2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 Nadia Yvette Chambers
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/irq_work.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/rwsem.h>
36 #include <linux/slab.h>
37 #include <linux/ctype.h>
38 #include <linux/init.h>
39 #include <linux/poll.h>
40 #include <linux/nmi.h>
44 #include "trace_output.h"
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
50 int ring_buffer_expanded;
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
59 static bool __read_mostly tracing_selftest_running;
62 * If a tracer is running, we do not want to run SELFTEST.
64 bool __read_mostly tracing_selftest_disabled;
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt[] = {
71 static struct tracer_flags dummy_tracer_flags = {
73 .opts = dummy_tracer_opt
76 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
82 * To prevent the comm cache from being overwritten when no
83 * tracing is active, only save the comm when a trace event
86 static DEFINE_PER_CPU(bool, trace_cmdline_save);
89 * When a reader is waiting for data, then this variable is
92 static bool trace_wakeup_needed;
94 static struct irq_work trace_work_wakeup;
97 * Kill all tracing for good (never come back).
98 * It is initialized to 1 but will turn to zero if the initialization
99 * of the tracer is successful. But that is the only place that sets
102 static int tracing_disabled = 1;
104 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
106 cpumask_var_t __read_mostly tracing_buffer_mask;
109 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
111 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
112 * is set, then ftrace_dump is called. This will output the contents
113 * of the ftrace buffers to the console. This is very useful for
114 * capturing traces that lead to crashes and outputing it to a
117 * It is default off, but you can enable it with either specifying
118 * "ftrace_dump_on_oops" in the kernel command line, or setting
119 * /proc/sys/kernel/ftrace_dump_on_oops
120 * Set 1 if you want to dump buffers of all CPUs
121 * Set 2 if you want to dump the buffer of the CPU that triggered oops
124 enum ftrace_dump_mode ftrace_dump_on_oops;
126 static int tracing_set_tracer(const char *buf);
128 #define MAX_TRACER_SIZE 100
129 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
130 static char *default_bootup_tracer;
132 static int __init set_cmdline_ftrace(char *str)
134 strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
135 default_bootup_tracer = bootup_tracer_buf;
136 /* We are using ftrace early, expand it */
137 ring_buffer_expanded = 1;
140 __setup("ftrace=", set_cmdline_ftrace);
142 static int __init set_ftrace_dump_on_oops(char *str)
144 if (*str++ != '=' || !*str) {
145 ftrace_dump_on_oops = DUMP_ALL;
149 if (!strcmp("orig_cpu", str)) {
150 ftrace_dump_on_oops = DUMP_ORIG;
156 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
159 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
160 static char *trace_boot_options __initdata;
162 static int __init set_trace_boot_options(char *str)
164 strncpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
165 trace_boot_options = trace_boot_options_buf;
168 __setup("trace_options=", set_trace_boot_options);
170 unsigned long long ns2usecs(cycle_t nsec)
178 * The global_trace is the descriptor that holds the tracing
179 * buffers for the live tracing. For each CPU, it contains
180 * a link list of pages that will store trace entries. The
181 * page descriptor of the pages in the memory is used to hold
182 * the link list by linking the lru item in the page descriptor
183 * to each of the pages in the buffer per CPU.
185 * For each active CPU there is a data field that holds the
186 * pages for the buffer for that CPU. Each CPU has the same number
187 * of pages allocated for its buffer.
189 static struct trace_array global_trace;
191 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
193 int filter_current_check_discard(struct ring_buffer *buffer,
194 struct ftrace_event_call *call, void *rec,
195 struct ring_buffer_event *event)
197 return filter_check_discard(call, rec, buffer, event);
199 EXPORT_SYMBOL_GPL(filter_current_check_discard);
201 cycle_t ftrace_now(int cpu)
205 /* Early boot up does not have a buffer yet */
206 if (!global_trace.buffer)
207 return trace_clock_local();
209 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
210 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
216 * The max_tr is used to snapshot the global_trace when a maximum
217 * latency is reached. Some tracers will use this to store a maximum
218 * trace while it continues examining live traces.
220 * The buffers for the max_tr are set up the same as the global_trace.
221 * When a snapshot is taken, the link list of the max_tr is swapped
222 * with the link list of the global_trace and the buffers are reset for
223 * the global_trace so the tracing can continue.
225 static struct trace_array max_tr;
227 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
229 int tracing_is_enabled(void)
231 return tracing_is_on();
235 * trace_buf_size is the size in bytes that is allocated
236 * for a buffer. Note, the number of bytes is always rounded
239 * This number is purposely set to a low number of 16384.
240 * If the dump on oops happens, it will be much appreciated
241 * to not have to wait for all that output. Anyway this can be
242 * boot time and run time configurable.
244 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
246 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
248 /* trace_types holds a link list of available tracers. */
249 static struct tracer *trace_types __read_mostly;
251 /* current_trace points to the tracer that is currently active */
252 static struct tracer *current_trace __read_mostly;
255 * trace_types_lock is used to protect the trace_types list.
257 static DEFINE_MUTEX(trace_types_lock);
260 * serialize the access of the ring buffer
262 * ring buffer serializes readers, but it is low level protection.
263 * The validity of the events (which returns by ring_buffer_peek() ..etc)
264 * are not protected by ring buffer.
266 * The content of events may become garbage if we allow other process consumes
267 * these events concurrently:
268 * A) the page of the consumed events may become a normal page
269 * (not reader page) in ring buffer, and this page will be rewrited
270 * by events producer.
271 * B) The page of the consumed events may become a page for splice_read,
272 * and this page will be returned to system.
274 * These primitives allow multi process access to different cpu ring buffer
277 * These primitives don't distinguish read-only and read-consume access.
278 * Multi read-only access are also serialized.
282 static DECLARE_RWSEM(all_cpu_access_lock);
283 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
285 static inline void trace_access_lock(int cpu)
287 if (cpu == TRACE_PIPE_ALL_CPU) {
288 /* gain it for accessing the whole ring buffer. */
289 down_write(&all_cpu_access_lock);
291 /* gain it for accessing a cpu ring buffer. */
293 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
294 down_read(&all_cpu_access_lock);
296 /* Secondly block other access to this @cpu ring buffer. */
297 mutex_lock(&per_cpu(cpu_access_lock, cpu));
301 static inline void trace_access_unlock(int cpu)
303 if (cpu == TRACE_PIPE_ALL_CPU) {
304 up_write(&all_cpu_access_lock);
306 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
307 up_read(&all_cpu_access_lock);
311 static inline void trace_access_lock_init(void)
315 for_each_possible_cpu(cpu)
316 mutex_init(&per_cpu(cpu_access_lock, cpu));
321 static DEFINE_MUTEX(access_lock);
323 static inline void trace_access_lock(int cpu)
326 mutex_lock(&access_lock);
329 static inline void trace_access_unlock(int cpu)
332 mutex_unlock(&access_lock);
335 static inline void trace_access_lock_init(void)
341 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
342 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
344 /* trace_flags holds trace_options default values */
345 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
346 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
347 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
348 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS;
350 static int trace_stop_count;
351 static DEFINE_RAW_SPINLOCK(tracing_start_lock);
354 * trace_wake_up - wake up tasks waiting for trace input
356 * Schedules a delayed work to wake up any task that is blocked on the
357 * trace_wait queue. These is used with trace_poll for tasks polling the
360 static void trace_wake_up(struct irq_work *work)
362 wake_up_all(&trace_wait);
367 * tracing_on - enable tracing buffers
369 * This function enables tracing buffers that may have been
370 * disabled with tracing_off.
372 void tracing_on(void)
374 if (global_trace.buffer)
375 ring_buffer_record_on(global_trace.buffer);
377 * This flag is only looked at when buffers haven't been
378 * allocated yet. We don't really care about the race
379 * between setting this flag and actually turning
382 global_trace.buffer_disabled = 0;
384 EXPORT_SYMBOL_GPL(tracing_on);
387 * tracing_off - turn off tracing buffers
389 * This function stops the tracing buffers from recording data.
390 * It does not disable any overhead the tracers themselves may
391 * be causing. This function simply causes all recording to
392 * the ring buffers to fail.
394 void tracing_off(void)
396 if (global_trace.buffer)
397 ring_buffer_record_off(global_trace.buffer);
399 * This flag is only looked at when buffers haven't been
400 * allocated yet. We don't really care about the race
401 * between setting this flag and actually turning
404 global_trace.buffer_disabled = 1;
406 EXPORT_SYMBOL_GPL(tracing_off);
409 * tracing_is_on - show state of ring buffers enabled
411 int tracing_is_on(void)
413 if (global_trace.buffer)
414 return ring_buffer_record_is_on(global_trace.buffer);
415 return !global_trace.buffer_disabled;
417 EXPORT_SYMBOL_GPL(tracing_is_on);
419 static int __init set_buf_size(char *str)
421 unsigned long buf_size;
425 buf_size = memparse(str, &str);
426 /* nr_entries can not be zero */
429 trace_buf_size = buf_size;
432 __setup("trace_buf_size=", set_buf_size);
434 static int __init set_tracing_thresh(char *str)
436 unsigned long threshold;
441 ret = kstrtoul(str, 0, &threshold);
444 tracing_thresh = threshold * 1000;
447 __setup("tracing_thresh=", set_tracing_thresh);
449 unsigned long nsecs_to_usecs(unsigned long nsecs)
454 /* These must match the bit postions in trace_iterator_flags */
455 static const char *trace_options[] = {
487 int in_ns; /* is this clock in nanoseconds? */
489 { trace_clock_local, "local", 1 },
490 { trace_clock_global, "global", 1 },
491 { trace_clock_counter, "counter", 0 },
498 * trace_parser_get_init - gets the buffer for trace parser
500 int trace_parser_get_init(struct trace_parser *parser, int size)
502 memset(parser, 0, sizeof(*parser));
504 parser->buffer = kmalloc(size, GFP_KERNEL);
513 * trace_parser_put - frees the buffer for trace parser
515 void trace_parser_put(struct trace_parser *parser)
517 kfree(parser->buffer);
521 * trace_get_user - reads the user input string separated by space
522 * (matched by isspace(ch))
524 * For each string found the 'struct trace_parser' is updated,
525 * and the function returns.
527 * Returns number of bytes read.
529 * See kernel/trace/trace.h for 'struct trace_parser' details.
531 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
532 size_t cnt, loff_t *ppos)
539 trace_parser_clear(parser);
541 ret = get_user(ch, ubuf++);
549 * The parser is not finished with the last write,
550 * continue reading the user input without skipping spaces.
553 /* skip white space */
554 while (cnt && isspace(ch)) {
555 ret = get_user(ch, ubuf++);
562 /* only spaces were written */
572 /* read the non-space input */
573 while (cnt && !isspace(ch)) {
574 if (parser->idx < parser->size - 1)
575 parser->buffer[parser->idx++] = ch;
580 ret = get_user(ch, ubuf++);
587 /* We either got finished input or we have to wait for another call. */
589 parser->buffer[parser->idx] = 0;
590 parser->cont = false;
593 parser->buffer[parser->idx++] = ch;
603 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
611 if (s->len <= s->readpos)
614 len = s->len - s->readpos;
617 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
627 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
631 if (s->len <= s->readpos)
634 len = s->len - s->readpos;
637 memcpy(buf, s->buffer + s->readpos, cnt);
644 * ftrace_max_lock is used to protect the swapping of buffers
645 * when taking a max snapshot. The buffers themselves are
646 * protected by per_cpu spinlocks. But the action of the swap
647 * needs its own lock.
649 * This is defined as a arch_spinlock_t in order to help
650 * with performance when lockdep debugging is enabled.
652 * It is also used in other places outside the update_max_tr
653 * so it needs to be defined outside of the
654 * CONFIG_TRACER_MAX_TRACE.
656 static arch_spinlock_t ftrace_max_lock =
657 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
659 unsigned long __read_mostly tracing_thresh;
661 #ifdef CONFIG_TRACER_MAX_TRACE
662 unsigned long __read_mostly tracing_max_latency;
665 * Copy the new maximum trace into the separate maximum-trace
666 * structure. (this way the maximum trace is permanently saved,
667 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
670 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
672 struct trace_array_cpu *data = tr->data[cpu];
673 struct trace_array_cpu *max_data;
676 max_tr.time_start = data->preempt_timestamp;
678 max_data = max_tr.data[cpu];
679 max_data->saved_latency = tracing_max_latency;
680 max_data->critical_start = data->critical_start;
681 max_data->critical_end = data->critical_end;
683 memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
684 max_data->pid = tsk->pid;
685 max_data->uid = task_uid(tsk);
686 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
687 max_data->policy = tsk->policy;
688 max_data->rt_priority = tsk->rt_priority;
690 /* record this tasks comm */
691 tracing_record_cmdline(tsk);
695 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
697 * @tsk: the task with the latency
698 * @cpu: The cpu that initiated the trace.
700 * Flip the buffers between the @tr and the max_tr and record information
701 * about which task was the cause of this latency.
704 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
706 struct ring_buffer *buf = tr->buffer;
708 if (trace_stop_count)
711 WARN_ON_ONCE(!irqs_disabled());
712 if (!current_trace->use_max_tr) {
716 arch_spin_lock(&ftrace_max_lock);
718 tr->buffer = max_tr.buffer;
721 __update_max_tr(tr, tsk, cpu);
722 arch_spin_unlock(&ftrace_max_lock);
726 * update_max_tr_single - only copy one trace over, and reset the rest
728 * @tsk - task with the latency
729 * @cpu - the cpu of the buffer to copy.
731 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
734 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
738 if (trace_stop_count)
741 WARN_ON_ONCE(!irqs_disabled());
742 if (!current_trace->use_max_tr) {
747 arch_spin_lock(&ftrace_max_lock);
749 ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
753 * We failed to swap the buffer due to a commit taking
754 * place on this CPU. We fail to record, but we reset
755 * the max trace buffer (no one writes directly to it)
756 * and flag that it failed.
758 trace_array_printk(&max_tr, _THIS_IP_,
759 "Failed to swap buffers due to commit in progress\n");
762 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
764 __update_max_tr(tr, tsk, cpu);
765 arch_spin_unlock(&ftrace_max_lock);
767 #endif /* CONFIG_TRACER_MAX_TRACE */
769 static void default_wait_pipe(struct trace_iterator *iter)
773 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
776 * The events can happen in critical sections where
777 * checking a work queue can cause deadlocks.
778 * After adding a task to the queue, this flag is set
779 * only to notify events to try to wake up the queue
782 * We don't clear it even if the buffer is no longer
783 * empty. The flag only causes the next event to run
784 * irq_work to do the work queue wake up. The worse
785 * that can happen if we race with !trace_empty() is that
786 * an event will cause an irq_work to try to wake up
789 * There's no reason to protect this flag either, as
790 * the work queue and irq_work logic will do the necessary
791 * synchronization for the wake ups. The only thing
792 * that is necessary is that the wake up happens after
793 * a task has been queued. It's OK for spurious wake ups.
795 trace_wakeup_needed = true;
797 if (trace_empty(iter))
800 finish_wait(&trace_wait, &wait);
804 * register_tracer - register a tracer with the ftrace system.
805 * @type - the plugin for the tracer
807 * Register a new plugin tracer.
809 int register_tracer(struct tracer *type)
815 pr_info("Tracer must have a name\n");
819 if (strlen(type->name) >= MAX_TRACER_SIZE) {
820 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
824 mutex_lock(&trace_types_lock);
826 tracing_selftest_running = true;
828 for (t = trace_types; t; t = t->next) {
829 if (strcmp(type->name, t->name) == 0) {
831 pr_info("Tracer %s already registered\n",
839 type->set_flag = &dummy_set_flag;
841 type->flags = &dummy_tracer_flags;
843 if (!type->flags->opts)
844 type->flags->opts = dummy_tracer_opt;
845 if (!type->wait_pipe)
846 type->wait_pipe = default_wait_pipe;
849 #ifdef CONFIG_FTRACE_STARTUP_TEST
850 if (type->selftest && !tracing_selftest_disabled) {
851 struct tracer *saved_tracer = current_trace;
852 struct trace_array *tr = &global_trace;
855 * Run a selftest on this tracer.
856 * Here we reset the trace buffer, and set the current
857 * tracer to be this tracer. The tracer can then run some
858 * internal tracing to verify that everything is in order.
859 * If we fail, we do not register this tracer.
861 tracing_reset_online_cpus(tr);
863 current_trace = type;
865 /* If we expanded the buffers, make sure the max is expanded too */
866 if (ring_buffer_expanded && type->use_max_tr)
867 ring_buffer_resize(max_tr.buffer, trace_buf_size,
868 RING_BUFFER_ALL_CPUS);
870 /* the test is responsible for initializing and enabling */
871 pr_info("Testing tracer %s: ", type->name);
872 ret = type->selftest(type, tr);
873 /* the test is responsible for resetting too */
874 current_trace = saved_tracer;
876 printk(KERN_CONT "FAILED!\n");
877 /* Add the warning after printing 'FAILED' */
881 /* Only reset on passing, to avoid touching corrupted buffers */
882 tracing_reset_online_cpus(tr);
884 /* Shrink the max buffer again */
885 if (ring_buffer_expanded && type->use_max_tr)
886 ring_buffer_resize(max_tr.buffer, 1,
887 RING_BUFFER_ALL_CPUS);
889 printk(KERN_CONT "PASSED\n");
893 type->next = trace_types;
897 tracing_selftest_running = false;
898 mutex_unlock(&trace_types_lock);
900 if (ret || !default_bootup_tracer)
903 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
906 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
907 /* Do we want this tracer to start on bootup? */
908 tracing_set_tracer(type->name);
909 default_bootup_tracer = NULL;
910 /* disable other selftests, since this will break it. */
911 tracing_selftest_disabled = 1;
912 #ifdef CONFIG_FTRACE_STARTUP_TEST
913 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
921 void tracing_reset(struct trace_array *tr, int cpu)
923 struct ring_buffer *buffer = tr->buffer;
925 ring_buffer_record_disable(buffer);
927 /* Make sure all commits have finished */
929 ring_buffer_reset_cpu(buffer, cpu);
931 ring_buffer_record_enable(buffer);
934 void tracing_reset_online_cpus(struct trace_array *tr)
936 struct ring_buffer *buffer = tr->buffer;
939 ring_buffer_record_disable(buffer);
941 /* Make sure all commits have finished */
944 tr->time_start = ftrace_now(tr->cpu);
946 for_each_online_cpu(cpu)
947 ring_buffer_reset_cpu(buffer, cpu);
949 ring_buffer_record_enable(buffer);
952 void tracing_reset_current(int cpu)
954 tracing_reset(&global_trace, cpu);
957 void tracing_reset_current_online_cpus(void)
959 tracing_reset_online_cpus(&global_trace);
962 #define SAVED_CMDLINES 128
963 #define NO_CMDLINE_MAP UINT_MAX
964 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
965 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
966 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
967 static int cmdline_idx;
968 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
970 /* temporary disable recording */
971 static atomic_t trace_record_cmdline_disabled __read_mostly;
973 static void trace_init_cmdlines(void)
975 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
976 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
980 int is_tracing_stopped(void)
982 return trace_stop_count;
986 * ftrace_off_permanent - disable all ftrace code permanently
988 * This should only be called when a serious anomally has
989 * been detected. This will turn off the function tracing,
990 * ring buffers, and other tracing utilites. It takes no
991 * locks and can be called from any context.
993 void ftrace_off_permanent(void)
995 tracing_disabled = 1;
997 tracing_off_permanent();
1001 * tracing_start - quick start of the tracer
1003 * If tracing is enabled but was stopped by tracing_stop,
1004 * this will start the tracer back up.
1006 void tracing_start(void)
1008 struct ring_buffer *buffer;
1009 unsigned long flags;
1011 if (tracing_disabled)
1014 raw_spin_lock_irqsave(&tracing_start_lock, flags);
1015 if (--trace_stop_count) {
1016 if (trace_stop_count < 0) {
1017 /* Someone screwed up their debugging */
1019 trace_stop_count = 0;
1024 /* Prevent the buffers from switching */
1025 arch_spin_lock(&ftrace_max_lock);
1027 buffer = global_trace.buffer;
1029 ring_buffer_record_enable(buffer);
1031 buffer = max_tr.buffer;
1033 ring_buffer_record_enable(buffer);
1035 arch_spin_unlock(&ftrace_max_lock);
1039 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1043 * tracing_stop - quick stop of the tracer
1045 * Light weight way to stop tracing. Use in conjunction with
1048 void tracing_stop(void)
1050 struct ring_buffer *buffer;
1051 unsigned long flags;
1054 raw_spin_lock_irqsave(&tracing_start_lock, flags);
1055 if (trace_stop_count++)
1058 /* Prevent the buffers from switching */
1059 arch_spin_lock(&ftrace_max_lock);
1061 buffer = global_trace.buffer;
1063 ring_buffer_record_disable(buffer);
1065 buffer = max_tr.buffer;
1067 ring_buffer_record_disable(buffer);
1069 arch_spin_unlock(&ftrace_max_lock);
1072 raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1075 void trace_stop_cmdline_recording(void);
1077 static void trace_save_cmdline(struct task_struct *tsk)
1081 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1085 * It's not the end of the world if we don't get
1086 * the lock, but we also don't want to spin
1087 * nor do we want to disable interrupts,
1088 * so if we miss here, then better luck next time.
1090 if (!arch_spin_trylock(&trace_cmdline_lock))
1093 idx = map_pid_to_cmdline[tsk->pid];
1094 if (idx == NO_CMDLINE_MAP) {
1095 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1098 * Check whether the cmdline buffer at idx has a pid
1099 * mapped. We are going to overwrite that entry so we
1100 * need to clear the map_pid_to_cmdline. Otherwise we
1101 * would read the new comm for the old pid.
1103 pid = map_cmdline_to_pid[idx];
1104 if (pid != NO_CMDLINE_MAP)
1105 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1107 map_cmdline_to_pid[idx] = tsk->pid;
1108 map_pid_to_cmdline[tsk->pid] = idx;
1113 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1115 arch_spin_unlock(&trace_cmdline_lock);
1118 void trace_find_cmdline(int pid, char comm[])
1123 strcpy(comm, "<idle>");
1127 if (WARN_ON_ONCE(pid < 0)) {
1128 strcpy(comm, "<XXX>");
1132 if (pid > PID_MAX_DEFAULT) {
1133 strcpy(comm, "<...>");
1138 arch_spin_lock(&trace_cmdline_lock);
1139 map = map_pid_to_cmdline[pid];
1140 if (map != NO_CMDLINE_MAP)
1141 strcpy(comm, saved_cmdlines[map]);
1143 strcpy(comm, "<...>");
1145 arch_spin_unlock(&trace_cmdline_lock);
1149 void tracing_record_cmdline(struct task_struct *tsk)
1151 if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
1154 if (!__this_cpu_read(trace_cmdline_save))
1157 __this_cpu_write(trace_cmdline_save, false);
1159 trace_save_cmdline(tsk);
1163 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1166 struct task_struct *tsk = current;
1168 entry->preempt_count = pc & 0xff;
1169 entry->pid = (tsk) ? tsk->pid : 0;
1172 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1173 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1175 TRACE_FLAG_IRQS_NOSUPPORT |
1177 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1178 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1179 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1181 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1183 struct ring_buffer_event *
1184 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1187 unsigned long flags, int pc)
1189 struct ring_buffer_event *event;
1191 event = ring_buffer_lock_reserve(buffer, len);
1192 if (event != NULL) {
1193 struct trace_entry *ent = ring_buffer_event_data(event);
1195 tracing_generic_entry_update(ent, flags, pc);
1203 __buffer_unlock_commit(struct ring_buffer *buffer, struct ring_buffer_event *event)
1205 __this_cpu_write(trace_cmdline_save, true);
1206 if (trace_wakeup_needed) {
1207 trace_wakeup_needed = false;
1208 /* irq_work_queue() supplies it's own memory barriers */
1209 irq_work_queue(&trace_work_wakeup);
1211 ring_buffer_unlock_commit(buffer, event);
1215 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1216 struct ring_buffer_event *event,
1217 unsigned long flags, int pc)
1219 __buffer_unlock_commit(buffer, event);
1221 ftrace_trace_stack(buffer, flags, 6, pc);
1222 ftrace_trace_userstack(buffer, flags, pc);
1225 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1226 struct ring_buffer_event *event,
1227 unsigned long flags, int pc)
1229 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1231 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit);
1233 struct ring_buffer_event *
1234 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1235 int type, unsigned long len,
1236 unsigned long flags, int pc)
1238 *current_rb = global_trace.buffer;
1239 return trace_buffer_lock_reserve(*current_rb,
1240 type, len, flags, pc);
1242 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1244 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1245 struct ring_buffer_event *event,
1246 unsigned long flags, int pc)
1248 __trace_buffer_unlock_commit(buffer, event, flags, pc);
1250 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1252 void trace_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1253 struct ring_buffer_event *event,
1254 unsigned long flags, int pc,
1255 struct pt_regs *regs)
1257 __buffer_unlock_commit(buffer, event);
1259 ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1260 ftrace_trace_userstack(buffer, flags, pc);
1262 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs);
1264 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1265 struct ring_buffer_event *event)
1267 ring_buffer_discard_commit(buffer, event);
1269 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1272 trace_function(struct trace_array *tr,
1273 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1276 struct ftrace_event_call *call = &event_function;
1277 struct ring_buffer *buffer = tr->buffer;
1278 struct ring_buffer_event *event;
1279 struct ftrace_entry *entry;
1281 /* If we are reading the ring buffer, don't trace */
1282 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1285 event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1289 entry = ring_buffer_event_data(event);
1291 entry->parent_ip = parent_ip;
1293 if (!filter_check_discard(call, entry, buffer, event))
1294 __buffer_unlock_commit(buffer, event);
1298 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1299 unsigned long ip, unsigned long parent_ip, unsigned long flags,
1302 if (likely(!atomic_read(&data->disabled)))
1303 trace_function(tr, ip, parent_ip, flags, pc);
1306 #ifdef CONFIG_STACKTRACE
1308 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1309 struct ftrace_stack {
1310 unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1313 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1314 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1316 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1317 unsigned long flags,
1318 int skip, int pc, struct pt_regs *regs)
1320 struct ftrace_event_call *call = &event_kernel_stack;
1321 struct ring_buffer_event *event;
1322 struct stack_entry *entry;
1323 struct stack_trace trace;
1325 int size = FTRACE_STACK_ENTRIES;
1327 trace.nr_entries = 0;
1331 * Since events can happen in NMIs there's no safe way to
1332 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1333 * or NMI comes in, it will just have to use the default
1334 * FTRACE_STACK_SIZE.
1336 preempt_disable_notrace();
1338 use_stack = ++__get_cpu_var(ftrace_stack_reserve);
1340 * We don't need any atomic variables, just a barrier.
1341 * If an interrupt comes in, we don't care, because it would
1342 * have exited and put the counter back to what we want.
1343 * We just need a barrier to keep gcc from moving things
1347 if (use_stack == 1) {
1348 trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1349 trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1352 save_stack_trace_regs(regs, &trace);
1354 save_stack_trace(&trace);
1356 if (trace.nr_entries > size)
1357 size = trace.nr_entries;
1359 /* From now on, use_stack is a boolean */
1362 size *= sizeof(unsigned long);
1364 event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1365 sizeof(*entry) + size, flags, pc);
1368 entry = ring_buffer_event_data(event);
1370 memset(&entry->caller, 0, size);
1373 memcpy(&entry->caller, trace.entries,
1374 trace.nr_entries * sizeof(unsigned long));
1376 trace.max_entries = FTRACE_STACK_ENTRIES;
1377 trace.entries = entry->caller;
1379 save_stack_trace_regs(regs, &trace);
1381 save_stack_trace(&trace);
1384 entry->size = trace.nr_entries;
1386 if (!filter_check_discard(call, entry, buffer, event))
1387 __buffer_unlock_commit(buffer, event);
1390 /* Again, don't let gcc optimize things here */
1392 __get_cpu_var(ftrace_stack_reserve)--;
1393 preempt_enable_notrace();
1397 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1398 int skip, int pc, struct pt_regs *regs)
1400 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1403 __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1406 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1409 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1412 __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1415 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1418 __ftrace_trace_stack(tr->buffer, flags, skip, pc, NULL);
1422 * trace_dump_stack - record a stack back trace in the trace buffer
1424 void trace_dump_stack(void)
1426 unsigned long flags;
1428 if (tracing_disabled || tracing_selftest_running)
1431 local_save_flags(flags);
1433 /* skipping 3 traces, seems to get us at the caller of this function */
1434 __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count(), NULL);
1437 static DEFINE_PER_CPU(int, user_stack_count);
1440 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1442 struct ftrace_event_call *call = &event_user_stack;
1443 struct ring_buffer_event *event;
1444 struct userstack_entry *entry;
1445 struct stack_trace trace;
1447 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1451 * NMIs can not handle page faults, even with fix ups.
1452 * The save user stack can (and often does) fault.
1454 if (unlikely(in_nmi()))
1458 * prevent recursion, since the user stack tracing may
1459 * trigger other kernel events.
1462 if (__this_cpu_read(user_stack_count))
1465 __this_cpu_inc(user_stack_count);
1467 event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1468 sizeof(*entry), flags, pc);
1470 goto out_drop_count;
1471 entry = ring_buffer_event_data(event);
1473 entry->tgid = current->tgid;
1474 memset(&entry->caller, 0, sizeof(entry->caller));
1476 trace.nr_entries = 0;
1477 trace.max_entries = FTRACE_STACK_ENTRIES;
1479 trace.entries = entry->caller;
1481 save_stack_trace_user(&trace);
1482 if (!filter_check_discard(call, entry, buffer, event))
1483 __buffer_unlock_commit(buffer, event);
1486 __this_cpu_dec(user_stack_count);
1492 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1494 ftrace_trace_userstack(tr, flags, preempt_count());
1498 #endif /* CONFIG_STACKTRACE */
1500 /* created for use with alloc_percpu */
1501 struct trace_buffer_struct {
1502 char buffer[TRACE_BUF_SIZE];
1505 static struct trace_buffer_struct *trace_percpu_buffer;
1506 static struct trace_buffer_struct *trace_percpu_sirq_buffer;
1507 static struct trace_buffer_struct *trace_percpu_irq_buffer;
1508 static struct trace_buffer_struct *trace_percpu_nmi_buffer;
1511 * The buffer used is dependent on the context. There is a per cpu
1512 * buffer for normal context, softirq contex, hard irq context and
1513 * for NMI context. Thise allows for lockless recording.
1515 * Note, if the buffers failed to be allocated, then this returns NULL
1517 static char *get_trace_buf(void)
1519 struct trace_buffer_struct *percpu_buffer;
1522 * If we have allocated per cpu buffers, then we do not
1523 * need to do any locking.
1526 percpu_buffer = trace_percpu_nmi_buffer;
1528 percpu_buffer = trace_percpu_irq_buffer;
1529 else if (in_softirq())
1530 percpu_buffer = trace_percpu_sirq_buffer;
1532 percpu_buffer = trace_percpu_buffer;
1537 return this_cpu_ptr(&percpu_buffer->buffer[0]);
1540 static int alloc_percpu_trace_buffer(void)
1542 struct trace_buffer_struct *buffers;
1543 struct trace_buffer_struct *sirq_buffers;
1544 struct trace_buffer_struct *irq_buffers;
1545 struct trace_buffer_struct *nmi_buffers;
1547 buffers = alloc_percpu(struct trace_buffer_struct);
1551 sirq_buffers = alloc_percpu(struct trace_buffer_struct);
1555 irq_buffers = alloc_percpu(struct trace_buffer_struct);
1559 nmi_buffers = alloc_percpu(struct trace_buffer_struct);
1563 trace_percpu_buffer = buffers;
1564 trace_percpu_sirq_buffer = sirq_buffers;
1565 trace_percpu_irq_buffer = irq_buffers;
1566 trace_percpu_nmi_buffer = nmi_buffers;
1571 free_percpu(irq_buffers);
1573 free_percpu(sirq_buffers);
1575 free_percpu(buffers);
1577 WARN(1, "Could not allocate percpu trace_printk buffer");
1581 static int buffers_allocated;
1583 void trace_printk_init_buffers(void)
1585 if (buffers_allocated)
1588 if (alloc_percpu_trace_buffer())
1591 pr_info("ftrace: Allocated trace_printk buffers\n");
1593 /* Expand the buffers to set size */
1594 tracing_update_buffers();
1596 buffers_allocated = 1;
1599 * trace_printk_init_buffers() can be called by modules.
1600 * If that happens, then we need to start cmdline recording
1601 * directly here. If the global_trace.buffer is already
1602 * allocated here, then this was called by module code.
1604 if (global_trace.buffer)
1605 tracing_start_cmdline_record();
1608 void trace_printk_start_comm(void)
1610 /* Start tracing comms if trace printk is set */
1611 if (!buffers_allocated)
1613 tracing_start_cmdline_record();
1616 static void trace_printk_start_stop_comm(int enabled)
1618 if (!buffers_allocated)
1622 tracing_start_cmdline_record();
1624 tracing_stop_cmdline_record();
1628 * trace_vbprintk - write binary msg to tracing buffer
1631 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1633 struct ftrace_event_call *call = &event_bprint;
1634 struct ring_buffer_event *event;
1635 struct ring_buffer *buffer;
1636 struct trace_array *tr = &global_trace;
1637 struct bprint_entry *entry;
1638 unsigned long flags;
1640 int len = 0, size, pc;
1642 if (unlikely(tracing_selftest_running || tracing_disabled))
1645 /* Don't pollute graph traces with trace_vprintk internals */
1646 pause_graph_tracing();
1648 pc = preempt_count();
1649 preempt_disable_notrace();
1651 tbuffer = get_trace_buf();
1657 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
1659 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
1662 local_save_flags(flags);
1663 size = sizeof(*entry) + sizeof(u32) * len;
1664 buffer = tr->buffer;
1665 event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1669 entry = ring_buffer_event_data(event);
1673 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
1674 if (!filter_check_discard(call, entry, buffer, event)) {
1675 __buffer_unlock_commit(buffer, event);
1676 ftrace_trace_stack(buffer, flags, 6, pc);
1680 preempt_enable_notrace();
1681 unpause_graph_tracing();
1685 EXPORT_SYMBOL_GPL(trace_vbprintk);
1687 int trace_array_printk(struct trace_array *tr,
1688 unsigned long ip, const char *fmt, ...)
1693 if (!(trace_flags & TRACE_ITER_PRINTK))
1697 ret = trace_array_vprintk(tr, ip, fmt, ap);
1702 int trace_array_vprintk(struct trace_array *tr,
1703 unsigned long ip, const char *fmt, va_list args)
1705 struct ftrace_event_call *call = &event_print;
1706 struct ring_buffer_event *event;
1707 struct ring_buffer *buffer;
1708 int len = 0, size, pc;
1709 struct print_entry *entry;
1710 unsigned long flags;
1713 if (tracing_disabled || tracing_selftest_running)
1716 /* Don't pollute graph traces with trace_vprintk internals */
1717 pause_graph_tracing();
1719 pc = preempt_count();
1720 preempt_disable_notrace();
1723 tbuffer = get_trace_buf();
1729 len = vsnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
1730 if (len > TRACE_BUF_SIZE)
1733 local_save_flags(flags);
1734 size = sizeof(*entry) + len + 1;
1735 buffer = tr->buffer;
1736 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1740 entry = ring_buffer_event_data(event);
1743 memcpy(&entry->buf, tbuffer, len);
1744 entry->buf[len] = '\0';
1745 if (!filter_check_discard(call, entry, buffer, event)) {
1746 __buffer_unlock_commit(buffer, event);
1747 ftrace_trace_stack(buffer, flags, 6, pc);
1750 preempt_enable_notrace();
1751 unpause_graph_tracing();
1756 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1758 return trace_array_vprintk(&global_trace, ip, fmt, args);
1760 EXPORT_SYMBOL_GPL(trace_vprintk);
1762 static void trace_iterator_increment(struct trace_iterator *iter)
1764 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
1768 ring_buffer_read(buf_iter, NULL);
1771 static struct trace_entry *
1772 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1773 unsigned long *lost_events)
1775 struct ring_buffer_event *event;
1776 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
1779 event = ring_buffer_iter_peek(buf_iter, ts);
1781 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1785 iter->ent_size = ring_buffer_event_length(event);
1786 return ring_buffer_event_data(event);
1792 static struct trace_entry *
1793 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1794 unsigned long *missing_events, u64 *ent_ts)
1796 struct ring_buffer *buffer = iter->tr->buffer;
1797 struct trace_entry *ent, *next = NULL;
1798 unsigned long lost_events = 0, next_lost = 0;
1799 int cpu_file = iter->cpu_file;
1800 u64 next_ts = 0, ts;
1806 * If we are in a per_cpu trace file, don't bother by iterating over
1807 * all cpu and peek directly.
1809 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1810 if (ring_buffer_empty_cpu(buffer, cpu_file))
1812 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1814 *ent_cpu = cpu_file;
1819 for_each_tracing_cpu(cpu) {
1821 if (ring_buffer_empty_cpu(buffer, cpu))
1824 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1827 * Pick the entry with the smallest timestamp:
1829 if (ent && (!next || ts < next_ts)) {
1833 next_lost = lost_events;
1834 next_size = iter->ent_size;
1838 iter->ent_size = next_size;
1841 *ent_cpu = next_cpu;
1847 *missing_events = next_lost;
1852 /* Find the next real entry, without updating the iterator itself */
1853 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1854 int *ent_cpu, u64 *ent_ts)
1856 return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1859 /* Find the next real entry, and increment the iterator to the next entry */
1860 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1862 iter->ent = __find_next_entry(iter, &iter->cpu,
1863 &iter->lost_events, &iter->ts);
1866 trace_iterator_increment(iter);
1868 return iter->ent ? iter : NULL;
1871 static void trace_consume(struct trace_iterator *iter)
1873 ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1874 &iter->lost_events);
1877 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1879 struct trace_iterator *iter = m->private;
1883 WARN_ON_ONCE(iter->leftover);
1887 /* can't go backwards */
1892 ent = trace_find_next_entry_inc(iter);
1896 while (ent && iter->idx < i)
1897 ent = trace_find_next_entry_inc(iter);
1904 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1906 struct trace_array *tr = iter->tr;
1907 struct ring_buffer_event *event;
1908 struct ring_buffer_iter *buf_iter;
1909 unsigned long entries = 0;
1912 tr->data[cpu]->skipped_entries = 0;
1914 buf_iter = trace_buffer_iter(iter, cpu);
1918 ring_buffer_iter_reset(buf_iter);
1921 * We could have the case with the max latency tracers
1922 * that a reset never took place on a cpu. This is evident
1923 * by the timestamp being before the start of the buffer.
1925 while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1926 if (ts >= iter->tr->time_start)
1929 ring_buffer_read(buf_iter, NULL);
1932 tr->data[cpu]->skipped_entries = entries;
1936 * The current tracer is copied to avoid a global locking
1939 static void *s_start(struct seq_file *m, loff_t *pos)
1941 struct trace_iterator *iter = m->private;
1942 static struct tracer *old_tracer;
1943 int cpu_file = iter->cpu_file;
1948 /* copy the tracer to avoid using a global lock all around */
1949 mutex_lock(&trace_types_lock);
1950 if (unlikely(old_tracer != current_trace && current_trace)) {
1951 old_tracer = current_trace;
1952 *iter->trace = *current_trace;
1954 mutex_unlock(&trace_types_lock);
1956 atomic_inc(&trace_record_cmdline_disabled);
1958 if (*pos != iter->pos) {
1963 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1964 for_each_tracing_cpu(cpu)
1965 tracing_iter_reset(iter, cpu);
1967 tracing_iter_reset(iter, cpu_file);
1970 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1975 * If we overflowed the seq_file before, then we want
1976 * to just reuse the trace_seq buffer again.
1982 p = s_next(m, p, &l);
1986 trace_event_read_lock();
1987 trace_access_lock(cpu_file);
1991 static void s_stop(struct seq_file *m, void *p)
1993 struct trace_iterator *iter = m->private;
1995 atomic_dec(&trace_record_cmdline_disabled);
1996 trace_access_unlock(iter->cpu_file);
1997 trace_event_read_unlock();
2001 get_total_entries(struct trace_array *tr, unsigned long *total, unsigned long *entries)
2003 unsigned long count;
2009 for_each_tracing_cpu(cpu) {
2010 count = ring_buffer_entries_cpu(tr->buffer, cpu);
2012 * If this buffer has skipped entries, then we hold all
2013 * entries for the trace and we need to ignore the
2014 * ones before the time stamp.
2016 if (tr->data[cpu]->skipped_entries) {
2017 count -= tr->data[cpu]->skipped_entries;
2018 /* total is the same as the entries */
2022 ring_buffer_overrun_cpu(tr->buffer, cpu);
2027 static void print_lat_help_header(struct seq_file *m)
2029 seq_puts(m, "# _------=> CPU# \n");
2030 seq_puts(m, "# / _-----=> irqs-off \n");
2031 seq_puts(m, "# | / _----=> need-resched \n");
2032 seq_puts(m, "# || / _---=> hardirq/softirq \n");
2033 seq_puts(m, "# ||| / _--=> preempt-depth \n");
2034 seq_puts(m, "# |||| / delay \n");
2035 seq_puts(m, "# cmd pid ||||| time | caller \n");
2036 seq_puts(m, "# \\ / ||||| \\ | / \n");
2039 static void print_event_info(struct trace_array *tr, struct seq_file *m)
2041 unsigned long total;
2042 unsigned long entries;
2044 get_total_entries(tr, &total, &entries);
2045 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2046 entries, total, num_online_cpus());
2050 static void print_func_help_header(struct trace_array *tr, struct seq_file *m)
2052 print_event_info(tr, m);
2053 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2054 seq_puts(m, "# | | | | |\n");
2057 static void print_func_help_header_irq(struct trace_array *tr, struct seq_file *m)
2059 print_event_info(tr, m);
2060 seq_puts(m, "# _-----=> irqs-off\n");
2061 seq_puts(m, "# / _----=> need-resched\n");
2062 seq_puts(m, "# | / _---=> hardirq/softirq\n");
2063 seq_puts(m, "# || / _--=> preempt-depth\n");
2064 seq_puts(m, "# ||| / delay\n");
2065 seq_puts(m, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2066 seq_puts(m, "# | | | |||| | |\n");
2070 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2072 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2073 struct trace_array *tr = iter->tr;
2074 struct trace_array_cpu *data = tr->data[tr->cpu];
2075 struct tracer *type = current_trace;
2076 unsigned long entries;
2077 unsigned long total;
2078 const char *name = "preemption";
2083 get_total_entries(tr, &total, &entries);
2085 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
2087 seq_puts(m, "# -----------------------------------"
2088 "---------------------------------\n");
2089 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2090 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2091 nsecs_to_usecs(data->saved_latency),
2095 #if defined(CONFIG_PREEMPT_NONE)
2097 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2099 #elif defined(CONFIG_PREEMPT)
2104 /* These are reserved for later use */
2107 seq_printf(m, " #P:%d)\n", num_online_cpus());
2111 seq_puts(m, "# -----------------\n");
2112 seq_printf(m, "# | task: %.16s-%d "
2113 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2114 data->comm, data->pid,
2115 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2116 data->policy, data->rt_priority);
2117 seq_puts(m, "# -----------------\n");
2119 if (data->critical_start) {
2120 seq_puts(m, "# => started at: ");
2121 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2122 trace_print_seq(m, &iter->seq);
2123 seq_puts(m, "\n# => ended at: ");
2124 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2125 trace_print_seq(m, &iter->seq);
2126 seq_puts(m, "\n#\n");
2132 static void test_cpu_buff_start(struct trace_iterator *iter)
2134 struct trace_seq *s = &iter->seq;
2136 if (!(trace_flags & TRACE_ITER_ANNOTATE))
2139 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2142 if (cpumask_test_cpu(iter->cpu, iter->started))
2145 if (iter->tr->data[iter->cpu]->skipped_entries)
2148 cpumask_set_cpu(iter->cpu, iter->started);
2150 /* Don't print started cpu buffer for the first entry of the trace */
2152 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2156 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2158 struct trace_seq *s = &iter->seq;
2159 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2160 struct trace_entry *entry;
2161 struct trace_event *event;
2165 test_cpu_buff_start(iter);
2167 event = ftrace_find_event(entry->type);
2169 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2170 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2171 if (!trace_print_lat_context(iter))
2174 if (!trace_print_context(iter))
2180 return event->funcs->trace(iter, sym_flags, event);
2182 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2185 return TRACE_TYPE_HANDLED;
2187 return TRACE_TYPE_PARTIAL_LINE;
2190 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2192 struct trace_seq *s = &iter->seq;
2193 struct trace_entry *entry;
2194 struct trace_event *event;
2198 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2199 if (!trace_seq_printf(s, "%d %d %llu ",
2200 entry->pid, iter->cpu, iter->ts))
2204 event = ftrace_find_event(entry->type);
2206 return event->funcs->raw(iter, 0, event);
2208 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2211 return TRACE_TYPE_HANDLED;
2213 return TRACE_TYPE_PARTIAL_LINE;
2216 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2218 struct trace_seq *s = &iter->seq;
2219 unsigned char newline = '\n';
2220 struct trace_entry *entry;
2221 struct trace_event *event;
2225 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2226 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2227 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2228 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2231 event = ftrace_find_event(entry->type);
2233 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2234 if (ret != TRACE_TYPE_HANDLED)
2238 SEQ_PUT_FIELD_RET(s, newline);
2240 return TRACE_TYPE_HANDLED;
2243 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2245 struct trace_seq *s = &iter->seq;
2246 struct trace_entry *entry;
2247 struct trace_event *event;
2251 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2252 SEQ_PUT_FIELD_RET(s, entry->pid);
2253 SEQ_PUT_FIELD_RET(s, iter->cpu);
2254 SEQ_PUT_FIELD_RET(s, iter->ts);
2257 event = ftrace_find_event(entry->type);
2258 return event ? event->funcs->binary(iter, 0, event) :
2262 int trace_empty(struct trace_iterator *iter)
2264 struct ring_buffer_iter *buf_iter;
2267 /* If we are looking at one CPU buffer, only check that one */
2268 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2269 cpu = iter->cpu_file;
2270 buf_iter = trace_buffer_iter(iter, cpu);
2272 if (!ring_buffer_iter_empty(buf_iter))
2275 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2281 for_each_tracing_cpu(cpu) {
2282 buf_iter = trace_buffer_iter(iter, cpu);
2284 if (!ring_buffer_iter_empty(buf_iter))
2287 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2295 /* Called with trace_event_read_lock() held. */
2296 enum print_line_t print_trace_line(struct trace_iterator *iter)
2298 enum print_line_t ret;
2300 if (iter->lost_events &&
2301 !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2302 iter->cpu, iter->lost_events))
2303 return TRACE_TYPE_PARTIAL_LINE;
2305 if (iter->trace && iter->trace->print_line) {
2306 ret = iter->trace->print_line(iter);
2307 if (ret != TRACE_TYPE_UNHANDLED)
2311 if (iter->ent->type == TRACE_BPRINT &&
2312 trace_flags & TRACE_ITER_PRINTK &&
2313 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2314 return trace_print_bprintk_msg_only(iter);
2316 if (iter->ent->type == TRACE_PRINT &&
2317 trace_flags & TRACE_ITER_PRINTK &&
2318 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2319 return trace_print_printk_msg_only(iter);
2321 if (trace_flags & TRACE_ITER_BIN)
2322 return print_bin_fmt(iter);
2324 if (trace_flags & TRACE_ITER_HEX)
2325 return print_hex_fmt(iter);
2327 if (trace_flags & TRACE_ITER_RAW)
2328 return print_raw_fmt(iter);
2330 return print_trace_fmt(iter);
2333 void trace_latency_header(struct seq_file *m)
2335 struct trace_iterator *iter = m->private;
2337 /* print nothing if the buffers are empty */
2338 if (trace_empty(iter))
2341 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2342 print_trace_header(m, iter);
2344 if (!(trace_flags & TRACE_ITER_VERBOSE))
2345 print_lat_help_header(m);
2348 void trace_default_header(struct seq_file *m)
2350 struct trace_iterator *iter = m->private;
2352 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2355 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2356 /* print nothing if the buffers are empty */
2357 if (trace_empty(iter))
2359 print_trace_header(m, iter);
2360 if (!(trace_flags & TRACE_ITER_VERBOSE))
2361 print_lat_help_header(m);
2363 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2364 if (trace_flags & TRACE_ITER_IRQ_INFO)
2365 print_func_help_header_irq(iter->tr, m);
2367 print_func_help_header(iter->tr, m);
2372 static void test_ftrace_alive(struct seq_file *m)
2374 if (!ftrace_is_dead())
2376 seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2377 seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2380 static int s_show(struct seq_file *m, void *v)
2382 struct trace_iterator *iter = v;
2385 if (iter->ent == NULL) {
2387 seq_printf(m, "# tracer: %s\n", iter->trace->name);
2389 test_ftrace_alive(m);
2391 if (iter->trace && iter->trace->print_header)
2392 iter->trace->print_header(m);
2394 trace_default_header(m);
2396 } else if (iter->leftover) {
2398 * If we filled the seq_file buffer earlier, we
2399 * want to just show it now.
2401 ret = trace_print_seq(m, &iter->seq);
2403 /* ret should this time be zero, but you never know */
2404 iter->leftover = ret;
2407 print_trace_line(iter);
2408 ret = trace_print_seq(m, &iter->seq);
2410 * If we overflow the seq_file buffer, then it will
2411 * ask us for this data again at start up.
2413 * ret is 0 if seq_file write succeeded.
2416 iter->leftover = ret;
2422 static const struct seq_operations tracer_seq_ops = {
2429 static struct trace_iterator *
2430 __tracing_open(struct inode *inode, struct file *file)
2432 long cpu_file = (long) inode->i_private;
2433 struct trace_iterator *iter;
2436 if (tracing_disabled)
2437 return ERR_PTR(-ENODEV);
2439 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
2441 return ERR_PTR(-ENOMEM);
2443 iter->buffer_iter = kzalloc(sizeof(*iter->buffer_iter) * num_possible_cpus(),
2445 if (!iter->buffer_iter)
2449 * We make a copy of the current tracer to avoid concurrent
2450 * changes on it while we are reading.
2452 mutex_lock(&trace_types_lock);
2453 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2458 *iter->trace = *current_trace;
2460 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2463 if (current_trace && current_trace->print_max)
2466 iter->tr = &global_trace;
2468 mutex_init(&iter->mutex);
2469 iter->cpu_file = cpu_file;
2471 /* Notify the tracer early; before we stop tracing. */
2472 if (iter->trace && iter->trace->open)
2473 iter->trace->open(iter);
2475 /* Annotate start of buffers if we had overruns */
2476 if (ring_buffer_overruns(iter->tr->buffer))
2477 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2479 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2480 if (trace_clocks[trace_clock_id].in_ns)
2481 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
2483 /* stop the trace while dumping */
2486 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2487 for_each_tracing_cpu(cpu) {
2488 iter->buffer_iter[cpu] =
2489 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2491 ring_buffer_read_prepare_sync();
2492 for_each_tracing_cpu(cpu) {
2493 ring_buffer_read_start(iter->buffer_iter[cpu]);
2494 tracing_iter_reset(iter, cpu);
2497 cpu = iter->cpu_file;
2498 iter->buffer_iter[cpu] =
2499 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2500 ring_buffer_read_prepare_sync();
2501 ring_buffer_read_start(iter->buffer_iter[cpu]);
2502 tracing_iter_reset(iter, cpu);
2505 mutex_unlock(&trace_types_lock);
2510 mutex_unlock(&trace_types_lock);
2512 kfree(iter->buffer_iter);
2514 seq_release_private(inode, file);
2515 return ERR_PTR(-ENOMEM);
2518 int tracing_open_generic(struct inode *inode, struct file *filp)
2520 if (tracing_disabled)
2523 filp->private_data = inode->i_private;
2527 static int tracing_release(struct inode *inode, struct file *file)
2529 struct seq_file *m = file->private_data;
2530 struct trace_iterator *iter;
2533 if (!(file->f_mode & FMODE_READ))
2538 mutex_lock(&trace_types_lock);
2539 for_each_tracing_cpu(cpu) {
2540 if (iter->buffer_iter[cpu])
2541 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2544 if (iter->trace && iter->trace->close)
2545 iter->trace->close(iter);
2547 /* reenable tracing if it was previously enabled */
2549 mutex_unlock(&trace_types_lock);
2551 mutex_destroy(&iter->mutex);
2552 free_cpumask_var(iter->started);
2554 kfree(iter->buffer_iter);
2555 seq_release_private(inode, file);
2559 static int tracing_open(struct inode *inode, struct file *file)
2561 struct trace_iterator *iter;
2564 /* If this file was open for write, then erase contents */
2565 if ((file->f_mode & FMODE_WRITE) &&
2566 (file->f_flags & O_TRUNC)) {
2567 long cpu = (long) inode->i_private;
2569 if (cpu == TRACE_PIPE_ALL_CPU)
2570 tracing_reset_online_cpus(&global_trace);
2572 tracing_reset(&global_trace, cpu);
2575 if (file->f_mode & FMODE_READ) {
2576 iter = __tracing_open(inode, file);
2578 ret = PTR_ERR(iter);
2579 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2580 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2586 t_next(struct seq_file *m, void *v, loff_t *pos)
2588 struct tracer *t = v;
2598 static void *t_start(struct seq_file *m, loff_t *pos)
2603 mutex_lock(&trace_types_lock);
2604 for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2610 static void t_stop(struct seq_file *m, void *p)
2612 mutex_unlock(&trace_types_lock);
2615 static int t_show(struct seq_file *m, void *v)
2617 struct tracer *t = v;
2622 seq_printf(m, "%s", t->name);
2631 static const struct seq_operations show_traces_seq_ops = {
2638 static int show_traces_open(struct inode *inode, struct file *file)
2640 if (tracing_disabled)
2643 return seq_open(file, &show_traces_seq_ops);
2647 tracing_write_stub(struct file *filp, const char __user *ubuf,
2648 size_t count, loff_t *ppos)
2653 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2655 if (file->f_mode & FMODE_READ)
2656 return seq_lseek(file, offset, origin);
2661 static const struct file_operations tracing_fops = {
2662 .open = tracing_open,
2664 .write = tracing_write_stub,
2665 .llseek = tracing_seek,
2666 .release = tracing_release,
2669 static const struct file_operations show_traces_fops = {
2670 .open = show_traces_open,
2672 .release = seq_release,
2673 .llseek = seq_lseek,
2677 * Only trace on a CPU if the bitmask is set:
2679 static cpumask_var_t tracing_cpumask;
2682 * The tracer itself will not take this lock, but still we want
2683 * to provide a consistent cpumask to user-space:
2685 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2688 * Temporary storage for the character representation of the
2689 * CPU bitmask (and one more byte for the newline):
2691 static char mask_str[NR_CPUS + 1];
2694 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2695 size_t count, loff_t *ppos)
2699 mutex_lock(&tracing_cpumask_update_lock);
2701 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2702 if (count - len < 2) {
2706 len += sprintf(mask_str + len, "\n");
2707 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2710 mutex_unlock(&tracing_cpumask_update_lock);
2716 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2717 size_t count, loff_t *ppos)
2720 cpumask_var_t tracing_cpumask_new;
2722 if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2725 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2729 mutex_lock(&tracing_cpumask_update_lock);
2731 local_irq_disable();
2732 arch_spin_lock(&ftrace_max_lock);
2733 for_each_tracing_cpu(cpu) {
2735 * Increase/decrease the disabled counter if we are
2736 * about to flip a bit in the cpumask:
2738 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2739 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2740 atomic_inc(&global_trace.data[cpu]->disabled);
2741 ring_buffer_record_disable_cpu(global_trace.buffer, cpu);
2743 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2744 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2745 atomic_dec(&global_trace.data[cpu]->disabled);
2746 ring_buffer_record_enable_cpu(global_trace.buffer, cpu);
2749 arch_spin_unlock(&ftrace_max_lock);
2752 cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2754 mutex_unlock(&tracing_cpumask_update_lock);
2755 free_cpumask_var(tracing_cpumask_new);
2760 free_cpumask_var(tracing_cpumask_new);
2765 static const struct file_operations tracing_cpumask_fops = {
2766 .open = tracing_open_generic,
2767 .read = tracing_cpumask_read,
2768 .write = tracing_cpumask_write,
2769 .llseek = generic_file_llseek,
2772 static int tracing_trace_options_show(struct seq_file *m, void *v)
2774 struct tracer_opt *trace_opts;
2778 mutex_lock(&trace_types_lock);
2779 tracer_flags = current_trace->flags->val;
2780 trace_opts = current_trace->flags->opts;
2782 for (i = 0; trace_options[i]; i++) {
2783 if (trace_flags & (1 << i))
2784 seq_printf(m, "%s\n", trace_options[i]);
2786 seq_printf(m, "no%s\n", trace_options[i]);
2789 for (i = 0; trace_opts[i].name; i++) {
2790 if (tracer_flags & trace_opts[i].bit)
2791 seq_printf(m, "%s\n", trace_opts[i].name);
2793 seq_printf(m, "no%s\n", trace_opts[i].name);
2795 mutex_unlock(&trace_types_lock);
2800 static int __set_tracer_option(struct tracer *trace,
2801 struct tracer_flags *tracer_flags,
2802 struct tracer_opt *opts, int neg)
2806 ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2811 tracer_flags->val &= ~opts->bit;
2813 tracer_flags->val |= opts->bit;
2817 /* Try to assign a tracer specific option */
2818 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2820 struct tracer_flags *tracer_flags = trace->flags;
2821 struct tracer_opt *opts = NULL;
2824 for (i = 0; tracer_flags->opts[i].name; i++) {
2825 opts = &tracer_flags->opts[i];
2827 if (strcmp(cmp, opts->name) == 0)
2828 return __set_tracer_option(trace, trace->flags,
2835 static void set_tracer_flags(unsigned int mask, int enabled)
2837 /* do nothing if flag is already set */
2838 if (!!(trace_flags & mask) == !!enabled)
2842 trace_flags |= mask;
2844 trace_flags &= ~mask;
2846 if (mask == TRACE_ITER_RECORD_CMD)
2847 trace_event_enable_cmd_record(enabled);
2849 if (mask == TRACE_ITER_OVERWRITE)
2850 ring_buffer_change_overwrite(global_trace.buffer, enabled);
2852 if (mask == TRACE_ITER_PRINTK)
2853 trace_printk_start_stop_comm(enabled);
2856 static int trace_set_options(char *option)
2863 cmp = strstrip(option);
2865 if (strncmp(cmp, "no", 2) == 0) {
2870 for (i = 0; trace_options[i]; i++) {
2871 if (strcmp(cmp, trace_options[i]) == 0) {
2872 set_tracer_flags(1 << i, !neg);
2877 /* If no option could be set, test the specific tracer options */
2878 if (!trace_options[i]) {
2879 mutex_lock(&trace_types_lock);
2880 ret = set_tracer_option(current_trace, cmp, neg);
2881 mutex_unlock(&trace_types_lock);
2888 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2889 size_t cnt, loff_t *ppos)
2893 if (cnt >= sizeof(buf))
2896 if (copy_from_user(&buf, ubuf, cnt))
2901 trace_set_options(buf);
2908 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2910 if (tracing_disabled)
2912 return single_open(file, tracing_trace_options_show, NULL);
2915 static const struct file_operations tracing_iter_fops = {
2916 .open = tracing_trace_options_open,
2918 .llseek = seq_lseek,
2919 .release = single_release,
2920 .write = tracing_trace_options_write,
2923 static const char readme_msg[] =
2924 "tracing mini-HOWTO:\n\n"
2925 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2926 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2927 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2928 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2930 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2931 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2933 "# cat /sys/kernel/debug/tracing/trace_options\n"
2934 "noprint-parent nosym-offset nosym-addr noverbose\n"
2935 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2936 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2937 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2938 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2942 tracing_readme_read(struct file *filp, char __user *ubuf,
2943 size_t cnt, loff_t *ppos)
2945 return simple_read_from_buffer(ubuf, cnt, ppos,
2946 readme_msg, strlen(readme_msg));
2949 static const struct file_operations tracing_readme_fops = {
2950 .open = tracing_open_generic,
2951 .read = tracing_readme_read,
2952 .llseek = generic_file_llseek,
2956 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2957 size_t cnt, loff_t *ppos)
2966 file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2970 buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2978 for (i = 0; i < SAVED_CMDLINES; i++) {
2981 pid = map_cmdline_to_pid[i];
2982 if (pid == -1 || pid == NO_CMDLINE_MAP)
2985 trace_find_cmdline(pid, buf_comm);
2986 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2991 len = simple_read_from_buffer(ubuf, cnt, ppos,
3000 static const struct file_operations tracing_saved_cmdlines_fops = {
3001 .open = tracing_open_generic,
3002 .read = tracing_saved_cmdlines_read,
3003 .llseek = generic_file_llseek,
3007 tracing_set_trace_read(struct file *filp, char __user *ubuf,
3008 size_t cnt, loff_t *ppos)
3010 char buf[MAX_TRACER_SIZE+2];
3013 mutex_lock(&trace_types_lock);
3015 r = sprintf(buf, "%s\n", current_trace->name);
3017 r = sprintf(buf, "\n");
3018 mutex_unlock(&trace_types_lock);
3020 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3023 int tracer_init(struct tracer *t, struct trace_array *tr)
3025 tracing_reset_online_cpus(tr);
3029 static void set_buffer_entries(struct trace_array *tr, unsigned long val)
3032 for_each_tracing_cpu(cpu)
3033 tr->data[cpu]->entries = val;
3036 /* resize @tr's buffer to the size of @size_tr's entries */
3037 static int resize_buffer_duplicate_size(struct trace_array *tr,
3038 struct trace_array *size_tr, int cpu_id)
3042 if (cpu_id == RING_BUFFER_ALL_CPUS) {
3043 for_each_tracing_cpu(cpu) {
3044 ret = ring_buffer_resize(tr->buffer,
3045 size_tr->data[cpu]->entries, cpu);
3048 tr->data[cpu]->entries = size_tr->data[cpu]->entries;
3051 ret = ring_buffer_resize(tr->buffer,
3052 size_tr->data[cpu_id]->entries, cpu_id);
3054 tr->data[cpu_id]->entries =
3055 size_tr->data[cpu_id]->entries;
3061 static int __tracing_resize_ring_buffer(unsigned long size, int cpu)
3066 * If kernel or user changes the size of the ring buffer
3067 * we use the size that was given, and we can forget about
3068 * expanding it later.
3070 ring_buffer_expanded = 1;
3072 /* May be called before buffers are initialized */
3073 if (!global_trace.buffer)
3076 ret = ring_buffer_resize(global_trace.buffer, size, cpu);
3080 if (!current_trace->use_max_tr)
3083 ret = ring_buffer_resize(max_tr.buffer, size, cpu);
3085 int r = resize_buffer_duplicate_size(&global_trace,
3086 &global_trace, cpu);
3089 * AARGH! We are left with different
3090 * size max buffer!!!!
3091 * The max buffer is our "snapshot" buffer.
3092 * When a tracer needs a snapshot (one of the
3093 * latency tracers), it swaps the max buffer
3094 * with the saved snap shot. We succeeded to
3095 * update the size of the main buffer, but failed to
3096 * update the size of the max buffer. But when we tried
3097 * to reset the main buffer to the original size, we
3098 * failed there too. This is very unlikely to
3099 * happen, but if it does, warn and kill all
3103 tracing_disabled = 1;
3108 if (cpu == RING_BUFFER_ALL_CPUS)
3109 set_buffer_entries(&max_tr, size);
3111 max_tr.data[cpu]->entries = size;
3114 if (cpu == RING_BUFFER_ALL_CPUS)
3115 set_buffer_entries(&global_trace, size);
3117 global_trace.data[cpu]->entries = size;
3122 static ssize_t tracing_resize_ring_buffer(unsigned long size, int cpu_id)
3126 mutex_lock(&trace_types_lock);
3128 if (cpu_id != RING_BUFFER_ALL_CPUS) {
3129 /* make sure, this cpu is enabled in the mask */
3130 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
3136 ret = __tracing_resize_ring_buffer(size, cpu_id);
3141 mutex_unlock(&trace_types_lock);
3148 * tracing_update_buffers - used by tracing facility to expand ring buffers
3150 * To save on memory when the tracing is never used on a system with it
3151 * configured in. The ring buffers are set to a minimum size. But once
3152 * a user starts to use the tracing facility, then they need to grow
3153 * to their default size.
3155 * This function is to be called when a tracer is about to be used.
3157 int tracing_update_buffers(void)
3161 mutex_lock(&trace_types_lock);
3162 if (!ring_buffer_expanded)
3163 ret = __tracing_resize_ring_buffer(trace_buf_size,
3164 RING_BUFFER_ALL_CPUS);
3165 mutex_unlock(&trace_types_lock);
3170 struct trace_option_dentry;
3172 static struct trace_option_dentry *
3173 create_trace_option_files(struct tracer *tracer);
3176 destroy_trace_option_files(struct trace_option_dentry *topts);
3178 static int tracing_set_tracer(const char *buf)
3180 static struct trace_option_dentry *topts;
3181 struct trace_array *tr = &global_trace;
3185 mutex_lock(&trace_types_lock);
3187 if (!ring_buffer_expanded) {
3188 ret = __tracing_resize_ring_buffer(trace_buf_size,
3189 RING_BUFFER_ALL_CPUS);
3195 for (t = trace_types; t; t = t->next) {
3196 if (strcmp(t->name, buf) == 0)
3203 if (t == current_trace)
3206 trace_branch_disable();
3207 if (current_trace && current_trace->reset)
3208 current_trace->reset(tr);
3209 if (current_trace && current_trace->use_max_tr) {
3211 * We don't free the ring buffer. instead, resize it because
3212 * The max_tr ring buffer has some state (e.g. ring->clock) and
3213 * we want preserve it.
3215 ring_buffer_resize(max_tr.buffer, 1, RING_BUFFER_ALL_CPUS);
3216 set_buffer_entries(&max_tr, 1);
3218 destroy_trace_option_files(topts);
3220 current_trace = &nop_trace;
3222 topts = create_trace_option_files(t);
3223 if (t->use_max_tr) {
3224 /* we need to make per cpu buffer sizes equivalent */
3225 ret = resize_buffer_duplicate_size(&max_tr, &global_trace,
3226 RING_BUFFER_ALL_CPUS);
3232 ret = tracer_init(t, tr);
3238 trace_branch_enable(tr);
3240 mutex_unlock(&trace_types_lock);
3246 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3247 size_t cnt, loff_t *ppos)
3249 char buf[MAX_TRACER_SIZE+1];
3256 if (cnt > MAX_TRACER_SIZE)
3257 cnt = MAX_TRACER_SIZE;
3259 if (copy_from_user(&buf, ubuf, cnt))
3264 /* strip ending whitespace. */
3265 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3268 err = tracing_set_tracer(buf);
3278 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3279 size_t cnt, loff_t *ppos)
3281 unsigned long *ptr = filp->private_data;
3285 r = snprintf(buf, sizeof(buf), "%ld\n",
3286 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3287 if (r > sizeof(buf))
3289 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3293 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3294 size_t cnt, loff_t *ppos)
3296 unsigned long *ptr = filp->private_data;
3300 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3309 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3311 long cpu_file = (long) inode->i_private;
3312 struct trace_iterator *iter;
3315 if (tracing_disabled)
3318 mutex_lock(&trace_types_lock);
3320 /* create a buffer to store the information to pass to userspace */
3321 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3328 * We make a copy of the current tracer to avoid concurrent
3329 * changes on it while we are reading.
3331 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3337 *iter->trace = *current_trace;
3339 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3344 /* trace pipe does not show start of buffer */
3345 cpumask_setall(iter->started);
3347 if (trace_flags & TRACE_ITER_LATENCY_FMT)
3348 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3350 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3351 if (trace_clocks[trace_clock_id].in_ns)
3352 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
3354 iter->cpu_file = cpu_file;
3355 iter->tr = &global_trace;
3356 mutex_init(&iter->mutex);
3357 filp->private_data = iter;
3359 if (iter->trace->pipe_open)
3360 iter->trace->pipe_open(iter);
3362 nonseekable_open(inode, filp);
3364 mutex_unlock(&trace_types_lock);
3370 mutex_unlock(&trace_types_lock);
3374 static int tracing_release_pipe(struct inode *inode, struct file *file)
3376 struct trace_iterator *iter = file->private_data;
3378 mutex_lock(&trace_types_lock);
3380 if (iter->trace->pipe_close)
3381 iter->trace->pipe_close(iter);
3383 mutex_unlock(&trace_types_lock);
3385 free_cpumask_var(iter->started);
3386 mutex_destroy(&iter->mutex);
3394 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3396 struct trace_iterator *iter = filp->private_data;
3398 if (trace_flags & TRACE_ITER_BLOCK) {
3400 * Always select as readable when in blocking mode
3402 return POLLIN | POLLRDNORM;
3404 if (!trace_empty(iter))
3405 return POLLIN | POLLRDNORM;
3406 poll_wait(filp, &trace_wait, poll_table);
3407 if (!trace_empty(iter))
3408 return POLLIN | POLLRDNORM;
3415 * This is a make-shift waitqueue.
3416 * A tracer might use this callback on some rare cases:
3418 * 1) the current tracer might hold the runqueue lock when it wakes up
3419 * a reader, hence a deadlock (sched, function, and function graph tracers)
3420 * 2) the function tracers, trace all functions, we don't want
3421 * the overhead of calling wake_up and friends
3422 * (and tracing them too)
3424 * Anyway, this is really very primitive wakeup.
3426 void poll_wait_pipe(struct trace_iterator *iter)
3428 set_current_state(TASK_INTERRUPTIBLE);
3429 /* sleep for 100 msecs, and try again. */
3430 schedule_timeout(HZ / 10);
3433 /* Must be called with trace_types_lock mutex held. */
3434 static int tracing_wait_pipe(struct file *filp)
3436 struct trace_iterator *iter = filp->private_data;
3438 while (trace_empty(iter)) {
3440 if ((filp->f_flags & O_NONBLOCK)) {
3444 mutex_unlock(&iter->mutex);
3446 iter->trace->wait_pipe(iter);
3448 mutex_lock(&iter->mutex);
3450 if (signal_pending(current))
3454 * We block until we read something and tracing is disabled.
3455 * We still block if tracing is disabled, but we have never
3456 * read anything. This allows a user to cat this file, and
3457 * then enable tracing. But after we have read something,
3458 * we give an EOF when tracing is again disabled.
3460 * iter->pos will be 0 if we haven't read anything.
3462 if (!tracing_is_enabled() && iter->pos)
3473 tracing_read_pipe(struct file *filp, char __user *ubuf,
3474 size_t cnt, loff_t *ppos)
3476 struct trace_iterator *iter = filp->private_data;
3477 static struct tracer *old_tracer;
3480 /* return any leftover data */
3481 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3485 trace_seq_init(&iter->seq);
3487 /* copy the tracer to avoid using a global lock all around */
3488 mutex_lock(&trace_types_lock);
3489 if (unlikely(old_tracer != current_trace && current_trace)) {
3490 old_tracer = current_trace;
3491 *iter->trace = *current_trace;
3493 mutex_unlock(&trace_types_lock);
3496 * Avoid more than one consumer on a single file descriptor
3497 * This is just a matter of traces coherency, the ring buffer itself
3500 mutex_lock(&iter->mutex);
3501 if (iter->trace->read) {
3502 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3508 sret = tracing_wait_pipe(filp);
3512 /* stop when tracing is finished */
3513 if (trace_empty(iter)) {
3518 if (cnt >= PAGE_SIZE)
3519 cnt = PAGE_SIZE - 1;
3521 /* reset all but tr, trace, and overruns */
3522 memset(&iter->seq, 0,
3523 sizeof(struct trace_iterator) -
3524 offsetof(struct trace_iterator, seq));
3527 trace_event_read_lock();
3528 trace_access_lock(iter->cpu_file);
3529 while (trace_find_next_entry_inc(iter) != NULL) {
3530 enum print_line_t ret;
3531 int len = iter->seq.len;
3533 ret = print_trace_line(iter);
3534 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3535 /* don't print partial lines */
3536 iter->seq.len = len;
3539 if (ret != TRACE_TYPE_NO_CONSUME)
3540 trace_consume(iter);
3542 if (iter->seq.len >= cnt)
3546 * Setting the full flag means we reached the trace_seq buffer
3547 * size and we should leave by partial output condition above.
3548 * One of the trace_seq_* functions is not used properly.
3550 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3553 trace_access_unlock(iter->cpu_file);
3554 trace_event_read_unlock();
3556 /* Now copy what we have to the user */
3557 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3558 if (iter->seq.readpos >= iter->seq.len)
3559 trace_seq_init(&iter->seq);
3562 * If there was nothing to send to user, in spite of consuming trace
3563 * entries, go back to wait for more entries.
3569 mutex_unlock(&iter->mutex);
3574 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3575 struct pipe_buffer *buf)
3577 __free_page(buf->page);
3580 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3583 __free_page(spd->pages[idx]);
3586 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3588 .map = generic_pipe_buf_map,
3589 .unmap = generic_pipe_buf_unmap,
3590 .confirm = generic_pipe_buf_confirm,
3591 .release = tracing_pipe_buf_release,
3592 .steal = generic_pipe_buf_steal,
3593 .get = generic_pipe_buf_get,
3597 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3602 /* Seq buffer is page-sized, exactly what we need. */
3604 count = iter->seq.len;
3605 ret = print_trace_line(iter);
3606 count = iter->seq.len - count;
3609 iter->seq.len -= count;
3612 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3613 iter->seq.len -= count;
3617 if (ret != TRACE_TYPE_NO_CONSUME)
3618 trace_consume(iter);
3620 if (!trace_find_next_entry_inc(iter)) {
3630 static ssize_t tracing_splice_read_pipe(struct file *filp,
3632 struct pipe_inode_info *pipe,
3636 struct page *pages_def[PIPE_DEF_BUFFERS];
3637 struct partial_page partial_def[PIPE_DEF_BUFFERS];
3638 struct trace_iterator *iter = filp->private_data;
3639 struct splice_pipe_desc spd = {
3641 .partial = partial_def,
3642 .nr_pages = 0, /* This gets updated below. */
3643 .nr_pages_max = PIPE_DEF_BUFFERS,
3645 .ops = &tracing_pipe_buf_ops,
3646 .spd_release = tracing_spd_release_pipe,
3648 static struct tracer *old_tracer;
3653 if (splice_grow_spd(pipe, &spd))
3656 /* copy the tracer to avoid using a global lock all around */
3657 mutex_lock(&trace_types_lock);
3658 if (unlikely(old_tracer != current_trace && current_trace)) {
3659 old_tracer = current_trace;
3660 *iter->trace = *current_trace;
3662 mutex_unlock(&trace_types_lock);
3664 mutex_lock(&iter->mutex);
3666 if (iter->trace->splice_read) {
3667 ret = iter->trace->splice_read(iter, filp,
3668 ppos, pipe, len, flags);
3673 ret = tracing_wait_pipe(filp);
3677 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3682 trace_event_read_lock();
3683 trace_access_lock(iter->cpu_file);
3685 /* Fill as many pages as possible. */
3686 for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3687 spd.pages[i] = alloc_page(GFP_KERNEL);
3691 rem = tracing_fill_pipe_page(rem, iter);
3693 /* Copy the data into the page, so we can start over. */
3694 ret = trace_seq_to_buffer(&iter->seq,
3695 page_address(spd.pages[i]),
3698 __free_page(spd.pages[i]);
3701 spd.partial[i].offset = 0;
3702 spd.partial[i].len = iter->seq.len;
3704 trace_seq_init(&iter->seq);
3707 trace_access_unlock(iter->cpu_file);
3708 trace_event_read_unlock();
3709 mutex_unlock(&iter->mutex);
3713 ret = splice_to_pipe(pipe, &spd);
3715 splice_shrink_spd(&spd);
3719 mutex_unlock(&iter->mutex);
3723 struct ftrace_entries_info {
3724 struct trace_array *tr;
3728 static int tracing_entries_open(struct inode *inode, struct file *filp)
3730 struct ftrace_entries_info *info;
3732 if (tracing_disabled)
3735 info = kzalloc(sizeof(*info), GFP_KERNEL);
3739 info->tr = &global_trace;
3740 info->cpu = (unsigned long)inode->i_private;
3742 filp->private_data = info;
3748 tracing_entries_read(struct file *filp, char __user *ubuf,
3749 size_t cnt, loff_t *ppos)
3751 struct ftrace_entries_info *info = filp->private_data;
3752 struct trace_array *tr = info->tr;
3757 mutex_lock(&trace_types_lock);
3759 if (info->cpu == RING_BUFFER_ALL_CPUS) {
3760 int cpu, buf_size_same;
3765 /* check if all cpu sizes are same */
3766 for_each_tracing_cpu(cpu) {
3767 /* fill in the size from first enabled cpu */
3769 size = tr->data[cpu]->entries;
3770 if (size != tr->data[cpu]->entries) {
3776 if (buf_size_same) {
3777 if (!ring_buffer_expanded)
3778 r = sprintf(buf, "%lu (expanded: %lu)\n",
3780 trace_buf_size >> 10);
3782 r = sprintf(buf, "%lu\n", size >> 10);
3784 r = sprintf(buf, "X\n");
3786 r = sprintf(buf, "%lu\n", tr->data[info->cpu]->entries >> 10);
3788 mutex_unlock(&trace_types_lock);
3790 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3795 tracing_entries_write(struct file *filp, const char __user *ubuf,
3796 size_t cnt, loff_t *ppos)
3798 struct ftrace_entries_info *info = filp->private_data;
3802 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3806 /* must have at least 1 entry */
3810 /* value is in KB */
3813 ret = tracing_resize_ring_buffer(val, info->cpu);
3823 tracing_entries_release(struct inode *inode, struct file *filp)
3825 struct ftrace_entries_info *info = filp->private_data;
3833 tracing_total_entries_read(struct file *filp, char __user *ubuf,
3834 size_t cnt, loff_t *ppos)
3836 struct trace_array *tr = filp->private_data;
3839 unsigned long size = 0, expanded_size = 0;
3841 mutex_lock(&trace_types_lock);
3842 for_each_tracing_cpu(cpu) {
3843 size += tr->data[cpu]->entries >> 10;
3844 if (!ring_buffer_expanded)
3845 expanded_size += trace_buf_size >> 10;
3847 if (ring_buffer_expanded)
3848 r = sprintf(buf, "%lu\n", size);
3850 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
3851 mutex_unlock(&trace_types_lock);
3853 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3857 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
3858 size_t cnt, loff_t *ppos)
3861 * There is no need to read what the user has written, this function
3862 * is just to make sure that there is no error when "echo" is used
3871 tracing_free_buffer_release(struct inode *inode, struct file *filp)
3873 /* disable tracing ? */
3874 if (trace_flags & TRACE_ITER_STOP_ON_FREE)
3876 /* resize the ring buffer to 0 */
3877 tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS);
3883 tracing_mark_write(struct file *filp, const char __user *ubuf,
3884 size_t cnt, loff_t *fpos)
3886 unsigned long addr = (unsigned long)ubuf;
3887 struct ring_buffer_event *event;
3888 struct ring_buffer *buffer;
3889 struct print_entry *entry;
3890 unsigned long irq_flags;
3891 struct page *pages[2];
3901 if (tracing_disabled)
3904 if (!(trace_flags & TRACE_ITER_MARKERS))
3907 if (cnt > TRACE_BUF_SIZE)
3908 cnt = TRACE_BUF_SIZE;
3911 * Userspace is injecting traces into the kernel trace buffer.
3912 * We want to be as non intrusive as possible.
3913 * To do so, we do not want to allocate any special buffers
3914 * or take any locks, but instead write the userspace data
3915 * straight into the ring buffer.
3917 * First we need to pin the userspace buffer into memory,
3918 * which, most likely it is, because it just referenced it.
3919 * But there's no guarantee that it is. By using get_user_pages_fast()
3920 * and kmap_atomic/kunmap_atomic() we can get access to the
3921 * pages directly. We then write the data directly into the
3924 BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
3926 /* check if we cross pages */
3927 if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
3930 offset = addr & (PAGE_SIZE - 1);
3933 ret = get_user_pages_fast(addr, nr_pages, 0, pages);
3934 if (ret < nr_pages) {
3936 put_page(pages[ret]);
3941 for (i = 0; i < nr_pages; i++)
3942 map_page[i] = kmap_atomic(pages[i]);
3944 local_save_flags(irq_flags);
3945 size = sizeof(*entry) + cnt + 2; /* possible \n added */
3946 buffer = global_trace.buffer;
3947 event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3948 irq_flags, preempt_count());
3950 /* Ring buffer disabled, return as if not open for write */
3955 entry = ring_buffer_event_data(event);
3956 entry->ip = _THIS_IP_;
3958 if (nr_pages == 2) {
3959 len = PAGE_SIZE - offset;
3960 memcpy(&entry->buf, map_page[0] + offset, len);
3961 memcpy(&entry->buf[len], map_page[1], cnt - len);
3963 memcpy(&entry->buf, map_page[0] + offset, cnt);
3965 if (entry->buf[cnt - 1] != '\n') {
3966 entry->buf[cnt] = '\n';
3967 entry->buf[cnt + 1] = '\0';
3969 entry->buf[cnt] = '\0';
3971 __buffer_unlock_commit(buffer, event);
3978 for (i = 0; i < nr_pages; i++){
3979 kunmap_atomic(map_page[i]);
3986 static int tracing_clock_show(struct seq_file *m, void *v)
3990 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3992 "%s%s%s%s", i ? " " : "",
3993 i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3994 i == trace_clock_id ? "]" : "");
4000 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
4001 size_t cnt, loff_t *fpos)
4004 const char *clockstr;
4007 if (cnt >= sizeof(buf))
4010 if (copy_from_user(&buf, ubuf, cnt))
4015 clockstr = strstrip(buf);
4017 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
4018 if (strcmp(trace_clocks[i].name, clockstr) == 0)
4021 if (i == ARRAY_SIZE(trace_clocks))
4026 mutex_lock(&trace_types_lock);
4028 ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
4030 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
4033 * New clock may not be consistent with the previous clock.
4034 * Reset the buffer so that it doesn't have incomparable timestamps.
4036 tracing_reset_online_cpus(&global_trace);
4038 tracing_reset_online_cpus(&max_tr);
4040 mutex_unlock(&trace_types_lock);
4047 static int tracing_clock_open(struct inode *inode, struct file *file)
4049 if (tracing_disabled)
4051 return single_open(file, tracing_clock_show, NULL);
4054 static const struct file_operations tracing_max_lat_fops = {
4055 .open = tracing_open_generic,
4056 .read = tracing_max_lat_read,
4057 .write = tracing_max_lat_write,
4058 .llseek = generic_file_llseek,
4061 static const struct file_operations set_tracer_fops = {
4062 .open = tracing_open_generic,
4063 .read = tracing_set_trace_read,
4064 .write = tracing_set_trace_write,
4065 .llseek = generic_file_llseek,
4068 static const struct file_operations tracing_pipe_fops = {
4069 .open = tracing_open_pipe,
4070 .poll = tracing_poll_pipe,
4071 .read = tracing_read_pipe,
4072 .splice_read = tracing_splice_read_pipe,
4073 .release = tracing_release_pipe,
4074 .llseek = no_llseek,
4077 static const struct file_operations tracing_entries_fops = {
4078 .open = tracing_entries_open,
4079 .read = tracing_entries_read,
4080 .write = tracing_entries_write,
4081 .release = tracing_entries_release,
4082 .llseek = generic_file_llseek,
4085 static const struct file_operations tracing_total_entries_fops = {
4086 .open = tracing_open_generic,
4087 .read = tracing_total_entries_read,
4088 .llseek = generic_file_llseek,
4091 static const struct file_operations tracing_free_buffer_fops = {
4092 .write = tracing_free_buffer_write,
4093 .release = tracing_free_buffer_release,
4096 static const struct file_operations tracing_mark_fops = {
4097 .open = tracing_open_generic,
4098 .write = tracing_mark_write,
4099 .llseek = generic_file_llseek,
4102 static const struct file_operations trace_clock_fops = {
4103 .open = tracing_clock_open,
4105 .llseek = seq_lseek,
4106 .release = single_release,
4107 .write = tracing_clock_write,
4110 struct ftrace_buffer_info {
4111 struct trace_array *tr;
4117 static int tracing_buffers_open(struct inode *inode, struct file *filp)
4119 int cpu = (int)(long)inode->i_private;
4120 struct ftrace_buffer_info *info;
4122 if (tracing_disabled)
4125 info = kzalloc(sizeof(*info), GFP_KERNEL);
4129 info->tr = &global_trace;
4132 /* Force reading ring buffer for first read */
4133 info->read = (unsigned int)-1;
4135 filp->private_data = info;
4137 return nonseekable_open(inode, filp);
4141 tracing_buffers_read(struct file *filp, char __user *ubuf,
4142 size_t count, loff_t *ppos)
4144 struct ftrace_buffer_info *info = filp->private_data;
4152 info->spare = ring_buffer_alloc_read_page(info->tr->buffer, info->cpu);
4156 /* Do we have previous read data to read? */
4157 if (info->read < PAGE_SIZE)
4160 trace_access_lock(info->cpu);
4161 ret = ring_buffer_read_page(info->tr->buffer,
4165 trace_access_unlock(info->cpu);
4172 size = PAGE_SIZE - info->read;
4176 ret = copy_to_user(ubuf, info->spare + info->read, size);
4187 static int tracing_buffers_release(struct inode *inode, struct file *file)
4189 struct ftrace_buffer_info *info = file->private_data;
4192 ring_buffer_free_read_page(info->tr->buffer, info->spare);
4199 struct ring_buffer *buffer;
4204 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4205 struct pipe_buffer *buf)
4207 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4212 ring_buffer_free_read_page(ref->buffer, ref->page);
4217 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4218 struct pipe_buffer *buf)
4220 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4225 /* Pipe buffer operations for a buffer. */
4226 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4228 .map = generic_pipe_buf_map,
4229 .unmap = generic_pipe_buf_unmap,
4230 .confirm = generic_pipe_buf_confirm,
4231 .release = buffer_pipe_buf_release,
4232 .steal = generic_pipe_buf_steal,
4233 .get = buffer_pipe_buf_get,
4237 * Callback from splice_to_pipe(), if we need to release some pages
4238 * at the end of the spd in case we error'ed out in filling the pipe.
4240 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4242 struct buffer_ref *ref =
4243 (struct buffer_ref *)spd->partial[i].private;
4248 ring_buffer_free_read_page(ref->buffer, ref->page);
4250 spd->partial[i].private = 0;
4254 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4255 struct pipe_inode_info *pipe, size_t len,
4258 struct ftrace_buffer_info *info = file->private_data;
4259 struct partial_page partial_def[PIPE_DEF_BUFFERS];
4260 struct page *pages_def[PIPE_DEF_BUFFERS];
4261 struct splice_pipe_desc spd = {
4263 .partial = partial_def,
4264 .nr_pages_max = PIPE_DEF_BUFFERS,
4266 .ops = &buffer_pipe_buf_ops,
4267 .spd_release = buffer_spd_release,
4269 struct buffer_ref *ref;
4270 int entries, size, i;
4273 if (splice_grow_spd(pipe, &spd))
4276 if (*ppos & (PAGE_SIZE - 1)) {
4281 if (len & (PAGE_SIZE - 1)) {
4282 if (len < PAGE_SIZE) {
4289 trace_access_lock(info->cpu);
4290 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4292 for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4296 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4301 ref->buffer = info->tr->buffer;
4302 ref->page = ring_buffer_alloc_read_page(ref->buffer, info->cpu);
4308 r = ring_buffer_read_page(ref->buffer, &ref->page,
4311 ring_buffer_free_read_page(ref->buffer, ref->page);
4317 * zero out any left over data, this is going to
4320 size = ring_buffer_page_len(ref->page);
4321 if (size < PAGE_SIZE)
4322 memset(ref->page + size, 0, PAGE_SIZE - size);
4324 page = virt_to_page(ref->page);
4326 spd.pages[i] = page;
4327 spd.partial[i].len = PAGE_SIZE;
4328 spd.partial[i].offset = 0;
4329 spd.partial[i].private = (unsigned long)ref;
4333 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4336 trace_access_unlock(info->cpu);
4339 /* did we read anything? */
4340 if (!spd.nr_pages) {
4341 if (flags & SPLICE_F_NONBLOCK)
4349 ret = splice_to_pipe(pipe, &spd);
4350 splice_shrink_spd(&spd);
4355 static const struct file_operations tracing_buffers_fops = {
4356 .open = tracing_buffers_open,
4357 .read = tracing_buffers_read,
4358 .release = tracing_buffers_release,
4359 .splice_read = tracing_buffers_splice_read,
4360 .llseek = no_llseek,
4364 tracing_stats_read(struct file *filp, char __user *ubuf,
4365 size_t count, loff_t *ppos)
4367 unsigned long cpu = (unsigned long)filp->private_data;
4368 struct trace_array *tr = &global_trace;
4369 struct trace_seq *s;
4371 unsigned long long t;
4372 unsigned long usec_rem;
4374 s = kmalloc(sizeof(*s), GFP_KERNEL);
4380 cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
4381 trace_seq_printf(s, "entries: %ld\n", cnt);
4383 cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
4384 trace_seq_printf(s, "overrun: %ld\n", cnt);
4386 cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
4387 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
4389 cnt = ring_buffer_bytes_cpu(tr->buffer, cpu);
4390 trace_seq_printf(s, "bytes: %ld\n", cnt);
4392 if (trace_clocks[trace_clock_id].in_ns) {
4393 /* local or global for trace_clock */
4394 t = ns2usecs(ring_buffer_oldest_event_ts(tr->buffer, cpu));
4395 usec_rem = do_div(t, USEC_PER_SEC);
4396 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
4399 t = ns2usecs(ring_buffer_time_stamp(tr->buffer, cpu));
4400 usec_rem = do_div(t, USEC_PER_SEC);
4401 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
4403 /* counter or tsc mode for trace_clock */
4404 trace_seq_printf(s, "oldest event ts: %llu\n",
4405 ring_buffer_oldest_event_ts(tr->buffer, cpu));
4407 trace_seq_printf(s, "now ts: %llu\n",
4408 ring_buffer_time_stamp(tr->buffer, cpu));
4411 cnt = ring_buffer_dropped_events_cpu(tr->buffer, cpu);
4412 trace_seq_printf(s, "dropped events: %ld\n", cnt);
4414 count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
4421 static const struct file_operations tracing_stats_fops = {
4422 .open = tracing_open_generic,
4423 .read = tracing_stats_read,
4424 .llseek = generic_file_llseek,
4427 #ifdef CONFIG_DYNAMIC_FTRACE
4429 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
4435 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
4436 size_t cnt, loff_t *ppos)
4438 static char ftrace_dyn_info_buffer[1024];
4439 static DEFINE_MUTEX(dyn_info_mutex);
4440 unsigned long *p = filp->private_data;
4441 char *buf = ftrace_dyn_info_buffer;
4442 int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
4445 mutex_lock(&dyn_info_mutex);
4446 r = sprintf(buf, "%ld ", *p);
4448 r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
4451 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4453 mutex_unlock(&dyn_info_mutex);
4458 static const struct file_operations tracing_dyn_info_fops = {
4459 .open = tracing_open_generic,
4460 .read = tracing_read_dyn_info,
4461 .llseek = generic_file_llseek,
4465 static struct dentry *d_tracer;
4467 struct dentry *tracing_init_dentry(void)
4474 if (!debugfs_initialized())
4477 d_tracer = debugfs_create_dir("tracing", NULL);
4479 if (!d_tracer && !once) {
4481 pr_warning("Could not create debugfs directory 'tracing'\n");
4488 static struct dentry *d_percpu;
4490 struct dentry *tracing_dentry_percpu(void)
4493 struct dentry *d_tracer;
4498 d_tracer = tracing_init_dentry();
4503 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4505 if (!d_percpu && !once) {
4507 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4514 static void tracing_init_debugfs_percpu(long cpu)
4516 struct dentry *d_percpu = tracing_dentry_percpu();
4517 struct dentry *d_cpu;
4518 char cpu_dir[30]; /* 30 characters should be more than enough */
4523 snprintf(cpu_dir, 30, "cpu%ld", cpu);
4524 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4526 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4530 /* per cpu trace_pipe */
4531 trace_create_file("trace_pipe", 0444, d_cpu,
4532 (void *) cpu, &tracing_pipe_fops);
4535 trace_create_file("trace", 0644, d_cpu,
4536 (void *) cpu, &tracing_fops);
4538 trace_create_file("trace_pipe_raw", 0444, d_cpu,
4539 (void *) cpu, &tracing_buffers_fops);
4541 trace_create_file("stats", 0444, d_cpu,
4542 (void *) cpu, &tracing_stats_fops);
4544 trace_create_file("buffer_size_kb", 0444, d_cpu,
4545 (void *) cpu, &tracing_entries_fops);
4548 #ifdef CONFIG_FTRACE_SELFTEST
4549 /* Let selftest have access to static functions in this file */
4550 #include "trace_selftest.c"
4553 struct trace_option_dentry {
4554 struct tracer_opt *opt;
4555 struct tracer_flags *flags;
4556 struct dentry *entry;
4560 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4563 struct trace_option_dentry *topt = filp->private_data;
4566 if (topt->flags->val & topt->opt->bit)
4571 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4575 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4578 struct trace_option_dentry *topt = filp->private_data;
4582 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4586 if (val != 0 && val != 1)
4589 if (!!(topt->flags->val & topt->opt->bit) != val) {
4590 mutex_lock(&trace_types_lock);
4591 ret = __set_tracer_option(current_trace, topt->flags,
4593 mutex_unlock(&trace_types_lock);
4604 static const struct file_operations trace_options_fops = {
4605 .open = tracing_open_generic,
4606 .read = trace_options_read,
4607 .write = trace_options_write,
4608 .llseek = generic_file_llseek,
4612 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4615 long index = (long)filp->private_data;
4618 if (trace_flags & (1 << index))
4623 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4627 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4630 long index = (long)filp->private_data;
4634 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4638 if (val != 0 && val != 1)
4640 set_tracer_flags(1 << index, val);
4647 static const struct file_operations trace_options_core_fops = {
4648 .open = tracing_open_generic,
4649 .read = trace_options_core_read,
4650 .write = trace_options_core_write,
4651 .llseek = generic_file_llseek,
4654 struct dentry *trace_create_file(const char *name,
4656 struct dentry *parent,
4658 const struct file_operations *fops)
4662 ret = debugfs_create_file(name, mode, parent, data, fops);
4664 pr_warning("Could not create debugfs '%s' entry\n", name);
4670 static struct dentry *trace_options_init_dentry(void)
4672 struct dentry *d_tracer;
4673 static struct dentry *t_options;
4678 d_tracer = tracing_init_dentry();
4682 t_options = debugfs_create_dir("options", d_tracer);
4684 pr_warning("Could not create debugfs directory 'options'\n");
4692 create_trace_option_file(struct trace_option_dentry *topt,
4693 struct tracer_flags *flags,
4694 struct tracer_opt *opt)
4696 struct dentry *t_options;
4698 t_options = trace_options_init_dentry();
4702 topt->flags = flags;
4705 topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4706 &trace_options_fops);
4710 static struct trace_option_dentry *
4711 create_trace_option_files(struct tracer *tracer)
4713 struct trace_option_dentry *topts;
4714 struct tracer_flags *flags;
4715 struct tracer_opt *opts;
4721 flags = tracer->flags;
4723 if (!flags || !flags->opts)
4728 for (cnt = 0; opts[cnt].name; cnt++)
4731 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4735 for (cnt = 0; opts[cnt].name; cnt++)
4736 create_trace_option_file(&topts[cnt], flags,
4743 destroy_trace_option_files(struct trace_option_dentry *topts)
4750 for (cnt = 0; topts[cnt].opt; cnt++) {
4751 if (topts[cnt].entry)
4752 debugfs_remove(topts[cnt].entry);
4758 static struct dentry *
4759 create_trace_option_core_file(const char *option, long index)
4761 struct dentry *t_options;
4763 t_options = trace_options_init_dentry();
4767 return trace_create_file(option, 0644, t_options, (void *)index,
4768 &trace_options_core_fops);
4771 static __init void create_trace_options_dir(void)
4773 struct dentry *t_options;
4776 t_options = trace_options_init_dentry();
4780 for (i = 0; trace_options[i]; i++)
4781 create_trace_option_core_file(trace_options[i], i);
4785 rb_simple_read(struct file *filp, char __user *ubuf,
4786 size_t cnt, loff_t *ppos)
4788 struct trace_array *tr = filp->private_data;
4789 struct ring_buffer *buffer = tr->buffer;
4794 r = ring_buffer_record_is_on(buffer);
4798 r = sprintf(buf, "%d\n", r);
4800 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4804 rb_simple_write(struct file *filp, const char __user *ubuf,
4805 size_t cnt, loff_t *ppos)
4807 struct trace_array *tr = filp->private_data;
4808 struct ring_buffer *buffer = tr->buffer;
4812 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4817 mutex_lock(&trace_types_lock);
4819 ring_buffer_record_on(buffer);
4820 if (current_trace->start)
4821 current_trace->start(tr);
4823 ring_buffer_record_off(buffer);
4824 if (current_trace->stop)
4825 current_trace->stop(tr);
4827 mutex_unlock(&trace_types_lock);
4835 static const struct file_operations rb_simple_fops = {
4836 .open = tracing_open_generic,
4837 .read = rb_simple_read,
4838 .write = rb_simple_write,
4839 .llseek = default_llseek,
4842 static __init int tracer_init_debugfs(void)
4844 struct dentry *d_tracer;
4847 trace_access_lock_init();
4849 d_tracer = tracing_init_dentry();
4851 trace_create_file("trace_options", 0644, d_tracer,
4852 NULL, &tracing_iter_fops);
4854 trace_create_file("tracing_cpumask", 0644, d_tracer,
4855 NULL, &tracing_cpumask_fops);
4857 trace_create_file("trace", 0644, d_tracer,
4858 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4860 trace_create_file("available_tracers", 0444, d_tracer,
4861 &global_trace, &show_traces_fops);
4863 trace_create_file("current_tracer", 0644, d_tracer,
4864 &global_trace, &set_tracer_fops);
4866 #ifdef CONFIG_TRACER_MAX_TRACE
4867 trace_create_file("tracing_max_latency", 0644, d_tracer,
4868 &tracing_max_latency, &tracing_max_lat_fops);
4871 trace_create_file("tracing_thresh", 0644, d_tracer,
4872 &tracing_thresh, &tracing_max_lat_fops);
4874 trace_create_file("README", 0444, d_tracer,
4875 NULL, &tracing_readme_fops);
4877 trace_create_file("trace_pipe", 0444, d_tracer,
4878 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4880 trace_create_file("buffer_size_kb", 0644, d_tracer,
4881 (void *) RING_BUFFER_ALL_CPUS, &tracing_entries_fops);
4883 trace_create_file("buffer_total_size_kb", 0444, d_tracer,
4884 &global_trace, &tracing_total_entries_fops);
4886 trace_create_file("free_buffer", 0644, d_tracer,
4887 &global_trace, &tracing_free_buffer_fops);
4889 trace_create_file("trace_marker", 0220, d_tracer,
4890 NULL, &tracing_mark_fops);
4892 trace_create_file("saved_cmdlines", 0444, d_tracer,
4893 NULL, &tracing_saved_cmdlines_fops);
4895 trace_create_file("trace_clock", 0644, d_tracer, NULL,
4898 trace_create_file("tracing_on", 0644, d_tracer,
4899 &global_trace, &rb_simple_fops);
4901 #ifdef CONFIG_DYNAMIC_FTRACE
4902 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4903 &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4906 create_trace_options_dir();
4908 for_each_tracing_cpu(cpu)
4909 tracing_init_debugfs_percpu(cpu);
4914 static int trace_panic_handler(struct notifier_block *this,
4915 unsigned long event, void *unused)
4917 if (ftrace_dump_on_oops)
4918 ftrace_dump(ftrace_dump_on_oops);
4922 static struct notifier_block trace_panic_notifier = {
4923 .notifier_call = trace_panic_handler,
4925 .priority = 150 /* priority: INT_MAX >= x >= 0 */
4928 static int trace_die_handler(struct notifier_block *self,
4934 if (ftrace_dump_on_oops)
4935 ftrace_dump(ftrace_dump_on_oops);
4943 static struct notifier_block trace_die_notifier = {
4944 .notifier_call = trace_die_handler,
4949 * printk is set to max of 1024, we really don't need it that big.
4950 * Nothing should be printing 1000 characters anyway.
4952 #define TRACE_MAX_PRINT 1000
4955 * Define here KERN_TRACE so that we have one place to modify
4956 * it if we decide to change what log level the ftrace dump
4959 #define KERN_TRACE KERN_EMERG
4962 trace_printk_seq(struct trace_seq *s)
4964 /* Probably should print a warning here. */
4968 /* should be zero ended, but we are paranoid. */
4969 s->buffer[s->len] = 0;
4971 printk(KERN_TRACE "%s", s->buffer);
4976 void trace_init_global_iter(struct trace_iterator *iter)
4978 iter->tr = &global_trace;
4979 iter->trace = current_trace;
4980 iter->cpu_file = TRACE_PIPE_ALL_CPU;
4984 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4986 static arch_spinlock_t ftrace_dump_lock =
4987 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4988 /* use static because iter can be a bit big for the stack */
4989 static struct trace_iterator iter;
4990 unsigned int old_userobj;
4991 static int dump_ran;
4992 unsigned long flags;
4996 local_irq_save(flags);
4997 arch_spin_lock(&ftrace_dump_lock);
5005 /* Did function tracer already get disabled? */
5006 if (ftrace_is_dead()) {
5007 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
5008 printk("# MAY BE MISSING FUNCTION EVENTS\n");
5011 if (disable_tracing)
5014 trace_init_global_iter(&iter);
5016 for_each_tracing_cpu(cpu) {
5017 atomic_inc(&iter.tr->data[cpu]->disabled);
5020 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
5022 /* don't look at user memory in panic mode */
5023 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
5025 /* Simulate the iterator */
5026 iter.tr = &global_trace;
5027 iter.trace = current_trace;
5029 switch (oops_dump_mode) {
5031 iter.cpu_file = TRACE_PIPE_ALL_CPU;
5034 iter.cpu_file = raw_smp_processor_id();
5039 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
5040 iter.cpu_file = TRACE_PIPE_ALL_CPU;
5043 printk(KERN_TRACE "Dumping ftrace buffer:\n");
5046 * We need to stop all tracing on all CPUS to read the
5047 * the next buffer. This is a bit expensive, but is
5048 * not done often. We fill all what we can read,
5049 * and then release the locks again.
5052 while (!trace_empty(&iter)) {
5055 printk(KERN_TRACE "---------------------------------\n");
5059 /* reset all but tr, trace, and overruns */
5060 memset(&iter.seq, 0,
5061 sizeof(struct trace_iterator) -
5062 offsetof(struct trace_iterator, seq));
5063 iter.iter_flags |= TRACE_FILE_LAT_FMT;
5066 if (trace_find_next_entry_inc(&iter) != NULL) {
5069 ret = print_trace_line(&iter);
5070 if (ret != TRACE_TYPE_NO_CONSUME)
5071 trace_consume(&iter);
5073 touch_nmi_watchdog();
5075 trace_printk_seq(&iter.seq);
5079 printk(KERN_TRACE " (ftrace buffer empty)\n");
5081 printk(KERN_TRACE "---------------------------------\n");
5084 /* Re-enable tracing if requested */
5085 if (!disable_tracing) {
5086 trace_flags |= old_userobj;
5088 for_each_tracing_cpu(cpu) {
5089 atomic_dec(&iter.tr->data[cpu]->disabled);
5095 arch_spin_unlock(&ftrace_dump_lock);
5096 local_irq_restore(flags);
5099 /* By default: disable tracing after the dump */
5100 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
5102 __ftrace_dump(true, oops_dump_mode);
5104 EXPORT_SYMBOL_GPL(ftrace_dump);
5106 __init static int tracer_alloc_buffers(void)
5109 enum ring_buffer_flags rb_flags;
5114 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
5117 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
5118 goto out_free_buffer_mask;
5120 /* Only allocate trace_printk buffers if a trace_printk exists */
5121 if (__stop___trace_bprintk_fmt != __start___trace_bprintk_fmt)
5122 /* Must be called before global_trace.buffer is allocated */
5123 trace_printk_init_buffers();
5125 /* To save memory, keep the ring buffer size to its minimum */
5126 if (ring_buffer_expanded)
5127 ring_buf_size = trace_buf_size;
5131 rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
5133 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
5134 cpumask_copy(tracing_cpumask, cpu_all_mask);
5136 /* TODO: make the number of buffers hot pluggable with CPUS */
5137 global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
5138 if (!global_trace.buffer) {
5139 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
5141 goto out_free_cpumask;
5143 if (global_trace.buffer_disabled)
5147 #ifdef CONFIG_TRACER_MAX_TRACE
5148 max_tr.buffer = ring_buffer_alloc(1, rb_flags);
5149 if (!max_tr.buffer) {
5150 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
5152 ring_buffer_free(global_trace.buffer);
5153 goto out_free_cpumask;
5157 /* Allocate the first page for all buffers */
5158 for_each_tracing_cpu(i) {
5159 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
5160 max_tr.data[i] = &per_cpu(max_tr_data, i);
5163 set_buffer_entries(&global_trace,
5164 ring_buffer_size(global_trace.buffer, 0));
5165 #ifdef CONFIG_TRACER_MAX_TRACE
5166 set_buffer_entries(&max_tr, 1);
5169 trace_init_cmdlines();
5170 init_irq_work(&trace_work_wakeup, trace_wake_up);
5172 register_tracer(&nop_trace);
5173 current_trace = &nop_trace;
5174 /* All seems OK, enable tracing */
5175 tracing_disabled = 0;
5177 atomic_notifier_chain_register(&panic_notifier_list,
5178 &trace_panic_notifier);
5180 register_die_notifier(&trace_die_notifier);
5182 while (trace_boot_options) {
5185 option = strsep(&trace_boot_options, ",");
5186 trace_set_options(option);
5192 free_cpumask_var(tracing_cpumask);
5193 out_free_buffer_mask:
5194 free_cpumask_var(tracing_buffer_mask);
5199 __init static int clear_boot_tracer(void)
5202 * The default tracer at boot buffer is an init section.
5203 * This function is called in lateinit. If we did not
5204 * find the boot tracer, then clear it out, to prevent
5205 * later registration from accessing the buffer that is
5206 * about to be freed.
5208 if (!default_bootup_tracer)
5211 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
5212 default_bootup_tracer);
5213 default_bootup_tracer = NULL;
5218 early_initcall(tracer_alloc_buffers);
5219 fs_initcall(tracer_init_debugfs);
5220 late_initcall(clear_boot_tracer);