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>
42 #include <linux/sched/rt.h>
45 #include "trace_output.h"
48 * On boot up, the ring buffer is set to the minimum size, so that
49 * we do not waste memory on systems that are not using tracing.
51 int ring_buffer_expanded;
54 * We need to change this state when a selftest is running.
55 * A selftest will lurk into the ring-buffer to count the
56 * entries inserted during the selftest although some concurrent
57 * insertions into the ring-buffer such as trace_printk could occurred
58 * at the same time, giving false positive or negative results.
60 static bool __read_mostly tracing_selftest_running;
63 * If a tracer is running, we do not want to run SELFTEST.
65 bool __read_mostly tracing_selftest_disabled;
67 /* For tracers that don't implement custom flags */
68 static struct tracer_opt dummy_tracer_opt[] = {
72 static struct tracer_flags dummy_tracer_flags = {
74 .opts = dummy_tracer_opt
77 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
83 * To prevent the comm cache from being overwritten when no
84 * tracing is active, only save the comm when a trace event
87 static DEFINE_PER_CPU(bool, trace_cmdline_save);
90 * When a reader is waiting for data, then this variable is
93 static bool trace_wakeup_needed;
95 static struct irq_work trace_work_wakeup;
98 * Kill all tracing for good (never come back).
99 * It is initialized to 1 but will turn to zero if the initialization
100 * of the tracer is successful. But that is the only place that sets
103 static int tracing_disabled = 1;
105 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
107 cpumask_var_t __read_mostly tracing_buffer_mask;
110 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
112 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
113 * is set, then ftrace_dump is called. This will output the contents
114 * of the ftrace buffers to the console. This is very useful for
115 * capturing traces that lead to crashes and outputing it to a
118 * It is default off, but you can enable it with either specifying
119 * "ftrace_dump_on_oops" in the kernel command line, or setting
120 * /proc/sys/kernel/ftrace_dump_on_oops
121 * Set 1 if you want to dump buffers of all CPUs
122 * Set 2 if you want to dump the buffer of the CPU that triggered oops
125 enum ftrace_dump_mode ftrace_dump_on_oops;
127 static int tracing_set_tracer(const char *buf);
129 #define MAX_TRACER_SIZE 100
130 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
131 static char *default_bootup_tracer;
133 static int __init set_cmdline_ftrace(char *str)
135 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
136 default_bootup_tracer = bootup_tracer_buf;
137 /* We are using ftrace early, expand it */
138 ring_buffer_expanded = 1;
141 __setup("ftrace=", set_cmdline_ftrace);
143 static int __init set_ftrace_dump_on_oops(char *str)
145 if (*str++ != '=' || !*str) {
146 ftrace_dump_on_oops = DUMP_ALL;
150 if (!strcmp("orig_cpu", str)) {
151 ftrace_dump_on_oops = DUMP_ORIG;
157 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
160 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
161 static char *trace_boot_options __initdata;
163 static int __init set_trace_boot_options(char *str)
165 strncpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
166 trace_boot_options = trace_boot_options_buf;
169 __setup("trace_options=", set_trace_boot_options);
171 unsigned long long ns2usecs(cycle_t nsec)
179 * The global_trace is the descriptor that holds the tracing
180 * buffers for the live tracing. For each CPU, it contains
181 * a link list of pages that will store trace entries. The
182 * page descriptor of the pages in the memory is used to hold
183 * the link list by linking the lru item in the page descriptor
184 * to each of the pages in the buffer per CPU.
186 * For each active CPU there is a data field that holds the
187 * pages for the buffer for that CPU. Each CPU has the same number
188 * of pages allocated for its buffer.
190 static struct trace_array global_trace;
192 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
194 int filter_current_check_discard(struct ring_buffer *buffer,
195 struct ftrace_event_call *call, void *rec,
196 struct ring_buffer_event *event)
198 return filter_check_discard(call, rec, buffer, event);
200 EXPORT_SYMBOL_GPL(filter_current_check_discard);
202 cycle_t ftrace_now(int cpu)
206 /* Early boot up does not have a buffer yet */
207 if (!global_trace.buffer)
208 return trace_clock_local();
210 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
211 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
217 * The max_tr is used to snapshot the global_trace when a maximum
218 * latency is reached. Some tracers will use this to store a maximum
219 * trace while it continues examining live traces.
221 * The buffers for the max_tr are set up the same as the global_trace.
222 * When a snapshot is taken, the link list of the max_tr is swapped
223 * with the link list of the global_trace and the buffers are reset for
224 * the global_trace so the tracing can continue.
226 static struct trace_array max_tr;
228 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
230 int tracing_is_enabled(void)
232 return tracing_is_on();
236 * trace_buf_size is the size in bytes that is allocated
237 * for a buffer. Note, the number of bytes is always rounded
240 * This number is purposely set to a low number of 16384.
241 * If the dump on oops happens, it will be much appreciated
242 * to not have to wait for all that output. Anyway this can be
243 * boot time and run time configurable.
245 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
247 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
249 /* trace_types holds a link list of available tracers. */
250 static struct tracer *trace_types __read_mostly;
252 /* current_trace points to the tracer that is currently active */
253 static struct tracer *current_trace __read_mostly = &nop_trace;
256 * trace_types_lock is used to protect the trace_types list.
258 static DEFINE_MUTEX(trace_types_lock);
261 * serialize the access of the ring buffer
263 * ring buffer serializes readers, but it is low level protection.
264 * The validity of the events (which returns by ring_buffer_peek() ..etc)
265 * are not protected by ring buffer.
267 * The content of events may become garbage if we allow other process consumes
268 * these events concurrently:
269 * A) the page of the consumed events may become a normal page
270 * (not reader page) in ring buffer, and this page will be rewrited
271 * by events producer.
272 * B) The page of the consumed events may become a page for splice_read,
273 * and this page will be returned to system.
275 * These primitives allow multi process access to different cpu ring buffer
278 * These primitives don't distinguish read-only and read-consume access.
279 * Multi read-only access are also serialized.
283 static DECLARE_RWSEM(all_cpu_access_lock);
284 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
286 static inline void trace_access_lock(int cpu)
288 if (cpu == TRACE_PIPE_ALL_CPU) {
289 /* gain it for accessing the whole ring buffer. */
290 down_write(&all_cpu_access_lock);
292 /* gain it for accessing a cpu ring buffer. */
294 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
295 down_read(&all_cpu_access_lock);
297 /* Secondly block other access to this @cpu ring buffer. */
298 mutex_lock(&per_cpu(cpu_access_lock, cpu));
302 static inline void trace_access_unlock(int cpu)
304 if (cpu == TRACE_PIPE_ALL_CPU) {
305 up_write(&all_cpu_access_lock);
307 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
308 up_read(&all_cpu_access_lock);
312 static inline void trace_access_lock_init(void)
316 for_each_possible_cpu(cpu)
317 mutex_init(&per_cpu(cpu_access_lock, cpu));
322 static DEFINE_MUTEX(access_lock);
324 static inline void trace_access_lock(int cpu)
327 mutex_lock(&access_lock);
330 static inline void trace_access_unlock(int cpu)
333 mutex_unlock(&access_lock);
336 static inline void trace_access_lock_init(void)
342 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
343 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
345 /* trace_flags holds trace_options default values */
346 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
347 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
348 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
349 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS;
351 static int trace_stop_count;
352 static DEFINE_RAW_SPINLOCK(tracing_start_lock);
355 * trace_wake_up - wake up tasks waiting for trace input
357 * Schedules a delayed work to wake up any task that is blocked on the
358 * trace_wait queue. These is used with trace_poll for tasks polling the
361 static void trace_wake_up(struct irq_work *work)
363 wake_up_all(&trace_wait);
368 * tracing_on - enable tracing buffers
370 * This function enables tracing buffers that may have been
371 * disabled with tracing_off.
373 void tracing_on(void)
375 if (global_trace.buffer)
376 ring_buffer_record_on(global_trace.buffer);
378 * This flag is only looked at when buffers haven't been
379 * allocated yet. We don't really care about the race
380 * between setting this flag and actually turning
383 global_trace.buffer_disabled = 0;
385 EXPORT_SYMBOL_GPL(tracing_on);
388 * tracing_off - turn off tracing buffers
390 * This function stops the tracing buffers from recording data.
391 * It does not disable any overhead the tracers themselves may
392 * be causing. This function simply causes all recording to
393 * the ring buffers to fail.
395 void tracing_off(void)
397 if (global_trace.buffer)
398 ring_buffer_record_off(global_trace.buffer);
400 * This flag is only looked at when buffers haven't been
401 * allocated yet. We don't really care about the race
402 * between setting this flag and actually turning
405 global_trace.buffer_disabled = 1;
407 EXPORT_SYMBOL_GPL(tracing_off);
410 * tracing_is_on - show state of ring buffers enabled
412 int tracing_is_on(void)
414 if (global_trace.buffer)
415 return ring_buffer_record_is_on(global_trace.buffer);
416 return !global_trace.buffer_disabled;
418 EXPORT_SYMBOL_GPL(tracing_is_on);
420 static int __init set_buf_size(char *str)
422 unsigned long buf_size;
426 buf_size = memparse(str, &str);
427 /* nr_entries can not be zero */
430 trace_buf_size = buf_size;
433 __setup("trace_buf_size=", set_buf_size);
435 static int __init set_tracing_thresh(char *str)
437 unsigned long threshold;
442 ret = kstrtoul(str, 0, &threshold);
445 tracing_thresh = threshold * 1000;
448 __setup("tracing_thresh=", set_tracing_thresh);
450 unsigned long nsecs_to_usecs(unsigned long nsecs)
455 /* These must match the bit postions in trace_iterator_flags */
456 static const char *trace_options[] = {
488 int in_ns; /* is this clock in nanoseconds? */
490 { trace_clock_local, "local", 1 },
491 { trace_clock_global, "global", 1 },
492 { trace_clock_counter, "counter", 0 },
499 * trace_parser_get_init - gets the buffer for trace parser
501 int trace_parser_get_init(struct trace_parser *parser, int size)
503 memset(parser, 0, sizeof(*parser));
505 parser->buffer = kmalloc(size, GFP_KERNEL);
514 * trace_parser_put - frees the buffer for trace parser
516 void trace_parser_put(struct trace_parser *parser)
518 kfree(parser->buffer);
522 * trace_get_user - reads the user input string separated by space
523 * (matched by isspace(ch))
525 * For each string found the 'struct trace_parser' is updated,
526 * and the function returns.
528 * Returns number of bytes read.
530 * See kernel/trace/trace.h for 'struct trace_parser' details.
532 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
533 size_t cnt, loff_t *ppos)
540 trace_parser_clear(parser);
542 ret = get_user(ch, ubuf++);
550 * The parser is not finished with the last write,
551 * continue reading the user input without skipping spaces.
554 /* skip white space */
555 while (cnt && isspace(ch)) {
556 ret = get_user(ch, ubuf++);
563 /* only spaces were written */
573 /* read the non-space input */
574 while (cnt && !isspace(ch)) {
575 if (parser->idx < parser->size - 1)
576 parser->buffer[parser->idx++] = ch;
581 ret = get_user(ch, ubuf++);
588 /* We either got finished input or we have to wait for another call. */
590 parser->buffer[parser->idx] = 0;
591 parser->cont = false;
594 parser->buffer[parser->idx++] = ch;
604 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
612 if (s->len <= s->readpos)
615 len = s->len - s->readpos;
618 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
628 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
632 if (s->len <= s->readpos)
635 len = s->len - s->readpos;
638 memcpy(buf, s->buffer + s->readpos, cnt);
645 * ftrace_max_lock is used to protect the swapping of buffers
646 * when taking a max snapshot. The buffers themselves are
647 * protected by per_cpu spinlocks. But the action of the swap
648 * needs its own lock.
650 * This is defined as a arch_spinlock_t in order to help
651 * with performance when lockdep debugging is enabled.
653 * It is also used in other places outside the update_max_tr
654 * so it needs to be defined outside of the
655 * CONFIG_TRACER_MAX_TRACE.
657 static arch_spinlock_t ftrace_max_lock =
658 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
660 unsigned long __read_mostly tracing_thresh;
662 #ifdef CONFIG_TRACER_MAX_TRACE
663 unsigned long __read_mostly tracing_max_latency;
666 * Copy the new maximum trace into the separate maximum-trace
667 * structure. (this way the maximum trace is permanently saved,
668 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
671 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
673 struct trace_array_cpu *data = tr->data[cpu];
674 struct trace_array_cpu *max_data;
677 max_tr.time_start = data->preempt_timestamp;
679 max_data = max_tr.data[cpu];
680 max_data->saved_latency = tracing_max_latency;
681 max_data->critical_start = data->critical_start;
682 max_data->critical_end = data->critical_end;
684 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
685 max_data->pid = tsk->pid;
686 max_data->uid = task_uid(tsk);
687 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
688 max_data->policy = tsk->policy;
689 max_data->rt_priority = tsk->rt_priority;
691 /* record this tasks comm */
692 tracing_record_cmdline(tsk);
696 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
698 * @tsk: the task with the latency
699 * @cpu: The cpu that initiated the trace.
701 * Flip the buffers between the @tr and the max_tr and record information
702 * about which task was the cause of this latency.
705 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
707 struct ring_buffer *buf = tr->buffer;
709 if (trace_stop_count)
712 WARN_ON_ONCE(!irqs_disabled());
714 if (!current_trace->allocated_snapshot) {
715 /* Only the nop tracer should hit this when disabling */
716 WARN_ON_ONCE(current_trace != &nop_trace);
720 arch_spin_lock(&ftrace_max_lock);
722 tr->buffer = max_tr.buffer;
725 __update_max_tr(tr, tsk, cpu);
726 arch_spin_unlock(&ftrace_max_lock);
730 * update_max_tr_single - only copy one trace over, and reset the rest
732 * @tsk - task with the latency
733 * @cpu - the cpu of the buffer to copy.
735 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
738 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
742 if (trace_stop_count)
745 WARN_ON_ONCE(!irqs_disabled());
746 if (WARN_ON_ONCE(!current_trace->allocated_snapshot))
749 arch_spin_lock(&ftrace_max_lock);
751 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
755 * We failed to swap the buffer due to a commit taking
756 * place on this CPU. We fail to record, but we reset
757 * the max trace buffer (no one writes directly to it)
758 * and flag that it failed.
760 trace_array_printk(&max_tr, _THIS_IP_,
761 "Failed to swap buffers due to commit in progress\n");
764 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
766 __update_max_tr(tr, tsk, cpu);
767 arch_spin_unlock(&ftrace_max_lock);
769 #endif /* CONFIG_TRACER_MAX_TRACE */
771 static void default_wait_pipe(struct trace_iterator *iter)
775 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
778 * The events can happen in critical sections where
779 * checking a work queue can cause deadlocks.
780 * After adding a task to the queue, this flag is set
781 * only to notify events to try to wake up the queue
784 * We don't clear it even if the buffer is no longer
785 * empty. The flag only causes the next event to run
786 * irq_work to do the work queue wake up. The worse
787 * that can happen if we race with !trace_empty() is that
788 * an event will cause an irq_work to try to wake up
791 * There's no reason to protect this flag either, as
792 * the work queue and irq_work logic will do the necessary
793 * synchronization for the wake ups. The only thing
794 * that is necessary is that the wake up happens after
795 * a task has been queued. It's OK for spurious wake ups.
797 trace_wakeup_needed = true;
799 if (trace_empty(iter))
802 finish_wait(&trace_wait, &wait);
806 * register_tracer - register a tracer with the ftrace system.
807 * @type - the plugin for the tracer
809 * Register a new plugin tracer.
811 int register_tracer(struct tracer *type)
817 pr_info("Tracer must have a name\n");
821 if (strlen(type->name) >= MAX_TRACER_SIZE) {
822 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
826 mutex_lock(&trace_types_lock);
828 tracing_selftest_running = true;
830 for (t = trace_types; t; t = t->next) {
831 if (strcmp(type->name, t->name) == 0) {
833 pr_info("Tracer %s already registered\n",
841 type->set_flag = &dummy_set_flag;
843 type->flags = &dummy_tracer_flags;
845 if (!type->flags->opts)
846 type->flags->opts = dummy_tracer_opt;
847 if (!type->wait_pipe)
848 type->wait_pipe = default_wait_pipe;
851 #ifdef CONFIG_FTRACE_STARTUP_TEST
852 if (type->selftest && !tracing_selftest_disabled) {
853 struct tracer *saved_tracer = current_trace;
854 struct trace_array *tr = &global_trace;
857 * Run a selftest on this tracer.
858 * Here we reset the trace buffer, and set the current
859 * tracer to be this tracer. The tracer can then run some
860 * internal tracing to verify that everything is in order.
861 * If we fail, we do not register this tracer.
863 tracing_reset_online_cpus(tr);
865 current_trace = type;
867 if (type->use_max_tr) {
868 /* If we expanded the buffers, make sure the max is expanded too */
869 if (ring_buffer_expanded)
870 ring_buffer_resize(max_tr.buffer, trace_buf_size,
871 RING_BUFFER_ALL_CPUS);
872 type->allocated_snapshot = true;
875 /* the test is responsible for initializing and enabling */
876 pr_info("Testing tracer %s: ", type->name);
877 ret = type->selftest(type, tr);
878 /* the test is responsible for resetting too */
879 current_trace = saved_tracer;
881 printk(KERN_CONT "FAILED!\n");
882 /* Add the warning after printing 'FAILED' */
886 /* Only reset on passing, to avoid touching corrupted buffers */
887 tracing_reset_online_cpus(tr);
889 if (type->use_max_tr) {
890 type->allocated_snapshot = false;
892 /* Shrink the max buffer again */
893 if (ring_buffer_expanded)
894 ring_buffer_resize(max_tr.buffer, 1,
895 RING_BUFFER_ALL_CPUS);
898 printk(KERN_CONT "PASSED\n");
902 type->next = trace_types;
906 tracing_selftest_running = false;
907 mutex_unlock(&trace_types_lock);
909 if (ret || !default_bootup_tracer)
912 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
915 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
916 /* Do we want this tracer to start on bootup? */
917 tracing_set_tracer(type->name);
918 default_bootup_tracer = NULL;
919 /* disable other selftests, since this will break it. */
920 tracing_selftest_disabled = 1;
921 #ifdef CONFIG_FTRACE_STARTUP_TEST
922 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
930 void tracing_reset(struct trace_array *tr, int cpu)
932 struct ring_buffer *buffer = tr->buffer;
937 ring_buffer_record_disable(buffer);
939 /* Make sure all commits have finished */
941 ring_buffer_reset_cpu(buffer, cpu);
943 ring_buffer_record_enable(buffer);
946 void tracing_reset_online_cpus(struct trace_array *tr)
948 struct ring_buffer *buffer = tr->buffer;
954 ring_buffer_record_disable(buffer);
956 /* Make sure all commits have finished */
959 tr->time_start = ftrace_now(tr->cpu);
961 for_each_online_cpu(cpu)
962 ring_buffer_reset_cpu(buffer, cpu);
964 ring_buffer_record_enable(buffer);
967 void tracing_reset_current(int cpu)
969 tracing_reset(&global_trace, cpu);
972 void tracing_reset_current_online_cpus(void)
974 tracing_reset_online_cpus(&global_trace);
977 #define SAVED_CMDLINES 128
978 #define NO_CMDLINE_MAP UINT_MAX
979 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
980 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
981 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
982 static int cmdline_idx;
983 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
985 /* temporary disable recording */
986 static atomic_t trace_record_cmdline_disabled __read_mostly;
988 static void trace_init_cmdlines(void)
990 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
991 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
995 int is_tracing_stopped(void)
997 return trace_stop_count;
1001 * ftrace_off_permanent - disable all ftrace code permanently
1003 * This should only be called when a serious anomally has
1004 * been detected. This will turn off the function tracing,
1005 * ring buffers, and other tracing utilites. It takes no
1006 * locks and can be called from any context.
1008 void ftrace_off_permanent(void)
1010 tracing_disabled = 1;
1012 tracing_off_permanent();
1016 * tracing_start - quick start of the tracer
1018 * If tracing is enabled but was stopped by tracing_stop,
1019 * this will start the tracer back up.
1021 void tracing_start(void)
1023 struct ring_buffer *buffer;
1024 unsigned long flags;
1026 if (tracing_disabled)
1029 raw_spin_lock_irqsave(&tracing_start_lock, flags);
1030 if (--trace_stop_count) {
1031 if (trace_stop_count < 0) {
1032 /* Someone screwed up their debugging */
1034 trace_stop_count = 0;
1039 /* Prevent the buffers from switching */
1040 arch_spin_lock(&ftrace_max_lock);
1042 buffer = global_trace.buffer;
1044 ring_buffer_record_enable(buffer);
1046 buffer = max_tr.buffer;
1048 ring_buffer_record_enable(buffer);
1050 arch_spin_unlock(&ftrace_max_lock);
1054 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1058 * tracing_stop - quick stop of the tracer
1060 * Light weight way to stop tracing. Use in conjunction with
1063 void tracing_stop(void)
1065 struct ring_buffer *buffer;
1066 unsigned long flags;
1069 raw_spin_lock_irqsave(&tracing_start_lock, flags);
1070 if (trace_stop_count++)
1073 /* Prevent the buffers from switching */
1074 arch_spin_lock(&ftrace_max_lock);
1076 buffer = global_trace.buffer;
1078 ring_buffer_record_disable(buffer);
1080 buffer = max_tr.buffer;
1082 ring_buffer_record_disable(buffer);
1084 arch_spin_unlock(&ftrace_max_lock);
1087 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1090 void trace_stop_cmdline_recording(void);
1092 static void trace_save_cmdline(struct task_struct *tsk)
1096 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1100 * It's not the end of the world if we don't get
1101 * the lock, but we also don't want to spin
1102 * nor do we want to disable interrupts,
1103 * so if we miss here, then better luck next time.
1105 if (!arch_spin_trylock(&trace_cmdline_lock))
1108 idx = map_pid_to_cmdline[tsk->pid];
1109 if (idx == NO_CMDLINE_MAP) {
1110 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1113 * Check whether the cmdline buffer at idx has a pid
1114 * mapped. We are going to overwrite that entry so we
1115 * need to clear the map_pid_to_cmdline. Otherwise we
1116 * would read the new comm for the old pid.
1118 pid = map_cmdline_to_pid[idx];
1119 if (pid != NO_CMDLINE_MAP)
1120 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1122 map_cmdline_to_pid[idx] = tsk->pid;
1123 map_pid_to_cmdline[tsk->pid] = idx;
1128 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1130 arch_spin_unlock(&trace_cmdline_lock);
1133 void trace_find_cmdline(int pid, char comm[])
1138 strcpy(comm, "<idle>");
1142 if (WARN_ON_ONCE(pid < 0)) {
1143 strcpy(comm, "<XXX>");
1147 if (pid > PID_MAX_DEFAULT) {
1148 strcpy(comm, "<...>");
1153 arch_spin_lock(&trace_cmdline_lock);
1154 map = map_pid_to_cmdline[pid];
1155 if (map != NO_CMDLINE_MAP)
1156 strcpy(comm, saved_cmdlines[map]);
1158 strcpy(comm, "<...>");
1160 arch_spin_unlock(&trace_cmdline_lock);
1164 void tracing_record_cmdline(struct task_struct *tsk)
1166 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1169 if (!__this_cpu_read(trace_cmdline_save))
1172 __this_cpu_write(trace_cmdline_save, false);
1174 trace_save_cmdline(tsk);
1178 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1181 struct task_struct *tsk = current;
1183 entry->preempt_count = pc & 0xff;
1184 entry->pid = (tsk) ? tsk->pid : 0;
1186 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1187 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1189 TRACE_FLAG_IRQS_NOSUPPORT |
1191 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1192 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1193 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1195 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1197 struct ring_buffer_event *
1198 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1201 unsigned long flags, int pc)
1203 struct ring_buffer_event *event;
1205 event = ring_buffer_lock_reserve(buffer, len);
1206 if (event != NULL) {
1207 struct trace_entry *ent = ring_buffer_event_data(event);
1209 tracing_generic_entry_update(ent, flags, pc);
1217 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1219 __this_cpu_write(trace_cmdline_save, true);
1220 if (trace_wakeup_needed) {
1221 trace_wakeup_needed = false;
1222 /* irq_work_queue() supplies it's own memory barriers */
1223 irq_work_queue(&trace_work_wakeup);
1225 ring_buffer_unlock_commit(buffer, event);
1229 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1230 struct ring_buffer_event *event,
1231 unsigned long flags, int pc)
1233 __buffer_unlock_commit(buffer, event);
1235 ftrace_trace_stack(buffer, flags, 6, pc);
1236 ftrace_trace_userstack(buffer, flags, pc);
1239 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1240 struct ring_buffer_event *event,
1241 unsigned long flags, int pc)
1243 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1245 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1247 struct ring_buffer_event *
1248 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1249 int type, unsigned long len,
1250 unsigned long flags, int pc)
1252 *current_rb = global_trace.buffer;
1253 return trace_buffer_lock_reserve(*current_rb,
1254 type, len, flags, pc);
1256 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1258 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1259 struct ring_buffer_event *event,
1260 unsigned long flags, int pc)
1262 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1264 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1266 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1267 struct ring_buffer_event *event,
1268 unsigned long flags, int pc,
1269 struct pt_regs *regs)
1271 __buffer_unlock_commit(buffer, event);
1273 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1274 ftrace_trace_userstack(buffer, flags, pc);
1276 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1278 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1279 struct ring_buffer_event *event)
1281 ring_buffer_discard_commit(buffer, event);
1283 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1286 trace_function(struct trace_array *tr,
1287 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1290 struct ftrace_event_call *call = &event_function;
1291 struct ring_buffer *buffer = tr->buffer;
1292 struct ring_buffer_event *event;
1293 struct ftrace_entry *entry;
1295 /* If we are reading the ring buffer, don't trace */
1296 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1299 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1303 entry = ring_buffer_event_data(event);
1305 entry->parent_ip = parent_ip;
1307 if (!filter_check_discard(call, entry, buffer, event))
1308 __buffer_unlock_commit(buffer, event);
1312 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1313 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1316 if (likely(!atomic_read(&data->disabled)))
1317 trace_function(tr, ip, parent_ip, flags, pc);
1320 #ifdef CONFIG_STACKTRACE
1322 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1323 struct ftrace_stack {
1324 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1327 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1328 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1330 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1331 unsigned long flags,
1332 int skip, int pc, struct pt_regs *regs)
1334 struct ftrace_event_call *call = &event_kernel_stack;
1335 struct ring_buffer_event *event;
1336 struct stack_entry *entry;
1337 struct stack_trace trace;
1339 int size = FTRACE_STACK_ENTRIES;
1341 trace.nr_entries = 0;
1345 * Since events can happen in NMIs there's no safe way to
1346 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1347 * or NMI comes in, it will just have to use the default
1348 * FTRACE_STACK_SIZE.
1350 preempt_disable_notrace();
1352 use_stack = __this_cpu_inc_return(ftrace_stack_reserve);
1354 * We don't need any atomic variables, just a barrier.
1355 * If an interrupt comes in, we don't care, because it would
1356 * have exited and put the counter back to what we want.
1357 * We just need a barrier to keep gcc from moving things
1361 if (use_stack == 1) {
1362 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1363 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1366 save_stack_trace_regs(regs, &trace);
1368 save_stack_trace(&trace);
1370 if (trace.nr_entries > size)
1371 size = trace.nr_entries;
1373 /* From now on, use_stack is a boolean */
1376 size *= sizeof(unsigned long);
1378 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1379 sizeof(*entry) + size, flags, pc);
1382 entry = ring_buffer_event_data(event);
1384 memset(&entry->caller, 0, size);
1387 memcpy(&entry->caller, trace.entries,
1388 trace.nr_entries * sizeof(unsigned long));
1390 trace.max_entries = FTRACE_STACK_ENTRIES;
1391 trace.entries = entry->caller;
1393 save_stack_trace_regs(regs, &trace);
1395 save_stack_trace(&trace);
1398 entry->size = trace.nr_entries;
1400 if (!filter_check_discard(call, entry, buffer, event))
1401 __buffer_unlock_commit(buffer, event);
1404 /* Again, don't let gcc optimize things here */
1406 __this_cpu_dec(ftrace_stack_reserve);
1407 preempt_enable_notrace();
1411 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1412 int skip, int pc, struct pt_regs *regs)
1414 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1417 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1420 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1423 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1426 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1429 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1432 __ftrace_trace_stack(tr->buffer, flags, skip, pc, NULL);
1436 * trace_dump_stack - record a stack back trace in the trace buffer
1438 void trace_dump_stack(void)
1440 unsigned long flags;
1442 if (tracing_disabled || tracing_selftest_running)
1445 local_save_flags(flags);
1447 /* skipping 3 traces, seems to get us at the caller of this function */
1448 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count(), NULL);
1451 static DEFINE_PER_CPU(int, user_stack_count);
1454 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1456 struct ftrace_event_call *call = &event_user_stack;
1457 struct ring_buffer_event *event;
1458 struct userstack_entry *entry;
1459 struct stack_trace trace;
1461 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1465 * NMIs can not handle page faults, even with fix ups.
1466 * The save user stack can (and often does) fault.
1468 if (unlikely(in_nmi()))
1472 * prevent recursion, since the user stack tracing may
1473 * trigger other kernel events.
1476 if (__this_cpu_read(user_stack_count))
1479 __this_cpu_inc(user_stack_count);
1481 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1482 sizeof(*entry), flags, pc);
1484 goto out_drop_count;
1485 entry = ring_buffer_event_data(event);
1487 entry->tgid = current->tgid;
1488 memset(&entry->caller, 0, sizeof(entry->caller));
1490 trace.nr_entries = 0;
1491 trace.max_entries = FTRACE_STACK_ENTRIES;
1493 trace.entries = entry->caller;
1495 save_stack_trace_user(&trace);
1496 if (!filter_check_discard(call, entry, buffer, event))
1497 __buffer_unlock_commit(buffer, event);
1500 __this_cpu_dec(user_stack_count);
1506 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1508 ftrace_trace_userstack(tr, flags, preempt_count());
1512 #endif /* CONFIG_STACKTRACE */
1514 /* created for use with alloc_percpu */
1515 struct trace_buffer_struct {
1516 char buffer[TRACE_BUF_SIZE];
1519 static struct trace_buffer_struct *trace_percpu_buffer;
1520 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1521 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1522 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1525 * The buffer used is dependent on the context. There is a per cpu
1526 * buffer for normal context, softirq contex, hard irq context and
1527 * for NMI context. Thise allows for lockless recording.
1529 * Note, if the buffers failed to be allocated, then this returns NULL
1531 static char *get_trace_buf(void)
1533 struct trace_buffer_struct *percpu_buffer;
1536 * If we have allocated per cpu buffers, then we do not
1537 * need to do any locking.
1540 percpu_buffer = trace_percpu_nmi_buffer;
1542 percpu_buffer = trace_percpu_irq_buffer;
1543 else if (in_softirq())
1544 percpu_buffer = trace_percpu_sirq_buffer;
1546 percpu_buffer = trace_percpu_buffer;
1551 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1554 static int alloc_percpu_trace_buffer(void)
1556 struct trace_buffer_struct *buffers;
1557 struct trace_buffer_struct *sirq_buffers;
1558 struct trace_buffer_struct *irq_buffers;
1559 struct trace_buffer_struct *nmi_buffers;
1561 buffers = alloc_percpu(struct trace_buffer_struct);
1565 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1569 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1573 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1577 trace_percpu_buffer = buffers;
1578 trace_percpu_sirq_buffer = sirq_buffers;
1579 trace_percpu_irq_buffer = irq_buffers;
1580 trace_percpu_nmi_buffer = nmi_buffers;
1585 free_percpu(irq_buffers);
1587 free_percpu(sirq_buffers);
1589 free_percpu(buffers);
1591 WARN(1, "Could not allocate percpu trace_printk buffer");
1595 static int buffers_allocated;
1597 void trace_printk_init_buffers(void)
1599 if (buffers_allocated)
1602 if (alloc_percpu_trace_buffer())
1605 pr_info("ftrace: Allocated trace_printk buffers\n");
1607 /* Expand the buffers to set size */
1608 tracing_update_buffers();
1610 buffers_allocated = 1;
1613 * trace_printk_init_buffers() can be called by modules.
1614 * If that happens, then we need to start cmdline recording
1615 * directly here. If the global_trace.buffer is already
1616 * allocated here, then this was called by module code.
1618 if (global_trace.buffer)
1619 tracing_start_cmdline_record();
1622 void trace_printk_start_comm(void)
1624 /* Start tracing comms if trace printk is set */
1625 if (!buffers_allocated)
1627 tracing_start_cmdline_record();
1630 static void trace_printk_start_stop_comm(int enabled)
1632 if (!buffers_allocated)
1636 tracing_start_cmdline_record();
1638 tracing_stop_cmdline_record();
1642 * trace_vbprintk - write binary msg to tracing buffer
1645 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1647 struct ftrace_event_call *call = &event_bprint;
1648 struct ring_buffer_event *event;
1649 struct ring_buffer *buffer;
1650 struct trace_array *tr = &global_trace;
1651 struct bprint_entry *entry;
1652 unsigned long flags;
1654 int len = 0, size, pc;
1656 if (unlikely(tracing_selftest_running || tracing_disabled))
1659 /* Don't pollute graph traces with trace_vprintk internals */
1660 pause_graph_tracing();
1662 pc = preempt_count();
1663 preempt_disable_notrace();
1665 tbuffer = get_trace_buf();
1671 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1673 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
1676 local_save_flags(flags);
1677 size = sizeof(*entry) + sizeof(u32) * len;
1678 buffer = tr->buffer;
1679 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1683 entry = ring_buffer_event_data(event);
1687 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
1688 if (!filter_check_discard(call, entry, buffer, event)) {
1689 __buffer_unlock_commit(buffer, event);
1690 ftrace_trace_stack(buffer, flags, 6, pc);
1694 preempt_enable_notrace();
1695 unpause_graph_tracing();
1699 EXPORT_SYMBOL_GPL(trace_vbprintk);
1701 int trace_array_printk(struct trace_array *tr,
1702 unsigned long ip, const char *fmt, ...)
1707 if (!(trace_flags & TRACE_ITER_PRINTK))
1711 ret = trace_array_vprintk(tr, ip, fmt, ap);
1716 int trace_array_vprintk(struct trace_array *tr,
1717 unsigned long ip, const char *fmt, va_list args)
1719 struct ftrace_event_call *call = &event_print;
1720 struct ring_buffer_event *event;
1721 struct ring_buffer *buffer;
1722 int len = 0, size, pc;
1723 struct print_entry *entry;
1724 unsigned long flags;
1727 if (tracing_disabled || tracing_selftest_running)
1730 /* Don't pollute graph traces with trace_vprintk internals */
1731 pause_graph_tracing();
1733 pc = preempt_count();
1734 preempt_disable_notrace();
1737 tbuffer = get_trace_buf();
1743 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
1744 if (len > TRACE_BUF_SIZE)
1747 local_save_flags(flags);
1748 size = sizeof(*entry) + len + 1;
1749 buffer = tr->buffer;
1750 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1754 entry = ring_buffer_event_data(event);
1757 memcpy(&entry->buf, tbuffer, len);
1758 entry->buf[len] = '\0';
1759 if (!filter_check_discard(call, entry, buffer, event)) {
1760 __buffer_unlock_commit(buffer, event);
1761 ftrace_trace_stack(buffer, flags, 6, pc);
1764 preempt_enable_notrace();
1765 unpause_graph_tracing();
1770 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1772 return trace_array_vprintk(&global_trace, ip, fmt, args);
1774 EXPORT_SYMBOL_GPL(trace_vprintk);
1776 static void trace_iterator_increment(struct trace_iterator *iter)
1778 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
1782 ring_buffer_read(buf_iter, NULL);
1785 static struct trace_entry *
1786 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1787 unsigned long *lost_events)
1789 struct ring_buffer_event *event;
1790 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
1793 event = ring_buffer_iter_peek(buf_iter, ts);
1795 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1799 iter->ent_size = ring_buffer_event_length(event);
1800 return ring_buffer_event_data(event);
1806 static struct trace_entry *
1807 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1808 unsigned long *missing_events, u64 *ent_ts)
1810 struct ring_buffer *buffer = iter->tr->buffer;
1811 struct trace_entry *ent, *next = NULL;
1812 unsigned long lost_events = 0, next_lost = 0;
1813 int cpu_file = iter->cpu_file;
1814 u64 next_ts = 0, ts;
1820 * If we are in a per_cpu trace file, don't bother by iterating over
1821 * all cpu and peek directly.
1823 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1824 if (ring_buffer_empty_cpu(buffer, cpu_file))
1826 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1828 *ent_cpu = cpu_file;
1833 for_each_tracing_cpu(cpu) {
1835 if (ring_buffer_empty_cpu(buffer, cpu))
1838 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1841 * Pick the entry with the smallest timestamp:
1843 if (ent && (!next || ts < next_ts)) {
1847 next_lost = lost_events;
1848 next_size = iter->ent_size;
1852 iter->ent_size = next_size;
1855 *ent_cpu = next_cpu;
1861 *missing_events = next_lost;
1866 /* Find the next real entry, without updating the iterator itself */
1867 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1868 int *ent_cpu, u64 *ent_ts)
1870 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1873 /* Find the next real entry, and increment the iterator to the next entry */
1874 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1876 iter->ent = __find_next_entry(iter, &iter->cpu,
1877 &iter->lost_events, &iter->ts);
1880 trace_iterator_increment(iter);
1882 return iter->ent ? iter : NULL;
1885 static void trace_consume(struct trace_iterator *iter)
1887 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1888 &iter->lost_events);
1891 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1893 struct trace_iterator *iter = m->private;
1897 WARN_ON_ONCE(iter->leftover);
1901 /* can't go backwards */
1906 ent = trace_find_next_entry_inc(iter);
1910 while (ent && iter->idx < i)
1911 ent = trace_find_next_entry_inc(iter);
1918 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1920 struct trace_array *tr = iter->tr;
1921 struct ring_buffer_event *event;
1922 struct ring_buffer_iter *buf_iter;
1923 unsigned long entries = 0;
1926 tr->data[cpu]->skipped_entries = 0;
1928 buf_iter = trace_buffer_iter(iter, cpu);
1932 ring_buffer_iter_reset(buf_iter);
1935 * We could have the case with the max latency tracers
1936 * that a reset never took place on a cpu. This is evident
1937 * by the timestamp being before the start of the buffer.
1939 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1940 if (ts >= iter->tr->time_start)
1943 ring_buffer_read(buf_iter, NULL);
1946 tr->data[cpu]->skipped_entries = entries;
1950 * The current tracer is copied to avoid a global locking
1953 static void *s_start(struct seq_file *m, loff_t *pos)
1955 struct trace_iterator *iter = m->private;
1956 int cpu_file = iter->cpu_file;
1962 * copy the tracer to avoid using a global lock all around.
1963 * iter->trace is a copy of current_trace, the pointer to the
1964 * name may be used instead of a strcmp(), as iter->trace->name
1965 * will point to the same string as current_trace->name.
1967 mutex_lock(&trace_types_lock);
1968 if (unlikely(current_trace && iter->trace->name != current_trace->name))
1969 *iter->trace = *current_trace;
1970 mutex_unlock(&trace_types_lock);
1972 if (iter->snapshot && iter->trace->use_max_tr)
1973 return ERR_PTR(-EBUSY);
1975 if (!iter->snapshot)
1976 atomic_inc(&trace_record_cmdline_disabled);
1978 if (*pos != iter->pos) {
1983 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1984 for_each_tracing_cpu(cpu)
1985 tracing_iter_reset(iter, cpu);
1987 tracing_iter_reset(iter, cpu_file);
1990 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1995 * If we overflowed the seq_file before, then we want
1996 * to just reuse the trace_seq buffer again.
2002 p = s_next(m, p, &l);
2006 trace_event_read_lock();
2007 trace_access_lock(cpu_file);
2011 static void s_stop(struct seq_file *m, void *p)
2013 struct trace_iterator *iter = m->private;
2015 if (iter->snapshot && iter->trace->use_max_tr)
2018 if (!iter->snapshot)
2019 atomic_dec(&trace_record_cmdline_disabled);
2020 trace_access_unlock(iter->cpu_file);
2021 trace_event_read_unlock();
2025 get_total_entries(struct trace_array *tr, unsigned long *total, unsigned long *entries)
2027 unsigned long count;
2033 for_each_tracing_cpu(cpu) {
2034 count = ring_buffer_entries_cpu(tr->buffer, cpu);
2036 * If this buffer has skipped entries, then we hold all
2037 * entries for the trace and we need to ignore the
2038 * ones before the time stamp.
2040 if (tr->data[cpu]->skipped_entries) {
2041 count -= tr->data[cpu]->skipped_entries;
2042 /* total is the same as the entries */
2046 ring_buffer_overrun_cpu(tr->buffer, cpu);
2051 static void print_lat_help_header(struct seq_file *m)
2053 seq_puts(m, "# _------=> CPU# \n");
2054 seq_puts(m, "# / _-----=> irqs-off \n");
2055 seq_puts(m, "# | / _----=> need-resched \n");
2056 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2057 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2058 seq_puts(m, "# |||| / delay \n");
2059 seq_puts(m, "# cmd pid ||||| time | caller \n");
2060 seq_puts(m, "# \\ / ||||| \\ | / \n");
2063 static void print_event_info(struct trace_array *tr, struct seq_file *m)
2065 unsigned long total;
2066 unsigned long entries;
2068 get_total_entries(tr, &total, &entries);
2069 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2070 entries, total, num_online_cpus());
2074 static void print_func_help_header(struct trace_array *tr, struct seq_file *m)
2076 print_event_info(tr, m);
2077 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2078 seq_puts(m, "# | | | | |\n");
2081 static void print_func_help_header_irq(struct trace_array *tr, struct seq_file *m)
2083 print_event_info(tr, m);
2084 seq_puts(m, "# _-----=> irqs-off\n");
2085 seq_puts(m, "# / _----=> need-resched\n");
2086 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2087 seq_puts(m, "# || / _--=> preempt-depth\n");
2088 seq_puts(m, "# ||| / delay\n");
2089 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2090 seq_puts(m, "# | | | |||| | |\n");
2094 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2096 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2097 struct trace_array *tr = iter->tr;
2098 struct trace_array_cpu *data = tr->data[tr->cpu];
2099 struct tracer *type = current_trace;
2100 unsigned long entries;
2101 unsigned long total;
2102 const char *name = "preemption";
2106 get_total_entries(tr, &total, &entries);
2108 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2110 seq_puts(m, "# -----------------------------------"
2111 "---------------------------------\n");
2112 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2113 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2114 nsecs_to_usecs(data->saved_latency),
2118 #if defined(CONFIG_PREEMPT_NONE)
2120 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2122 #elif defined(CONFIG_PREEMPT)
2127 /* These are reserved for later use */
2130 seq_printf(m, " #P:%d)\n", num_online_cpus());
2134 seq_puts(m, "# -----------------\n");
2135 seq_printf(m, "# | task: %.16s-%d "
2136 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2137 data->comm, data->pid,
2138 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2139 data->policy, data->rt_priority);
2140 seq_puts(m, "# -----------------\n");
2142 if (data->critical_start) {
2143 seq_puts(m, "# => started at: ");
2144 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2145 trace_print_seq(m, &iter->seq);
2146 seq_puts(m, "\n# => ended at: ");
2147 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2148 trace_print_seq(m, &iter->seq);
2149 seq_puts(m, "\n#\n");
2155 static void test_cpu_buff_start(struct trace_iterator *iter)
2157 struct trace_seq *s = &iter->seq;
2159 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2162 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2165 if (cpumask_test_cpu(iter->cpu, iter->started))
2168 if (iter->tr->data[iter->cpu]->skipped_entries)
2171 cpumask_set_cpu(iter->cpu, iter->started);
2173 /* Don't print started cpu buffer for the first entry of the trace */
2175 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2179 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2181 struct trace_seq *s = &iter->seq;
2182 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2183 struct trace_entry *entry;
2184 struct trace_event *event;
2188 test_cpu_buff_start(iter);
2190 event = ftrace_find_event(entry->type);
2192 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2193 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2194 if (!trace_print_lat_context(iter))
2197 if (!trace_print_context(iter))
2203 return event->funcs->trace(iter, sym_flags, event);
2205 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2208 return TRACE_TYPE_HANDLED;
2210 return TRACE_TYPE_PARTIAL_LINE;
2213 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2215 struct trace_seq *s = &iter->seq;
2216 struct trace_entry *entry;
2217 struct trace_event *event;
2221 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2222 if (!trace_seq_printf(s, "%d %d %llu ",
2223 entry->pid, iter->cpu, iter->ts))
2227 event = ftrace_find_event(entry->type);
2229 return event->funcs->raw(iter, 0, event);
2231 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2234 return TRACE_TYPE_HANDLED;
2236 return TRACE_TYPE_PARTIAL_LINE;
2239 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2241 struct trace_seq *s = &iter->seq;
2242 unsigned char newline = '\n';
2243 struct trace_entry *entry;
2244 struct trace_event *event;
2248 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2249 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2250 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2251 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2254 event = ftrace_find_event(entry->type);
2256 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2257 if (ret != TRACE_TYPE_HANDLED)
2261 SEQ_PUT_FIELD_RET(s, newline);
2263 return TRACE_TYPE_HANDLED;
2266 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2268 struct trace_seq *s = &iter->seq;
2269 struct trace_entry *entry;
2270 struct trace_event *event;
2274 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2275 SEQ_PUT_FIELD_RET(s, entry->pid);
2276 SEQ_PUT_FIELD_RET(s, iter->cpu);
2277 SEQ_PUT_FIELD_RET(s, iter->ts);
2280 event = ftrace_find_event(entry->type);
2281 return event ? event->funcs->binary(iter, 0, event) :
2285 int trace_empty(struct trace_iterator *iter)
2287 struct ring_buffer_iter *buf_iter;
2290 /* If we are looking at one CPU buffer, only check that one */
2291 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2292 cpu = iter->cpu_file;
2293 buf_iter = trace_buffer_iter(iter, cpu);
2295 if (!ring_buffer_iter_empty(buf_iter))
2298 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2304 for_each_tracing_cpu(cpu) {
2305 buf_iter = trace_buffer_iter(iter, cpu);
2307 if (!ring_buffer_iter_empty(buf_iter))
2310 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2318 /* Called with trace_event_read_lock() held. */
2319 enum print_line_t print_trace_line(struct trace_iterator *iter)
2321 enum print_line_t ret;
2323 if (iter->lost_events &&
2324 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2325 iter->cpu, iter->lost_events))
2326 return TRACE_TYPE_PARTIAL_LINE;
2328 if (iter->trace && iter->trace->print_line) {
2329 ret = iter->trace->print_line(iter);
2330 if (ret != TRACE_TYPE_UNHANDLED)
2334 if (iter->ent->type == TRACE_BPRINT &&
2335 trace_flags & TRACE_ITER_PRINTK &&
2336 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2337 return trace_print_bprintk_msg_only(iter);
2339 if (iter->ent->type == TRACE_PRINT &&
2340 trace_flags & TRACE_ITER_PRINTK &&
2341 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2342 return trace_print_printk_msg_only(iter);
2344 if (trace_flags & TRACE_ITER_BIN)
2345 return print_bin_fmt(iter);
2347 if (trace_flags & TRACE_ITER_HEX)
2348 return print_hex_fmt(iter);
2350 if (trace_flags & TRACE_ITER_RAW)
2351 return print_raw_fmt(iter);
2353 return print_trace_fmt(iter);
2356 void trace_latency_header(struct seq_file *m)
2358 struct trace_iterator *iter = m->private;
2360 /* print nothing if the buffers are empty */
2361 if (trace_empty(iter))
2364 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2365 print_trace_header(m, iter);
2367 if (!(trace_flags & TRACE_ITER_VERBOSE))
2368 print_lat_help_header(m);
2371 void trace_default_header(struct seq_file *m)
2373 struct trace_iterator *iter = m->private;
2375 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2378 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2379 /* print nothing if the buffers are empty */
2380 if (trace_empty(iter))
2382 print_trace_header(m, iter);
2383 if (!(trace_flags & TRACE_ITER_VERBOSE))
2384 print_lat_help_header(m);
2386 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2387 if (trace_flags & TRACE_ITER_IRQ_INFO)
2388 print_func_help_header_irq(iter->tr, m);
2390 print_func_help_header(iter->tr, m);
2395 static void test_ftrace_alive(struct seq_file *m)
2397 if (!ftrace_is_dead())
2399 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2400 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2403 #ifdef CONFIG_TRACER_MAX_TRACE
2404 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
2406 if (iter->trace->allocated_snapshot)
2407 seq_printf(m, "#\n# * Snapshot is allocated *\n#\n");
2409 seq_printf(m, "#\n# * Snapshot is freed *\n#\n");
2411 seq_printf(m, "# Snapshot commands:\n");
2412 seq_printf(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2413 seq_printf(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2414 seq_printf(m, "# Takes a snapshot of the main buffer.\n");
2415 seq_printf(m, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate)\n");
2416 seq_printf(m, "# (Doesn't have to be '2' works with any number that\n");
2417 seq_printf(m, "# is not a '0' or '1')\n");
2420 /* Should never be called */
2421 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
2424 static int s_show(struct seq_file *m, void *v)
2426 struct trace_iterator *iter = v;
2429 if (iter->ent == NULL) {
2431 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2433 test_ftrace_alive(m);
2435 if (iter->snapshot && trace_empty(iter))
2436 print_snapshot_help(m, iter);
2437 else if (iter->trace && iter->trace->print_header)
2438 iter->trace->print_header(m);
2440 trace_default_header(m);
2442 } else if (iter->leftover) {
2444 * If we filled the seq_file buffer earlier, we
2445 * want to just show it now.
2447 ret = trace_print_seq(m, &iter->seq);
2449 /* ret should this time be zero, but you never know */
2450 iter->leftover = ret;
2453 print_trace_line(iter);
2454 ret = trace_print_seq(m, &iter->seq);
2456 * If we overflow the seq_file buffer, then it will
2457 * ask us for this data again at start up.
2459 * ret is 0 if seq_file write succeeded.
2462 iter->leftover = ret;
2468 static const struct seq_operations tracer_seq_ops = {
2475 static struct trace_iterator *
2476 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
2478 long cpu_file = (long) inode->i_private;
2479 struct trace_iterator *iter;
2482 if (tracing_disabled)
2483 return ERR_PTR(-ENODEV);
2485 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2487 return ERR_PTR(-ENOMEM);
2489 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2491 if (!iter->buffer_iter)
2495 * We make a copy of the current tracer to avoid concurrent
2496 * changes on it while we are reading.
2498 mutex_lock(&trace_types_lock);
2499 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2503 *iter->trace = *current_trace;
2505 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2508 if (current_trace->print_max || snapshot)
2511 iter->tr = &global_trace;
2512 iter->snapshot = snapshot;
2514 mutex_init(&iter->mutex);
2515 iter->cpu_file = cpu_file;
2517 /* Notify the tracer early; before we stop tracing. */
2518 if (iter->trace && iter->trace->open)
2519 iter->trace->open(iter);
2521 /* Annotate start of buffers if we had overruns */
2522 if (ring_buffer_overruns(iter->tr->buffer))
2523 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2525 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2526 if (trace_clocks[trace_clock_id].in_ns)
2527 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2529 /* stop the trace while dumping if we are not opening "snapshot" */
2530 if (!iter->snapshot)
2533 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2534 for_each_tracing_cpu(cpu) {
2535 iter->buffer_iter[cpu] =
2536 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2538 ring_buffer_read_prepare_sync();
2539 for_each_tracing_cpu(cpu) {
2540 ring_buffer_read_start(iter->buffer_iter[cpu]);
2541 tracing_iter_reset(iter, cpu);
2544 cpu = iter->cpu_file;
2545 iter->buffer_iter[cpu] =
2546 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2547 ring_buffer_read_prepare_sync();
2548 ring_buffer_read_start(iter->buffer_iter[cpu]);
2549 tracing_iter_reset(iter, cpu);
2552 mutex_unlock(&trace_types_lock);
2557 mutex_unlock(&trace_types_lock);
2559 kfree(iter->buffer_iter);
2561 seq_release_private(inode, file);
2562 return ERR_PTR(-ENOMEM);
2565 int tracing_open_generic(struct inode *inode, struct file *filp)
2567 if (tracing_disabled)
2570 filp->private_data = inode->i_private;
2574 static int tracing_release(struct inode *inode, struct file *file)
2576 struct seq_file *m = file->private_data;
2577 struct trace_iterator *iter;
2580 if (!(file->f_mode & FMODE_READ))
2585 mutex_lock(&trace_types_lock);
2586 for_each_tracing_cpu(cpu) {
2587 if (iter->buffer_iter[cpu])
2588 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2591 if (iter->trace && iter->trace->close)
2592 iter->trace->close(iter);
2594 if (!iter->snapshot)
2595 /* reenable tracing if it was previously enabled */
2597 mutex_unlock(&trace_types_lock);
2599 mutex_destroy(&iter->mutex);
2600 free_cpumask_var(iter->started);
2602 kfree(iter->buffer_iter);
2603 seq_release_private(inode, file);
2607 static int tracing_open(struct inode *inode, struct file *file)
2609 struct trace_iterator *iter;
2612 /* If this file was open for write, then erase contents */
2613 if ((file->f_mode & FMODE_WRITE) &&
2614 (file->f_flags & O_TRUNC)) {
2615 long cpu = (long) inode->i_private;
2617 if (cpu == TRACE_PIPE_ALL_CPU)
2618 tracing_reset_online_cpus(&global_trace);
2620 tracing_reset(&global_trace, cpu);
2623 if (file->f_mode & FMODE_READ) {
2624 iter = __tracing_open(inode, file, false);
2626 ret = PTR_ERR(iter);
2627 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2628 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2634 t_next(struct seq_file *m, void *v, loff_t *pos)
2636 struct tracer *t = v;
2646 static void *t_start(struct seq_file *m, loff_t *pos)
2651 mutex_lock(&trace_types_lock);
2652 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2658 static void t_stop(struct seq_file *m, void *p)
2660 mutex_unlock(&trace_types_lock);
2663 static int t_show(struct seq_file *m, void *v)
2665 struct tracer *t = v;
2670 seq_printf(m, "%s", t->name);
2679 static const struct seq_operations show_traces_seq_ops = {
2686 static int show_traces_open(struct inode *inode, struct file *file)
2688 if (tracing_disabled)
2691 return seq_open(file, &show_traces_seq_ops);
2695 tracing_write_stub(struct file *filp, const char __user *ubuf,
2696 size_t count, loff_t *ppos)
2701 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2703 if (file->f_mode & FMODE_READ)
2704 return seq_lseek(file, offset, origin);
2709 static const struct file_operations tracing_fops = {
2710 .open = tracing_open,
2712 .write = tracing_write_stub,
2713 .llseek = tracing_seek,
2714 .release = tracing_release,
2717 static const struct file_operations show_traces_fops = {
2718 .open = show_traces_open,
2720 .release = seq_release,
2721 .llseek = seq_lseek,
2725 * Only trace on a CPU if the bitmask is set:
2727 static cpumask_var_t tracing_cpumask;
2730 * The tracer itself will not take this lock, but still we want
2731 * to provide a consistent cpumask to user-space:
2733 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2736 * Temporary storage for the character representation of the
2737 * CPU bitmask (and one more byte for the newline):
2739 static char mask_str[NR_CPUS + 1];
2742 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2743 size_t count, loff_t *ppos)
2747 mutex_lock(&tracing_cpumask_update_lock);
2749 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2750 if (count - len < 2) {
2754 len += sprintf(mask_str + len, "\n");
2755 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2758 mutex_unlock(&tracing_cpumask_update_lock);
2764 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2765 size_t count, loff_t *ppos)
2768 cpumask_var_t tracing_cpumask_new;
2770 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2773 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2777 mutex_lock(&tracing_cpumask_update_lock);
2779 local_irq_disable();
2780 arch_spin_lock(&ftrace_max_lock);
2781 for_each_tracing_cpu(cpu) {
2783 * Increase/decrease the disabled counter if we are
2784 * about to flip a bit in the cpumask:
2786 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2787 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2788 atomic_inc(&global_trace.data[cpu]->disabled);
2789 ring_buffer_record_disable_cpu(global_trace.buffer, cpu);
2791 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2792 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2793 atomic_dec(&global_trace.data[cpu]->disabled);
2794 ring_buffer_record_enable_cpu(global_trace.buffer, cpu);
2797 arch_spin_unlock(&ftrace_max_lock);
2800 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2802 mutex_unlock(&tracing_cpumask_update_lock);
2803 free_cpumask_var(tracing_cpumask_new);
2808 free_cpumask_var(tracing_cpumask_new);
2813 static const struct file_operations tracing_cpumask_fops = {
2814 .open = tracing_open_generic,
2815 .read = tracing_cpumask_read,
2816 .write = tracing_cpumask_write,
2817 .llseek = generic_file_llseek,
2820 static int tracing_trace_options_show(struct seq_file *m, void *v)
2822 struct tracer_opt *trace_opts;
2826 mutex_lock(&trace_types_lock);
2827 tracer_flags = current_trace->flags->val;
2828 trace_opts = current_trace->flags->opts;
2830 for (i = 0; trace_options[i]; i++) {
2831 if (trace_flags & (1 << i))
2832 seq_printf(m, "%s\n", trace_options[i]);
2834 seq_printf(m, "no%s\n", trace_options[i]);
2837 for (i = 0; trace_opts[i].name; i++) {
2838 if (tracer_flags & trace_opts[i].bit)
2839 seq_printf(m, "%s\n", trace_opts[i].name);
2841 seq_printf(m, "no%s\n", trace_opts[i].name);
2843 mutex_unlock(&trace_types_lock);
2848 static int __set_tracer_option(struct tracer *trace,
2849 struct tracer_flags *tracer_flags,
2850 struct tracer_opt *opts, int neg)
2854 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2859 tracer_flags->val &= ~opts->bit;
2861 tracer_flags->val |= opts->bit;
2865 /* Try to assign a tracer specific option */
2866 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2868 struct tracer_flags *tracer_flags = trace->flags;
2869 struct tracer_opt *opts = NULL;
2872 for (i = 0; tracer_flags->opts[i].name; i++) {
2873 opts = &tracer_flags->opts[i];
2875 if (strcmp(cmp, opts->name) == 0)
2876 return __set_tracer_option(trace, trace->flags,
2883 static void set_tracer_flags(unsigned int mask, int enabled)
2885 /* do nothing if flag is already set */
2886 if (!!(trace_flags & mask) == !!enabled)
2890 trace_flags |= mask;
2892 trace_flags &= ~mask;
2894 if (mask == TRACE_ITER_RECORD_CMD)
2895 trace_event_enable_cmd_record(enabled);
2897 if (mask == TRACE_ITER_OVERWRITE)
2898 ring_buffer_change_overwrite(global_trace.buffer, enabled);
2900 if (mask == TRACE_ITER_PRINTK)
2901 trace_printk_start_stop_comm(enabled);
2904 static int trace_set_options(char *option)
2911 cmp = strstrip(option);
2913 if (strncmp(cmp, "no", 2) == 0) {
2918 for (i = 0; trace_options[i]; i++) {
2919 if (strcmp(cmp, trace_options[i]) == 0) {
2920 set_tracer_flags(1 << i, !neg);
2925 /* If no option could be set, test the specific tracer options */
2926 if (!trace_options[i]) {
2927 mutex_lock(&trace_types_lock);
2928 ret = set_tracer_option(current_trace, cmp, neg);
2929 mutex_unlock(&trace_types_lock);
2936 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2937 size_t cnt, loff_t *ppos)
2941 if (cnt >= sizeof(buf))
2944 if (copy_from_user(&buf, ubuf, cnt))
2949 trace_set_options(buf);
2956 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2958 if (tracing_disabled)
2960 return single_open(file, tracing_trace_options_show, NULL);
2963 static const struct file_operations tracing_iter_fops = {
2964 .open = tracing_trace_options_open,
2966 .llseek = seq_lseek,
2967 .release = single_release,
2968 .write = tracing_trace_options_write,
2971 static const char readme_msg[] =
2972 "tracing mini-HOWTO:\n\n"
2973 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2974 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2975 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2976 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2978 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2979 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2981 "# cat /sys/kernel/debug/tracing/trace_options\n"
2982 "noprint-parent nosym-offset nosym-addr noverbose\n"
2983 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2984 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2985 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2986 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2990 tracing_readme_read(struct file *filp, char __user *ubuf,
2991 size_t cnt, loff_t *ppos)
2993 return simple_read_from_buffer(ubuf, cnt, ppos,
2994 readme_msg, strlen(readme_msg));
2997 static const struct file_operations tracing_readme_fops = {
2998 .open = tracing_open_generic,
2999 .read = tracing_readme_read,
3000 .llseek = generic_file_llseek,
3004 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
3005 size_t cnt, loff_t *ppos)
3014 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
3018 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
3026 for (i = 0; i < SAVED_CMDLINES; i++) {
3029 pid = map_cmdline_to_pid[i];
3030 if (pid == -1 || pid == NO_CMDLINE_MAP)
3033 trace_find_cmdline(pid, buf_comm);
3034 r = sprintf(buf, "%d %s\n", pid, buf_comm);
3039 len = simple_read_from_buffer(ubuf, cnt, ppos,
3048 static const struct file_operations tracing_saved_cmdlines_fops = {
3049 .open = tracing_open_generic,
3050 .read = tracing_saved_cmdlines_read,
3051 .llseek = generic_file_llseek,
3055 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3056 size_t cnt, loff_t *ppos)
3058 char buf[MAX_TRACER_SIZE+2];
3061 mutex_lock(&trace_types_lock);
3062 r = sprintf(buf, "%s\n", current_trace->name);
3063 mutex_unlock(&trace_types_lock);
3065 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3068 int tracer_init(struct tracer *t, struct trace_array *tr)
3070 tracing_reset_online_cpus(tr);
3074 static void set_buffer_entries(struct trace_array *tr, unsigned long val)
3077 for_each_tracing_cpu(cpu)
3078 tr->data[cpu]->entries = val;
3081 /* resize @tr's buffer to the size of @size_tr's entries */
3082 static int resize_buffer_duplicate_size(struct trace_array *tr,
3083 struct trace_array *size_tr, int cpu_id)
3087 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3088 for_each_tracing_cpu(cpu) {
3089 ret = ring_buffer_resize(tr->buffer,
3090 size_tr->data[cpu]->entries, cpu);
3093 tr->data[cpu]->entries = size_tr->data[cpu]->entries;
3096 ret = ring_buffer_resize(tr->buffer,
3097 size_tr->data[cpu_id]->entries, cpu_id);
3099 tr->data[cpu_id]->entries =
3100 size_tr->data[cpu_id]->entries;
3106 static int __tracing_resize_ring_buffer(unsigned long size, int cpu)
3111 * If kernel or user changes the size of the ring buffer
3112 * we use the size that was given, and we can forget about
3113 * expanding it later.
3115 ring_buffer_expanded = 1;
3117 /* May be called before buffers are initialized */
3118 if (!global_trace.buffer)
3121 ret = ring_buffer_resize(global_trace.buffer, size, cpu);
3125 if (!current_trace->use_max_tr)
3128 ret = ring_buffer_resize(max_tr.buffer, size, cpu);
3130 int r = resize_buffer_duplicate_size(&global_trace,
3131 &global_trace, cpu);
3134 * AARGH! We are left with different
3135 * size max buffer!!!!
3136 * The max buffer is our "snapshot" buffer.
3137 * When a tracer needs a snapshot (one of the
3138 * latency tracers), it swaps the max buffer
3139 * with the saved snap shot. We succeeded to
3140 * update the size of the main buffer, but failed to
3141 * update the size of the max buffer. But when we tried
3142 * to reset the main buffer to the original size, we
3143 * failed there too. This is very unlikely to
3144 * happen, but if it does, warn and kill all
3148 tracing_disabled = 1;
3153 if (cpu == RING_BUFFER_ALL_CPUS)
3154 set_buffer_entries(&max_tr, size);
3156 max_tr.data[cpu]->entries = size;
3159 if (cpu == RING_BUFFER_ALL_CPUS)
3160 set_buffer_entries(&global_trace, size);
3162 global_trace.data[cpu]->entries = size;
3167 static ssize_t tracing_resize_ring_buffer(unsigned long size, int cpu_id)
3171 mutex_lock(&trace_types_lock);
3173 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3174 /* make sure, this cpu is enabled in the mask */
3175 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3181 ret = __tracing_resize_ring_buffer(size, cpu_id);
3186 mutex_unlock(&trace_types_lock);
3193 * tracing_update_buffers - used by tracing facility to expand ring buffers
3195 * To save on memory when the tracing is never used on a system with it
3196 * configured in. The ring buffers are set to a minimum size. But once
3197 * a user starts to use the tracing facility, then they need to grow
3198 * to their default size.
3200 * This function is to be called when a tracer is about to be used.
3202 int tracing_update_buffers(void)
3206 mutex_lock(&trace_types_lock);
3207 if (!ring_buffer_expanded)
3208 ret = __tracing_resize_ring_buffer(trace_buf_size,
3209 RING_BUFFER_ALL_CPUS);
3210 mutex_unlock(&trace_types_lock);
3215 struct trace_option_dentry;
3217 static struct trace_option_dentry *
3218 create_trace_option_files(struct tracer *tracer);
3221 destroy_trace_option_files(struct trace_option_dentry *topts);
3223 static int tracing_set_tracer(const char *buf)
3225 static struct trace_option_dentry *topts;
3226 struct trace_array *tr = &global_trace;
3231 mutex_lock(&trace_types_lock);
3233 if (!ring_buffer_expanded) {
3234 ret = __tracing_resize_ring_buffer(trace_buf_size,
3235 RING_BUFFER_ALL_CPUS);
3241 for (t = trace_types; t; t = t->next) {
3242 if (strcmp(t->name, buf) == 0)
3249 if (t == current_trace)
3252 trace_branch_disable();
3253 if (current_trace->reset)
3254 current_trace->reset(tr);
3256 had_max_tr = current_trace->allocated_snapshot;
3257 current_trace = &nop_trace;
3259 if (had_max_tr && !t->use_max_tr) {
3261 * We need to make sure that the update_max_tr sees that
3262 * current_trace changed to nop_trace to keep it from
3263 * swapping the buffers after we resize it.
3264 * The update_max_tr is called from interrupts disabled
3265 * so a synchronized_sched() is sufficient.
3267 synchronize_sched();
3269 * We don't free the ring buffer. instead, resize it because
3270 * The max_tr ring buffer has some state (e.g. ring->clock) and
3271 * we want preserve it.
3273 ring_buffer_resize(max_tr.buffer, 1, RING_BUFFER_ALL_CPUS);
3274 set_buffer_entries(&max_tr, 1);
3275 tracing_reset_online_cpus(&max_tr);
3276 current_trace->allocated_snapshot = false;
3278 destroy_trace_option_files(topts);
3280 topts = create_trace_option_files(t);
3281 if (t->use_max_tr && !had_max_tr) {
3282 /* we need to make per cpu buffer sizes equivalent */
3283 ret = resize_buffer_duplicate_size(&max_tr, &global_trace,
3284 RING_BUFFER_ALL_CPUS);
3287 t->allocated_snapshot = true;
3291 ret = tracer_init(t, tr);
3297 trace_branch_enable(tr);
3299 mutex_unlock(&trace_types_lock);
3305 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3306 size_t cnt, loff_t *ppos)
3308 char buf[MAX_TRACER_SIZE+1];
3315 if (cnt > MAX_TRACER_SIZE)
3316 cnt = MAX_TRACER_SIZE;
3318 if (copy_from_user(&buf, ubuf, cnt))
3323 /* strip ending whitespace. */
3324 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3327 err = tracing_set_tracer(buf);
3337 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3338 size_t cnt, loff_t *ppos)
3340 unsigned long *ptr = filp->private_data;
3344 r = snprintf(buf, sizeof(buf), "%ld\n",
3345 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3346 if (r > sizeof(buf))
3348 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3352 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3353 size_t cnt, loff_t *ppos)
3355 unsigned long *ptr = filp->private_data;
3359 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3368 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3370 long cpu_file = (long) inode->i_private;
3371 struct trace_iterator *iter;
3374 if (tracing_disabled)
3377 mutex_lock(&trace_types_lock);
3379 /* create a buffer to store the information to pass to userspace */
3380 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3387 * We make a copy of the current tracer to avoid concurrent
3388 * changes on it while we are reading.
3390 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3395 *iter->trace = *current_trace;
3397 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3402 /* trace pipe does not show start of buffer */
3403 cpumask_setall(iter->started);
3405 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3406 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3408 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3409 if (trace_clocks[trace_clock_id].in_ns)
3410 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3412 iter->cpu_file = cpu_file;
3413 iter->tr = &global_trace;
3414 mutex_init(&iter->mutex);
3415 filp->private_data = iter;
3417 if (iter->trace->pipe_open)
3418 iter->trace->pipe_open(iter);
3420 nonseekable_open(inode, filp);
3422 mutex_unlock(&trace_types_lock);
3428 mutex_unlock(&trace_types_lock);
3432 static int tracing_release_pipe(struct inode *inode, struct file *file)
3434 struct trace_iterator *iter = file->private_data;
3436 mutex_lock(&trace_types_lock);
3438 if (iter->trace->pipe_close)
3439 iter->trace->pipe_close(iter);
3441 mutex_unlock(&trace_types_lock);
3443 free_cpumask_var(iter->started);
3444 mutex_destroy(&iter->mutex);
3452 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3454 struct trace_iterator *iter = filp->private_data;
3456 if (trace_flags & TRACE_ITER_BLOCK) {
3458 * Always select as readable when in blocking mode
3460 return POLLIN | POLLRDNORM;
3462 if (!trace_empty(iter))
3463 return POLLIN | POLLRDNORM;
3464 poll_wait(filp, &trace_wait, poll_table);
3465 if (!trace_empty(iter))
3466 return POLLIN | POLLRDNORM;
3473 * This is a make-shift waitqueue.
3474 * A tracer might use this callback on some rare cases:
3476 * 1) the current tracer might hold the runqueue lock when it wakes up
3477 * a reader, hence a deadlock (sched, function, and function graph tracers)
3478 * 2) the function tracers, trace all functions, we don't want
3479 * the overhead of calling wake_up and friends
3480 * (and tracing them too)
3482 * Anyway, this is really very primitive wakeup.
3484 void poll_wait_pipe(struct trace_iterator *iter)
3486 set_current_state(TASK_INTERRUPTIBLE);
3487 /* sleep for 100 msecs, and try again. */
3488 schedule_timeout(HZ / 10);
3491 /* Must be called with trace_types_lock mutex held. */
3492 static int tracing_wait_pipe(struct file *filp)
3494 struct trace_iterator *iter = filp->private_data;
3496 while (trace_empty(iter)) {
3498 if ((filp->f_flags & O_NONBLOCK)) {
3502 mutex_unlock(&iter->mutex);
3504 iter->trace->wait_pipe(iter);
3506 mutex_lock(&iter->mutex);
3508 if (signal_pending(current))
3512 * We block until we read something and tracing is disabled.
3513 * We still block if tracing is disabled, but we have never
3514 * read anything. This allows a user to cat this file, and
3515 * then enable tracing. But after we have read something,
3516 * we give an EOF when tracing is again disabled.
3518 * iter->pos will be 0 if we haven't read anything.
3520 if (!tracing_is_enabled() && iter->pos)
3531 tracing_read_pipe(struct file *filp, char __user *ubuf,
3532 size_t cnt, loff_t *ppos)
3534 struct trace_iterator *iter = filp->private_data;
3537 /* return any leftover data */
3538 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3542 trace_seq_init(&iter->seq);
3544 /* copy the tracer to avoid using a global lock all around */
3545 mutex_lock(&trace_types_lock);
3546 if (unlikely(iter->trace->name != current_trace->name))
3547 *iter->trace = *current_trace;
3548 mutex_unlock(&trace_types_lock);
3551 * Avoid more than one consumer on a single file descriptor
3552 * This is just a matter of traces coherency, the ring buffer itself
3555 mutex_lock(&iter->mutex);
3556 if (iter->trace->read) {
3557 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3563 sret = tracing_wait_pipe(filp);
3567 /* stop when tracing is finished */
3568 if (trace_empty(iter)) {
3573 if (cnt >= PAGE_SIZE)
3574 cnt = PAGE_SIZE - 1;
3576 /* reset all but tr, trace, and overruns */
3577 memset(&iter->seq, 0,
3578 sizeof(struct trace_iterator) -
3579 offsetof(struct trace_iterator, seq));
3582 trace_event_read_lock();
3583 trace_access_lock(iter->cpu_file);
3584 while (trace_find_next_entry_inc(iter) != NULL) {
3585 enum print_line_t ret;
3586 int len = iter->seq.len;
3588 ret = print_trace_line(iter);
3589 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3590 /* don't print partial lines */
3591 iter->seq.len = len;
3594 if (ret != TRACE_TYPE_NO_CONSUME)
3595 trace_consume(iter);
3597 if (iter->seq.len >= cnt)
3601 * Setting the full flag means we reached the trace_seq buffer
3602 * size and we should leave by partial output condition above.
3603 * One of the trace_seq_* functions is not used properly.
3605 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3608 trace_access_unlock(iter->cpu_file);
3609 trace_event_read_unlock();
3611 /* Now copy what we have to the user */
3612 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3613 if (iter->seq.readpos >= iter->seq.len)
3614 trace_seq_init(&iter->seq);
3617 * If there was nothing to send to user, in spite of consuming trace
3618 * entries, go back to wait for more entries.
3624 mutex_unlock(&iter->mutex);
3629 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3630 struct pipe_buffer *buf)
3632 __free_page(buf->page);
3635 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3638 __free_page(spd->pages[idx]);
3641 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3643 .map = generic_pipe_buf_map,
3644 .unmap = generic_pipe_buf_unmap,
3645 .confirm = generic_pipe_buf_confirm,
3646 .release = tracing_pipe_buf_release,
3647 .steal = generic_pipe_buf_steal,
3648 .get = generic_pipe_buf_get,
3652 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3657 /* Seq buffer is page-sized, exactly what we need. */
3659 count = iter->seq.len;
3660 ret = print_trace_line(iter);
3661 count = iter->seq.len - count;
3664 iter->seq.len -= count;
3667 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3668 iter->seq.len -= count;
3672 if (ret != TRACE_TYPE_NO_CONSUME)
3673 trace_consume(iter);
3675 if (!trace_find_next_entry_inc(iter)) {
3685 static ssize_t tracing_splice_read_pipe(struct file *filp,
3687 struct pipe_inode_info *pipe,
3691 struct page *pages_def[PIPE_DEF_BUFFERS];
3692 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3693 struct trace_iterator *iter = filp->private_data;
3694 struct splice_pipe_desc spd = {
3696 .partial = partial_def,
3697 .nr_pages = 0, /* This gets updated below. */
3698 .nr_pages_max = PIPE_DEF_BUFFERS,
3700 .ops = &tracing_pipe_buf_ops,
3701 .spd_release = tracing_spd_release_pipe,
3707 if (splice_grow_spd(pipe, &spd))
3710 /* copy the tracer to avoid using a global lock all around */
3711 mutex_lock(&trace_types_lock);
3712 if (unlikely(iter->trace->name != current_trace->name))
3713 *iter->trace = *current_trace;
3714 mutex_unlock(&trace_types_lock);
3716 mutex_lock(&iter->mutex);
3718 if (iter->trace->splice_read) {
3719 ret = iter->trace->splice_read(iter, filp,
3720 ppos, pipe, len, flags);
3725 ret = tracing_wait_pipe(filp);
3729 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3734 trace_event_read_lock();
3735 trace_access_lock(iter->cpu_file);
3737 /* Fill as many pages as possible. */
3738 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3739 spd.pages[i] = alloc_page(GFP_KERNEL);
3743 rem = tracing_fill_pipe_page(rem, iter);
3745 /* Copy the data into the page, so we can start over. */
3746 ret = trace_seq_to_buffer(&iter->seq,
3747 page_address(spd.pages[i]),
3750 __free_page(spd.pages[i]);
3753 spd.partial[i].offset = 0;
3754 spd.partial[i].len = iter->seq.len;
3756 trace_seq_init(&iter->seq);
3759 trace_access_unlock(iter->cpu_file);
3760 trace_event_read_unlock();
3761 mutex_unlock(&iter->mutex);
3765 ret = splice_to_pipe(pipe, &spd);
3767 splice_shrink_spd(&spd);
3771 mutex_unlock(&iter->mutex);
3775 struct ftrace_entries_info {
3776 struct trace_array *tr;
3780 static int tracing_entries_open(struct inode *inode, struct file *filp)
3782 struct ftrace_entries_info *info;
3784 if (tracing_disabled)
3787 info = kzalloc(sizeof(*info), GFP_KERNEL);
3791 info->tr = &global_trace;
3792 info->cpu = (unsigned long)inode->i_private;
3794 filp->private_data = info;
3800 tracing_entries_read(struct file *filp, char __user *ubuf,
3801 size_t cnt, loff_t *ppos)
3803 struct ftrace_entries_info *info = filp->private_data;
3804 struct trace_array *tr = info->tr;
3809 mutex_lock(&trace_types_lock);
3811 if (info->cpu == RING_BUFFER_ALL_CPUS) {
3812 int cpu, buf_size_same;
3817 /* check if all cpu sizes are same */
3818 for_each_tracing_cpu(cpu) {
3819 /* fill in the size from first enabled cpu */
3821 size = tr->data[cpu]->entries;
3822 if (size != tr->data[cpu]->entries) {
3828 if (buf_size_same) {
3829 if (!ring_buffer_expanded)
3830 r = sprintf(buf, "%lu (expanded: %lu)\n",
3832 trace_buf_size >> 10);
3834 r = sprintf(buf, "%lu\n", size >> 10);
3836 r = sprintf(buf, "X\n");
3838 r = sprintf(buf, "%lu\n", tr->data[info->cpu]->entries >> 10);
3840 mutex_unlock(&trace_types_lock);
3842 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3847 tracing_entries_write(struct file *filp, const char __user *ubuf,
3848 size_t cnt, loff_t *ppos)
3850 struct ftrace_entries_info *info = filp->private_data;
3854 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3858 /* must have at least 1 entry */
3862 /* value is in KB */
3865 ret = tracing_resize_ring_buffer(val, info->cpu);
3875 tracing_entries_release(struct inode *inode, struct file *filp)
3877 struct ftrace_entries_info *info = filp->private_data;
3885 tracing_total_entries_read(struct file *filp, char __user *ubuf,
3886 size_t cnt, loff_t *ppos)
3888 struct trace_array *tr = filp->private_data;
3891 unsigned long size = 0, expanded_size = 0;
3893 mutex_lock(&trace_types_lock);
3894 for_each_tracing_cpu(cpu) {
3895 size += tr->data[cpu]->entries >> 10;
3896 if (!ring_buffer_expanded)
3897 expanded_size += trace_buf_size >> 10;
3899 if (ring_buffer_expanded)
3900 r = sprintf(buf, "%lu\n", size);
3902 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
3903 mutex_unlock(&trace_types_lock);
3905 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3909 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
3910 size_t cnt, loff_t *ppos)
3913 * There is no need to read what the user has written, this function
3914 * is just to make sure that there is no error when "echo" is used
3923 tracing_free_buffer_release(struct inode *inode, struct file *filp)
3925 /* disable tracing ? */
3926 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
3928 /* resize the ring buffer to 0 */
3929 tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS);
3935 tracing_mark_write(struct file *filp, const char __user *ubuf,
3936 size_t cnt, loff_t *fpos)
3938 unsigned long addr = (unsigned long)ubuf;
3939 struct ring_buffer_event *event;
3940 struct ring_buffer *buffer;
3941 struct print_entry *entry;
3942 unsigned long irq_flags;
3943 struct page *pages[2];
3953 if (tracing_disabled)
3956 if (!(trace_flags & TRACE_ITER_MARKERS))
3959 if (cnt > TRACE_BUF_SIZE)
3960 cnt = TRACE_BUF_SIZE;
3963 * Userspace is injecting traces into the kernel trace buffer.
3964 * We want to be as non intrusive as possible.
3965 * To do so, we do not want to allocate any special buffers
3966 * or take any locks, but instead write the userspace data
3967 * straight into the ring buffer.
3969 * First we need to pin the userspace buffer into memory,
3970 * which, most likely it is, because it just referenced it.
3971 * But there's no guarantee that it is. By using get_user_pages_fast()
3972 * and kmap_atomic/kunmap_atomic() we can get access to the
3973 * pages directly. We then write the data directly into the
3976 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
3978 /* check if we cross pages */
3979 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
3982 offset = addr & (PAGE_SIZE - 1);
3985 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
3986 if (ret < nr_pages) {
3988 put_page(pages[ret]);
3993 for (i = 0; i < nr_pages; i++)
3994 map_page[i] = kmap_atomic(pages[i]);
3996 local_save_flags(irq_flags);
3997 size = sizeof(*entry) + cnt + 2; /* possible \n added */
3998 buffer = global_trace.buffer;
3999 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
4000 irq_flags, preempt_count());
4002 /* Ring buffer disabled, return as if not open for write */
4007 entry = ring_buffer_event_data(event);
4008 entry->ip = _THIS_IP_;
4010 if (nr_pages == 2) {
4011 len = PAGE_SIZE - offset;
4012 memcpy(&entry->buf, map_page[0] + offset, len);
4013 memcpy(&entry->buf[len], map_page[1], cnt - len);
4015 memcpy(&entry->buf, map_page[0] + offset, cnt);
4017 if (entry->buf[cnt - 1] != '\n') {
4018 entry->buf[cnt] = '\n';
4019 entry->buf[cnt + 1] = '\0';
4021 entry->buf[cnt] = '\0';
4023 __buffer_unlock_commit(buffer, event);
4030 for (i = 0; i < nr_pages; i++){
4031 kunmap_atomic(map_page[i]);
4038 static int tracing_clock_show(struct seq_file *m, void *v)
4042 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
4044 "%s%s%s%s", i ? " " : "",
4045 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
4046 i == trace_clock_id ? "]" : "");
4052 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4053 size_t cnt, loff_t *fpos)
4056 const char *clockstr;
4059 if (cnt >= sizeof(buf))
4062 if (copy_from_user(&buf, ubuf, cnt))
4067 clockstr = strstrip(buf);
4069 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4070 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4073 if (i == ARRAY_SIZE(trace_clocks))
4078 mutex_lock(&trace_types_lock);
4080 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
4082 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
4085 * New clock may not be consistent with the previous clock.
4086 * Reset the buffer so that it doesn't have incomparable timestamps.
4088 tracing_reset_online_cpus(&global_trace);
4089 tracing_reset_online_cpus(&max_tr);
4091 mutex_unlock(&trace_types_lock);
4098 static int tracing_clock_open(struct inode *inode, struct file *file)
4100 if (tracing_disabled)
4102 return single_open(file, tracing_clock_show, NULL);
4105 #ifdef CONFIG_TRACER_SNAPSHOT
4106 static int tracing_snapshot_open(struct inode *inode, struct file *file)
4108 struct trace_iterator *iter;
4111 if (file->f_mode & FMODE_READ) {
4112 iter = __tracing_open(inode, file, true);
4114 ret = PTR_ERR(iter);
4120 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
4126 ret = tracing_update_buffers();
4130 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4134 mutex_lock(&trace_types_lock);
4136 if (current_trace->use_max_tr) {
4143 if (current_trace->allocated_snapshot) {
4144 /* free spare buffer */
4145 ring_buffer_resize(max_tr.buffer, 1,
4146 RING_BUFFER_ALL_CPUS);
4147 set_buffer_entries(&max_tr, 1);
4148 tracing_reset_online_cpus(&max_tr);
4149 current_trace->allocated_snapshot = false;
4153 if (!current_trace->allocated_snapshot) {
4154 /* allocate spare buffer */
4155 ret = resize_buffer_duplicate_size(&max_tr,
4156 &global_trace, RING_BUFFER_ALL_CPUS);
4159 current_trace->allocated_snapshot = true;
4162 local_irq_disable();
4163 /* Now, we're going to swap */
4164 update_max_tr(&global_trace, current, smp_processor_id());
4168 if (current_trace->allocated_snapshot)
4169 tracing_reset_online_cpus(&max_tr);
4178 mutex_unlock(&trace_types_lock);
4181 #endif /* CONFIG_TRACER_SNAPSHOT */
4184 static const struct file_operations tracing_max_lat_fops = {
4185 .open = tracing_open_generic,
4186 .read = tracing_max_lat_read,
4187 .write = tracing_max_lat_write,
4188 .llseek = generic_file_llseek,
4191 static const struct file_operations set_tracer_fops = {
4192 .open = tracing_open_generic,
4193 .read = tracing_set_trace_read,
4194 .write = tracing_set_trace_write,
4195 .llseek = generic_file_llseek,
4198 static const struct file_operations tracing_pipe_fops = {
4199 .open = tracing_open_pipe,
4200 .poll = tracing_poll_pipe,
4201 .read = tracing_read_pipe,
4202 .splice_read = tracing_splice_read_pipe,
4203 .release = tracing_release_pipe,
4204 .llseek = no_llseek,
4207 static const struct file_operations tracing_entries_fops = {
4208 .open = tracing_entries_open,
4209 .read = tracing_entries_read,
4210 .write = tracing_entries_write,
4211 .release = tracing_entries_release,
4212 .llseek = generic_file_llseek,
4215 static const struct file_operations tracing_total_entries_fops = {
4216 .open = tracing_open_generic,
4217 .read = tracing_total_entries_read,
4218 .llseek = generic_file_llseek,
4221 static const struct file_operations tracing_free_buffer_fops = {
4222 .write = tracing_free_buffer_write,
4223 .release = tracing_free_buffer_release,
4226 static const struct file_operations tracing_mark_fops = {
4227 .open = tracing_open_generic,
4228 .write = tracing_mark_write,
4229 .llseek = generic_file_llseek,
4232 static const struct file_operations trace_clock_fops = {
4233 .open = tracing_clock_open,
4235 .llseek = seq_lseek,
4236 .release = single_release,
4237 .write = tracing_clock_write,
4240 #ifdef CONFIG_TRACER_SNAPSHOT
4241 static const struct file_operations snapshot_fops = {
4242 .open = tracing_snapshot_open,
4244 .write = tracing_snapshot_write,
4245 .llseek = tracing_seek,
4246 .release = tracing_release,
4248 #endif /* CONFIG_TRACER_SNAPSHOT */
4250 struct ftrace_buffer_info {
4251 struct trace_array *tr;
4257 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4259 int cpu = (int)(long)inode->i_private;
4260 struct ftrace_buffer_info *info;
4262 if (tracing_disabled)
4265 info = kzalloc(sizeof(*info), GFP_KERNEL);
4269 info->tr = &global_trace;
4272 /* Force reading ring buffer for first read */
4273 info->read = (unsigned int)-1;
4275 filp->private_data = info;
4277 return nonseekable_open(inode, filp);
4281 tracing_buffers_read(struct file *filp, char __user *ubuf,
4282 size_t count, loff_t *ppos)
4284 struct ftrace_buffer_info *info = filp->private_data;
4292 info->spare = ring_buffer_alloc_read_page(info->tr->buffer, info->cpu);
4296 /* Do we have previous read data to read? */
4297 if (info->read < PAGE_SIZE)
4300 trace_access_lock(info->cpu);
4301 ret = ring_buffer_read_page(info->tr->buffer,
4305 trace_access_unlock(info->cpu);
4312 size = PAGE_SIZE - info->read;
4316 ret = copy_to_user(ubuf, info->spare + info->read, size);
4327 static int tracing_buffers_release(struct inode *inode, struct file *file)
4329 struct ftrace_buffer_info *info = file->private_data;
4332 ring_buffer_free_read_page(info->tr->buffer, info->spare);
4339 struct ring_buffer *buffer;
4344 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4345 struct pipe_buffer *buf)
4347 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4352 ring_buffer_free_read_page(ref->buffer, ref->page);
4357 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4358 struct pipe_buffer *buf)
4360 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4365 /* Pipe buffer operations for a buffer. */
4366 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4368 .map = generic_pipe_buf_map,
4369 .unmap = generic_pipe_buf_unmap,
4370 .confirm = generic_pipe_buf_confirm,
4371 .release = buffer_pipe_buf_release,
4372 .steal = generic_pipe_buf_steal,
4373 .get = buffer_pipe_buf_get,
4377 * Callback from splice_to_pipe(), if we need to release some pages
4378 * at the end of the spd in case we error'ed out in filling the pipe.
4380 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4382 struct buffer_ref *ref =
4383 (struct buffer_ref *)spd->partial[i].private;
4388 ring_buffer_free_read_page(ref->buffer, ref->page);
4390 spd->partial[i].private = 0;
4394 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4395 struct pipe_inode_info *pipe, size_t len,
4398 struct ftrace_buffer_info *info = file->private_data;
4399 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4400 struct page *pages_def[PIPE_DEF_BUFFERS];
4401 struct splice_pipe_desc spd = {
4403 .partial = partial_def,
4404 .nr_pages_max = PIPE_DEF_BUFFERS,
4406 .ops = &buffer_pipe_buf_ops,
4407 .spd_release = buffer_spd_release,
4409 struct buffer_ref *ref;
4410 int entries, size, i;
4413 if (splice_grow_spd(pipe, &spd))
4416 if (*ppos & (PAGE_SIZE - 1)) {
4421 if (len & (PAGE_SIZE - 1)) {
4422 if (len < PAGE_SIZE) {
4429 trace_access_lock(info->cpu);
4430 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4432 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4436 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4441 ref->buffer = info->tr->buffer;
4442 ref->page = ring_buffer_alloc_read_page(ref->buffer, info->cpu);
4448 r = ring_buffer_read_page(ref->buffer, &ref->page,
4451 ring_buffer_free_read_page(ref->buffer, ref->page);
4457 * zero out any left over data, this is going to
4460 size = ring_buffer_page_len(ref->page);
4461 if (size < PAGE_SIZE)
4462 memset(ref->page + size, 0, PAGE_SIZE - size);
4464 page = virt_to_page(ref->page);
4466 spd.pages[i] = page;
4467 spd.partial[i].len = PAGE_SIZE;
4468 spd.partial[i].offset = 0;
4469 spd.partial[i].private = (unsigned long)ref;
4473 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4476 trace_access_unlock(info->cpu);
4479 /* did we read anything? */
4480 if (!spd.nr_pages) {
4481 if (flags & SPLICE_F_NONBLOCK)
4489 ret = splice_to_pipe(pipe, &spd);
4490 splice_shrink_spd(&spd);
4495 static const struct file_operations tracing_buffers_fops = {
4496 .open = tracing_buffers_open,
4497 .read = tracing_buffers_read,
4498 .release = tracing_buffers_release,
4499 .splice_read = tracing_buffers_splice_read,
4500 .llseek = no_llseek,
4504 tracing_stats_read(struct file *filp, char __user *ubuf,
4505 size_t count, loff_t *ppos)
4507 unsigned long cpu = (unsigned long)filp->private_data;
4508 struct trace_array *tr = &global_trace;
4509 struct trace_seq *s;
4511 unsigned long long t;
4512 unsigned long usec_rem;
4514 s = kmalloc(sizeof(*s), GFP_KERNEL);
4520 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
4521 trace_seq_printf(s, "entries: %ld\n", cnt);
4523 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
4524 trace_seq_printf(s, "overrun: %ld\n", cnt);
4526 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
4527 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
4529 cnt = ring_buffer_bytes_cpu(tr->buffer, cpu);
4530 trace_seq_printf(s, "bytes: %ld\n", cnt);
4532 if (trace_clocks[trace_clock_id].in_ns) {
4533 /* local or global for trace_clock */
4534 t = ns2usecs(ring_buffer_oldest_event_ts(tr->buffer, cpu));
4535 usec_rem = do_div(t, USEC_PER_SEC);
4536 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
4539 t = ns2usecs(ring_buffer_time_stamp(tr->buffer, cpu));
4540 usec_rem = do_div(t, USEC_PER_SEC);
4541 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
4543 /* counter or tsc mode for trace_clock */
4544 trace_seq_printf(s, "oldest event ts: %llu\n",
4545 ring_buffer_oldest_event_ts(tr->buffer, cpu));
4547 trace_seq_printf(s, "now ts: %llu\n",
4548 ring_buffer_time_stamp(tr->buffer, cpu));
4551 cnt = ring_buffer_dropped_events_cpu(tr->buffer, cpu);
4552 trace_seq_printf(s, "dropped events: %ld\n", cnt);
4554 cnt = ring_buffer_read_events_cpu(tr->buffer, cpu);
4555 trace_seq_printf(s, "read events: %ld\n", cnt);
4557 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
4564 static const struct file_operations tracing_stats_fops = {
4565 .open = tracing_open_generic,
4566 .read = tracing_stats_read,
4567 .llseek = generic_file_llseek,
4570 #ifdef CONFIG_DYNAMIC_FTRACE
4572 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
4578 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
4579 size_t cnt, loff_t *ppos)
4581 static char ftrace_dyn_info_buffer[1024];
4582 static DEFINE_MUTEX(dyn_info_mutex);
4583 unsigned long *p = filp->private_data;
4584 char *buf = ftrace_dyn_info_buffer;
4585 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
4588 mutex_lock(&dyn_info_mutex);
4589 r = sprintf(buf, "%ld ", *p);
4591 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
4594 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4596 mutex_unlock(&dyn_info_mutex);
4601 static const struct file_operations tracing_dyn_info_fops = {
4602 .open = tracing_open_generic,
4603 .read = tracing_read_dyn_info,
4604 .llseek = generic_file_llseek,
4608 static struct dentry *d_tracer;
4610 struct dentry *tracing_init_dentry(void)
4617 if (!debugfs_initialized())
4620 d_tracer = debugfs_create_dir("tracing", NULL);
4622 if (!d_tracer && !once) {
4624 pr_warning("Could not create debugfs directory 'tracing'\n");
4631 static struct dentry *d_percpu;
4633 static struct dentry *tracing_dentry_percpu(void)
4636 struct dentry *d_tracer;
4641 d_tracer = tracing_init_dentry();
4646 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4648 if (!d_percpu && !once) {
4650 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4657 static void tracing_init_debugfs_percpu(long cpu)
4659 struct dentry *d_percpu = tracing_dentry_percpu();
4660 struct dentry *d_cpu;
4661 char cpu_dir[30]; /* 30 characters should be more than enough */
4666 snprintf(cpu_dir, 30, "cpu%ld", cpu);
4667 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4669 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4673 /* per cpu trace_pipe */
4674 trace_create_file("trace_pipe", 0444, d_cpu,
4675 (void *) cpu, &tracing_pipe_fops);
4678 trace_create_file("trace", 0644, d_cpu,
4679 (void *) cpu, &tracing_fops);
4681 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4682 (void *) cpu, &tracing_buffers_fops);
4684 trace_create_file("stats", 0444, d_cpu,
4685 (void *) cpu, &tracing_stats_fops);
4687 trace_create_file("buffer_size_kb", 0444, d_cpu,
4688 (void *) cpu, &tracing_entries_fops);
4691 #ifdef CONFIG_FTRACE_SELFTEST
4692 /* Let selftest have access to static functions in this file */
4693 #include "trace_selftest.c"
4696 struct trace_option_dentry {
4697 struct tracer_opt *opt;
4698 struct tracer_flags *flags;
4699 struct dentry *entry;
4703 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4706 struct trace_option_dentry *topt = filp->private_data;
4709 if (topt->flags->val & topt->opt->bit)
4714 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4718 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4721 struct trace_option_dentry *topt = filp->private_data;
4725 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4729 if (val != 0 && val != 1)
4732 if (!!(topt->flags->val & topt->opt->bit) != val) {
4733 mutex_lock(&trace_types_lock);
4734 ret = __set_tracer_option(current_trace, topt->flags,
4736 mutex_unlock(&trace_types_lock);
4747 static const struct file_operations trace_options_fops = {
4748 .open = tracing_open_generic,
4749 .read = trace_options_read,
4750 .write = trace_options_write,
4751 .llseek = generic_file_llseek,
4755 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4758 long index = (long)filp->private_data;
4761 if (trace_flags & (1 << index))
4766 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4770 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4773 long index = (long)filp->private_data;
4777 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4781 if (val != 0 && val != 1)
4783 set_tracer_flags(1 << index, val);
4790 static const struct file_operations trace_options_core_fops = {
4791 .open = tracing_open_generic,
4792 .read = trace_options_core_read,
4793 .write = trace_options_core_write,
4794 .llseek = generic_file_llseek,
4797 struct dentry *trace_create_file(const char *name,
4799 struct dentry *parent,
4801 const struct file_operations *fops)
4805 ret = debugfs_create_file(name, mode, parent, data, fops);
4807 pr_warning("Could not create debugfs '%s' entry\n", name);
4813 static struct dentry *trace_options_init_dentry(void)
4815 struct dentry *d_tracer;
4816 static struct dentry *t_options;
4821 d_tracer = tracing_init_dentry();
4825 t_options = debugfs_create_dir("options", d_tracer);
4827 pr_warning("Could not create debugfs directory 'options'\n");
4835 create_trace_option_file(struct trace_option_dentry *topt,
4836 struct tracer_flags *flags,
4837 struct tracer_opt *opt)
4839 struct dentry *t_options;
4841 t_options = trace_options_init_dentry();
4845 topt->flags = flags;
4848 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4849 &trace_options_fops);
4853 static struct trace_option_dentry *
4854 create_trace_option_files(struct tracer *tracer)
4856 struct trace_option_dentry *topts;
4857 struct tracer_flags *flags;
4858 struct tracer_opt *opts;
4864 flags = tracer->flags;
4866 if (!flags || !flags->opts)
4871 for (cnt = 0; opts[cnt].name; cnt++)
4874 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4878 for (cnt = 0; opts[cnt].name; cnt++)
4879 create_trace_option_file(&topts[cnt], flags,
4886 destroy_trace_option_files(struct trace_option_dentry *topts)
4893 for (cnt = 0; topts[cnt].opt; cnt++) {
4894 if (topts[cnt].entry)
4895 debugfs_remove(topts[cnt].entry);
4901 static struct dentry *
4902 create_trace_option_core_file(const char *option, long index)
4904 struct dentry *t_options;
4906 t_options = trace_options_init_dentry();
4910 return trace_create_file(option, 0644, t_options, (void *)index,
4911 &trace_options_core_fops);
4914 static __init void create_trace_options_dir(void)
4916 struct dentry *t_options;
4919 t_options = trace_options_init_dentry();
4923 for (i = 0; trace_options[i]; i++)
4924 create_trace_option_core_file(trace_options[i], i);
4928 rb_simple_read(struct file *filp, char __user *ubuf,
4929 size_t cnt, loff_t *ppos)
4931 struct trace_array *tr = filp->private_data;
4932 struct ring_buffer *buffer = tr->buffer;
4937 r = ring_buffer_record_is_on(buffer);
4941 r = sprintf(buf, "%d\n", r);
4943 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4947 rb_simple_write(struct file *filp, const char __user *ubuf,
4948 size_t cnt, loff_t *ppos)
4950 struct trace_array *tr = filp->private_data;
4951 struct ring_buffer *buffer = tr->buffer;
4955 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4960 mutex_lock(&trace_types_lock);
4962 ring_buffer_record_on(buffer);
4963 if (current_trace->start)
4964 current_trace->start(tr);
4966 ring_buffer_record_off(buffer);
4967 if (current_trace->stop)
4968 current_trace->stop(tr);
4970 mutex_unlock(&trace_types_lock);
4978 static const struct file_operations rb_simple_fops = {
4979 .open = tracing_open_generic,
4980 .read = rb_simple_read,
4981 .write = rb_simple_write,
4982 .llseek = default_llseek,
4985 static __init int tracer_init_debugfs(void)
4987 struct dentry *d_tracer;
4990 trace_access_lock_init();
4992 d_tracer = tracing_init_dentry();
4994 trace_create_file("trace_options", 0644, d_tracer,
4995 NULL, &tracing_iter_fops);
4997 trace_create_file("tracing_cpumask", 0644, d_tracer,
4998 NULL, &tracing_cpumask_fops);
5000 trace_create_file("trace", 0644, d_tracer,
5001 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
5003 trace_create_file("available_tracers", 0444, d_tracer,
5004 &global_trace, &show_traces_fops);
5006 trace_create_file("current_tracer", 0644, d_tracer,
5007 &global_trace, &set_tracer_fops);
5009 #ifdef CONFIG_TRACER_MAX_TRACE
5010 trace_create_file("tracing_max_latency", 0644, d_tracer,
5011 &tracing_max_latency, &tracing_max_lat_fops);
5014 trace_create_file("tracing_thresh", 0644, d_tracer,
5015 &tracing_thresh, &tracing_max_lat_fops);
5017 trace_create_file("README", 0444, d_tracer,
5018 NULL, &tracing_readme_fops);
5020 trace_create_file("trace_pipe", 0444, d_tracer,
5021 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
5023 trace_create_file("buffer_size_kb", 0644, d_tracer,
5024 (void *) RING_BUFFER_ALL_CPUS, &tracing_entries_fops);
5026 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
5027 &global_trace, &tracing_total_entries_fops);
5029 trace_create_file("free_buffer", 0644, d_tracer,
5030 &global_trace, &tracing_free_buffer_fops);
5032 trace_create_file("trace_marker", 0220, d_tracer,
5033 NULL, &tracing_mark_fops);
5035 trace_create_file("saved_cmdlines", 0444, d_tracer,
5036 NULL, &tracing_saved_cmdlines_fops);
5038 trace_create_file("trace_clock", 0644, d_tracer, NULL,
5041 trace_create_file("tracing_on", 0644, d_tracer,
5042 &global_trace, &rb_simple_fops);
5044 #ifdef CONFIG_DYNAMIC_FTRACE
5045 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
5046 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
5049 #ifdef CONFIG_TRACER_SNAPSHOT
5050 trace_create_file("snapshot", 0644, d_tracer,
5051 (void *) TRACE_PIPE_ALL_CPU, &snapshot_fops);
5054 create_trace_options_dir();
5056 for_each_tracing_cpu(cpu)
5057 tracing_init_debugfs_percpu(cpu);
5062 static int trace_panic_handler(struct notifier_block *this,
5063 unsigned long event, void *unused)
5065 if (ftrace_dump_on_oops)
5066 ftrace_dump(ftrace_dump_on_oops);
5070 static struct notifier_block trace_panic_notifier = {
5071 .notifier_call = trace_panic_handler,
5073 .priority = 150 /* priority: INT_MAX >= x >= 0 */
5076 static int trace_die_handler(struct notifier_block *self,
5082 if (ftrace_dump_on_oops)
5083 ftrace_dump(ftrace_dump_on_oops);
5091 static struct notifier_block trace_die_notifier = {
5092 .notifier_call = trace_die_handler,
5097 * printk is set to max of 1024, we really don't need it that big.
5098 * Nothing should be printing 1000 characters anyway.
5100 #define TRACE_MAX_PRINT 1000
5103 * Define here KERN_TRACE so that we have one place to modify
5104 * it if we decide to change what log level the ftrace dump
5107 #define KERN_TRACE KERN_EMERG
5110 trace_printk_seq(struct trace_seq *s)
5112 /* Probably should print a warning here. */
5116 /* should be zero ended, but we are paranoid. */
5117 s->buffer[s->len] = 0;
5119 printk(KERN_TRACE "%s", s->buffer);
5124 void trace_init_global_iter(struct trace_iterator *iter)
5126 iter->tr = &global_trace;
5127 iter->trace = current_trace;
5128 iter->cpu_file = TRACE_PIPE_ALL_CPU;
5132 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
5134 static arch_spinlock_t ftrace_dump_lock =
5135 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
5136 /* use static because iter can be a bit big for the stack */
5137 static struct trace_iterator iter;
5138 unsigned int old_userobj;
5139 static int dump_ran;
5140 unsigned long flags;
5144 local_irq_save(flags);
5145 arch_spin_lock(&ftrace_dump_lock);
5153 /* Did function tracer already get disabled? */
5154 if (ftrace_is_dead()) {
5155 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
5156 printk("# MAY BE MISSING FUNCTION EVENTS\n");
5159 if (disable_tracing)
5162 /* Simulate the iterator */
5163 trace_init_global_iter(&iter);
5165 for_each_tracing_cpu(cpu) {
5166 atomic_inc(&iter.tr->data[cpu]->disabled);
5169 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
5171 /* don't look at user memory in panic mode */
5172 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
5174 switch (oops_dump_mode) {
5176 iter.cpu_file = TRACE_PIPE_ALL_CPU;
5179 iter.cpu_file = raw_smp_processor_id();
5184 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
5185 iter.cpu_file = TRACE_PIPE_ALL_CPU;
5188 printk(KERN_TRACE "Dumping ftrace buffer:\n");
5191 * We need to stop all tracing on all CPUS to read the
5192 * the next buffer. This is a bit expensive, but is
5193 * not done often. We fill all what we can read,
5194 * and then release the locks again.
5197 while (!trace_empty(&iter)) {
5200 printk(KERN_TRACE "---------------------------------\n");
5204 /* reset all but tr, trace, and overruns */
5205 memset(&iter.seq, 0,
5206 sizeof(struct trace_iterator) -
5207 offsetof(struct trace_iterator, seq));
5208 iter.iter_flags |= TRACE_FILE_LAT_FMT;
5211 if (trace_find_next_entry_inc(&iter) != NULL) {
5214 ret = print_trace_line(&iter);
5215 if (ret != TRACE_TYPE_NO_CONSUME)
5216 trace_consume(&iter);
5218 touch_nmi_watchdog();
5220 trace_printk_seq(&iter.seq);
5224 printk(KERN_TRACE " (ftrace buffer empty)\n");
5226 printk(KERN_TRACE "---------------------------------\n");
5229 /* Re-enable tracing if requested */
5230 if (!disable_tracing) {
5231 trace_flags |= old_userobj;
5233 for_each_tracing_cpu(cpu) {
5234 atomic_dec(&iter.tr->data[cpu]->disabled);
5240 arch_spin_unlock(&ftrace_dump_lock);
5241 local_irq_restore(flags);
5244 /* By default: disable tracing after the dump */
5245 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
5247 __ftrace_dump(true, oops_dump_mode);
5249 EXPORT_SYMBOL_GPL(ftrace_dump);
5251 __init static int tracer_alloc_buffers(void)
5254 enum ring_buffer_flags rb_flags;
5259 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
5262 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
5263 goto out_free_buffer_mask;
5265 /* Only allocate trace_printk buffers if a trace_printk exists */
5266 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
5267 /* Must be called before global_trace.buffer is allocated */
5268 trace_printk_init_buffers();
5270 /* To save memory, keep the ring buffer size to its minimum */
5271 if (ring_buffer_expanded)
5272 ring_buf_size = trace_buf_size;
5276 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5278 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
5279 cpumask_copy(tracing_cpumask, cpu_all_mask);
5281 /* TODO: make the number of buffers hot pluggable with CPUS */
5282 global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
5283 if (!global_trace.buffer) {
5284 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
5286 goto out_free_cpumask;
5288 if (global_trace.buffer_disabled)
5292 #ifdef CONFIG_TRACER_MAX_TRACE
5293 max_tr.buffer = ring_buffer_alloc(1, rb_flags);
5294 if (!max_tr.buffer) {
5295 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
5297 ring_buffer_free(global_trace.buffer);
5298 goto out_free_cpumask;
5302 /* Allocate the first page for all buffers */
5303 for_each_tracing_cpu(i) {
5304 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
5305 max_tr.data[i] = &per_cpu(max_tr_data, i);
5308 set_buffer_entries(&global_trace,
5309 ring_buffer_size(global_trace.buffer, 0));
5310 #ifdef CONFIG_TRACER_MAX_TRACE
5311 set_buffer_entries(&max_tr, 1);
5314 trace_init_cmdlines();
5315 init_irq_work(&trace_work_wakeup, trace_wake_up);
5317 register_tracer(&nop_trace);
5319 /* All seems OK, enable tracing */
5320 tracing_disabled = 0;
5322 atomic_notifier_chain_register(&panic_notifier_list,
5323 &trace_panic_notifier);
5325 register_die_notifier(&trace_die_notifier);
5327 while (trace_boot_options) {
5330 option = strsep(&trace_boot_options, ",");
5331 trace_set_options(option);
5337 free_cpumask_var(tracing_cpumask);
5338 out_free_buffer_mask:
5339 free_cpumask_var(tracing_buffer_mask);
5344 __init static int clear_boot_tracer(void)
5347 * The default tracer at boot buffer is an init section.
5348 * This function is called in lateinit. If we did not
5349 * find the boot tracer, then clear it out, to prevent
5350 * later registration from accessing the buffer that is
5351 * about to be freed.
5353 if (!default_bootup_tracer)
5356 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
5357 default_bootup_tracer);
5358 default_bootup_tracer = NULL;
5363 early_initcall(tracer_alloc_buffers);
5364 fs_initcall(tracer_init_debugfs);
5365 late_initcall(clear_boot_tracer);