr->current_trace->flags->val;
3282 trace_opts = tr->current_trace->flags->opts;
3284 for (i = 0; trace_options[i]; i++) {
3285 if (trace_flags & (1 << i))
3286 seq_printf(m, "%s\n", trace_options[i]);
3288 seq_printf(m, "no%s\n", trace_options[i]);
3291 for (i = 0; trace_opts[i].name; i++) {
3292 if (tracer_flags & trace_opts[i].bit)
3293 seq_printf(m, "%s\n", trace_opts[i].name);
3295 seq_printf(m, "no%s\n", trace_opts[i].name);
3297 mutex_unlock(&trace_types_lock);
3302 static int __set_tracer_option(struct tracer *trace,
3303 struct tracer_flags *tracer_flags,
3304 struct tracer_opt *opts, int neg)
3308 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
3313 tracer_flags->val &= ~opts->bit;
3315 tracer_flags->val |= opts->bit;
3319 /* Try to assign a tracer specific option */
3320 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
3322 struct tracer_flags *tracer_flags = trace->flags;
3323 struct tracer_opt *opts = NULL;
3326 for (i = 0; tracer_flags->opts[i].name; i++) {
3327 opts = &tracer_flags->opts[i];
3329 if (strcmp(cmp, opts->name) == 0)
3330 return __set_tracer_option(trace, trace->flags,
3337 /* Some tracers require overwrite to stay enabled */
3338 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
3340 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
3346 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
3348 /* do nothing if flag is already set */
3349 if (!!(trace_flags & mask) == !!enabled)
3352 /* Give the tracer a chance to approve the change */
3353 if (tr->current_trace->flag_changed)
3354 if (tr->current_trace->flag_changed(tr->current_trace, mask, !!enabled))
3358 trace_flags |= mask;
3360 trace_flags &= ~mask;
3362 if (mask == TRACE_ITER_RECORD_CMD)
3363 trace_event_enable_cmd_record(enabled);
3365 if (mask == TRACE_ITER_OVERWRITE) {
3366 ring_buffer_change_overwrite(tr->trace_buffer.buffer, enabled);
3367 #ifdef CONFIG_TRACER_MAX_TRACE
3368 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
3372 if (mask == TRACE_ITER_PRINTK)
3373 trace_printk_start_stop_comm(enabled);
3378 static int trace_set_options(struct trace_array *tr, char *option)
3385 cmp = strstrip(option);
3387 if (strncmp(cmp, "no", 2) == 0) {
3392 mutex_lock(&trace_types_lock);
3394 for (i = 0; trace_options[i]; i++) {
3395 if (strcmp(cmp, trace_options[i]) == 0) {
3396 ret = set_tracer_flag(tr, 1 << i, !neg);
3401 /* If no option could be set, test the specific tracer options */
3402 if (!trace_options[i])
3403 ret = set_tracer_option(tr->current_trace, cmp, neg);
3405 mutex_unlock(&trace_types_lock);
3411 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
3412 size_t cnt, loff_t *ppos)
3414 struct seq_file *m = filp->private_data;
3415 struct trace_array *tr = m->private;
3419 if (cnt >= sizeof(buf))
3422 if (copy_from_user(&buf, ubuf, cnt))
3427 ret = trace_set_options(tr, buf);
3436 static int tracing_trace_options_open(struct inode *inode, struct file *file)
3438 struct trace_array *tr = inode->i_private;
3441 if (tracing_disabled)
3444 if (trace_array_get(tr) < 0)
3447 ret = single_open(file, tracing_trace_options_show, inode->i_private);
3449 trace_array_put(tr);
3454 static const struct file_operations tracing_iter_fops = {
3455 .open = tracing_trace_options_open,
3457 .llseek = seq_lseek,
3458 .release = tracing_single_release_tr,
3459 .write = tracing_trace_options_write,
3462 static const char readme_msg[] =
3463 "tracing mini-HOWTO:\n\n"
3464 "# echo 0 > tracing_on : quick way to disable tracing\n"
3465 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3466 " Important files:\n"
3467 " trace\t\t\t- The static contents of the buffer\n"
3468 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3469 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3470 " current_tracer\t- function and latency tracers\n"
3471 " available_tracers\t- list of configured tracers for current_tracer\n"
3472 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3473 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3474 " trace_clock\t\t-change the clock used to order events\n"
3475 " local: Per cpu clock but may not be synced across CPUs\n"
3476 " global: Synced across CPUs but slows tracing down.\n"
3477 " counter: Not a clock, but just an increment\n"
3478 " uptime: Jiffy counter from time of boot\n"
3479 " perf: Same clock that perf events use\n"
3480 #ifdef CONFIG_X86_64
3481 " x86-tsc: TSC cycle counter\n"
3483 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3484 " tracing_cpumask\t- Limit which CPUs to trace\n"
3485 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3486 "\t\t\t Remove sub-buffer with rmdir\n"
3487 " trace_options\t\t- Set format or modify how tracing happens\n"
3488 "\t\t\t Disable an option by adding a suffix 'no' to the option name\n"
3489 #ifdef CONFIG_DYNAMIC_FTRACE
3490 "\n available_filter_functions - list of functions that can be filtered on\n"
3491 " set_ftrace_filter\t- echo function name in here to only trace these functions\n"
3492 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3493 " modules: Can select a group via module\n"
3494 " Format: :mod:<module-name>\n"
3495 " example: echo :mod:ext3 > set_ftrace_filter\n"
3496 " triggers: a command to perform when function is hit\n"
3497 " Format: <function>:<trigger>[:count]\n"
3498 " trigger: traceon, traceoff\n"
3499 " enable_event:<system>:<event>\n"
3500 " disable_event:<system>:<event>\n"
3501 #ifdef CONFIG_STACKTRACE
3504 #ifdef CONFIG_TRACER_SNAPSHOT
3507 " example: echo do_fault:traceoff > set_ftrace_filter\n"
3508 " echo do_trap:traceoff:3 > set_ftrace_filter\n"
3509 " The first one will disable tracing every time do_fault is hit\n"
3510 " The second will disable tracing at most 3 times when do_trap is hit\n"
3511 " The first time do trap is hit and it disables tracing, the counter\n"
3512 " will decrement to 2. If tracing is already disabled, the counter\n"
3513 " will not decrement. It only decrements when the trigger did work\n"
3514 " To remove trigger without count:\n"
3515 " echo '!<function>:<trigger> > set_ftrace_filter\n"
3516 " To remove trigger with a count:\n"
3517 " echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3518 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3519 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3520 " modules: Can select a group via module command :mod:\n"
3521 " Does not accept triggers\n"
3522 #endif /* CONFIG_DYNAMIC_FTRACE */
3523 #ifdef CONFIG_FUNCTION_TRACER
3524 " set_ftrace_pid\t- Write pid(s) to only function trace those pids (function)\n"
3526 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3527 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3528 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3530 #ifdef CONFIG_TRACER_SNAPSHOT
3531 "\n snapshot\t\t- Like 'trace' but shows the content of the static snapshot buffer\n"
3532 "\t\t\t Read the contents for more information\n"
3534 #ifdef CONFIG_STACKTRACE
3535 " stack_trace\t\t- Shows the max stack trace when active\n"
3536 " stack_max_size\t- Shows current max stack size that was traced\n"
3537 "\t\t\t Write into this file to reset the max size (trigger a new trace)\n"
3538 #ifdef CONFIG_DYNAMIC_FTRACE
3539 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace traces\n"
3541 #endif /* CONFIG_STACKTRACE */
3545 tracing_readme_read(struct file *filp, char __user *ubuf,
3546 size_t cnt, loff_t *ppos)
3548 return simple_read_from_buffer(ubuf, cnt, ppos,
3549 readme_msg, strlen(readme_msg));
3552 static const struct file_operations tracing_readme_fops = {
3553 .open = tracing_open_generic,
3554 .read = tracing_readme_read,
3555 .llseek = generic_file_llseek,
3559 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3560 size_t cnt, loff_t *ppos)
3569 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3573 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3581 for (i = 0; i < SAVED_CMDLINES; i++) {
3584 pid = map_cmdline_to_pid[i];
3585 if (pid == -1 || pid == NO_CMDLINE_MAP)
3588 trace_find_cmdline(pid, buf_comm);
3589 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3594 len = simple_read_from_buffer(ubuf, cnt, ppos,
3603 static const struct file_operations tracing_saved_cmdlines_fops = {
3604 .open = tracing_open_generic,
3605 .read = tracing_saved_cmdlines_read,
3606 .llseek = generic_file_llseek,
3610 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3611 size_t cnt, loff_t *ppos)
3613 struct trace_array *tr = filp->private_data;
3614 char buf[MAX_TRACER_SIZE+2];
3617 mutex_lock(&trace_types_lock);
3618 r = sprintf(buf, "%s\n", tr->current_trace->name);
3619 mutex_unlock(&trace_types_lock);
3621 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3624 int tracer_init(struct tracer *t, struct trace_array *tr)
3626 tracing_reset_online_cpus(&tr->trace_buffer);
3630 static void set_buffer_entries(struct trace_buffer *buf, unsigned long val)
3634 for_each_tracing_cpu(cpu)
3635 per_cpu_ptr(buf->data, cpu)->entries = val;
3638 #ifdef CONFIG_TRACER_MAX_TRACE
3639 /* resize @tr's buffer to the size of @size_tr's entries */
3640 static int resize_buffer_duplicate_size(struct trace_buffer *trace_buf,
3641 struct trace_buffer *size_buf, int cpu_id)
3645 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3646 for_each_tracing_cpu(cpu) {
3647 ret = ring_buffer_resize(trace_buf->buffer,
3648 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
3651 per_cpu_ptr(trace_buf->data, cpu)->entries =
3652 per_cpu_ptr(size_buf->data, cpu)->entries;
3655 ret = ring_buffer_resize(trace_buf->buffer,
3656 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
3658 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
3659 per_cpu_ptr(size_buf->data, cpu_id)->entries;
3664 #endif /* CONFIG_TRACER_MAX_TRACE */
3666 static int __tracing_resize_ring_buffer(struct trace_array *tr,
3667 unsigned long size, int cpu)
3672 * If kernel or user changes the size of the ring buffer
3673 * we use the size that was given, and we can forget about
3674 * expanding it later.
3676 ring_buffer_expanded = true;
3678 /* May be called before buffers are initialized */
3679 if (!tr->trace_buffer.buffer)
3682 ret = ring_buffer_resize(tr->trace_buffer.buffer, size, cpu);
3686 #ifdef CONFIG_TRACER_MAX_TRACE
3687 if (!(tr->flags & TRACE_ARRAY_FL_GLOBAL) ||
3688 !tr->current_trace->use_max_tr)
3691 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
3693 int r = resize_buffer_duplicate_size(&tr->trace_buffer,
3694 &tr->trace_buffer, cpu);
3697 * AARGH! We are left with different
3698 * size max buffer!!!!
3699 * The max buffer is our "snapshot" buffer.
3700 * When a tracer needs a snapshot (one of the
3701 * latency tracers), it swaps the max buffer
3702 * with the saved snap shot. We succeeded to
3703 * update the size of the main buffer, but failed to
3704 * update the size of the max buffer. But when we tried
3705 * to reset the main buffer to the original size, we
3706 * failed there too. This is very unlikely to
3707 * happen, but if it does, warn and kill all
3711 tracing_disabled = 1;
3716 if (cpu == RING_BUFFER_ALL_CPUS)
3717 set_buffer_entries(&tr->max_buffer, size);
3719 per_cpu_ptr(tr->max_buffer.data, cpu)->entries = size;
3722 #endif /* CONFIG_TRACER_MAX_TRACE */
3724 if (cpu == RING_BUFFER_ALL_CPUS)
3725 set_buffer_entries(&tr->trace_buffer, size);
3727 per_cpu_ptr(tr->trace_buffer.data, cpu)->entries = size;
3732 static ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
3733 unsigned long size, int cpu_id)
3737 mutex_lock(&trace_types_lock);
3739 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3740 /* make sure, this cpu is enabled in the mask */
3741 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3747 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
3752 mutex_unlock(&trace_types_lock);
3759 * tracing_update_buffers - used by tracing facility to expand ring buffers
3761 * To save on memory when the tracing is never used on a system with it
3762 * configured in. The ring buffers are set to a minimum size. But once
3763 * a user starts to use the tracing facility, then they need to grow
3764 * to their default size.
3766 * This function is to be called when a tracer is about to be used.
3768 int tracing_update_buffers(void)
3772 mutex_lock(&trace_types_lock);
3773 if (!ring_buffer_expanded)
3774 ret = __tracing_resize_ring_buffer(&global_trace, trace_buf_size,
3775 RING_BUFFER_ALL_CPUS);
3776 mutex_unlock(&trace_types_lock);
3781 struct trace_option_dentry;
3783 static struct trace_option_dentry *
3784 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
3787 destroy_trace_option_files(struct trace_option_dentry *topts);
3789 static int tracing_set_tracer(const char *buf)
3791 static struct trace_option_dentry *topts;
3792 struct trace_array *tr = &global_trace;
3794 #ifdef CONFIG_TRACER_MAX_TRACE
3799 mutex_lock(&trace_types_lock);
3801 if (!ring_buffer_expanded) {
3802 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
3803 RING_BUFFER_ALL_CPUS);
3809 for (t = trace_types; t; t = t->next) {
3810 if (strcmp(t->name, buf) == 0)
3817 if (t == tr->current_trace)
3820 trace_branch_disable();
3822 tr->current_trace->enabled = false;
3824 if (tr->current_trace->reset)
3825 tr->current_trace->reset(tr);
3827 /* Current trace needs to be nop_trace before synchronize_sched */
3828 tr->current_trace = &nop_trace;
3830 #ifdef CONFIG_TRACER_MAX_TRACE
3831 had_max_tr = tr->allocated_snapshot;
3833 if (had_max_tr && !t->use_max_tr) {
3835 * We need to make sure that the update_max_tr sees that
3836 * current_trace changed to nop_trace to keep it from
3837 * swapping the buffers after we resize it.
3838 * The update_max_tr is called from interrupts disabled
3839 * so a synchronized_sched() is sufficient.
3841 synchronize_sched();
3845 destroy_trace_option_files(topts);
3847 topts = create_trace_option_files(tr, t);
3849 #ifdef CONFIG_TRACER_MAX_TRACE
3850 if (t->use_max_tr && !had_max_tr) {
3851 ret = alloc_snapshot(tr);
3858 ret = tracer_init(t, tr);
3863 tr->current_trace = t;
3864 tr->current_trace->enabled = true;
3865 trace_branch_enable(tr);
3867 mutex_unlock(&trace_types_lock);
3873 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3874 size_t cnt, loff_t *ppos)
3876 char buf[MAX_TRACER_SIZE+1];
3883 if (cnt > MAX_TRACER_SIZE)
3884 cnt = MAX_TRACER_SIZE;
3886 if (copy_from_user(&buf, ubuf, cnt))
3891 /* strip ending whitespace. */
3892 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3895 err = tracing_set_tracer(buf);
3905 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3906 size_t cnt, loff_t *ppos)
3908 unsigned long *ptr = filp->private_data;
3912 r = snprintf(buf, sizeof(buf), "%ld\n",
3913 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3914 if (r > sizeof(buf))
3916 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3920 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3921 size_t cnt, loff_t *ppos)
3923 unsigned long *ptr = filp->private_data;
3927 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3936 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3938 struct trace_array *tr = inode->i_private;
3939 struct trace_iterator *iter;
3942 if (tracing_disabled)
3945 if (trace_array_get(tr) < 0)
3948 mutex_lock(&trace_types_lock);
3950 /* create a buffer to store the information to pass to userspace */
3951 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3954 __trace_array_put(tr);
3959 * We make a copy of the current tracer to avoid concurrent
3960 * changes on it while we are reading.
3962 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3967 *iter->trace = *tr->current_trace;
3969 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3974 /* trace pipe does not show start of buffer */
3975 cpumask_setall(iter->started);
3977 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3978 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3980 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3981 if (trace_clocks[tr->clock_id].in_ns)
3982 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3985 iter->trace_buffer = &tr->trace_buffer;
3986 iter->cpu_file = tracing_get_cpu(inode);
3987 mutex_init(&iter->mutex);
3988 filp->private_data = iter;
3990 if (iter->trace->pipe_open)
3991 iter->trace->pipe_open(iter);
3993 nonseekable_open(inode, filp);
3995 mutex_unlock(&trace_types_lock);
4001 __trace_array_put(tr);
4002 mutex_unlock(&trace_types_lock);
4006 static int tracing_release_pipe(struct inode *inode, struct file *file)
4008 struct trace_iterator *iter = file->private_data;
4009 struct trace_array *tr = inode->i_private;
4011 mutex_lock(&trace_types_lock);
4013 if (iter->trace->pipe_close)
4014 iter->trace->pipe_close(iter);
4016 mutex_unlock(&trace_types_lock);
4018 free_cpumask_var(iter->started);
4019 mutex_destroy(&iter->mutex);
4023 trace_array_put(tr);
4029 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
4031 /* Iterators are static, they should be filled or empty */
4032 if (trace_buffer_iter(iter, iter->cpu_file))
4033 return POLLIN | POLLRDNORM;
4035 if (trace_flags & TRACE_ITER_BLOCK)
4037 * Always select as readable when in blocking mode
4039 return POLLIN | POLLRDNORM;
4041 return ring_buffer_poll_wait(iter->trace_buffer->buffer, iter->cpu_file,
4046 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
4048 struct trace_iterator *iter = filp->private_data;
4050 return trace_poll(iter, filp, poll_table);
4054 * This is a make-shift waitqueue.
4055 * A tracer might use this callback on some rare cases:
4057 * 1) the current tracer might hold the runqueue lock when it wakes up
4058 * a reader, hence a deadlock (sched, function, and function graph tracers)
4059 * 2) the function tracers, trace all functions, we don't want
4060 * the overhead of calling wake_up and friends
4061 * (and tracing them too)
4063 * Anyway, this is really very primitive wakeup.
4065 int poll_wait_pipe(struct trace_iterator *iter)
4067 set_current_state(TASK_INTERRUPTIBLE);
4068 /* sleep for 100 msecs, and try again. */
4069 schedule_timeout(HZ / 10);
4073 /* Must be called with trace_types_lock mutex held. */
4074 static int tracing_wait_pipe(struct file *filp)
4076 struct trace_iterator *iter = filp->private_data;
4079 while (trace_empty(iter)) {
4081 if ((filp->f_flags & O_NONBLOCK)) {
4085 mutex_unlock(&iter->mutex);
4087 ret = iter->trace->wait_pipe(iter);
4089 mutex_lock(&iter->mutex);
4094 if (signal_pending(current))
4098 * We block until we read something and tracing is disabled.
4099 * We still block if tracing is disabled, but we have never
4100 * read anything. This allows a user to cat this file, and
4101 * then enable tracing. But after we have read something,
4102 * we give an EOF when tracing is again disabled.
4104 * iter->pos will be 0 if we haven't read anything.
4106 if (!tracing_is_on() && iter->pos)
4117 tracing_read_pipe(struct file *filp, char __user *ubuf,
4118 size_t cnt, loff_t *ppos)
4120 struct trace_iterator *iter = filp->private_data;
4121 struct trace_array *tr = iter->tr;
4124 /* return any leftover data */
4125 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4129 trace_seq_init(&iter->seq);
4131 /* copy the tracer to avoid using a global lock all around */
4132 mutex_lock(&trace_types_lock);
4133 if (unlikely(iter->trace->name != tr->current_trace->name))
4134 *iter->trace = *tr->current_trace;
4135 mutex_unlock(&trace_types_lock);
4138 * Avoid more than one consumer on a single file descriptor
4139 * This is just a matter of traces coherency, the ring buffer itself
4142 mutex_lock(&iter->mutex);
4143 if (iter->trace->read) {
4144 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
4150 sret = tracing_wait_pipe(filp);
4154 /* stop when tracing is finished */
4155 if (trace_empty(iter)) {
4160 if (cnt >= PAGE_SIZE)
4161 cnt = PAGE_SIZE - 1;
4163 /* reset all but tr, trace, and overruns */
4164 memset(&iter->seq, 0,
4165 sizeof(struct trace_iterator) -
4166 offsetof(struct trace_iterator, seq));
4167 cpumask_clear(iter->started);
4170 trace_event_read_lock();
4171 trace_access_lock(iter->cpu_file);
4172 while (trace_find_next_entry_inc(iter) != NULL) {
4173 enum print_line_t ret;
4174 int len = iter->seq.len;
4176 ret = print_trace_line(iter);
4177 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4178 /* don't print partial lines */
4179 iter->seq.len = len;
4182 if (ret != TRACE_TYPE_NO_CONSUME)
4183 trace_consume(iter);
4185 if (iter->seq.len >= cnt)
4189 * Setting the full flag means we reached the trace_seq buffer
4190 * size and we should leave by partial output condition above.
4191 * One of the trace_seq_* functions is not used properly.
4193 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
4196 trace_access_unlock(iter->cpu_file);
4197 trace_event_read_unlock();
4199 /* Now copy what we have to the user */
4200 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
4201 if (iter->seq.readpos >= iter->seq.len)
4202 trace_seq_init(&iter->seq);
4205 * If there was nothing to send to user, in spite of consuming trace
4206 * entries, go back to wait for more entries.
4212 mutex_unlock(&iter->mutex);
4217 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
4218 struct pipe_buffer *buf)
4220 __free_page(buf->page);
4223 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
4226 __free_page(spd->pages[idx]);
4229 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
4231 .map = generic_pipe_buf_map,
4232 .unmap = generic_pipe_buf_unmap,
4233 .confirm = generic_pipe_buf_confirm,
4234 .release = tracing_pipe_buf_release,
4235 .steal = generic_pipe_buf_steal,
4236 .get = generic_pipe_buf_get,
4240 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
4245 /* Seq buffer is page-sized, exactly what we need. */
4247 count = iter->seq.len;
4248 ret = print_trace_line(iter);
4249 count = iter->seq.len - count;
4252 iter->seq.len -= count;
4255 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4256 iter->seq.len -= count;
4260 if (ret != TRACE_TYPE_NO_CONSUME)
4261 trace_consume(iter);
4263 if (!trace_find_next_entry_inc(iter)) {
4273 static ssize_t tracing_splice_read_pipe(struct file *filp,
4275 struct pipe_inode_info *pipe,
4279 struct page *pages_def[PIPE_DEF_BUFFERS];
4280 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4281 struct trace_iterator *iter = filp->private_data;
4282 struct splice_pipe_desc spd = {
4284 .partial = partial_def,
4285 .nr_pages = 0, /* This gets updated below. */
4286 .nr_pages_max = PIPE_DEF_BUFFERS,
4288 .ops = &tracing_pipe_buf_ops,
4289 .spd_release = tracing_spd_release_pipe,
4291 struct trace_array *tr = iter->tr;
4296 if (splice_grow_spd(pipe, &spd))
4299 /* copy the tracer to avoid using a global lock all around */
4300 mutex_lock(&trace_types_lock);
4301 if (unlikely(iter->trace->name != tr->current_trace->name))
4302 *iter->trace = *tr->current_trace;
4303 mutex_unlock(&trace_types_lock);
4305 mutex_lock(&iter->mutex);
4307 if (iter->trace->splice_read) {
4308 ret = iter->trace->splice_read(iter, filp,
4309 ppos, pipe, len, flags);
4314 ret = tracing_wait_pipe(filp);
4318 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
4323 trace_event_read_lock();
4324 trace_access_lock(iter->cpu_file);
4326 /* Fill as many pages as possible. */
4327 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
4328 spd.pages[i] = alloc_page(GFP_KERNEL);
4332 rem = tracing_fill_pipe_page(rem, iter);
4334 /* Copy the data into the page, so we can start over. */
4335 ret = trace_seq_to_buffer(&iter->seq,
4336 page_address(spd.pages[i]),
4339 __free_page(spd.pages[i]);
4342 spd.partial[i].offset = 0;
4343 spd.partial[i].len = iter->seq.len;
4345 trace_seq_init(&iter->seq);
4348 trace_access_unlock(iter->cpu_file);
4349 trace_event_read_unlock();
4350 mutex_unlock(&iter->mutex);
4354 ret = splice_to_pipe(pipe, &spd);
4356 splice_shrink_spd(&spd);
4360 mutex_unlock(&iter->mutex);
4365 tracing_entries_read(struct file *filp, char __user *ubuf,
4366 size_t cnt, loff_t *ppos)
4368 struct inode *inode = file_inode(filp);
4369 struct trace_array *tr = inode->i_private;
4370 int cpu = tracing_get_cpu(inode);
4375 mutex_lock(&trace_types_lock);
4377 if (cpu == RING_BUFFER_ALL_CPUS) {
4378 int cpu, buf_size_same;
4383 /* check if all cpu sizes are same */
4384 for_each_tracing_cpu(cpu) {
4385 /* fill in the size from first enabled cpu */
4387 size = per_cpu_ptr(tr->trace_buffer.data, cpu)->entries;
4388 if (size != per_cpu_ptr(tr->trace_buffer.data, cpu)->entries) {
4394 if (buf_size_same) {
4395 if (!ring_buffer_expanded)
4396 r = sprintf(buf, "%lu (expanded: %lu)\n",
4398 trace_buf_size >> 10);
4400 r = sprintf(buf, "%lu\n", size >> 10);
4402 r = sprintf(buf, "X\n");
4404 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10);
4406 mutex_unlock(&trace_types_lock);
4408 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4413 tracing_entries_write(struct file *filp, const char __user *ubuf,
4414 size_t cnt, loff_t *ppos)
4416 struct inode *inode = file_inode(filp);
4417 struct trace_array *tr = inode->i_private;
4421 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4425 /* must have at least 1 entry */
4429 /* value is in KB */
4431 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
4441 tracing_total_entries_read(struct file *filp, char __user *ubuf,
4442 size_t cnt, loff_t *ppos)
4444 struct trace_array *tr = filp->private_data;
4447 unsigned long size = 0, expanded_size = 0;
4449 mutex_lock(&trace_types_lock);
4450 for_each_tracing_cpu(cpu) {
4451 size += per_cpu_ptr(tr->trace_buffer.data, cpu)->entries >> 10;
4452 if (!ring_buffer_expanded)
4453 expanded_size += trace_buf_size >> 10;
4455 if (ring_buffer_expanded)
4456 r = sprintf(buf, "%lu\n", size);
4458 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
4459 mutex_unlock(&trace_types_lock);
4461 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4465 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
4466 size_t cnt, loff_t *ppos)
4469 * There is no need to read what the user has written, this function
4470 * is just to make sure that there is no error when "echo" is used
4479 tracing_free_buffer_release(struct inode *inode, struct file *filp)
4481 struct trace_array *tr = inode->i_private;
4483 /* disable tracing ? */
4484 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
4485 tracer_tracing_off(tr);
4486 /* resize the ring buffer to 0 */
4487 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
4489 trace_array_put(tr);
4495 tracing_mark_write(struct file *filp, const char __user *ubuf,
4496 size_t cnt, loff_t *fpos)
4498 unsigned long addr = (unsigned long)ubuf;
4499 struct trace_array *tr = filp->private_data;
4500 struct ring_buffer_event *event;
4501 struct ring_buffer *buffer;
4502 struct print_entry *entry;
4503 unsigned long irq_flags;
4504 struct page *pages[2];
4514 if (tracing_disabled)
4517 if (!(trace_flags & TRACE_ITER_MARKERS))
4520 if (cnt > TRACE_BUF_SIZE)
4521 cnt = TRACE_BUF_SIZE;
4524 * Userspace is injecting traces into the kernel trace buffer.
4525 * We want to be as non intrusive as possible.
4526 * To do so, we do not want to allocate any special buffers
4527 * or take any locks, but instead write the userspace data
4528 * straight into the ring buffer.
4530 * First we need to pin the userspace buffer into memory,
4531 * which, most likely it is, because it just referenced it.
4532 * But there's no guarantee that it is. By using get_user_pages_fast()
4533 * and kmap_atomic/kunmap_atomic() we can get access to the
4534 * pages directly. We then write the data directly into the
4537 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
4539 /* check if we cross pages */
4540 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
4543 offset = addr & (PAGE_SIZE - 1);
4546 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
4547 if (ret < nr_pages) {
4549 put_page(pages[ret]);
4554 for (i = 0; i < nr_pages; i++)
4555 map_page[i] = kmap_atomic(pages[i]);
4557 local_save_flags(irq_flags);
4558 size = sizeof(*entry) + cnt + 2; /* possible \n added */
4559 buffer = tr->trace_buffer.buffer;
4560 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4561 irq_flags, preempt_count());
4563 /* Ring buffer disabled, return as if not open for write */
4568 entry = ring_buffer_event_data(event);
4569 entry->ip = _THIS_IP_;
4571 if (nr_pages == 2) {
4572 len = PAGE_SIZE - offset;
4573 memcpy(&entry->buf, map_page[0] + offset, len);
4574 memcpy(&entry->buf[len], map_page[1], cnt - len);
4576 memcpy(&entry->buf, map_page[0] + offset, cnt);
4578 if (entry->buf[cnt - 1] != '\n') {
4579 entry->buf[cnt] = '\n';
4580 entry->buf[cnt + 1] = '\0';
4582 entry->buf[cnt] = '\0';
4584 __buffer_unlock_commit(buffer, event);
4591 for (i = 0; i < nr_pages; i++){
4592 kunmap_atomic(map_page[i]);
4599 static int tracing_clock_show(struct seq_file *m, void *v)
4601 struct trace_array *tr = m->private;
4604 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4606 "%s%s%s%s", i ? " " : "",
4607 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
4608 i == tr->clock_id ? "]" : "");
4614 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4615 size_t cnt, loff_t *fpos)
4617 struct seq_file *m = filp->private_data;
4618 struct trace_array *tr = m->private;
4620 const char *clockstr;
4623 if (cnt >= sizeof(buf))
4626 if (copy_from_user(&buf, ubuf, cnt))
4631 clockstr = strstrip(buf);
4633 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4634 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4637 if (i == ARRAY_SIZE(trace_clocks))
4640 mutex_lock(&trace_types_lock);
4644 ring_buffer_set_clock(tr->trace_buffer.buffer, trace_clocks[i].func);
4647 * New clock may not be consistent with the previous clock.
4648 * Reset the buffer so that it doesn't have incomparable timestamps.
4650 tracing_reset_online_cpus(&tr->trace_buffer);
4652 #ifdef CONFIG_TRACER_MAX_TRACE
4653 if (tr->flags & TRACE_ARRAY_FL_GLOBAL && tr->max_buffer.buffer)
4654 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
4655 tracing_reset_online_cpus(&tr->max_buffer);
4658 mutex_unlock(&trace_types_lock);
4665 static int tracing_clock_open(struct inode *inode, struct file *file)
4667 struct trace_array *tr = inode->i_private;
4670 if (tracing_disabled)
4673 if (trace_array_get(tr))
4676 ret = single_open(file, tracing_clock_show, inode->i_private);
4678 trace_array_put(tr);
4683 struct ftrace_buffer_info {
4684 struct trace_iterator iter;
4689 #ifdef CONFIG_TRACER_SNAPSHOT
4690 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4692 struct trace_array *tr = inode->i_private;
4693 struct trace_iterator *iter;
4697 if (trace_array_get(tr) < 0)
4700 if (file->f_mode & FMODE_READ) {
4701 iter = __tracing_open(inode, file, true);
4703 ret = PTR_ERR(iter);
4705 /* Writes still need the seq_file to hold the private data */
4707 m = kzalloc(sizeof(*m), GFP_KERNEL);
4710 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
4718 iter->trace_buffer = &tr->max_buffer;
4719 iter->cpu_file = tracing_get_cpu(inode);
4721 file->private_data = m;
4725 trace_array_put(tr);
4731 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4734 struct seq_file *m = filp->private_data;
4735 struct trace_iterator *iter = m->private;
4736 struct trace_array *tr = iter->tr;
4740 ret = tracing_update_buffers();
4744 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4748 mutex_lock(&trace_types_lock);
4750 if (tr->current_trace->use_max_tr) {
4757 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4761 if (tr->allocated_snapshot)
4765 /* Only allow per-cpu swap if the ring buffer supports it */
4766 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4767 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4772 if (!tr->allocated_snapshot) {
4773 ret = alloc_snapshot(tr);
4777 local_irq_disable();
4778 /* Now, we're going to swap */
4779 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4780 update_max_tr(tr, current, smp_processor_id());
4782 update_max_tr_single(tr, current, iter->cpu_file);
4786 if (tr->allocated_snapshot) {
4787 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4788 tracing_reset_online_cpus(&tr->max_buffer);
4790 tracing_reset(&tr->max_buffer, iter->cpu_file);
4800 mutex_unlock(&trace_types_lock);
4804 static int tracing_snapshot_release(struct inode *inode, struct file *file)
4806 struct seq_file *m = file->private_data;
4809 ret = tracing_release(inode, file);
4811 if (file->f_mode & FMODE_READ)
4814 /* If write only, the seq_file is just a stub */
4822 static int tracing_buffers_open(struct inode *inode, struct file *filp);
4823 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
4824 size_t count, loff_t *ppos);
4825 static int tracing_buffers_release(struct inode *inode, struct file *file);
4826 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4827 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
4829 static int snapshot_raw_open(struct inode *inode, struct file *filp)
4831 struct ftrace_buffer_info *info;
4834 ret = tracing_buffers_open(inode, filp);
4838 info = filp->private_data;
4840 if (info->iter.trace->use_max_tr) {
4841 tracing_buffers_release(inode, filp);
4845 info->iter.snapshot = true;
4846 info->iter.trace_buffer = &info->iter.tr->max_buffer;
4851 #endif /* CONFIG_TRACER_SNAPSHOT */
4854 static const struct file_operations tracing_max_lat_fops = {
4855 .open = tracing_open_generic,
4856 .read = tracing_max_lat_read,
4857 .write = tracing_max_lat_write,
4858 .llseek = generic_file_llseek,
4861 static const struct file_operations set_tracer_fops = {
4862 .open = tracing_open_generic,
4863 .read = tracing_set_trace_read,
4864 .write = tracing_set_trace_write,
4865 .llseek = generic_file_llseek,
4868 static const struct file_operations tracing_pipe_fops = {
4869 .open = tracing_open_pipe,
4870 .poll = tracing_poll_pipe,
4871 .read = tracing_read_pipe,
4872 .splice_read = tracing_splice_read_pipe,
4873 .release = tracing_release_pipe,
4874 .llseek = no_llseek,
4877 static const struct file_operations tracing_entries_fops = {
4878 .open = tracing_open_generic_tr,
4879 .read = tracing_entries_read,
4880 .write = tracing_entries_write,
4881 .llseek = generic_file_llseek,
4882 .release = tracing_release_generic_tr,
4885 static const struct file_operations tracing_total_entries_fops = {
4886 .open = tracing_open_generic_tr,
4887 .read = tracing_total_entries_read,
4888 .llseek = generic_file_llseek,
4889 .release = tracing_release_generic_tr,
4892 static const struct file_operations tracing_free_buffer_fops = {
4893 .open = tracing_open_generic_tr,
4894 .write = tracing_free_buffer_write,
4895 .release = tracing_free_buffer_release,
4898 static const struct file_operations tracing_mark_fops = {
4899 .open = tracing_open_generic_tr,
4900 .write = tracing_mark_write,
4901 .llseek = generic_file_llseek,
4902 .release = tracing_release_generic_tr,
4905 static const struct file_operations trace_clock_fops = {
4906 .open = tracing_clock_open,
4908 .llseek = seq_lseek,
4909 .release = tracing_single_release_tr,
4910 .write = tracing_clock_write,
4913 #ifdef CONFIG_TRACER_SNAPSHOT
4914 static const struct file_operations snapshot_fops = {
4915 .open = tracing_snapshot_open,
4917 .write = tracing_snapshot_write,
4918 .llseek = tracing_seek,
4919 .release = tracing_snapshot_release,
4922 static const struct file_operations snapshot_raw_fops = {
4923 .open = snapshot_raw_open,
4924 .read = tracing_buffers_read,
4925 .release = tracing_buffers_release,
4926 .splice_read = tracing_buffers_splice_read,
4927 .llseek = no_llseek,
4930 #endif /* CONFIG_TRACER_SNAPSHOT */
4932 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4934 struct trace_array *tr = inode->i_private;
4935 struct ftrace_buffer_info *info;
4938 if (tracing_disabled)
4941 if (trace_array_get(tr) < 0)
4944 info = kzalloc(sizeof(*info), GFP_KERNEL);
4946 trace_array_put(tr);
4950 mutex_lock(&trace_types_lock);
4953 info->iter.cpu_file = tracing_get_cpu(inode);
4954 info->iter.trace = tr->current_trace;
4955 info->iter.trace_buffer = &tr->trace_buffer;
4957 /* Force reading ring buffer for first read */
4958 info->read = (unsigned int)-1;
4960 filp->private_data = info;
4962 mutex_unlock(&trace_types_lock);
4964 ret = nonseekable_open(inode, filp);
4966 trace_array_put(tr);
4972 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
4974 struct ftrace_buffer_info *info = filp->private_data;
4975 struct trace_iterator *iter = &info->iter;
4977 return trace_poll(iter, filp, poll_table);
4981 tracing_buffers_read(struct file *filp, char __user *ubuf,
4982 size_t count, loff_t *ppos)
4984 struct ftrace_buffer_info *info = filp->private_data;
4985 struct trace_iterator *iter = &info->iter;
4992 mutex_lock(&trace_types_lock);
4994 #ifdef CONFIG_TRACER_MAX_TRACE
4995 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5002 info->spare = ring_buffer_alloc_read_page(iter->trace_buffer->buffer,
5008 /* Do we have previous read data to read? */
5009 if (info->read < PAGE_SIZE)
5013 trace_access_lock(iter->cpu_file);
5014 ret = ring_buffer_read_page(iter->trace_buffer->buffer,
5018 trace_access_unlock(iter->cpu_file);
5021 if (trace_empty(iter)) {
5022 if ((filp->f_flags & O_NONBLOCK)) {
5026 mutex_unlock(&trace_types_lock);
5027 ret = iter->trace->wait_pipe(iter);
5028 mutex_lock(&trace_types_lock);
5033 if (signal_pending(current)) {
5045 size = PAGE_SIZE - info->read;
5049 ret = copy_to_user(ubuf, info->spare + info->read, size);
5060 mutex_unlock(&trace_types_lock);
5065 static int tracing_buffers_release(struct inode *inode, struct file *file)
5067 struct ftrace_buffer_info *info = file->private_data;
5068 struct trace_iterator *iter = &info->iter;
5070 mutex_lock(&trace_types_lock);
5072 __trace_array_put(iter->tr);
5075 ring_buffer_free_read_page(iter->trace_buffer->buffer, info->spare);
5078 mutex_unlock(&trace_types_lock);
5084 struct ring_buffer *buffer;
5089 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
5090 struct pipe_buffer *buf)
5092 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5097 ring_buffer_free_read_page(ref->buffer, ref->page);
5102 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
5103 struct pipe_buffer *buf)
5105 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
5110 /* Pipe buffer operations for a buffer. */
5111 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
5113 .map = generic_pipe_buf_map,
5114 .unmap = generic_pipe_buf_unmap,
5115 .confirm = generic_pipe_buf_confirm,
5116 .release = buffer_pipe_buf_release,
5117 .steal = generic_pipe_buf_steal,
5118 .get = buffer_pipe_buf_get,
5122 * Callback from splice_to_pipe(), if we need to release some pages
5123 * at the end of the spd in case we error'ed out in filling the pipe.
5125 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
5127 struct buffer_ref *ref =
5128 (struct buffer_ref *)spd->partial[i].private;
5133 ring_buffer_free_read_page(ref->buffer, ref->page);
5135 spd->partial[i].private = 0;
5139 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
5140 struct pipe_inode_info *pipe, size_t len,
5143 struct ftrace_buffer_info *info = file->private_data;
5144 struct trace_iterator *iter = &info->iter;
5145 struct partial_page partial_def[PIPE_DEF_BUFFERS];
5146 struct page *pages_def[PIPE_DEF_BUFFERS];
5147 struct splice_pipe_desc spd = {
5149 .partial = partial_def,
5150 .nr_pages_max = PIPE_DEF_BUFFERS,
5152 .ops = &buffer_pipe_buf_ops,
5153 .spd_release = buffer_spd_release,
5155 struct buffer_ref *ref;
5156 int entries, size, i;
5159 mutex_lock(&trace_types_lock);
5161 #ifdef CONFIG_TRACER_MAX_TRACE
5162 if (iter->snapshot && iter->tr->current_trace->use_max_tr) {
5168 if (splice_grow_spd(pipe, &spd)) {
5173 if (*ppos & (PAGE_SIZE - 1)) {
5178 if (len & (PAGE_SIZE - 1)) {
5179 if (len < PAGE_SIZE) {
5187 trace_access_lock(iter->cpu_file);
5188 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5190 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
5194 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
5199 ref->buffer = iter->trace_buffer->buffer;
5200 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
5206 r = ring_buffer_read_page(ref->buffer, &ref->page,
5207 len, iter->cpu_file, 1);
5209 ring_buffer_free_read_page(ref->buffer, ref->page);
5215 * zero out any left over data, this is going to
5218 size = ring_buffer_page_len(ref->page);
5219 if (size < PAGE_SIZE)
5220 memset(ref->page + size, 0, PAGE_SIZE - size);
5222 page = virt_to_page(ref->page);
5224 spd.pages[i] = page;
5225 spd.partial[i].len = PAGE_SIZE;
5226 spd.partial[i].offset = 0;
5227 spd.partial[i].private = (unsigned long)ref;
5231 entries = ring_buffer_entries_cpu(iter->trace_buffer->buffer, iter->cpu_file);
5234 trace_access_unlock(iter->cpu_file);
5237 /* did we read anything? */
5238 if (!spd.nr_pages) {
5239 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK)) {
5243 mutex_unlock(&trace_types_lock);
5244 ret = iter->trace->wait_pipe(iter);
5245 mutex_lock(&trace_types_lock);
5248 if (signal_pending(current)) {
5255 ret = splice_to_pipe(pipe, &spd);
5256 splice_shrink_spd(&spd);
5258 mutex_unlock(&trace_types_lock);
5263 static const struct file_operations tracing_buffers_fops = {
5264 .open = tracing_buffers_open,
5265 .read = tracing_buffers_read,
5266 .poll = tracing_buffers_poll,
5267 .release = tracing_buffers_release,
5268 .splice_read = tracing_buffers_splice_read,
5269 .llseek = no_llseek,
5273 tracing_stats_read(struct file *filp, char __user *ubuf,
5274 size_t count, loff_t *ppos)
5276 struct inode *inode = file_inode(filp);
5277 struct trace_array *tr = inode->i_private;
5278 struct trace_buffer *trace_buf = &tr->trace_buffer;
5279 int cpu = tracing_get_cpu(inode);
5280 struct trace_seq *s;
5282 unsigned long long t;
5283 unsigned long usec_rem;
5285 s = kmalloc(sizeof(*s), GFP_KERNEL);
5291 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
5292 trace_seq_printf(s, "entries: %ld\n", cnt);
5294 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
5295 trace_seq_printf(s, "overrun: %ld\n", cnt);
5297 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
5298 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
5300 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
5301 trace_seq_printf(s, "bytes: %ld\n", cnt);
5303 if (trace_clocks[tr->clock_id].in_ns) {
5304 /* local or global for trace_clock */
5305 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5306 usec_rem = do_div(t, USEC_PER_SEC);
5307 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
5310 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer, cpu));
5311 usec_rem = do_div(t, USEC_PER_SEC);
5312 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
5314 /* counter or tsc mode for trace_clock */
5315 trace_seq_printf(s, "oldest event ts: %llu\n",
5316 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
5318 trace_seq_printf(s, "now ts: %llu\n",
5319 ring_buffer_time_stamp(trace_buf->buffer, cpu));
5322 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
5323 trace_seq_printf(s, "dropped events: %ld\n", cnt);
5325 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
5326 trace_seq_printf(s, "read events: %ld\n", cnt);
5328 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
5335 static const struct file_operations tracing_stats_fops = {
5336 .open = tracing_open_generic_tr,
5337 .read = tracing_stats_read,
5338 .llseek = generic_file_llseek,
5339 .release = tracing_release_generic_tr,
5342 #ifdef CONFIG_DYNAMIC_FTRACE
5344 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
5350 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
5351 size_t cnt, loff_t *ppos)
5353 static char ftrace_dyn_info_buffer[1024];
5354 static DEFINE_MUTEX(dyn_info_mutex);
5355 unsigned long *p = filp->private_data;
5356 char *buf = ftrace_dyn_info_buffer;
5357 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
5360 mutex_lock(&dyn_info_mutex);
5361 r = sprintf(buf, "%ld ", *p);
5363 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
5366 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5368 mutex_unlock(&dyn_info_mutex);
5373 static const struct file_operations tracing_dyn_info_fops = {
5374 .open = tracing_open_generic,
5375 .read = tracing_read_dyn_info,
5376 .llseek = generic_file_llseek,
5378 #endif /* CONFIG_DYNAMIC_FTRACE */
5380 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5382 ftrace_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5388 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip, void **data)
5390 unsigned long *count = (long *)data;
5402 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
5403 struct ftrace_probe_ops *ops, void *data)
5405 long count = (long)data;
5407 seq_printf(m, "%ps:", (void *)ip);
5409 seq_printf(m, "snapshot");
5412 seq_printf(m, ":unlimited\n");
5414 seq_printf(m, ":count=%ld\n", count);
5419 static struct ftrace_probe_ops snapshot_probe_ops = {
5420 .func = ftrace_snapshot,
5421 .print = ftrace_snapshot_print,
5424 static struct ftrace_probe_ops snapshot_count_probe_ops = {
5425 .func = ftrace_count_snapshot,
5426 .print = ftrace_snapshot_print,
5430 ftrace_trace_snapshot_callback(struct ftrace_hash *hash,
5431 char *glob, char *cmd, char *param, int enable)
5433 struct ftrace_probe_ops *ops;
5434 void *count = (void *)-1;
5438 /* hash funcs only work with set_ftrace_filter */
5442 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
5444 if (glob[0] == '!') {
5445 unregister_ftrace_function_probe_func(glob+1, ops);
5452 number = strsep(¶m, ":");
5454 if (!strlen(number))
5458 * We use the callback data field (which is a pointer)
5461 ret = kstrtoul(number, 0, (unsigned long *)&count);
5466 ret = register_ftrace_function_probe(glob, ops, count);
5469 alloc_snapshot(&global_trace);
5471 return ret < 0 ? ret : 0;
5474 static struct ftrace_func_command ftrace_snapshot_cmd = {
5476 .func = ftrace_trace_snapshot_callback,
5479 static int register_snapshot_cmd(void)
5481 return register_ftrace_command(&ftrace_snapshot_cmd);
5484 static inline int register_snapshot_cmd(void) { return 0; }
5485 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5487 struct dentry *tracing_init_dentry_tr(struct trace_array *tr)
5492 if (!debugfs_initialized())
5495 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
5496 tr->dir = debugfs_create_dir("tracing", NULL);
5499 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5504 struct dentry *tracing_init_dentry(void)
5506 return tracing_init_dentry_tr(&global_trace);
5509 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
5511 struct dentry *d_tracer;
5514 return tr->percpu_dir;
5516 d_tracer = tracing_init_dentry_tr(tr);
5520 tr->percpu_dir = debugfs_create_dir("per_cpu", d_tracer);
5522 WARN_ONCE(!tr->percpu_dir,
5523 "Could not create debugfs directory 'per_cpu/%d'\n", cpu);
5525 return tr->percpu_dir;
5528 static struct dentry *
5529 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
5530 void *data, long cpu, const struct file_operations *fops)
5532 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
5534 if (ret) /* See tracing_get_cpu() */
5535 ret->d_inode->i_cdev = (void *)(cpu + 1);
5540 tracing_init_debugfs_percpu(struct trace_array *tr, long cpu)
5542 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
5543 struct dentry *d_cpu;
5544 char cpu_dir[30]; /* 30 characters should be more than enough */
5549 snprintf(cpu_dir, 30, "cpu%ld", cpu);
5550 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
5552 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
5556 /* per cpu trace_pipe */
5557 trace_create_cpu_file("trace_pipe", 0444, d_cpu,
5558 tr, cpu, &tracing_pipe_fops);
5561 trace_create_cpu_file("trace", 0644, d_cpu,
5562 tr, cpu, &tracing_fops);
5564 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu,
5565 tr, cpu, &tracing_buffers_fops);
5567 trace_create_cpu_file("stats", 0444, d_cpu,
5568 tr, cpu, &tracing_stats_fops);
5570 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu,
5571 tr, cpu, &tracing_entries_fops);
5573 #ifdef CONFIG_TRACER_SNAPSHOT
5574 trace_create_cpu_file("snapshot", 0644, d_cpu,
5575 tr, cpu, &snapshot_fops);
5577 trace_create_cpu_file("snapshot_raw", 0444, d_cpu,
5578 tr, cpu, &snapshot_raw_fops);
5582 #ifdef CONFIG_FTRACE_SELFTEST
5583 /* Let selftest have access to static functions in this file */
5584 #include "trace_selftest.c"
5587 struct trace_option_dentry {
5588 struct tracer_opt *opt;
5589 struct tracer_flags *flags;
5590 struct trace_array *tr;
5591 struct dentry *entry;
5595 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
5598 struct trace_option_dentry *topt = filp->private_data;
5601 if (topt->flags->val & topt->opt->bit)
5606 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5610 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
5613 struct trace_option_dentry *topt = filp->private_data;
5617 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5621 if (val != 0 && val != 1)
5624 if (!!(topt->flags->val & topt->opt->bit) != val) {
5625 mutex_lock(&trace_types_lock);
5626 ret = __set_tracer_option(topt->tr->current_trace, topt->flags,
5628 mutex_unlock(&trace_types_lock);
5639 static const struct file_operations trace_options_fops = {
5640 .open = tracing_open_generic,
5641 .read = trace_options_read,
5642 .write = trace_options_write,
5643 .llseek = generic_file_llseek,
5647 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
5650 long index = (long)filp->private_data;
5653 if (trace_flags & (1 << index))
5658 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
5662 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
5665 struct trace_array *tr = &global_trace;
5666 long index = (long)filp->private_data;
5670 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5674 if (val != 0 && val != 1)
5677 mutex_lock(&trace_types_lock);
5678 ret = set_tracer_flag(tr, 1 << index, val);
5679 mutex_unlock(&trace_types_lock);
5689 static const struct file_operations trace_options_core_fops = {
5690 .open = tracing_open_generic,
5691 .read = trace_options_core_read,
5692 .write = trace_options_core_write,
5693 .llseek = generic_file_llseek,
5696 struct dentry *trace_create_file(const char *name,
5698 struct dentry *parent,
5700 const struct file_operations *fops)
5704 ret = debugfs_create_file(name, mode, parent, data, fops);
5706 pr_warning("Could not create debugfs '%s' entry\n", name);
5712 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
5714 struct dentry *d_tracer;
5719 d_tracer = tracing_init_dentry_tr(tr);
5723 tr->options = debugfs_create_dir("options", d_tracer);
5725 pr_warning("Could not create debugfs directory 'options'\n");
5733 create_trace_option_file(struct trace_array *tr,
5734 struct trace_option_dentry *topt,
5735 struct tracer_flags *flags,
5736 struct tracer_opt *opt)
5738 struct dentry *t_options;
5740 t_options = trace_options_init_dentry(tr);
5744 topt->flags = flags;
5748 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
5749 &trace_options_fops);
5753 static struct trace_option_dentry *
5754 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
5756 struct trace_option_dentry *topts;
5757 struct tracer_flags *flags;
5758 struct tracer_opt *opts;
5764 flags = tracer->flags;
5766 if (!flags || !flags->opts)
5771 for (cnt = 0; opts[cnt].name; cnt++)
5774 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
5778 for (cnt = 0; opts[cnt].name; cnt++)
5779 create_trace_option_file(tr, &topts[cnt], flags,
5786 destroy_trace_option_files(struct trace_option_dentry *topts)
5793 for (cnt = 0; topts[cnt].opt; cnt++) {
5794 if (topts[cnt].entry)
5795 debugfs_remove(topts[cnt].entry);
5801 static struct dentry *
5802 create_trace_option_core_file(struct trace_array *tr,
5803 const char *option, long index)
5805 struct dentry *t_options;
5807 t_options = trace_options_init_dentry(tr);
5811 return trace_create_file(option, 0644, t_options, (void *)index,
5812 &trace_options_core_fops);
5815 static __init void create_trace_options_dir(struct trace_array *tr)
5817 struct dentry *t_options;
5820 t_options = trace_options_init_dentry(tr);
5824 for (i = 0; trace_options[i]; i++)
5825 create_trace_option_core_file(tr, trace_options[i], i);
5829 rb_simple_read(struct file *filp, char __user *ubuf,
5830 size_t cnt, loff_t *ppos)
5832 struct trace_array *tr = filp->private_data;
5836 r = tracer_tracing_is_on(tr);
5837 r = sprintf(buf, "%d\n", r);
5839 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
5843 rb_simple_write(struct file *filp, const char __user *ubuf,
5844 size_t cnt, loff_t *ppos)
5846 struct trace_array *tr = filp->private_data;
5847 struct ring_buffer *buffer = tr->trace_buffer.buffer;
5851 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
5856 mutex_lock(&trace_types_lock);
5858 tracer_tracing_on(tr);
5859 if (tr->current_trace->start)
5860 tr->current_trace->start(tr);
5862 tracer_tracing_off(tr);
5863 if (tr->current_trace->stop)
5864 tr->current_trace->stop(tr);
5866 mutex_unlock(&trace_types_lock);
5874 static const struct file_operations rb_simple_fops = {
5875 .open = tracing_open_generic_tr,
5876 .read = rb_simple_read,
5877 .write = rb_simple_write,
5878 .release = tracing_release_generic_tr,
5879 .llseek = default_llseek,
5882 struct dentry *trace_instance_dir;
5885 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer);
5887 static void init_trace_buffers(struct trace_array *tr, struct trace_buffer *buf)
5891 for_each_tracing_cpu(cpu) {
5892 memset(per_cpu_ptr(buf->data, cpu), 0, sizeof(struct trace_array_cpu));
5893 per_cpu_ptr(buf->data, cpu)->trace_cpu.cpu = cpu;
5894 per_cpu_ptr(buf->data, cpu)->trace_cpu.tr = tr;
5899 allocate_trace_buffer(struct trace_array *tr, struct trace_buffer *buf, int size)
5901 enum ring_buffer_flags rb_flags;
5903 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5907 buf->buffer = ring_buffer_alloc(size, rb_flags);
5911 buf->data = alloc_percpu(struct trace_array_cpu);
5913 ring_buffer_free(buf->buffer);
5917 init_trace_buffers(tr, buf);
5919 /* Allocate the first page for all buffers */
5920 set_buffer_entries(&tr->trace_buffer,
5921 ring_buffer_size(tr->trace_buffer.buffer, 0));
5926 static int allocate_trace_buffers(struct trace_array *tr, int size)
5930 ret = allocate_trace_buffer(tr, &tr->trace_buffer, size);
5934 #ifdef CONFIG_TRACER_MAX_TRACE
5935 ret = allocate_trace_buffer(tr, &tr->max_buffer,
5936 allocate_snapshot ? size : 1);
5938 ring_buffer_free(tr->trace_buffer.buffer);
5939 free_percpu(tr->trace_buffer.data);
5942 tr->allocated_snapshot = allocate_snapshot;
5945 * Only the top level trace array gets its snapshot allocated
5946 * from the kernel command line.
5948 allocate_snapshot = false;
5953 static int new_instance_create(const char *name)
5955 struct trace_array *tr;
5958 mutex_lock(&trace_types_lock);
5961 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
5962 if (tr->name && strcmp(tr->name, name) == 0)
5967 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
5971 tr->name = kstrdup(name, GFP_KERNEL);
5975 raw_spin_lock_init(&tr->start_lock);
5977 tr->current_trace = &nop_trace;
5979 INIT_LIST_HEAD(&tr->systems);
5980 INIT_LIST_HEAD(&tr->events);
5982 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
5985 /* Holder for file callbacks */
5986 tr->trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
5987 tr->trace_cpu.tr = tr;
5989 tr->dir = debugfs_create_dir(name, trace_instance_dir);
5993 ret = event_trace_add_tracer(tr->dir, tr);
5995 debugfs_remove_recursive(tr->dir);
5999 init_tracer_debugfs(tr, tr->dir);
6001 list_add(&tr->list, &ftrace_trace_arrays);
6003 mutex_unlock(&trace_types_lock);
6008 if (tr->trace_buffer.buffer)
6009 ring_buffer_free(tr->trace_buffer.buffer);
6014 mutex_unlock(&trace_types_lock);
6020 static int instance_delete(const char *name)
6022 struct trace_array *tr;
6026 mutex_lock(&trace_types_lock);
6029 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6030 if (tr->name && strcmp(tr->name, name) == 0) {
6042 list_del(&tr->list);
6044 event_trace_del_tracer(tr);
6045 debugfs_remove_recursive(tr->dir);
6046 free_percpu(tr->trace_buffer.data);
6047 ring_buffer_free(tr->trace_buffer.buffer);
6055 mutex_unlock(&trace_types_lock);
6060 static int instance_mkdir (struct inode *inode, struct dentry *dentry, umode_t mode)
6062 struct dentry *parent;
6065 /* Paranoid: Make sure the parent is the "instances" directory */
6066 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6067 if (WARN_ON_ONCE(parent != trace_instance_dir))
6071 * The inode mutex is locked, but debugfs_create_dir() will also
6072 * take the mutex. As the instances directory can not be destroyed
6073 * or changed in any other way, it is safe to unlock it, and
6074 * let the dentry try. If two users try to make the same dir at
6075 * the same time, then the new_instance_create() will determine the
6078 mutex_unlock(&inode->i_mutex);
6080 ret = new_instance_create(dentry->d_iname);
6082 mutex_lock(&inode->i_mutex);
6087 static int instance_rmdir(struct inode *inode, struct dentry *dentry)
6089 struct dentry *parent;
6092 /* Paranoid: Make sure the parent is the "instances" directory */
6093 parent = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
6094 if (WARN_ON_ONCE(parent != trace_instance_dir))
6097 /* The caller did a dget() on dentry */
6098 mutex_unlock(&dentry->d_inode->i_mutex);
6101 * The inode mutex is locked, but debugfs_create_dir() will also
6102 * take the mutex. As the instances directory can not be destroyed
6103 * or changed in any other way, it is safe to unlock it, and
6104 * let the dentry try. If two users try to make the same dir at
6105 * the same time, then the instance_delete() will determine the
6108 mutex_unlock(&inode->i_mutex);
6110 ret = instance_delete(dentry->d_iname);
6112 mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT);
6113 mutex_lock(&dentry->d_inode->i_mutex);
6118 static const struct inode_operations instance_dir_inode_operations = {
6119 .lookup = simple_lookup,
6120 .mkdir = instance_mkdir,
6121 .rmdir = instance_rmdir,
6124 static __init void create_trace_instances(struct dentry *d_tracer)
6126 trace_instance_dir = debugfs_create_dir("instances", d_tracer);
6127 if (WARN_ON(!trace_instance_dir))
6130 /* Hijack the dir inode operations, to allow mkdir */
6131 trace_instance_dir->d_inode->i_op = &instance_dir_inode_operations;
6135 init_tracer_debugfs(struct trace_array *tr, struct dentry *d_tracer)
6139 trace_create_file("trace_options", 0644, d_tracer,
6140 tr, &tracing_iter_fops);
6142 trace_create_file("trace", 0644, d_tracer,
6145 trace_create_file("trace_pipe", 0444, d_tracer,
6146 tr, &tracing_pipe_fops);
6148 trace_create_file("buffer_size_kb", 0644, d_tracer,
6149 tr, &tracing_entries_fops);
6151 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
6152 tr, &tracing_total_entries_fops);
6154 trace_create_file("free_buffer", 0644, d_tracer,
6155 tr, &tracing_free_buffer_fops);
6157 trace_create_file("trace_marker", 0220, d_tracer,
6158 tr, &tracing_mark_fops);
6160 trace_create_file("trace_clock", 0644, d_tracer, tr,
6163 trace_create_file("tracing_on", 0644, d_tracer,
6164 tr, &rb_simple_fops);
6166 #ifdef CONFIG_TRACER_SNAPSHOT
6167 trace_create_file("snapshot", 0644, d_tracer,
6168 tr, &snapshot_fops);
6171 for_each_tracing_cpu(cpu)
6172 tracing_init_debugfs_percpu(tr, cpu);
6176 static __init int tracer_init_debugfs(void)
6178 struct dentry *d_tracer;
6180 trace_access_lock_init();
6182 d_tracer = tracing_init_dentry();
6186 init_tracer_debugfs(&global_trace, d_tracer);
6188 trace_create_file("tracing_cpumask", 0644, d_tracer,
6189 &global_trace, &tracing_cpumask_fops);
6191 trace_create_file("available_tracers", 0444, d_tracer,
6192 &global_trace, &show_traces_fops);
6194 trace_create_file("current_tracer", 0644, d_tracer,
6195 &global_trace, &set_tracer_fops);
6197 #ifdef CONFIG_TRACER_MAX_TRACE
6198 trace_create_file("tracing_max_latency", 0644, d_tracer,
6199 &tracing_max_latency, &tracing_max_lat_fops);
6202 trace_create_file("tracing_thresh", 0644, d_tracer,
6203 &tracing_thresh, &tracing_max_lat_fops);
6205 trace_create_file("README", 0444, d_tracer,
6206 NULL, &tracing_readme_fops);
6208 trace_create_file("saved_cmdlines", 0444, d_tracer,
6209 NULL, &tracing_saved_cmdlines_fops);
6211 #ifdef CONFIG_DYNAMIC_FTRACE
6212 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
6213 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
6216 create_trace_instances(d_tracer);
6218 create_trace_options_dir(&global_trace);
6223 static int trace_panic_handler(struct notifier_block *this,
6224 unsigned long event, void *unused)
6226 if (ftrace_dump_on_oops)
6227 ftrace_dump(ftrace_dump_on_oops);
6231 static struct notifier_block trace_panic_notifier = {
6232 .notifier_call = trace_panic_handler,
6234 .priority = 150 /* priority: INT_MAX >= x >= 0 */
6237 static int trace_die_handler(struct notifier_block *self,
6243 if (ftrace_dump_on_oops)
6244 ftrace_dump(ftrace_dump_on_oops);
6252 static struct notifier_block trace_die_notifier = {
6253 .notifier_call = trace_die_handler,
6258 * printk is set to max of 1024, we really don't need it that big.
6259 * Nothing should be printing 1000 characters anyway.
6261 #define TRACE_MAX_PRINT 1000
6264 * Define here KERN_TRACE so that we have one place to modify
6265 * it if we decide to change what log level the ftrace dump
6268 #define KERN_TRACE KERN_EMERG
6271 trace_printk_seq(struct trace_seq *s)
6273 /* Probably should print a warning here. */
6274 if (s->len >= TRACE_MAX_PRINT)
6275 s->len = TRACE_MAX_PRINT;
6277 /* should be zero ended, but we are paranoid. */
6278 s->buffer[s->len] = 0;
6280 printk(KERN_TRACE "%s", s->buffer);
6285 void trace_init_global_iter(struct trace_iterator *iter)
6287 iter->tr = &global_trace;
6288 iter->trace = iter->tr->current_trace;
6289 iter->cpu_file = RING_BUFFER_ALL_CPUS;
6290 iter->trace_buffer = &global_trace.trace_buffer;
6293 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
6295 /* use static because iter can be a bit big for the stack */
6296 static struct trace_iterator iter;
6297 static atomic_t dump_running;
6298 unsigned int old_userobj;
6299 unsigned long flags;
6302 /* Only allow one dump user at a time. */
6303 if (atomic_inc_return(&dump_running) != 1) {
6304 atomic_dec(&dump_running);
6309 * Always turn off tracing when we dump.
6310 * We don't need to show trace output of what happens
6311 * between multiple crashes.
6313 * If the user does a sysrq-z, then they can re-enable
6314 * tracing with echo 1 > tracing_on.
6318 local_irq_save(flags);
6320 /* Simulate the iterator */
6321 trace_init_global_iter(&iter);
6323 for_each_tracing_cpu(cpu) {
6324 atomic_inc(&per_cpu_ptr(iter.tr->trace_buffer.data, cpu)->disabled);
6327 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
6329 /* don't look at user memory in panic mode */
6330 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
6332 switch (oops_dump_mode) {
6334 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6337 iter.cpu_file = raw_smp_processor_id();
6342 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
6343 iter.cpu_file = RING_BUFFER_ALL_CPUS;
6346 printk(KERN_TRACE "Dumping ftrace buffer:\n");
6348 /* Did function tracer already get disabled? */
6349 if (ftrace_is_dead()) {
6350 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6351 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6355 * We need to stop all tracing on all CPUS to read the
6356 * the next buffer. This is a bit expensive, but is
6357 * not done often. We fill all what we can read,
6358 * and then release the locks again.
6361 while (!trace_empty(&iter)) {
6364 printk(KERN_TRACE "---------------------------------\n");
6368 /* reset all but tr, trace, and overruns */
6369 memset(&iter.seq, 0,
6370 sizeof(struct trace_iterator) -
6371 offsetof(struct trace_iterator, seq));
6372 iter.iter_flags |= TRACE_FILE_LAT_FMT;
6375 if (trace_find_next_entry_inc(&iter) != NULL) {
6378 ret = print_trace_line(&iter);
6379 if (ret != TRACE_TYPE_NO_CONSUME)
6380 trace_consume(&iter);
6382 touch_nmi_watchdog();
6384 trace_printk_seq(&iter.seq);
6388 printk(KERN_TRACE " (ftrace buffer empty)\n");
6390 printk(KERN_TRACE "---------------------------------\n");
6393 trace_flags |= old_userobj;
6395 for_each_tracing_cpu(cpu) {
6396 atomic_dec(&per_cpu_ptr(iter.trace_buffer->data, cpu)->disabled);
6398 atomic_dec(&dump_running);
6399 local_irq_restore(flags);
6401 EXPORT_SYMBOL_GPL(ftrace_dump);
6403 __init static int tracer_alloc_buffers(void)
6409 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
6412 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
6413 goto out_free_buffer_mask;
6415 /* Only allocate trace_printk buffers if a trace_printk exists */
6416 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
6417 /* Must be called before global_trace.buffer is allocated */
6418 trace_printk_init_buffers();
6420 /* To save memory, keep the ring buffer size to its minimum */
6421 if (ring_buffer_expanded)
6422 ring_buf_size = trace_buf_size;
6426 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
6427 cpumask_copy(tracing_cpumask, cpu_all_mask);
6429 raw_spin_lock_init(&global_trace.start_lock);
6431 /* TODO: make the number of buffers hot pluggable with CPUS */
6432 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
6433 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
6435 goto out_free_cpumask;
6438 if (global_trace.buffer_disabled)
6441 trace_init_cmdlines();
6444 * register_tracer() might reference current_trace, so it
6445 * needs to be set before we register anything. This is
6446 * just a bootstrap of current_trace anyway.
6448 global_trace.current_trace = &nop_trace;
6450 register_tracer(&nop_trace);
6452 /* All seems OK, enable tracing */
6453 tracing_disabled = 0;
6455 atomic_notifier_chain_register(&panic_notifier_list,
6456 &trace_panic_notifier);
6458 register_die_notifier(&trace_die_notifier);
6460 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
6462 /* Holder for file callbacks */
6463 global_trace.trace_cpu.cpu = RING_BUFFER_ALL_CPUS;
6464 global_trace.trace_cpu.tr = &global_trace;
6466 INIT_LIST_HEAD(&global_trace.systems);
6467 INIT_LIST_HEAD(&global_trace.events);
6468 list_add(&global_trace.list, &ftrace_trace_arrays);
6470 while (trace_boot_options) {
6473 option = strsep(&trace_boot_options, ",");
6474 trace_set_options(&global_trace, option);
6477 register_snapshot_cmd();
6482 free_percpu(global_trace.trace_buffer.data);
6483 #ifdef CONFIG_TRACER_MAX_TRACE
6484 free_percpu(global_trace.max_buffer.data);
6486 free_cpumask_var(tracing_cpumask);
6487 out_free_buffer_mask:
6488 free_cpumask_var(tracing_buffer_mask);
6493 __init static int clear_boot_tracer(void)
6496 * The default tracer at boot buffer is an init section.
6497 * This function is called in lateinit. If we did not
6498 * find the boot tracer, then clear it out, to prevent
6499 * later registration from accessing the buffer that is
6500 * about to be freed.
6502 if (!default_bootup_tracer)
6505 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
6506 default_bootup_tracer);
6507 default_bootup_tracer = NULL;
6512 early_initcall(tracer_alloc_buffers);
6513 fs_initcall(tracer_init_debugfs);
6514 late_initcall(clear_boot_tracer);