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"
45 * On boot up, the ring buffer is set to the minimum size, so that
46 * we do not waste memory on systems that are not using tracing.
48 int ring_buffer_expanded
;
51 * We need to change this state when a selftest is running.
52 * A selftest will lurk into the ring-buffer to count the
53 * entries inserted during the selftest although some concurrent
54 * insertions into the ring-buffer such as trace_printk could occurred
55 * at the same time, giving false positive or negative results.
57 static bool __read_mostly tracing_selftest_running
;
60 * If a tracer is running, we do not want to run SELFTEST.
62 bool __read_mostly tracing_selftest_disabled
;
64 /* For tracers that don't implement custom flags */
65 static struct tracer_opt dummy_tracer_opt
[] = {
69 static struct tracer_flags dummy_tracer_flags
= {
71 .opts
= dummy_tracer_opt
74 static int dummy_set_flag(u32 old_flags
, u32 bit
, int set
)
80 * Kill all tracing for good (never come back).
81 * It is initialized to 1 but will turn to zero if the initialization
82 * of the tracer is successful. But that is the only place that sets
85 static int tracing_disabled
= 1;
87 DEFINE_PER_CPU(int, ftrace_cpu_disabled
);
89 static inline void ftrace_disable_cpu(void)
92 __this_cpu_inc(ftrace_cpu_disabled
);
95 static inline void ftrace_enable_cpu(void)
97 __this_cpu_dec(ftrace_cpu_disabled
);
101 cpumask_var_t __read_mostly tracing_buffer_mask
;
104 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
106 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
107 * is set, then ftrace_dump is called. This will output the contents
108 * of the ftrace buffers to the console. This is very useful for
109 * capturing traces that lead to crashes and outputing it to a
112 * It is default off, but you can enable it with either specifying
113 * "ftrace_dump_on_oops" in the kernel command line, or setting
114 * /proc/sys/kernel/ftrace_dump_on_oops
115 * Set 1 if you want to dump buffers of all CPUs
116 * Set 2 if you want to dump the buffer of the CPU that triggered oops
119 enum ftrace_dump_mode ftrace_dump_on_oops
;
121 static int tracing_set_tracer(const char *buf
);
123 #define MAX_TRACER_SIZE 100
124 static char bootup_tracer_buf
[MAX_TRACER_SIZE
] __initdata
;
125 static char *default_bootup_tracer
;
127 static int __init
set_cmdline_ftrace(char *str
)
129 strncpy(bootup_tracer_buf
, str
, MAX_TRACER_SIZE
);
130 default_bootup_tracer
= bootup_tracer_buf
;
131 /* We are using ftrace early, expand it */
132 ring_buffer_expanded
= 1;
135 __setup("ftrace=", set_cmdline_ftrace
);
137 static int __init
set_ftrace_dump_on_oops(char *str
)
139 if (*str
++ != '=' || !*str
) {
140 ftrace_dump_on_oops
= DUMP_ALL
;
144 if (!strcmp("orig_cpu", str
)) {
145 ftrace_dump_on_oops
= DUMP_ORIG
;
151 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops
);
153 unsigned long long ns2usecs(cycle_t nsec
)
161 * The global_trace is the descriptor that holds the tracing
162 * buffers for the live tracing. For each CPU, it contains
163 * a link list of pages that will store trace entries. The
164 * page descriptor of the pages in the memory is used to hold
165 * the link list by linking the lru item in the page descriptor
166 * to each of the pages in the buffer per CPU.
168 * For each active CPU there is a data field that holds the
169 * pages for the buffer for that CPU. Each CPU has the same number
170 * of pages allocated for its buffer.
172 static struct trace_array global_trace
;
174 static DEFINE_PER_CPU(struct trace_array_cpu
, global_trace_cpu
);
176 int filter_current_check_discard(struct ring_buffer
*buffer
,
177 struct ftrace_event_call
*call
, void *rec
,
178 struct ring_buffer_event
*event
)
180 return filter_check_discard(call
, rec
, buffer
, event
);
182 EXPORT_SYMBOL_GPL(filter_current_check_discard
);
184 cycle_t
ftrace_now(int cpu
)
188 /* Early boot up does not have a buffer yet */
189 if (!global_trace
.buffer
)
190 return trace_clock_local();
192 ts
= ring_buffer_time_stamp(global_trace
.buffer
, cpu
);
193 ring_buffer_normalize_time_stamp(global_trace
.buffer
, cpu
, &ts
);
199 * The max_tr is used to snapshot the global_trace when a maximum
200 * latency is reached. Some tracers will use this to store a maximum
201 * trace while it continues examining live traces.
203 * The buffers for the max_tr are set up the same as the global_trace.
204 * When a snapshot is taken, the link list of the max_tr is swapped
205 * with the link list of the global_trace and the buffers are reset for
206 * the global_trace so the tracing can continue.
208 static struct trace_array max_tr
;
210 static DEFINE_PER_CPU(struct trace_array_cpu
, max_tr_data
);
212 /* tracer_enabled is used to toggle activation of a tracer */
213 static int tracer_enabled
= 1;
216 * tracing_is_enabled - return tracer_enabled status
218 * This function is used by other tracers to know the status
219 * of the tracer_enabled flag. Tracers may use this function
220 * to know if it should enable their features when starting
221 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
223 int tracing_is_enabled(void)
225 return tracer_enabled
;
229 * trace_buf_size is the size in bytes that is allocated
230 * for a buffer. Note, the number of bytes is always rounded
233 * This number is purposely set to a low number of 16384.
234 * If the dump on oops happens, it will be much appreciated
235 * to not have to wait for all that output. Anyway this can be
236 * boot time and run time configurable.
238 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
240 static unsigned long trace_buf_size
= TRACE_BUF_SIZE_DEFAULT
;
242 /* trace_types holds a link list of available tracers. */
243 static struct tracer
*trace_types __read_mostly
;
245 /* current_trace points to the tracer that is currently active */
246 static struct tracer
*current_trace __read_mostly
;
249 * trace_types_lock is used to protect the trace_types list.
251 static DEFINE_MUTEX(trace_types_lock
);
254 * serialize the access of the ring buffer
256 * ring buffer serializes readers, but it is low level protection.
257 * The validity of the events (which returns by ring_buffer_peek() ..etc)
258 * are not protected by ring buffer.
260 * The content of events may become garbage if we allow other process consumes
261 * these events concurrently:
262 * A) the page of the consumed events may become a normal page
263 * (not reader page) in ring buffer, and this page will be rewrited
264 * by events producer.
265 * B) The page of the consumed events may become a page for splice_read,
266 * and this page will be returned to system.
268 * These primitives allow multi process access to different cpu ring buffer
271 * These primitives don't distinguish read-only and read-consume access.
272 * Multi read-only access are also serialized.
276 static DECLARE_RWSEM(all_cpu_access_lock
);
277 static DEFINE_PER_CPU(struct mutex
, cpu_access_lock
);
279 static inline void trace_access_lock(int cpu
)
281 if (cpu
== TRACE_PIPE_ALL_CPU
) {
282 /* gain it for accessing the whole ring buffer. */
283 down_write(&all_cpu_access_lock
);
285 /* gain it for accessing a cpu ring buffer. */
287 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
288 down_read(&all_cpu_access_lock
);
290 /* Secondly block other access to this @cpu ring buffer. */
291 mutex_lock(&per_cpu(cpu_access_lock
, cpu
));
295 static inline void trace_access_unlock(int cpu
)
297 if (cpu
== TRACE_PIPE_ALL_CPU
) {
298 up_write(&all_cpu_access_lock
);
300 mutex_unlock(&per_cpu(cpu_access_lock
, cpu
));
301 up_read(&all_cpu_access_lock
);
305 static inline void trace_access_lock_init(void)
309 for_each_possible_cpu(cpu
)
310 mutex_init(&per_cpu(cpu_access_lock
, cpu
));
315 static DEFINE_MUTEX(access_lock
);
317 static inline void trace_access_lock(int cpu
)
320 mutex_lock(&access_lock
);
323 static inline void trace_access_unlock(int cpu
)
326 mutex_unlock(&access_lock
);
329 static inline void trace_access_lock_init(void)
335 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
336 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
338 /* trace_flags holds trace_options default values */
339 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
| TRACE_ITER_PRINTK
|
340 TRACE_ITER_ANNOTATE
| TRACE_ITER_CONTEXT_INFO
| TRACE_ITER_SLEEP_TIME
|
341 TRACE_ITER_GRAPH_TIME
| TRACE_ITER_RECORD_CMD
| TRACE_ITER_OVERWRITE
|
344 static int trace_stop_count
;
345 static DEFINE_RAW_SPINLOCK(tracing_start_lock
);
347 static void wakeup_work_handler(struct work_struct
*work
)
349 wake_up(&trace_wait
);
352 static DECLARE_DELAYED_WORK(wakeup_work
, wakeup_work_handler
);
355 * trace_wake_up - wake up tasks waiting for trace input
357 * Schedules a delayed work to wake up any task that is blocked on the
358 * trace_wait queue. These is used with trace_poll for tasks polling the
361 void trace_wake_up(void)
363 const unsigned long delay
= msecs_to_jiffies(2);
365 if (trace_flags
& TRACE_ITER_BLOCK
)
367 schedule_delayed_work(&wakeup_work
, delay
);
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
[] = {
438 { trace_clock_local
, "local" },
439 { trace_clock_global
, "global" },
440 { trace_clock_counter
, "counter" },
446 * trace_parser_get_init - gets the buffer for trace parser
448 int trace_parser_get_init(struct trace_parser
*parser
, int size
)
450 memset(parser
, 0, sizeof(*parser
));
452 parser
->buffer
= kmalloc(size
, GFP_KERNEL
);
461 * trace_parser_put - frees the buffer for trace parser
463 void trace_parser_put(struct trace_parser
*parser
)
465 kfree(parser
->buffer
);
469 * trace_get_user - reads the user input string separated by space
470 * (matched by isspace(ch))
472 * For each string found the 'struct trace_parser' is updated,
473 * and the function returns.
475 * Returns number of bytes read.
477 * See kernel/trace/trace.h for 'struct trace_parser' details.
479 int trace_get_user(struct trace_parser
*parser
, const char __user
*ubuf
,
480 size_t cnt
, loff_t
*ppos
)
487 trace_parser_clear(parser
);
489 ret
= get_user(ch
, ubuf
++);
497 * The parser is not finished with the last write,
498 * continue reading the user input without skipping spaces.
501 /* skip white space */
502 while (cnt
&& isspace(ch
)) {
503 ret
= get_user(ch
, ubuf
++);
510 /* only spaces were written */
520 /* read the non-space input */
521 while (cnt
&& !isspace(ch
)) {
522 if (parser
->idx
< parser
->size
- 1)
523 parser
->buffer
[parser
->idx
++] = ch
;
528 ret
= get_user(ch
, ubuf
++);
535 /* We either got finished input or we have to wait for another call. */
537 parser
->buffer
[parser
->idx
] = 0;
538 parser
->cont
= false;
541 parser
->buffer
[parser
->idx
++] = ch
;
551 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
559 if (s
->len
<= s
->readpos
)
562 len
= s
->len
- s
->readpos
;
565 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
575 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
580 if (s
->len
<= s
->readpos
)
583 len
= s
->len
- s
->readpos
;
586 ret
= memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
595 * ftrace_max_lock is used to protect the swapping of buffers
596 * when taking a max snapshot. The buffers themselves are
597 * protected by per_cpu spinlocks. But the action of the swap
598 * needs its own lock.
600 * This is defined as a arch_spinlock_t in order to help
601 * with performance when lockdep debugging is enabled.
603 * It is also used in other places outside the update_max_tr
604 * so it needs to be defined outside of the
605 * CONFIG_TRACER_MAX_TRACE.
607 static arch_spinlock_t ftrace_max_lock
=
608 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
610 unsigned long __read_mostly tracing_thresh
;
612 #ifdef CONFIG_TRACER_MAX_TRACE
613 unsigned long __read_mostly tracing_max_latency
;
616 * Copy the new maximum trace into the separate maximum-trace
617 * structure. (this way the maximum trace is permanently saved,
618 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
621 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
623 struct trace_array_cpu
*data
= tr
->data
[cpu
];
624 struct trace_array_cpu
*max_data
;
627 max_tr
.time_start
= data
->preempt_timestamp
;
629 max_data
= max_tr
.data
[cpu
];
630 max_data
->saved_latency
= tracing_max_latency
;
631 max_data
->critical_start
= data
->critical_start
;
632 max_data
->critical_end
= data
->critical_end
;
634 memcpy(max_data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
635 max_data
->pid
= tsk
->pid
;
636 max_data
->uid
= task_uid(tsk
);
637 max_data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
638 max_data
->policy
= tsk
->policy
;
639 max_data
->rt_priority
= tsk
->rt_priority
;
641 /* record this tasks comm */
642 tracing_record_cmdline(tsk
);
646 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
648 * @tsk: the task with the latency
649 * @cpu: The cpu that initiated the trace.
651 * Flip the buffers between the @tr and the max_tr and record information
652 * about which task was the cause of this latency.
655 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
657 struct ring_buffer
*buf
= tr
->buffer
;
659 if (trace_stop_count
)
662 WARN_ON_ONCE(!irqs_disabled());
663 if (!current_trace
->use_max_tr
) {
667 arch_spin_lock(&ftrace_max_lock
);
669 tr
->buffer
= max_tr
.buffer
;
672 __update_max_tr(tr
, tsk
, cpu
);
673 arch_spin_unlock(&ftrace_max_lock
);
677 * update_max_tr_single - only copy one trace over, and reset the rest
679 * @tsk - task with the latency
680 * @cpu - the cpu of the buffer to copy.
682 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
685 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
689 if (trace_stop_count
)
692 WARN_ON_ONCE(!irqs_disabled());
693 if (!current_trace
->use_max_tr
) {
698 arch_spin_lock(&ftrace_max_lock
);
700 ftrace_disable_cpu();
702 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
706 * We failed to swap the buffer due to a commit taking
707 * place on this CPU. We fail to record, but we reset
708 * the max trace buffer (no one writes directly to it)
709 * and flag that it failed.
711 trace_array_printk(&max_tr
, _THIS_IP_
,
712 "Failed to swap buffers due to commit in progress\n");
717 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
&& ret
!= -EBUSY
);
719 __update_max_tr(tr
, tsk
, cpu
);
720 arch_spin_unlock(&ftrace_max_lock
);
722 #endif /* CONFIG_TRACER_MAX_TRACE */
725 * register_tracer - register a tracer with the ftrace system.
726 * @type - the plugin for the tracer
728 * Register a new plugin tracer.
730 int register_tracer(struct tracer
*type
)
731 __releases(kernel_lock
)
732 __acquires(kernel_lock
)
738 pr_info("Tracer must have a name\n");
742 if (strlen(type
->name
) >= MAX_TRACER_SIZE
) {
743 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE
);
747 mutex_lock(&trace_types_lock
);
749 tracing_selftest_running
= true;
751 for (t
= trace_types
; t
; t
= t
->next
) {
752 if (strcmp(type
->name
, t
->name
) == 0) {
754 pr_info("Tracer %s already registered\n",
762 type
->set_flag
= &dummy_set_flag
;
764 type
->flags
= &dummy_tracer_flags
;
766 if (!type
->flags
->opts
)
767 type
->flags
->opts
= dummy_tracer_opt
;
768 if (!type
->wait_pipe
)
769 type
->wait_pipe
= default_wait_pipe
;
772 #ifdef CONFIG_FTRACE_STARTUP_TEST
773 if (type
->selftest
&& !tracing_selftest_disabled
) {
774 struct tracer
*saved_tracer
= current_trace
;
775 struct trace_array
*tr
= &global_trace
;
778 * Run a selftest on this tracer.
779 * Here we reset the trace buffer, and set the current
780 * tracer to be this tracer. The tracer can then run some
781 * internal tracing to verify that everything is in order.
782 * If we fail, we do not register this tracer.
784 tracing_reset_online_cpus(tr
);
786 current_trace
= type
;
788 /* If we expanded the buffers, make sure the max is expanded too */
789 if (ring_buffer_expanded
&& type
->use_max_tr
)
790 ring_buffer_resize(max_tr
.buffer
, trace_buf_size
);
792 /* the test is responsible for initializing and enabling */
793 pr_info("Testing tracer %s: ", type
->name
);
794 ret
= type
->selftest(type
, tr
);
795 /* the test is responsible for resetting too */
796 current_trace
= saved_tracer
;
798 printk(KERN_CONT
"FAILED!\n");
801 /* Only reset on passing, to avoid touching corrupted buffers */
802 tracing_reset_online_cpus(tr
);
804 /* Shrink the max buffer again */
805 if (ring_buffer_expanded
&& type
->use_max_tr
)
806 ring_buffer_resize(max_tr
.buffer
, 1);
808 printk(KERN_CONT
"PASSED\n");
812 type
->next
= trace_types
;
816 tracing_selftest_running
= false;
817 mutex_unlock(&trace_types_lock
);
819 if (ret
|| !default_bootup_tracer
)
822 if (strncmp(default_bootup_tracer
, type
->name
, MAX_TRACER_SIZE
))
825 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
826 /* Do we want this tracer to start on bootup? */
827 tracing_set_tracer(type
->name
);
828 default_bootup_tracer
= NULL
;
829 /* disable other selftests, since this will break it. */
830 tracing_selftest_disabled
= 1;
831 #ifdef CONFIG_FTRACE_STARTUP_TEST
832 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
840 void unregister_tracer(struct tracer
*type
)
844 mutex_lock(&trace_types_lock
);
845 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
849 pr_info("Tracer %s not registered\n", type
->name
);
855 if (type
== current_trace
&& tracer_enabled
) {
858 if (current_trace
->stop
)
859 current_trace
->stop(&global_trace
);
860 current_trace
= &nop_trace
;
863 mutex_unlock(&trace_types_lock
);
866 static void __tracing_reset(struct ring_buffer
*buffer
, int cpu
)
868 ftrace_disable_cpu();
869 ring_buffer_reset_cpu(buffer
, cpu
);
873 void tracing_reset(struct trace_array
*tr
, int cpu
)
875 struct ring_buffer
*buffer
= tr
->buffer
;
877 ring_buffer_record_disable(buffer
);
879 /* Make sure all commits have finished */
881 __tracing_reset(buffer
, cpu
);
883 ring_buffer_record_enable(buffer
);
886 void tracing_reset_online_cpus(struct trace_array
*tr
)
888 struct ring_buffer
*buffer
= tr
->buffer
;
891 ring_buffer_record_disable(buffer
);
893 /* Make sure all commits have finished */
896 tr
->time_start
= ftrace_now(tr
->cpu
);
898 for_each_online_cpu(cpu
)
899 __tracing_reset(buffer
, cpu
);
901 ring_buffer_record_enable(buffer
);
904 void tracing_reset_current(int cpu
)
906 tracing_reset(&global_trace
, cpu
);
909 void tracing_reset_current_online_cpus(void)
911 tracing_reset_online_cpus(&global_trace
);
914 #define SAVED_CMDLINES 128
915 #define NO_CMDLINE_MAP UINT_MAX
916 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
917 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
918 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
919 static int cmdline_idx
;
920 static arch_spinlock_t trace_cmdline_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
922 /* temporary disable recording */
923 static atomic_t trace_record_cmdline_disabled __read_mostly
;
925 static void trace_init_cmdlines(void)
927 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
928 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
932 int is_tracing_stopped(void)
934 return trace_stop_count
;
938 * ftrace_off_permanent - disable all ftrace code permanently
940 * This should only be called when a serious anomally has
941 * been detected. This will turn off the function tracing,
942 * ring buffers, and other tracing utilites. It takes no
943 * locks and can be called from any context.
945 void ftrace_off_permanent(void)
947 tracing_disabled
= 1;
949 tracing_off_permanent();
953 * tracing_start - quick start of the tracer
955 * If tracing is enabled but was stopped by tracing_stop,
956 * this will start the tracer back up.
958 void tracing_start(void)
960 struct ring_buffer
*buffer
;
963 if (tracing_disabled
)
966 raw_spin_lock_irqsave(&tracing_start_lock
, flags
);
967 if (--trace_stop_count
) {
968 if (trace_stop_count
< 0) {
969 /* Someone screwed up their debugging */
971 trace_stop_count
= 0;
976 /* Prevent the buffers from switching */
977 arch_spin_lock(&ftrace_max_lock
);
979 buffer
= global_trace
.buffer
;
981 ring_buffer_record_enable(buffer
);
983 buffer
= max_tr
.buffer
;
985 ring_buffer_record_enable(buffer
);
987 arch_spin_unlock(&ftrace_max_lock
);
991 raw_spin_unlock_irqrestore(&tracing_start_lock
, flags
);
995 * tracing_stop - quick stop of the tracer
997 * Light weight way to stop tracing. Use in conjunction with
1000 void tracing_stop(void)
1002 struct ring_buffer
*buffer
;
1003 unsigned long flags
;
1006 raw_spin_lock_irqsave(&tracing_start_lock
, flags
);
1007 if (trace_stop_count
++)
1010 /* Prevent the buffers from switching */
1011 arch_spin_lock(&ftrace_max_lock
);
1013 buffer
= global_trace
.buffer
;
1015 ring_buffer_record_disable(buffer
);
1017 buffer
= max_tr
.buffer
;
1019 ring_buffer_record_disable(buffer
);
1021 arch_spin_unlock(&ftrace_max_lock
);
1024 raw_spin_unlock_irqrestore(&tracing_start_lock
, flags
);
1027 void trace_stop_cmdline_recording(void);
1029 static void trace_save_cmdline(struct task_struct
*tsk
)
1033 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
1037 * It's not the end of the world if we don't get
1038 * the lock, but we also don't want to spin
1039 * nor do we want to disable interrupts,
1040 * so if we miss here, then better luck next time.
1042 if (!arch_spin_trylock(&trace_cmdline_lock
))
1045 idx
= map_pid_to_cmdline
[tsk
->pid
];
1046 if (idx
== NO_CMDLINE_MAP
) {
1047 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
1050 * Check whether the cmdline buffer at idx has a pid
1051 * mapped. We are going to overwrite that entry so we
1052 * need to clear the map_pid_to_cmdline. Otherwise we
1053 * would read the new comm for the old pid.
1055 pid
= map_cmdline_to_pid
[idx
];
1056 if (pid
!= NO_CMDLINE_MAP
)
1057 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
1059 map_cmdline_to_pid
[idx
] = tsk
->pid
;
1060 map_pid_to_cmdline
[tsk
->pid
] = idx
;
1065 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
1067 arch_spin_unlock(&trace_cmdline_lock
);
1070 void trace_find_cmdline(int pid
, char comm
[])
1075 strcpy(comm
, "<idle>");
1079 if (WARN_ON_ONCE(pid
< 0)) {
1080 strcpy(comm
, "<XXX>");
1084 if (pid
> PID_MAX_DEFAULT
) {
1085 strcpy(comm
, "<...>");
1090 arch_spin_lock(&trace_cmdline_lock
);
1091 map
= map_pid_to_cmdline
[pid
];
1092 if (map
!= NO_CMDLINE_MAP
)
1093 strcpy(comm
, saved_cmdlines
[map
]);
1095 strcpy(comm
, "<...>");
1097 arch_spin_unlock(&trace_cmdline_lock
);
1101 void tracing_record_cmdline(struct task_struct
*tsk
)
1103 if (atomic_read(&trace_record_cmdline_disabled
) || !tracer_enabled
||
1107 trace_save_cmdline(tsk
);
1111 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
1114 struct task_struct
*tsk
= current
;
1116 entry
->preempt_count
= pc
& 0xff;
1117 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
1120 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1121 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
1123 TRACE_FLAG_IRQS_NOSUPPORT
|
1125 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
1126 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
1127 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
1129 EXPORT_SYMBOL_GPL(tracing_generic_entry_update
);
1131 struct ring_buffer_event
*
1132 trace_buffer_lock_reserve(struct ring_buffer
*buffer
,
1135 unsigned long flags
, int pc
)
1137 struct ring_buffer_event
*event
;
1139 event
= ring_buffer_lock_reserve(buffer
, len
);
1140 if (event
!= NULL
) {
1141 struct trace_entry
*ent
= ring_buffer_event_data(event
);
1143 tracing_generic_entry_update(ent
, flags
, pc
);
1151 __trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1152 struct ring_buffer_event
*event
,
1153 unsigned long flags
, int pc
,
1156 ring_buffer_unlock_commit(buffer
, event
);
1158 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1159 ftrace_trace_userstack(buffer
, flags
, pc
);
1165 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1166 struct ring_buffer_event
*event
,
1167 unsigned long flags
, int pc
)
1169 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1172 struct ring_buffer_event
*
1173 trace_current_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1174 int type
, unsigned long len
,
1175 unsigned long flags
, int pc
)
1177 *current_rb
= global_trace
.buffer
;
1178 return trace_buffer_lock_reserve(*current_rb
,
1179 type
, len
, flags
, pc
);
1181 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve
);
1183 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
1184 struct ring_buffer_event
*event
,
1185 unsigned long flags
, int pc
)
1187 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1189 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit
);
1191 void trace_nowake_buffer_unlock_commit(struct ring_buffer
*buffer
,
1192 struct ring_buffer_event
*event
,
1193 unsigned long flags
, int pc
)
1195 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 0);
1197 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit
);
1199 void trace_nowake_buffer_unlock_commit_regs(struct ring_buffer
*buffer
,
1200 struct ring_buffer_event
*event
,
1201 unsigned long flags
, int pc
,
1202 struct pt_regs
*regs
)
1204 ring_buffer_unlock_commit(buffer
, event
);
1206 ftrace_trace_stack_regs(buffer
, flags
, 0, pc
, regs
);
1207 ftrace_trace_userstack(buffer
, flags
, pc
);
1209 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit_regs
);
1211 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
1212 struct ring_buffer_event
*event
)
1214 ring_buffer_discard_commit(buffer
, event
);
1216 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit
);
1219 trace_function(struct trace_array
*tr
,
1220 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1223 struct ftrace_event_call
*call
= &event_function
;
1224 struct ring_buffer
*buffer
= tr
->buffer
;
1225 struct ring_buffer_event
*event
;
1226 struct ftrace_entry
*entry
;
1228 /* If we are reading the ring buffer, don't trace */
1229 if (unlikely(__this_cpu_read(ftrace_cpu_disabled
)))
1232 event
= trace_buffer_lock_reserve(buffer
, TRACE_FN
, sizeof(*entry
),
1236 entry
= ring_buffer_event_data(event
);
1238 entry
->parent_ip
= parent_ip
;
1240 if (!filter_check_discard(call
, entry
, buffer
, event
))
1241 ring_buffer_unlock_commit(buffer
, event
);
1245 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1246 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1249 if (likely(!atomic_read(&data
->disabled
)))
1250 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
1253 #ifdef CONFIG_STACKTRACE
1255 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1256 struct ftrace_stack
{
1257 unsigned long calls
[FTRACE_STACK_MAX_ENTRIES
];
1260 static DEFINE_PER_CPU(struct ftrace_stack
, ftrace_stack
);
1261 static DEFINE_PER_CPU(int, ftrace_stack_reserve
);
1263 static void __ftrace_trace_stack(struct ring_buffer
*buffer
,
1264 unsigned long flags
,
1265 int skip
, int pc
, struct pt_regs
*regs
)
1267 struct ftrace_event_call
*call
= &event_kernel_stack
;
1268 struct ring_buffer_event
*event
;
1269 struct stack_entry
*entry
;
1270 struct stack_trace trace
;
1272 int size
= FTRACE_STACK_ENTRIES
;
1274 trace
.nr_entries
= 0;
1278 * Since events can happen in NMIs there's no safe way to
1279 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1280 * or NMI comes in, it will just have to use the default
1281 * FTRACE_STACK_SIZE.
1283 preempt_disable_notrace();
1285 use_stack
= ++__get_cpu_var(ftrace_stack_reserve
);
1287 * We don't need any atomic variables, just a barrier.
1288 * If an interrupt comes in, we don't care, because it would
1289 * have exited and put the counter back to what we want.
1290 * We just need a barrier to keep gcc from moving things
1294 if (use_stack
== 1) {
1295 trace
.entries
= &__get_cpu_var(ftrace_stack
).calls
[0];
1296 trace
.max_entries
= FTRACE_STACK_MAX_ENTRIES
;
1299 save_stack_trace_regs(regs
, &trace
);
1301 save_stack_trace(&trace
);
1303 if (trace
.nr_entries
> size
)
1304 size
= trace
.nr_entries
;
1306 /* From now on, use_stack is a boolean */
1309 size
*= sizeof(unsigned long);
1311 event
= trace_buffer_lock_reserve(buffer
, TRACE_STACK
,
1312 sizeof(*entry
) + size
, flags
, pc
);
1315 entry
= ring_buffer_event_data(event
);
1317 memset(&entry
->caller
, 0, size
);
1320 memcpy(&entry
->caller
, trace
.entries
,
1321 trace
.nr_entries
* sizeof(unsigned long));
1323 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1324 trace
.entries
= entry
->caller
;
1326 save_stack_trace_regs(regs
, &trace
);
1328 save_stack_trace(&trace
);
1331 entry
->size
= trace
.nr_entries
;
1333 if (!filter_check_discard(call
, entry
, buffer
, event
))
1334 ring_buffer_unlock_commit(buffer
, event
);
1337 /* Again, don't let gcc optimize things here */
1339 __get_cpu_var(ftrace_stack_reserve
)--;
1340 preempt_enable_notrace();
1344 void ftrace_trace_stack_regs(struct ring_buffer
*buffer
, unsigned long flags
,
1345 int skip
, int pc
, struct pt_regs
*regs
)
1347 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1350 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, regs
);
1353 void ftrace_trace_stack(struct ring_buffer
*buffer
, unsigned long flags
,
1356 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1359 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, NULL
);
1362 void __trace_stack(struct trace_array
*tr
, unsigned long flags
, int skip
,
1365 __ftrace_trace_stack(tr
->buffer
, flags
, skip
, pc
, NULL
);
1369 * trace_dump_stack - record a stack back trace in the trace buffer
1371 void trace_dump_stack(void)
1373 unsigned long flags
;
1375 if (tracing_disabled
|| tracing_selftest_running
)
1378 local_save_flags(flags
);
1380 /* skipping 3 traces, seems to get us at the caller of this function */
1381 __ftrace_trace_stack(global_trace
.buffer
, flags
, 3, preempt_count(), NULL
);
1384 static DEFINE_PER_CPU(int, user_stack_count
);
1387 ftrace_trace_userstack(struct ring_buffer
*buffer
, unsigned long flags
, int pc
)
1389 struct ftrace_event_call
*call
= &event_user_stack
;
1390 struct ring_buffer_event
*event
;
1391 struct userstack_entry
*entry
;
1392 struct stack_trace trace
;
1394 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1398 * NMIs can not handle page faults, even with fix ups.
1399 * The save user stack can (and often does) fault.
1401 if (unlikely(in_nmi()))
1405 * prevent recursion, since the user stack tracing may
1406 * trigger other kernel events.
1409 if (__this_cpu_read(user_stack_count
))
1412 __this_cpu_inc(user_stack_count
);
1414 event
= trace_buffer_lock_reserve(buffer
, TRACE_USER_STACK
,
1415 sizeof(*entry
), flags
, pc
);
1417 goto out_drop_count
;
1418 entry
= ring_buffer_event_data(event
);
1420 entry
->tgid
= current
->tgid
;
1421 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1423 trace
.nr_entries
= 0;
1424 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1426 trace
.entries
= entry
->caller
;
1428 save_stack_trace_user(&trace
);
1429 if (!filter_check_discard(call
, entry
, buffer
, event
))
1430 ring_buffer_unlock_commit(buffer
, event
);
1433 __this_cpu_dec(user_stack_count
);
1439 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1441 ftrace_trace_userstack(tr
, flags
, preempt_count());
1445 #endif /* CONFIG_STACKTRACE */
1448 * trace_vbprintk - write binary msg to tracing buffer
1451 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1453 static arch_spinlock_t trace_buf_lock
=
1454 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
1455 static u32 trace_buf
[TRACE_BUF_SIZE
];
1457 struct ftrace_event_call
*call
= &event_bprint
;
1458 struct ring_buffer_event
*event
;
1459 struct ring_buffer
*buffer
;
1460 struct trace_array
*tr
= &global_trace
;
1461 struct trace_array_cpu
*data
;
1462 struct bprint_entry
*entry
;
1463 unsigned long flags
;
1465 int cpu
, len
= 0, size
, pc
;
1467 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1470 /* Don't pollute graph traces with trace_vprintk internals */
1471 pause_graph_tracing();
1473 pc
= preempt_count();
1474 preempt_disable_notrace();
1475 cpu
= raw_smp_processor_id();
1476 data
= tr
->data
[cpu
];
1478 disable
= atomic_inc_return(&data
->disabled
);
1479 if (unlikely(disable
!= 1))
1482 /* Lockdep uses trace_printk for lock tracing */
1483 local_irq_save(flags
);
1484 arch_spin_lock(&trace_buf_lock
);
1485 len
= vbin_printf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1487 if (len
> TRACE_BUF_SIZE
|| len
< 0)
1490 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1491 buffer
= tr
->buffer
;
1492 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
1496 entry
= ring_buffer_event_data(event
);
1500 memcpy(entry
->buf
, trace_buf
, sizeof(u32
) * len
);
1501 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1502 ring_buffer_unlock_commit(buffer
, event
);
1503 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1507 arch_spin_unlock(&trace_buf_lock
);
1508 local_irq_restore(flags
);
1511 atomic_dec_return(&data
->disabled
);
1512 preempt_enable_notrace();
1513 unpause_graph_tracing();
1517 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1519 int trace_array_printk(struct trace_array
*tr
,
1520 unsigned long ip
, const char *fmt
, ...)
1525 if (!(trace_flags
& TRACE_ITER_PRINTK
))
1529 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
1534 int trace_array_vprintk(struct trace_array
*tr
,
1535 unsigned long ip
, const char *fmt
, va_list args
)
1537 static arch_spinlock_t trace_buf_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
1538 static char trace_buf
[TRACE_BUF_SIZE
];
1540 struct ftrace_event_call
*call
= &event_print
;
1541 struct ring_buffer_event
*event
;
1542 struct ring_buffer
*buffer
;
1543 struct trace_array_cpu
*data
;
1544 int cpu
, len
= 0, size
, pc
;
1545 struct print_entry
*entry
;
1546 unsigned long irq_flags
;
1549 if (tracing_disabled
|| tracing_selftest_running
)
1552 pc
= preempt_count();
1553 preempt_disable_notrace();
1554 cpu
= raw_smp_processor_id();
1555 data
= tr
->data
[cpu
];
1557 disable
= atomic_inc_return(&data
->disabled
);
1558 if (unlikely(disable
!= 1))
1561 pause_graph_tracing();
1562 raw_local_irq_save(irq_flags
);
1563 arch_spin_lock(&trace_buf_lock
);
1564 len
= vsnprintf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1566 size
= sizeof(*entry
) + len
+ 1;
1567 buffer
= tr
->buffer
;
1568 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
1572 entry
= ring_buffer_event_data(event
);
1575 memcpy(&entry
->buf
, trace_buf
, len
);
1576 entry
->buf
[len
] = '\0';
1577 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1578 ring_buffer_unlock_commit(buffer
, event
);
1579 ftrace_trace_stack(buffer
, irq_flags
, 6, pc
);
1583 arch_spin_unlock(&trace_buf_lock
);
1584 raw_local_irq_restore(irq_flags
);
1585 unpause_graph_tracing();
1587 atomic_dec_return(&data
->disabled
);
1588 preempt_enable_notrace();
1593 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1595 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
1597 EXPORT_SYMBOL_GPL(trace_vprintk
);
1599 static void trace_iterator_increment(struct trace_iterator
*iter
)
1601 /* Don't allow ftrace to trace into the ring buffers */
1602 ftrace_disable_cpu();
1605 if (iter
->buffer_iter
[iter
->cpu
])
1606 ring_buffer_read(iter
->buffer_iter
[iter
->cpu
], NULL
);
1608 ftrace_enable_cpu();
1611 static struct trace_entry
*
1612 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
,
1613 unsigned long *lost_events
)
1615 struct ring_buffer_event
*event
;
1616 struct ring_buffer_iter
*buf_iter
= iter
->buffer_iter
[cpu
];
1618 /* Don't allow ftrace to trace into the ring buffers */
1619 ftrace_disable_cpu();
1622 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1624 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
,
1627 ftrace_enable_cpu();
1630 iter
->ent_size
= ring_buffer_event_length(event
);
1631 return ring_buffer_event_data(event
);
1637 static struct trace_entry
*
1638 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
,
1639 unsigned long *missing_events
, u64
*ent_ts
)
1641 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1642 struct trace_entry
*ent
, *next
= NULL
;
1643 unsigned long lost_events
= 0, next_lost
= 0;
1644 int cpu_file
= iter
->cpu_file
;
1645 u64 next_ts
= 0, ts
;
1650 * If we are in a per_cpu trace file, don't bother by iterating over
1651 * all cpu and peek directly.
1653 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1654 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1656 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
, missing_events
);
1658 *ent_cpu
= cpu_file
;
1663 for_each_tracing_cpu(cpu
) {
1665 if (ring_buffer_empty_cpu(buffer
, cpu
))
1668 ent
= peek_next_entry(iter
, cpu
, &ts
, &lost_events
);
1671 * Pick the entry with the smallest timestamp:
1673 if (ent
&& (!next
|| ts
< next_ts
)) {
1677 next_lost
= lost_events
;
1682 *ent_cpu
= next_cpu
;
1688 *missing_events
= next_lost
;
1693 /* Find the next real entry, without updating the iterator itself */
1694 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1695 int *ent_cpu
, u64
*ent_ts
)
1697 return __find_next_entry(iter
, ent_cpu
, NULL
, ent_ts
);
1700 /* Find the next real entry, and increment the iterator to the next entry */
1701 void *trace_find_next_entry_inc(struct trace_iterator
*iter
)
1703 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
,
1704 &iter
->lost_events
, &iter
->ts
);
1707 trace_iterator_increment(iter
);
1709 return iter
->ent
? iter
: NULL
;
1712 static void trace_consume(struct trace_iterator
*iter
)
1714 /* Don't allow ftrace to trace into the ring buffers */
1715 ftrace_disable_cpu();
1716 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
,
1717 &iter
->lost_events
);
1718 ftrace_enable_cpu();
1721 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1723 struct trace_iterator
*iter
= m
->private;
1727 WARN_ON_ONCE(iter
->leftover
);
1731 /* can't go backwards */
1736 ent
= trace_find_next_entry_inc(iter
);
1740 while (ent
&& iter
->idx
< i
)
1741 ent
= trace_find_next_entry_inc(iter
);
1748 void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
1750 struct trace_array
*tr
= iter
->tr
;
1751 struct ring_buffer_event
*event
;
1752 struct ring_buffer_iter
*buf_iter
;
1753 unsigned long entries
= 0;
1756 tr
->data
[cpu
]->skipped_entries
= 0;
1758 if (!iter
->buffer_iter
[cpu
])
1761 buf_iter
= iter
->buffer_iter
[cpu
];
1762 ring_buffer_iter_reset(buf_iter
);
1765 * We could have the case with the max latency tracers
1766 * that a reset never took place on a cpu. This is evident
1767 * by the timestamp being before the start of the buffer.
1769 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
1770 if (ts
>= iter
->tr
->time_start
)
1773 ring_buffer_read(buf_iter
, NULL
);
1776 tr
->data
[cpu
]->skipped_entries
= entries
;
1780 * The current tracer is copied to avoid a global locking
1783 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1785 struct trace_iterator
*iter
= m
->private;
1786 static struct tracer
*old_tracer
;
1787 int cpu_file
= iter
->cpu_file
;
1792 /* copy the tracer to avoid using a global lock all around */
1793 mutex_lock(&trace_types_lock
);
1794 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1795 old_tracer
= current_trace
;
1796 *iter
->trace
= *current_trace
;
1798 mutex_unlock(&trace_types_lock
);
1800 atomic_inc(&trace_record_cmdline_disabled
);
1802 if (*pos
!= iter
->pos
) {
1807 ftrace_disable_cpu();
1809 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1810 for_each_tracing_cpu(cpu
)
1811 tracing_iter_reset(iter
, cpu
);
1813 tracing_iter_reset(iter
, cpu_file
);
1815 ftrace_enable_cpu();
1818 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1823 * If we overflowed the seq_file before, then we want
1824 * to just reuse the trace_seq buffer again.
1830 p
= s_next(m
, p
, &l
);
1834 trace_event_read_lock();
1835 trace_access_lock(cpu_file
);
1839 static void s_stop(struct seq_file
*m
, void *p
)
1841 struct trace_iterator
*iter
= m
->private;
1843 atomic_dec(&trace_record_cmdline_disabled
);
1844 trace_access_unlock(iter
->cpu_file
);
1845 trace_event_read_unlock();
1849 get_total_entries(struct trace_array
*tr
, unsigned long *total
, unsigned long *entries
)
1851 unsigned long count
;
1857 for_each_tracing_cpu(cpu
) {
1858 count
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
1860 * If this buffer has skipped entries, then we hold all
1861 * entries for the trace and we need to ignore the
1862 * ones before the time stamp.
1864 if (tr
->data
[cpu
]->skipped_entries
) {
1865 count
-= tr
->data
[cpu
]->skipped_entries
;
1866 /* total is the same as the entries */
1870 ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
1875 static void print_lat_help_header(struct seq_file
*m
)
1877 seq_puts(m
, "# _------=> CPU# \n");
1878 seq_puts(m
, "# / _-----=> irqs-off \n");
1879 seq_puts(m
, "# | / _----=> need-resched \n");
1880 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1881 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1882 seq_puts(m
, "# |||| / delay \n");
1883 seq_puts(m
, "# cmd pid ||||| time | caller \n");
1884 seq_puts(m
, "# \\ / ||||| \\ | / \n");
1887 static void print_event_info(struct trace_array
*tr
, struct seq_file
*m
)
1889 unsigned long total
;
1890 unsigned long entries
;
1892 get_total_entries(tr
, &total
, &entries
);
1893 seq_printf(m
, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
1894 entries
, total
, num_online_cpus());
1898 static void print_func_help_header(struct trace_array
*tr
, struct seq_file
*m
)
1900 print_event_info(tr
, m
);
1901 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1902 seq_puts(m
, "# | | | | |\n");
1905 static void print_func_help_header_irq(struct trace_array
*tr
, struct seq_file
*m
)
1907 print_event_info(tr
, m
);
1908 seq_puts(m
, "# _-----=> irqs-off\n");
1909 seq_puts(m
, "# / _----=> need-resched\n");
1910 seq_puts(m
, "# | / _---=> hardirq/softirq\n");
1911 seq_puts(m
, "# || / _--=> preempt-depth\n");
1912 seq_puts(m
, "# ||| / delay\n");
1913 seq_puts(m
, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
1914 seq_puts(m
, "# | | | |||| | |\n");
1918 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1920 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1921 struct trace_array
*tr
= iter
->tr
;
1922 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1923 struct tracer
*type
= current_trace
;
1924 unsigned long entries
;
1925 unsigned long total
;
1926 const char *name
= "preemption";
1931 get_total_entries(tr
, &total
, &entries
);
1933 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
1935 seq_puts(m
, "# -----------------------------------"
1936 "---------------------------------\n");
1937 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1938 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1939 nsecs_to_usecs(data
->saved_latency
),
1943 #if defined(CONFIG_PREEMPT_NONE)
1945 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1947 #elif defined(CONFIG_PREEMPT)
1952 /* These are reserved for later use */
1955 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1959 seq_puts(m
, "# -----------------\n");
1960 seq_printf(m
, "# | task: %.16s-%d "
1961 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1962 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1963 data
->policy
, data
->rt_priority
);
1964 seq_puts(m
, "# -----------------\n");
1966 if (data
->critical_start
) {
1967 seq_puts(m
, "# => started at: ");
1968 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1969 trace_print_seq(m
, &iter
->seq
);
1970 seq_puts(m
, "\n# => ended at: ");
1971 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1972 trace_print_seq(m
, &iter
->seq
);
1973 seq_puts(m
, "\n#\n");
1979 static void test_cpu_buff_start(struct trace_iterator
*iter
)
1981 struct trace_seq
*s
= &iter
->seq
;
1983 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
1986 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
1989 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
1992 if (iter
->tr
->data
[iter
->cpu
]->skipped_entries
)
1995 cpumask_set_cpu(iter
->cpu
, iter
->started
);
1997 /* Don't print started cpu buffer for the first entry of the trace */
1999 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
2003 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
2005 struct trace_seq
*s
= &iter
->seq
;
2006 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2007 struct trace_entry
*entry
;
2008 struct trace_event
*event
;
2012 test_cpu_buff_start(iter
);
2014 event
= ftrace_find_event(entry
->type
);
2016 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2017 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2018 if (!trace_print_lat_context(iter
))
2021 if (!trace_print_context(iter
))
2027 return event
->funcs
->trace(iter
, sym_flags
, event
);
2029 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
2032 return TRACE_TYPE_HANDLED
;
2034 return TRACE_TYPE_PARTIAL_LINE
;
2037 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
2039 struct trace_seq
*s
= &iter
->seq
;
2040 struct trace_entry
*entry
;
2041 struct trace_event
*event
;
2045 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2046 if (!trace_seq_printf(s
, "%d %d %llu ",
2047 entry
->pid
, iter
->cpu
, iter
->ts
))
2051 event
= ftrace_find_event(entry
->type
);
2053 return event
->funcs
->raw(iter
, 0, event
);
2055 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
2058 return TRACE_TYPE_HANDLED
;
2060 return TRACE_TYPE_PARTIAL_LINE
;
2063 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
2065 struct trace_seq
*s
= &iter
->seq
;
2066 unsigned char newline
= '\n';
2067 struct trace_entry
*entry
;
2068 struct trace_event
*event
;
2072 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2073 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
2074 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
2075 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
2078 event
= ftrace_find_event(entry
->type
);
2080 enum print_line_t ret
= event
->funcs
->hex(iter
, 0, event
);
2081 if (ret
!= TRACE_TYPE_HANDLED
)
2085 SEQ_PUT_FIELD_RET(s
, newline
);
2087 return TRACE_TYPE_HANDLED
;
2090 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
2092 struct trace_seq
*s
= &iter
->seq
;
2093 struct trace_entry
*entry
;
2094 struct trace_event
*event
;
2098 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2099 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
2100 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
2101 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
2104 event
= ftrace_find_event(entry
->type
);
2105 return event
? event
->funcs
->binary(iter
, 0, event
) :
2109 int trace_empty(struct trace_iterator
*iter
)
2113 /* If we are looking at one CPU buffer, only check that one */
2114 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
2115 cpu
= iter
->cpu_file
;
2116 if (iter
->buffer_iter
[cpu
]) {
2117 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
2120 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2126 for_each_tracing_cpu(cpu
) {
2127 if (iter
->buffer_iter
[cpu
]) {
2128 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
2131 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2139 /* Called with trace_event_read_lock() held. */
2140 enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
2142 enum print_line_t ret
;
2144 if (iter
->lost_events
&&
2145 !trace_seq_printf(&iter
->seq
, "CPU:%d [LOST %lu EVENTS]\n",
2146 iter
->cpu
, iter
->lost_events
))
2147 return TRACE_TYPE_PARTIAL_LINE
;
2149 if (iter
->trace
&& iter
->trace
->print_line
) {
2150 ret
= iter
->trace
->print_line(iter
);
2151 if (ret
!= TRACE_TYPE_UNHANDLED
)
2155 if (iter
->ent
->type
== TRACE_BPRINT
&&
2156 trace_flags
& TRACE_ITER_PRINTK
&&
2157 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2158 return trace_print_bprintk_msg_only(iter
);
2160 if (iter
->ent
->type
== TRACE_PRINT
&&
2161 trace_flags
& TRACE_ITER_PRINTK
&&
2162 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2163 return trace_print_printk_msg_only(iter
);
2165 if (trace_flags
& TRACE_ITER_BIN
)
2166 return print_bin_fmt(iter
);
2168 if (trace_flags
& TRACE_ITER_HEX
)
2169 return print_hex_fmt(iter
);
2171 if (trace_flags
& TRACE_ITER_RAW
)
2172 return print_raw_fmt(iter
);
2174 return print_trace_fmt(iter
);
2177 void trace_latency_header(struct seq_file
*m
)
2179 struct trace_iterator
*iter
= m
->private;
2181 /* print nothing if the buffers are empty */
2182 if (trace_empty(iter
))
2185 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
2186 print_trace_header(m
, iter
);
2188 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2189 print_lat_help_header(m
);
2192 void trace_default_header(struct seq_file
*m
)
2194 struct trace_iterator
*iter
= m
->private;
2196 if (!(trace_flags
& TRACE_ITER_CONTEXT_INFO
))
2199 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2200 /* print nothing if the buffers are empty */
2201 if (trace_empty(iter
))
2203 print_trace_header(m
, iter
);
2204 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2205 print_lat_help_header(m
);
2207 if (!(trace_flags
& TRACE_ITER_VERBOSE
)) {
2208 if (trace_flags
& TRACE_ITER_IRQ_INFO
)
2209 print_func_help_header_irq(iter
->tr
, m
);
2211 print_func_help_header(iter
->tr
, m
);
2216 static void test_ftrace_alive(struct seq_file
*m
)
2218 if (!ftrace_is_dead())
2220 seq_printf(m
, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2221 seq_printf(m
, "# MAY BE MISSING FUNCTION EVENTS\n");
2224 static int s_show(struct seq_file
*m
, void *v
)
2226 struct trace_iterator
*iter
= v
;
2229 if (iter
->ent
== NULL
) {
2231 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
2233 test_ftrace_alive(m
);
2235 if (iter
->trace
&& iter
->trace
->print_header
)
2236 iter
->trace
->print_header(m
);
2238 trace_default_header(m
);
2240 } else if (iter
->leftover
) {
2242 * If we filled the seq_file buffer earlier, we
2243 * want to just show it now.
2245 ret
= trace_print_seq(m
, &iter
->seq
);
2247 /* ret should this time be zero, but you never know */
2248 iter
->leftover
= ret
;
2251 print_trace_line(iter
);
2252 ret
= trace_print_seq(m
, &iter
->seq
);
2254 * If we overflow the seq_file buffer, then it will
2255 * ask us for this data again at start up.
2257 * ret is 0 if seq_file write succeeded.
2260 iter
->leftover
= ret
;
2266 static const struct seq_operations tracer_seq_ops
= {
2273 static struct trace_iterator
*
2274 __tracing_open(struct inode
*inode
, struct file
*file
)
2276 long cpu_file
= (long) inode
->i_private
;
2277 void *fail_ret
= ERR_PTR(-ENOMEM
);
2278 struct trace_iterator
*iter
;
2282 if (tracing_disabled
)
2283 return ERR_PTR(-ENODEV
);
2285 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2287 return ERR_PTR(-ENOMEM
);
2290 * We make a copy of the current tracer to avoid concurrent
2291 * changes on it while we are reading.
2293 mutex_lock(&trace_types_lock
);
2294 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2299 *iter
->trace
= *current_trace
;
2301 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2304 if (current_trace
&& current_trace
->print_max
)
2307 iter
->tr
= &global_trace
;
2309 mutex_init(&iter
->mutex
);
2310 iter
->cpu_file
= cpu_file
;
2312 /* Notify the tracer early; before we stop tracing. */
2313 if (iter
->trace
&& iter
->trace
->open
)
2314 iter
->trace
->open(iter
);
2316 /* Annotate start of buffers if we had overruns */
2317 if (ring_buffer_overruns(iter
->tr
->buffer
))
2318 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2320 /* stop the trace while dumping */
2323 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
2324 for_each_tracing_cpu(cpu
) {
2325 iter
->buffer_iter
[cpu
] =
2326 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2328 ring_buffer_read_prepare_sync();
2329 for_each_tracing_cpu(cpu
) {
2330 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2331 tracing_iter_reset(iter
, cpu
);
2334 cpu
= iter
->cpu_file
;
2335 iter
->buffer_iter
[cpu
] =
2336 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2337 ring_buffer_read_prepare_sync();
2338 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2339 tracing_iter_reset(iter
, cpu
);
2342 ret
= seq_open(file
, &tracer_seq_ops
);
2344 fail_ret
= ERR_PTR(ret
);
2348 m
= file
->private_data
;
2351 mutex_unlock(&trace_types_lock
);
2356 for_each_tracing_cpu(cpu
) {
2357 if (iter
->buffer_iter
[cpu
])
2358 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2360 free_cpumask_var(iter
->started
);
2363 mutex_unlock(&trace_types_lock
);
2370 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2372 if (tracing_disabled
)
2375 filp
->private_data
= inode
->i_private
;
2379 static int tracing_release(struct inode
*inode
, struct file
*file
)
2381 struct seq_file
*m
= file
->private_data
;
2382 struct trace_iterator
*iter
;
2385 if (!(file
->f_mode
& FMODE_READ
))
2390 mutex_lock(&trace_types_lock
);
2391 for_each_tracing_cpu(cpu
) {
2392 if (iter
->buffer_iter
[cpu
])
2393 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2396 if (iter
->trace
&& iter
->trace
->close
)
2397 iter
->trace
->close(iter
);
2399 /* reenable tracing if it was previously enabled */
2401 mutex_unlock(&trace_types_lock
);
2403 seq_release(inode
, file
);
2404 mutex_destroy(&iter
->mutex
);
2405 free_cpumask_var(iter
->started
);
2411 static int tracing_open(struct inode
*inode
, struct file
*file
)
2413 struct trace_iterator
*iter
;
2416 /* If this file was open for write, then erase contents */
2417 if ((file
->f_mode
& FMODE_WRITE
) &&
2418 (file
->f_flags
& O_TRUNC
)) {
2419 long cpu
= (long) inode
->i_private
;
2421 if (cpu
== TRACE_PIPE_ALL_CPU
)
2422 tracing_reset_online_cpus(&global_trace
);
2424 tracing_reset(&global_trace
, cpu
);
2427 if (file
->f_mode
& FMODE_READ
) {
2428 iter
= __tracing_open(inode
, file
);
2430 ret
= PTR_ERR(iter
);
2431 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2432 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2438 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2440 struct tracer
*t
= v
;
2450 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2455 mutex_lock(&trace_types_lock
);
2456 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2462 static void t_stop(struct seq_file
*m
, void *p
)
2464 mutex_unlock(&trace_types_lock
);
2467 static int t_show(struct seq_file
*m
, void *v
)
2469 struct tracer
*t
= v
;
2474 seq_printf(m
, "%s", t
->name
);
2483 static const struct seq_operations show_traces_seq_ops
= {
2490 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2492 if (tracing_disabled
)
2495 return seq_open(file
, &show_traces_seq_ops
);
2499 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2500 size_t count
, loff_t
*ppos
)
2505 static loff_t
tracing_seek(struct file
*file
, loff_t offset
, int origin
)
2507 if (file
->f_mode
& FMODE_READ
)
2508 return seq_lseek(file
, offset
, origin
);
2513 static const struct file_operations tracing_fops
= {
2514 .open
= tracing_open
,
2516 .write
= tracing_write_stub
,
2517 .llseek
= tracing_seek
,
2518 .release
= tracing_release
,
2521 static const struct file_operations show_traces_fops
= {
2522 .open
= show_traces_open
,
2524 .release
= seq_release
,
2525 .llseek
= seq_lseek
,
2529 * Only trace on a CPU if the bitmask is set:
2531 static cpumask_var_t tracing_cpumask
;
2534 * The tracer itself will not take this lock, but still we want
2535 * to provide a consistent cpumask to user-space:
2537 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2540 * Temporary storage for the character representation of the
2541 * CPU bitmask (and one more byte for the newline):
2543 static char mask_str
[NR_CPUS
+ 1];
2546 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2547 size_t count
, loff_t
*ppos
)
2551 mutex_lock(&tracing_cpumask_update_lock
);
2553 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2554 if (count
- len
< 2) {
2558 len
+= sprintf(mask_str
+ len
, "\n");
2559 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2562 mutex_unlock(&tracing_cpumask_update_lock
);
2568 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2569 size_t count
, loff_t
*ppos
)
2572 cpumask_var_t tracing_cpumask_new
;
2574 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2577 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2581 mutex_lock(&tracing_cpumask_update_lock
);
2583 local_irq_disable();
2584 arch_spin_lock(&ftrace_max_lock
);
2585 for_each_tracing_cpu(cpu
) {
2587 * Increase/decrease the disabled counter if we are
2588 * about to flip a bit in the cpumask:
2590 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2591 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2592 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2594 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2595 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2596 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2599 arch_spin_unlock(&ftrace_max_lock
);
2602 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2604 mutex_unlock(&tracing_cpumask_update_lock
);
2605 free_cpumask_var(tracing_cpumask_new
);
2610 free_cpumask_var(tracing_cpumask_new
);
2615 static const struct file_operations tracing_cpumask_fops
= {
2616 .open
= tracing_open_generic
,
2617 .read
= tracing_cpumask_read
,
2618 .write
= tracing_cpumask_write
,
2619 .llseek
= generic_file_llseek
,
2622 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
2624 struct tracer_opt
*trace_opts
;
2628 mutex_lock(&trace_types_lock
);
2629 tracer_flags
= current_trace
->flags
->val
;
2630 trace_opts
= current_trace
->flags
->opts
;
2632 for (i
= 0; trace_options
[i
]; i
++) {
2633 if (trace_flags
& (1 << i
))
2634 seq_printf(m
, "%s\n", trace_options
[i
]);
2636 seq_printf(m
, "no%s\n", trace_options
[i
]);
2639 for (i
= 0; trace_opts
[i
].name
; i
++) {
2640 if (tracer_flags
& trace_opts
[i
].bit
)
2641 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
2643 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
2645 mutex_unlock(&trace_types_lock
);
2650 static int __set_tracer_option(struct tracer
*trace
,
2651 struct tracer_flags
*tracer_flags
,
2652 struct tracer_opt
*opts
, int neg
)
2656 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
2661 tracer_flags
->val
&= ~opts
->bit
;
2663 tracer_flags
->val
|= opts
->bit
;
2667 /* Try to assign a tracer specific option */
2668 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2670 struct tracer_flags
*tracer_flags
= trace
->flags
;
2671 struct tracer_opt
*opts
= NULL
;
2674 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
2675 opts
= &tracer_flags
->opts
[i
];
2677 if (strcmp(cmp
, opts
->name
) == 0)
2678 return __set_tracer_option(trace
, trace
->flags
,
2685 static void set_tracer_flags(unsigned int mask
, int enabled
)
2687 /* do nothing if flag is already set */
2688 if (!!(trace_flags
& mask
) == !!enabled
)
2692 trace_flags
|= mask
;
2694 trace_flags
&= ~mask
;
2696 if (mask
== TRACE_ITER_RECORD_CMD
)
2697 trace_event_enable_cmd_record(enabled
);
2699 if (mask
== TRACE_ITER_OVERWRITE
)
2700 ring_buffer_change_overwrite(global_trace
.buffer
, enabled
);
2704 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2705 size_t cnt
, loff_t
*ppos
)
2713 if (cnt
>= sizeof(buf
))
2716 if (copy_from_user(&buf
, ubuf
, cnt
))
2720 cmp
= strstrip(buf
);
2722 if (strncmp(cmp
, "no", 2) == 0) {
2727 for (i
= 0; trace_options
[i
]; i
++) {
2728 if (strcmp(cmp
, trace_options
[i
]) == 0) {
2729 set_tracer_flags(1 << i
, !neg
);
2734 /* If no option could be set, test the specific tracer options */
2735 if (!trace_options
[i
]) {
2736 mutex_lock(&trace_types_lock
);
2737 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2738 mutex_unlock(&trace_types_lock
);
2748 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
2750 if (tracing_disabled
)
2752 return single_open(file
, tracing_trace_options_show
, NULL
);
2755 static const struct file_operations tracing_iter_fops
= {
2756 .open
= tracing_trace_options_open
,
2758 .llseek
= seq_lseek
,
2759 .release
= single_release
,
2760 .write
= tracing_trace_options_write
,
2763 static const char readme_msg
[] =
2764 "tracing mini-HOWTO:\n\n"
2765 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2766 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2767 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2768 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2770 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2771 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2773 "# cat /sys/kernel/debug/tracing/trace_options\n"
2774 "noprint-parent nosym-offset nosym-addr noverbose\n"
2775 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2776 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2777 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2778 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2782 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2783 size_t cnt
, loff_t
*ppos
)
2785 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2786 readme_msg
, strlen(readme_msg
));
2789 static const struct file_operations tracing_readme_fops
= {
2790 .open
= tracing_open_generic
,
2791 .read
= tracing_readme_read
,
2792 .llseek
= generic_file_llseek
,
2796 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
2797 size_t cnt
, loff_t
*ppos
)
2806 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
2810 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
2818 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
2821 pid
= map_cmdline_to_pid
[i
];
2822 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
2825 trace_find_cmdline(pid
, buf_comm
);
2826 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
2831 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
2840 static const struct file_operations tracing_saved_cmdlines_fops
= {
2841 .open
= tracing_open_generic
,
2842 .read
= tracing_saved_cmdlines_read
,
2843 .llseek
= generic_file_llseek
,
2847 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2848 size_t cnt
, loff_t
*ppos
)
2853 r
= sprintf(buf
, "%u\n", tracer_enabled
);
2854 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2858 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2859 size_t cnt
, loff_t
*ppos
)
2861 struct trace_array
*tr
= filp
->private_data
;
2865 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
2871 mutex_lock(&trace_types_lock
);
2872 if (tracer_enabled
^ val
) {
2874 /* Only need to warn if this is used to change the state */
2875 WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on");
2879 if (current_trace
->start
)
2880 current_trace
->start(tr
);
2885 if (current_trace
->stop
)
2886 current_trace
->stop(tr
);
2889 mutex_unlock(&trace_types_lock
);
2897 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2898 size_t cnt
, loff_t
*ppos
)
2900 char buf
[MAX_TRACER_SIZE
+2];
2903 mutex_lock(&trace_types_lock
);
2905 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2907 r
= sprintf(buf
, "\n");
2908 mutex_unlock(&trace_types_lock
);
2910 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2913 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
2915 tracing_reset_online_cpus(tr
);
2919 static int __tracing_resize_ring_buffer(unsigned long size
)
2924 * If kernel or user changes the size of the ring buffer
2925 * we use the size that was given, and we can forget about
2926 * expanding it later.
2928 ring_buffer_expanded
= 1;
2930 ret
= ring_buffer_resize(global_trace
.buffer
, size
);
2934 if (!current_trace
->use_max_tr
)
2937 ret
= ring_buffer_resize(max_tr
.buffer
, size
);
2941 r
= ring_buffer_resize(global_trace
.buffer
,
2942 global_trace
.entries
);
2945 * AARGH! We are left with different
2946 * size max buffer!!!!
2947 * The max buffer is our "snapshot" buffer.
2948 * When a tracer needs a snapshot (one of the
2949 * latency tracers), it swaps the max buffer
2950 * with the saved snap shot. We succeeded to
2951 * update the size of the main buffer, but failed to
2952 * update the size of the max buffer. But when we tried
2953 * to reset the main buffer to the original size, we
2954 * failed there too. This is very unlikely to
2955 * happen, but if it does, warn and kill all
2959 tracing_disabled
= 1;
2964 max_tr
.entries
= size
;
2966 global_trace
.entries
= size
;
2971 static ssize_t
tracing_resize_ring_buffer(unsigned long size
)
2973 int cpu
, ret
= size
;
2975 mutex_lock(&trace_types_lock
);
2979 /* disable all cpu buffers */
2980 for_each_tracing_cpu(cpu
) {
2981 if (global_trace
.data
[cpu
])
2982 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2983 if (max_tr
.data
[cpu
])
2984 atomic_inc(&max_tr
.data
[cpu
]->disabled
);
2987 if (size
!= global_trace
.entries
)
2988 ret
= __tracing_resize_ring_buffer(size
);
2993 for_each_tracing_cpu(cpu
) {
2994 if (global_trace
.data
[cpu
])
2995 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2996 if (max_tr
.data
[cpu
])
2997 atomic_dec(&max_tr
.data
[cpu
]->disabled
);
3001 mutex_unlock(&trace_types_lock
);
3008 * tracing_update_buffers - used by tracing facility to expand ring buffers
3010 * To save on memory when the tracing is never used on a system with it
3011 * configured in. The ring buffers are set to a minimum size. But once
3012 * a user starts to use the tracing facility, then they need to grow
3013 * to their default size.
3015 * This function is to be called when a tracer is about to be used.
3017 int tracing_update_buffers(void)
3021 mutex_lock(&trace_types_lock
);
3022 if (!ring_buffer_expanded
)
3023 ret
= __tracing_resize_ring_buffer(trace_buf_size
);
3024 mutex_unlock(&trace_types_lock
);
3029 struct trace_option_dentry
;
3031 static struct trace_option_dentry
*
3032 create_trace_option_files(struct tracer
*tracer
);
3035 destroy_trace_option_files(struct trace_option_dentry
*topts
);
3037 static int tracing_set_tracer(const char *buf
)
3039 static struct trace_option_dentry
*topts
;
3040 struct trace_array
*tr
= &global_trace
;
3044 mutex_lock(&trace_types_lock
);
3046 if (!ring_buffer_expanded
) {
3047 ret
= __tracing_resize_ring_buffer(trace_buf_size
);
3053 for (t
= trace_types
; t
; t
= t
->next
) {
3054 if (strcmp(t
->name
, buf
) == 0)
3061 if (t
== current_trace
)
3064 trace_branch_disable();
3065 if (current_trace
&& current_trace
->reset
)
3066 current_trace
->reset(tr
);
3067 if (current_trace
&& current_trace
->use_max_tr
) {
3069 * We don't free the ring buffer. instead, resize it because
3070 * The max_tr ring buffer has some state (e.g. ring->clock) and
3071 * we want preserve it.
3073 ring_buffer_resize(max_tr
.buffer
, 1);
3076 destroy_trace_option_files(topts
);
3080 topts
= create_trace_option_files(current_trace
);
3081 if (current_trace
->use_max_tr
) {
3082 ret
= ring_buffer_resize(max_tr
.buffer
, global_trace
.entries
);
3085 max_tr
.entries
= global_trace
.entries
;
3089 ret
= tracer_init(t
, tr
);
3094 trace_branch_enable(tr
);
3096 mutex_unlock(&trace_types_lock
);
3102 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
3103 size_t cnt
, loff_t
*ppos
)
3105 char buf
[MAX_TRACER_SIZE
+1];
3112 if (cnt
> MAX_TRACER_SIZE
)
3113 cnt
= MAX_TRACER_SIZE
;
3115 if (copy_from_user(&buf
, ubuf
, cnt
))
3120 /* strip ending whitespace. */
3121 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
3124 err
= tracing_set_tracer(buf
);
3134 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
3135 size_t cnt
, loff_t
*ppos
)
3137 unsigned long *ptr
= filp
->private_data
;
3141 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
3142 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
3143 if (r
> sizeof(buf
))
3145 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3149 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
3150 size_t cnt
, loff_t
*ppos
)
3152 unsigned long *ptr
= filp
->private_data
;
3156 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3165 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
3167 long cpu_file
= (long) inode
->i_private
;
3168 struct trace_iterator
*iter
;
3171 if (tracing_disabled
)
3174 mutex_lock(&trace_types_lock
);
3176 /* create a buffer to store the information to pass to userspace */
3177 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
3184 * We make a copy of the current tracer to avoid concurrent
3185 * changes on it while we are reading.
3187 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
3193 *iter
->trace
= *current_trace
;
3195 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
3200 /* trace pipe does not show start of buffer */
3201 cpumask_setall(iter
->started
);
3203 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
3204 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
3206 iter
->cpu_file
= cpu_file
;
3207 iter
->tr
= &global_trace
;
3208 mutex_init(&iter
->mutex
);
3209 filp
->private_data
= iter
;
3211 if (iter
->trace
->pipe_open
)
3212 iter
->trace
->pipe_open(iter
);
3214 nonseekable_open(inode
, filp
);
3216 mutex_unlock(&trace_types_lock
);
3222 mutex_unlock(&trace_types_lock
);
3226 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
3228 struct trace_iterator
*iter
= file
->private_data
;
3230 mutex_lock(&trace_types_lock
);
3232 if (iter
->trace
->pipe_close
)
3233 iter
->trace
->pipe_close(iter
);
3235 mutex_unlock(&trace_types_lock
);
3237 free_cpumask_var(iter
->started
);
3238 mutex_destroy(&iter
->mutex
);
3246 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
3248 struct trace_iterator
*iter
= filp
->private_data
;
3250 if (trace_flags
& TRACE_ITER_BLOCK
) {
3252 * Always select as readable when in blocking mode
3254 return POLLIN
| POLLRDNORM
;
3256 if (!trace_empty(iter
))
3257 return POLLIN
| POLLRDNORM
;
3258 poll_wait(filp
, &trace_wait
, poll_table
);
3259 if (!trace_empty(iter
))
3260 return POLLIN
| POLLRDNORM
;
3267 void default_wait_pipe(struct trace_iterator
*iter
)
3271 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
3273 if (trace_empty(iter
))
3276 finish_wait(&trace_wait
, &wait
);
3280 * This is a make-shift waitqueue.
3281 * A tracer might use this callback on some rare cases:
3283 * 1) the current tracer might hold the runqueue lock when it wakes up
3284 * a reader, hence a deadlock (sched, function, and function graph tracers)
3285 * 2) the function tracers, trace all functions, we don't want
3286 * the overhead of calling wake_up and friends
3287 * (and tracing them too)
3289 * Anyway, this is really very primitive wakeup.
3291 void poll_wait_pipe(struct trace_iterator
*iter
)
3293 set_current_state(TASK_INTERRUPTIBLE
);
3294 /* sleep for 100 msecs, and try again. */
3295 schedule_timeout(HZ
/ 10);
3298 /* Must be called with trace_types_lock mutex held. */
3299 static int tracing_wait_pipe(struct file
*filp
)
3301 struct trace_iterator
*iter
= filp
->private_data
;
3303 while (trace_empty(iter
)) {
3305 if ((filp
->f_flags
& O_NONBLOCK
)) {
3309 mutex_unlock(&iter
->mutex
);
3311 iter
->trace
->wait_pipe(iter
);
3313 mutex_lock(&iter
->mutex
);
3315 if (signal_pending(current
))
3319 * We block until we read something and tracing is disabled.
3320 * We still block if tracing is disabled, but we have never
3321 * read anything. This allows a user to cat this file, and
3322 * then enable tracing. But after we have read something,
3323 * we give an EOF when tracing is again disabled.
3325 * iter->pos will be 0 if we haven't read anything.
3327 if (!tracer_enabled
&& iter
->pos
)
3338 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
3339 size_t cnt
, loff_t
*ppos
)
3341 struct trace_iterator
*iter
= filp
->private_data
;
3342 static struct tracer
*old_tracer
;
3345 /* return any leftover data */
3346 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3350 trace_seq_init(&iter
->seq
);
3352 /* copy the tracer to avoid using a global lock all around */
3353 mutex_lock(&trace_types_lock
);
3354 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3355 old_tracer
= current_trace
;
3356 *iter
->trace
= *current_trace
;
3358 mutex_unlock(&trace_types_lock
);
3361 * Avoid more than one consumer on a single file descriptor
3362 * This is just a matter of traces coherency, the ring buffer itself
3365 mutex_lock(&iter
->mutex
);
3366 if (iter
->trace
->read
) {
3367 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
3373 sret
= tracing_wait_pipe(filp
);
3377 /* stop when tracing is finished */
3378 if (trace_empty(iter
)) {
3383 if (cnt
>= PAGE_SIZE
)
3384 cnt
= PAGE_SIZE
- 1;
3386 /* reset all but tr, trace, and overruns */
3387 memset(&iter
->seq
, 0,
3388 sizeof(struct trace_iterator
) -
3389 offsetof(struct trace_iterator
, seq
));
3392 trace_event_read_lock();
3393 trace_access_lock(iter
->cpu_file
);
3394 while (trace_find_next_entry_inc(iter
) != NULL
) {
3395 enum print_line_t ret
;
3396 int len
= iter
->seq
.len
;
3398 ret
= print_trace_line(iter
);
3399 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3400 /* don't print partial lines */
3401 iter
->seq
.len
= len
;
3404 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3405 trace_consume(iter
);
3407 if (iter
->seq
.len
>= cnt
)
3411 * Setting the full flag means we reached the trace_seq buffer
3412 * size and we should leave by partial output condition above.
3413 * One of the trace_seq_* functions is not used properly.
3415 WARN_ONCE(iter
->seq
.full
, "full flag set for trace type %d",
3418 trace_access_unlock(iter
->cpu_file
);
3419 trace_event_read_unlock();
3421 /* Now copy what we have to the user */
3422 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3423 if (iter
->seq
.readpos
>= iter
->seq
.len
)
3424 trace_seq_init(&iter
->seq
);
3427 * If there was nothing to send to user, in spite of consuming trace
3428 * entries, go back to wait for more entries.
3434 mutex_unlock(&iter
->mutex
);
3439 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
3440 struct pipe_buffer
*buf
)
3442 __free_page(buf
->page
);
3445 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
3448 __free_page(spd
->pages
[idx
]);
3451 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
3453 .map
= generic_pipe_buf_map
,
3454 .unmap
= generic_pipe_buf_unmap
,
3455 .confirm
= generic_pipe_buf_confirm
,
3456 .release
= tracing_pipe_buf_release
,
3457 .steal
= generic_pipe_buf_steal
,
3458 .get
= generic_pipe_buf_get
,
3462 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
3467 /* Seq buffer is page-sized, exactly what we need. */
3469 count
= iter
->seq
.len
;
3470 ret
= print_trace_line(iter
);
3471 count
= iter
->seq
.len
- count
;
3474 iter
->seq
.len
-= count
;
3477 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3478 iter
->seq
.len
-= count
;
3482 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3483 trace_consume(iter
);
3485 if (!trace_find_next_entry_inc(iter
)) {
3495 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3497 struct pipe_inode_info
*pipe
,
3501 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3502 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3503 struct trace_iterator
*iter
= filp
->private_data
;
3504 struct splice_pipe_desc spd
= {
3506 .partial
= partial_def
,
3507 .nr_pages
= 0, /* This gets updated below. */
3509 .ops
= &tracing_pipe_buf_ops
,
3510 .spd_release
= tracing_spd_release_pipe
,
3512 static struct tracer
*old_tracer
;
3517 if (splice_grow_spd(pipe
, &spd
))
3520 /* copy the tracer to avoid using a global lock all around */
3521 mutex_lock(&trace_types_lock
);
3522 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3523 old_tracer
= current_trace
;
3524 *iter
->trace
= *current_trace
;
3526 mutex_unlock(&trace_types_lock
);
3528 mutex_lock(&iter
->mutex
);
3530 if (iter
->trace
->splice_read
) {
3531 ret
= iter
->trace
->splice_read(iter
, filp
,
3532 ppos
, pipe
, len
, flags
);
3537 ret
= tracing_wait_pipe(filp
);
3541 if (!iter
->ent
&& !trace_find_next_entry_inc(iter
)) {
3546 trace_event_read_lock();
3547 trace_access_lock(iter
->cpu_file
);
3549 /* Fill as many pages as possible. */
3550 for (i
= 0, rem
= len
; i
< pipe
->buffers
&& rem
; i
++) {
3551 spd
.pages
[i
] = alloc_page(GFP_KERNEL
);
3555 rem
= tracing_fill_pipe_page(rem
, iter
);
3557 /* Copy the data into the page, so we can start over. */
3558 ret
= trace_seq_to_buffer(&iter
->seq
,
3559 page_address(spd
.pages
[i
]),
3562 __free_page(spd
.pages
[i
]);
3565 spd
.partial
[i
].offset
= 0;
3566 spd
.partial
[i
].len
= iter
->seq
.len
;
3568 trace_seq_init(&iter
->seq
);
3571 trace_access_unlock(iter
->cpu_file
);
3572 trace_event_read_unlock();
3573 mutex_unlock(&iter
->mutex
);
3577 ret
= splice_to_pipe(pipe
, &spd
);
3579 splice_shrink_spd(pipe
, &spd
);
3583 mutex_unlock(&iter
->mutex
);
3588 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3589 size_t cnt
, loff_t
*ppos
)
3591 struct trace_array
*tr
= filp
->private_data
;
3595 mutex_lock(&trace_types_lock
);
3596 if (!ring_buffer_expanded
)
3597 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3599 trace_buf_size
>> 10);
3601 r
= sprintf(buf
, "%lu\n", tr
->entries
>> 10);
3602 mutex_unlock(&trace_types_lock
);
3604 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3608 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3609 size_t cnt
, loff_t
*ppos
)
3614 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3618 /* must have at least 1 entry */
3622 /* value is in KB */
3625 ret
= tracing_resize_ring_buffer(val
);
3635 tracing_total_entries_read(struct file
*filp
, char __user
*ubuf
,
3636 size_t cnt
, loff_t
*ppos
)
3638 struct trace_array
*tr
= filp
->private_data
;
3641 unsigned long size
= 0, expanded_size
= 0;
3643 mutex_lock(&trace_types_lock
);
3644 for_each_tracing_cpu(cpu
) {
3645 size
+= tr
->entries
>> 10;
3646 if (!ring_buffer_expanded
)
3647 expanded_size
+= trace_buf_size
>> 10;
3649 if (ring_buffer_expanded
)
3650 r
= sprintf(buf
, "%lu\n", size
);
3652 r
= sprintf(buf
, "%lu (expanded: %lu)\n", size
, expanded_size
);
3653 mutex_unlock(&trace_types_lock
);
3655 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3659 tracing_free_buffer_write(struct file
*filp
, const char __user
*ubuf
,
3660 size_t cnt
, loff_t
*ppos
)
3663 * There is no need to read what the user has written, this function
3664 * is just to make sure that there is no error when "echo" is used
3673 tracing_free_buffer_release(struct inode
*inode
, struct file
*filp
)
3675 /* disable tracing ? */
3676 if (trace_flags
& TRACE_ITER_STOP_ON_FREE
)
3678 /* resize the ring buffer to 0 */
3679 tracing_resize_ring_buffer(0);
3685 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3686 size_t cnt
, loff_t
*fpos
)
3688 unsigned long addr
= (unsigned long)ubuf
;
3689 struct ring_buffer_event
*event
;
3690 struct ring_buffer
*buffer
;
3691 struct print_entry
*entry
;
3692 unsigned long irq_flags
;
3693 struct page
*pages
[2];
3703 if (tracing_disabled
)
3706 if (cnt
> TRACE_BUF_SIZE
)
3707 cnt
= TRACE_BUF_SIZE
;
3710 * Userspace is injecting traces into the kernel trace buffer.
3711 * We want to be as non intrusive as possible.
3712 * To do so, we do not want to allocate any special buffers
3713 * or take any locks, but instead write the userspace data
3714 * straight into the ring buffer.
3716 * First we need to pin the userspace buffer into memory,
3717 * which, most likely it is, because it just referenced it.
3718 * But there's no guarantee that it is. By using get_user_pages_fast()
3719 * and kmap_atomic/kunmap_atomic() we can get access to the
3720 * pages directly. We then write the data directly into the
3723 BUILD_BUG_ON(TRACE_BUF_SIZE
>= PAGE_SIZE
);
3725 /* check if we cross pages */
3726 if ((addr
& PAGE_MASK
) != ((addr
+ cnt
) & PAGE_MASK
))
3729 offset
= addr
& (PAGE_SIZE
- 1);
3732 ret
= get_user_pages_fast(addr
, nr_pages
, 0, pages
);
3733 if (ret
< nr_pages
) {
3735 put_page(pages
[ret
]);
3740 page1
= kmap_atomic(pages
[0]);
3742 page2
= kmap_atomic(pages
[1]);
3744 local_save_flags(irq_flags
);
3745 size
= sizeof(*entry
) + cnt
+ 2; /* possible \n added */
3746 buffer
= global_trace
.buffer
;
3747 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
3748 irq_flags
, preempt_count());
3750 /* Ring buffer disabled, return as if not open for write */
3755 entry
= ring_buffer_event_data(event
);
3756 entry
->ip
= _THIS_IP_
;
3758 if (nr_pages
== 2) {
3759 len
= PAGE_SIZE
- offset
;
3760 memcpy(&entry
->buf
, page1
+ offset
, len
);
3761 memcpy(&entry
->buf
[len
], page2
, cnt
- len
);
3763 memcpy(&entry
->buf
, page1
+ offset
, cnt
);
3765 if (entry
->buf
[cnt
- 1] != '\n') {
3766 entry
->buf
[cnt
] = '\n';
3767 entry
->buf
[cnt
+ 1] = '\0';
3769 entry
->buf
[cnt
] = '\0';
3771 ring_buffer_unlock_commit(buffer
, event
);
3779 kunmap_atomic(page2
);
3780 kunmap_atomic(page1
);
3781 while (nr_pages
> 0)
3782 put_page(pages
[--nr_pages
]);
3787 static int tracing_clock_show(struct seq_file
*m
, void *v
)
3791 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
3793 "%s%s%s%s", i
? " " : "",
3794 i
== trace_clock_id
? "[" : "", trace_clocks
[i
].name
,
3795 i
== trace_clock_id
? "]" : "");
3801 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
3802 size_t cnt
, loff_t
*fpos
)
3805 const char *clockstr
;
3808 if (cnt
>= sizeof(buf
))
3811 if (copy_from_user(&buf
, ubuf
, cnt
))
3816 clockstr
= strstrip(buf
);
3818 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
3819 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
3822 if (i
== ARRAY_SIZE(trace_clocks
))
3827 mutex_lock(&trace_types_lock
);
3829 ring_buffer_set_clock(global_trace
.buffer
, trace_clocks
[i
].func
);
3831 ring_buffer_set_clock(max_tr
.buffer
, trace_clocks
[i
].func
);
3833 mutex_unlock(&trace_types_lock
);
3840 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
3842 if (tracing_disabled
)
3844 return single_open(file
, tracing_clock_show
, NULL
);
3847 static const struct file_operations tracing_max_lat_fops
= {
3848 .open
= tracing_open_generic
,
3849 .read
= tracing_max_lat_read
,
3850 .write
= tracing_max_lat_write
,
3851 .llseek
= generic_file_llseek
,
3854 static const struct file_operations tracing_ctrl_fops
= {
3855 .open
= tracing_open_generic
,
3856 .read
= tracing_ctrl_read
,
3857 .write
= tracing_ctrl_write
,
3858 .llseek
= generic_file_llseek
,
3861 static const struct file_operations set_tracer_fops
= {
3862 .open
= tracing_open_generic
,
3863 .read
= tracing_set_trace_read
,
3864 .write
= tracing_set_trace_write
,
3865 .llseek
= generic_file_llseek
,
3868 static const struct file_operations tracing_pipe_fops
= {
3869 .open
= tracing_open_pipe
,
3870 .poll
= tracing_poll_pipe
,
3871 .read
= tracing_read_pipe
,
3872 .splice_read
= tracing_splice_read_pipe
,
3873 .release
= tracing_release_pipe
,
3874 .llseek
= no_llseek
,
3877 static const struct file_operations tracing_entries_fops
= {
3878 .open
= tracing_open_generic
,
3879 .read
= tracing_entries_read
,
3880 .write
= tracing_entries_write
,
3881 .llseek
= generic_file_llseek
,
3884 static const struct file_operations tracing_total_entries_fops
= {
3885 .open
= tracing_open_generic
,
3886 .read
= tracing_total_entries_read
,
3887 .llseek
= generic_file_llseek
,
3890 static const struct file_operations tracing_free_buffer_fops
= {
3891 .write
= tracing_free_buffer_write
,
3892 .release
= tracing_free_buffer_release
,
3895 static const struct file_operations tracing_mark_fops
= {
3896 .open
= tracing_open_generic
,
3897 .write
= tracing_mark_write
,
3898 .llseek
= generic_file_llseek
,
3901 static const struct file_operations trace_clock_fops
= {
3902 .open
= tracing_clock_open
,
3904 .llseek
= seq_lseek
,
3905 .release
= single_release
,
3906 .write
= tracing_clock_write
,
3909 struct ftrace_buffer_info
{
3910 struct trace_array
*tr
;
3916 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
3918 int cpu
= (int)(long)inode
->i_private
;
3919 struct ftrace_buffer_info
*info
;
3921 if (tracing_disabled
)
3924 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3928 info
->tr
= &global_trace
;
3931 /* Force reading ring buffer for first read */
3932 info
->read
= (unsigned int)-1;
3934 filp
->private_data
= info
;
3936 return nonseekable_open(inode
, filp
);
3940 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
3941 size_t count
, loff_t
*ppos
)
3943 struct ftrace_buffer_info
*info
= filp
->private_data
;
3951 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
, info
->cpu
);
3955 /* Do we have previous read data to read? */
3956 if (info
->read
< PAGE_SIZE
)
3959 trace_access_lock(info
->cpu
);
3960 ret
= ring_buffer_read_page(info
->tr
->buffer
,
3964 trace_access_unlock(info
->cpu
);
3971 size
= PAGE_SIZE
- info
->read
;
3975 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
3986 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
3988 struct ftrace_buffer_info
*info
= file
->private_data
;
3991 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
3998 struct ring_buffer
*buffer
;
4003 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
4004 struct pipe_buffer
*buf
)
4006 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
4011 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4016 static int buffer_pipe_buf_steal(struct pipe_inode_info
*pipe
,
4017 struct pipe_buffer
*buf
)
4022 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
4023 struct pipe_buffer
*buf
)
4025 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
4030 /* Pipe buffer operations for a buffer. */
4031 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
4033 .map
= generic_pipe_buf_map
,
4034 .unmap
= generic_pipe_buf_unmap
,
4035 .confirm
= generic_pipe_buf_confirm
,
4036 .release
= buffer_pipe_buf_release
,
4037 .steal
= buffer_pipe_buf_steal
,
4038 .get
= buffer_pipe_buf_get
,
4042 * Callback from splice_to_pipe(), if we need to release some pages
4043 * at the end of the spd in case we error'ed out in filling the pipe.
4045 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
4047 struct buffer_ref
*ref
=
4048 (struct buffer_ref
*)spd
->partial
[i
].private;
4053 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4055 spd
->partial
[i
].private = 0;
4059 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
4060 struct pipe_inode_info
*pipe
, size_t len
,
4063 struct ftrace_buffer_info
*info
= file
->private_data
;
4064 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
4065 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
4066 struct splice_pipe_desc spd
= {
4068 .partial
= partial_def
,
4070 .ops
= &buffer_pipe_buf_ops
,
4071 .spd_release
= buffer_spd_release
,
4073 struct buffer_ref
*ref
;
4074 int entries
, size
, i
;
4077 if (splice_grow_spd(pipe
, &spd
))
4080 if (*ppos
& (PAGE_SIZE
- 1)) {
4081 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
4086 if (len
& (PAGE_SIZE
- 1)) {
4087 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
4088 if (len
< PAGE_SIZE
) {
4095 trace_access_lock(info
->cpu
);
4096 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4098 for (i
= 0; i
< pipe
->buffers
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
4102 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
4107 ref
->buffer
= info
->tr
->buffer
;
4108 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
, info
->cpu
);
4114 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
4117 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4123 * zero out any left over data, this is going to
4126 size
= ring_buffer_page_len(ref
->page
);
4127 if (size
< PAGE_SIZE
)
4128 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
4130 page
= virt_to_page(ref
->page
);
4132 spd
.pages
[i
] = page
;
4133 spd
.partial
[i
].len
= PAGE_SIZE
;
4134 spd
.partial
[i
].offset
= 0;
4135 spd
.partial
[i
].private = (unsigned long)ref
;
4139 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4142 trace_access_unlock(info
->cpu
);
4145 /* did we read anything? */
4146 if (!spd
.nr_pages
) {
4147 if (flags
& SPLICE_F_NONBLOCK
)
4155 ret
= splice_to_pipe(pipe
, &spd
);
4156 splice_shrink_spd(pipe
, &spd
);
4161 static const struct file_operations tracing_buffers_fops
= {
4162 .open
= tracing_buffers_open
,
4163 .read
= tracing_buffers_read
,
4164 .release
= tracing_buffers_release
,
4165 .splice_read
= tracing_buffers_splice_read
,
4166 .llseek
= no_llseek
,
4170 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
4171 size_t count
, loff_t
*ppos
)
4173 unsigned long cpu
= (unsigned long)filp
->private_data
;
4174 struct trace_array
*tr
= &global_trace
;
4175 struct trace_seq
*s
;
4177 unsigned long long t
;
4178 unsigned long usec_rem
;
4180 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
4186 cnt
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
4187 trace_seq_printf(s
, "entries: %ld\n", cnt
);
4189 cnt
= ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
4190 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
4192 cnt
= ring_buffer_commit_overrun_cpu(tr
->buffer
, cpu
);
4193 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
4195 cnt
= ring_buffer_bytes_cpu(tr
->buffer
, cpu
);
4196 trace_seq_printf(s
, "bytes: %ld\n", cnt
);
4198 t
= ns2usecs(ring_buffer_oldest_event_ts(tr
->buffer
, cpu
));
4199 usec_rem
= do_div(t
, USEC_PER_SEC
);
4200 trace_seq_printf(s
, "oldest event ts: %5llu.%06lu\n", t
, usec_rem
);
4202 t
= ns2usecs(ring_buffer_time_stamp(tr
->buffer
, cpu
));
4203 usec_rem
= do_div(t
, USEC_PER_SEC
);
4204 trace_seq_printf(s
, "now ts: %5llu.%06lu\n", t
, usec_rem
);
4206 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
4213 static const struct file_operations tracing_stats_fops
= {
4214 .open
= tracing_open_generic
,
4215 .read
= tracing_stats_read
,
4216 .llseek
= generic_file_llseek
,
4219 #ifdef CONFIG_DYNAMIC_FTRACE
4221 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
4227 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
4228 size_t cnt
, loff_t
*ppos
)
4230 static char ftrace_dyn_info_buffer
[1024];
4231 static DEFINE_MUTEX(dyn_info_mutex
);
4232 unsigned long *p
= filp
->private_data
;
4233 char *buf
= ftrace_dyn_info_buffer
;
4234 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
4237 mutex_lock(&dyn_info_mutex
);
4238 r
= sprintf(buf
, "%ld ", *p
);
4240 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
4243 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4245 mutex_unlock(&dyn_info_mutex
);
4250 static const struct file_operations tracing_dyn_info_fops
= {
4251 .open
= tracing_open_generic
,
4252 .read
= tracing_read_dyn_info
,
4253 .llseek
= generic_file_llseek
,
4257 static struct dentry
*d_tracer
;
4259 struct dentry
*tracing_init_dentry(void)
4266 if (!debugfs_initialized())
4269 d_tracer
= debugfs_create_dir("tracing", NULL
);
4271 if (!d_tracer
&& !once
) {
4273 pr_warning("Could not create debugfs directory 'tracing'\n");
4280 static struct dentry
*d_percpu
;
4282 struct dentry
*tracing_dentry_percpu(void)
4285 struct dentry
*d_tracer
;
4290 d_tracer
= tracing_init_dentry();
4295 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
4297 if (!d_percpu
&& !once
) {
4299 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4306 static void tracing_init_debugfs_percpu(long cpu
)
4308 struct dentry
*d_percpu
= tracing_dentry_percpu();
4309 struct dentry
*d_cpu
;
4310 char cpu_dir
[30]; /* 30 characters should be more than enough */
4312 snprintf(cpu_dir
, 30, "cpu%ld", cpu
);
4313 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
4315 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
4319 /* per cpu trace_pipe */
4320 trace_create_file("trace_pipe", 0444, d_cpu
,
4321 (void *) cpu
, &tracing_pipe_fops
);
4324 trace_create_file("trace", 0644, d_cpu
,
4325 (void *) cpu
, &tracing_fops
);
4327 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
4328 (void *) cpu
, &tracing_buffers_fops
);
4330 trace_create_file("stats", 0444, d_cpu
,
4331 (void *) cpu
, &tracing_stats_fops
);
4334 #ifdef CONFIG_FTRACE_SELFTEST
4335 /* Let selftest have access to static functions in this file */
4336 #include "trace_selftest.c"
4339 struct trace_option_dentry
{
4340 struct tracer_opt
*opt
;
4341 struct tracer_flags
*flags
;
4342 struct dentry
*entry
;
4346 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4349 struct trace_option_dentry
*topt
= filp
->private_data
;
4352 if (topt
->flags
->val
& topt
->opt
->bit
)
4357 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4361 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4364 struct trace_option_dentry
*topt
= filp
->private_data
;
4368 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4372 if (val
!= 0 && val
!= 1)
4375 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
4376 mutex_lock(&trace_types_lock
);
4377 ret
= __set_tracer_option(current_trace
, topt
->flags
,
4379 mutex_unlock(&trace_types_lock
);
4390 static const struct file_operations trace_options_fops
= {
4391 .open
= tracing_open_generic
,
4392 .read
= trace_options_read
,
4393 .write
= trace_options_write
,
4394 .llseek
= generic_file_llseek
,
4398 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4401 long index
= (long)filp
->private_data
;
4404 if (trace_flags
& (1 << index
))
4409 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4413 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4416 long index
= (long)filp
->private_data
;
4420 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4424 if (val
!= 0 && val
!= 1)
4426 set_tracer_flags(1 << index
, val
);
4433 static const struct file_operations trace_options_core_fops
= {
4434 .open
= tracing_open_generic
,
4435 .read
= trace_options_core_read
,
4436 .write
= trace_options_core_write
,
4437 .llseek
= generic_file_llseek
,
4440 struct dentry
*trace_create_file(const char *name
,
4442 struct dentry
*parent
,
4444 const struct file_operations
*fops
)
4448 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
4450 pr_warning("Could not create debugfs '%s' entry\n", name
);
4456 static struct dentry
*trace_options_init_dentry(void)
4458 struct dentry
*d_tracer
;
4459 static struct dentry
*t_options
;
4464 d_tracer
= tracing_init_dentry();
4468 t_options
= debugfs_create_dir("options", d_tracer
);
4470 pr_warning("Could not create debugfs directory 'options'\n");
4478 create_trace_option_file(struct trace_option_dentry
*topt
,
4479 struct tracer_flags
*flags
,
4480 struct tracer_opt
*opt
)
4482 struct dentry
*t_options
;
4484 t_options
= trace_options_init_dentry();
4488 topt
->flags
= flags
;
4491 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
4492 &trace_options_fops
);
4496 static struct trace_option_dentry
*
4497 create_trace_option_files(struct tracer
*tracer
)
4499 struct trace_option_dentry
*topts
;
4500 struct tracer_flags
*flags
;
4501 struct tracer_opt
*opts
;
4507 flags
= tracer
->flags
;
4509 if (!flags
|| !flags
->opts
)
4514 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4517 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
4521 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4522 create_trace_option_file(&topts
[cnt
], flags
,
4529 destroy_trace_option_files(struct trace_option_dentry
*topts
)
4536 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
4537 if (topts
[cnt
].entry
)
4538 debugfs_remove(topts
[cnt
].entry
);
4544 static struct dentry
*
4545 create_trace_option_core_file(const char *option
, long index
)
4547 struct dentry
*t_options
;
4549 t_options
= trace_options_init_dentry();
4553 return trace_create_file(option
, 0644, t_options
, (void *)index
,
4554 &trace_options_core_fops
);
4557 static __init
void create_trace_options_dir(void)
4559 struct dentry
*t_options
;
4562 t_options
= trace_options_init_dentry();
4566 for (i
= 0; trace_options
[i
]; i
++)
4567 create_trace_option_core_file(trace_options
[i
], i
);
4570 static __init
int tracer_init_debugfs(void)
4572 struct dentry
*d_tracer
;
4575 trace_access_lock_init();
4577 d_tracer
= tracing_init_dentry();
4579 trace_create_file("tracing_enabled", 0644, d_tracer
,
4580 &global_trace
, &tracing_ctrl_fops
);
4582 trace_create_file("trace_options", 0644, d_tracer
,
4583 NULL
, &tracing_iter_fops
);
4585 trace_create_file("tracing_cpumask", 0644, d_tracer
,
4586 NULL
, &tracing_cpumask_fops
);
4588 trace_create_file("trace", 0644, d_tracer
,
4589 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
4591 trace_create_file("available_tracers", 0444, d_tracer
,
4592 &global_trace
, &show_traces_fops
);
4594 trace_create_file("current_tracer", 0644, d_tracer
,
4595 &global_trace
, &set_tracer_fops
);
4597 #ifdef CONFIG_TRACER_MAX_TRACE
4598 trace_create_file("tracing_max_latency", 0644, d_tracer
,
4599 &tracing_max_latency
, &tracing_max_lat_fops
);
4602 trace_create_file("tracing_thresh", 0644, d_tracer
,
4603 &tracing_thresh
, &tracing_max_lat_fops
);
4605 trace_create_file("README", 0444, d_tracer
,
4606 NULL
, &tracing_readme_fops
);
4608 trace_create_file("trace_pipe", 0444, d_tracer
,
4609 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
4611 trace_create_file("buffer_size_kb", 0644, d_tracer
,
4612 &global_trace
, &tracing_entries_fops
);
4614 trace_create_file("buffer_total_size_kb", 0444, d_tracer
,
4615 &global_trace
, &tracing_total_entries_fops
);
4617 trace_create_file("free_buffer", 0644, d_tracer
,
4618 &global_trace
, &tracing_free_buffer_fops
);
4620 trace_create_file("trace_marker", 0220, d_tracer
,
4621 NULL
, &tracing_mark_fops
);
4623 trace_create_file("saved_cmdlines", 0444, d_tracer
,
4624 NULL
, &tracing_saved_cmdlines_fops
);
4626 trace_create_file("trace_clock", 0644, d_tracer
, NULL
,
4629 #ifdef CONFIG_DYNAMIC_FTRACE
4630 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
4631 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
4634 create_trace_options_dir();
4636 for_each_tracing_cpu(cpu
)
4637 tracing_init_debugfs_percpu(cpu
);
4642 static int trace_panic_handler(struct notifier_block
*this,
4643 unsigned long event
, void *unused
)
4645 if (ftrace_dump_on_oops
)
4646 ftrace_dump(ftrace_dump_on_oops
);
4650 static struct notifier_block trace_panic_notifier
= {
4651 .notifier_call
= trace_panic_handler
,
4653 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
4656 static int trace_die_handler(struct notifier_block
*self
,
4662 if (ftrace_dump_on_oops
)
4663 ftrace_dump(ftrace_dump_on_oops
);
4671 static struct notifier_block trace_die_notifier
= {
4672 .notifier_call
= trace_die_handler
,
4677 * printk is set to max of 1024, we really don't need it that big.
4678 * Nothing should be printing 1000 characters anyway.
4680 #define TRACE_MAX_PRINT 1000
4683 * Define here KERN_TRACE so that we have one place to modify
4684 * it if we decide to change what log level the ftrace dump
4687 #define KERN_TRACE KERN_EMERG
4690 trace_printk_seq(struct trace_seq
*s
)
4692 /* Probably should print a warning here. */
4696 /* should be zero ended, but we are paranoid. */
4697 s
->buffer
[s
->len
] = 0;
4699 printk(KERN_TRACE
"%s", s
->buffer
);
4704 void trace_init_global_iter(struct trace_iterator
*iter
)
4706 iter
->tr
= &global_trace
;
4707 iter
->trace
= current_trace
;
4708 iter
->cpu_file
= TRACE_PIPE_ALL_CPU
;
4712 __ftrace_dump(bool disable_tracing
, enum ftrace_dump_mode oops_dump_mode
)
4714 static arch_spinlock_t ftrace_dump_lock
=
4715 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
4716 /* use static because iter can be a bit big for the stack */
4717 static struct trace_iterator iter
;
4718 unsigned int old_userobj
;
4719 static int dump_ran
;
4720 unsigned long flags
;
4724 local_irq_save(flags
);
4725 arch_spin_lock(&ftrace_dump_lock
);
4733 /* Did function tracer already get disabled? */
4734 if (ftrace_is_dead()) {
4735 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
4736 printk("# MAY BE MISSING FUNCTION EVENTS\n");
4739 if (disable_tracing
)
4742 trace_init_global_iter(&iter
);
4744 for_each_tracing_cpu(cpu
) {
4745 atomic_inc(&iter
.tr
->data
[cpu
]->disabled
);
4748 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
4750 /* don't look at user memory in panic mode */
4751 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
4753 /* Simulate the iterator */
4754 iter
.tr
= &global_trace
;
4755 iter
.trace
= current_trace
;
4757 switch (oops_dump_mode
) {
4759 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4762 iter
.cpu_file
= raw_smp_processor_id();
4767 printk(KERN_TRACE
"Bad dumping mode, switching to all CPUs dump\n");
4768 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4771 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
4774 * We need to stop all tracing on all CPUS to read the
4775 * the next buffer. This is a bit expensive, but is
4776 * not done often. We fill all what we can read,
4777 * and then release the locks again.
4780 while (!trace_empty(&iter
)) {
4783 printk(KERN_TRACE
"---------------------------------\n");
4787 /* reset all but tr, trace, and overruns */
4788 memset(&iter
.seq
, 0,
4789 sizeof(struct trace_iterator
) -
4790 offsetof(struct trace_iterator
, seq
));
4791 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
4794 if (trace_find_next_entry_inc(&iter
) != NULL
) {
4797 ret
= print_trace_line(&iter
);
4798 if (ret
!= TRACE_TYPE_NO_CONSUME
)
4799 trace_consume(&iter
);
4802 trace_printk_seq(&iter
.seq
);
4806 printk(KERN_TRACE
" (ftrace buffer empty)\n");
4808 printk(KERN_TRACE
"---------------------------------\n");
4811 /* Re-enable tracing if requested */
4812 if (!disable_tracing
) {
4813 trace_flags
|= old_userobj
;
4815 for_each_tracing_cpu(cpu
) {
4816 atomic_dec(&iter
.tr
->data
[cpu
]->disabled
);
4822 arch_spin_unlock(&ftrace_dump_lock
);
4823 local_irq_restore(flags
);
4826 /* By default: disable tracing after the dump */
4827 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode
)
4829 __ftrace_dump(true, oops_dump_mode
);
4831 EXPORT_SYMBOL_GPL(ftrace_dump
);
4833 __init
static int tracer_alloc_buffers(void)
4836 enum ring_buffer_flags rb_flags
;
4841 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
4844 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
4845 goto out_free_buffer_mask
;
4847 /* To save memory, keep the ring buffer size to its minimum */
4848 if (ring_buffer_expanded
)
4849 ring_buf_size
= trace_buf_size
;
4853 rb_flags
= trace_flags
& TRACE_ITER_OVERWRITE
? RB_FL_OVERWRITE
: 0;
4855 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
4856 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
4858 /* TODO: make the number of buffers hot pluggable with CPUS */
4859 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
, rb_flags
);
4860 if (!global_trace
.buffer
) {
4861 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
4863 goto out_free_cpumask
;
4865 global_trace
.entries
= ring_buffer_size(global_trace
.buffer
);
4868 #ifdef CONFIG_TRACER_MAX_TRACE
4869 max_tr
.buffer
= ring_buffer_alloc(1, rb_flags
);
4870 if (!max_tr
.buffer
) {
4871 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
4873 ring_buffer_free(global_trace
.buffer
);
4874 goto out_free_cpumask
;
4879 /* Allocate the first page for all buffers */
4880 for_each_tracing_cpu(i
) {
4881 global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
4882 max_tr
.data
[i
] = &per_cpu(max_tr_data
, i
);
4885 trace_init_cmdlines();
4887 register_tracer(&nop_trace
);
4888 current_trace
= &nop_trace
;
4889 /* All seems OK, enable tracing */
4890 tracing_disabled
= 0;
4892 atomic_notifier_chain_register(&panic_notifier_list
,
4893 &trace_panic_notifier
);
4895 register_die_notifier(&trace_die_notifier
);
4900 free_cpumask_var(tracing_cpumask
);
4901 out_free_buffer_mask
:
4902 free_cpumask_var(tracing_buffer_mask
);
4907 __init
static int clear_boot_tracer(void)
4910 * The default tracer at boot buffer is an init section.
4911 * This function is called in lateinit. If we did not
4912 * find the boot tracer, then clear it out, to prevent
4913 * later registration from accessing the buffer that is
4914 * about to be freed.
4916 if (!default_bootup_tracer
)
4919 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
4920 default_bootup_tracer
);
4921 default_bootup_tracer
= NULL
;
4926 early_initcall(tracer_alloc_buffers
);
4927 fs_initcall(tracer_init_debugfs
);
4928 late_initcall(clear_boot_tracer
);