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 <linux/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/ctype.h>
35 #include <linux/init.h>
36 #include <linux/poll.h>
37 #include <linux/gfp.h>
41 #include "trace_output.h"
43 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
45 unsigned long __read_mostly tracing_max_latency
;
46 unsigned long __read_mostly tracing_thresh
;
49 * On boot up, the ring buffer is set to the minimum size, so that
50 * we do not waste memory on systems that are not using tracing.
52 static int ring_buffer_expanded
;
55 * We need to change this state when a selftest is running.
56 * A selftest will lurk into the ring-buffer to count the
57 * entries inserted during the selftest although some concurrent
58 * insertions into the ring-buffer such as trace_printk could occurred
59 * at the same time, giving false positive or negative results.
61 static bool __read_mostly tracing_selftest_running
;
64 * If a tracer is running, we do not want to run SELFTEST.
66 static bool __read_mostly tracing_selftest_disabled
;
68 /* For tracers that don't implement custom flags */
69 static struct tracer_opt dummy_tracer_opt
[] = {
73 static struct tracer_flags dummy_tracer_flags
= {
75 .opts
= dummy_tracer_opt
78 static int dummy_set_flag(u32 old_flags
, u32 bit
, int set
)
84 * Kill all tracing for good (never come back).
85 * It is initialized to 1 but will turn to zero if the initialization
86 * of the tracer is successful. But that is the only place that sets
89 static int tracing_disabled
= 1;
91 static DEFINE_PER_CPU(local_t
, ftrace_cpu_disabled
);
93 static inline void ftrace_disable_cpu(void)
96 local_inc(&__get_cpu_var(ftrace_cpu_disabled
));
99 static inline void ftrace_enable_cpu(void)
101 local_dec(&__get_cpu_var(ftrace_cpu_disabled
));
105 static cpumask_var_t __read_mostly tracing_buffer_mask
;
107 /* Define which cpu buffers are currently read in trace_pipe */
108 static cpumask_var_t tracing_reader_cpumask
;
110 #define for_each_tracing_cpu(cpu) \
111 for_each_cpu(cpu, tracing_buffer_mask)
114 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
116 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
117 * is set, then ftrace_dump is called. This will output the contents
118 * of the ftrace buffers to the console. This is very useful for
119 * capturing traces that lead to crashes and outputing it to a
122 * It is default off, but you can enable it with either specifying
123 * "ftrace_dump_on_oops" in the kernel command line, or setting
124 * /proc/sys/kernel/ftrace_dump_on_oops to true.
126 int ftrace_dump_on_oops
;
128 static int tracing_set_tracer(const char *buf
);
130 #define BOOTUP_TRACER_SIZE 100
131 static char bootup_tracer_buf
[BOOTUP_TRACER_SIZE
] __initdata
;
132 static char *default_bootup_tracer
;
134 static int __init
set_ftrace(char *str
)
136 strncpy(bootup_tracer_buf
, str
, BOOTUP_TRACER_SIZE
);
137 default_bootup_tracer
= bootup_tracer_buf
;
138 /* We are using ftrace early, expand it */
139 ring_buffer_expanded
= 1;
142 __setup("ftrace=", set_ftrace
);
144 static int __init
set_ftrace_dump_on_oops(char *str
)
146 ftrace_dump_on_oops
= 1;
149 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops
);
151 unsigned long long ns2usecs(cycle_t nsec
)
159 * The global_trace is the descriptor that holds the tracing
160 * buffers for the live tracing. For each CPU, it contains
161 * a link list of pages that will store trace entries. The
162 * page descriptor of the pages in the memory is used to hold
163 * the link list by linking the lru item in the page descriptor
164 * to each of the pages in the buffer per CPU.
166 * For each active CPU there is a data field that holds the
167 * pages for the buffer for that CPU. Each CPU has the same number
168 * of pages allocated for its buffer.
170 static struct trace_array global_trace
;
172 static DEFINE_PER_CPU(struct trace_array_cpu
, global_trace_cpu
);
174 cycle_t
ftrace_now(int cpu
)
178 /* Early boot up does not have a buffer yet */
179 if (!global_trace
.buffer
)
180 return trace_clock_local();
182 ts
= ring_buffer_time_stamp(global_trace
.buffer
, cpu
);
183 ring_buffer_normalize_time_stamp(global_trace
.buffer
, cpu
, &ts
);
189 * The max_tr is used to snapshot the global_trace when a maximum
190 * latency is reached. Some tracers will use this to store a maximum
191 * trace while it continues examining live traces.
193 * The buffers for the max_tr are set up the same as the global_trace.
194 * When a snapshot is taken, the link list of the max_tr is swapped
195 * with the link list of the global_trace and the buffers are reset for
196 * the global_trace so the tracing can continue.
198 static struct trace_array max_tr
;
200 static DEFINE_PER_CPU(struct trace_array_cpu
, max_data
);
202 /* tracer_enabled is used to toggle activation of a tracer */
203 static int tracer_enabled
= 1;
206 * tracing_is_enabled - return tracer_enabled status
208 * This function is used by other tracers to know the status
209 * of the tracer_enabled flag. Tracers may use this function
210 * to know if it should enable their features when starting
211 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
213 int tracing_is_enabled(void)
215 return tracer_enabled
;
219 * trace_buf_size is the size in bytes that is allocated
220 * for a buffer. Note, the number of bytes is always rounded
223 * This number is purposely set to a low number of 16384.
224 * If the dump on oops happens, it will be much appreciated
225 * to not have to wait for all that output. Anyway this can be
226 * boot time and run time configurable.
228 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
230 static unsigned long trace_buf_size
= TRACE_BUF_SIZE_DEFAULT
;
232 /* trace_types holds a link list of available tracers. */
233 static struct tracer
*trace_types __read_mostly
;
235 /* current_trace points to the tracer that is currently active */
236 static struct tracer
*current_trace __read_mostly
;
239 * max_tracer_type_len is used to simplify the allocating of
240 * buffers to read userspace tracer names. We keep track of
241 * the longest tracer name registered.
243 static int max_tracer_type_len
;
246 * trace_types_lock is used to protect the trace_types list.
247 * This lock is also used to keep user access serialized.
248 * Accesses from userspace will grab this lock while userspace
249 * activities happen inside the kernel.
251 static DEFINE_MUTEX(trace_types_lock
);
253 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
254 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
256 /* trace_flags holds trace_options default values */
257 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
| TRACE_ITER_PRINTK
|
258 TRACE_ITER_ANNOTATE
| TRACE_ITER_CONTEXT_INFO
| TRACE_ITER_SLEEP_TIME
|
259 TRACE_ITER_GRAPH_TIME
;
262 * trace_wake_up - wake up tasks waiting for trace input
264 * Simply wakes up any task that is blocked on the trace_wait
265 * queue. These is used with trace_poll for tasks polling the trace.
267 void trace_wake_up(void)
270 * The runqueue_is_locked() can fail, but this is the best we
273 if (!(trace_flags
& TRACE_ITER_BLOCK
) && !runqueue_is_locked())
274 wake_up(&trace_wait
);
277 static int __init
set_buf_size(char *str
)
279 unsigned long buf_size
;
284 ret
= strict_strtoul(str
, 0, &buf_size
);
285 /* nr_entries can not be zero */
286 if (ret
< 0 || buf_size
== 0)
288 trace_buf_size
= buf_size
;
291 __setup("trace_buf_size=", set_buf_size
);
293 unsigned long nsecs_to_usecs(unsigned long nsecs
)
298 /* These must match the bit postions in trace_iterator_flags */
299 static const char *trace_options
[] = {
326 * ftrace_max_lock is used to protect the swapping of buffers
327 * when taking a max snapshot. The buffers themselves are
328 * protected by per_cpu spinlocks. But the action of the swap
329 * needs its own lock.
331 * This is defined as a raw_spinlock_t in order to help
332 * with performance when lockdep debugging is enabled.
334 static raw_spinlock_t ftrace_max_lock
=
335 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
338 * Copy the new maximum trace into the separate maximum-trace
339 * structure. (this way the maximum trace is permanently saved,
340 * for later retrieval via /debugfs/tracing/latency_trace)
343 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
345 struct trace_array_cpu
*data
= tr
->data
[cpu
];
348 max_tr
.time_start
= data
->preempt_timestamp
;
350 data
= max_tr
.data
[cpu
];
351 data
->saved_latency
= tracing_max_latency
;
353 memcpy(data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
354 data
->pid
= tsk
->pid
;
355 data
->uid
= task_uid(tsk
);
356 data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
357 data
->policy
= tsk
->policy
;
358 data
->rt_priority
= tsk
->rt_priority
;
360 /* record this tasks comm */
361 tracing_record_cmdline(tsk
);
364 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
372 if (s
->len
<= s
->readpos
)
375 len
= s
->len
- s
->readpos
;
378 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
388 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
393 if (s
->len
<= s
->readpos
)
396 len
= s
->len
- s
->readpos
;
399 ret
= memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
408 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
410 * @tsk: the task with the latency
411 * @cpu: The cpu that initiated the trace.
413 * Flip the buffers between the @tr and the max_tr and record information
414 * about which task was the cause of this latency.
417 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
419 struct ring_buffer
*buf
= tr
->buffer
;
421 WARN_ON_ONCE(!irqs_disabled());
422 __raw_spin_lock(&ftrace_max_lock
);
424 tr
->buffer
= max_tr
.buffer
;
427 ftrace_disable_cpu();
428 ring_buffer_reset(tr
->buffer
);
431 __update_max_tr(tr
, tsk
, cpu
);
432 __raw_spin_unlock(&ftrace_max_lock
);
436 * update_max_tr_single - only copy one trace over, and reset the rest
438 * @tsk - task with the latency
439 * @cpu - the cpu of the buffer to copy.
441 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
444 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
448 WARN_ON_ONCE(!irqs_disabled());
449 __raw_spin_lock(&ftrace_max_lock
);
451 ftrace_disable_cpu();
453 ring_buffer_reset(max_tr
.buffer
);
454 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
458 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
);
460 __update_max_tr(tr
, tsk
, cpu
);
461 __raw_spin_unlock(&ftrace_max_lock
);
465 * register_tracer - register a tracer with the ftrace system.
466 * @type - the plugin for the tracer
468 * Register a new plugin tracer.
470 int register_tracer(struct tracer
*type
)
471 __releases(kernel_lock
)
472 __acquires(kernel_lock
)
479 pr_info("Tracer must have a name\n");
484 * When this gets called we hold the BKL which means that
485 * preemption is disabled. Various trace selftests however
486 * need to disable and enable preemption for successful tests.
487 * So we drop the BKL here and grab it after the tests again.
490 mutex_lock(&trace_types_lock
);
492 tracing_selftest_running
= true;
494 for (t
= trace_types
; t
; t
= t
->next
) {
495 if (strcmp(type
->name
, t
->name
) == 0) {
497 pr_info("Trace %s already registered\n",
505 type
->set_flag
= &dummy_set_flag
;
507 type
->flags
= &dummy_tracer_flags
;
509 if (!type
->flags
->opts
)
510 type
->flags
->opts
= dummy_tracer_opt
;
511 if (!type
->wait_pipe
)
512 type
->wait_pipe
= default_wait_pipe
;
515 #ifdef CONFIG_FTRACE_STARTUP_TEST
516 if (type
->selftest
&& !tracing_selftest_disabled
) {
517 struct tracer
*saved_tracer
= current_trace
;
518 struct trace_array
*tr
= &global_trace
;
522 * Run a selftest on this tracer.
523 * Here we reset the trace buffer, and set the current
524 * tracer to be this tracer. The tracer can then run some
525 * internal tracing to verify that everything is in order.
526 * If we fail, we do not register this tracer.
528 for_each_tracing_cpu(i
)
529 tracing_reset(tr
, i
);
531 current_trace
= type
;
532 /* the test is responsible for initializing and enabling */
533 pr_info("Testing tracer %s: ", type
->name
);
534 ret
= type
->selftest(type
, tr
);
535 /* the test is responsible for resetting too */
536 current_trace
= saved_tracer
;
538 printk(KERN_CONT
"FAILED!\n");
541 /* Only reset on passing, to avoid touching corrupted buffers */
542 for_each_tracing_cpu(i
)
543 tracing_reset(tr
, i
);
545 printk(KERN_CONT
"PASSED\n");
549 type
->next
= trace_types
;
551 len
= strlen(type
->name
);
552 if (len
> max_tracer_type_len
)
553 max_tracer_type_len
= len
;
556 tracing_selftest_running
= false;
557 mutex_unlock(&trace_types_lock
);
559 if (ret
|| !default_bootup_tracer
)
562 if (strncmp(default_bootup_tracer
, type
->name
, BOOTUP_TRACER_SIZE
))
565 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
566 /* Do we want this tracer to start on bootup? */
567 tracing_set_tracer(type
->name
);
568 default_bootup_tracer
= NULL
;
569 /* disable other selftests, since this will break it. */
570 tracing_selftest_disabled
= 1;
571 #ifdef CONFIG_FTRACE_STARTUP_TEST
572 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
581 void unregister_tracer(struct tracer
*type
)
586 mutex_lock(&trace_types_lock
);
587 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
591 pr_info("Trace %s not registered\n", type
->name
);
597 if (type
== current_trace
&& tracer_enabled
) {
600 if (current_trace
->stop
)
601 current_trace
->stop(&global_trace
);
602 current_trace
= &nop_trace
;
605 if (strlen(type
->name
) != max_tracer_type_len
)
608 max_tracer_type_len
= 0;
609 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
610 len
= strlen((*t
)->name
);
611 if (len
> max_tracer_type_len
)
612 max_tracer_type_len
= len
;
615 mutex_unlock(&trace_types_lock
);
618 void tracing_reset(struct trace_array
*tr
, int cpu
)
620 ftrace_disable_cpu();
621 ring_buffer_reset_cpu(tr
->buffer
, cpu
);
625 void tracing_reset_online_cpus(struct trace_array
*tr
)
629 tr
->time_start
= ftrace_now(tr
->cpu
);
631 for_each_online_cpu(cpu
)
632 tracing_reset(tr
, cpu
);
635 #define SAVED_CMDLINES 128
636 #define NO_CMDLINE_MAP UINT_MAX
637 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
638 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
639 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
640 static int cmdline_idx
;
641 static raw_spinlock_t trace_cmdline_lock
= __RAW_SPIN_LOCK_UNLOCKED
;
643 /* temporary disable recording */
644 static atomic_t trace_record_cmdline_disabled __read_mostly
;
646 static void trace_init_cmdlines(void)
648 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
649 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
653 static int trace_stop_count
;
654 static DEFINE_SPINLOCK(tracing_start_lock
);
657 * ftrace_off_permanent - disable all ftrace code permanently
659 * This should only be called when a serious anomally has
660 * been detected. This will turn off the function tracing,
661 * ring buffers, and other tracing utilites. It takes no
662 * locks and can be called from any context.
664 void ftrace_off_permanent(void)
666 tracing_disabled
= 1;
668 tracing_off_permanent();
672 * tracing_start - quick start of the tracer
674 * If tracing is enabled but was stopped by tracing_stop,
675 * this will start the tracer back up.
677 void tracing_start(void)
679 struct ring_buffer
*buffer
;
682 if (tracing_disabled
)
685 spin_lock_irqsave(&tracing_start_lock
, flags
);
686 if (--trace_stop_count
) {
687 if (trace_stop_count
< 0) {
688 /* Someone screwed up their debugging */
690 trace_stop_count
= 0;
696 buffer
= global_trace
.buffer
;
698 ring_buffer_record_enable(buffer
);
700 buffer
= max_tr
.buffer
;
702 ring_buffer_record_enable(buffer
);
706 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
710 * tracing_stop - quick stop of the tracer
712 * Light weight way to stop tracing. Use in conjunction with
715 void tracing_stop(void)
717 struct ring_buffer
*buffer
;
721 spin_lock_irqsave(&tracing_start_lock
, flags
);
722 if (trace_stop_count
++)
725 buffer
= global_trace
.buffer
;
727 ring_buffer_record_disable(buffer
);
729 buffer
= max_tr
.buffer
;
731 ring_buffer_record_disable(buffer
);
734 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
737 void trace_stop_cmdline_recording(void);
739 static void trace_save_cmdline(struct task_struct
*tsk
)
743 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
747 * It's not the end of the world if we don't get
748 * the lock, but we also don't want to spin
749 * nor do we want to disable interrupts,
750 * so if we miss here, then better luck next time.
752 if (!__raw_spin_trylock(&trace_cmdline_lock
))
755 idx
= map_pid_to_cmdline
[tsk
->pid
];
756 if (idx
== NO_CMDLINE_MAP
) {
757 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
760 * Check whether the cmdline buffer at idx has a pid
761 * mapped. We are going to overwrite that entry so we
762 * need to clear the map_pid_to_cmdline. Otherwise we
763 * would read the new comm for the old pid.
765 pid
= map_cmdline_to_pid
[idx
];
766 if (pid
!= NO_CMDLINE_MAP
)
767 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
769 map_cmdline_to_pid
[idx
] = tsk
->pid
;
770 map_pid_to_cmdline
[tsk
->pid
] = idx
;
775 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
777 __raw_spin_unlock(&trace_cmdline_lock
);
780 void trace_find_cmdline(int pid
, char comm
[])
785 strcpy(comm
, "<idle>");
789 if (pid
> PID_MAX_DEFAULT
) {
790 strcpy(comm
, "<...>");
794 __raw_spin_lock(&trace_cmdline_lock
);
795 map
= map_pid_to_cmdline
[pid
];
796 if (map
!= NO_CMDLINE_MAP
)
797 strcpy(comm
, saved_cmdlines
[map
]);
799 strcpy(comm
, "<...>");
801 __raw_spin_unlock(&trace_cmdline_lock
);
804 void tracing_record_cmdline(struct task_struct
*tsk
)
806 if (atomic_read(&trace_record_cmdline_disabled
) || !tracer_enabled
||
810 trace_save_cmdline(tsk
);
814 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
817 struct task_struct
*tsk
= current
;
819 entry
->preempt_count
= pc
& 0xff;
820 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
821 entry
->tgid
= (tsk
) ? tsk
->tgid
: 0;
823 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
824 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
826 TRACE_FLAG_IRQS_NOSUPPORT
|
828 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
829 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
830 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
833 struct ring_buffer_event
*trace_buffer_lock_reserve(struct trace_array
*tr
,
836 unsigned long flags
, int pc
)
838 struct ring_buffer_event
*event
;
840 event
= ring_buffer_lock_reserve(tr
->buffer
, len
);
842 struct trace_entry
*ent
= ring_buffer_event_data(event
);
844 tracing_generic_entry_update(ent
, flags
, pc
);
850 static void ftrace_trace_stack(struct trace_array
*tr
,
851 unsigned long flags
, int skip
, int pc
);
852 static void ftrace_trace_userstack(struct trace_array
*tr
,
853 unsigned long flags
, int pc
);
855 static inline void __trace_buffer_unlock_commit(struct trace_array
*tr
,
856 struct ring_buffer_event
*event
,
857 unsigned long flags
, int pc
,
860 ring_buffer_unlock_commit(tr
->buffer
, event
);
862 ftrace_trace_stack(tr
, flags
, 6, pc
);
863 ftrace_trace_userstack(tr
, flags
, pc
);
869 void trace_buffer_unlock_commit(struct trace_array
*tr
,
870 struct ring_buffer_event
*event
,
871 unsigned long flags
, int pc
)
873 __trace_buffer_unlock_commit(tr
, event
, flags
, pc
, 1);
876 struct ring_buffer_event
*
877 trace_current_buffer_lock_reserve(unsigned char type
, unsigned long len
,
878 unsigned long flags
, int pc
)
880 return trace_buffer_lock_reserve(&global_trace
,
881 type
, len
, flags
, pc
);
884 void trace_current_buffer_unlock_commit(struct ring_buffer_event
*event
,
885 unsigned long flags
, int pc
)
887 return __trace_buffer_unlock_commit(&global_trace
, event
, flags
, pc
, 1);
890 void trace_nowake_buffer_unlock_commit(struct ring_buffer_event
*event
,
891 unsigned long flags
, int pc
)
893 return __trace_buffer_unlock_commit(&global_trace
, event
, flags
, pc
, 0);
897 trace_function(struct trace_array
*tr
,
898 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
901 struct ftrace_event_call
*call
= &event_function
;
902 struct ring_buffer_event
*event
;
903 struct ftrace_entry
*entry
;
905 /* If we are reading the ring buffer, don't trace */
906 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
909 event
= trace_buffer_lock_reserve(tr
, TRACE_FN
, sizeof(*entry
),
913 entry
= ring_buffer_event_data(event
);
915 entry
->parent_ip
= parent_ip
;
917 filter_check_discard(call
, entry
, event
);
919 ring_buffer_unlock_commit(tr
->buffer
, event
);
922 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
923 static int __trace_graph_entry(struct trace_array
*tr
,
924 struct ftrace_graph_ent
*trace
,
928 struct ftrace_event_call
*call
= &event_funcgraph_entry
;
929 struct ring_buffer_event
*event
;
930 struct ftrace_graph_ent_entry
*entry
;
932 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
935 event
= trace_buffer_lock_reserve(&global_trace
, TRACE_GRAPH_ENT
,
936 sizeof(*entry
), flags
, pc
);
939 entry
= ring_buffer_event_data(event
);
940 entry
->graph_ent
= *trace
;
941 filter_check_discard(call
, entry
, event
);
942 ring_buffer_unlock_commit(global_trace
.buffer
, event
);
947 static void __trace_graph_return(struct trace_array
*tr
,
948 struct ftrace_graph_ret
*trace
,
952 struct ftrace_event_call
*call
= &event_funcgraph_exit
;
953 struct ring_buffer_event
*event
;
954 struct ftrace_graph_ret_entry
*entry
;
956 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
959 event
= trace_buffer_lock_reserve(&global_trace
, TRACE_GRAPH_RET
,
960 sizeof(*entry
), flags
, pc
);
963 entry
= ring_buffer_event_data(event
);
965 filter_check_discard(call
, entry
, event
);
966 ring_buffer_unlock_commit(global_trace
.buffer
, event
);
971 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
972 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
975 if (likely(!atomic_read(&data
->disabled
)))
976 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
979 static void __ftrace_trace_stack(struct trace_array
*tr
,
983 #ifdef CONFIG_STACKTRACE
984 struct ftrace_event_call
*call
= &event_kernel_stack
;
985 struct ring_buffer_event
*event
;
986 struct stack_entry
*entry
;
987 struct stack_trace trace
;
989 event
= trace_buffer_lock_reserve(tr
, TRACE_STACK
,
990 sizeof(*entry
), flags
, pc
);
993 entry
= ring_buffer_event_data(event
);
994 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
996 trace
.nr_entries
= 0;
997 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
999 trace
.entries
= entry
->caller
;
1001 save_stack_trace(&trace
);
1002 filter_check_discard(call
, entry
, event
);
1003 ring_buffer_unlock_commit(tr
->buffer
, event
);
1007 static void ftrace_trace_stack(struct trace_array
*tr
,
1008 unsigned long flags
,
1011 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1014 __ftrace_trace_stack(tr
, flags
, skip
, pc
);
1017 void __trace_stack(struct trace_array
*tr
,
1018 unsigned long flags
,
1021 __ftrace_trace_stack(tr
, flags
, skip
, pc
);
1024 static void ftrace_trace_userstack(struct trace_array
*tr
,
1025 unsigned long flags
, int pc
)
1027 #ifdef CONFIG_STACKTRACE
1028 struct ftrace_event_call
*call
= &event_user_stack
;
1029 struct ring_buffer_event
*event
;
1030 struct userstack_entry
*entry
;
1031 struct stack_trace trace
;
1033 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1036 event
= trace_buffer_lock_reserve(tr
, TRACE_USER_STACK
,
1037 sizeof(*entry
), flags
, pc
);
1040 entry
= ring_buffer_event_data(event
);
1042 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1044 trace
.nr_entries
= 0;
1045 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1047 trace
.entries
= entry
->caller
;
1049 save_stack_trace_user(&trace
);
1050 filter_check_discard(call
, entry
, event
);
1051 ring_buffer_unlock_commit(tr
->buffer
, event
);
1056 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1058 ftrace_trace_userstack(tr
, flags
, preempt_count());
1063 ftrace_trace_special(void *__tr
,
1064 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
,
1067 struct ftrace_event_call
*call
= &event_special
;
1068 struct ring_buffer_event
*event
;
1069 struct trace_array
*tr
= __tr
;
1070 struct special_entry
*entry
;
1072 event
= trace_buffer_lock_reserve(tr
, TRACE_SPECIAL
,
1073 sizeof(*entry
), 0, pc
);
1076 entry
= ring_buffer_event_data(event
);
1080 filter_check_discard(call
, entry
, event
);
1081 trace_buffer_unlock_commit(tr
, event
, 0, pc
);
1085 __trace_special(void *__tr
, void *__data
,
1086 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1088 ftrace_trace_special(__tr
, arg1
, arg2
, arg3
, preempt_count());
1092 tracing_sched_switch_trace(struct trace_array
*tr
,
1093 struct task_struct
*prev
,
1094 struct task_struct
*next
,
1095 unsigned long flags
, int pc
)
1097 struct ftrace_event_call
*call
= &event_context_switch
;
1098 struct ring_buffer_event
*event
;
1099 struct ctx_switch_entry
*entry
;
1101 event
= trace_buffer_lock_reserve(tr
, TRACE_CTX
,
1102 sizeof(*entry
), flags
, pc
);
1105 entry
= ring_buffer_event_data(event
);
1106 entry
->prev_pid
= prev
->pid
;
1107 entry
->prev_prio
= prev
->prio
;
1108 entry
->prev_state
= prev
->state
;
1109 entry
->next_pid
= next
->pid
;
1110 entry
->next_prio
= next
->prio
;
1111 entry
->next_state
= next
->state
;
1112 entry
->next_cpu
= task_cpu(next
);
1114 filter_check_discard(call
, entry
, event
);
1116 trace_buffer_unlock_commit(tr
, event
, flags
, pc
);
1120 tracing_sched_wakeup_trace(struct trace_array
*tr
,
1121 struct task_struct
*wakee
,
1122 struct task_struct
*curr
,
1123 unsigned long flags
, int pc
)
1125 struct ftrace_event_call
*call
= &event_wakeup
;
1126 struct ring_buffer_event
*event
;
1127 struct ctx_switch_entry
*entry
;
1129 event
= trace_buffer_lock_reserve(tr
, TRACE_WAKE
,
1130 sizeof(*entry
), flags
, pc
);
1133 entry
= ring_buffer_event_data(event
);
1134 entry
->prev_pid
= curr
->pid
;
1135 entry
->prev_prio
= curr
->prio
;
1136 entry
->prev_state
= curr
->state
;
1137 entry
->next_pid
= wakee
->pid
;
1138 entry
->next_prio
= wakee
->prio
;
1139 entry
->next_state
= wakee
->state
;
1140 entry
->next_cpu
= task_cpu(wakee
);
1142 filter_check_discard(call
, entry
, event
);
1144 ring_buffer_unlock_commit(tr
->buffer
, event
);
1145 ftrace_trace_stack(tr
, flags
, 6, pc
);
1146 ftrace_trace_userstack(tr
, flags
, pc
);
1150 ftrace_special(unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1152 struct trace_array
*tr
= &global_trace
;
1153 struct trace_array_cpu
*data
;
1154 unsigned long flags
;
1158 if (tracing_disabled
)
1161 pc
= preempt_count();
1162 local_irq_save(flags
);
1163 cpu
= raw_smp_processor_id();
1164 data
= tr
->data
[cpu
];
1166 if (likely(atomic_inc_return(&data
->disabled
) == 1))
1167 ftrace_trace_special(tr
, arg1
, arg2
, arg3
, pc
);
1169 atomic_dec(&data
->disabled
);
1170 local_irq_restore(flags
);
1173 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1174 int trace_graph_entry(struct ftrace_graph_ent
*trace
)
1176 struct trace_array
*tr
= &global_trace
;
1177 struct trace_array_cpu
*data
;
1178 unsigned long flags
;
1184 if (!ftrace_trace_task(current
))
1187 if (!ftrace_graph_addr(trace
->func
))
1190 local_irq_save(flags
);
1191 cpu
= raw_smp_processor_id();
1192 data
= tr
->data
[cpu
];
1193 disabled
= atomic_inc_return(&data
->disabled
);
1194 if (likely(disabled
== 1)) {
1195 pc
= preempt_count();
1196 ret
= __trace_graph_entry(tr
, trace
, flags
, pc
);
1200 /* Only do the atomic if it is not already set */
1201 if (!test_tsk_trace_graph(current
))
1202 set_tsk_trace_graph(current
);
1204 atomic_dec(&data
->disabled
);
1205 local_irq_restore(flags
);
1210 void trace_graph_return(struct ftrace_graph_ret
*trace
)
1212 struct trace_array
*tr
= &global_trace
;
1213 struct trace_array_cpu
*data
;
1214 unsigned long flags
;
1219 local_irq_save(flags
);
1220 cpu
= raw_smp_processor_id();
1221 data
= tr
->data
[cpu
];
1222 disabled
= atomic_inc_return(&data
->disabled
);
1223 if (likely(disabled
== 1)) {
1224 pc
= preempt_count();
1225 __trace_graph_return(tr
, trace
, flags
, pc
);
1228 clear_tsk_trace_graph(current
);
1229 atomic_dec(&data
->disabled
);
1230 local_irq_restore(flags
);
1232 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1236 * trace_vbprintk - write binary msg to tracing buffer
1239 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1241 static raw_spinlock_t trace_buf_lock
=
1242 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
1243 static u32 trace_buf
[TRACE_BUF_SIZE
];
1245 struct ftrace_event_call
*call
= &event_bprint
;
1246 struct ring_buffer_event
*event
;
1247 struct trace_array
*tr
= &global_trace
;
1248 struct trace_array_cpu
*data
;
1249 struct bprint_entry
*entry
;
1250 unsigned long flags
;
1252 int cpu
, len
= 0, size
, pc
;
1254 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1257 /* Don't pollute graph traces with trace_vprintk internals */
1258 pause_graph_tracing();
1260 pc
= preempt_count();
1261 resched
= ftrace_preempt_disable();
1262 cpu
= raw_smp_processor_id();
1263 data
= tr
->data
[cpu
];
1265 if (unlikely(atomic_read(&data
->disabled
)))
1268 /* Lockdep uses trace_printk for lock tracing */
1269 local_irq_save(flags
);
1270 __raw_spin_lock(&trace_buf_lock
);
1271 len
= vbin_printf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1273 if (len
> TRACE_BUF_SIZE
|| len
< 0)
1276 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1277 event
= trace_buffer_lock_reserve(tr
, TRACE_BPRINT
, size
, flags
, pc
);
1280 entry
= ring_buffer_event_data(event
);
1284 memcpy(entry
->buf
, trace_buf
, sizeof(u32
) * len
);
1285 filter_check_discard(call
, entry
, event
);
1286 ring_buffer_unlock_commit(tr
->buffer
, event
);
1289 __raw_spin_unlock(&trace_buf_lock
);
1290 local_irq_restore(flags
);
1293 ftrace_preempt_enable(resched
);
1294 unpause_graph_tracing();
1298 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1300 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1302 static raw_spinlock_t trace_buf_lock
= __RAW_SPIN_LOCK_UNLOCKED
;
1303 static char trace_buf
[TRACE_BUF_SIZE
];
1305 struct ftrace_event_call
*call
= &event_print
;
1306 struct ring_buffer_event
*event
;
1307 struct trace_array
*tr
= &global_trace
;
1308 struct trace_array_cpu
*data
;
1309 int cpu
, len
= 0, size
, pc
;
1310 struct print_entry
*entry
;
1311 unsigned long irq_flags
;
1313 if (tracing_disabled
|| tracing_selftest_running
)
1316 pc
= preempt_count();
1317 preempt_disable_notrace();
1318 cpu
= raw_smp_processor_id();
1319 data
= tr
->data
[cpu
];
1321 if (unlikely(atomic_read(&data
->disabled
)))
1324 pause_graph_tracing();
1325 raw_local_irq_save(irq_flags
);
1326 __raw_spin_lock(&trace_buf_lock
);
1327 len
= vsnprintf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1329 len
= min(len
, TRACE_BUF_SIZE
-1);
1332 size
= sizeof(*entry
) + len
+ 1;
1333 event
= trace_buffer_lock_reserve(tr
, TRACE_PRINT
, size
, irq_flags
, pc
);
1336 entry
= ring_buffer_event_data(event
);
1339 memcpy(&entry
->buf
, trace_buf
, len
);
1340 entry
->buf
[len
] = 0;
1341 filter_check_discard(call
, entry
, event
);
1342 ring_buffer_unlock_commit(tr
->buffer
, event
);
1345 __raw_spin_unlock(&trace_buf_lock
);
1346 raw_local_irq_restore(irq_flags
);
1347 unpause_graph_tracing();
1349 preempt_enable_notrace();
1353 EXPORT_SYMBOL_GPL(trace_vprintk
);
1355 enum trace_file_type
{
1356 TRACE_FILE_LAT_FMT
= 1,
1357 TRACE_FILE_ANNOTATE
= 2,
1360 static void trace_iterator_increment(struct trace_iterator
*iter
)
1362 /* Don't allow ftrace to trace into the ring buffers */
1363 ftrace_disable_cpu();
1366 if (iter
->buffer_iter
[iter
->cpu
])
1367 ring_buffer_read(iter
->buffer_iter
[iter
->cpu
], NULL
);
1369 ftrace_enable_cpu();
1372 static struct trace_entry
*
1373 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
)
1375 struct ring_buffer_event
*event
;
1376 struct ring_buffer_iter
*buf_iter
= iter
->buffer_iter
[cpu
];
1378 /* Don't allow ftrace to trace into the ring buffers */
1379 ftrace_disable_cpu();
1382 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1384 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
);
1386 ftrace_enable_cpu();
1388 return event
? ring_buffer_event_data(event
) : NULL
;
1391 static struct trace_entry
*
1392 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
, u64
*ent_ts
)
1394 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1395 struct trace_entry
*ent
, *next
= NULL
;
1396 int cpu_file
= iter
->cpu_file
;
1397 u64 next_ts
= 0, ts
;
1402 * If we are in a per_cpu trace file, don't bother by iterating over
1403 * all cpu and peek directly.
1405 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1406 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1408 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
);
1410 *ent_cpu
= cpu_file
;
1415 for_each_tracing_cpu(cpu
) {
1417 if (ring_buffer_empty_cpu(buffer
, cpu
))
1420 ent
= peek_next_entry(iter
, cpu
, &ts
);
1423 * Pick the entry with the smallest timestamp:
1425 if (ent
&& (!next
|| ts
< next_ts
)) {
1433 *ent_cpu
= next_cpu
;
1441 /* Find the next real entry, without updating the iterator itself */
1442 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1443 int *ent_cpu
, u64
*ent_ts
)
1445 return __find_next_entry(iter
, ent_cpu
, ent_ts
);
1448 /* Find the next real entry, and increment the iterator to the next entry */
1449 static void *find_next_entry_inc(struct trace_iterator
*iter
)
1451 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
, &iter
->ts
);
1454 trace_iterator_increment(iter
);
1456 return iter
->ent
? iter
: NULL
;
1459 static void trace_consume(struct trace_iterator
*iter
)
1461 /* Don't allow ftrace to trace into the ring buffers */
1462 ftrace_disable_cpu();
1463 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
);
1464 ftrace_enable_cpu();
1467 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1469 struct trace_iterator
*iter
= m
->private;
1475 /* can't go backwards */
1480 ent
= find_next_entry_inc(iter
);
1484 while (ent
&& iter
->idx
< i
)
1485 ent
= find_next_entry_inc(iter
);
1493 * No necessary locking here. The worst thing which can
1494 * happen is loosing events consumed at the same time
1495 * by a trace_pipe reader.
1496 * Other than that, we don't risk to crash the ring buffer
1497 * because it serializes the readers.
1499 * The current tracer is copied to avoid a global locking
1502 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1504 struct trace_iterator
*iter
= m
->private;
1505 static struct tracer
*old_tracer
;
1506 int cpu_file
= iter
->cpu_file
;
1511 /* copy the tracer to avoid using a global lock all around */
1512 mutex_lock(&trace_types_lock
);
1513 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1514 old_tracer
= current_trace
;
1515 *iter
->trace
= *current_trace
;
1517 mutex_unlock(&trace_types_lock
);
1519 atomic_inc(&trace_record_cmdline_disabled
);
1521 if (*pos
!= iter
->pos
) {
1526 ftrace_disable_cpu();
1528 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1529 for_each_tracing_cpu(cpu
)
1530 ring_buffer_iter_reset(iter
->buffer_iter
[cpu
]);
1532 ring_buffer_iter_reset(iter
->buffer_iter
[cpu_file
]);
1535 ftrace_enable_cpu();
1537 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1542 p
= s_next(m
, p
, &l
);
1548 static void s_stop(struct seq_file
*m
, void *p
)
1550 atomic_dec(&trace_record_cmdline_disabled
);
1553 static void print_lat_help_header(struct seq_file
*m
)
1555 seq_puts(m
, "# _------=> CPU# \n");
1556 seq_puts(m
, "# / _-----=> irqs-off \n");
1557 seq_puts(m
, "# | / _----=> need-resched \n");
1558 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1559 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1560 seq_puts(m
, "# |||| / \n");
1561 seq_puts(m
, "# ||||| delay \n");
1562 seq_puts(m
, "# cmd pid ||||| time | caller \n");
1563 seq_puts(m
, "# \\ / ||||| \\ | / \n");
1566 static void print_func_help_header(struct seq_file
*m
)
1568 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1569 seq_puts(m
, "# | | | | |\n");
1574 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1576 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1577 struct trace_array
*tr
= iter
->tr
;
1578 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1579 struct tracer
*type
= current_trace
;
1580 unsigned long total
;
1581 unsigned long entries
;
1582 const char *name
= "preemption";
1587 entries
= ring_buffer_entries(iter
->tr
->buffer
);
1589 ring_buffer_overruns(iter
->tr
->buffer
);
1591 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
1593 seq_puts(m
, "# -----------------------------------"
1594 "---------------------------------\n");
1595 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1596 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1597 nsecs_to_usecs(data
->saved_latency
),
1601 #if defined(CONFIG_PREEMPT_NONE)
1603 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1605 #elif defined(CONFIG_PREEMPT)
1610 /* These are reserved for later use */
1613 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1617 seq_puts(m
, "# -----------------\n");
1618 seq_printf(m
, "# | task: %.16s-%d "
1619 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1620 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1621 data
->policy
, data
->rt_priority
);
1622 seq_puts(m
, "# -----------------\n");
1624 if (data
->critical_start
) {
1625 seq_puts(m
, "# => started at: ");
1626 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1627 trace_print_seq(m
, &iter
->seq
);
1628 seq_puts(m
, "\n# => ended at: ");
1629 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1630 trace_print_seq(m
, &iter
->seq
);
1637 static void test_cpu_buff_start(struct trace_iterator
*iter
)
1639 struct trace_seq
*s
= &iter
->seq
;
1641 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
1644 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
1647 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
1650 cpumask_set_cpu(iter
->cpu
, iter
->started
);
1652 /* Don't print started cpu buffer for the first entry of the trace */
1654 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
1658 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
1660 struct trace_seq
*s
= &iter
->seq
;
1661 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1662 struct trace_entry
*entry
;
1663 struct trace_event
*event
;
1667 test_cpu_buff_start(iter
);
1669 event
= ftrace_find_event(entry
->type
);
1671 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1672 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1673 if (!trace_print_lat_context(iter
))
1676 if (!trace_print_context(iter
))
1682 return event
->trace(iter
, sym_flags
);
1684 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
1687 return TRACE_TYPE_HANDLED
;
1689 return TRACE_TYPE_PARTIAL_LINE
;
1692 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
1694 struct trace_seq
*s
= &iter
->seq
;
1695 struct trace_entry
*entry
;
1696 struct trace_event
*event
;
1700 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1701 if (!trace_seq_printf(s
, "%d %d %llu ",
1702 entry
->pid
, iter
->cpu
, iter
->ts
))
1706 event
= ftrace_find_event(entry
->type
);
1708 return event
->raw(iter
, 0);
1710 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
1713 return TRACE_TYPE_HANDLED
;
1715 return TRACE_TYPE_PARTIAL_LINE
;
1718 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
1720 struct trace_seq
*s
= &iter
->seq
;
1721 unsigned char newline
= '\n';
1722 struct trace_entry
*entry
;
1723 struct trace_event
*event
;
1727 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1728 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1729 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1730 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
1733 event
= ftrace_find_event(entry
->type
);
1735 enum print_line_t ret
= event
->hex(iter
, 0);
1736 if (ret
!= TRACE_TYPE_HANDLED
)
1740 SEQ_PUT_FIELD_RET(s
, newline
);
1742 return TRACE_TYPE_HANDLED
;
1745 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
1747 struct trace_seq
*s
= &iter
->seq
;
1748 struct trace_entry
*entry
;
1749 struct trace_event
*event
;
1753 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1754 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1755 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
1756 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
1759 event
= ftrace_find_event(entry
->type
);
1760 return event
? event
->binary(iter
, 0) : TRACE_TYPE_HANDLED
;
1763 static int trace_empty(struct trace_iterator
*iter
)
1767 /* If we are looking at one CPU buffer, only check that one */
1768 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
1769 cpu
= iter
->cpu_file
;
1770 if (iter
->buffer_iter
[cpu
]) {
1771 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1774 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1780 for_each_tracing_cpu(cpu
) {
1781 if (iter
->buffer_iter
[cpu
]) {
1782 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1785 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1793 static enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
1795 enum print_line_t ret
;
1797 if (iter
->trace
&& iter
->trace
->print_line
) {
1798 ret
= iter
->trace
->print_line(iter
);
1799 if (ret
!= TRACE_TYPE_UNHANDLED
)
1803 if (iter
->ent
->type
== TRACE_BPRINT
&&
1804 trace_flags
& TRACE_ITER_PRINTK
&&
1805 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
1806 return trace_print_bprintk_msg_only(iter
);
1808 if (iter
->ent
->type
== TRACE_PRINT
&&
1809 trace_flags
& TRACE_ITER_PRINTK
&&
1810 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
1811 return trace_print_printk_msg_only(iter
);
1813 if (trace_flags
& TRACE_ITER_BIN
)
1814 return print_bin_fmt(iter
);
1816 if (trace_flags
& TRACE_ITER_HEX
)
1817 return print_hex_fmt(iter
);
1819 if (trace_flags
& TRACE_ITER_RAW
)
1820 return print_raw_fmt(iter
);
1822 return print_trace_fmt(iter
);
1825 static int s_show(struct seq_file
*m
, void *v
)
1827 struct trace_iterator
*iter
= v
;
1829 if (iter
->ent
== NULL
) {
1831 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
1834 if (iter
->trace
&& iter
->trace
->print_header
)
1835 iter
->trace
->print_header(m
);
1836 else if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1837 /* print nothing if the buffers are empty */
1838 if (trace_empty(iter
))
1840 print_trace_header(m
, iter
);
1841 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1842 print_lat_help_header(m
);
1844 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1845 print_func_help_header(m
);
1848 print_trace_line(iter
);
1849 trace_print_seq(m
, &iter
->seq
);
1855 static struct seq_operations tracer_seq_ops
= {
1862 static struct trace_iterator
*
1863 __tracing_open(struct inode
*inode
, struct file
*file
)
1865 long cpu_file
= (long) inode
->i_private
;
1866 void *fail_ret
= ERR_PTR(-ENOMEM
);
1867 struct trace_iterator
*iter
;
1871 if (tracing_disabled
)
1872 return ERR_PTR(-ENODEV
);
1874 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
1876 return ERR_PTR(-ENOMEM
);
1879 * We make a copy of the current tracer to avoid concurrent
1880 * changes on it while we are reading.
1882 mutex_lock(&trace_types_lock
);
1883 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
1888 *iter
->trace
= *current_trace
;
1890 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
1893 cpumask_clear(iter
->started
);
1895 if (current_trace
&& current_trace
->print_max
)
1898 iter
->tr
= &global_trace
;
1900 mutex_init(&iter
->mutex
);
1901 iter
->cpu_file
= cpu_file
;
1903 /* Notify the tracer early; before we stop tracing. */
1904 if (iter
->trace
&& iter
->trace
->open
)
1905 iter
->trace
->open(iter
);
1907 /* Annotate start of buffers if we had overruns */
1908 if (ring_buffer_overruns(iter
->tr
->buffer
))
1909 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
1911 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
1912 for_each_tracing_cpu(cpu
) {
1914 iter
->buffer_iter
[cpu
] =
1915 ring_buffer_read_start(iter
->tr
->buffer
, cpu
);
1918 cpu
= iter
->cpu_file
;
1919 iter
->buffer_iter
[cpu
] =
1920 ring_buffer_read_start(iter
->tr
->buffer
, cpu
);
1923 /* TODO stop tracer */
1924 ret
= seq_open(file
, &tracer_seq_ops
);
1926 fail_ret
= ERR_PTR(ret
);
1930 m
= file
->private_data
;
1933 /* stop the trace while dumping */
1936 mutex_unlock(&trace_types_lock
);
1941 for_each_tracing_cpu(cpu
) {
1942 if (iter
->buffer_iter
[cpu
])
1943 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
1945 free_cpumask_var(iter
->started
);
1947 mutex_unlock(&trace_types_lock
);
1954 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
1956 if (tracing_disabled
)
1959 filp
->private_data
= inode
->i_private
;
1963 static int tracing_release(struct inode
*inode
, struct file
*file
)
1965 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
1966 struct trace_iterator
*iter
;
1969 if (!(file
->f_mode
& FMODE_READ
))
1974 mutex_lock(&trace_types_lock
);
1975 for_each_tracing_cpu(cpu
) {
1976 if (iter
->buffer_iter
[cpu
])
1977 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
1980 if (iter
->trace
&& iter
->trace
->close
)
1981 iter
->trace
->close(iter
);
1983 /* reenable tracing if it was previously enabled */
1985 mutex_unlock(&trace_types_lock
);
1987 seq_release(inode
, file
);
1988 mutex_destroy(&iter
->mutex
);
1989 free_cpumask_var(iter
->started
);
1995 static int tracing_open(struct inode
*inode
, struct file
*file
)
1997 struct trace_iterator
*iter
;
2000 /* If this file was open for write, then erase contents */
2001 if ((file
->f_mode
& FMODE_WRITE
) &&
2002 !(file
->f_flags
& O_APPEND
)) {
2003 long cpu
= (long) inode
->i_private
;
2005 if (cpu
== TRACE_PIPE_ALL_CPU
)
2006 tracing_reset_online_cpus(&global_trace
);
2008 tracing_reset(&global_trace
, cpu
);
2011 if (file
->f_mode
& FMODE_READ
) {
2012 iter
= __tracing_open(inode
, file
);
2014 ret
= PTR_ERR(iter
);
2015 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2016 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2022 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2024 struct tracer
*t
= m
->private;
2036 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2038 struct tracer
*t
= m
->private;
2041 mutex_lock(&trace_types_lock
);
2042 for (; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2048 static void t_stop(struct seq_file
*m
, void *p
)
2050 mutex_unlock(&trace_types_lock
);
2053 static int t_show(struct seq_file
*m
, void *v
)
2055 struct tracer
*t
= v
;
2060 seq_printf(m
, "%s", t
->name
);
2069 static struct seq_operations show_traces_seq_ops
= {
2076 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2080 if (tracing_disabled
)
2083 ret
= seq_open(file
, &show_traces_seq_ops
);
2085 struct seq_file
*m
= file
->private_data
;
2086 m
->private = trace_types
;
2093 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2094 size_t count
, loff_t
*ppos
)
2099 static const struct file_operations tracing_fops
= {
2100 .open
= tracing_open
,
2102 .write
= tracing_write_stub
,
2103 .llseek
= seq_lseek
,
2104 .release
= tracing_release
,
2107 static const struct file_operations show_traces_fops
= {
2108 .open
= show_traces_open
,
2110 .release
= seq_release
,
2114 * Only trace on a CPU if the bitmask is set:
2116 static cpumask_var_t tracing_cpumask
;
2119 * The tracer itself will not take this lock, but still we want
2120 * to provide a consistent cpumask to user-space:
2122 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2125 * Temporary storage for the character representation of the
2126 * CPU bitmask (and one more byte for the newline):
2128 static char mask_str
[NR_CPUS
+ 1];
2131 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2132 size_t count
, loff_t
*ppos
)
2136 mutex_lock(&tracing_cpumask_update_lock
);
2138 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2139 if (count
- len
< 2) {
2143 len
+= sprintf(mask_str
+ len
, "\n");
2144 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2147 mutex_unlock(&tracing_cpumask_update_lock
);
2153 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2154 size_t count
, loff_t
*ppos
)
2157 cpumask_var_t tracing_cpumask_new
;
2159 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2162 mutex_lock(&tracing_cpumask_update_lock
);
2163 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2167 local_irq_disable();
2168 __raw_spin_lock(&ftrace_max_lock
);
2169 for_each_tracing_cpu(cpu
) {
2171 * Increase/decrease the disabled counter if we are
2172 * about to flip a bit in the cpumask:
2174 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2175 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2176 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2178 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2179 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2180 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2183 __raw_spin_unlock(&ftrace_max_lock
);
2186 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2188 mutex_unlock(&tracing_cpumask_update_lock
);
2189 free_cpumask_var(tracing_cpumask_new
);
2194 mutex_unlock(&tracing_cpumask_update_lock
);
2195 free_cpumask_var(tracing_cpumask
);
2200 static const struct file_operations tracing_cpumask_fops
= {
2201 .open
= tracing_open_generic
,
2202 .read
= tracing_cpumask_read
,
2203 .write
= tracing_cpumask_write
,
2207 tracing_trace_options_read(struct file
*filp
, char __user
*ubuf
,
2208 size_t cnt
, loff_t
*ppos
)
2210 struct tracer_opt
*trace_opts
;
2218 /* calculate max size */
2219 for (i
= 0; trace_options
[i
]; i
++) {
2220 len
+= strlen(trace_options
[i
]);
2221 len
+= 3; /* "no" and newline */
2224 mutex_lock(&trace_types_lock
);
2225 tracer_flags
= current_trace
->flags
->val
;
2226 trace_opts
= current_trace
->flags
->opts
;
2229 * Increase the size with names of options specific
2230 * of the current tracer.
2232 for (i
= 0; trace_opts
[i
].name
; i
++) {
2233 len
+= strlen(trace_opts
[i
].name
);
2234 len
+= 3; /* "no" and newline */
2237 /* +2 for \n and \0 */
2238 buf
= kmalloc(len
+ 2, GFP_KERNEL
);
2240 mutex_unlock(&trace_types_lock
);
2244 for (i
= 0; trace_options
[i
]; i
++) {
2245 if (trace_flags
& (1 << i
))
2246 r
+= sprintf(buf
+ r
, "%s\n", trace_options
[i
]);
2248 r
+= sprintf(buf
+ r
, "no%s\n", trace_options
[i
]);
2251 for (i
= 0; trace_opts
[i
].name
; i
++) {
2252 if (tracer_flags
& trace_opts
[i
].bit
)
2253 r
+= sprintf(buf
+ r
, "%s\n",
2254 trace_opts
[i
].name
);
2256 r
+= sprintf(buf
+ r
, "no%s\n",
2257 trace_opts
[i
].name
);
2259 mutex_unlock(&trace_types_lock
);
2261 WARN_ON(r
>= len
+ 2);
2263 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2269 /* Try to assign a tracer specific option */
2270 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2272 struct tracer_flags
*trace_flags
= trace
->flags
;
2273 struct tracer_opt
*opts
= NULL
;
2277 for (i
= 0; trace_flags
->opts
[i
].name
; i
++) {
2278 opts
= &trace_flags
->opts
[i
];
2279 len
= strlen(opts
->name
);
2281 if (strncmp(cmp
, opts
->name
, len
) == 0) {
2282 ret
= trace
->set_flag(trace_flags
->val
,
2288 if (!trace_flags
->opts
[i
].name
)
2291 /* Refused to handle */
2296 trace_flags
->val
&= ~opts
->bit
;
2298 trace_flags
->val
|= opts
->bit
;
2303 static void set_tracer_flags(unsigned int mask
, int enabled
)
2305 /* do nothing if flag is already set */
2306 if (!!(trace_flags
& mask
) == !!enabled
)
2310 trace_flags
|= mask
;
2312 trace_flags
&= ~mask
;
2314 if (mask
== TRACE_ITER_GLOBAL_CLK
) {
2318 func
= trace_clock_global
;
2320 func
= trace_clock_local
;
2322 mutex_lock(&trace_types_lock
);
2323 ring_buffer_set_clock(global_trace
.buffer
, func
);
2326 ring_buffer_set_clock(max_tr
.buffer
, func
);
2327 mutex_unlock(&trace_types_lock
);
2332 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2333 size_t cnt
, loff_t
*ppos
)
2341 if (cnt
>= sizeof(buf
))
2344 if (copy_from_user(&buf
, ubuf
, cnt
))
2349 if (strncmp(buf
, "no", 2) == 0) {
2354 for (i
= 0; trace_options
[i
]; i
++) {
2355 int len
= strlen(trace_options
[i
]);
2357 if (strncmp(cmp
, trace_options
[i
], len
) == 0) {
2358 set_tracer_flags(1 << i
, !neg
);
2363 /* If no option could be set, test the specific tracer options */
2364 if (!trace_options
[i
]) {
2365 mutex_lock(&trace_types_lock
);
2366 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2367 mutex_unlock(&trace_types_lock
);
2377 static const struct file_operations tracing_iter_fops
= {
2378 .open
= tracing_open_generic
,
2379 .read
= tracing_trace_options_read
,
2380 .write
= tracing_trace_options_write
,
2383 static const char readme_msg
[] =
2384 "tracing mini-HOWTO:\n\n"
2386 "# mount -t debugfs nodev /debug\n\n"
2387 "# cat /debug/tracing/available_tracers\n"
2388 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2389 "# cat /debug/tracing/current_tracer\n"
2391 "# echo sched_switch > /debug/tracing/current_tracer\n"
2392 "# cat /debug/tracing/current_tracer\n"
2394 "# cat /debug/tracing/trace_options\n"
2395 "noprint-parent nosym-offset nosym-addr noverbose\n"
2396 "# echo print-parent > /debug/tracing/trace_options\n"
2397 "# echo 1 > /debug/tracing/tracing_enabled\n"
2398 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2399 "echo 0 > /debug/tracing/tracing_enabled\n"
2403 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2404 size_t cnt
, loff_t
*ppos
)
2406 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2407 readme_msg
, strlen(readme_msg
));
2410 static const struct file_operations tracing_readme_fops
= {
2411 .open
= tracing_open_generic
,
2412 .read
= tracing_readme_read
,
2416 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2417 size_t cnt
, loff_t
*ppos
)
2422 r
= sprintf(buf
, "%u\n", tracer_enabled
);
2423 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2427 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2428 size_t cnt
, loff_t
*ppos
)
2430 struct trace_array
*tr
= filp
->private_data
;
2435 if (cnt
>= sizeof(buf
))
2438 if (copy_from_user(&buf
, ubuf
, cnt
))
2443 ret
= strict_strtoul(buf
, 10, &val
);
2449 mutex_lock(&trace_types_lock
);
2450 if (tracer_enabled
^ val
) {
2453 if (current_trace
->start
)
2454 current_trace
->start(tr
);
2459 if (current_trace
->stop
)
2460 current_trace
->stop(tr
);
2463 mutex_unlock(&trace_types_lock
);
2471 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2472 size_t cnt
, loff_t
*ppos
)
2474 char buf
[max_tracer_type_len
+2];
2477 mutex_lock(&trace_types_lock
);
2479 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2481 r
= sprintf(buf
, "\n");
2482 mutex_unlock(&trace_types_lock
);
2484 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2487 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
2489 tracing_reset_online_cpus(tr
);
2493 static int tracing_resize_ring_buffer(unsigned long size
)
2498 * If kernel or user changes the size of the ring buffer
2499 * we use the size that was given, and we can forget about
2500 * expanding it later.
2502 ring_buffer_expanded
= 1;
2504 ret
= ring_buffer_resize(global_trace
.buffer
, size
);
2508 ret
= ring_buffer_resize(max_tr
.buffer
, size
);
2512 r
= ring_buffer_resize(global_trace
.buffer
,
2513 global_trace
.entries
);
2516 * AARGH! We are left with different
2517 * size max buffer!!!!
2518 * The max buffer is our "snapshot" buffer.
2519 * When a tracer needs a snapshot (one of the
2520 * latency tracers), it swaps the max buffer
2521 * with the saved snap shot. We succeeded to
2522 * update the size of the main buffer, but failed to
2523 * update the size of the max buffer. But when we tried
2524 * to reset the main buffer to the original size, we
2525 * failed there too. This is very unlikely to
2526 * happen, but if it does, warn and kill all
2530 tracing_disabled
= 1;
2535 global_trace
.entries
= size
;
2541 * tracing_update_buffers - used by tracing facility to expand ring buffers
2543 * To save on memory when the tracing is never used on a system with it
2544 * configured in. The ring buffers are set to a minimum size. But once
2545 * a user starts to use the tracing facility, then they need to grow
2546 * to their default size.
2548 * This function is to be called when a tracer is about to be used.
2550 int tracing_update_buffers(void)
2554 mutex_lock(&trace_types_lock
);
2555 if (!ring_buffer_expanded
)
2556 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2557 mutex_unlock(&trace_types_lock
);
2562 struct trace_option_dentry
;
2564 static struct trace_option_dentry
*
2565 create_trace_option_files(struct tracer
*tracer
);
2568 destroy_trace_option_files(struct trace_option_dentry
*topts
);
2570 static int tracing_set_tracer(const char *buf
)
2572 static struct trace_option_dentry
*topts
;
2573 struct trace_array
*tr
= &global_trace
;
2577 mutex_lock(&trace_types_lock
);
2579 if (!ring_buffer_expanded
) {
2580 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2586 for (t
= trace_types
; t
; t
= t
->next
) {
2587 if (strcmp(t
->name
, buf
) == 0)
2594 if (t
== current_trace
)
2597 trace_branch_disable();
2598 if (current_trace
&& current_trace
->reset
)
2599 current_trace
->reset(tr
);
2601 destroy_trace_option_files(topts
);
2605 topts
= create_trace_option_files(current_trace
);
2608 ret
= tracer_init(t
, tr
);
2613 trace_branch_enable(tr
);
2615 mutex_unlock(&trace_types_lock
);
2621 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2622 size_t cnt
, loff_t
*ppos
)
2624 char buf
[max_tracer_type_len
+1];
2631 if (cnt
> max_tracer_type_len
)
2632 cnt
= max_tracer_type_len
;
2634 if (copy_from_user(&buf
, ubuf
, cnt
))
2639 /* strip ending whitespace. */
2640 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2643 err
= tracing_set_tracer(buf
);
2653 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2654 size_t cnt
, loff_t
*ppos
)
2656 unsigned long *ptr
= filp
->private_data
;
2660 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
2661 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2662 if (r
> sizeof(buf
))
2664 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2668 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2669 size_t cnt
, loff_t
*ppos
)
2671 unsigned long *ptr
= filp
->private_data
;
2676 if (cnt
>= sizeof(buf
))
2679 if (copy_from_user(&buf
, ubuf
, cnt
))
2684 ret
= strict_strtoul(buf
, 10, &val
);
2693 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2695 long cpu_file
= (long) inode
->i_private
;
2696 struct trace_iterator
*iter
;
2699 if (tracing_disabled
)
2702 mutex_lock(&trace_types_lock
);
2704 /* We only allow one reader per cpu */
2705 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
2706 if (!cpumask_empty(tracing_reader_cpumask
)) {
2710 cpumask_setall(tracing_reader_cpumask
);
2712 if (!cpumask_test_cpu(cpu_file
, tracing_reader_cpumask
))
2713 cpumask_set_cpu(cpu_file
, tracing_reader_cpumask
);
2720 /* create a buffer to store the information to pass to userspace */
2721 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2728 * We make a copy of the current tracer to avoid concurrent
2729 * changes on it while we are reading.
2731 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2737 *iter
->trace
= *current_trace
;
2739 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
2744 /* trace pipe does not show start of buffer */
2745 cpumask_setall(iter
->started
);
2747 iter
->cpu_file
= cpu_file
;
2748 iter
->tr
= &global_trace
;
2749 mutex_init(&iter
->mutex
);
2750 filp
->private_data
= iter
;
2752 if (iter
->trace
->pipe_open
)
2753 iter
->trace
->pipe_open(iter
);
2756 mutex_unlock(&trace_types_lock
);
2762 mutex_unlock(&trace_types_lock
);
2766 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
2768 struct trace_iterator
*iter
= file
->private_data
;
2770 mutex_lock(&trace_types_lock
);
2772 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
)
2773 cpumask_clear(tracing_reader_cpumask
);
2775 cpumask_clear_cpu(iter
->cpu_file
, tracing_reader_cpumask
);
2777 mutex_unlock(&trace_types_lock
);
2779 free_cpumask_var(iter
->started
);
2780 mutex_destroy(&iter
->mutex
);
2788 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
2790 struct trace_iterator
*iter
= filp
->private_data
;
2792 if (trace_flags
& TRACE_ITER_BLOCK
) {
2794 * Always select as readable when in blocking mode
2796 return POLLIN
| POLLRDNORM
;
2798 if (!trace_empty(iter
))
2799 return POLLIN
| POLLRDNORM
;
2800 poll_wait(filp
, &trace_wait
, poll_table
);
2801 if (!trace_empty(iter
))
2802 return POLLIN
| POLLRDNORM
;
2809 void default_wait_pipe(struct trace_iterator
*iter
)
2813 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
2815 if (trace_empty(iter
))
2818 finish_wait(&trace_wait
, &wait
);
2822 * This is a make-shift waitqueue.
2823 * A tracer might use this callback on some rare cases:
2825 * 1) the current tracer might hold the runqueue lock when it wakes up
2826 * a reader, hence a deadlock (sched, function, and function graph tracers)
2827 * 2) the function tracers, trace all functions, we don't want
2828 * the overhead of calling wake_up and friends
2829 * (and tracing them too)
2831 * Anyway, this is really very primitive wakeup.
2833 void poll_wait_pipe(struct trace_iterator
*iter
)
2835 set_current_state(TASK_INTERRUPTIBLE
);
2836 /* sleep for 100 msecs, and try again. */
2837 schedule_timeout(HZ
/ 10);
2840 /* Must be called with trace_types_lock mutex held. */
2841 static int tracing_wait_pipe(struct file
*filp
)
2843 struct trace_iterator
*iter
= filp
->private_data
;
2845 while (trace_empty(iter
)) {
2847 if ((filp
->f_flags
& O_NONBLOCK
)) {
2851 mutex_unlock(&iter
->mutex
);
2853 iter
->trace
->wait_pipe(iter
);
2855 mutex_lock(&iter
->mutex
);
2857 if (signal_pending(current
))
2861 * We block until we read something and tracing is disabled.
2862 * We still block if tracing is disabled, but we have never
2863 * read anything. This allows a user to cat this file, and
2864 * then enable tracing. But after we have read something,
2865 * we give an EOF when tracing is again disabled.
2867 * iter->pos will be 0 if we haven't read anything.
2869 if (!tracer_enabled
&& iter
->pos
)
2880 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
2881 size_t cnt
, loff_t
*ppos
)
2883 struct trace_iterator
*iter
= filp
->private_data
;
2884 static struct tracer
*old_tracer
;
2887 /* return any leftover data */
2888 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
2892 trace_seq_init(&iter
->seq
);
2894 /* copy the tracer to avoid using a global lock all around */
2895 mutex_lock(&trace_types_lock
);
2896 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
2897 old_tracer
= current_trace
;
2898 *iter
->trace
= *current_trace
;
2900 mutex_unlock(&trace_types_lock
);
2903 * Avoid more than one consumer on a single file descriptor
2904 * This is just a matter of traces coherency, the ring buffer itself
2907 mutex_lock(&iter
->mutex
);
2908 if (iter
->trace
->read
) {
2909 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
2915 sret
= tracing_wait_pipe(filp
);
2919 /* stop when tracing is finished */
2920 if (trace_empty(iter
)) {
2925 if (cnt
>= PAGE_SIZE
)
2926 cnt
= PAGE_SIZE
- 1;
2928 /* reset all but tr, trace, and overruns */
2929 memset(&iter
->seq
, 0,
2930 sizeof(struct trace_iterator
) -
2931 offsetof(struct trace_iterator
, seq
));
2934 while (find_next_entry_inc(iter
) != NULL
) {
2935 enum print_line_t ret
;
2936 int len
= iter
->seq
.len
;
2938 ret
= print_trace_line(iter
);
2939 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
2940 /* don't print partial lines */
2941 iter
->seq
.len
= len
;
2944 if (ret
!= TRACE_TYPE_NO_CONSUME
)
2945 trace_consume(iter
);
2947 if (iter
->seq
.len
>= cnt
)
2951 /* Now copy what we have to the user */
2952 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
2953 if (iter
->seq
.readpos
>= iter
->seq
.len
)
2954 trace_seq_init(&iter
->seq
);
2957 * If there was nothing to send to user, inspite of consuming trace
2958 * entries, go back to wait for more entries.
2964 mutex_unlock(&iter
->mutex
);
2969 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
2970 struct pipe_buffer
*buf
)
2972 __free_page(buf
->page
);
2975 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
2978 __free_page(spd
->pages
[idx
]);
2981 static struct pipe_buf_operations tracing_pipe_buf_ops
= {
2983 .map
= generic_pipe_buf_map
,
2984 .unmap
= generic_pipe_buf_unmap
,
2985 .confirm
= generic_pipe_buf_confirm
,
2986 .release
= tracing_pipe_buf_release
,
2987 .steal
= generic_pipe_buf_steal
,
2988 .get
= generic_pipe_buf_get
,
2992 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
2997 /* Seq buffer is page-sized, exactly what we need. */
2999 count
= iter
->seq
.len
;
3000 ret
= print_trace_line(iter
);
3001 count
= iter
->seq
.len
- count
;
3004 iter
->seq
.len
-= count
;
3007 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3008 iter
->seq
.len
-= count
;
3012 trace_consume(iter
);
3014 if (!find_next_entry_inc(iter
)) {
3024 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3026 struct pipe_inode_info
*pipe
,
3030 struct page
*pages
[PIPE_BUFFERS
];
3031 struct partial_page partial
[PIPE_BUFFERS
];
3032 struct trace_iterator
*iter
= filp
->private_data
;
3033 struct splice_pipe_desc spd
= {
3036 .nr_pages
= 0, /* This gets updated below. */
3038 .ops
= &tracing_pipe_buf_ops
,
3039 .spd_release
= tracing_spd_release_pipe
,
3041 static struct tracer
*old_tracer
;
3046 /* copy the tracer to avoid using a global lock all around */
3047 mutex_lock(&trace_types_lock
);
3048 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3049 old_tracer
= current_trace
;
3050 *iter
->trace
= *current_trace
;
3052 mutex_unlock(&trace_types_lock
);
3054 mutex_lock(&iter
->mutex
);
3056 if (iter
->trace
->splice_read
) {
3057 ret
= iter
->trace
->splice_read(iter
, filp
,
3058 ppos
, pipe
, len
, flags
);
3063 ret
= tracing_wait_pipe(filp
);
3067 if (!iter
->ent
&& !find_next_entry_inc(iter
)) {
3072 /* Fill as many pages as possible. */
3073 for (i
= 0, rem
= len
; i
< PIPE_BUFFERS
&& rem
; i
++) {
3074 pages
[i
] = alloc_page(GFP_KERNEL
);
3078 rem
= tracing_fill_pipe_page(rem
, iter
);
3080 /* Copy the data into the page, so we can start over. */
3081 ret
= trace_seq_to_buffer(&iter
->seq
,
3082 page_address(pages
[i
]),
3085 __free_page(pages
[i
]);
3088 partial
[i
].offset
= 0;
3089 partial
[i
].len
= iter
->seq
.len
;
3091 trace_seq_init(&iter
->seq
);
3094 mutex_unlock(&iter
->mutex
);
3098 return splice_to_pipe(pipe
, &spd
);
3101 mutex_unlock(&iter
->mutex
);
3107 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3108 size_t cnt
, loff_t
*ppos
)
3110 struct trace_array
*tr
= filp
->private_data
;
3114 mutex_lock(&trace_types_lock
);
3115 if (!ring_buffer_expanded
)
3116 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3118 trace_buf_size
>> 10);
3120 r
= sprintf(buf
, "%lu\n", tr
->entries
>> 10);
3121 mutex_unlock(&trace_types_lock
);
3123 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3127 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3128 size_t cnt
, loff_t
*ppos
)
3134 if (cnt
>= sizeof(buf
))
3137 if (copy_from_user(&buf
, ubuf
, cnt
))
3142 ret
= strict_strtoul(buf
, 10, &val
);
3146 /* must have at least 1 entry */
3150 mutex_lock(&trace_types_lock
);
3154 /* disable all cpu buffers */
3155 for_each_tracing_cpu(cpu
) {
3156 if (global_trace
.data
[cpu
])
3157 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
3158 if (max_tr
.data
[cpu
])
3159 atomic_inc(&max_tr
.data
[cpu
]->disabled
);
3162 /* value is in KB */
3165 if (val
!= global_trace
.entries
) {
3166 ret
= tracing_resize_ring_buffer(val
);
3175 /* If check pages failed, return ENOMEM */
3176 if (tracing_disabled
)
3179 for_each_tracing_cpu(cpu
) {
3180 if (global_trace
.data
[cpu
])
3181 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
3182 if (max_tr
.data
[cpu
])
3183 atomic_dec(&max_tr
.data
[cpu
]->disabled
);
3187 max_tr
.entries
= global_trace
.entries
;
3188 mutex_unlock(&trace_types_lock
);
3193 static int mark_printk(const char *fmt
, ...)
3197 va_start(args
, fmt
);
3198 ret
= trace_vprintk(0, fmt
, args
);
3204 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3205 size_t cnt
, loff_t
*fpos
)
3210 if (tracing_disabled
)
3213 if (cnt
> TRACE_BUF_SIZE
)
3214 cnt
= TRACE_BUF_SIZE
;
3216 buf
= kmalloc(cnt
+ 1, GFP_KERNEL
);
3220 if (copy_from_user(buf
, ubuf
, cnt
)) {
3225 /* Cut from the first nil or newline. */
3227 end
= strchr(buf
, '\n');
3231 cnt
= mark_printk("%s\n", buf
);
3238 static const struct file_operations tracing_max_lat_fops
= {
3239 .open
= tracing_open_generic
,
3240 .read
= tracing_max_lat_read
,
3241 .write
= tracing_max_lat_write
,
3244 static const struct file_operations tracing_ctrl_fops
= {
3245 .open
= tracing_open_generic
,
3246 .read
= tracing_ctrl_read
,
3247 .write
= tracing_ctrl_write
,
3250 static const struct file_operations set_tracer_fops
= {
3251 .open
= tracing_open_generic
,
3252 .read
= tracing_set_trace_read
,
3253 .write
= tracing_set_trace_write
,
3256 static const struct file_operations tracing_pipe_fops
= {
3257 .open
= tracing_open_pipe
,
3258 .poll
= tracing_poll_pipe
,
3259 .read
= tracing_read_pipe
,
3260 .splice_read
= tracing_splice_read_pipe
,
3261 .release
= tracing_release_pipe
,
3264 static const struct file_operations tracing_entries_fops
= {
3265 .open
= tracing_open_generic
,
3266 .read
= tracing_entries_read
,
3267 .write
= tracing_entries_write
,
3270 static const struct file_operations tracing_mark_fops
= {
3271 .open
= tracing_open_generic
,
3272 .write
= tracing_mark_write
,
3275 struct ftrace_buffer_info
{
3276 struct trace_array
*tr
;
3282 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
3284 int cpu
= (int)(long)inode
->i_private
;
3285 struct ftrace_buffer_info
*info
;
3287 if (tracing_disabled
)
3290 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3294 info
->tr
= &global_trace
;
3297 /* Force reading ring buffer for first read */
3298 info
->read
= (unsigned int)-1;
3300 filp
->private_data
= info
;
3302 return nonseekable_open(inode
, filp
);
3306 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
3307 size_t count
, loff_t
*ppos
)
3309 struct ftrace_buffer_info
*info
= filp
->private_data
;
3318 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
);
3322 /* Do we have previous read data to read? */
3323 if (info
->read
< PAGE_SIZE
)
3328 ret
= ring_buffer_read_page(info
->tr
->buffer
,
3335 pos
= ring_buffer_page_len(info
->spare
);
3337 if (pos
< PAGE_SIZE
)
3338 memset(info
->spare
+ pos
, 0, PAGE_SIZE
- pos
);
3341 size
= PAGE_SIZE
- info
->read
;
3345 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
3356 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
3358 struct ftrace_buffer_info
*info
= file
->private_data
;
3361 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
3368 struct ring_buffer
*buffer
;
3373 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
3374 struct pipe_buffer
*buf
)
3376 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3381 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3386 static int buffer_pipe_buf_steal(struct pipe_inode_info
*pipe
,
3387 struct pipe_buffer
*buf
)
3392 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
3393 struct pipe_buffer
*buf
)
3395 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3400 /* Pipe buffer operations for a buffer. */
3401 static struct pipe_buf_operations buffer_pipe_buf_ops
= {
3403 .map
= generic_pipe_buf_map
,
3404 .unmap
= generic_pipe_buf_unmap
,
3405 .confirm
= generic_pipe_buf_confirm
,
3406 .release
= buffer_pipe_buf_release
,
3407 .steal
= buffer_pipe_buf_steal
,
3408 .get
= buffer_pipe_buf_get
,
3412 * Callback from splice_to_pipe(), if we need to release some pages
3413 * at the end of the spd in case we error'ed out in filling the pipe.
3415 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
3417 struct buffer_ref
*ref
=
3418 (struct buffer_ref
*)spd
->partial
[i
].private;
3423 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3425 spd
->partial
[i
].private = 0;
3429 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
3430 struct pipe_inode_info
*pipe
, size_t len
,
3433 struct ftrace_buffer_info
*info
= file
->private_data
;
3434 struct partial_page partial
[PIPE_BUFFERS
];
3435 struct page
*pages
[PIPE_BUFFERS
];
3436 struct splice_pipe_desc spd
= {
3440 .ops
= &buffer_pipe_buf_ops
,
3441 .spd_release
= buffer_spd_release
,
3443 struct buffer_ref
*ref
;
3447 if (*ppos
& (PAGE_SIZE
- 1)) {
3448 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3452 if (len
& (PAGE_SIZE
- 1)) {
3453 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3454 if (len
< PAGE_SIZE
)
3459 for (i
= 0; i
< PIPE_BUFFERS
&& len
; i
++, len
-= PAGE_SIZE
) {
3463 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
3467 ref
->buffer
= info
->tr
->buffer
;
3468 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
);
3474 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
3477 ring_buffer_free_read_page(ref
->buffer
,
3484 * zero out any left over data, this is going to
3487 size
= ring_buffer_page_len(ref
->page
);
3488 if (size
< PAGE_SIZE
)
3489 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
3491 page
= virt_to_page(ref
->page
);
3493 spd
.pages
[i
] = page
;
3494 spd
.partial
[i
].len
= PAGE_SIZE
;
3495 spd
.partial
[i
].offset
= 0;
3496 spd
.partial
[i
].private = (unsigned long)ref
;
3503 /* did we read anything? */
3504 if (!spd
.nr_pages
) {
3505 if (flags
& SPLICE_F_NONBLOCK
)
3513 ret
= splice_to_pipe(pipe
, &spd
);
3518 static const struct file_operations tracing_buffers_fops
= {
3519 .open
= tracing_buffers_open
,
3520 .read
= tracing_buffers_read
,
3521 .release
= tracing_buffers_release
,
3522 .splice_read
= tracing_buffers_splice_read
,
3523 .llseek
= no_llseek
,
3526 #ifdef CONFIG_DYNAMIC_FTRACE
3528 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
3534 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
3535 size_t cnt
, loff_t
*ppos
)
3537 static char ftrace_dyn_info_buffer
[1024];
3538 static DEFINE_MUTEX(dyn_info_mutex
);
3539 unsigned long *p
= filp
->private_data
;
3540 char *buf
= ftrace_dyn_info_buffer
;
3541 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
3544 mutex_lock(&dyn_info_mutex
);
3545 r
= sprintf(buf
, "%ld ", *p
);
3547 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
3550 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3552 mutex_unlock(&dyn_info_mutex
);
3557 static const struct file_operations tracing_dyn_info_fops
= {
3558 .open
= tracing_open_generic
,
3559 .read
= tracing_read_dyn_info
,
3563 static struct dentry
*d_tracer
;
3565 struct dentry
*tracing_init_dentry(void)
3572 if (!debugfs_initialized())
3575 d_tracer
= debugfs_create_dir("tracing", NULL
);
3577 if (!d_tracer
&& !once
) {
3579 pr_warning("Could not create debugfs directory 'tracing'\n");
3586 static struct dentry
*d_percpu
;
3588 struct dentry
*tracing_dentry_percpu(void)
3591 struct dentry
*d_tracer
;
3596 d_tracer
= tracing_init_dentry();
3601 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
3603 if (!d_percpu
&& !once
) {
3605 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3612 static void tracing_init_debugfs_percpu(long cpu
)
3614 struct dentry
*d_percpu
= tracing_dentry_percpu();
3615 struct dentry
*d_cpu
;
3616 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3619 if (cpu
> 999 || cpu
< 0)
3622 sprintf(cpu_dir
, "cpu%ld", cpu
);
3623 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
3625 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
3629 /* per cpu trace_pipe */
3630 trace_create_file("trace_pipe", 0444, d_cpu
,
3631 (void *) cpu
, &tracing_pipe_fops
);
3634 trace_create_file("trace", 0644, d_cpu
,
3635 (void *) cpu
, &tracing_fops
);
3637 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
3638 (void *) cpu
, &tracing_buffers_fops
);
3641 #ifdef CONFIG_FTRACE_SELFTEST
3642 /* Let selftest have access to static functions in this file */
3643 #include "trace_selftest.c"
3646 struct trace_option_dentry
{
3647 struct tracer_opt
*opt
;
3648 struct tracer_flags
*flags
;
3649 struct dentry
*entry
;
3653 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
3656 struct trace_option_dentry
*topt
= filp
->private_data
;
3659 if (topt
->flags
->val
& topt
->opt
->bit
)
3664 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
3668 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
3671 struct trace_option_dentry
*topt
= filp
->private_data
;
3676 if (cnt
>= sizeof(buf
))
3679 if (copy_from_user(&buf
, ubuf
, cnt
))
3684 ret
= strict_strtoul(buf
, 10, &val
);
3691 /* do nothing if already cleared */
3692 if (!(topt
->flags
->val
& topt
->opt
->bit
))
3695 mutex_lock(&trace_types_lock
);
3696 if (current_trace
->set_flag
)
3697 ret
= current_trace
->set_flag(topt
->flags
->val
,
3699 mutex_unlock(&trace_types_lock
);
3702 topt
->flags
->val
&= ~topt
->opt
->bit
;
3705 /* do nothing if already set */
3706 if (topt
->flags
->val
& topt
->opt
->bit
)
3709 mutex_lock(&trace_types_lock
);
3710 if (current_trace
->set_flag
)
3711 ret
= current_trace
->set_flag(topt
->flags
->val
,
3713 mutex_unlock(&trace_types_lock
);
3716 topt
->flags
->val
|= topt
->opt
->bit
;
3729 static const struct file_operations trace_options_fops
= {
3730 .open
= tracing_open_generic
,
3731 .read
= trace_options_read
,
3732 .write
= trace_options_write
,
3736 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
3739 long index
= (long)filp
->private_data
;
3742 if (trace_flags
& (1 << index
))
3747 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
3751 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
3754 long index
= (long)filp
->private_data
;
3759 if (cnt
>= sizeof(buf
))
3762 if (copy_from_user(&buf
, ubuf
, cnt
))
3767 ret
= strict_strtoul(buf
, 10, &val
);
3773 trace_flags
&= ~(1 << index
);
3776 trace_flags
|= 1 << index
;
3788 static const struct file_operations trace_options_core_fops
= {
3789 .open
= tracing_open_generic
,
3790 .read
= trace_options_core_read
,
3791 .write
= trace_options_core_write
,
3794 struct dentry
*trace_create_file(const char *name
,
3796 struct dentry
*parent
,
3798 const struct file_operations
*fops
)
3802 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
3804 pr_warning("Could not create debugfs '%s' entry\n", name
);
3810 static struct dentry
*trace_options_init_dentry(void)
3812 struct dentry
*d_tracer
;
3813 static struct dentry
*t_options
;
3818 d_tracer
= tracing_init_dentry();
3822 t_options
= debugfs_create_dir("options", d_tracer
);
3824 pr_warning("Could not create debugfs directory 'options'\n");
3832 create_trace_option_file(struct trace_option_dentry
*topt
,
3833 struct tracer_flags
*flags
,
3834 struct tracer_opt
*opt
)
3836 struct dentry
*t_options
;
3838 t_options
= trace_options_init_dentry();
3842 topt
->flags
= flags
;
3845 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
3846 &trace_options_fops
);
3850 static struct trace_option_dentry
*
3851 create_trace_option_files(struct tracer
*tracer
)
3853 struct trace_option_dentry
*topts
;
3854 struct tracer_flags
*flags
;
3855 struct tracer_opt
*opts
;
3861 flags
= tracer
->flags
;
3863 if (!flags
|| !flags
->opts
)
3868 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
3871 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
3875 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
3876 create_trace_option_file(&topts
[cnt
], flags
,
3883 destroy_trace_option_files(struct trace_option_dentry
*topts
)
3890 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
3891 if (topts
[cnt
].entry
)
3892 debugfs_remove(topts
[cnt
].entry
);
3898 static struct dentry
*
3899 create_trace_option_core_file(const char *option
, long index
)
3901 struct dentry
*t_options
;
3903 t_options
= trace_options_init_dentry();
3907 return trace_create_file(option
, 0644, t_options
, (void *)index
,
3908 &trace_options_core_fops
);
3911 static __init
void create_trace_options_dir(void)
3913 struct dentry
*t_options
;
3916 t_options
= trace_options_init_dentry();
3920 for (i
= 0; trace_options
[i
]; i
++)
3921 create_trace_option_core_file(trace_options
[i
], i
);
3924 static __init
int tracer_init_debugfs(void)
3926 struct dentry
*d_tracer
;
3929 d_tracer
= tracing_init_dentry();
3931 trace_create_file("tracing_enabled", 0644, d_tracer
,
3932 &global_trace
, &tracing_ctrl_fops
);
3934 trace_create_file("trace_options", 0644, d_tracer
,
3935 NULL
, &tracing_iter_fops
);
3937 trace_create_file("tracing_cpumask", 0644, d_tracer
,
3938 NULL
, &tracing_cpumask_fops
);
3940 trace_create_file("trace", 0644, d_tracer
,
3941 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
3943 trace_create_file("available_tracers", 0444, d_tracer
,
3944 &global_trace
, &show_traces_fops
);
3946 trace_create_file("current_tracer", 0444, d_tracer
,
3947 &global_trace
, &set_tracer_fops
);
3949 trace_create_file("tracing_max_latency", 0644, d_tracer
,
3950 &tracing_max_latency
, &tracing_max_lat_fops
);
3952 trace_create_file("tracing_thresh", 0644, d_tracer
,
3953 &tracing_thresh
, &tracing_max_lat_fops
);
3955 trace_create_file("README", 0644, d_tracer
,
3956 NULL
, &tracing_readme_fops
);
3958 trace_create_file("trace_pipe", 0444, d_tracer
,
3959 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
3961 trace_create_file("buffer_size_kb", 0644, d_tracer
,
3962 &global_trace
, &tracing_entries_fops
);
3964 trace_create_file("trace_marker", 0220, d_tracer
,
3965 NULL
, &tracing_mark_fops
);
3967 #ifdef CONFIG_DYNAMIC_FTRACE
3968 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
3969 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
3971 #ifdef CONFIG_SYSPROF_TRACER
3972 init_tracer_sysprof_debugfs(d_tracer
);
3975 create_trace_options_dir();
3977 for_each_tracing_cpu(cpu
)
3978 tracing_init_debugfs_percpu(cpu
);
3983 static int trace_panic_handler(struct notifier_block
*this,
3984 unsigned long event
, void *unused
)
3986 if (ftrace_dump_on_oops
)
3991 static struct notifier_block trace_panic_notifier
= {
3992 .notifier_call
= trace_panic_handler
,
3994 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
3997 static int trace_die_handler(struct notifier_block
*self
,
4003 if (ftrace_dump_on_oops
)
4012 static struct notifier_block trace_die_notifier
= {
4013 .notifier_call
= trace_die_handler
,
4018 * printk is set to max of 1024, we really don't need it that big.
4019 * Nothing should be printing 1000 characters anyway.
4021 #define TRACE_MAX_PRINT 1000
4024 * Define here KERN_TRACE so that we have one place to modify
4025 * it if we decide to change what log level the ftrace dump
4028 #define KERN_TRACE KERN_EMERG
4031 trace_printk_seq(struct trace_seq
*s
)
4033 /* Probably should print a warning here. */
4037 /* should be zero ended, but we are paranoid. */
4038 s
->buffer
[s
->len
] = 0;
4040 printk(KERN_TRACE
"%s", s
->buffer
);
4045 static void __ftrace_dump(bool disable_tracing
)
4047 static DEFINE_SPINLOCK(ftrace_dump_lock
);
4048 /* use static because iter can be a bit big for the stack */
4049 static struct trace_iterator iter
;
4050 unsigned int old_userobj
;
4051 static int dump_ran
;
4052 unsigned long flags
;
4056 spin_lock_irqsave(&ftrace_dump_lock
, flags
);
4064 if (disable_tracing
)
4067 for_each_tracing_cpu(cpu
) {
4068 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
4071 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
4073 /* don't look at user memory in panic mode */
4074 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
4076 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
4078 /* Simulate the iterator */
4079 iter
.tr
= &global_trace
;
4080 iter
.trace
= current_trace
;
4081 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4084 * We need to stop all tracing on all CPUS to read the
4085 * the next buffer. This is a bit expensive, but is
4086 * not done often. We fill all what we can read,
4087 * and then release the locks again.
4090 while (!trace_empty(&iter
)) {
4093 printk(KERN_TRACE
"---------------------------------\n");
4097 /* reset all but tr, trace, and overruns */
4098 memset(&iter
.seq
, 0,
4099 sizeof(struct trace_iterator
) -
4100 offsetof(struct trace_iterator
, seq
));
4101 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
4104 if (find_next_entry_inc(&iter
) != NULL
) {
4105 print_trace_line(&iter
);
4106 trace_consume(&iter
);
4109 trace_printk_seq(&iter
.seq
);
4113 printk(KERN_TRACE
" (ftrace buffer empty)\n");
4115 printk(KERN_TRACE
"---------------------------------\n");
4117 /* Re-enable tracing if requested */
4118 if (!disable_tracing
) {
4119 trace_flags
|= old_userobj
;
4121 for_each_tracing_cpu(cpu
) {
4122 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
4128 spin_unlock_irqrestore(&ftrace_dump_lock
, flags
);
4131 /* By default: disable tracing after the dump */
4132 void ftrace_dump(void)
4134 __ftrace_dump(true);
4137 __init
static int tracer_alloc_buffers(void)
4139 struct trace_array_cpu
*data
;
4144 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
4147 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
4148 goto out_free_buffer_mask
;
4150 if (!alloc_cpumask_var(&tracing_reader_cpumask
, GFP_KERNEL
))
4151 goto out_free_tracing_cpumask
;
4153 /* To save memory, keep the ring buffer size to its minimum */
4154 if (ring_buffer_expanded
)
4155 ring_buf_size
= trace_buf_size
;
4159 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
4160 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
4161 cpumask_clear(tracing_reader_cpumask
);
4163 /* TODO: make the number of buffers hot pluggable with CPUS */
4164 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
,
4165 TRACE_BUFFER_FLAGS
);
4166 if (!global_trace
.buffer
) {
4167 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
4169 goto out_free_cpumask
;
4171 global_trace
.entries
= ring_buffer_size(global_trace
.buffer
);
4174 #ifdef CONFIG_TRACER_MAX_TRACE
4175 max_tr
.buffer
= ring_buffer_alloc(ring_buf_size
,
4176 TRACE_BUFFER_FLAGS
);
4177 if (!max_tr
.buffer
) {
4178 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
4180 ring_buffer_free(global_trace
.buffer
);
4181 goto out_free_cpumask
;
4183 max_tr
.entries
= ring_buffer_size(max_tr
.buffer
);
4184 WARN_ON(max_tr
.entries
!= global_trace
.entries
);
4187 /* Allocate the first page for all buffers */
4188 for_each_tracing_cpu(i
) {
4189 data
= global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
4190 max_tr
.data
[i
] = &per_cpu(max_data
, i
);
4193 trace_init_cmdlines();
4195 register_tracer(&nop_trace
);
4196 current_trace
= &nop_trace
;
4197 #ifdef CONFIG_BOOT_TRACER
4198 register_tracer(&boot_tracer
);
4200 /* All seems OK, enable tracing */
4201 tracing_disabled
= 0;
4203 atomic_notifier_chain_register(&panic_notifier_list
,
4204 &trace_panic_notifier
);
4206 register_die_notifier(&trace_die_notifier
);
4211 free_cpumask_var(tracing_reader_cpumask
);
4212 out_free_tracing_cpumask
:
4213 free_cpumask_var(tracing_cpumask
);
4214 out_free_buffer_mask
:
4215 free_cpumask_var(tracing_buffer_mask
);
4220 __init
static int clear_boot_tracer(void)
4223 * The default tracer at boot buffer is an init section.
4224 * This function is called in lateinit. If we did not
4225 * find the boot tracer, then clear it out, to prevent
4226 * later registration from accessing the buffer that is
4227 * about to be freed.
4229 if (!default_bootup_tracer
)
4232 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
4233 default_bootup_tracer
);
4234 default_bootup_tracer
= NULL
;
4239 early_initcall(tracer_alloc_buffers
);
4240 fs_initcall(tracer_init_debugfs
);
4241 late_initcall(clear_boot_tracer
);