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 William Lee Irwin III
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
42 #include "trace_output.h"
44 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
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 * Kill all tracing for good (never come back).
83 * It is initialized to 1 but will turn to zero if the initialization
84 * of the tracer is successful. But that is the only place that sets
87 static int tracing_disabled
= 1;
89 DEFINE_PER_CPU(int, ftrace_cpu_disabled
);
91 static inline void ftrace_disable_cpu(void)
94 __this_cpu_inc(ftrace_cpu_disabled
);
97 static inline void ftrace_enable_cpu(void)
99 __this_cpu_dec(ftrace_cpu_disabled
);
103 cpumask_var_t __read_mostly tracing_buffer_mask
;
106 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
108 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
109 * is set, then ftrace_dump is called. This will output the contents
110 * of the ftrace buffers to the console. This is very useful for
111 * capturing traces that lead to crashes and outputing it to a
114 * It is default off, but you can enable it with either specifying
115 * "ftrace_dump_on_oops" in the kernel command line, or setting
116 * /proc/sys/kernel/ftrace_dump_on_oops
117 * Set 1 if you want to dump buffers of all CPUs
118 * Set 2 if you want to dump the buffer of the CPU that triggered oops
121 enum ftrace_dump_mode ftrace_dump_on_oops
;
123 static int tracing_set_tracer(const char *buf
);
125 #define MAX_TRACER_SIZE 100
126 static char bootup_tracer_buf
[MAX_TRACER_SIZE
] __initdata
;
127 static char *default_bootup_tracer
;
129 static int __init
set_cmdline_ftrace(char *str
)
131 strncpy(bootup_tracer_buf
, str
, MAX_TRACER_SIZE
);
132 default_bootup_tracer
= bootup_tracer_buf
;
133 /* We are using ftrace early, expand it */
134 ring_buffer_expanded
= 1;
137 __setup("ftrace=", set_cmdline_ftrace
);
139 static int __init
set_ftrace_dump_on_oops(char *str
)
141 if (*str
++ != '=' || !*str
) {
142 ftrace_dump_on_oops
= DUMP_ALL
;
146 if (!strcmp("orig_cpu", str
)) {
147 ftrace_dump_on_oops
= DUMP_ORIG
;
153 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops
);
155 unsigned long long ns2usecs(cycle_t nsec
)
163 * The global_trace is the descriptor that holds the tracing
164 * buffers for the live tracing. For each CPU, it contains
165 * a link list of pages that will store trace entries. The
166 * page descriptor of the pages in the memory is used to hold
167 * the link list by linking the lru item in the page descriptor
168 * to each of the pages in the buffer per CPU.
170 * For each active CPU there is a data field that holds the
171 * pages for the buffer for that CPU. Each CPU has the same number
172 * of pages allocated for its buffer.
174 static struct trace_array global_trace
;
176 static DEFINE_PER_CPU(struct trace_array_cpu
, global_trace_cpu
);
178 int filter_current_check_discard(struct ring_buffer
*buffer
,
179 struct ftrace_event_call
*call
, void *rec
,
180 struct ring_buffer_event
*event
)
182 return filter_check_discard(call
, rec
, buffer
, event
);
184 EXPORT_SYMBOL_GPL(filter_current_check_discard
);
186 cycle_t
ftrace_now(int cpu
)
190 /* Early boot up does not have a buffer yet */
191 if (!global_trace
.buffer
)
192 return trace_clock_local();
194 ts
= ring_buffer_time_stamp(global_trace
.buffer
, cpu
);
195 ring_buffer_normalize_time_stamp(global_trace
.buffer
, cpu
, &ts
);
201 * The max_tr is used to snapshot the global_trace when a maximum
202 * latency is reached. Some tracers will use this to store a maximum
203 * trace while it continues examining live traces.
205 * The buffers for the max_tr are set up the same as the global_trace.
206 * When a snapshot is taken, the link list of the max_tr is swapped
207 * with the link list of the global_trace and the buffers are reset for
208 * the global_trace so the tracing can continue.
210 static struct trace_array max_tr
;
212 static DEFINE_PER_CPU(struct trace_array_cpu
, max_tr_data
);
214 /* tracer_enabled is used to toggle activation of a tracer */
215 static int tracer_enabled
= 1;
218 * tracing_is_enabled - return tracer_enabled status
220 * This function is used by other tracers to know the status
221 * of the tracer_enabled flag. Tracers may use this function
222 * to know if it should enable their features when starting
223 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
225 int tracing_is_enabled(void)
227 return tracer_enabled
;
231 * trace_buf_size is the size in bytes that is allocated
232 * for a buffer. Note, the number of bytes is always rounded
235 * This number is purposely set to a low number of 16384.
236 * If the dump on oops happens, it will be much appreciated
237 * to not have to wait for all that output. Anyway this can be
238 * boot time and run time configurable.
240 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
242 static unsigned long trace_buf_size
= TRACE_BUF_SIZE_DEFAULT
;
244 /* trace_types holds a link list of available tracers. */
245 static struct tracer
*trace_types __read_mostly
;
247 /* current_trace points to the tracer that is currently active */
248 static struct tracer
*current_trace __read_mostly
;
251 * trace_types_lock is used to protect the trace_types list.
253 static DEFINE_MUTEX(trace_types_lock
);
256 * serialize the access of the ring buffer
258 * ring buffer serializes readers, but it is low level protection.
259 * The validity of the events (which returns by ring_buffer_peek() ..etc)
260 * are not protected by ring buffer.
262 * The content of events may become garbage if we allow other process consumes
263 * these events concurrently:
264 * A) the page of the consumed events may become a normal page
265 * (not reader page) in ring buffer, and this page will be rewrited
266 * by events producer.
267 * B) The page of the consumed events may become a page for splice_read,
268 * and this page will be returned to system.
270 * These primitives allow multi process access to different cpu ring buffer
273 * These primitives don't distinguish read-only and read-consume access.
274 * Multi read-only access are also serialized.
278 static DECLARE_RWSEM(all_cpu_access_lock
);
279 static DEFINE_PER_CPU(struct mutex
, cpu_access_lock
);
281 static inline void trace_access_lock(int cpu
)
283 if (cpu
== TRACE_PIPE_ALL_CPU
) {
284 /* gain it for accessing the whole ring buffer. */
285 down_write(&all_cpu_access_lock
);
287 /* gain it for accessing a cpu ring buffer. */
289 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
290 down_read(&all_cpu_access_lock
);
292 /* Secondly block other access to this @cpu ring buffer. */
293 mutex_lock(&per_cpu(cpu_access_lock
, cpu
));
297 static inline void trace_access_unlock(int cpu
)
299 if (cpu
== TRACE_PIPE_ALL_CPU
) {
300 up_write(&all_cpu_access_lock
);
302 mutex_unlock(&per_cpu(cpu_access_lock
, cpu
));
303 up_read(&all_cpu_access_lock
);
307 static inline void trace_access_lock_init(void)
311 for_each_possible_cpu(cpu
)
312 mutex_init(&per_cpu(cpu_access_lock
, cpu
));
317 static DEFINE_MUTEX(access_lock
);
319 static inline void trace_access_lock(int cpu
)
322 mutex_lock(&access_lock
);
325 static inline void trace_access_unlock(int cpu
)
328 mutex_unlock(&access_lock
);
331 static inline void trace_access_lock_init(void)
337 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
338 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
340 /* trace_flags holds trace_options default values */
341 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
| TRACE_ITER_PRINTK
|
342 TRACE_ITER_ANNOTATE
| TRACE_ITER_CONTEXT_INFO
| TRACE_ITER_SLEEP_TIME
|
343 TRACE_ITER_GRAPH_TIME
| TRACE_ITER_RECORD_CMD
;
345 static int trace_stop_count
;
346 static DEFINE_SPINLOCK(tracing_start_lock
);
349 * trace_wake_up - wake up tasks waiting for trace input
351 * Simply wakes up any task that is blocked on the trace_wait
352 * queue. These is used with trace_poll for tasks polling the trace.
354 void trace_wake_up(void)
358 if (trace_flags
& TRACE_ITER_BLOCK
)
361 * The runqueue_is_locked() can fail, but this is the best we
365 if (!runqueue_is_locked(cpu
))
366 wake_up(&trace_wait
);
370 static int __init
set_buf_size(char *str
)
372 unsigned long buf_size
;
376 buf_size
= memparse(str
, &str
);
377 /* nr_entries can not be zero */
380 trace_buf_size
= buf_size
;
383 __setup("trace_buf_size=", set_buf_size
);
385 static int __init
set_tracing_thresh(char *str
)
387 unsigned long threshhold
;
392 ret
= strict_strtoul(str
, 0, &threshhold
);
395 tracing_thresh
= threshhold
* 1000;
398 __setup("tracing_thresh=", set_tracing_thresh
);
400 unsigned long nsecs_to_usecs(unsigned long nsecs
)
405 /* These must match the bit postions in trace_iterator_flags */
406 static const char *trace_options
[] = {
435 { trace_clock_local
, "local" },
436 { trace_clock_global
, "global" },
442 * trace_parser_get_init - gets the buffer for trace parser
444 int trace_parser_get_init(struct trace_parser
*parser
, int size
)
446 memset(parser
, 0, sizeof(*parser
));
448 parser
->buffer
= kmalloc(size
, GFP_KERNEL
);
457 * trace_parser_put - frees the buffer for trace parser
459 void trace_parser_put(struct trace_parser
*parser
)
461 kfree(parser
->buffer
);
465 * trace_get_user - reads the user input string separated by space
466 * (matched by isspace(ch))
468 * For each string found the 'struct trace_parser' is updated,
469 * and the function returns.
471 * Returns number of bytes read.
473 * See kernel/trace/trace.h for 'struct trace_parser' details.
475 int trace_get_user(struct trace_parser
*parser
, const char __user
*ubuf
,
476 size_t cnt
, loff_t
*ppos
)
483 trace_parser_clear(parser
);
485 ret
= get_user(ch
, ubuf
++);
493 * The parser is not finished with the last write,
494 * continue reading the user input without skipping spaces.
497 /* skip white space */
498 while (cnt
&& isspace(ch
)) {
499 ret
= get_user(ch
, ubuf
++);
506 /* only spaces were written */
516 /* read the non-space input */
517 while (cnt
&& !isspace(ch
)) {
518 if (parser
->idx
< parser
->size
- 1)
519 parser
->buffer
[parser
->idx
++] = ch
;
524 ret
= get_user(ch
, ubuf
++);
531 /* We either got finished input or we have to wait for another call. */
533 parser
->buffer
[parser
->idx
] = 0;
534 parser
->cont
= false;
537 parser
->buffer
[parser
->idx
++] = ch
;
547 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
555 if (s
->len
<= s
->readpos
)
558 len
= s
->len
- s
->readpos
;
561 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
571 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
576 if (s
->len
<= s
->readpos
)
579 len
= s
->len
- s
->readpos
;
582 ret
= memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
591 * ftrace_max_lock is used to protect the swapping of buffers
592 * when taking a max snapshot. The buffers themselves are
593 * protected by per_cpu spinlocks. But the action of the swap
594 * needs its own lock.
596 * This is defined as a arch_spinlock_t in order to help
597 * with performance when lockdep debugging is enabled.
599 * It is also used in other places outside the update_max_tr
600 * so it needs to be defined outside of the
601 * CONFIG_TRACER_MAX_TRACE.
603 static arch_spinlock_t ftrace_max_lock
=
604 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
606 unsigned long __read_mostly tracing_thresh
;
608 #ifdef CONFIG_TRACER_MAX_TRACE
609 unsigned long __read_mostly tracing_max_latency
;
612 * Copy the new maximum trace into the separate maximum-trace
613 * structure. (this way the maximum trace is permanently saved,
614 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
617 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
619 struct trace_array_cpu
*data
= tr
->data
[cpu
];
620 struct trace_array_cpu
*max_data
;
623 max_tr
.time_start
= data
->preempt_timestamp
;
625 max_data
= max_tr
.data
[cpu
];
626 max_data
->saved_latency
= tracing_max_latency
;
627 max_data
->critical_start
= data
->critical_start
;
628 max_data
->critical_end
= data
->critical_end
;
630 memcpy(max_data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
631 max_data
->pid
= tsk
->pid
;
632 max_data
->uid
= task_uid(tsk
);
633 max_data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
634 max_data
->policy
= tsk
->policy
;
635 max_data
->rt_priority
= tsk
->rt_priority
;
637 /* record this tasks comm */
638 tracing_record_cmdline(tsk
);
642 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
644 * @tsk: the task with the latency
645 * @cpu: The cpu that initiated the trace.
647 * Flip the buffers between the @tr and the max_tr and record information
648 * about which task was the cause of this latency.
651 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
653 struct ring_buffer
*buf
= tr
->buffer
;
655 if (trace_stop_count
)
658 WARN_ON_ONCE(!irqs_disabled());
659 if (!current_trace
->use_max_tr
) {
663 arch_spin_lock(&ftrace_max_lock
);
665 tr
->buffer
= max_tr
.buffer
;
668 __update_max_tr(tr
, tsk
, cpu
);
669 arch_spin_unlock(&ftrace_max_lock
);
673 * update_max_tr_single - only copy one trace over, and reset the rest
675 * @tsk - task with the latency
676 * @cpu - the cpu of the buffer to copy.
678 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
681 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
685 if (trace_stop_count
)
688 WARN_ON_ONCE(!irqs_disabled());
689 if (!current_trace
->use_max_tr
) {
694 arch_spin_lock(&ftrace_max_lock
);
696 ftrace_disable_cpu();
698 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
702 * We failed to swap the buffer due to a commit taking
703 * place on this CPU. We fail to record, but we reset
704 * the max trace buffer (no one writes directly to it)
705 * and flag that it failed.
707 trace_array_printk(&max_tr
, _THIS_IP_
,
708 "Failed to swap buffers due to commit in progress\n");
713 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
&& ret
!= -EBUSY
);
715 __update_max_tr(tr
, tsk
, cpu
);
716 arch_spin_unlock(&ftrace_max_lock
);
718 #endif /* CONFIG_TRACER_MAX_TRACE */
721 * register_tracer - register a tracer with the ftrace system.
722 * @type - the plugin for the tracer
724 * Register a new plugin tracer.
726 int register_tracer(struct tracer
*type
)
727 __releases(kernel_lock
)
728 __acquires(kernel_lock
)
734 pr_info("Tracer must have a name\n");
738 if (strlen(type
->name
) >= MAX_TRACER_SIZE
) {
739 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE
);
743 mutex_lock(&trace_types_lock
);
745 tracing_selftest_running
= true;
747 for (t
= trace_types
; t
; t
= t
->next
) {
748 if (strcmp(type
->name
, t
->name
) == 0) {
750 pr_info("Tracer %s already registered\n",
758 type
->set_flag
= &dummy_set_flag
;
760 type
->flags
= &dummy_tracer_flags
;
762 if (!type
->flags
->opts
)
763 type
->flags
->opts
= dummy_tracer_opt
;
764 if (!type
->wait_pipe
)
765 type
->wait_pipe
= default_wait_pipe
;
768 #ifdef CONFIG_FTRACE_STARTUP_TEST
769 if (type
->selftest
&& !tracing_selftest_disabled
) {
770 struct tracer
*saved_tracer
= current_trace
;
771 struct trace_array
*tr
= &global_trace
;
774 * Run a selftest on this tracer.
775 * Here we reset the trace buffer, and set the current
776 * tracer to be this tracer. The tracer can then run some
777 * internal tracing to verify that everything is in order.
778 * If we fail, we do not register this tracer.
780 tracing_reset_online_cpus(tr
);
782 current_trace
= type
;
783 /* the test is responsible for initializing and enabling */
784 pr_info("Testing tracer %s: ", type
->name
);
785 ret
= type
->selftest(type
, tr
);
786 /* the test is responsible for resetting too */
787 current_trace
= saved_tracer
;
789 printk(KERN_CONT
"FAILED!\n");
792 /* Only reset on passing, to avoid touching corrupted buffers */
793 tracing_reset_online_cpus(tr
);
795 printk(KERN_CONT
"PASSED\n");
799 type
->next
= trace_types
;
803 tracing_selftest_running
= false;
804 mutex_unlock(&trace_types_lock
);
806 if (ret
|| !default_bootup_tracer
)
809 if (strncmp(default_bootup_tracer
, type
->name
, MAX_TRACER_SIZE
))
812 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
813 /* Do we want this tracer to start on bootup? */
814 tracing_set_tracer(type
->name
);
815 default_bootup_tracer
= NULL
;
816 /* disable other selftests, since this will break it. */
817 tracing_selftest_disabled
= 1;
818 #ifdef CONFIG_FTRACE_STARTUP_TEST
819 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
827 void unregister_tracer(struct tracer
*type
)
831 mutex_lock(&trace_types_lock
);
832 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
836 pr_info("Tracer %s not registered\n", type
->name
);
842 if (type
== current_trace
&& tracer_enabled
) {
845 if (current_trace
->stop
)
846 current_trace
->stop(&global_trace
);
847 current_trace
= &nop_trace
;
850 mutex_unlock(&trace_types_lock
);
853 static void __tracing_reset(struct ring_buffer
*buffer
, int cpu
)
855 ftrace_disable_cpu();
856 ring_buffer_reset_cpu(buffer
, cpu
);
860 void tracing_reset(struct trace_array
*tr
, int cpu
)
862 struct ring_buffer
*buffer
= tr
->buffer
;
864 ring_buffer_record_disable(buffer
);
866 /* Make sure all commits have finished */
868 __tracing_reset(buffer
, cpu
);
870 ring_buffer_record_enable(buffer
);
873 void tracing_reset_online_cpus(struct trace_array
*tr
)
875 struct ring_buffer
*buffer
= tr
->buffer
;
878 ring_buffer_record_disable(buffer
);
880 /* Make sure all commits have finished */
883 tr
->time_start
= ftrace_now(tr
->cpu
);
885 for_each_online_cpu(cpu
)
886 __tracing_reset(buffer
, cpu
);
888 ring_buffer_record_enable(buffer
);
891 void tracing_reset_current(int cpu
)
893 tracing_reset(&global_trace
, cpu
);
896 void tracing_reset_current_online_cpus(void)
898 tracing_reset_online_cpus(&global_trace
);
901 #define SAVED_CMDLINES 128
902 #define NO_CMDLINE_MAP UINT_MAX
903 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
904 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
905 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
906 static int cmdline_idx
;
907 static arch_spinlock_t trace_cmdline_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
909 /* temporary disable recording */
910 static atomic_t trace_record_cmdline_disabled __read_mostly
;
912 static void trace_init_cmdlines(void)
914 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
915 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
919 int is_tracing_stopped(void)
921 return trace_stop_count
;
925 * ftrace_off_permanent - disable all ftrace code permanently
927 * This should only be called when a serious anomally has
928 * been detected. This will turn off the function tracing,
929 * ring buffers, and other tracing utilites. It takes no
930 * locks and can be called from any context.
932 void ftrace_off_permanent(void)
934 tracing_disabled
= 1;
936 tracing_off_permanent();
940 * tracing_start - quick start of the tracer
942 * If tracing is enabled but was stopped by tracing_stop,
943 * this will start the tracer back up.
945 void tracing_start(void)
947 struct ring_buffer
*buffer
;
950 if (tracing_disabled
)
953 spin_lock_irqsave(&tracing_start_lock
, flags
);
954 if (--trace_stop_count
) {
955 if (trace_stop_count
< 0) {
956 /* Someone screwed up their debugging */
958 trace_stop_count
= 0;
963 /* Prevent the buffers from switching */
964 arch_spin_lock(&ftrace_max_lock
);
966 buffer
= global_trace
.buffer
;
968 ring_buffer_record_enable(buffer
);
970 buffer
= max_tr
.buffer
;
972 ring_buffer_record_enable(buffer
);
974 arch_spin_unlock(&ftrace_max_lock
);
978 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
982 * tracing_stop - quick stop of the tracer
984 * Light weight way to stop tracing. Use in conjunction with
987 void tracing_stop(void)
989 struct ring_buffer
*buffer
;
993 spin_lock_irqsave(&tracing_start_lock
, flags
);
994 if (trace_stop_count
++)
997 /* Prevent the buffers from switching */
998 arch_spin_lock(&ftrace_max_lock
);
1000 buffer
= global_trace
.buffer
;
1002 ring_buffer_record_disable(buffer
);
1004 buffer
= max_tr
.buffer
;
1006 ring_buffer_record_disable(buffer
);
1008 arch_spin_unlock(&ftrace_max_lock
);
1011 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
1014 void trace_stop_cmdline_recording(void);
1016 static void trace_save_cmdline(struct task_struct
*tsk
)
1020 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
1024 * It's not the end of the world if we don't get
1025 * the lock, but we also don't want to spin
1026 * nor do we want to disable interrupts,
1027 * so if we miss here, then better luck next time.
1029 if (!arch_spin_trylock(&trace_cmdline_lock
))
1032 idx
= map_pid_to_cmdline
[tsk
->pid
];
1033 if (idx
== NO_CMDLINE_MAP
) {
1034 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
1037 * Check whether the cmdline buffer at idx has a pid
1038 * mapped. We are going to overwrite that entry so we
1039 * need to clear the map_pid_to_cmdline. Otherwise we
1040 * would read the new comm for the old pid.
1042 pid
= map_cmdline_to_pid
[idx
];
1043 if (pid
!= NO_CMDLINE_MAP
)
1044 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
1046 map_cmdline_to_pid
[idx
] = tsk
->pid
;
1047 map_pid_to_cmdline
[tsk
->pid
] = idx
;
1052 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
1054 arch_spin_unlock(&trace_cmdline_lock
);
1057 void trace_find_cmdline(int pid
, char comm
[])
1062 strcpy(comm
, "<idle>");
1066 if (WARN_ON_ONCE(pid
< 0)) {
1067 strcpy(comm
, "<XXX>");
1071 if (pid
> PID_MAX_DEFAULT
) {
1072 strcpy(comm
, "<...>");
1077 arch_spin_lock(&trace_cmdline_lock
);
1078 map
= map_pid_to_cmdline
[pid
];
1079 if (map
!= NO_CMDLINE_MAP
)
1080 strcpy(comm
, saved_cmdlines
[map
]);
1082 strcpy(comm
, "<...>");
1084 arch_spin_unlock(&trace_cmdline_lock
);
1088 void tracing_record_cmdline(struct task_struct
*tsk
)
1090 if (atomic_read(&trace_record_cmdline_disabled
) || !tracer_enabled
||
1094 trace_save_cmdline(tsk
);
1098 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
1101 struct task_struct
*tsk
= current
;
1103 entry
->preempt_count
= pc
& 0xff;
1104 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
1105 entry
->lock_depth
= (tsk
) ? tsk
->lock_depth
: 0;
1107 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1108 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
1110 TRACE_FLAG_IRQS_NOSUPPORT
|
1112 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
1113 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
1114 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
1116 EXPORT_SYMBOL_GPL(tracing_generic_entry_update
);
1118 struct ring_buffer_event
*
1119 trace_buffer_lock_reserve(struct ring_buffer
*buffer
,
1122 unsigned long flags
, int pc
)
1124 struct ring_buffer_event
*event
;
1126 event
= ring_buffer_lock_reserve(buffer
, len
);
1127 if (event
!= NULL
) {
1128 struct trace_entry
*ent
= ring_buffer_event_data(event
);
1130 tracing_generic_entry_update(ent
, flags
, pc
);
1138 __trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1139 struct ring_buffer_event
*event
,
1140 unsigned long flags
, int pc
,
1143 ring_buffer_unlock_commit(buffer
, event
);
1145 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1146 ftrace_trace_userstack(buffer
, flags
, pc
);
1152 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1153 struct ring_buffer_event
*event
,
1154 unsigned long flags
, int pc
)
1156 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1159 struct ring_buffer_event
*
1160 trace_current_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1161 int type
, unsigned long len
,
1162 unsigned long flags
, int pc
)
1164 *current_rb
= global_trace
.buffer
;
1165 return trace_buffer_lock_reserve(*current_rb
,
1166 type
, len
, flags
, pc
);
1168 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve
);
1170 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
1171 struct ring_buffer_event
*event
,
1172 unsigned long flags
, int pc
)
1174 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1176 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit
);
1178 void trace_nowake_buffer_unlock_commit(struct ring_buffer
*buffer
,
1179 struct ring_buffer_event
*event
,
1180 unsigned long flags
, int pc
)
1182 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 0);
1184 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit
);
1186 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
1187 struct ring_buffer_event
*event
)
1189 ring_buffer_discard_commit(buffer
, event
);
1191 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit
);
1194 trace_function(struct trace_array
*tr
,
1195 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1198 struct ftrace_event_call
*call
= &event_function
;
1199 struct ring_buffer
*buffer
= tr
->buffer
;
1200 struct ring_buffer_event
*event
;
1201 struct ftrace_entry
*entry
;
1203 /* If we are reading the ring buffer, don't trace */
1204 if (unlikely(__this_cpu_read(ftrace_cpu_disabled
)))
1207 event
= trace_buffer_lock_reserve(buffer
, TRACE_FN
, sizeof(*entry
),
1211 entry
= ring_buffer_event_data(event
);
1213 entry
->parent_ip
= parent_ip
;
1215 if (!filter_check_discard(call
, entry
, buffer
, event
))
1216 ring_buffer_unlock_commit(buffer
, event
);
1220 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1221 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1224 if (likely(!atomic_read(&data
->disabled
)))
1225 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
1228 #ifdef CONFIG_STACKTRACE
1229 static void __ftrace_trace_stack(struct ring_buffer
*buffer
,
1230 unsigned long flags
,
1233 struct ftrace_event_call
*call
= &event_kernel_stack
;
1234 struct ring_buffer_event
*event
;
1235 struct stack_entry
*entry
;
1236 struct stack_trace trace
;
1238 event
= trace_buffer_lock_reserve(buffer
, TRACE_STACK
,
1239 sizeof(*entry
), flags
, pc
);
1242 entry
= ring_buffer_event_data(event
);
1243 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1245 trace
.nr_entries
= 0;
1246 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1248 trace
.entries
= entry
->caller
;
1250 save_stack_trace(&trace
);
1251 if (!filter_check_discard(call
, entry
, buffer
, event
))
1252 ring_buffer_unlock_commit(buffer
, event
);
1255 void ftrace_trace_stack(struct ring_buffer
*buffer
, unsigned long flags
,
1258 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1261 __ftrace_trace_stack(buffer
, flags
, skip
, pc
);
1264 void __trace_stack(struct trace_array
*tr
, unsigned long flags
, int skip
,
1267 __ftrace_trace_stack(tr
->buffer
, flags
, skip
, pc
);
1271 * trace_dump_stack - record a stack back trace in the trace buffer
1273 void trace_dump_stack(void)
1275 unsigned long flags
;
1277 if (tracing_disabled
|| tracing_selftest_running
)
1280 local_save_flags(flags
);
1282 /* skipping 3 traces, seems to get us at the caller of this function */
1283 __ftrace_trace_stack(global_trace
.buffer
, flags
, 3, preempt_count());
1287 ftrace_trace_userstack(struct ring_buffer
*buffer
, unsigned long flags
, int pc
)
1289 struct ftrace_event_call
*call
= &event_user_stack
;
1290 struct ring_buffer_event
*event
;
1291 struct userstack_entry
*entry
;
1292 struct stack_trace trace
;
1294 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1298 * NMIs can not handle page faults, even with fix ups.
1299 * The save user stack can (and often does) fault.
1301 if (unlikely(in_nmi()))
1304 event
= trace_buffer_lock_reserve(buffer
, TRACE_USER_STACK
,
1305 sizeof(*entry
), flags
, pc
);
1308 entry
= ring_buffer_event_data(event
);
1310 entry
->tgid
= current
->tgid
;
1311 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1313 trace
.nr_entries
= 0;
1314 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1316 trace
.entries
= entry
->caller
;
1318 save_stack_trace_user(&trace
);
1319 if (!filter_check_discard(call
, entry
, buffer
, event
))
1320 ring_buffer_unlock_commit(buffer
, event
);
1324 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1326 ftrace_trace_userstack(tr
, flags
, preempt_count());
1330 #endif /* CONFIG_STACKTRACE */
1333 * trace_vbprintk - write binary msg to tracing buffer
1336 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1338 static arch_spinlock_t trace_buf_lock
=
1339 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
1340 static u32 trace_buf
[TRACE_BUF_SIZE
];
1342 struct ftrace_event_call
*call
= &event_bprint
;
1343 struct ring_buffer_event
*event
;
1344 struct ring_buffer
*buffer
;
1345 struct trace_array
*tr
= &global_trace
;
1346 struct trace_array_cpu
*data
;
1347 struct bprint_entry
*entry
;
1348 unsigned long flags
;
1350 int cpu
, len
= 0, size
, pc
;
1352 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1355 /* Don't pollute graph traces with trace_vprintk internals */
1356 pause_graph_tracing();
1358 pc
= preempt_count();
1359 preempt_disable_notrace();
1360 cpu
= raw_smp_processor_id();
1361 data
= tr
->data
[cpu
];
1363 disable
= atomic_inc_return(&data
->disabled
);
1364 if (unlikely(disable
!= 1))
1367 /* Lockdep uses trace_printk for lock tracing */
1368 local_irq_save(flags
);
1369 arch_spin_lock(&trace_buf_lock
);
1370 len
= vbin_printf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1372 if (len
> TRACE_BUF_SIZE
|| len
< 0)
1375 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1376 buffer
= tr
->buffer
;
1377 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
1381 entry
= ring_buffer_event_data(event
);
1385 memcpy(entry
->buf
, trace_buf
, sizeof(u32
) * len
);
1386 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1387 ring_buffer_unlock_commit(buffer
, event
);
1388 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1392 arch_spin_unlock(&trace_buf_lock
);
1393 local_irq_restore(flags
);
1396 atomic_dec_return(&data
->disabled
);
1397 preempt_enable_notrace();
1398 unpause_graph_tracing();
1402 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1404 int trace_array_printk(struct trace_array
*tr
,
1405 unsigned long ip
, const char *fmt
, ...)
1410 if (!(trace_flags
& TRACE_ITER_PRINTK
))
1414 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
1419 int trace_array_vprintk(struct trace_array
*tr
,
1420 unsigned long ip
, const char *fmt
, va_list args
)
1422 static arch_spinlock_t trace_buf_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
1423 static char trace_buf
[TRACE_BUF_SIZE
];
1425 struct ftrace_event_call
*call
= &event_print
;
1426 struct ring_buffer_event
*event
;
1427 struct ring_buffer
*buffer
;
1428 struct trace_array_cpu
*data
;
1429 int cpu
, len
= 0, size
, pc
;
1430 struct print_entry
*entry
;
1431 unsigned long irq_flags
;
1434 if (tracing_disabled
|| tracing_selftest_running
)
1437 pc
= preempt_count();
1438 preempt_disable_notrace();
1439 cpu
= raw_smp_processor_id();
1440 data
= tr
->data
[cpu
];
1442 disable
= atomic_inc_return(&data
->disabled
);
1443 if (unlikely(disable
!= 1))
1446 pause_graph_tracing();
1447 raw_local_irq_save(irq_flags
);
1448 arch_spin_lock(&trace_buf_lock
);
1449 len
= vsnprintf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1451 size
= sizeof(*entry
) + len
+ 1;
1452 buffer
= tr
->buffer
;
1453 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
1457 entry
= ring_buffer_event_data(event
);
1460 memcpy(&entry
->buf
, trace_buf
, len
);
1461 entry
->buf
[len
] = '\0';
1462 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1463 ring_buffer_unlock_commit(buffer
, event
);
1464 ftrace_trace_stack(buffer
, irq_flags
, 6, pc
);
1468 arch_spin_unlock(&trace_buf_lock
);
1469 raw_local_irq_restore(irq_flags
);
1470 unpause_graph_tracing();
1472 atomic_dec_return(&data
->disabled
);
1473 preempt_enable_notrace();
1478 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1480 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
1482 EXPORT_SYMBOL_GPL(trace_vprintk
);
1484 static void trace_iterator_increment(struct trace_iterator
*iter
)
1486 /* Don't allow ftrace to trace into the ring buffers */
1487 ftrace_disable_cpu();
1490 if (iter
->buffer_iter
[iter
->cpu
])
1491 ring_buffer_read(iter
->buffer_iter
[iter
->cpu
], NULL
);
1493 ftrace_enable_cpu();
1496 static struct trace_entry
*
1497 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
,
1498 unsigned long *lost_events
)
1500 struct ring_buffer_event
*event
;
1501 struct ring_buffer_iter
*buf_iter
= iter
->buffer_iter
[cpu
];
1503 /* Don't allow ftrace to trace into the ring buffers */
1504 ftrace_disable_cpu();
1507 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1509 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
,
1512 ftrace_enable_cpu();
1514 return event
? ring_buffer_event_data(event
) : NULL
;
1517 static struct trace_entry
*
1518 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
,
1519 unsigned long *missing_events
, u64
*ent_ts
)
1521 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1522 struct trace_entry
*ent
, *next
= NULL
;
1523 unsigned long lost_events
= 0, next_lost
= 0;
1524 int cpu_file
= iter
->cpu_file
;
1525 u64 next_ts
= 0, ts
;
1530 * If we are in a per_cpu trace file, don't bother by iterating over
1531 * all cpu and peek directly.
1533 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1534 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1536 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
, missing_events
);
1538 *ent_cpu
= cpu_file
;
1543 for_each_tracing_cpu(cpu
) {
1545 if (ring_buffer_empty_cpu(buffer
, cpu
))
1548 ent
= peek_next_entry(iter
, cpu
, &ts
, &lost_events
);
1551 * Pick the entry with the smallest timestamp:
1553 if (ent
&& (!next
|| ts
< next_ts
)) {
1557 next_lost
= lost_events
;
1562 *ent_cpu
= next_cpu
;
1568 *missing_events
= next_lost
;
1573 /* Find the next real entry, without updating the iterator itself */
1574 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1575 int *ent_cpu
, u64
*ent_ts
)
1577 return __find_next_entry(iter
, ent_cpu
, NULL
, ent_ts
);
1580 /* Find the next real entry, and increment the iterator to the next entry */
1581 void *trace_find_next_entry_inc(struct trace_iterator
*iter
)
1583 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
,
1584 &iter
->lost_events
, &iter
->ts
);
1587 trace_iterator_increment(iter
);
1589 return iter
->ent
? iter
: NULL
;
1592 static void trace_consume(struct trace_iterator
*iter
)
1594 /* Don't allow ftrace to trace into the ring buffers */
1595 ftrace_disable_cpu();
1596 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
,
1597 &iter
->lost_events
);
1598 ftrace_enable_cpu();
1601 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1603 struct trace_iterator
*iter
= m
->private;
1607 WARN_ON_ONCE(iter
->leftover
);
1611 /* can't go backwards */
1616 ent
= trace_find_next_entry_inc(iter
);
1620 while (ent
&& iter
->idx
< i
)
1621 ent
= trace_find_next_entry_inc(iter
);
1628 void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
1630 struct trace_array
*tr
= iter
->tr
;
1631 struct ring_buffer_event
*event
;
1632 struct ring_buffer_iter
*buf_iter
;
1633 unsigned long entries
= 0;
1636 tr
->data
[cpu
]->skipped_entries
= 0;
1638 if (!iter
->buffer_iter
[cpu
])
1641 buf_iter
= iter
->buffer_iter
[cpu
];
1642 ring_buffer_iter_reset(buf_iter
);
1645 * We could have the case with the max latency tracers
1646 * that a reset never took place on a cpu. This is evident
1647 * by the timestamp being before the start of the buffer.
1649 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
1650 if (ts
>= iter
->tr
->time_start
)
1653 ring_buffer_read(buf_iter
, NULL
);
1656 tr
->data
[cpu
]->skipped_entries
= entries
;
1660 * The current tracer is copied to avoid a global locking
1663 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1665 struct trace_iterator
*iter
= m
->private;
1666 static struct tracer
*old_tracer
;
1667 int cpu_file
= iter
->cpu_file
;
1672 /* copy the tracer to avoid using a global lock all around */
1673 mutex_lock(&trace_types_lock
);
1674 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1675 old_tracer
= current_trace
;
1676 *iter
->trace
= *current_trace
;
1678 mutex_unlock(&trace_types_lock
);
1680 atomic_inc(&trace_record_cmdline_disabled
);
1682 if (*pos
!= iter
->pos
) {
1687 ftrace_disable_cpu();
1689 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1690 for_each_tracing_cpu(cpu
)
1691 tracing_iter_reset(iter
, cpu
);
1693 tracing_iter_reset(iter
, cpu_file
);
1695 ftrace_enable_cpu();
1698 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1703 * If we overflowed the seq_file before, then we want
1704 * to just reuse the trace_seq buffer again.
1710 p
= s_next(m
, p
, &l
);
1714 trace_event_read_lock();
1715 trace_access_lock(cpu_file
);
1719 static void s_stop(struct seq_file
*m
, void *p
)
1721 struct trace_iterator
*iter
= m
->private;
1723 atomic_dec(&trace_record_cmdline_disabled
);
1724 trace_access_unlock(iter
->cpu_file
);
1725 trace_event_read_unlock();
1728 static void print_lat_help_header(struct seq_file
*m
)
1730 seq_puts(m
, "# _------=> CPU# \n");
1731 seq_puts(m
, "# / _-----=> irqs-off \n");
1732 seq_puts(m
, "# | / _----=> need-resched \n");
1733 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1734 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1735 seq_puts(m
, "# |||| /_--=> lock-depth \n");
1736 seq_puts(m
, "# |||||/ delay \n");
1737 seq_puts(m
, "# cmd pid |||||| time | caller \n");
1738 seq_puts(m
, "# \\ / |||||| \\ | / \n");
1741 static void print_func_help_header(struct seq_file
*m
)
1743 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1744 seq_puts(m
, "# | | | | |\n");
1749 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1751 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1752 struct trace_array
*tr
= iter
->tr
;
1753 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1754 struct tracer
*type
= current_trace
;
1755 unsigned long entries
= 0;
1756 unsigned long total
= 0;
1757 unsigned long count
;
1758 const char *name
= "preemption";
1765 for_each_tracing_cpu(cpu
) {
1766 count
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
1768 * If this buffer has skipped entries, then we hold all
1769 * entries for the trace and we need to ignore the
1770 * ones before the time stamp.
1772 if (tr
->data
[cpu
]->skipped_entries
) {
1773 count
-= tr
->data
[cpu
]->skipped_entries
;
1774 /* total is the same as the entries */
1778 ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
1782 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
1784 seq_puts(m
, "# -----------------------------------"
1785 "---------------------------------\n");
1786 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1787 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1788 nsecs_to_usecs(data
->saved_latency
),
1792 #if defined(CONFIG_PREEMPT_NONE)
1794 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1796 #elif defined(CONFIG_PREEMPT)
1801 /* These are reserved for later use */
1804 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1808 seq_puts(m
, "# -----------------\n");
1809 seq_printf(m
, "# | task: %.16s-%d "
1810 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1811 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1812 data
->policy
, data
->rt_priority
);
1813 seq_puts(m
, "# -----------------\n");
1815 if (data
->critical_start
) {
1816 seq_puts(m
, "# => started at: ");
1817 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1818 trace_print_seq(m
, &iter
->seq
);
1819 seq_puts(m
, "\n# => ended at: ");
1820 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1821 trace_print_seq(m
, &iter
->seq
);
1822 seq_puts(m
, "\n#\n");
1828 static void test_cpu_buff_start(struct trace_iterator
*iter
)
1830 struct trace_seq
*s
= &iter
->seq
;
1832 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
1835 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
1838 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
1841 if (iter
->tr
->data
[iter
->cpu
]->skipped_entries
)
1844 cpumask_set_cpu(iter
->cpu
, iter
->started
);
1846 /* Don't print started cpu buffer for the first entry of the trace */
1848 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
1852 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
1854 struct trace_seq
*s
= &iter
->seq
;
1855 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1856 struct trace_entry
*entry
;
1857 struct trace_event
*event
;
1861 test_cpu_buff_start(iter
);
1863 event
= ftrace_find_event(entry
->type
);
1865 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1866 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1867 if (!trace_print_lat_context(iter
))
1870 if (!trace_print_context(iter
))
1876 return event
->funcs
->trace(iter
, sym_flags
, event
);
1878 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
1881 return TRACE_TYPE_HANDLED
;
1883 return TRACE_TYPE_PARTIAL_LINE
;
1886 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
1888 struct trace_seq
*s
= &iter
->seq
;
1889 struct trace_entry
*entry
;
1890 struct trace_event
*event
;
1894 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1895 if (!trace_seq_printf(s
, "%d %d %llu ",
1896 entry
->pid
, iter
->cpu
, iter
->ts
))
1900 event
= ftrace_find_event(entry
->type
);
1902 return event
->funcs
->raw(iter
, 0, event
);
1904 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
1907 return TRACE_TYPE_HANDLED
;
1909 return TRACE_TYPE_PARTIAL_LINE
;
1912 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
1914 struct trace_seq
*s
= &iter
->seq
;
1915 unsigned char newline
= '\n';
1916 struct trace_entry
*entry
;
1917 struct trace_event
*event
;
1921 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1922 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1923 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1924 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
1927 event
= ftrace_find_event(entry
->type
);
1929 enum print_line_t ret
= event
->funcs
->hex(iter
, 0, event
);
1930 if (ret
!= TRACE_TYPE_HANDLED
)
1934 SEQ_PUT_FIELD_RET(s
, newline
);
1936 return TRACE_TYPE_HANDLED
;
1939 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
1941 struct trace_seq
*s
= &iter
->seq
;
1942 struct trace_entry
*entry
;
1943 struct trace_event
*event
;
1947 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1948 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1949 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
1950 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
1953 event
= ftrace_find_event(entry
->type
);
1954 return event
? event
->funcs
->binary(iter
, 0, event
) :
1958 int trace_empty(struct trace_iterator
*iter
)
1962 /* If we are looking at one CPU buffer, only check that one */
1963 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
1964 cpu
= iter
->cpu_file
;
1965 if (iter
->buffer_iter
[cpu
]) {
1966 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1969 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1975 for_each_tracing_cpu(cpu
) {
1976 if (iter
->buffer_iter
[cpu
]) {
1977 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1980 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1988 /* Called with trace_event_read_lock() held. */
1989 enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
1991 enum print_line_t ret
;
1993 if (iter
->lost_events
)
1994 trace_seq_printf(&iter
->seq
, "CPU:%d [LOST %lu EVENTS]\n",
1995 iter
->cpu
, iter
->lost_events
);
1997 if (iter
->trace
&& iter
->trace
->print_line
) {
1998 ret
= iter
->trace
->print_line(iter
);
1999 if (ret
!= TRACE_TYPE_UNHANDLED
)
2003 if (iter
->ent
->type
== TRACE_BPRINT
&&
2004 trace_flags
& TRACE_ITER_PRINTK
&&
2005 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2006 return trace_print_bprintk_msg_only(iter
);
2008 if (iter
->ent
->type
== TRACE_PRINT
&&
2009 trace_flags
& TRACE_ITER_PRINTK
&&
2010 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2011 return trace_print_printk_msg_only(iter
);
2013 if (trace_flags
& TRACE_ITER_BIN
)
2014 return print_bin_fmt(iter
);
2016 if (trace_flags
& TRACE_ITER_HEX
)
2017 return print_hex_fmt(iter
);
2019 if (trace_flags
& TRACE_ITER_RAW
)
2020 return print_raw_fmt(iter
);
2022 return print_trace_fmt(iter
);
2025 void trace_default_header(struct seq_file
*m
)
2027 struct trace_iterator
*iter
= m
->private;
2029 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2030 /* print nothing if the buffers are empty */
2031 if (trace_empty(iter
))
2033 print_trace_header(m
, iter
);
2034 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2035 print_lat_help_header(m
);
2037 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2038 print_func_help_header(m
);
2042 static int s_show(struct seq_file
*m
, void *v
)
2044 struct trace_iterator
*iter
= v
;
2047 if (iter
->ent
== NULL
) {
2049 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
2052 if (iter
->trace
&& iter
->trace
->print_header
)
2053 iter
->trace
->print_header(m
);
2055 trace_default_header(m
);
2057 } else if (iter
->leftover
) {
2059 * If we filled the seq_file buffer earlier, we
2060 * want to just show it now.
2062 ret
= trace_print_seq(m
, &iter
->seq
);
2064 /* ret should this time be zero, but you never know */
2065 iter
->leftover
= ret
;
2068 print_trace_line(iter
);
2069 ret
= trace_print_seq(m
, &iter
->seq
);
2071 * If we overflow the seq_file buffer, then it will
2072 * ask us for this data again at start up.
2074 * ret is 0 if seq_file write succeeded.
2077 iter
->leftover
= ret
;
2083 static const struct seq_operations tracer_seq_ops
= {
2090 static struct trace_iterator
*
2091 __tracing_open(struct inode
*inode
, struct file
*file
)
2093 long cpu_file
= (long) inode
->i_private
;
2094 void *fail_ret
= ERR_PTR(-ENOMEM
);
2095 struct trace_iterator
*iter
;
2099 if (tracing_disabled
)
2100 return ERR_PTR(-ENODEV
);
2102 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2104 return ERR_PTR(-ENOMEM
);
2107 * We make a copy of the current tracer to avoid concurrent
2108 * changes on it while we are reading.
2110 mutex_lock(&trace_types_lock
);
2111 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2116 *iter
->trace
= *current_trace
;
2118 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2121 if (current_trace
&& current_trace
->print_max
)
2124 iter
->tr
= &global_trace
;
2126 mutex_init(&iter
->mutex
);
2127 iter
->cpu_file
= cpu_file
;
2129 /* Notify the tracer early; before we stop tracing. */
2130 if (iter
->trace
&& iter
->trace
->open
)
2131 iter
->trace
->open(iter
);
2133 /* Annotate start of buffers if we had overruns */
2134 if (ring_buffer_overruns(iter
->tr
->buffer
))
2135 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2137 /* stop the trace while dumping */
2140 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
2141 for_each_tracing_cpu(cpu
) {
2142 iter
->buffer_iter
[cpu
] =
2143 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2145 ring_buffer_read_prepare_sync();
2146 for_each_tracing_cpu(cpu
) {
2147 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2148 tracing_iter_reset(iter
, cpu
);
2151 cpu
= iter
->cpu_file
;
2152 iter
->buffer_iter
[cpu
] =
2153 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2154 ring_buffer_read_prepare_sync();
2155 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2156 tracing_iter_reset(iter
, cpu
);
2159 ret
= seq_open(file
, &tracer_seq_ops
);
2161 fail_ret
= ERR_PTR(ret
);
2165 m
= file
->private_data
;
2168 mutex_unlock(&trace_types_lock
);
2173 for_each_tracing_cpu(cpu
) {
2174 if (iter
->buffer_iter
[cpu
])
2175 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2177 free_cpumask_var(iter
->started
);
2180 mutex_unlock(&trace_types_lock
);
2187 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2189 if (tracing_disabled
)
2192 filp
->private_data
= inode
->i_private
;
2196 static int tracing_release(struct inode
*inode
, struct file
*file
)
2198 struct seq_file
*m
= file
->private_data
;
2199 struct trace_iterator
*iter
;
2202 if (!(file
->f_mode
& FMODE_READ
))
2207 mutex_lock(&trace_types_lock
);
2208 for_each_tracing_cpu(cpu
) {
2209 if (iter
->buffer_iter
[cpu
])
2210 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2213 if (iter
->trace
&& iter
->trace
->close
)
2214 iter
->trace
->close(iter
);
2216 /* reenable tracing if it was previously enabled */
2218 mutex_unlock(&trace_types_lock
);
2220 seq_release(inode
, file
);
2221 mutex_destroy(&iter
->mutex
);
2222 free_cpumask_var(iter
->started
);
2228 static int tracing_open(struct inode
*inode
, struct file
*file
)
2230 struct trace_iterator
*iter
;
2233 /* If this file was open for write, then erase contents */
2234 if ((file
->f_mode
& FMODE_WRITE
) &&
2235 (file
->f_flags
& O_TRUNC
)) {
2236 long cpu
= (long) inode
->i_private
;
2238 if (cpu
== TRACE_PIPE_ALL_CPU
)
2239 tracing_reset_online_cpus(&global_trace
);
2241 tracing_reset(&global_trace
, cpu
);
2244 if (file
->f_mode
& FMODE_READ
) {
2245 iter
= __tracing_open(inode
, file
);
2247 ret
= PTR_ERR(iter
);
2248 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2249 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2255 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2257 struct tracer
*t
= v
;
2267 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2272 mutex_lock(&trace_types_lock
);
2273 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2279 static void t_stop(struct seq_file
*m
, void *p
)
2281 mutex_unlock(&trace_types_lock
);
2284 static int t_show(struct seq_file
*m
, void *v
)
2286 struct tracer
*t
= v
;
2291 seq_printf(m
, "%s", t
->name
);
2300 static const struct seq_operations show_traces_seq_ops
= {
2307 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2309 if (tracing_disabled
)
2312 return seq_open(file
, &show_traces_seq_ops
);
2316 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2317 size_t count
, loff_t
*ppos
)
2322 static const struct file_operations tracing_fops
= {
2323 .open
= tracing_open
,
2325 .write
= tracing_write_stub
,
2326 .llseek
= seq_lseek
,
2327 .release
= tracing_release
,
2330 static const struct file_operations show_traces_fops
= {
2331 .open
= show_traces_open
,
2333 .release
= seq_release
,
2334 .llseek
= seq_lseek
,
2338 * Only trace on a CPU if the bitmask is set:
2340 static cpumask_var_t tracing_cpumask
;
2343 * The tracer itself will not take this lock, but still we want
2344 * to provide a consistent cpumask to user-space:
2346 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2349 * Temporary storage for the character representation of the
2350 * CPU bitmask (and one more byte for the newline):
2352 static char mask_str
[NR_CPUS
+ 1];
2355 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2356 size_t count
, loff_t
*ppos
)
2360 mutex_lock(&tracing_cpumask_update_lock
);
2362 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2363 if (count
- len
< 2) {
2367 len
+= sprintf(mask_str
+ len
, "\n");
2368 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2371 mutex_unlock(&tracing_cpumask_update_lock
);
2377 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2378 size_t count
, loff_t
*ppos
)
2381 cpumask_var_t tracing_cpumask_new
;
2383 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2386 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2390 mutex_lock(&tracing_cpumask_update_lock
);
2392 local_irq_disable();
2393 arch_spin_lock(&ftrace_max_lock
);
2394 for_each_tracing_cpu(cpu
) {
2396 * Increase/decrease the disabled counter if we are
2397 * about to flip a bit in the cpumask:
2399 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2400 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2401 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2403 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2404 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2405 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2408 arch_spin_unlock(&ftrace_max_lock
);
2411 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2413 mutex_unlock(&tracing_cpumask_update_lock
);
2414 free_cpumask_var(tracing_cpumask_new
);
2419 free_cpumask_var(tracing_cpumask_new
);
2424 static const struct file_operations tracing_cpumask_fops
= {
2425 .open
= tracing_open_generic
,
2426 .read
= tracing_cpumask_read
,
2427 .write
= tracing_cpumask_write
,
2428 .llseek
= generic_file_llseek
,
2431 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
2433 struct tracer_opt
*trace_opts
;
2437 mutex_lock(&trace_types_lock
);
2438 tracer_flags
= current_trace
->flags
->val
;
2439 trace_opts
= current_trace
->flags
->opts
;
2441 for (i
= 0; trace_options
[i
]; i
++) {
2442 if (trace_flags
& (1 << i
))
2443 seq_printf(m
, "%s\n", trace_options
[i
]);
2445 seq_printf(m
, "no%s\n", trace_options
[i
]);
2448 for (i
= 0; trace_opts
[i
].name
; i
++) {
2449 if (tracer_flags
& trace_opts
[i
].bit
)
2450 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
2452 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
2454 mutex_unlock(&trace_types_lock
);
2459 static int __set_tracer_option(struct tracer
*trace
,
2460 struct tracer_flags
*tracer_flags
,
2461 struct tracer_opt
*opts
, int neg
)
2465 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
2470 tracer_flags
->val
&= ~opts
->bit
;
2472 tracer_flags
->val
|= opts
->bit
;
2476 /* Try to assign a tracer specific option */
2477 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2479 struct tracer_flags
*tracer_flags
= trace
->flags
;
2480 struct tracer_opt
*opts
= NULL
;
2483 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
2484 opts
= &tracer_flags
->opts
[i
];
2486 if (strcmp(cmp
, opts
->name
) == 0)
2487 return __set_tracer_option(trace
, trace
->flags
,
2494 static void set_tracer_flags(unsigned int mask
, int enabled
)
2496 /* do nothing if flag is already set */
2497 if (!!(trace_flags
& mask
) == !!enabled
)
2501 trace_flags
|= mask
;
2503 trace_flags
&= ~mask
;
2505 if (mask
== TRACE_ITER_RECORD_CMD
)
2506 trace_event_enable_cmd_record(enabled
);
2510 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2511 size_t cnt
, loff_t
*ppos
)
2519 if (cnt
>= sizeof(buf
))
2522 if (copy_from_user(&buf
, ubuf
, cnt
))
2526 cmp
= strstrip(buf
);
2528 if (strncmp(cmp
, "no", 2) == 0) {
2533 for (i
= 0; trace_options
[i
]; i
++) {
2534 if (strcmp(cmp
, trace_options
[i
]) == 0) {
2535 set_tracer_flags(1 << i
, !neg
);
2540 /* If no option could be set, test the specific tracer options */
2541 if (!trace_options
[i
]) {
2542 mutex_lock(&trace_types_lock
);
2543 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2544 mutex_unlock(&trace_types_lock
);
2554 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
2556 if (tracing_disabled
)
2558 return single_open(file
, tracing_trace_options_show
, NULL
);
2561 static const struct file_operations tracing_iter_fops
= {
2562 .open
= tracing_trace_options_open
,
2564 .llseek
= seq_lseek
,
2565 .release
= single_release
,
2566 .write
= tracing_trace_options_write
,
2569 static const char readme_msg
[] =
2570 "tracing mini-HOWTO:\n\n"
2571 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2572 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2573 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2574 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2576 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2577 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2579 "# cat /sys/kernel/debug/tracing/trace_options\n"
2580 "noprint-parent nosym-offset nosym-addr noverbose\n"
2581 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2582 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2583 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2584 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2588 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2589 size_t cnt
, loff_t
*ppos
)
2591 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2592 readme_msg
, strlen(readme_msg
));
2595 static const struct file_operations tracing_readme_fops
= {
2596 .open
= tracing_open_generic
,
2597 .read
= tracing_readme_read
,
2598 .llseek
= generic_file_llseek
,
2602 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
2603 size_t cnt
, loff_t
*ppos
)
2612 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
2616 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
2624 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
2627 pid
= map_cmdline_to_pid
[i
];
2628 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
2631 trace_find_cmdline(pid
, buf_comm
);
2632 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
2637 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
2646 static const struct file_operations tracing_saved_cmdlines_fops
= {
2647 .open
= tracing_open_generic
,
2648 .read
= tracing_saved_cmdlines_read
,
2649 .llseek
= generic_file_llseek
,
2653 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2654 size_t cnt
, loff_t
*ppos
)
2659 r
= sprintf(buf
, "%u\n", tracer_enabled
);
2660 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2664 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2665 size_t cnt
, loff_t
*ppos
)
2667 struct trace_array
*tr
= filp
->private_data
;
2672 if (cnt
>= sizeof(buf
))
2675 if (copy_from_user(&buf
, ubuf
, cnt
))
2680 ret
= strict_strtoul(buf
, 10, &val
);
2686 mutex_lock(&trace_types_lock
);
2687 if (tracer_enabled
^ val
) {
2690 if (current_trace
->start
)
2691 current_trace
->start(tr
);
2696 if (current_trace
->stop
)
2697 current_trace
->stop(tr
);
2700 mutex_unlock(&trace_types_lock
);
2708 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2709 size_t cnt
, loff_t
*ppos
)
2711 char buf
[MAX_TRACER_SIZE
+2];
2714 mutex_lock(&trace_types_lock
);
2716 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2718 r
= sprintf(buf
, "\n");
2719 mutex_unlock(&trace_types_lock
);
2721 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2724 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
2726 tracing_reset_online_cpus(tr
);
2730 static int tracing_resize_ring_buffer(unsigned long size
)
2735 * If kernel or user changes the size of the ring buffer
2736 * we use the size that was given, and we can forget about
2737 * expanding it later.
2739 ring_buffer_expanded
= 1;
2741 ret
= ring_buffer_resize(global_trace
.buffer
, size
);
2745 if (!current_trace
->use_max_tr
)
2748 ret
= ring_buffer_resize(max_tr
.buffer
, size
);
2752 r
= ring_buffer_resize(global_trace
.buffer
,
2753 global_trace
.entries
);
2756 * AARGH! We are left with different
2757 * size max buffer!!!!
2758 * The max buffer is our "snapshot" buffer.
2759 * When a tracer needs a snapshot (one of the
2760 * latency tracers), it swaps the max buffer
2761 * with the saved snap shot. We succeeded to
2762 * update the size of the main buffer, but failed to
2763 * update the size of the max buffer. But when we tried
2764 * to reset the main buffer to the original size, we
2765 * failed there too. This is very unlikely to
2766 * happen, but if it does, warn and kill all
2770 tracing_disabled
= 1;
2775 max_tr
.entries
= size
;
2777 global_trace
.entries
= size
;
2784 * tracing_update_buffers - used by tracing facility to expand ring buffers
2786 * To save on memory when the tracing is never used on a system with it
2787 * configured in. The ring buffers are set to a minimum size. But once
2788 * a user starts to use the tracing facility, then they need to grow
2789 * to their default size.
2791 * This function is to be called when a tracer is about to be used.
2793 int tracing_update_buffers(void)
2797 mutex_lock(&trace_types_lock
);
2798 if (!ring_buffer_expanded
)
2799 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2800 mutex_unlock(&trace_types_lock
);
2805 struct trace_option_dentry
;
2807 static struct trace_option_dentry
*
2808 create_trace_option_files(struct tracer
*tracer
);
2811 destroy_trace_option_files(struct trace_option_dentry
*topts
);
2813 static int tracing_set_tracer(const char *buf
)
2815 static struct trace_option_dentry
*topts
;
2816 struct trace_array
*tr
= &global_trace
;
2820 mutex_lock(&trace_types_lock
);
2822 if (!ring_buffer_expanded
) {
2823 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2829 for (t
= trace_types
; t
; t
= t
->next
) {
2830 if (strcmp(t
->name
, buf
) == 0)
2837 if (t
== current_trace
)
2840 trace_branch_disable();
2841 if (current_trace
&& current_trace
->reset
)
2842 current_trace
->reset(tr
);
2843 if (current_trace
&& current_trace
->use_max_tr
) {
2845 * We don't free the ring buffer. instead, resize it because
2846 * The max_tr ring buffer has some state (e.g. ring->clock) and
2847 * we want preserve it.
2849 ring_buffer_resize(max_tr
.buffer
, 1);
2852 destroy_trace_option_files(topts
);
2856 topts
= create_trace_option_files(current_trace
);
2857 if (current_trace
->use_max_tr
) {
2858 ret
= ring_buffer_resize(max_tr
.buffer
, global_trace
.entries
);
2861 max_tr
.entries
= global_trace
.entries
;
2865 ret
= tracer_init(t
, tr
);
2870 trace_branch_enable(tr
);
2872 mutex_unlock(&trace_types_lock
);
2878 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2879 size_t cnt
, loff_t
*ppos
)
2881 char buf
[MAX_TRACER_SIZE
+1];
2888 if (cnt
> MAX_TRACER_SIZE
)
2889 cnt
= MAX_TRACER_SIZE
;
2891 if (copy_from_user(&buf
, ubuf
, cnt
))
2896 /* strip ending whitespace. */
2897 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2900 err
= tracing_set_tracer(buf
);
2910 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2911 size_t cnt
, loff_t
*ppos
)
2913 unsigned long *ptr
= filp
->private_data
;
2917 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
2918 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2919 if (r
> sizeof(buf
))
2921 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2925 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2926 size_t cnt
, loff_t
*ppos
)
2928 unsigned long *ptr
= filp
->private_data
;
2933 if (cnt
>= sizeof(buf
))
2936 if (copy_from_user(&buf
, ubuf
, cnt
))
2941 ret
= strict_strtoul(buf
, 10, &val
);
2950 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2952 long cpu_file
= (long) inode
->i_private
;
2953 struct trace_iterator
*iter
;
2956 if (tracing_disabled
)
2959 mutex_lock(&trace_types_lock
);
2961 /* create a buffer to store the information to pass to userspace */
2962 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2969 * We make a copy of the current tracer to avoid concurrent
2970 * changes on it while we are reading.
2972 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2978 *iter
->trace
= *current_trace
;
2980 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
2985 /* trace pipe does not show start of buffer */
2986 cpumask_setall(iter
->started
);
2988 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2989 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2991 iter
->cpu_file
= cpu_file
;
2992 iter
->tr
= &global_trace
;
2993 mutex_init(&iter
->mutex
);
2994 filp
->private_data
= iter
;
2996 if (iter
->trace
->pipe_open
)
2997 iter
->trace
->pipe_open(iter
);
2999 nonseekable_open(inode
, filp
);
3001 mutex_unlock(&trace_types_lock
);
3007 mutex_unlock(&trace_types_lock
);
3011 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
3013 struct trace_iterator
*iter
= file
->private_data
;
3015 mutex_lock(&trace_types_lock
);
3017 if (iter
->trace
->pipe_close
)
3018 iter
->trace
->pipe_close(iter
);
3020 mutex_unlock(&trace_types_lock
);
3022 free_cpumask_var(iter
->started
);
3023 mutex_destroy(&iter
->mutex
);
3031 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
3033 struct trace_iterator
*iter
= filp
->private_data
;
3035 if (trace_flags
& TRACE_ITER_BLOCK
) {
3037 * Always select as readable when in blocking mode
3039 return POLLIN
| POLLRDNORM
;
3041 if (!trace_empty(iter
))
3042 return POLLIN
| POLLRDNORM
;
3043 poll_wait(filp
, &trace_wait
, poll_table
);
3044 if (!trace_empty(iter
))
3045 return POLLIN
| POLLRDNORM
;
3052 void default_wait_pipe(struct trace_iterator
*iter
)
3056 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
3058 if (trace_empty(iter
))
3061 finish_wait(&trace_wait
, &wait
);
3065 * This is a make-shift waitqueue.
3066 * A tracer might use this callback on some rare cases:
3068 * 1) the current tracer might hold the runqueue lock when it wakes up
3069 * a reader, hence a deadlock (sched, function, and function graph tracers)
3070 * 2) the function tracers, trace all functions, we don't want
3071 * the overhead of calling wake_up and friends
3072 * (and tracing them too)
3074 * Anyway, this is really very primitive wakeup.
3076 void poll_wait_pipe(struct trace_iterator
*iter
)
3078 set_current_state(TASK_INTERRUPTIBLE
);
3079 /* sleep for 100 msecs, and try again. */
3080 schedule_timeout(HZ
/ 10);
3083 /* Must be called with trace_types_lock mutex held. */
3084 static int tracing_wait_pipe(struct file
*filp
)
3086 struct trace_iterator
*iter
= filp
->private_data
;
3088 while (trace_empty(iter
)) {
3090 if ((filp
->f_flags
& O_NONBLOCK
)) {
3094 mutex_unlock(&iter
->mutex
);
3096 iter
->trace
->wait_pipe(iter
);
3098 mutex_lock(&iter
->mutex
);
3100 if (signal_pending(current
))
3104 * We block until we read something and tracing is disabled.
3105 * We still block if tracing is disabled, but we have never
3106 * read anything. This allows a user to cat this file, and
3107 * then enable tracing. But after we have read something,
3108 * we give an EOF when tracing is again disabled.
3110 * iter->pos will be 0 if we haven't read anything.
3112 if (!tracer_enabled
&& iter
->pos
)
3123 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
3124 size_t cnt
, loff_t
*ppos
)
3126 struct trace_iterator
*iter
= filp
->private_data
;
3127 static struct tracer
*old_tracer
;
3130 /* return any leftover data */
3131 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3135 trace_seq_init(&iter
->seq
);
3137 /* copy the tracer to avoid using a global lock all around */
3138 mutex_lock(&trace_types_lock
);
3139 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3140 old_tracer
= current_trace
;
3141 *iter
->trace
= *current_trace
;
3143 mutex_unlock(&trace_types_lock
);
3146 * Avoid more than one consumer on a single file descriptor
3147 * This is just a matter of traces coherency, the ring buffer itself
3150 mutex_lock(&iter
->mutex
);
3151 if (iter
->trace
->read
) {
3152 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
3158 sret
= tracing_wait_pipe(filp
);
3162 /* stop when tracing is finished */
3163 if (trace_empty(iter
)) {
3168 if (cnt
>= PAGE_SIZE
)
3169 cnt
= PAGE_SIZE
- 1;
3171 /* reset all but tr, trace, and overruns */
3172 memset(&iter
->seq
, 0,
3173 sizeof(struct trace_iterator
) -
3174 offsetof(struct trace_iterator
, seq
));
3177 trace_event_read_lock();
3178 trace_access_lock(iter
->cpu_file
);
3179 while (trace_find_next_entry_inc(iter
) != NULL
) {
3180 enum print_line_t ret
;
3181 int len
= iter
->seq
.len
;
3183 ret
= print_trace_line(iter
);
3184 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3185 /* don't print partial lines */
3186 iter
->seq
.len
= len
;
3189 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3190 trace_consume(iter
);
3192 if (iter
->seq
.len
>= cnt
)
3195 trace_access_unlock(iter
->cpu_file
);
3196 trace_event_read_unlock();
3198 /* Now copy what we have to the user */
3199 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3200 if (iter
->seq
.readpos
>= iter
->seq
.len
)
3201 trace_seq_init(&iter
->seq
);
3204 * If there was nothing to send to user, inspite of consuming trace
3205 * entries, go back to wait for more entries.
3211 mutex_unlock(&iter
->mutex
);
3216 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
3217 struct pipe_buffer
*buf
)
3219 __free_page(buf
->page
);
3222 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
3225 __free_page(spd
->pages
[idx
]);
3228 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
3230 .map
= generic_pipe_buf_map
,
3231 .unmap
= generic_pipe_buf_unmap
,
3232 .confirm
= generic_pipe_buf_confirm
,
3233 .release
= tracing_pipe_buf_release
,
3234 .steal
= generic_pipe_buf_steal
,
3235 .get
= generic_pipe_buf_get
,
3239 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
3244 /* Seq buffer is page-sized, exactly what we need. */
3246 count
= iter
->seq
.len
;
3247 ret
= print_trace_line(iter
);
3248 count
= iter
->seq
.len
- count
;
3251 iter
->seq
.len
-= count
;
3254 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3255 iter
->seq
.len
-= count
;
3259 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3260 trace_consume(iter
);
3262 if (!trace_find_next_entry_inc(iter
)) {
3272 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3274 struct pipe_inode_info
*pipe
,
3278 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3279 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3280 struct trace_iterator
*iter
= filp
->private_data
;
3281 struct splice_pipe_desc spd
= {
3283 .partial
= partial_def
,
3284 .nr_pages
= 0, /* This gets updated below. */
3286 .ops
= &tracing_pipe_buf_ops
,
3287 .spd_release
= tracing_spd_release_pipe
,
3289 static struct tracer
*old_tracer
;
3294 if (splice_grow_spd(pipe
, &spd
))
3297 /* copy the tracer to avoid using a global lock all around */
3298 mutex_lock(&trace_types_lock
);
3299 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3300 old_tracer
= current_trace
;
3301 *iter
->trace
= *current_trace
;
3303 mutex_unlock(&trace_types_lock
);
3305 mutex_lock(&iter
->mutex
);
3307 if (iter
->trace
->splice_read
) {
3308 ret
= iter
->trace
->splice_read(iter
, filp
,
3309 ppos
, pipe
, len
, flags
);
3314 ret
= tracing_wait_pipe(filp
);
3318 if (!iter
->ent
&& !trace_find_next_entry_inc(iter
)) {
3323 trace_event_read_lock();
3324 trace_access_lock(iter
->cpu_file
);
3326 /* Fill as many pages as possible. */
3327 for (i
= 0, rem
= len
; i
< pipe
->buffers
&& rem
; i
++) {
3328 spd
.pages
[i
] = alloc_page(GFP_KERNEL
);
3332 rem
= tracing_fill_pipe_page(rem
, iter
);
3334 /* Copy the data into the page, so we can start over. */
3335 ret
= trace_seq_to_buffer(&iter
->seq
,
3336 page_address(spd
.pages
[i
]),
3339 __free_page(spd
.pages
[i
]);
3342 spd
.partial
[i
].offset
= 0;
3343 spd
.partial
[i
].len
= iter
->seq
.len
;
3345 trace_seq_init(&iter
->seq
);
3348 trace_access_unlock(iter
->cpu_file
);
3349 trace_event_read_unlock();
3350 mutex_unlock(&iter
->mutex
);
3354 ret
= splice_to_pipe(pipe
, &spd
);
3356 splice_shrink_spd(pipe
, &spd
);
3360 mutex_unlock(&iter
->mutex
);
3365 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3366 size_t cnt
, loff_t
*ppos
)
3368 struct trace_array
*tr
= filp
->private_data
;
3372 mutex_lock(&trace_types_lock
);
3373 if (!ring_buffer_expanded
)
3374 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3376 trace_buf_size
>> 10);
3378 r
= sprintf(buf
, "%lu\n", tr
->entries
>> 10);
3379 mutex_unlock(&trace_types_lock
);
3381 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3385 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3386 size_t cnt
, loff_t
*ppos
)
3392 if (cnt
>= sizeof(buf
))
3395 if (copy_from_user(&buf
, ubuf
, cnt
))
3400 ret
= strict_strtoul(buf
, 10, &val
);
3404 /* must have at least 1 entry */
3408 mutex_lock(&trace_types_lock
);
3412 /* disable all cpu buffers */
3413 for_each_tracing_cpu(cpu
) {
3414 if (global_trace
.data
[cpu
])
3415 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
3416 if (max_tr
.data
[cpu
])
3417 atomic_inc(&max_tr
.data
[cpu
]->disabled
);
3420 /* value is in KB */
3423 if (val
!= global_trace
.entries
) {
3424 ret
= tracing_resize_ring_buffer(val
);
3433 /* If check pages failed, return ENOMEM */
3434 if (tracing_disabled
)
3437 for_each_tracing_cpu(cpu
) {
3438 if (global_trace
.data
[cpu
])
3439 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
3440 if (max_tr
.data
[cpu
])
3441 atomic_dec(&max_tr
.data
[cpu
]->disabled
);
3445 mutex_unlock(&trace_types_lock
);
3450 static int mark_printk(const char *fmt
, ...)
3454 va_start(args
, fmt
);
3455 ret
= trace_vprintk(0, fmt
, args
);
3461 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3462 size_t cnt
, loff_t
*fpos
)
3467 if (tracing_disabled
)
3470 if (cnt
> TRACE_BUF_SIZE
)
3471 cnt
= TRACE_BUF_SIZE
;
3473 buf
= kmalloc(cnt
+ 2, GFP_KERNEL
);
3477 if (copy_from_user(buf
, ubuf
, cnt
)) {
3481 if (buf
[cnt
-1] != '\n') {
3487 written
= mark_printk("%s", buf
);
3491 /* don't tell userspace we wrote more - it might confuse them */
3498 static int tracing_clock_show(struct seq_file
*m
, void *v
)
3502 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
3504 "%s%s%s%s", i
? " " : "",
3505 i
== trace_clock_id
? "[" : "", trace_clocks
[i
].name
,
3506 i
== trace_clock_id
? "]" : "");
3512 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
3513 size_t cnt
, loff_t
*fpos
)
3516 const char *clockstr
;
3519 if (cnt
>= sizeof(buf
))
3522 if (copy_from_user(&buf
, ubuf
, cnt
))
3527 clockstr
= strstrip(buf
);
3529 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
3530 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
3533 if (i
== ARRAY_SIZE(trace_clocks
))
3538 mutex_lock(&trace_types_lock
);
3540 ring_buffer_set_clock(global_trace
.buffer
, trace_clocks
[i
].func
);
3542 ring_buffer_set_clock(max_tr
.buffer
, trace_clocks
[i
].func
);
3544 mutex_unlock(&trace_types_lock
);
3551 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
3553 if (tracing_disabled
)
3555 return single_open(file
, tracing_clock_show
, NULL
);
3558 static const struct file_operations tracing_max_lat_fops
= {
3559 .open
= tracing_open_generic
,
3560 .read
= tracing_max_lat_read
,
3561 .write
= tracing_max_lat_write
,
3562 .llseek
= generic_file_llseek
,
3565 static const struct file_operations tracing_ctrl_fops
= {
3566 .open
= tracing_open_generic
,
3567 .read
= tracing_ctrl_read
,
3568 .write
= tracing_ctrl_write
,
3569 .llseek
= generic_file_llseek
,
3572 static const struct file_operations set_tracer_fops
= {
3573 .open
= tracing_open_generic
,
3574 .read
= tracing_set_trace_read
,
3575 .write
= tracing_set_trace_write
,
3576 .llseek
= generic_file_llseek
,
3579 static const struct file_operations tracing_pipe_fops
= {
3580 .open
= tracing_open_pipe
,
3581 .poll
= tracing_poll_pipe
,
3582 .read
= tracing_read_pipe
,
3583 .splice_read
= tracing_splice_read_pipe
,
3584 .release
= tracing_release_pipe
,
3585 .llseek
= no_llseek
,
3588 static const struct file_operations tracing_entries_fops
= {
3589 .open
= tracing_open_generic
,
3590 .read
= tracing_entries_read
,
3591 .write
= tracing_entries_write
,
3592 .llseek
= generic_file_llseek
,
3595 static const struct file_operations tracing_mark_fops
= {
3596 .open
= tracing_open_generic
,
3597 .write
= tracing_mark_write
,
3598 .llseek
= generic_file_llseek
,
3601 static const struct file_operations trace_clock_fops
= {
3602 .open
= tracing_clock_open
,
3604 .llseek
= seq_lseek
,
3605 .release
= single_release
,
3606 .write
= tracing_clock_write
,
3609 struct ftrace_buffer_info
{
3610 struct trace_array
*tr
;
3616 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
3618 int cpu
= (int)(long)inode
->i_private
;
3619 struct ftrace_buffer_info
*info
;
3621 if (tracing_disabled
)
3624 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3628 info
->tr
= &global_trace
;
3631 /* Force reading ring buffer for first read */
3632 info
->read
= (unsigned int)-1;
3634 filp
->private_data
= info
;
3636 return nonseekable_open(inode
, filp
);
3640 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
3641 size_t count
, loff_t
*ppos
)
3643 struct ftrace_buffer_info
*info
= filp
->private_data
;
3651 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
);
3655 /* Do we have previous read data to read? */
3656 if (info
->read
< PAGE_SIZE
)
3661 trace_access_lock(info
->cpu
);
3662 ret
= ring_buffer_read_page(info
->tr
->buffer
,
3666 trace_access_unlock(info
->cpu
);
3671 size
= PAGE_SIZE
- info
->read
;
3675 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
3686 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
3688 struct ftrace_buffer_info
*info
= file
->private_data
;
3691 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
3698 struct ring_buffer
*buffer
;
3703 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
3704 struct pipe_buffer
*buf
)
3706 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3711 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3716 static int buffer_pipe_buf_steal(struct pipe_inode_info
*pipe
,
3717 struct pipe_buffer
*buf
)
3722 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
3723 struct pipe_buffer
*buf
)
3725 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3730 /* Pipe buffer operations for a buffer. */
3731 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
3733 .map
= generic_pipe_buf_map
,
3734 .unmap
= generic_pipe_buf_unmap
,
3735 .confirm
= generic_pipe_buf_confirm
,
3736 .release
= buffer_pipe_buf_release
,
3737 .steal
= buffer_pipe_buf_steal
,
3738 .get
= buffer_pipe_buf_get
,
3742 * Callback from splice_to_pipe(), if we need to release some pages
3743 * at the end of the spd in case we error'ed out in filling the pipe.
3745 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
3747 struct buffer_ref
*ref
=
3748 (struct buffer_ref
*)spd
->partial
[i
].private;
3753 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3755 spd
->partial
[i
].private = 0;
3759 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
3760 struct pipe_inode_info
*pipe
, size_t len
,
3763 struct ftrace_buffer_info
*info
= file
->private_data
;
3764 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3765 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3766 struct splice_pipe_desc spd
= {
3768 .partial
= partial_def
,
3770 .ops
= &buffer_pipe_buf_ops
,
3771 .spd_release
= buffer_spd_release
,
3773 struct buffer_ref
*ref
;
3774 int entries
, size
, i
;
3777 if (splice_grow_spd(pipe
, &spd
))
3780 if (*ppos
& (PAGE_SIZE
- 1)) {
3781 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3786 if (len
& (PAGE_SIZE
- 1)) {
3787 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3788 if (len
< PAGE_SIZE
) {
3795 trace_access_lock(info
->cpu
);
3796 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
3798 for (i
= 0; i
< pipe
->buffers
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
3802 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
3807 ref
->buffer
= info
->tr
->buffer
;
3808 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
);
3814 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
3817 ring_buffer_free_read_page(ref
->buffer
,
3824 * zero out any left over data, this is going to
3827 size
= ring_buffer_page_len(ref
->page
);
3828 if (size
< PAGE_SIZE
)
3829 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
3831 page
= virt_to_page(ref
->page
);
3833 spd
.pages
[i
] = page
;
3834 spd
.partial
[i
].len
= PAGE_SIZE
;
3835 spd
.partial
[i
].offset
= 0;
3836 spd
.partial
[i
].private = (unsigned long)ref
;
3840 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
3843 trace_access_unlock(info
->cpu
);
3846 /* did we read anything? */
3847 if (!spd
.nr_pages
) {
3848 if (flags
& SPLICE_F_NONBLOCK
)
3856 ret
= splice_to_pipe(pipe
, &spd
);
3857 splice_shrink_spd(pipe
, &spd
);
3862 static const struct file_operations tracing_buffers_fops
= {
3863 .open
= tracing_buffers_open
,
3864 .read
= tracing_buffers_read
,
3865 .release
= tracing_buffers_release
,
3866 .splice_read
= tracing_buffers_splice_read
,
3867 .llseek
= no_llseek
,
3871 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
3872 size_t count
, loff_t
*ppos
)
3874 unsigned long cpu
= (unsigned long)filp
->private_data
;
3875 struct trace_array
*tr
= &global_trace
;
3876 struct trace_seq
*s
;
3879 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
3885 cnt
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
3886 trace_seq_printf(s
, "entries: %ld\n", cnt
);
3888 cnt
= ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
3889 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
3891 cnt
= ring_buffer_commit_overrun_cpu(tr
->buffer
, cpu
);
3892 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
3894 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
3901 static const struct file_operations tracing_stats_fops
= {
3902 .open
= tracing_open_generic
,
3903 .read
= tracing_stats_read
,
3904 .llseek
= generic_file_llseek
,
3907 #ifdef CONFIG_DYNAMIC_FTRACE
3909 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
3915 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
3916 size_t cnt
, loff_t
*ppos
)
3918 static char ftrace_dyn_info_buffer
[1024];
3919 static DEFINE_MUTEX(dyn_info_mutex
);
3920 unsigned long *p
= filp
->private_data
;
3921 char *buf
= ftrace_dyn_info_buffer
;
3922 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
3925 mutex_lock(&dyn_info_mutex
);
3926 r
= sprintf(buf
, "%ld ", *p
);
3928 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
3931 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3933 mutex_unlock(&dyn_info_mutex
);
3938 static const struct file_operations tracing_dyn_info_fops
= {
3939 .open
= tracing_open_generic
,
3940 .read
= tracing_read_dyn_info
,
3941 .llseek
= generic_file_llseek
,
3945 static struct dentry
*d_tracer
;
3947 struct dentry
*tracing_init_dentry(void)
3954 if (!debugfs_initialized())
3957 d_tracer
= debugfs_create_dir("tracing", NULL
);
3959 if (!d_tracer
&& !once
) {
3961 pr_warning("Could not create debugfs directory 'tracing'\n");
3968 static struct dentry
*d_percpu
;
3970 struct dentry
*tracing_dentry_percpu(void)
3973 struct dentry
*d_tracer
;
3978 d_tracer
= tracing_init_dentry();
3983 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
3985 if (!d_percpu
&& !once
) {
3987 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3994 static void tracing_init_debugfs_percpu(long cpu
)
3996 struct dentry
*d_percpu
= tracing_dentry_percpu();
3997 struct dentry
*d_cpu
;
3998 char cpu_dir
[30]; /* 30 characters should be more than enough */
4000 snprintf(cpu_dir
, 30, "cpu%ld", cpu
);
4001 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
4003 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
4007 /* per cpu trace_pipe */
4008 trace_create_file("trace_pipe", 0444, d_cpu
,
4009 (void *) cpu
, &tracing_pipe_fops
);
4012 trace_create_file("trace", 0644, d_cpu
,
4013 (void *) cpu
, &tracing_fops
);
4015 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
4016 (void *) cpu
, &tracing_buffers_fops
);
4018 trace_create_file("stats", 0444, d_cpu
,
4019 (void *) cpu
, &tracing_stats_fops
);
4022 #ifdef CONFIG_FTRACE_SELFTEST
4023 /* Let selftest have access to static functions in this file */
4024 #include "trace_selftest.c"
4027 struct trace_option_dentry
{
4028 struct tracer_opt
*opt
;
4029 struct tracer_flags
*flags
;
4030 struct dentry
*entry
;
4034 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4037 struct trace_option_dentry
*topt
= filp
->private_data
;
4040 if (topt
->flags
->val
& topt
->opt
->bit
)
4045 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4049 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4052 struct trace_option_dentry
*topt
= filp
->private_data
;
4057 if (cnt
>= sizeof(buf
))
4060 if (copy_from_user(&buf
, ubuf
, cnt
))
4065 ret
= strict_strtoul(buf
, 10, &val
);
4069 if (val
!= 0 && val
!= 1)
4072 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
4073 mutex_lock(&trace_types_lock
);
4074 ret
= __set_tracer_option(current_trace
, topt
->flags
,
4076 mutex_unlock(&trace_types_lock
);
4087 static const struct file_operations trace_options_fops
= {
4088 .open
= tracing_open_generic
,
4089 .read
= trace_options_read
,
4090 .write
= trace_options_write
,
4091 .llseek
= generic_file_llseek
,
4095 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4098 long index
= (long)filp
->private_data
;
4101 if (trace_flags
& (1 << index
))
4106 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4110 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4113 long index
= (long)filp
->private_data
;
4118 if (cnt
>= sizeof(buf
))
4121 if (copy_from_user(&buf
, ubuf
, cnt
))
4126 ret
= strict_strtoul(buf
, 10, &val
);
4130 if (val
!= 0 && val
!= 1)
4132 set_tracer_flags(1 << index
, val
);
4139 static const struct file_operations trace_options_core_fops
= {
4140 .open
= tracing_open_generic
,
4141 .read
= trace_options_core_read
,
4142 .write
= trace_options_core_write
,
4143 .llseek
= generic_file_llseek
,
4146 struct dentry
*trace_create_file(const char *name
,
4148 struct dentry
*parent
,
4150 const struct file_operations
*fops
)
4154 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
4156 pr_warning("Could not create debugfs '%s' entry\n", name
);
4162 static struct dentry
*trace_options_init_dentry(void)
4164 struct dentry
*d_tracer
;
4165 static struct dentry
*t_options
;
4170 d_tracer
= tracing_init_dentry();
4174 t_options
= debugfs_create_dir("options", d_tracer
);
4176 pr_warning("Could not create debugfs directory 'options'\n");
4184 create_trace_option_file(struct trace_option_dentry
*topt
,
4185 struct tracer_flags
*flags
,
4186 struct tracer_opt
*opt
)
4188 struct dentry
*t_options
;
4190 t_options
= trace_options_init_dentry();
4194 topt
->flags
= flags
;
4197 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
4198 &trace_options_fops
);
4202 static struct trace_option_dentry
*
4203 create_trace_option_files(struct tracer
*tracer
)
4205 struct trace_option_dentry
*topts
;
4206 struct tracer_flags
*flags
;
4207 struct tracer_opt
*opts
;
4213 flags
= tracer
->flags
;
4215 if (!flags
|| !flags
->opts
)
4220 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4223 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
4227 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4228 create_trace_option_file(&topts
[cnt
], flags
,
4235 destroy_trace_option_files(struct trace_option_dentry
*topts
)
4242 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
4243 if (topts
[cnt
].entry
)
4244 debugfs_remove(topts
[cnt
].entry
);
4250 static struct dentry
*
4251 create_trace_option_core_file(const char *option
, long index
)
4253 struct dentry
*t_options
;
4255 t_options
= trace_options_init_dentry();
4259 return trace_create_file(option
, 0644, t_options
, (void *)index
,
4260 &trace_options_core_fops
);
4263 static __init
void create_trace_options_dir(void)
4265 struct dentry
*t_options
;
4268 t_options
= trace_options_init_dentry();
4272 for (i
= 0; trace_options
[i
]; i
++)
4273 create_trace_option_core_file(trace_options
[i
], i
);
4276 static __init
int tracer_init_debugfs(void)
4278 struct dentry
*d_tracer
;
4281 trace_access_lock_init();
4283 d_tracer
= tracing_init_dentry();
4285 trace_create_file("tracing_enabled", 0644, d_tracer
,
4286 &global_trace
, &tracing_ctrl_fops
);
4288 trace_create_file("trace_options", 0644, d_tracer
,
4289 NULL
, &tracing_iter_fops
);
4291 trace_create_file("tracing_cpumask", 0644, d_tracer
,
4292 NULL
, &tracing_cpumask_fops
);
4294 trace_create_file("trace", 0644, d_tracer
,
4295 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
4297 trace_create_file("available_tracers", 0444, d_tracer
,
4298 &global_trace
, &show_traces_fops
);
4300 trace_create_file("current_tracer", 0644, d_tracer
,
4301 &global_trace
, &set_tracer_fops
);
4303 #ifdef CONFIG_TRACER_MAX_TRACE
4304 trace_create_file("tracing_max_latency", 0644, d_tracer
,
4305 &tracing_max_latency
, &tracing_max_lat_fops
);
4308 trace_create_file("tracing_thresh", 0644, d_tracer
,
4309 &tracing_thresh
, &tracing_max_lat_fops
);
4311 trace_create_file("README", 0444, d_tracer
,
4312 NULL
, &tracing_readme_fops
);
4314 trace_create_file("trace_pipe", 0444, d_tracer
,
4315 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
4317 trace_create_file("buffer_size_kb", 0644, d_tracer
,
4318 &global_trace
, &tracing_entries_fops
);
4320 trace_create_file("trace_marker", 0220, d_tracer
,
4321 NULL
, &tracing_mark_fops
);
4323 trace_create_file("saved_cmdlines", 0444, d_tracer
,
4324 NULL
, &tracing_saved_cmdlines_fops
);
4326 trace_create_file("trace_clock", 0644, d_tracer
, NULL
,
4329 #ifdef CONFIG_DYNAMIC_FTRACE
4330 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
4331 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
4334 create_trace_options_dir();
4336 for_each_tracing_cpu(cpu
)
4337 tracing_init_debugfs_percpu(cpu
);
4342 static int trace_panic_handler(struct notifier_block
*this,
4343 unsigned long event
, void *unused
)
4345 if (ftrace_dump_on_oops
)
4346 ftrace_dump(ftrace_dump_on_oops
);
4350 static struct notifier_block trace_panic_notifier
= {
4351 .notifier_call
= trace_panic_handler
,
4353 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
4356 static int trace_die_handler(struct notifier_block
*self
,
4362 if (ftrace_dump_on_oops
)
4363 ftrace_dump(ftrace_dump_on_oops
);
4371 static struct notifier_block trace_die_notifier
= {
4372 .notifier_call
= trace_die_handler
,
4377 * printk is set to max of 1024, we really don't need it that big.
4378 * Nothing should be printing 1000 characters anyway.
4380 #define TRACE_MAX_PRINT 1000
4383 * Define here KERN_TRACE so that we have one place to modify
4384 * it if we decide to change what log level the ftrace dump
4387 #define KERN_TRACE KERN_EMERG
4390 trace_printk_seq(struct trace_seq
*s
)
4392 /* Probably should print a warning here. */
4396 /* should be zero ended, but we are paranoid. */
4397 s
->buffer
[s
->len
] = 0;
4399 printk(KERN_TRACE
"%s", s
->buffer
);
4404 void trace_init_global_iter(struct trace_iterator
*iter
)
4406 iter
->tr
= &global_trace
;
4407 iter
->trace
= current_trace
;
4408 iter
->cpu_file
= TRACE_PIPE_ALL_CPU
;
4412 __ftrace_dump(bool disable_tracing
, enum ftrace_dump_mode oops_dump_mode
)
4414 static arch_spinlock_t ftrace_dump_lock
=
4415 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
4416 /* use static because iter can be a bit big for the stack */
4417 static struct trace_iterator iter
;
4418 unsigned int old_userobj
;
4419 static int dump_ran
;
4420 unsigned long flags
;
4424 local_irq_save(flags
);
4425 arch_spin_lock(&ftrace_dump_lock
);
4433 if (disable_tracing
)
4436 trace_init_global_iter(&iter
);
4438 for_each_tracing_cpu(cpu
) {
4439 atomic_inc(&iter
.tr
->data
[cpu
]->disabled
);
4442 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
4444 /* don't look at user memory in panic mode */
4445 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
4447 /* Simulate the iterator */
4448 iter
.tr
= &global_trace
;
4449 iter
.trace
= current_trace
;
4451 switch (oops_dump_mode
) {
4453 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4456 iter
.cpu_file
= raw_smp_processor_id();
4461 printk(KERN_TRACE
"Bad dumping mode, switching to all CPUs dump\n");
4462 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4465 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
4468 * We need to stop all tracing on all CPUS to read the
4469 * the next buffer. This is a bit expensive, but is
4470 * not done often. We fill all what we can read,
4471 * and then release the locks again.
4474 while (!trace_empty(&iter
)) {
4477 printk(KERN_TRACE
"---------------------------------\n");
4481 /* reset all but tr, trace, and overruns */
4482 memset(&iter
.seq
, 0,
4483 sizeof(struct trace_iterator
) -
4484 offsetof(struct trace_iterator
, seq
));
4485 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
4488 if (trace_find_next_entry_inc(&iter
) != NULL
) {
4491 ret
= print_trace_line(&iter
);
4492 if (ret
!= TRACE_TYPE_NO_CONSUME
)
4493 trace_consume(&iter
);
4496 trace_printk_seq(&iter
.seq
);
4500 printk(KERN_TRACE
" (ftrace buffer empty)\n");
4502 printk(KERN_TRACE
"---------------------------------\n");
4505 /* Re-enable tracing if requested */
4506 if (!disable_tracing
) {
4507 trace_flags
|= old_userobj
;
4509 for_each_tracing_cpu(cpu
) {
4510 atomic_dec(&iter
.tr
->data
[cpu
]->disabled
);
4516 arch_spin_unlock(&ftrace_dump_lock
);
4517 local_irq_restore(flags
);
4520 /* By default: disable tracing after the dump */
4521 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode
)
4523 __ftrace_dump(true, oops_dump_mode
);
4526 __init
static int tracer_alloc_buffers(void)
4532 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
4535 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
4536 goto out_free_buffer_mask
;
4538 /* To save memory, keep the ring buffer size to its minimum */
4539 if (ring_buffer_expanded
)
4540 ring_buf_size
= trace_buf_size
;
4544 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
4545 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
4547 /* TODO: make the number of buffers hot pluggable with CPUS */
4548 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
,
4549 TRACE_BUFFER_FLAGS
);
4550 if (!global_trace
.buffer
) {
4551 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
4553 goto out_free_cpumask
;
4555 global_trace
.entries
= ring_buffer_size(global_trace
.buffer
);
4558 #ifdef CONFIG_TRACER_MAX_TRACE
4559 max_tr
.buffer
= ring_buffer_alloc(1, TRACE_BUFFER_FLAGS
);
4560 if (!max_tr
.buffer
) {
4561 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
4563 ring_buffer_free(global_trace
.buffer
);
4564 goto out_free_cpumask
;
4569 /* Allocate the first page for all buffers */
4570 for_each_tracing_cpu(i
) {
4571 global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
4572 max_tr
.data
[i
] = &per_cpu(max_tr_data
, i
);
4575 trace_init_cmdlines();
4577 register_tracer(&nop_trace
);
4578 current_trace
= &nop_trace
;
4579 /* All seems OK, enable tracing */
4580 tracing_disabled
= 0;
4582 atomic_notifier_chain_register(&panic_notifier_list
,
4583 &trace_panic_notifier
);
4585 register_die_notifier(&trace_die_notifier
);
4590 free_cpumask_var(tracing_cpumask
);
4591 out_free_buffer_mask
:
4592 free_cpumask_var(tracing_buffer_mask
);
4597 __init
static int clear_boot_tracer(void)
4600 * The default tracer at boot buffer is an init section.
4601 * This function is called in lateinit. If we did not
4602 * find the boot tracer, then clear it out, to prevent
4603 * later registration from accessing the buffer that is
4604 * about to be freed.
4606 if (!default_bootup_tracer
)
4609 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
4610 default_bootup_tracer
);
4611 default_bootup_tracer
= NULL
;
4616 early_initcall(tracer_alloc_buffers
);
4617 fs_initcall(tracer_init_debugfs
);
4618 late_initcall(clear_boot_tracer
);