2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 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/irq_work.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/rwsem.h>
36 #include <linux/slab.h>
37 #include <linux/ctype.h>
38 #include <linux/init.h>
39 #include <linux/poll.h>
40 #include <linux/nmi.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 int 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 * When a reader is waiting for data, then this variable is
92 static bool trace_wakeup_needed;
94 static struct irq_work trace_work_wakeup;
97 * Kill all tracing for good (never come back).
98 * It is initialized to 1 but will turn to zero if the initialization
99 * of the tracer is successful. But that is the only place that sets
102 static int tracing_disabled = 1;
104 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
106 cpumask_var_t __read_mostly tracing_buffer_mask;
109 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
111 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
112 * is set, then ftrace_dump is called. This will output the contents
113 * of the ftrace buffers to the console. This is very useful for
114 * capturing traces that lead to crashes and outputing it to a
117 * It is default off, but you can enable it with either specifying
118 * "ftrace_dump_on_oops" in the kernel command line, or setting
119 * /proc/sys/kernel/ftrace_dump_on_oops
120 * Set 1 if you want to dump buffers of all CPUs
121 * Set 2 if you want to dump the buffer of the CPU that triggered oops
124 enum ftrace_dump_mode ftrace_dump_on_oops;
126 static int tracing_set_tracer(const char *buf);
128 #define MAX_TRACER_SIZE 100
129 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
130 static char *default_bootup_tracer;
132 static int __init set_cmdline_ftrace(char *str)
134 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
135 default_bootup_tracer = bootup_tracer_buf;
136 /* We are using ftrace early, expand it */
137 ring_buffer_expanded = 1;
140 __setup("ftrace=", set_cmdline_ftrace);
142 static int __init set_ftrace_dump_on_oops(char *str)
144 if (*str++ != '=' || !*str) {
145 ftrace_dump_on_oops = DUMP_ALL;
149 if (!strcmp("orig_cpu", str)) {
150 ftrace_dump_on_oops = DUMP_ORIG;
156 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
159 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
160 static char *trace_boot_options __initdata;
162 static int __init set_trace_boot_options(char *str)
164 strncpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
165 trace_boot_options = trace_boot_options_buf;
168 __setup("trace_options=", set_trace_boot_options);
170 unsigned long long ns2usecs(cycle_t nsec)
178 * The global_trace is the descriptor that holds the tracing
179 * buffers for the live tracing. For each CPU, it contains
180 * a link list of pages that will store trace entries. The
181 * page descriptor of the pages in the memory is used to hold
182 * the link list by linking the lru item in the page descriptor
183 * to each of the pages in the buffer per CPU.
185 * For each active CPU there is a data field that holds the
186 * pages for the buffer for that CPU. Each CPU has the same number
187 * of pages allocated for its buffer.
189 static struct trace_array global_trace;
191 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
193 int filter_current_check_discard(struct ring_buffer *buffer,
194 struct ftrace_event_call *call, void *rec,
195 struct ring_buffer_event *event)
197 return filter_check_discard(call, rec, buffer, event);
199 EXPORT_SYMBOL_GPL(filter_current_check_discard);
201 cycle_t ftrace_now(int cpu)
205 /* Early boot up does not have a buffer yet */
206 if (!global_trace.buffer)
207 return trace_clock_local();
209 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
210 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
216 * The max_tr is used to snapshot the global_trace when a maximum
217 * latency is reached. Some tracers will use this to store a maximum
218 * trace while it continues examining live traces.
220 * The buffers for the max_tr are set up the same as the global_trace.
221 * When a snapshot is taken, the link list of the max_tr is swapped
222 * with the link list of the global_trace and the buffers are reset for
223 * the global_trace so the tracing can continue.
225 static struct trace_array max_tr;
227 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
229 int tracing_is_enabled(void)
231 return tracing_is_on();
235 * trace_buf_size is the size in bytes that is allocated
236 * for a buffer. Note, the number of bytes is always rounded
239 * This number is purposely set to a low number of 16384.
240 * If the dump on oops happens, it will be much appreciated
241 * to not have to wait for all that output. Anyway this can be
242 * boot time and run time configurable.
244 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
246 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
248 /* trace_types holds a link list of available tracers. */
249 static struct tracer *trace_types __read_mostly;
251 /* current_trace points to the tracer that is currently active */
252 static struct tracer *current_trace __read_mostly = &nop_trace;
255 * trace_types_lock is used to protect the trace_types list.
257 static DEFINE_MUTEX(trace_types_lock);
260 * serialize the access of the ring buffer
262 * ring buffer serializes readers, but it is low level protection.
263 * The validity of the events (which returns by ring_buffer_peek() ..etc)
264 * are not protected by ring buffer.
266 * The content of events may become garbage if we allow other process consumes
267 * these events concurrently:
268 * A) the page of the consumed events may become a normal page
269 * (not reader page) in ring buffer, and this page will be rewrited
270 * by events producer.
271 * B) The page of the consumed events may become a page for splice_read,
272 * and this page will be returned to system.
274 * These primitives allow multi process access to different cpu ring buffer
277 * These primitives don't distinguish read-only and read-consume access.
278 * Multi read-only access are also serialized.
282 static DECLARE_RWSEM(all_cpu_access_lock);
283 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
285 static inline void trace_access_lock(int cpu)
287 if (cpu == TRACE_PIPE_ALL_CPU) {
288 /* gain it for accessing the whole ring buffer. */
289 down_write(&all_cpu_access_lock);
291 /* gain it for accessing a cpu ring buffer. */
293 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
294 down_read(&all_cpu_access_lock);
296 /* Secondly block other access to this @cpu ring buffer. */
297 mutex_lock(&per_cpu(cpu_access_lock, cpu));
301 static inline void trace_access_unlock(int cpu)
303 if (cpu == TRACE_PIPE_ALL_CPU) {
304 up_write(&all_cpu_access_lock);
306 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
307 up_read(&all_cpu_access_lock);
311 static inline void trace_access_lock_init(void)
315 for_each_possible_cpu(cpu)
316 mutex_init(&per_cpu(cpu_access_lock, cpu));
321 static DEFINE_MUTEX(access_lock);
323 static inline void trace_access_lock(int cpu)
326 mutex_lock(&access_lock);
329 static inline void trace_access_unlock(int cpu)
332 mutex_unlock(&access_lock);
335 static inline void trace_access_lock_init(void)
341 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
342 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
344 /* trace_flags holds trace_options default values */
345 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
346 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
347 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
348 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS;
350 static int trace_stop_count;
351 static DEFINE_RAW_SPINLOCK(tracing_start_lock);
354 * trace_wake_up - wake up tasks waiting for trace input
356 * Schedules a delayed work to wake up any task that is blocked on the
357 * trace_wait queue. These is used with trace_poll for tasks polling the
360 static void trace_wake_up(struct irq_work *work)
362 wake_up_all(&trace_wait);
367 * tracing_on - enable tracing buffers
369 * This function enables tracing buffers that may have been
370 * disabled with tracing_off.
372 void tracing_on(void)
374 if (global_trace.buffer)
375 ring_buffer_record_on(global_trace.buffer);
377 * This flag is only looked at when buffers haven't been
378 * allocated yet. We don't really care about the race
379 * between setting this flag and actually turning
382 global_trace.buffer_disabled = 0;
384 EXPORT_SYMBOL_GPL(tracing_on);
387 * tracing_off - turn off tracing buffers
389 * This function stops the tracing buffers from recording data.
390 * It does not disable any overhead the tracers themselves may
391 * be causing. This function simply causes all recording to
392 * the ring buffers to fail.
394 void tracing_off(void)
396 if (global_trace.buffer)
397 ring_buffer_record_off(global_trace.buffer);
399 * This flag is only looked at when buffers haven't been
400 * allocated yet. We don't really care about the race
401 * between setting this flag and actually turning
404 global_trace.buffer_disabled = 1;
406 EXPORT_SYMBOL_GPL(tracing_off);
409 * tracing_is_on - show state of ring buffers enabled
411 int tracing_is_on(void)
413 if (global_trace.buffer)
414 return ring_buffer_record_is_on(global_trace.buffer);
415 return !global_trace.buffer_disabled;
417 EXPORT_SYMBOL_GPL(tracing_is_on);
419 static int __init set_buf_size(char *str)
421 unsigned long buf_size;
425 buf_size = memparse(str, &str);
426 /* nr_entries can not be zero */
429 trace_buf_size = buf_size;
432 __setup("trace_buf_size=", set_buf_size);
434 static int __init set_tracing_thresh(char *str)
436 unsigned long threshold;
441 ret = kstrtoul(str, 0, &threshold);
444 tracing_thresh = threshold * 1000;
447 __setup("tracing_thresh=", set_tracing_thresh);
449 unsigned long nsecs_to_usecs(unsigned long nsecs)
454 /* These must match the bit postions in trace_iterator_flags */
455 static const char *trace_options[] = {
487 int in_ns; /* is this clock in nanoseconds? */
489 { trace_clock_local, "local", 1 },
490 { trace_clock_global, "global", 1 },
491 { trace_clock_counter, "counter", 0 },
498 * trace_parser_get_init - gets the buffer for trace parser
500 int trace_parser_get_init(struct trace_parser *parser, int size)
502 memset(parser, 0, sizeof(*parser));
504 parser->buffer = kmalloc(size, GFP_KERNEL);
513 * trace_parser_put - frees the buffer for trace parser
515 void trace_parser_put(struct trace_parser *parser)
517 kfree(parser->buffer);
521 * trace_get_user - reads the user input string separated by space
522 * (matched by isspace(ch))
524 * For each string found the 'struct trace_parser' is updated,
525 * and the function returns.
527 * Returns number of bytes read.
529 * See kernel/trace/trace.h for 'struct trace_parser' details.
531 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
532 size_t cnt, loff_t *ppos)
539 trace_parser_clear(parser);
541 ret = get_user(ch, ubuf++);
549 * The parser is not finished with the last write,
550 * continue reading the user input without skipping spaces.
553 /* skip white space */
554 while (cnt && isspace(ch)) {
555 ret = get_user(ch, ubuf++);
562 /* only spaces were written */
572 /* read the non-space input */
573 while (cnt && !isspace(ch)) {
574 if (parser->idx < parser->size - 1)
575 parser->buffer[parser->idx++] = ch;
580 ret = get_user(ch, ubuf++);
587 /* We either got finished input or we have to wait for another call. */
589 parser->buffer[parser->idx] = 0;
590 parser->cont = false;
593 parser->buffer[parser->idx++] = ch;
603 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
611 if (s->len <= s->readpos)
614 len = s->len - s->readpos;
617 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
627 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
631 if (s->len <= s->readpos)
634 len = s->len - s->readpos;
637 memcpy(buf, s->buffer + s->readpos, cnt);
644 * ftrace_max_lock is used to protect the swapping of buffers
645 * when taking a max snapshot. The buffers themselves are
646 * protected by per_cpu spinlocks. But the action of the swap
647 * needs its own lock.
649 * This is defined as a arch_spinlock_t in order to help
650 * with performance when lockdep debugging is enabled.
652 * It is also used in other places outside the update_max_tr
653 * so it needs to be defined outside of the
654 * CONFIG_TRACER_MAX_TRACE.
656 static arch_spinlock_t ftrace_max_lock =
657 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
659 unsigned long __read_mostly tracing_thresh;
661 #ifdef CONFIG_TRACER_MAX_TRACE
662 unsigned long __read_mostly tracing_max_latency;
665 * Copy the new maximum trace into the separate maximum-trace
666 * structure. (this way the maximum trace is permanently saved,
667 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
670 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
672 struct trace_array_cpu *data = tr->data[cpu];
673 struct trace_array_cpu *max_data;
676 max_tr.time_start = data->preempt_timestamp;
678 max_data = max_tr.data[cpu];
679 max_data->saved_latency = tracing_max_latency;
680 max_data->critical_start = data->critical_start;
681 max_data->critical_end = data->critical_end;
683 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
684 max_data->pid = tsk->pid;
685 max_data->uid = task_uid(tsk);
686 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
687 max_data->policy = tsk->policy;
688 max_data->rt_priority = tsk->rt_priority;
690 /* record this tasks comm */
691 tracing_record_cmdline(tsk);
695 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
697 * @tsk: the task with the latency
698 * @cpu: The cpu that initiated the trace.
700 * Flip the buffers between the @tr and the max_tr and record information
701 * about which task was the cause of this latency.
704 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
706 struct ring_buffer *buf = tr->buffer;
708 if (trace_stop_count)
711 WARN_ON_ONCE(!irqs_disabled());
713 if (!current_trace->allocated_snapshot) {
714 /* Only the nop tracer should hit this when disabling */
715 WARN_ON_ONCE(current_trace != &nop_trace);
719 arch_spin_lock(&ftrace_max_lock);
721 tr->buffer = max_tr.buffer;
724 __update_max_tr(tr, tsk, cpu);
725 arch_spin_unlock(&ftrace_max_lock);
729 * update_max_tr_single - only copy one trace over, and reset the rest
731 * @tsk - task with the latency
732 * @cpu - the cpu of the buffer to copy.
734 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
737 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
741 if (trace_stop_count)
744 WARN_ON_ONCE(!irqs_disabled());
745 if (WARN_ON_ONCE(!current_trace->allocated_snapshot))
748 arch_spin_lock(&ftrace_max_lock);
750 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
754 * We failed to swap the buffer due to a commit taking
755 * place on this CPU. We fail to record, but we reset
756 * the max trace buffer (no one writes directly to it)
757 * and flag that it failed.
759 trace_array_printk(&max_tr, _THIS_IP_,
760 "Failed to swap buffers due to commit in progress\n");
763 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
765 __update_max_tr(tr, tsk, cpu);
766 arch_spin_unlock(&ftrace_max_lock);
768 #endif /* CONFIG_TRACER_MAX_TRACE */
770 static void default_wait_pipe(struct trace_iterator *iter)
774 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
777 * The events can happen in critical sections where
778 * checking a work queue can cause deadlocks.
779 * After adding a task to the queue, this flag is set
780 * only to notify events to try to wake up the queue
783 * We don't clear it even if the buffer is no longer
784 * empty. The flag only causes the next event to run
785 * irq_work to do the work queue wake up. The worse
786 * that can happen if we race with !trace_empty() is that
787 * an event will cause an irq_work to try to wake up
790 * There's no reason to protect this flag either, as
791 * the work queue and irq_work logic will do the necessary
792 * synchronization for the wake ups. The only thing
793 * that is necessary is that the wake up happens after
794 * a task has been queued. It's OK for spurious wake ups.
796 trace_wakeup_needed = true;
798 if (trace_empty(iter))
801 finish_wait(&trace_wait, &wait);
805 * register_tracer - register a tracer with the ftrace system.
806 * @type - the plugin for the tracer
808 * Register a new plugin tracer.
810 int register_tracer(struct tracer *type)
816 pr_info("Tracer must have a name\n");
820 if (strlen(type->name) >= MAX_TRACER_SIZE) {
821 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
825 mutex_lock(&trace_types_lock);
827 tracing_selftest_running = true;
829 for (t = trace_types; t; t = t->next) {
830 if (strcmp(type->name, t->name) == 0) {
832 pr_info("Tracer %s already registered\n",
840 type->set_flag = &dummy_set_flag;
842 type->flags = &dummy_tracer_flags;
844 if (!type->flags->opts)
845 type->flags->opts = dummy_tracer_opt;
846 if (!type->wait_pipe)
847 type->wait_pipe = default_wait_pipe;
850 #ifdef CONFIG_FTRACE_STARTUP_TEST
851 if (type->selftest && !tracing_selftest_disabled) {
852 struct tracer *saved_tracer = current_trace;
853 struct trace_array *tr = &global_trace;
856 * Run a selftest on this tracer.
857 * Here we reset the trace buffer, and set the current
858 * tracer to be this tracer. The tracer can then run some
859 * internal tracing to verify that everything is in order.
860 * If we fail, we do not register this tracer.
862 tracing_reset_online_cpus(tr);
864 current_trace = type;
866 if (type->use_max_tr) {
867 /* If we expanded the buffers, make sure the max is expanded too */
868 if (ring_buffer_expanded)
869 ring_buffer_resize(max_tr.buffer, trace_buf_size,
870 RING_BUFFER_ALL_CPUS);
871 type->allocated_snapshot = true;
874 /* the test is responsible for initializing and enabling */
875 pr_info("Testing tracer %s: ", type->name);
876 ret = type->selftest(type, tr);
877 /* the test is responsible for resetting too */
878 current_trace = saved_tracer;
880 printk(KERN_CONT "FAILED!\n");
881 /* Add the warning after printing 'FAILED' */
885 /* Only reset on passing, to avoid touching corrupted buffers */
886 tracing_reset_online_cpus(tr);
888 if (type->use_max_tr) {
889 type->allocated_snapshot = false;
891 /* Shrink the max buffer again */
892 if (ring_buffer_expanded)
893 ring_buffer_resize(max_tr.buffer, 1,
894 RING_BUFFER_ALL_CPUS);
897 printk(KERN_CONT "PASSED\n");
901 type->next = trace_types;
905 tracing_selftest_running = false;
906 mutex_unlock(&trace_types_lock);
908 if (ret || !default_bootup_tracer)
911 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
914 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
915 /* Do we want this tracer to start on bootup? */
916 tracing_set_tracer(type->name);
917 default_bootup_tracer = NULL;
918 /* disable other selftests, since this will break it. */
919 tracing_selftest_disabled = 1;
920 #ifdef CONFIG_FTRACE_STARTUP_TEST
921 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
929 void tracing_reset(struct trace_array *tr, int cpu)
931 struct ring_buffer *buffer = tr->buffer;
936 ring_buffer_record_disable(buffer);
938 /* Make sure all commits have finished */
940 ring_buffer_reset_cpu(buffer, cpu);
942 ring_buffer_record_enable(buffer);
945 void tracing_reset_online_cpus(struct trace_array *tr)
947 struct ring_buffer *buffer = tr->buffer;
953 ring_buffer_record_disable(buffer);
955 /* Make sure all commits have finished */
958 tr->time_start = ftrace_now(tr->cpu);
960 for_each_online_cpu(cpu)
961 ring_buffer_reset_cpu(buffer, cpu);
963 ring_buffer_record_enable(buffer);
966 void tracing_reset_current(int cpu)
968 tracing_reset(&global_trace, cpu);
971 void tracing_reset_current_online_cpus(void)
973 tracing_reset_online_cpus(&global_trace);
976 #define SAVED_CMDLINES 128
977 #define NO_CMDLINE_MAP UINT_MAX
978 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
979 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
980 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
981 static int cmdline_idx;
982 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
984 /* temporary disable recording */
985 static atomic_t trace_record_cmdline_disabled __read_mostly;
987 static void trace_init_cmdlines(void)
989 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
990 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
994 int is_tracing_stopped(void)
996 return trace_stop_count;
1000 * ftrace_off_permanent - disable all ftrace code permanently
1002 * This should only be called when a serious anomally has
1003 * been detected. This will turn off the function tracing,
1004 * ring buffers, and other tracing utilites. It takes no
1005 * locks and can be called from any context.
1007 void ftrace_off_permanent(void)
1009 tracing_disabled = 1;
1011 tracing_off_permanent();
1015 * tracing_start - quick start of the tracer
1017 * If tracing is enabled but was stopped by tracing_stop,
1018 * this will start the tracer back up.
1020 void tracing_start(void)
1022 struct ring_buffer *buffer;
1023 unsigned long flags;
1025 if (tracing_disabled)
1028 raw_spin_lock_irqsave(&tracing_start_lock, flags);
1029 if (--trace_stop_count) {
1030 if (trace_stop_count < 0) {
1031 /* Someone screwed up their debugging */
1033 trace_stop_count = 0;
1038 /* Prevent the buffers from switching */
1039 arch_spin_lock(&ftrace_max_lock);
1041 buffer = global_trace.buffer;
1043 ring_buffer_record_enable(buffer);
1045 buffer = max_tr.buffer;
1047 ring_buffer_record_enable(buffer);
1049 arch_spin_unlock(&ftrace_max_lock);
1053 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1057 * tracing_stop - quick stop of the tracer
1059 * Light weight way to stop tracing. Use in conjunction with
1062 void tracing_stop(void)
1064 struct ring_buffer *buffer;
1065 unsigned long flags;
1068 raw_spin_lock_irqsave(&tracing_start_lock, flags);
1069 if (trace_stop_count++)
1072 /* Prevent the buffers from switching */
1073 arch_spin_lock(&ftrace_max_lock);
1075 buffer = global_trace.buffer;
1077 ring_buffer_record_disable(buffer);
1079 buffer = max_tr.buffer;
1081 ring_buffer_record_disable(buffer);
1083 arch_spin_unlock(&ftrace_max_lock);
1086 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1089 void trace_stop_cmdline_recording(void);
1091 static void trace_save_cmdline(struct task_struct *tsk)
1095 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1099 * It's not the end of the world if we don't get
1100 * the lock, but we also don't want to spin
1101 * nor do we want to disable interrupts,
1102 * so if we miss here, then better luck next time.
1104 if (!arch_spin_trylock(&trace_cmdline_lock))
1107 idx = map_pid_to_cmdline[tsk->pid];
1108 if (idx == NO_CMDLINE_MAP) {
1109 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1112 * Check whether the cmdline buffer at idx has a pid
1113 * mapped. We are going to overwrite that entry so we
1114 * need to clear the map_pid_to_cmdline. Otherwise we
1115 * would read the new comm for the old pid.
1117 pid = map_cmdline_to_pid[idx];
1118 if (pid != NO_CMDLINE_MAP)
1119 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1121 map_cmdline_to_pid[idx] = tsk->pid;
1122 map_pid_to_cmdline[tsk->pid] = idx;
1127 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1129 arch_spin_unlock(&trace_cmdline_lock);
1132 void trace_find_cmdline(int pid, char comm[])
1137 strcpy(comm, "<idle>");
1141 if (WARN_ON_ONCE(pid < 0)) {
1142 strcpy(comm, "<XXX>");
1146 if (pid > PID_MAX_DEFAULT) {
1147 strcpy(comm, "<...>");
1152 arch_spin_lock(&trace_cmdline_lock);
1153 map = map_pid_to_cmdline[pid];
1154 if (map != NO_CMDLINE_MAP)
1155 strcpy(comm, saved_cmdlines[map]);
1157 strcpy(comm, "<...>");
1159 arch_spin_unlock(&trace_cmdline_lock);
1163 void tracing_record_cmdline(struct task_struct *tsk)
1165 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1168 if (!__this_cpu_read(trace_cmdline_save))
1171 __this_cpu_write(trace_cmdline_save, false);
1173 trace_save_cmdline(tsk);
1177 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1180 struct task_struct *tsk = current;
1182 entry->preempt_count = pc & 0xff;
1183 entry->pid = (tsk) ? tsk->pid : 0;
1185 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1186 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1188 TRACE_FLAG_IRQS_NOSUPPORT |
1190 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1191 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1192 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1194 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1196 struct ring_buffer_event *
1197 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1200 unsigned long flags, int pc)
1202 struct ring_buffer_event *event;
1204 event = ring_buffer_lock_reserve(buffer, len);
1205 if (event != NULL) {
1206 struct trace_entry *ent = ring_buffer_event_data(event);
1208 tracing_generic_entry_update(ent, flags, pc);
1216 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1218 __this_cpu_write(trace_cmdline_save, true);
1219 if (trace_wakeup_needed) {
1220 trace_wakeup_needed = false;
1221 /* irq_work_queue() supplies it's own memory barriers */
1222 irq_work_queue(&trace_work_wakeup);
1224 ring_buffer_unlock_commit(buffer, event);
1228 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1229 struct ring_buffer_event *event,
1230 unsigned long flags, int pc)
1232 __buffer_unlock_commit(buffer, event);
1234 ftrace_trace_stack(buffer, flags, 6, pc);
1235 ftrace_trace_userstack(buffer, flags, pc);
1238 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1239 struct ring_buffer_event *event,
1240 unsigned long flags, int pc)
1242 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1244 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1246 struct ring_buffer_event *
1247 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1248 int type, unsigned long len,
1249 unsigned long flags, int pc)
1251 *current_rb = global_trace.buffer;
1252 return trace_buffer_lock_reserve(*current_rb,
1253 type, len, flags, pc);
1255 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1257 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1258 struct ring_buffer_event *event,
1259 unsigned long flags, int pc)
1261 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1263 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1265 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1266 struct ring_buffer_event *event,
1267 unsigned long flags, int pc,
1268 struct pt_regs *regs)
1270 __buffer_unlock_commit(buffer, event);
1272 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1273 ftrace_trace_userstack(buffer, flags, pc);
1275 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1277 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1278 struct ring_buffer_event *event)
1280 ring_buffer_discard_commit(buffer, event);
1282 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1285 trace_function(struct trace_array *tr,
1286 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1289 struct ftrace_event_call *call = &event_function;
1290 struct ring_buffer *buffer = tr->buffer;
1291 struct ring_buffer_event *event;
1292 struct ftrace_entry *entry;
1294 /* If we are reading the ring buffer, don't trace */
1295 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1298 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1302 entry = ring_buffer_event_data(event);
1304 entry->parent_ip = parent_ip;
1306 if (!filter_check_discard(call, entry, buffer, event))
1307 __buffer_unlock_commit(buffer, event);
1311 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1312 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1315 if (likely(!atomic_read(&data->disabled)))
1316 trace_function(tr, ip, parent_ip, flags, pc);
1319 #ifdef CONFIG_STACKTRACE
1321 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1322 struct ftrace_stack {
1323 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1326 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1327 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1329 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1330 unsigned long flags,
1331 int skip, int pc, struct pt_regs *regs)
1333 struct ftrace_event_call *call = &event_kernel_stack;
1334 struct ring_buffer_event *event;
1335 struct stack_entry *entry;
1336 struct stack_trace trace;
1338 int size = FTRACE_STACK_ENTRIES;
1340 trace.nr_entries = 0;
1344 * Since events can happen in NMIs there's no safe way to
1345 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1346 * or NMI comes in, it will just have to use the default
1347 * FTRACE_STACK_SIZE.
1349 preempt_disable_notrace();
1351 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1353 * We don't need any atomic variables, just a barrier.
1354 * If an interrupt comes in, we don't care, because it would
1355 * have exited and put the counter back to what we want.
1356 * We just need a barrier to keep gcc from moving things
1360 if (use_stack == 1) {
1361 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1362 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1365 save_stack_trace_regs(regs, &trace);
1367 save_stack_trace(&trace);
1369 if (trace.nr_entries > size)
1370 size = trace.nr_entries;
1372 /* From now on, use_stack is a boolean */
1375 size *= sizeof(unsigned long);
1377 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1378 sizeof(*entry) + size, flags, pc);
1381 entry = ring_buffer_event_data(event);
1383 memset(&entry->caller, 0, size);
1386 memcpy(&entry->caller, trace.entries,
1387 trace.nr_entries * sizeof(unsigned long));
1389 trace.max_entries = FTRACE_STACK_ENTRIES;
1390 trace.entries = entry->caller;
1392 save_stack_trace_regs(regs, &trace);
1394 save_stack_trace(&trace);
1397 entry->size = trace.nr_entries;
1399 if (!filter_check_discard(call, entry, buffer, event))
1400 __buffer_unlock_commit(buffer, event);
1403 /* Again, don't let gcc optimize things here */
1405 __this_cpu_dec(ftrace_stack_reserve);
1406 preempt_enable_notrace();
1410 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1411 int skip, int pc, struct pt_regs *regs)
1413 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1416 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1419 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1422 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1425 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1428 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1431 __ftrace_trace_stack(tr->buffer, flags, skip, pc, NULL);
1435 * trace_dump_stack - record a stack back trace in the trace buffer
1437 void trace_dump_stack(void)
1439 unsigned long flags;
1441 if (tracing_disabled || tracing_selftest_running)
1444 local_save_flags(flags);
1446 /* skipping 3 traces, seems to get us at the caller of this function */
1447 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count(), NULL);
1450 static DEFINE_PER_CPU(int, user_stack_count);
1453 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1455 struct ftrace_event_call *call = &event_user_stack;
1456 struct ring_buffer_event *event;
1457 struct userstack_entry *entry;
1458 struct stack_trace trace;
1460 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1464 * NMIs can not handle page faults, even with fix ups.
1465 * The save user stack can (and often does) fault.
1467 if (unlikely(in_nmi()))
1471 * prevent recursion, since the user stack tracing may
1472 * trigger other kernel events.
1475 if (__this_cpu_read(user_stack_count))
1478 __this_cpu_inc(user_stack_count);
1480 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1481 sizeof(*entry), flags, pc);
1483 goto out_drop_count;
1484 entry = ring_buffer_event_data(event);
1486 entry->tgid = current->tgid;
1487 memset(&entry->caller, 0, sizeof(entry->caller));
1489 trace.nr_entries = 0;
1490 trace.max_entries = FTRACE_STACK_ENTRIES;
1492 trace.entries = entry->caller;
1494 save_stack_trace_user(&trace);
1495 if (!filter_check_discard(call, entry, buffer, event))
1496 __buffer_unlock_commit(buffer, event);
1499 __this_cpu_dec(user_stack_count);
1505 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1507 ftrace_trace_userstack(tr, flags, preempt_count());
1511 #endif /* CONFIG_STACKTRACE */
1513 /* created for use with alloc_percpu */
1514 struct trace_buffer_struct {
1515 char buffer[TRACE_BUF_SIZE];
1518 static struct trace_buffer_struct *trace_percpu_buffer;
1519 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1520 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1521 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1524 * The buffer used is dependent on the context. There is a per cpu
1525 * buffer for normal context, softirq contex, hard irq context and
1526 * for NMI context. Thise allows for lockless recording.
1528 * Note, if the buffers failed to be allocated, then this returns NULL
1530 static char *get_trace_buf(void)
1532 struct trace_buffer_struct *percpu_buffer;
1535 * If we have allocated per cpu buffers, then we do not
1536 * need to do any locking.
1539 percpu_buffer = trace_percpu_nmi_buffer;
1541 percpu_buffer = trace_percpu_irq_buffer;
1542 else if (in_softirq())
1543 percpu_buffer = trace_percpu_sirq_buffer;
1545 percpu_buffer = trace_percpu_buffer;
1550 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1553 static int alloc_percpu_trace_buffer(void)
1555 struct trace_buffer_struct *buffers;
1556 struct trace_buffer_struct *sirq_buffers;
1557 struct trace_buffer_struct *irq_buffers;
1558 struct trace_buffer_struct *nmi_buffers;
1560 buffers = alloc_percpu(struct trace_buffer_struct);
1564 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1568 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1572 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1576 trace_percpu_buffer = buffers;
1577 trace_percpu_sirq_buffer = sirq_buffers;
1578 trace_percpu_irq_buffer = irq_buffers;
1579 trace_percpu_nmi_buffer = nmi_buffers;
1584 free_percpu(irq_buffers);
1586 free_percpu(sirq_buffers);
1588 free_percpu(buffers);
1590 WARN(1, "Could not allocate percpu trace_printk buffer");
1594 static int buffers_allocated;
1596 void trace_printk_init_buffers(void)
1598 if (buffers_allocated)
1601 if (alloc_percpu_trace_buffer())
1604 pr_info("ftrace: Allocated trace_printk buffers\n");
1606 /* Expand the buffers to set size */
1607 tracing_update_buffers();
1609 buffers_allocated = 1;
1612 * trace_printk_init_buffers() can be called by modules.
1613 * If that happens, then we need to start cmdline recording
1614 * directly here. If the global_trace.buffer is already
1615 * allocated here, then this was called by module code.
1617 if (global_trace.buffer)
1618 tracing_start_cmdline_record();
1621 void trace_printk_start_comm(void)
1623 /* Start tracing comms if trace printk is set */
1624 if (!buffers_allocated)
1626 tracing_start_cmdline_record();
1629 static void trace_printk_start_stop_comm(int enabled)
1631 if (!buffers_allocated)
1635 tracing_start_cmdline_record();
1637 tracing_stop_cmdline_record();
1641 * trace_vbprintk - write binary msg to tracing buffer
1644 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1646 struct ftrace_event_call *call = &event_bprint;
1647 struct ring_buffer_event *event;
1648 struct ring_buffer *buffer;
1649 struct trace_array *tr = &global_trace;
1650 struct bprint_entry *entry;
1651 unsigned long flags;
1653 int len = 0, size, pc;
1655 if (unlikely(tracing_selftest_running || tracing_disabled))
1658 /* Don't pollute graph traces with trace_vprintk internals */
1659 pause_graph_tracing();
1661 pc = preempt_count();
1662 preempt_disable_notrace();
1664 tbuffer = get_trace_buf();
1670 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1672 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
1675 local_save_flags(flags);
1676 size = sizeof(*entry) + sizeof(u32) * len;
1677 buffer = tr->buffer;
1678 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1682 entry = ring_buffer_event_data(event);
1686 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
1687 if (!filter_check_discard(call, entry, buffer, event)) {
1688 __buffer_unlock_commit(buffer, event);
1689 ftrace_trace_stack(buffer, flags, 6, pc);
1693 preempt_enable_notrace();
1694 unpause_graph_tracing();
1698 EXPORT_SYMBOL_GPL(trace_vbprintk);
1700 int trace_array_printk(struct trace_array *tr,
1701 unsigned long ip, const char *fmt, ...)
1706 if (!(trace_flags & TRACE_ITER_PRINTK))
1710 ret = trace_array_vprintk(tr, ip, fmt, ap);
1715 int trace_array_vprintk(struct trace_array *tr,
1716 unsigned long ip, const char *fmt, va_list args)
1718 struct ftrace_event_call *call = &event_print;
1719 struct ring_buffer_event *event;
1720 struct ring_buffer *buffer;
1721 int len = 0, size, pc;
1722 struct print_entry *entry;
1723 unsigned long flags;
1726 if (tracing_disabled || tracing_selftest_running)
1729 /* Don't pollute graph traces with trace_vprintk internals */
1730 pause_graph_tracing();
1732 pc = preempt_count();
1733 preempt_disable_notrace();
1736 tbuffer = get_trace_buf();
1742 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
1743 if (len > TRACE_BUF_SIZE)
1746 local_save_flags(flags);
1747 size = sizeof(*entry) + len + 1;
1748 buffer = tr->buffer;
1749 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1753 entry = ring_buffer_event_data(event);
1756 memcpy(&entry->buf, tbuffer, len);
1757 entry->buf[len] = '\0';
1758 if (!filter_check_discard(call, entry, buffer, event)) {
1759 __buffer_unlock_commit(buffer, event);
1760 ftrace_trace_stack(buffer, flags, 6, pc);
1763 preempt_enable_notrace();
1764 unpause_graph_tracing();
1769 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1771 return trace_array_vprintk(&global_trace, ip, fmt, args);
1773 EXPORT_SYMBOL_GPL(trace_vprintk);
1775 static void trace_iterator_increment(struct trace_iterator *iter)
1777 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
1781 ring_buffer_read(buf_iter, NULL);
1784 static struct trace_entry *
1785 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1786 unsigned long *lost_events)
1788 struct ring_buffer_event *event;
1789 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
1792 event = ring_buffer_iter_peek(buf_iter, ts);
1794 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1798 iter->ent_size = ring_buffer_event_length(event);
1799 return ring_buffer_event_data(event);
1805 static struct trace_entry *
1806 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1807 unsigned long *missing_events, u64 *ent_ts)
1809 struct ring_buffer *buffer = iter->tr->buffer;
1810 struct trace_entry *ent, *next = NULL;
1811 unsigned long lost_events = 0, next_lost = 0;
1812 int cpu_file = iter->cpu_file;
1813 u64 next_ts = 0, ts;
1819 * If we are in a per_cpu trace file, don't bother by iterating over
1820 * all cpu and peek directly.
1822 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1823 if (ring_buffer_empty_cpu(buffer, cpu_file))
1825 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1827 *ent_cpu = cpu_file;
1832 for_each_tracing_cpu(cpu) {
1834 if (ring_buffer_empty_cpu(buffer, cpu))
1837 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1840 * Pick the entry with the smallest timestamp:
1842 if (ent && (!next || ts < next_ts)) {
1846 next_lost = lost_events;
1847 next_size = iter->ent_size;
1851 iter->ent_size = next_size;
1854 *ent_cpu = next_cpu;
1860 *missing_events = next_lost;
1865 /* Find the next real entry, without updating the iterator itself */
1866 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1867 int *ent_cpu, u64 *ent_ts)
1869 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1872 /* Find the next real entry, and increment the iterator to the next entry */
1873 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1875 iter->ent = __find_next_entry(iter, &iter->cpu,
1876 &iter->lost_events, &iter->ts);
1879 trace_iterator_increment(iter);
1881 return iter->ent ? iter : NULL;
1884 static void trace_consume(struct trace_iterator *iter)
1886 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1887 &iter->lost_events);
1890 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1892 struct trace_iterator *iter = m->private;
1896 WARN_ON_ONCE(iter->leftover);
1900 /* can't go backwards */
1905 ent = trace_find_next_entry_inc(iter);
1909 while (ent && iter->idx < i)
1910 ent = trace_find_next_entry_inc(iter);
1917 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1919 struct trace_array *tr = iter->tr;
1920 struct ring_buffer_event *event;
1921 struct ring_buffer_iter *buf_iter;
1922 unsigned long entries = 0;
1925 tr->data[cpu]->skipped_entries = 0;
1927 buf_iter = trace_buffer_iter(iter, cpu);
1931 ring_buffer_iter_reset(buf_iter);
1934 * We could have the case with the max latency tracers
1935 * that a reset never took place on a cpu. This is evident
1936 * by the timestamp being before the start of the buffer.
1938 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1939 if (ts >= iter->tr->time_start)
1942 ring_buffer_read(buf_iter, NULL);
1945 tr->data[cpu]->skipped_entries = entries;
1949 * The current tracer is copied to avoid a global locking
1952 static void *s_start(struct seq_file *m, loff_t *pos)
1954 struct trace_iterator *iter = m->private;
1955 int cpu_file = iter->cpu_file;
1961 * copy the tracer to avoid using a global lock all around.
1962 * iter->trace is a copy of current_trace, the pointer to the
1963 * name may be used instead of a strcmp(), as iter->trace->name
1964 * will point to the same string as current_trace->name.
1966 mutex_lock(&trace_types_lock);
1967 if (unlikely(current_trace && iter->trace->name != current_trace->name))
1968 *iter->trace = *current_trace;
1969 mutex_unlock(&trace_types_lock);
1971 if (iter->snapshot && iter->trace->use_max_tr)
1972 return ERR_PTR(-EBUSY);
1974 if (!iter->snapshot)
1975 atomic_inc(&trace_record_cmdline_disabled);
1977 if (*pos != iter->pos) {
1982 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1983 for_each_tracing_cpu(cpu)
1984 tracing_iter_reset(iter, cpu);
1986 tracing_iter_reset(iter, cpu_file);
1989 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1994 * If we overflowed the seq_file before, then we want
1995 * to just reuse the trace_seq buffer again.
2001 p = s_next(m, p, &l);
2005 trace_event_read_lock();
2006 trace_access_lock(cpu_file);
2010 static void s_stop(struct seq_file *m, void *p)
2012 struct trace_iterator *iter = m->private;
2014 if (iter->snapshot && iter->trace->use_max_tr)
2017 if (!iter->snapshot)
2018 atomic_dec(&trace_record_cmdline_disabled);
2019 trace_access_unlock(iter->cpu_file);
2020 trace_event_read_unlock();
2024 get_total_entries(struct trace_array *tr, unsigned long *total, unsigned long *entries)
2026 unsigned long count;
2032 for_each_tracing_cpu(cpu) {
2033 count = ring_buffer_entries_cpu(tr->buffer, cpu);
2035 * If this buffer has skipped entries, then we hold all
2036 * entries for the trace and we need to ignore the
2037 * ones before the time stamp.
2039 if (tr->data[cpu]->skipped_entries) {
2040 count -= tr->data[cpu]->skipped_entries;
2041 /* total is the same as the entries */
2045 ring_buffer_overrun_cpu(tr->buffer, cpu);
2050 static void print_lat_help_header(struct seq_file *m)
2052 seq_puts(m, "# _------=> CPU# \n");
2053 seq_puts(m, "# / _-----=> irqs-off \n");
2054 seq_puts(m, "# | / _----=> need-resched \n");
2055 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2056 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2057 seq_puts(m, "# |||| / delay \n");
2058 seq_puts(m, "# cmd pid ||||| time | caller \n");
2059 seq_puts(m, "# \\ / ||||| \\ | / \n");
2062 static void print_event_info(struct trace_array *tr, struct seq_file *m)
2064 unsigned long total;
2065 unsigned long entries;
2067 get_total_entries(tr, &total, &entries);
2068 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2069 entries, total, num_online_cpus());
2073 static void print_func_help_header(struct trace_array *tr, struct seq_file *m)
2075 print_event_info(tr, m);
2076 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2077 seq_puts(m, "# | | | | |\n");
2080 static void print_func_help_header_irq(struct trace_array *tr, struct seq_file *m)
2082 print_event_info(tr, m);
2083 seq_puts(m, "# _-----=> irqs-off\n");
2084 seq_puts(m, "# / _----=> need-resched\n");
2085 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2086 seq_puts(m, "# || / _--=> preempt-depth\n");
2087 seq_puts(m, "# ||| / delay\n");
2088 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2089 seq_puts(m, "# | | | |||| | |\n");
2093 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2095 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2096 struct trace_array *tr = iter->tr;
2097 struct trace_array_cpu *data = tr->data[tr->cpu];
2098 struct tracer *type = current_trace;
2099 unsigned long entries;
2100 unsigned long total;
2101 const char *name = "preemption";
2105 get_total_entries(tr, &total, &entries);
2107 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2109 seq_puts(m, "# -----------------------------------"
2110 "---------------------------------\n");
2111 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2112 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2113 nsecs_to_usecs(data->saved_latency),
2117 #if defined(CONFIG_PREEMPT_NONE)
2119 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2121 #elif defined(CONFIG_PREEMPT)
2126 /* These are reserved for later use */
2129 seq_printf(m, " #P:%d)\n", num_online_cpus());
2133 seq_puts(m, "# -----------------\n");
2134 seq_printf(m, "# | task: %.16s-%d "
2135 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2136 data->comm, data->pid,
2137 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2138 data->policy, data->rt_priority);
2139 seq_puts(m, "# -----------------\n");
2141 if (data->critical_start) {
2142 seq_puts(m, "# => started at: ");
2143 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2144 trace_print_seq(m, &iter->seq);
2145 seq_puts(m, "\n# => ended at: ");
2146 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2147 trace_print_seq(m, &iter->seq);
2148 seq_puts(m, "\n#\n");
2154 static void test_cpu_buff_start(struct trace_iterator *iter)
2156 struct trace_seq *s = &iter->seq;
2158 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2161 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2164 if (cpumask_test_cpu(iter->cpu, iter->started))
2167 if (iter->tr->data[iter->cpu]->skipped_entries)
2170 cpumask_set_cpu(iter->cpu, iter->started);
2172 /* Don't print started cpu buffer for the first entry of the trace */
2174 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2178 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2180 struct trace_seq *s = &iter->seq;
2181 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2182 struct trace_entry *entry;
2183 struct trace_event *event;
2187 test_cpu_buff_start(iter);
2189 event = ftrace_find_event(entry->type);
2191 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2192 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2193 if (!trace_print_lat_context(iter))
2196 if (!trace_print_context(iter))
2202 return event->funcs->trace(iter, sym_flags, event);
2204 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2207 return TRACE_TYPE_HANDLED;
2209 return TRACE_TYPE_PARTIAL_LINE;
2212 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2214 struct trace_seq *s = &iter->seq;
2215 struct trace_entry *entry;
2216 struct trace_event *event;
2220 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2221 if (!trace_seq_printf(s, "%d %d %llu ",
2222 entry->pid, iter->cpu, iter->ts))
2226 event = ftrace_find_event(entry->type);
2228 return event->funcs->raw(iter, 0, event);
2230 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2233 return TRACE_TYPE_HANDLED;
2235 return TRACE_TYPE_PARTIAL_LINE;
2238 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2240 struct trace_seq *s = &iter->seq;
2241 unsigned char newline = '\n';
2242 struct trace_entry *entry;
2243 struct trace_event *event;
2247 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2248 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2249 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2250 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2253 event = ftrace_find_event(entry->type);
2255 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2256 if (ret != TRACE_TYPE_HANDLED)
2260 SEQ_PUT_FIELD_RET(s, newline);
2262 return TRACE_TYPE_HANDLED;
2265 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2267 struct trace_seq *s = &iter->seq;
2268 struct trace_entry *entry;
2269 struct trace_event *event;
2273 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2274 SEQ_PUT_FIELD_RET(s, entry->pid);
2275 SEQ_PUT_FIELD_RET(s, iter->cpu);
2276 SEQ_PUT_FIELD_RET(s, iter->ts);
2279 event = ftrace_find_event(entry->type);
2280 return event ? event->funcs->binary(iter, 0, event) :
2284 int trace_empty(struct trace_iterator *iter)
2286 struct ring_buffer_iter *buf_iter;
2289 /* If we are looking at one CPU buffer, only check that one */
2290 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2291 cpu = iter->cpu_file;
2292 buf_iter = trace_buffer_iter(iter, cpu);
2294 if (!ring_buffer_iter_empty(buf_iter))
2297 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2303 for_each_tracing_cpu(cpu) {
2304 buf_iter = trace_buffer_iter(iter, cpu);
2306 if (!ring_buffer_iter_empty(buf_iter))
2309 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2317 /* Called with trace_event_read_lock() held. */
2318 enum print_line_t print_trace_line(struct trace_iterator *iter)
2320 enum print_line_t ret;
2322 if (iter->lost_events &&
2323 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2324 iter->cpu, iter->lost_events))
2325 return TRACE_TYPE_PARTIAL_LINE;
2327 if (iter->trace && iter->trace->print_line) {
2328 ret = iter->trace->print_line(iter);
2329 if (ret != TRACE_TYPE_UNHANDLED)
2333 if (iter->ent->type == TRACE_BPRINT &&
2334 trace_flags & TRACE_ITER_PRINTK &&
2335 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2336 return trace_print_bprintk_msg_only(iter);
2338 if (iter->ent->type == TRACE_PRINT &&
2339 trace_flags & TRACE_ITER_PRINTK &&
2340 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2341 return trace_print_printk_msg_only(iter);
2343 if (trace_flags & TRACE_ITER_BIN)
2344 return print_bin_fmt(iter);
2346 if (trace_flags & TRACE_ITER_HEX)
2347 return print_hex_fmt(iter);
2349 if (trace_flags & TRACE_ITER_RAW)
2350 return print_raw_fmt(iter);
2352 return print_trace_fmt(iter);
2355 void trace_latency_header(struct seq_file *m)
2357 struct trace_iterator *iter = m->private;
2359 /* print nothing if the buffers are empty */
2360 if (trace_empty(iter))
2363 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2364 print_trace_header(m, iter);
2366 if (!(trace_flags & TRACE_ITER_VERBOSE))
2367 print_lat_help_header(m);
2370 void trace_default_header(struct seq_file *m)
2372 struct trace_iterator *iter = m->private;
2374 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2377 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2378 /* print nothing if the buffers are empty */
2379 if (trace_empty(iter))
2381 print_trace_header(m, iter);
2382 if (!(trace_flags & TRACE_ITER_VERBOSE))
2383 print_lat_help_header(m);
2385 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2386 if (trace_flags & TRACE_ITER_IRQ_INFO)
2387 print_func_help_header_irq(iter->tr, m);
2389 print_func_help_header(iter->tr, m);
2394 static void test_ftrace_alive(struct seq_file *m)
2396 if (!ftrace_is_dead())
2398 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2399 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2402 static int s_show(struct seq_file *m, void *v)
2404 struct trace_iterator *iter = v;
2407 if (iter->ent == NULL) {
2409 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2411 test_ftrace_alive(m);
2413 if (iter->trace && iter->trace->print_header)
2414 iter->trace->print_header(m);
2416 trace_default_header(m);
2418 } else if (iter->leftover) {
2420 * If we filled the seq_file buffer earlier, we
2421 * want to just show it now.
2423 ret = trace_print_seq(m, &iter->seq);
2425 /* ret should this time be zero, but you never know */
2426 iter->leftover = ret;
2429 print_trace_line(iter);
2430 ret = trace_print_seq(m, &iter->seq);
2432 * If we overflow the seq_file buffer, then it will
2433 * ask us for this data again at start up.
2435 * ret is 0 if seq_file write succeeded.
2438 iter->leftover = ret;
2444 static const struct seq_operations tracer_seq_ops = {
2451 static struct trace_iterator *
2452 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
2454 long cpu_file = (long) inode->i_private;
2455 struct trace_iterator *iter;
2458 if (tracing_disabled)
2459 return ERR_PTR(-ENODEV);
2461 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2463 return ERR_PTR(-ENOMEM);
2465 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2467 if (!iter->buffer_iter)
2471 * We make a copy of the current tracer to avoid concurrent
2472 * changes on it while we are reading.
2474 mutex_lock(&trace_types_lock);
2475 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2479 *iter->trace = *current_trace;
2481 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2484 if (current_trace->print_max || snapshot)
2487 iter->tr = &global_trace;
2488 iter->snapshot = snapshot;
2490 mutex_init(&iter->mutex);
2491 iter->cpu_file = cpu_file;
2493 /* Notify the tracer early; before we stop tracing. */
2494 if (iter->trace && iter->trace->open)
2495 iter->trace->open(iter);
2497 /* Annotate start of buffers if we had overruns */
2498 if (ring_buffer_overruns(iter->tr->buffer))
2499 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2501 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2502 if (trace_clocks[trace_clock_id].in_ns)
2503 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2505 /* stop the trace while dumping if we are not opening "snapshot" */
2506 if (!iter->snapshot)
2509 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2510 for_each_tracing_cpu(cpu) {
2511 iter->buffer_iter[cpu] =
2512 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2514 ring_buffer_read_prepare_sync();
2515 for_each_tracing_cpu(cpu) {
2516 ring_buffer_read_start(iter->buffer_iter[cpu]);
2517 tracing_iter_reset(iter, cpu);
2520 cpu = iter->cpu_file;
2521 iter->buffer_iter[cpu] =
2522 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2523 ring_buffer_read_prepare_sync();
2524 ring_buffer_read_start(iter->buffer_iter[cpu]);
2525 tracing_iter_reset(iter, cpu);
2528 mutex_unlock(&trace_types_lock);
2533 mutex_unlock(&trace_types_lock);
2535 kfree(iter->buffer_iter);
2537 seq_release_private(inode, file);
2538 return ERR_PTR(-ENOMEM);
2541 int tracing_open_generic(struct inode *inode, struct file *filp)
2543 if (tracing_disabled)
2546 filp->private_data = inode->i_private;
2550 static int tracing_release(struct inode *inode, struct file *file)
2552 struct seq_file *m = file->private_data;
2553 struct trace_iterator *iter;
2556 if (!(file->f_mode & FMODE_READ))
2561 mutex_lock(&trace_types_lock);
2562 for_each_tracing_cpu(cpu) {
2563 if (iter->buffer_iter[cpu])
2564 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2567 if (iter->trace && iter->trace->close)
2568 iter->trace->close(iter);
2570 if (!iter->snapshot)
2571 /* reenable tracing if it was previously enabled */
2573 mutex_unlock(&trace_types_lock);
2575 mutex_destroy(&iter->mutex);
2576 free_cpumask_var(iter->started);
2578 kfree(iter->buffer_iter);
2579 seq_release_private(inode, file);
2583 static int tracing_open(struct inode *inode, struct file *file)
2585 struct trace_iterator *iter;
2588 /* If this file was open for write, then erase contents */
2589 if ((file->f_mode & FMODE_WRITE) &&
2590 (file->f_flags & O_TRUNC)) {
2591 long cpu = (long) inode->i_private;
2593 if (cpu == TRACE_PIPE_ALL_CPU)
2594 tracing_reset_online_cpus(&global_trace);
2596 tracing_reset(&global_trace, cpu);
2599 if (file->f_mode & FMODE_READ) {
2600 iter = __tracing_open(inode, file, false);
2602 ret = PTR_ERR(iter);
2603 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2604 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2610 t_next(struct seq_file *m, void *v, loff_t *pos)
2612 struct tracer *t = v;
2622 static void *t_start(struct seq_file *m, loff_t *pos)
2627 mutex_lock(&trace_types_lock);
2628 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2634 static void t_stop(struct seq_file *m, void *p)
2636 mutex_unlock(&trace_types_lock);
2639 static int t_show(struct seq_file *m, void *v)
2641 struct tracer *t = v;
2646 seq_printf(m, "%s", t->name);
2655 static const struct seq_operations show_traces_seq_ops = {
2662 static int show_traces_open(struct inode *inode, struct file *file)
2664 if (tracing_disabled)
2667 return seq_open(file, &show_traces_seq_ops);
2671 tracing_write_stub(struct file *filp, const char __user *ubuf,
2672 size_t count, loff_t *ppos)
2677 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2679 if (file->f_mode & FMODE_READ)
2680 return seq_lseek(file, offset, origin);
2685 static const struct file_operations tracing_fops = {
2686 .open = tracing_open,
2688 .write = tracing_write_stub,
2689 .llseek = tracing_seek,
2690 .release = tracing_release,
2693 static const struct file_operations show_traces_fops = {
2694 .open = show_traces_open,
2696 .release = seq_release,
2697 .llseek = seq_lseek,
2701 * Only trace on a CPU if the bitmask is set:
2703 static cpumask_var_t tracing_cpumask;
2706 * The tracer itself will not take this lock, but still we want
2707 * to provide a consistent cpumask to user-space:
2709 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2712 * Temporary storage for the character representation of the
2713 * CPU bitmask (and one more byte for the newline):
2715 static char mask_str[NR_CPUS + 1];
2718 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2719 size_t count, loff_t *ppos)
2723 mutex_lock(&tracing_cpumask_update_lock);
2725 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2726 if (count - len < 2) {
2730 len += sprintf(mask_str + len, "\n");
2731 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2734 mutex_unlock(&tracing_cpumask_update_lock);
2740 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2741 size_t count, loff_t *ppos)
2744 cpumask_var_t tracing_cpumask_new;
2746 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2749 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2753 mutex_lock(&tracing_cpumask_update_lock);
2755 local_irq_disable();
2756 arch_spin_lock(&ftrace_max_lock);
2757 for_each_tracing_cpu(cpu) {
2759 * Increase/decrease the disabled counter if we are
2760 * about to flip a bit in the cpumask:
2762 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2763 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2764 atomic_inc(&global_trace.data[cpu]->disabled);
2765 ring_buffer_record_disable_cpu(global_trace.buffer, cpu);
2767 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2768 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2769 atomic_dec(&global_trace.data[cpu]->disabled);
2770 ring_buffer_record_enable_cpu(global_trace.buffer, cpu);
2773 arch_spin_unlock(&ftrace_max_lock);
2776 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2778 mutex_unlock(&tracing_cpumask_update_lock);
2779 free_cpumask_var(tracing_cpumask_new);
2784 free_cpumask_var(tracing_cpumask_new);
2789 static const struct file_operations tracing_cpumask_fops = {
2790 .open = tracing_open_generic,
2791 .read = tracing_cpumask_read,
2792 .write = tracing_cpumask_write,
2793 .llseek = generic_file_llseek,
2796 static int tracing_trace_options_show(struct seq_file *m, void *v)
2798 struct tracer_opt *trace_opts;
2802 mutex_lock(&trace_types_lock);
2803 tracer_flags = current_trace->flags->val;
2804 trace_opts = current_trace->flags->opts;
2806 for (i = 0; trace_options[i]; i++) {
2807 if (trace_flags & (1 << i))
2808 seq_printf(m, "%s\n", trace_options[i]);
2810 seq_printf(m, "no%s\n", trace_options[i]);
2813 for (i = 0; trace_opts[i].name; i++) {
2814 if (tracer_flags & trace_opts[i].bit)
2815 seq_printf(m, "%s\n", trace_opts[i].name);
2817 seq_printf(m, "no%s\n", trace_opts[i].name);
2819 mutex_unlock(&trace_types_lock);
2824 static int __set_tracer_option(struct tracer *trace,
2825 struct tracer_flags *tracer_flags,
2826 struct tracer_opt *opts, int neg)
2830 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2835 tracer_flags->val &= ~opts->bit;
2837 tracer_flags->val |= opts->bit;
2841 /* Try to assign a tracer specific option */
2842 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2844 struct tracer_flags *tracer_flags = trace->flags;
2845 struct tracer_opt *opts = NULL;
2848 for (i = 0; tracer_flags->opts[i].name; i++) {
2849 opts = &tracer_flags->opts[i];
2851 if (strcmp(cmp, opts->name) == 0)
2852 return __set_tracer_option(trace, trace->flags,
2859 static void set_tracer_flags(unsigned int mask, int enabled)
2861 /* do nothing if flag is already set */
2862 if (!!(trace_flags & mask) == !!enabled)
2866 trace_flags |= mask;
2868 trace_flags &= ~mask;
2870 if (mask == TRACE_ITER_RECORD_CMD)
2871 trace_event_enable_cmd_record(enabled);
2873 if (mask == TRACE_ITER_OVERWRITE)
2874 ring_buffer_change_overwrite(global_trace.buffer, enabled);
2876 if (mask == TRACE_ITER_PRINTK)
2877 trace_printk_start_stop_comm(enabled);
2880 static int trace_set_options(char *option)
2887 cmp = strstrip(option);
2889 if (strncmp(cmp, "no", 2) == 0) {
2894 for (i = 0; trace_options[i]; i++) {
2895 if (strcmp(cmp, trace_options[i]) == 0) {
2896 set_tracer_flags(1 << i, !neg);
2901 /* If no option could be set, test the specific tracer options */
2902 if (!trace_options[i]) {
2903 mutex_lock(&trace_types_lock);
2904 ret = set_tracer_option(current_trace, cmp, neg);
2905 mutex_unlock(&trace_types_lock);
2912 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2913 size_t cnt, loff_t *ppos)
2917 if (cnt >= sizeof(buf))
2920 if (copy_from_user(&buf, ubuf, cnt))
2925 trace_set_options(buf);
2932 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2934 if (tracing_disabled)
2936 return single_open(file, tracing_trace_options_show, NULL);
2939 static const struct file_operations tracing_iter_fops = {
2940 .open = tracing_trace_options_open,
2942 .llseek = seq_lseek,
2943 .release = single_release,
2944 .write = tracing_trace_options_write,
2947 static const char readme_msg[] =
2948 "tracing mini-HOWTO:\n\n"
2949 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2950 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2951 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2952 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2954 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2955 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2957 "# cat /sys/kernel/debug/tracing/trace_options\n"
2958 "noprint-parent nosym-offset nosym-addr noverbose\n"
2959 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2960 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2961 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2962 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2966 tracing_readme_read(struct file *filp, char __user *ubuf,
2967 size_t cnt, loff_t *ppos)
2969 return simple_read_from_buffer(ubuf, cnt, ppos,
2970 readme_msg, strlen(readme_msg));
2973 static const struct file_operations tracing_readme_fops = {
2974 .open = tracing_open_generic,
2975 .read = tracing_readme_read,
2976 .llseek = generic_file_llseek,
2980 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2981 size_t cnt, loff_t *ppos)
2990 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2994 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3002 for (i = 0; i < SAVED_CMDLINES; i++) {
3005 pid = map_cmdline_to_pid[i];
3006 if (pid == -1 || pid == NO_CMDLINE_MAP)
3009 trace_find_cmdline(pid, buf_comm);
3010 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3015 len = simple_read_from_buffer(ubuf, cnt, ppos,
3024 static const struct file_operations tracing_saved_cmdlines_fops = {
3025 .open = tracing_open_generic,
3026 .read = tracing_saved_cmdlines_read,
3027 .llseek = generic_file_llseek,
3031 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3032 size_t cnt, loff_t *ppos)
3034 char buf[MAX_TRACER_SIZE+2];
3037 mutex_lock(&trace_types_lock);
3038 r = sprintf(buf, "%s\n", current_trace->name);
3039 mutex_unlock(&trace_types_lock);
3041 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3044 int tracer_init(struct tracer *t, struct trace_array *tr)
3046 tracing_reset_online_cpus(tr);
3050 static void set_buffer_entries(struct trace_array *tr, unsigned long val)
3053 for_each_tracing_cpu(cpu)
3054 tr->data[cpu]->entries = val;
3057 /* resize @tr's buffer to the size of @size_tr's entries */
3058 static int resize_buffer_duplicate_size(struct trace_array *tr,
3059 struct trace_array *size_tr, int cpu_id)
3063 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3064 for_each_tracing_cpu(cpu) {
3065 ret = ring_buffer_resize(tr->buffer,
3066 size_tr->data[cpu]->entries, cpu);
3069 tr->data[cpu]->entries = size_tr->data[cpu]->entries;
3072 ret = ring_buffer_resize(tr->buffer,
3073 size_tr->data[cpu_id]->entries, cpu_id);
3075 tr->data[cpu_id]->entries =
3076 size_tr->data[cpu_id]->entries;
3082 static int __tracing_resize_ring_buffer(unsigned long size, int cpu)
3087 * If kernel or user changes the size of the ring buffer
3088 * we use the size that was given, and we can forget about
3089 * expanding it later.
3091 ring_buffer_expanded = 1;
3093 /* May be called before buffers are initialized */
3094 if (!global_trace.buffer)
3097 ret = ring_buffer_resize(global_trace.buffer, size, cpu);
3101 if (!current_trace->use_max_tr)
3104 ret = ring_buffer_resize(max_tr.buffer, size, cpu);
3106 int r = resize_buffer_duplicate_size(&global_trace,
3107 &global_trace, cpu);
3110 * AARGH! We are left with different
3111 * size max buffer!!!!
3112 * The max buffer is our "snapshot" buffer.
3113 * When a tracer needs a snapshot (one of the
3114 * latency tracers), it swaps the max buffer
3115 * with the saved snap shot. We succeeded to
3116 * update the size of the main buffer, but failed to
3117 * update the size of the max buffer. But when we tried
3118 * to reset the main buffer to the original size, we
3119 * failed there too. This is very unlikely to
3120 * happen, but if it does, warn and kill all
3124 tracing_disabled = 1;
3129 if (cpu == RING_BUFFER_ALL_CPUS)
3130 set_buffer_entries(&max_tr, size);
3132 max_tr.data[cpu]->entries = size;
3135 if (cpu == RING_BUFFER_ALL_CPUS)
3136 set_buffer_entries(&global_trace, size);
3138 global_trace.data[cpu]->entries = size;
3143 static ssize_t tracing_resize_ring_buffer(unsigned long size, int cpu_id)
3147 mutex_lock(&trace_types_lock);
3149 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3150 /* make sure, this cpu is enabled in the mask */
3151 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3157 ret = __tracing_resize_ring_buffer(size, cpu_id);
3162 mutex_unlock(&trace_types_lock);
3169 * tracing_update_buffers - used by tracing facility to expand ring buffers
3171 * To save on memory when the tracing is never used on a system with it
3172 * configured in. The ring buffers are set to a minimum size. But once
3173 * a user starts to use the tracing facility, then they need to grow
3174 * to their default size.
3176 * This function is to be called when a tracer is about to be used.
3178 int tracing_update_buffers(void)
3182 mutex_lock(&trace_types_lock);
3183 if (!ring_buffer_expanded)
3184 ret = __tracing_resize_ring_buffer(trace_buf_size,
3185 RING_BUFFER_ALL_CPUS);
3186 mutex_unlock(&trace_types_lock);
3191 struct trace_option_dentry;
3193 static struct trace_option_dentry *
3194 create_trace_option_files(struct tracer *tracer);
3197 destroy_trace_option_files(struct trace_option_dentry *topts);
3199 static int tracing_set_tracer(const char *buf)
3201 static struct trace_option_dentry *topts;
3202 struct trace_array *tr = &global_trace;
3207 mutex_lock(&trace_types_lock);
3209 if (!ring_buffer_expanded) {
3210 ret = __tracing_resize_ring_buffer(trace_buf_size,
3211 RING_BUFFER_ALL_CPUS);
3217 for (t = trace_types; t; t = t->next) {
3218 if (strcmp(t->name, buf) == 0)
3225 if (t == current_trace)
3228 trace_branch_disable();
3229 if (current_trace->reset)
3230 current_trace->reset(tr);
3232 had_max_tr = current_trace->allocated_snapshot;
3233 current_trace = &nop_trace;
3235 if (had_max_tr && !t->use_max_tr) {
3237 * We need to make sure that the update_max_tr sees that
3238 * current_trace changed to nop_trace to keep it from
3239 * swapping the buffers after we resize it.
3240 * The update_max_tr is called from interrupts disabled
3241 * so a synchronized_sched() is sufficient.
3243 synchronize_sched();
3245 * We don't free the ring buffer. instead, resize it because
3246 * The max_tr ring buffer has some state (e.g. ring->clock) and
3247 * we want preserve it.
3249 ring_buffer_resize(max_tr.buffer, 1, RING_BUFFER_ALL_CPUS);
3250 set_buffer_entries(&max_tr, 1);
3251 tracing_reset_online_cpus(&max_tr);
3252 current_trace->allocated_snapshot = false;
3254 destroy_trace_option_files(topts);
3256 topts = create_trace_option_files(t);
3257 if (t->use_max_tr && !had_max_tr) {
3258 /* we need to make per cpu buffer sizes equivalent */
3259 ret = resize_buffer_duplicate_size(&max_tr, &global_trace,
3260 RING_BUFFER_ALL_CPUS);
3263 t->allocated_snapshot = true;
3267 ret = tracer_init(t, tr);
3273 trace_branch_enable(tr);
3275 mutex_unlock(&trace_types_lock);
3281 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3282 size_t cnt, loff_t *ppos)
3284 char buf[MAX_TRACER_SIZE+1];
3291 if (cnt > MAX_TRACER_SIZE)
3292 cnt = MAX_TRACER_SIZE;
3294 if (copy_from_user(&buf, ubuf, cnt))
3299 /* strip ending whitespace. */
3300 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3303 err = tracing_set_tracer(buf);
3313 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3314 size_t cnt, loff_t *ppos)
3316 unsigned long *ptr = filp->private_data;
3320 r = snprintf(buf, sizeof(buf), "%ld\n",
3321 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3322 if (r > sizeof(buf))
3324 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3328 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3329 size_t cnt, loff_t *ppos)
3331 unsigned long *ptr = filp->private_data;
3335 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3344 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3346 long cpu_file = (long) inode->i_private;
3347 struct trace_iterator *iter;
3350 if (tracing_disabled)
3353 mutex_lock(&trace_types_lock);
3355 /* create a buffer to store the information to pass to userspace */
3356 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3363 * We make a copy of the current tracer to avoid concurrent
3364 * changes on it while we are reading.
3366 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3371 *iter->trace = *current_trace;
3373 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3378 /* trace pipe does not show start of buffer */
3379 cpumask_setall(iter->started);
3381 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3382 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3384 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3385 if (trace_clocks[trace_clock_id].in_ns)
3386 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3388 iter->cpu_file = cpu_file;
3389 iter->tr = &global_trace;
3390 mutex_init(&iter->mutex);
3391 filp->private_data = iter;
3393 if (iter->trace->pipe_open)
3394 iter->trace->pipe_open(iter);
3396 nonseekable_open(inode, filp);
3398 mutex_unlock(&trace_types_lock);
3404 mutex_unlock(&trace_types_lock);
3408 static int tracing_release_pipe(struct inode *inode, struct file *file)
3410 struct trace_iterator *iter = file->private_data;
3412 mutex_lock(&trace_types_lock);
3414 if (iter->trace->pipe_close)
3415 iter->trace->pipe_close(iter);
3417 mutex_unlock(&trace_types_lock);
3419 free_cpumask_var(iter->started);
3420 mutex_destroy(&iter->mutex);
3428 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3430 struct trace_iterator *iter = filp->private_data;
3432 if (trace_flags & TRACE_ITER_BLOCK) {
3434 * Always select as readable when in blocking mode
3436 return POLLIN | POLLRDNORM;
3438 if (!trace_empty(iter))
3439 return POLLIN | POLLRDNORM;
3440 poll_wait(filp, &trace_wait, poll_table);
3441 if (!trace_empty(iter))
3442 return POLLIN | POLLRDNORM;
3449 * This is a make-shift waitqueue.
3450 * A tracer might use this callback on some rare cases:
3452 * 1) the current tracer might hold the runqueue lock when it wakes up
3453 * a reader, hence a deadlock (sched, function, and function graph tracers)
3454 * 2) the function tracers, trace all functions, we don't want
3455 * the overhead of calling wake_up and friends
3456 * (and tracing them too)
3458 * Anyway, this is really very primitive wakeup.
3460 void poll_wait_pipe(struct trace_iterator *iter)
3462 set_current_state(TASK_INTERRUPTIBLE);
3463 /* sleep for 100 msecs, and try again. */
3464 schedule_timeout(HZ / 10);
3467 /* Must be called with trace_types_lock mutex held. */
3468 static int tracing_wait_pipe(struct file *filp)
3470 struct trace_iterator *iter = filp->private_data;
3472 while (trace_empty(iter)) {
3474 if ((filp->f_flags & O_NONBLOCK)) {
3478 mutex_unlock(&iter->mutex);
3480 iter->trace->wait_pipe(iter);
3482 mutex_lock(&iter->mutex);
3484 if (signal_pending(current))
3488 * We block until we read something and tracing is disabled.
3489 * We still block if tracing is disabled, but we have never
3490 * read anything. This allows a user to cat this file, and
3491 * then enable tracing. But after we have read something,
3492 * we give an EOF when tracing is again disabled.
3494 * iter->pos will be 0 if we haven't read anything.
3496 if (!tracing_is_enabled() && iter->pos)
3507 tracing_read_pipe(struct file *filp, char __user *ubuf,
3508 size_t cnt, loff_t *ppos)
3510 struct trace_iterator *iter = filp->private_data;
3513 /* return any leftover data */
3514 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3518 trace_seq_init(&iter->seq);
3520 /* copy the tracer to avoid using a global lock all around */
3521 mutex_lock(&trace_types_lock);
3522 if (unlikely(iter->trace->name != current_trace->name))
3523 *iter->trace = *current_trace;
3524 mutex_unlock(&trace_types_lock);
3527 * Avoid more than one consumer on a single file descriptor
3528 * This is just a matter of traces coherency, the ring buffer itself
3531 mutex_lock(&iter->mutex);
3532 if (iter->trace->read) {
3533 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3539 sret = tracing_wait_pipe(filp);
3543 /* stop when tracing is finished */
3544 if (trace_empty(iter)) {
3549 if (cnt >= PAGE_SIZE)
3550 cnt = PAGE_SIZE - 1;
3552 /* reset all but tr, trace, and overruns */
3553 memset(&iter->seq, 0,
3554 sizeof(struct trace_iterator) -
3555 offsetof(struct trace_iterator, seq));
3558 trace_event_read_lock();
3559 trace_access_lock(iter->cpu_file);
3560 while (trace_find_next_entry_inc(iter) != NULL) {
3561 enum print_line_t ret;
3562 int len = iter->seq.len;
3564 ret = print_trace_line(iter);
3565 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3566 /* don't print partial lines */
3567 iter->seq.len = len;
3570 if (ret != TRACE_TYPE_NO_CONSUME)
3571 trace_consume(iter);
3573 if (iter->seq.len >= cnt)
3577 * Setting the full flag means we reached the trace_seq buffer
3578 * size and we should leave by partial output condition above.
3579 * One of the trace_seq_* functions is not used properly.
3581 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3584 trace_access_unlock(iter->cpu_file);
3585 trace_event_read_unlock();
3587 /* Now copy what we have to the user */
3588 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3589 if (iter->seq.readpos >= iter->seq.len)
3590 trace_seq_init(&iter->seq);
3593 * If there was nothing to send to user, in spite of consuming trace
3594 * entries, go back to wait for more entries.
3600 mutex_unlock(&iter->mutex);
3605 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3606 struct pipe_buffer *buf)
3608 __free_page(buf->page);
3611 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3614 __free_page(spd->pages[idx]);
3617 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3619 .map = generic_pipe_buf_map,
3620 .unmap = generic_pipe_buf_unmap,
3621 .confirm = generic_pipe_buf_confirm,
3622 .release = tracing_pipe_buf_release,
3623 .steal = generic_pipe_buf_steal,
3624 .get = generic_pipe_buf_get,
3628 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3633 /* Seq buffer is page-sized, exactly what we need. */
3635 count = iter->seq.len;
3636 ret = print_trace_line(iter);
3637 count = iter->seq.len - count;
3640 iter->seq.len -= count;
3643 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3644 iter->seq.len -= count;
3648 if (ret != TRACE_TYPE_NO_CONSUME)
3649 trace_consume(iter);
3651 if (!trace_find_next_entry_inc(iter)) {
3661 static ssize_t tracing_splice_read_pipe(struct file *filp,
3663 struct pipe_inode_info *pipe,
3667 struct page *pages_def[PIPE_DEF_BUFFERS];
3668 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3669 struct trace_iterator *iter = filp->private_data;
3670 struct splice_pipe_desc spd = {
3672 .partial = partial_def,
3673 .nr_pages = 0, /* This gets updated below. */
3674 .nr_pages_max = PIPE_DEF_BUFFERS,
3676 .ops = &tracing_pipe_buf_ops,
3677 .spd_release = tracing_spd_release_pipe,
3683 if (splice_grow_spd(pipe, &spd))
3686 /* copy the tracer to avoid using a global lock all around */
3687 mutex_lock(&trace_types_lock);
3688 if (unlikely(iter->trace->name != current_trace->name))
3689 *iter->trace = *current_trace;
3690 mutex_unlock(&trace_types_lock);
3692 mutex_lock(&iter->mutex);
3694 if (iter->trace->splice_read) {
3695 ret = iter->trace->splice_read(iter, filp,
3696 ppos, pipe, len, flags);
3701 ret = tracing_wait_pipe(filp);
3705 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3710 trace_event_read_lock();
3711 trace_access_lock(iter->cpu_file);
3713 /* Fill as many pages as possible. */
3714 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3715 spd.pages[i] = alloc_page(GFP_KERNEL);
3719 rem = tracing_fill_pipe_page(rem, iter);
3721 /* Copy the data into the page, so we can start over. */
3722 ret = trace_seq_to_buffer(&iter->seq,
3723 page_address(spd.pages[i]),
3726 __free_page(spd.pages[i]);
3729 spd.partial[i].offset = 0;
3730 spd.partial[i].len = iter->seq.len;
3732 trace_seq_init(&iter->seq);
3735 trace_access_unlock(iter->cpu_file);
3736 trace_event_read_unlock();
3737 mutex_unlock(&iter->mutex);
3741 ret = splice_to_pipe(pipe, &spd);
3743 splice_shrink_spd(&spd);
3747 mutex_unlock(&iter->mutex);
3751 struct ftrace_entries_info {
3752 struct trace_array *tr;
3756 static int tracing_entries_open(struct inode *inode, struct file *filp)
3758 struct ftrace_entries_info *info;
3760 if (tracing_disabled)
3763 info = kzalloc(sizeof(*info), GFP_KERNEL);
3767 info->tr = &global_trace;
3768 info->cpu = (unsigned long)inode->i_private;
3770 filp->private_data = info;
3776 tracing_entries_read(struct file *filp, char __user *ubuf,
3777 size_t cnt, loff_t *ppos)
3779 struct ftrace_entries_info *info = filp->private_data;
3780 struct trace_array *tr = info->tr;
3785 mutex_lock(&trace_types_lock);
3787 if (info->cpu == RING_BUFFER_ALL_CPUS) {
3788 int cpu, buf_size_same;
3793 /* check if all cpu sizes are same */
3794 for_each_tracing_cpu(cpu) {
3795 /* fill in the size from first enabled cpu */
3797 size = tr->data[cpu]->entries;
3798 if (size != tr->data[cpu]->entries) {
3804 if (buf_size_same) {
3805 if (!ring_buffer_expanded)
3806 r = sprintf(buf, "%lu (expanded: %lu)\n",
3808 trace_buf_size >> 10);
3810 r = sprintf(buf, "%lu\n", size >> 10);
3812 r = sprintf(buf, "X\n");
3814 r = sprintf(buf, "%lu\n", tr->data[info->cpu]->entries >> 10);
3816 mutex_unlock(&trace_types_lock);
3818 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3823 tracing_entries_write(struct file *filp, const char __user *ubuf,
3824 size_t cnt, loff_t *ppos)
3826 struct ftrace_entries_info *info = filp->private_data;
3830 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3834 /* must have at least 1 entry */
3838 /* value is in KB */
3841 ret = tracing_resize_ring_buffer(val, info->cpu);
3851 tracing_entries_release(struct inode *inode, struct file *filp)
3853 struct ftrace_entries_info *info = filp->private_data;
3861 tracing_total_entries_read(struct file *filp, char __user *ubuf,
3862 size_t cnt, loff_t *ppos)
3864 struct trace_array *tr = filp->private_data;
3867 unsigned long size = 0, expanded_size = 0;
3869 mutex_lock(&trace_types_lock);
3870 for_each_tracing_cpu(cpu) {
3871 size += tr->data[cpu]->entries >> 10;
3872 if (!ring_buffer_expanded)
3873 expanded_size += trace_buf_size >> 10;
3875 if (ring_buffer_expanded)
3876 r = sprintf(buf, "%lu\n", size);
3878 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
3879 mutex_unlock(&trace_types_lock);
3881 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3885 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
3886 size_t cnt, loff_t *ppos)
3889 * There is no need to read what the user has written, this function
3890 * is just to make sure that there is no error when "echo" is used
3899 tracing_free_buffer_release(struct inode *inode, struct file *filp)
3901 /* disable tracing ? */
3902 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
3904 /* resize the ring buffer to 0 */
3905 tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS);
3911 tracing_mark_write(struct file *filp, const char __user *ubuf,
3912 size_t cnt, loff_t *fpos)
3914 unsigned long addr = (unsigned long)ubuf;
3915 struct ring_buffer_event *event;
3916 struct ring_buffer *buffer;
3917 struct print_entry *entry;
3918 unsigned long irq_flags;
3919 struct page *pages[2];
3929 if (tracing_disabled)
3932 if (!(trace_flags & TRACE_ITER_MARKERS))
3935 if (cnt > TRACE_BUF_SIZE)
3936 cnt = TRACE_BUF_SIZE;
3939 * Userspace is injecting traces into the kernel trace buffer.
3940 * We want to be as non intrusive as possible.
3941 * To do so, we do not want to allocate any special buffers
3942 * or take any locks, but instead write the userspace data
3943 * straight into the ring buffer.
3945 * First we need to pin the userspace buffer into memory,
3946 * which, most likely it is, because it just referenced it.
3947 * But there's no guarantee that it is. By using get_user_pages_fast()
3948 * and kmap_atomic/kunmap_atomic() we can get access to the
3949 * pages directly. We then write the data directly into the
3952 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
3954 /* check if we cross pages */
3955 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
3958 offset = addr & (PAGE_SIZE - 1);
3961 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
3962 if (ret < nr_pages) {
3964 put_page(pages[ret]);
3969 for (i = 0; i < nr_pages; i++)
3970 map_page[i] = kmap_atomic(pages[i]);
3972 local_save_flags(irq_flags);
3973 size = sizeof(*entry) + cnt + 2; /* possible \n added */
3974 buffer = global_trace.buffer;
3975 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3976 irq_flags, preempt_count());
3978 /* Ring buffer disabled, return as if not open for write */
3983 entry = ring_buffer_event_data(event);
3984 entry->ip = _THIS_IP_;
3986 if (nr_pages == 2) {
3987 len = PAGE_SIZE - offset;
3988 memcpy(&entry->buf, map_page[0] + offset, len);
3989 memcpy(&entry->buf[len], map_page[1], cnt - len);
3991 memcpy(&entry->buf, map_page[0] + offset, cnt);
3993 if (entry->buf[cnt - 1] != '\n') {
3994 entry->buf[cnt] = '\n';
3995 entry->buf[cnt + 1] = '\0';
3997 entry->buf[cnt] = '\0';
3999 __buffer_unlock_commit(buffer, event);
4006 for (i = 0; i < nr_pages; i++){
4007 kunmap_atomic(map_page[i]);
4014 static int tracing_clock_show(struct seq_file *m, void *v)
4018 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4020 "%s%s%s%s", i ? " " : "",
4021 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
4022 i == trace_clock_id ? "]" : "");
4028 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4029 size_t cnt, loff_t *fpos)
4032 const char *clockstr;
4035 if (cnt >= sizeof(buf))
4038 if (copy_from_user(&buf, ubuf, cnt))
4043 clockstr = strstrip(buf);
4045 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4046 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4049 if (i == ARRAY_SIZE(trace_clocks))
4054 mutex_lock(&trace_types_lock);
4056 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
4058 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
4061 * New clock may not be consistent with the previous clock.
4062 * Reset the buffer so that it doesn't have incomparable timestamps.
4064 tracing_reset_online_cpus(&global_trace);
4065 tracing_reset_online_cpus(&max_tr);
4067 mutex_unlock(&trace_types_lock);
4074 static int tracing_clock_open(struct inode *inode, struct file *file)
4076 if (tracing_disabled)
4078 return single_open(file, tracing_clock_show, NULL);
4081 #ifdef CONFIG_TRACER_SNAPSHOT
4082 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4084 struct trace_iterator *iter;
4087 if (file->f_mode & FMODE_READ) {
4088 iter = __tracing_open(inode, file, true);
4090 ret = PTR_ERR(iter);
4096 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4102 ret = tracing_update_buffers();
4106 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4110 mutex_lock(&trace_types_lock);
4112 if (current_trace->use_max_tr) {
4119 if (current_trace->allocated_snapshot) {
4120 /* free spare buffer */
4121 ring_buffer_resize(max_tr.buffer, 1,
4122 RING_BUFFER_ALL_CPUS);
4123 set_buffer_entries(&max_tr, 1);
4124 tracing_reset_online_cpus(&max_tr);
4125 current_trace->allocated_snapshot = false;
4129 if (!current_trace->allocated_snapshot) {
4130 /* allocate spare buffer */
4131 ret = resize_buffer_duplicate_size(&max_tr,
4132 &global_trace, RING_BUFFER_ALL_CPUS);
4135 current_trace->allocated_snapshot = true;
4138 local_irq_disable();
4139 /* Now, we're going to swap */
4140 update_max_tr(&global_trace, current, smp_processor_id());
4144 if (current_trace->allocated_snapshot)
4145 tracing_reset_online_cpus(&max_tr);
4156 mutex_unlock(&trace_types_lock);
4159 #endif /* CONFIG_TRACER_SNAPSHOT */
4162 static const struct file_operations tracing_max_lat_fops = {
4163 .open = tracing_open_generic,
4164 .read = tracing_max_lat_read,
4165 .write = tracing_max_lat_write,
4166 .llseek = generic_file_llseek,
4169 static const struct file_operations set_tracer_fops = {
4170 .open = tracing_open_generic,
4171 .read = tracing_set_trace_read,
4172 .write = tracing_set_trace_write,
4173 .llseek = generic_file_llseek,
4176 static const struct file_operations tracing_pipe_fops = {
4177 .open = tracing_open_pipe,
4178 .poll = tracing_poll_pipe,
4179 .read = tracing_read_pipe,
4180 .splice_read = tracing_splice_read_pipe,
4181 .release = tracing_release_pipe,
4182 .llseek = no_llseek,
4185 static const struct file_operations tracing_entries_fops = {
4186 .open = tracing_entries_open,
4187 .read = tracing_entries_read,
4188 .write = tracing_entries_write,
4189 .release = tracing_entries_release,
4190 .llseek = generic_file_llseek,
4193 static const struct file_operations tracing_total_entries_fops = {
4194 .open = tracing_open_generic,
4195 .read = tracing_total_entries_read,
4196 .llseek = generic_file_llseek,
4199 static const struct file_operations tracing_free_buffer_fops = {
4200 .write = tracing_free_buffer_write,
4201 .release = tracing_free_buffer_release,
4204 static const struct file_operations tracing_mark_fops = {
4205 .open = tracing_open_generic,
4206 .write = tracing_mark_write,
4207 .llseek = generic_file_llseek,
4210 static const struct file_operations trace_clock_fops = {
4211 .open = tracing_clock_open,
4213 .llseek = seq_lseek,
4214 .release = single_release,
4215 .write = tracing_clock_write,
4218 #ifdef CONFIG_TRACER_SNAPSHOT
4219 static const struct file_operations snapshot_fops = {
4220 .open = tracing_snapshot_open,
4222 .write = tracing_snapshot_write,
4223 .llseek = tracing_seek,
4224 .release = tracing_release,
4226 #endif /* CONFIG_TRACER_SNAPSHOT */
4228 struct ftrace_buffer_info {
4229 struct trace_array *tr;
4235 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4237 int cpu = (int)(long)inode->i_private;
4238 struct ftrace_buffer_info *info;
4240 if (tracing_disabled)
4243 info = kzalloc(sizeof(*info), GFP_KERNEL);
4247 info->tr = &global_trace;
4250 /* Force reading ring buffer for first read */
4251 info->read = (unsigned int)-1;
4253 filp->private_data = info;
4255 return nonseekable_open(inode, filp);
4259 tracing_buffers_read(struct file *filp, char __user *ubuf,
4260 size_t count, loff_t *ppos)
4262 struct ftrace_buffer_info *info = filp->private_data;
4270 info->spare = ring_buffer_alloc_read_page(info->tr->buffer, info->cpu);
4274 /* Do we have previous read data to read? */
4275 if (info->read < PAGE_SIZE)
4278 trace_access_lock(info->cpu);
4279 ret = ring_buffer_read_page(info->tr->buffer,
4283 trace_access_unlock(info->cpu);
4290 size = PAGE_SIZE - info->read;
4294 ret = copy_to_user(ubuf, info->spare + info->read, size);
4305 static int tracing_buffers_release(struct inode *inode, struct file *file)
4307 struct ftrace_buffer_info *info = file->private_data;
4310 ring_buffer_free_read_page(info->tr->buffer, info->spare);
4317 struct ring_buffer *buffer;
4322 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4323 struct pipe_buffer *buf)
4325 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4330 ring_buffer_free_read_page(ref->buffer, ref->page);
4335 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4336 struct pipe_buffer *buf)
4338 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4343 /* Pipe buffer operations for a buffer. */
4344 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4346 .map = generic_pipe_buf_map,
4347 .unmap = generic_pipe_buf_unmap,
4348 .confirm = generic_pipe_buf_confirm,
4349 .release = buffer_pipe_buf_release,
4350 .steal = generic_pipe_buf_steal,
4351 .get = buffer_pipe_buf_get,
4355 * Callback from splice_to_pipe(), if we need to release some pages
4356 * at the end of the spd in case we error'ed out in filling the pipe.
4358 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4360 struct buffer_ref *ref =
4361 (struct buffer_ref *)spd->partial[i].private;
4366 ring_buffer_free_read_page(ref->buffer, ref->page);
4368 spd->partial[i].private = 0;
4372 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4373 struct pipe_inode_info *pipe, size_t len,
4376 struct ftrace_buffer_info *info = file->private_data;
4377 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4378 struct page *pages_def[PIPE_DEF_BUFFERS];
4379 struct splice_pipe_desc spd = {
4381 .partial = partial_def,
4382 .nr_pages_max = PIPE_DEF_BUFFERS,
4384 .ops = &buffer_pipe_buf_ops,
4385 .spd_release = buffer_spd_release,
4387 struct buffer_ref *ref;
4388 int entries, size, i;
4391 if (splice_grow_spd(pipe, &spd))
4394 if (*ppos & (PAGE_SIZE - 1)) {
4399 if (len & (PAGE_SIZE - 1)) {
4400 if (len < PAGE_SIZE) {
4407 trace_access_lock(info->cpu);
4408 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4410 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4414 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4419 ref->buffer = info->tr->buffer;
4420 ref->page = ring_buffer_alloc_read_page(ref->buffer, info->cpu);
4426 r = ring_buffer_read_page(ref->buffer, &ref->page,
4429 ring_buffer_free_read_page(ref->buffer, ref->page);
4435 * zero out any left over data, this is going to
4438 size = ring_buffer_page_len(ref->page);
4439 if (size < PAGE_SIZE)
4440 memset(ref->page + size, 0, PAGE_SIZE - size);
4442 page = virt_to_page(ref->page);
4444 spd.pages[i] = page;
4445 spd.partial[i].len = PAGE_SIZE;
4446 spd.partial[i].offset = 0;
4447 spd.partial[i].private = (unsigned long)ref;
4451 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4454 trace_access_unlock(info->cpu);
4457 /* did we read anything? */
4458 if (!spd.nr_pages) {
4459 if (flags & SPLICE_F_NONBLOCK)
4467 ret = splice_to_pipe(pipe, &spd);
4468 splice_shrink_spd(&spd);
4473 static const struct file_operations tracing_buffers_fops = {
4474 .open = tracing_buffers_open,
4475 .read = tracing_buffers_read,
4476 .release = tracing_buffers_release,
4477 .splice_read = tracing_buffers_splice_read,
4478 .llseek = no_llseek,
4482 tracing_stats_read(struct file *filp, char __user *ubuf,
4483 size_t count, loff_t *ppos)
4485 unsigned long cpu = (unsigned long)filp->private_data;
4486 struct trace_array *tr = &global_trace;
4487 struct trace_seq *s;
4489 unsigned long long t;
4490 unsigned long usec_rem;
4492 s = kmalloc(sizeof(*s), GFP_KERNEL);
4498 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
4499 trace_seq_printf(s, "entries: %ld\n", cnt);
4501 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
4502 trace_seq_printf(s, "overrun: %ld\n", cnt);
4504 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
4505 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
4507 cnt = ring_buffer_bytes_cpu(tr->buffer, cpu);
4508 trace_seq_printf(s, "bytes: %ld\n", cnt);
4510 if (trace_clocks[trace_clock_id].in_ns) {
4511 /* local or global for trace_clock */
4512 t = ns2usecs(ring_buffer_oldest_event_ts(tr->buffer, cpu));
4513 usec_rem = do_div(t, USEC_PER_SEC);
4514 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
4517 t = ns2usecs(ring_buffer_time_stamp(tr->buffer, cpu));
4518 usec_rem = do_div(t, USEC_PER_SEC);
4519 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
4521 /* counter or tsc mode for trace_clock */
4522 trace_seq_printf(s, "oldest event ts: %llu\n",
4523 ring_buffer_oldest_event_ts(tr->buffer, cpu));
4525 trace_seq_printf(s, "now ts: %llu\n",
4526 ring_buffer_time_stamp(tr->buffer, cpu));
4529 cnt = ring_buffer_dropped_events_cpu(tr->buffer, cpu);
4530 trace_seq_printf(s, "dropped events: %ld\n", cnt);
4532 cnt = ring_buffer_read_events_cpu(tr->buffer, cpu);
4533 trace_seq_printf(s, "read events: %ld\n", cnt);
4535 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
4542 static const struct file_operations tracing_stats_fops = {
4543 .open = tracing_open_generic,
4544 .read = tracing_stats_read,
4545 .llseek = generic_file_llseek,
4548 #ifdef CONFIG_DYNAMIC_FTRACE
4550 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
4556 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
4557 size_t cnt, loff_t *ppos)
4559 static char ftrace_dyn_info_buffer[1024];
4560 static DEFINE_MUTEX(dyn_info_mutex);
4561 unsigned long *p = filp->private_data;
4562 char *buf = ftrace_dyn_info_buffer;
4563 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
4566 mutex_lock(&dyn_info_mutex);
4567 r = sprintf(buf, "%ld ", *p);
4569 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
4572 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4574 mutex_unlock(&dyn_info_mutex);
4579 static const struct file_operations tracing_dyn_info_fops = {
4580 .open = tracing_open_generic,
4581 .read = tracing_read_dyn_info,
4582 .llseek = generic_file_llseek,
4586 static struct dentry *d_tracer;
4588 struct dentry *tracing_init_dentry(void)
4595 if (!debugfs_initialized())
4598 d_tracer = debugfs_create_dir("tracing", NULL);
4600 if (!d_tracer && !once) {
4602 pr_warning("Could not create debugfs directory 'tracing'\n");
4609 static struct dentry *d_percpu;
4611 static struct dentry *tracing_dentry_percpu(void)
4614 struct dentry *d_tracer;
4619 d_tracer = tracing_init_dentry();
4624 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4626 if (!d_percpu && !once) {
4628 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4635 static void tracing_init_debugfs_percpu(long cpu)
4637 struct dentry *d_percpu = tracing_dentry_percpu();
4638 struct dentry *d_cpu;
4639 char cpu_dir[30]; /* 30 characters should be more than enough */
4644 snprintf(cpu_dir, 30, "cpu%ld", cpu);
4645 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4647 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4651 /* per cpu trace_pipe */
4652 trace_create_file("trace_pipe", 0444, d_cpu,
4653 (void *) cpu, &tracing_pipe_fops);
4656 trace_create_file("trace", 0644, d_cpu,
4657 (void *) cpu, &tracing_fops);
4659 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4660 (void *) cpu, &tracing_buffers_fops);
4662 trace_create_file("stats", 0444, d_cpu,
4663 (void *) cpu, &tracing_stats_fops);
4665 trace_create_file("buffer_size_kb", 0444, d_cpu,
4666 (void *) cpu, &tracing_entries_fops);
4669 #ifdef CONFIG_FTRACE_SELFTEST
4670 /* Let selftest have access to static functions in this file */
4671 #include "trace_selftest.c"
4674 struct trace_option_dentry {
4675 struct tracer_opt *opt;
4676 struct tracer_flags *flags;
4677 struct dentry *entry;
4681 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4684 struct trace_option_dentry *topt = filp->private_data;
4687 if (topt->flags->val & topt->opt->bit)
4692 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4696 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4699 struct trace_option_dentry *topt = filp->private_data;
4703 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4707 if (val != 0 && val != 1)
4710 if (!!(topt->flags->val & topt->opt->bit) != val) {
4711 mutex_lock(&trace_types_lock);
4712 ret = __set_tracer_option(current_trace, topt->flags,
4714 mutex_unlock(&trace_types_lock);
4725 static const struct file_operations trace_options_fops = {
4726 .open = tracing_open_generic,
4727 .read = trace_options_read,
4728 .write = trace_options_write,
4729 .llseek = generic_file_llseek,
4733 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4736 long index = (long)filp->private_data;
4739 if (trace_flags & (1 << index))
4744 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4748 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4751 long index = (long)filp->private_data;
4755 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4759 if (val != 0 && val != 1)
4761 set_tracer_flags(1 << index, val);
4768 static const struct file_operations trace_options_core_fops = {
4769 .open = tracing_open_generic,
4770 .read = trace_options_core_read,
4771 .write = trace_options_core_write,
4772 .llseek = generic_file_llseek,
4775 struct dentry *trace_create_file(const char *name,
4777 struct dentry *parent,
4779 const struct file_operations *fops)
4783 ret = debugfs_create_file(name, mode, parent, data, fops);
4785 pr_warning("Could not create debugfs '%s' entry\n", name);
4791 static struct dentry *trace_options_init_dentry(void)
4793 struct dentry *d_tracer;
4794 static struct dentry *t_options;
4799 d_tracer = tracing_init_dentry();
4803 t_options = debugfs_create_dir("options", d_tracer);
4805 pr_warning("Could not create debugfs directory 'options'\n");
4813 create_trace_option_file(struct trace_option_dentry *topt,
4814 struct tracer_flags *flags,
4815 struct tracer_opt *opt)
4817 struct dentry *t_options;
4819 t_options = trace_options_init_dentry();
4823 topt->flags = flags;
4826 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4827 &trace_options_fops);
4831 static struct trace_option_dentry *
4832 create_trace_option_files(struct tracer *tracer)
4834 struct trace_option_dentry *topts;
4835 struct tracer_flags *flags;
4836 struct tracer_opt *opts;
4842 flags = tracer->flags;
4844 if (!flags || !flags->opts)
4849 for (cnt = 0; opts[cnt].name; cnt++)
4852 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4856 for (cnt = 0; opts[cnt].name; cnt++)
4857 create_trace_option_file(&topts[cnt], flags,
4864 destroy_trace_option_files(struct trace_option_dentry *topts)
4871 for (cnt = 0; topts[cnt].opt; cnt++) {
4872 if (topts[cnt].entry)
4873 debugfs_remove(topts[cnt].entry);
4879 static struct dentry *
4880 create_trace_option_core_file(const char *option, long index)
4882 struct dentry *t_options;
4884 t_options = trace_options_init_dentry();
4888 return trace_create_file(option, 0644, t_options, (void *)index,
4889 &trace_options_core_fops);
4892 static __init void create_trace_options_dir(void)
4894 struct dentry *t_options;
4897 t_options = trace_options_init_dentry();
4901 for (i = 0; trace_options[i]; i++)
4902 create_trace_option_core_file(trace_options[i], i);
4906 rb_simple_read(struct file *filp, char __user *ubuf,
4907 size_t cnt, loff_t *ppos)
4909 struct trace_array *tr = filp->private_data;
4910 struct ring_buffer *buffer = tr->buffer;
4915 r = ring_buffer_record_is_on(buffer);
4919 r = sprintf(buf, "%d\n", r);
4921 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4925 rb_simple_write(struct file *filp, const char __user *ubuf,
4926 size_t cnt, loff_t *ppos)
4928 struct trace_array *tr = filp->private_data;
4929 struct ring_buffer *buffer = tr->buffer;
4933 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4938 mutex_lock(&trace_types_lock);
4940 ring_buffer_record_on(buffer);
4941 if (current_trace->start)
4942 current_trace->start(tr);
4944 ring_buffer_record_off(buffer);
4945 if (current_trace->stop)
4946 current_trace->stop(tr);
4948 mutex_unlock(&trace_types_lock);
4956 static const struct file_operations rb_simple_fops = {
4957 .open = tracing_open_generic,
4958 .read = rb_simple_read,
4959 .write = rb_simple_write,
4960 .llseek = default_llseek,
4963 static __init int tracer_init_debugfs(void)
4965 struct dentry *d_tracer;
4968 trace_access_lock_init();
4970 d_tracer = tracing_init_dentry();
4972 trace_create_file("trace_options", 0644, d_tracer,
4973 NULL, &tracing_iter_fops);
4975 trace_create_file("tracing_cpumask", 0644, d_tracer,
4976 NULL, &tracing_cpumask_fops);
4978 trace_create_file("trace", 0644, d_tracer,
4979 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4981 trace_create_file("available_tracers", 0444, d_tracer,
4982 &global_trace, &show_traces_fops);
4984 trace_create_file("current_tracer", 0644, d_tracer,
4985 &global_trace, &set_tracer_fops);
4987 #ifdef CONFIG_TRACER_MAX_TRACE
4988 trace_create_file("tracing_max_latency", 0644, d_tracer,
4989 &tracing_max_latency, &tracing_max_lat_fops);
4992 trace_create_file("tracing_thresh", 0644, d_tracer,
4993 &tracing_thresh, &tracing_max_lat_fops);
4995 trace_create_file("README", 0444, d_tracer,
4996 NULL, &tracing_readme_fops);
4998 trace_create_file("trace_pipe", 0444, d_tracer,
4999 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
5001 trace_create_file("buffer_size_kb", 0644, d_tracer,
5002 (void *) RING_BUFFER_ALL_CPUS, &tracing_entries_fops);
5004 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
5005 &global_trace, &tracing_total_entries_fops);
5007 trace_create_file("free_buffer", 0644, d_tracer,
5008 &global_trace, &tracing_free_buffer_fops);
5010 trace_create_file("trace_marker", 0220, d_tracer,
5011 NULL, &tracing_mark_fops);
5013 trace_create_file("saved_cmdlines", 0444, d_tracer,
5014 NULL, &tracing_saved_cmdlines_fops);
5016 trace_create_file("trace_clock", 0644, d_tracer, NULL,
5019 trace_create_file("tracing_on", 0644, d_tracer,
5020 &global_trace, &rb_simple_fops);
5022 #ifdef CONFIG_DYNAMIC_FTRACE
5023 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
5024 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
5027 #ifdef CONFIG_TRACER_SNAPSHOT
5028 trace_create_file("snapshot", 0644, d_tracer,
5029 (void *) TRACE_PIPE_ALL_CPU, &snapshot_fops);
5032 create_trace_options_dir();
5034 for_each_tracing_cpu(cpu)
5035 tracing_init_debugfs_percpu(cpu);
5040 static int trace_panic_handler(struct notifier_block *this,
5041 unsigned long event, void *unused)
5043 if (ftrace_dump_on_oops)
5044 ftrace_dump(ftrace_dump_on_oops);
5048 static struct notifier_block trace_panic_notifier = {
5049 .notifier_call = trace_panic_handler,
5051 .priority = 150 /* priority: INT_MAX >= x >= 0 */
5054 static int trace_die_handler(struct notifier_block *self,
5060 if (ftrace_dump_on_oops)
5061 ftrace_dump(ftrace_dump_on_oops);
5069 static struct notifier_block trace_die_notifier = {
5070 .notifier_call = trace_die_handler,
5075 * printk is set to max of 1024, we really don't need it that big.
5076 * Nothing should be printing 1000 characters anyway.
5078 #define TRACE_MAX_PRINT 1000
5081 * Define here KERN_TRACE so that we have one place to modify
5082 * it if we decide to change what log level the ftrace dump
5085 #define KERN_TRACE KERN_EMERG
5088 trace_printk_seq(struct trace_seq *s)
5090 /* Probably should print a warning here. */
5094 /* should be zero ended, but we are paranoid. */
5095 s->buffer[s->len] = 0;
5097 printk(KERN_TRACE "%s", s->buffer);
5102 void trace_init_global_iter(struct trace_iterator *iter)
5104 iter->tr = &global_trace;
5105 iter->trace = current_trace;
5106 iter->cpu_file = TRACE_PIPE_ALL_CPU;
5110 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
5112 static arch_spinlock_t ftrace_dump_lock =
5113 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
5114 /* use static because iter can be a bit big for the stack */
5115 static struct trace_iterator iter;
5116 unsigned int old_userobj;
5117 static int dump_ran;
5118 unsigned long flags;
5122 local_irq_save(flags);
5123 arch_spin_lock(&ftrace_dump_lock);
5131 /* Did function tracer already get disabled? */
5132 if (ftrace_is_dead()) {
5133 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
5134 printk("# MAY BE MISSING FUNCTION EVENTS\n");
5137 if (disable_tracing)
5140 /* Simulate the iterator */
5141 trace_init_global_iter(&iter);
5143 for_each_tracing_cpu(cpu) {
5144 atomic_inc(&iter.tr->data[cpu]->disabled);
5147 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
5149 /* don't look at user memory in panic mode */
5150 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
5152 switch (oops_dump_mode) {
5154 iter.cpu_file = TRACE_PIPE_ALL_CPU;
5157 iter.cpu_file = raw_smp_processor_id();
5162 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
5163 iter.cpu_file = TRACE_PIPE_ALL_CPU;
5166 printk(KERN_TRACE "Dumping ftrace buffer:\n");
5169 * We need to stop all tracing on all CPUS to read the
5170 * the next buffer. This is a bit expensive, but is
5171 * not done often. We fill all what we can read,
5172 * and then release the locks again.
5175 while (!trace_empty(&iter)) {
5178 printk(KERN_TRACE "---------------------------------\n");
5182 /* reset all but tr, trace, and overruns */
5183 memset(&iter.seq, 0,
5184 sizeof(struct trace_iterator) -
5185 offsetof(struct trace_iterator, seq));
5186 iter.iter_flags |= TRACE_FILE_LAT_FMT;
5189 if (trace_find_next_entry_inc(&iter) != NULL) {
5192 ret = print_trace_line(&iter);
5193 if (ret != TRACE_TYPE_NO_CONSUME)
5194 trace_consume(&iter);
5196 touch_nmi_watchdog();
5198 trace_printk_seq(&iter.seq);
5202 printk(KERN_TRACE " (ftrace buffer empty)\n");
5204 printk(KERN_TRACE "---------------------------------\n");
5207 /* Re-enable tracing if requested */
5208 if (!disable_tracing) {
5209 trace_flags |= old_userobj;
5211 for_each_tracing_cpu(cpu) {
5212 atomic_dec(&iter.tr->data[cpu]->disabled);
5218 arch_spin_unlock(&ftrace_dump_lock);
5219 local_irq_restore(flags);
5222 /* By default: disable tracing after the dump */
5223 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
5225 __ftrace_dump(true, oops_dump_mode);
5227 EXPORT_SYMBOL_GPL(ftrace_dump);
5229 __init static int tracer_alloc_buffers(void)
5232 enum ring_buffer_flags rb_flags;
5237 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
5240 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
5241 goto out_free_buffer_mask;
5243 /* Only allocate trace_printk buffers if a trace_printk exists */
5244 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
5245 /* Must be called before global_trace.buffer is allocated */
5246 trace_printk_init_buffers();
5248 /* To save memory, keep the ring buffer size to its minimum */
5249 if (ring_buffer_expanded)
5250 ring_buf_size = trace_buf_size;
5254 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5256 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
5257 cpumask_copy(tracing_cpumask, cpu_all_mask);
5259 /* TODO: make the number of buffers hot pluggable with CPUS */
5260 global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
5261 if (!global_trace.buffer) {
5262 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
5264 goto out_free_cpumask;
5266 if (global_trace.buffer_disabled)
5270 #ifdef CONFIG_TRACER_MAX_TRACE
5271 max_tr.buffer = ring_buffer_alloc(1, rb_flags);
5272 if (!max_tr.buffer) {
5273 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
5275 ring_buffer_free(global_trace.buffer);
5276 goto out_free_cpumask;
5280 /* Allocate the first page for all buffers */
5281 for_each_tracing_cpu(i) {
5282 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
5283 max_tr.data[i] = &per_cpu(max_tr_data, i);
5286 set_buffer_entries(&global_trace,
5287 ring_buffer_size(global_trace.buffer, 0));
5288 #ifdef CONFIG_TRACER_MAX_TRACE
5289 set_buffer_entries(&max_tr, 1);
5292 trace_init_cmdlines();
5293 init_irq_work(&trace_work_wakeup, trace_wake_up);
5295 register_tracer(&nop_trace);
5297 /* All seems OK, enable tracing */
5298 tracing_disabled = 0;
5300 atomic_notifier_chain_register(&panic_notifier_list,
5301 &trace_panic_notifier);
5303 register_die_notifier(&trace_die_notifier);
5305 while (trace_boot_options) {
5308 option = strsep(&trace_boot_options, ",");
5309 trace_set_options(option);
5315 free_cpumask_var(tracing_cpumask);
5316 out_free_buffer_mask:
5317 free_cpumask_var(tracing_buffer_mask);
5322 __init static int clear_boot_tracer(void)
5325 * The default tracer at boot buffer is an init section.
5326 * This function is called in lateinit. If we did not
5327 * find the boot tracer, then clear it out, to prevent
5328 * later registration from accessing the buffer that is
5329 * about to be freed.
5331 if (!default_bootup_tracer)
5334 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
5335 default_bootup_tracer);
5336 default_bootup_tracer = NULL;
5341 early_initcall(tracer_alloc_buffers);
5342 fs_initcall(tracer_init_debugfs);
5343 late_initcall(clear_boot_tracer);