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/ctype.h>
34 #include <linux/init.h>
35 #include <linux/poll.h>
36 #include <linux/gfp.h>
40 #include "trace_output.h"
42 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
44 unsigned long __read_mostly tracing_max_latency
;
45 unsigned long __read_mostly tracing_thresh
;
48 * On boot up, the ring buffer is set to the minimum size, so that
49 * we do not waste memory on systems that are not using tracing.
51 static int ring_buffer_expanded
;
54 * We need to change this state when a selftest is running.
55 * A selftest will lurk into the ring-buffer to count the
56 * entries inserted during the selftest although some concurrent
57 * insertions into the ring-buffer such as trace_printk could occurred
58 * at the same time, giving false positive or negative results.
60 static bool __read_mostly tracing_selftest_running
;
63 * If a tracer is running, we do not want to run SELFTEST.
65 static bool __read_mostly tracing_selftest_disabled
;
67 /* For tracers that don't implement custom flags */
68 static struct tracer_opt dummy_tracer_opt
[] = {
72 static struct tracer_flags dummy_tracer_flags
= {
74 .opts
= dummy_tracer_opt
77 static int dummy_set_flag(u32 old_flags
, u32 bit
, int set
)
83 * Kill all tracing for good (never come back).
84 * It is initialized to 1 but will turn to zero if the initialization
85 * of the tracer is successful. But that is the only place that sets
88 static int tracing_disabled
= 1;
90 static DEFINE_PER_CPU(local_t
, ftrace_cpu_disabled
);
92 static inline void ftrace_disable_cpu(void)
95 local_inc(&__get_cpu_var(ftrace_cpu_disabled
));
98 static inline void ftrace_enable_cpu(void)
100 local_dec(&__get_cpu_var(ftrace_cpu_disabled
));
104 static cpumask_var_t __read_mostly tracing_buffer_mask
;
106 /* Define which cpu buffers are currently read in trace_pipe */
107 static cpumask_var_t tracing_reader_cpumask
;
109 #define for_each_tracing_cpu(cpu) \
110 for_each_cpu(cpu, tracing_buffer_mask)
113 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
115 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
116 * is set, then ftrace_dump is called. This will output the contents
117 * of the ftrace buffers to the console. This is very useful for
118 * capturing traces that lead to crashes and outputing it to a
121 * It is default off, but you can enable it with either specifying
122 * "ftrace_dump_on_oops" in the kernel command line, or setting
123 * /proc/sys/kernel/ftrace_dump_on_oops to true.
125 int ftrace_dump_on_oops
;
127 static int tracing_set_tracer(const char *buf
);
129 #define BOOTUP_TRACER_SIZE 100
130 static char bootup_tracer_buf
[BOOTUP_TRACER_SIZE
] __initdata
;
131 static char *default_bootup_tracer
;
133 static int __init
set_ftrace(char *str
)
135 strncpy(bootup_tracer_buf
, str
, BOOTUP_TRACER_SIZE
);
136 default_bootup_tracer
= bootup_tracer_buf
;
137 /* We are using ftrace early, expand it */
138 ring_buffer_expanded
= 1;
141 __setup("ftrace=", set_ftrace
);
143 static int __init
set_ftrace_dump_on_oops(char *str
)
145 ftrace_dump_on_oops
= 1;
148 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops
);
151 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
;
261 * trace_wake_up - wake up tasks waiting for trace input
263 * Simply wakes up any task that is blocked on the trace_wait
264 * queue. These is used with trace_poll for tasks polling the trace.
266 void trace_wake_up(void)
269 * The runqueue_is_locked() can fail, but this is the best we
272 if (!(trace_flags
& TRACE_ITER_BLOCK
) && !runqueue_is_locked())
273 wake_up(&trace_wait
);
276 static int __init
set_buf_size(char *str
)
278 unsigned long buf_size
;
283 ret
= strict_strtoul(str
, 0, &buf_size
);
284 /* nr_entries can not be zero */
285 if (ret
< 0 || buf_size
== 0)
287 trace_buf_size
= buf_size
;
290 __setup("trace_buf_size=", set_buf_size
);
292 unsigned long nsecs_to_usecs(unsigned long nsecs
)
297 /* These must match the bit postions in trace_iterator_flags */
298 static const char *trace_options
[] = {
324 * ftrace_max_lock is used to protect the swapping of buffers
325 * when taking a max snapshot. The buffers themselves are
326 * protected by per_cpu spinlocks. But the action of the swap
327 * needs its own lock.
329 * This is defined as a raw_spinlock_t in order to help
330 * with performance when lockdep debugging is enabled.
332 static raw_spinlock_t ftrace_max_lock
=
333 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
336 * Copy the new maximum trace into the separate maximum-trace
337 * structure. (this way the maximum trace is permanently saved,
338 * for later retrieval via /debugfs/tracing/latency_trace)
341 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
343 struct trace_array_cpu
*data
= tr
->data
[cpu
];
346 max_tr
.time_start
= data
->preempt_timestamp
;
348 data
= max_tr
.data
[cpu
];
349 data
->saved_latency
= tracing_max_latency
;
351 memcpy(data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
352 data
->pid
= tsk
->pid
;
353 data
->uid
= task_uid(tsk
);
354 data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
355 data
->policy
= tsk
->policy
;
356 data
->rt_priority
= tsk
->rt_priority
;
358 /* record this tasks comm */
359 tracing_record_cmdline(tsk
);
362 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
370 if (s
->len
<= s
->readpos
)
373 len
= s
->len
- s
->readpos
;
376 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
386 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
391 if (s
->len
<= s
->readpos
)
394 len
= s
->len
- s
->readpos
;
397 ret
= memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
406 trace_print_seq(struct seq_file
*m
, struct trace_seq
*s
)
408 int len
= s
->len
>= PAGE_SIZE
? PAGE_SIZE
- 1 : s
->len
;
411 seq_puts(m
, s
->buffer
);
417 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
419 * @tsk: the task with the latency
420 * @cpu: The cpu that initiated the trace.
422 * Flip the buffers between the @tr and the max_tr and record information
423 * about which task was the cause of this latency.
426 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
428 struct ring_buffer
*buf
= tr
->buffer
;
430 WARN_ON_ONCE(!irqs_disabled());
431 __raw_spin_lock(&ftrace_max_lock
);
433 tr
->buffer
= max_tr
.buffer
;
436 ftrace_disable_cpu();
437 ring_buffer_reset(tr
->buffer
);
440 __update_max_tr(tr
, tsk
, cpu
);
441 __raw_spin_unlock(&ftrace_max_lock
);
445 * update_max_tr_single - only copy one trace over, and reset the rest
447 * @tsk - task with the latency
448 * @cpu - the cpu of the buffer to copy.
450 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
453 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
457 WARN_ON_ONCE(!irqs_disabled());
458 __raw_spin_lock(&ftrace_max_lock
);
460 ftrace_disable_cpu();
462 ring_buffer_reset(max_tr
.buffer
);
463 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
467 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
);
469 __update_max_tr(tr
, tsk
, cpu
);
470 __raw_spin_unlock(&ftrace_max_lock
);
474 * register_tracer - register a tracer with the ftrace system.
475 * @type - the plugin for the tracer
477 * Register a new plugin tracer.
479 int register_tracer(struct tracer
*type
)
480 __releases(kernel_lock
)
481 __acquires(kernel_lock
)
488 pr_info("Tracer must have a name\n");
493 * When this gets called we hold the BKL which means that
494 * preemption is disabled. Various trace selftests however
495 * need to disable and enable preemption for successful tests.
496 * So we drop the BKL here and grab it after the tests again.
499 mutex_lock(&trace_types_lock
);
501 tracing_selftest_running
= true;
503 for (t
= trace_types
; t
; t
= t
->next
) {
504 if (strcmp(type
->name
, t
->name
) == 0) {
506 pr_info("Trace %s already registered\n",
514 type
->set_flag
= &dummy_set_flag
;
516 type
->flags
= &dummy_tracer_flags
;
518 if (!type
->flags
->opts
)
519 type
->flags
->opts
= dummy_tracer_opt
;
520 if (!type
->wait_pipe
)
521 type
->wait_pipe
= default_wait_pipe
;
524 #ifdef CONFIG_FTRACE_STARTUP_TEST
525 if (type
->selftest
&& !tracing_selftest_disabled
) {
526 struct tracer
*saved_tracer
= current_trace
;
527 struct trace_array
*tr
= &global_trace
;
531 * Run a selftest on this tracer.
532 * Here we reset the trace buffer, and set the current
533 * tracer to be this tracer. The tracer can then run some
534 * internal tracing to verify that everything is in order.
535 * If we fail, we do not register this tracer.
537 for_each_tracing_cpu(i
)
538 tracing_reset(tr
, i
);
540 current_trace
= type
;
541 /* the test is responsible for initializing and enabling */
542 pr_info("Testing tracer %s: ", type
->name
);
543 ret
= type
->selftest(type
, tr
);
544 /* the test is responsible for resetting too */
545 current_trace
= saved_tracer
;
547 printk(KERN_CONT
"FAILED!\n");
550 /* Only reset on passing, to avoid touching corrupted buffers */
551 for_each_tracing_cpu(i
)
552 tracing_reset(tr
, i
);
554 printk(KERN_CONT
"PASSED\n");
558 type
->next
= trace_types
;
560 len
= strlen(type
->name
);
561 if (len
> max_tracer_type_len
)
562 max_tracer_type_len
= len
;
565 tracing_selftest_running
= false;
566 mutex_unlock(&trace_types_lock
);
568 if (ret
|| !default_bootup_tracer
)
571 if (strncmp(default_bootup_tracer
, type
->name
, BOOTUP_TRACER_SIZE
))
574 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
575 /* Do we want this tracer to start on bootup? */
576 tracing_set_tracer(type
->name
);
577 default_bootup_tracer
= NULL
;
578 /* disable other selftests, since this will break it. */
579 tracing_selftest_disabled
= 1;
580 #ifdef CONFIG_FTRACE_STARTUP_TEST
581 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
590 void unregister_tracer(struct tracer
*type
)
595 mutex_lock(&trace_types_lock
);
596 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
600 pr_info("Trace %s not registered\n", type
->name
);
606 if (type
== current_trace
&& tracer_enabled
) {
609 if (current_trace
->stop
)
610 current_trace
->stop(&global_trace
);
611 current_trace
= &nop_trace
;
614 if (strlen(type
->name
) != max_tracer_type_len
)
617 max_tracer_type_len
= 0;
618 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
619 len
= strlen((*t
)->name
);
620 if (len
> max_tracer_type_len
)
621 max_tracer_type_len
= len
;
624 mutex_unlock(&trace_types_lock
);
627 void tracing_reset(struct trace_array
*tr
, int cpu
)
629 ftrace_disable_cpu();
630 ring_buffer_reset_cpu(tr
->buffer
, cpu
);
634 void tracing_reset_online_cpus(struct trace_array
*tr
)
638 tr
->time_start
= ftrace_now(tr
->cpu
);
640 for_each_online_cpu(cpu
)
641 tracing_reset(tr
, cpu
);
644 #define SAVED_CMDLINES 128
645 #define NO_CMDLINE_MAP UINT_MAX
646 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
647 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
648 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
649 static int cmdline_idx
;
650 static raw_spinlock_t trace_cmdline_lock
= __RAW_SPIN_LOCK_UNLOCKED
;
652 /* temporary disable recording */
653 static atomic_t trace_record_cmdline_disabled __read_mostly
;
655 static void trace_init_cmdlines(void)
657 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
658 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
662 static int trace_stop_count
;
663 static DEFINE_SPINLOCK(tracing_start_lock
);
666 * ftrace_off_permanent - disable all ftrace code permanently
668 * This should only be called when a serious anomally has
669 * been detected. This will turn off the function tracing,
670 * ring buffers, and other tracing utilites. It takes no
671 * locks and can be called from any context.
673 void ftrace_off_permanent(void)
675 tracing_disabled
= 1;
677 tracing_off_permanent();
681 * tracing_start - quick start of the tracer
683 * If tracing is enabled but was stopped by tracing_stop,
684 * this will start the tracer back up.
686 void tracing_start(void)
688 struct ring_buffer
*buffer
;
691 if (tracing_disabled
)
694 spin_lock_irqsave(&tracing_start_lock
, flags
);
695 if (--trace_stop_count
) {
696 if (trace_stop_count
< 0) {
697 /* Someone screwed up their debugging */
699 trace_stop_count
= 0;
705 buffer
= global_trace
.buffer
;
707 ring_buffer_record_enable(buffer
);
709 buffer
= max_tr
.buffer
;
711 ring_buffer_record_enable(buffer
);
715 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
719 * tracing_stop - quick stop of the tracer
721 * Light weight way to stop tracing. Use in conjunction with
724 void tracing_stop(void)
726 struct ring_buffer
*buffer
;
730 spin_lock_irqsave(&tracing_start_lock
, flags
);
731 if (trace_stop_count
++)
734 buffer
= global_trace
.buffer
;
736 ring_buffer_record_disable(buffer
);
738 buffer
= max_tr
.buffer
;
740 ring_buffer_record_disable(buffer
);
743 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
746 void trace_stop_cmdline_recording(void);
748 static void trace_save_cmdline(struct task_struct
*tsk
)
752 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
756 * It's not the end of the world if we don't get
757 * the lock, but we also don't want to spin
758 * nor do we want to disable interrupts,
759 * so if we miss here, then better luck next time.
761 if (!__raw_spin_trylock(&trace_cmdline_lock
))
764 idx
= map_pid_to_cmdline
[tsk
->pid
];
765 if (idx
== NO_CMDLINE_MAP
) {
766 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
769 * Check whether the cmdline buffer at idx has a pid
770 * mapped. We are going to overwrite that entry so we
771 * need to clear the map_pid_to_cmdline. Otherwise we
772 * would read the new comm for the old pid.
774 pid
= map_cmdline_to_pid
[idx
];
775 if (pid
!= NO_CMDLINE_MAP
)
776 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
778 map_cmdline_to_pid
[idx
] = tsk
->pid
;
779 map_pid_to_cmdline
[tsk
->pid
] = idx
;
784 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
786 __raw_spin_unlock(&trace_cmdline_lock
);
789 void trace_find_cmdline(int pid
, char comm
[])
794 strcpy(comm
, "<idle>");
798 if (pid
> PID_MAX_DEFAULT
) {
799 strcpy(comm
, "<...>");
803 __raw_spin_lock(&trace_cmdline_lock
);
804 map
= map_pid_to_cmdline
[pid
];
805 if (map
!= NO_CMDLINE_MAP
)
806 strcpy(comm
, saved_cmdlines
[map
]);
808 strcpy(comm
, "<...>");
810 __raw_spin_unlock(&trace_cmdline_lock
);
813 void tracing_record_cmdline(struct task_struct
*tsk
)
815 if (atomic_read(&trace_record_cmdline_disabled
) || !tracer_enabled
||
819 trace_save_cmdline(tsk
);
823 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
826 struct task_struct
*tsk
= current
;
828 entry
->preempt_count
= pc
& 0xff;
829 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
830 entry
->tgid
= (tsk
) ? tsk
->tgid
: 0;
832 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
833 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
835 TRACE_FLAG_IRQS_NOSUPPORT
|
837 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
838 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
839 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
842 struct ring_buffer_event
*trace_buffer_lock_reserve(struct trace_array
*tr
,
845 unsigned long flags
, int pc
)
847 struct ring_buffer_event
*event
;
849 event
= ring_buffer_lock_reserve(tr
->buffer
, len
);
851 struct trace_entry
*ent
= ring_buffer_event_data(event
);
853 tracing_generic_entry_update(ent
, flags
, pc
);
859 static void ftrace_trace_stack(struct trace_array
*tr
,
860 unsigned long flags
, int skip
, int pc
);
861 static void ftrace_trace_userstack(struct trace_array
*tr
,
862 unsigned long flags
, int pc
);
864 static inline void __trace_buffer_unlock_commit(struct trace_array
*tr
,
865 struct ring_buffer_event
*event
,
866 unsigned long flags
, int pc
,
869 ring_buffer_unlock_commit(tr
->buffer
, event
);
871 ftrace_trace_stack(tr
, flags
, 6, pc
);
872 ftrace_trace_userstack(tr
, flags
, pc
);
878 void trace_buffer_unlock_commit(struct trace_array
*tr
,
879 struct ring_buffer_event
*event
,
880 unsigned long flags
, int pc
)
882 __trace_buffer_unlock_commit(tr
, event
, flags
, pc
, 1);
885 struct ring_buffer_event
*
886 trace_current_buffer_lock_reserve(unsigned char type
, unsigned long len
,
887 unsigned long flags
, int pc
)
889 return trace_buffer_lock_reserve(&global_trace
,
890 type
, len
, flags
, pc
);
893 void trace_current_buffer_unlock_commit(struct ring_buffer_event
*event
,
894 unsigned long flags
, int pc
)
896 return __trace_buffer_unlock_commit(&global_trace
, event
, flags
, pc
, 1);
899 void trace_nowake_buffer_unlock_commit(struct ring_buffer_event
*event
,
900 unsigned long flags
, int pc
)
902 return __trace_buffer_unlock_commit(&global_trace
, event
, flags
, pc
, 0);
906 trace_function(struct trace_array
*tr
,
907 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
910 struct ring_buffer_event
*event
;
911 struct ftrace_entry
*entry
;
913 /* If we are reading the ring buffer, don't trace */
914 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
917 event
= trace_buffer_lock_reserve(tr
, TRACE_FN
, sizeof(*entry
),
921 entry
= ring_buffer_event_data(event
);
923 entry
->parent_ip
= parent_ip
;
924 ring_buffer_unlock_commit(tr
->buffer
, event
);
927 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
928 static int __trace_graph_entry(struct trace_array
*tr
,
929 struct ftrace_graph_ent
*trace
,
933 struct ring_buffer_event
*event
;
934 struct ftrace_graph_ent_entry
*entry
;
936 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
939 event
= trace_buffer_lock_reserve(&global_trace
, TRACE_GRAPH_ENT
,
940 sizeof(*entry
), flags
, pc
);
943 entry
= ring_buffer_event_data(event
);
944 entry
->graph_ent
= *trace
;
945 ring_buffer_unlock_commit(global_trace
.buffer
, event
);
950 static void __trace_graph_return(struct trace_array
*tr
,
951 struct ftrace_graph_ret
*trace
,
955 struct ring_buffer_event
*event
;
956 struct ftrace_graph_ret_entry
*entry
;
958 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
961 event
= trace_buffer_lock_reserve(&global_trace
, TRACE_GRAPH_RET
,
962 sizeof(*entry
), flags
, pc
);
965 entry
= ring_buffer_event_data(event
);
967 ring_buffer_unlock_commit(global_trace
.buffer
, event
);
972 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
973 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
976 if (likely(!atomic_read(&data
->disabled
)))
977 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
980 static void __ftrace_trace_stack(struct trace_array
*tr
,
984 #ifdef CONFIG_STACKTRACE
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 ring_buffer_unlock_commit(tr
->buffer
, event
);
1006 static void ftrace_trace_stack(struct trace_array
*tr
,
1007 unsigned long flags
,
1010 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1013 __ftrace_trace_stack(tr
, flags
, skip
, pc
);
1016 void __trace_stack(struct trace_array
*tr
,
1017 unsigned long flags
,
1020 __ftrace_trace_stack(tr
, flags
, skip
, pc
);
1023 static void ftrace_trace_userstack(struct trace_array
*tr
,
1024 unsigned long flags
, int pc
)
1026 #ifdef CONFIG_STACKTRACE
1027 struct ring_buffer_event
*event
;
1028 struct userstack_entry
*entry
;
1029 struct stack_trace trace
;
1031 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1034 event
= trace_buffer_lock_reserve(tr
, TRACE_USER_STACK
,
1035 sizeof(*entry
), flags
, pc
);
1038 entry
= ring_buffer_event_data(event
);
1040 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1042 trace
.nr_entries
= 0;
1043 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1045 trace
.entries
= entry
->caller
;
1047 save_stack_trace_user(&trace
);
1048 ring_buffer_unlock_commit(tr
->buffer
, event
);
1053 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1055 ftrace_trace_userstack(tr
, flags
, preempt_count());
1060 ftrace_trace_special(void *__tr
,
1061 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
,
1064 struct ring_buffer_event
*event
;
1065 struct trace_array
*tr
= __tr
;
1066 struct special_entry
*entry
;
1068 event
= trace_buffer_lock_reserve(tr
, TRACE_SPECIAL
,
1069 sizeof(*entry
), 0, pc
);
1072 entry
= ring_buffer_event_data(event
);
1076 trace_buffer_unlock_commit(tr
, event
, 0, pc
);
1080 __trace_special(void *__tr
, void *__data
,
1081 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1083 ftrace_trace_special(__tr
, arg1
, arg2
, arg3
, preempt_count());
1087 tracing_sched_switch_trace(struct trace_array
*tr
,
1088 struct task_struct
*prev
,
1089 struct task_struct
*next
,
1090 unsigned long flags
, int pc
)
1092 struct ring_buffer_event
*event
;
1093 struct ctx_switch_entry
*entry
;
1095 event
= trace_buffer_lock_reserve(tr
, TRACE_CTX
,
1096 sizeof(*entry
), flags
, pc
);
1099 entry
= ring_buffer_event_data(event
);
1100 entry
->prev_pid
= prev
->pid
;
1101 entry
->prev_prio
= prev
->prio
;
1102 entry
->prev_state
= prev
->state
;
1103 entry
->next_pid
= next
->pid
;
1104 entry
->next_prio
= next
->prio
;
1105 entry
->next_state
= next
->state
;
1106 entry
->next_cpu
= task_cpu(next
);
1107 trace_buffer_unlock_commit(tr
, event
, flags
, pc
);
1111 tracing_sched_wakeup_trace(struct trace_array
*tr
,
1112 struct task_struct
*wakee
,
1113 struct task_struct
*curr
,
1114 unsigned long flags
, int pc
)
1116 struct ring_buffer_event
*event
;
1117 struct ctx_switch_entry
*entry
;
1119 event
= trace_buffer_lock_reserve(tr
, TRACE_WAKE
,
1120 sizeof(*entry
), flags
, pc
);
1123 entry
= ring_buffer_event_data(event
);
1124 entry
->prev_pid
= curr
->pid
;
1125 entry
->prev_prio
= curr
->prio
;
1126 entry
->prev_state
= curr
->state
;
1127 entry
->next_pid
= wakee
->pid
;
1128 entry
->next_prio
= wakee
->prio
;
1129 entry
->next_state
= wakee
->state
;
1130 entry
->next_cpu
= task_cpu(wakee
);
1132 ring_buffer_unlock_commit(tr
->buffer
, event
);
1133 ftrace_trace_stack(tr
, flags
, 6, pc
);
1134 ftrace_trace_userstack(tr
, flags
, pc
);
1138 ftrace_special(unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1140 struct trace_array
*tr
= &global_trace
;
1141 struct trace_array_cpu
*data
;
1142 unsigned long flags
;
1146 if (tracing_disabled
)
1149 pc
= preempt_count();
1150 local_irq_save(flags
);
1151 cpu
= raw_smp_processor_id();
1152 data
= tr
->data
[cpu
];
1154 if (likely(atomic_inc_return(&data
->disabled
) == 1))
1155 ftrace_trace_special(tr
, arg1
, arg2
, arg3
, pc
);
1157 atomic_dec(&data
->disabled
);
1158 local_irq_restore(flags
);
1161 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1162 int trace_graph_entry(struct ftrace_graph_ent
*trace
)
1164 struct trace_array
*tr
= &global_trace
;
1165 struct trace_array_cpu
*data
;
1166 unsigned long flags
;
1172 if (!ftrace_trace_task(current
))
1175 if (!ftrace_graph_addr(trace
->func
))
1178 local_irq_save(flags
);
1179 cpu
= raw_smp_processor_id();
1180 data
= tr
->data
[cpu
];
1181 disabled
= atomic_inc_return(&data
->disabled
);
1182 if (likely(disabled
== 1)) {
1183 pc
= preempt_count();
1184 ret
= __trace_graph_entry(tr
, trace
, flags
, pc
);
1188 /* Only do the atomic if it is not already set */
1189 if (!test_tsk_trace_graph(current
))
1190 set_tsk_trace_graph(current
);
1192 atomic_dec(&data
->disabled
);
1193 local_irq_restore(flags
);
1198 void trace_graph_return(struct ftrace_graph_ret
*trace
)
1200 struct trace_array
*tr
= &global_trace
;
1201 struct trace_array_cpu
*data
;
1202 unsigned long flags
;
1207 local_irq_save(flags
);
1208 cpu
= raw_smp_processor_id();
1209 data
= tr
->data
[cpu
];
1210 disabled
= atomic_inc_return(&data
->disabled
);
1211 if (likely(disabled
== 1)) {
1212 pc
= preempt_count();
1213 __trace_graph_return(tr
, trace
, flags
, pc
);
1216 clear_tsk_trace_graph(current
);
1217 atomic_dec(&data
->disabled
);
1218 local_irq_restore(flags
);
1220 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1224 * trace_vbprintk - write binary msg to tracing buffer
1227 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1229 static raw_spinlock_t trace_buf_lock
=
1230 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
1231 static u32 trace_buf
[TRACE_BUF_SIZE
];
1233 struct ring_buffer_event
*event
;
1234 struct trace_array
*tr
= &global_trace
;
1235 struct trace_array_cpu
*data
;
1236 struct bprint_entry
*entry
;
1237 unsigned long flags
;
1239 int cpu
, len
= 0, size
, pc
;
1241 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1244 /* Don't pollute graph traces with trace_vprintk internals */
1245 pause_graph_tracing();
1247 pc
= preempt_count();
1248 resched
= ftrace_preempt_disable();
1249 cpu
= raw_smp_processor_id();
1250 data
= tr
->data
[cpu
];
1252 if (unlikely(atomic_read(&data
->disabled
)))
1255 /* Lockdep uses trace_printk for lock tracing */
1256 local_irq_save(flags
);
1257 __raw_spin_lock(&trace_buf_lock
);
1258 len
= vbin_printf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1260 if (len
> TRACE_BUF_SIZE
|| len
< 0)
1263 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1264 event
= trace_buffer_lock_reserve(tr
, TRACE_BPRINT
, size
, flags
, pc
);
1267 entry
= ring_buffer_event_data(event
);
1271 memcpy(entry
->buf
, trace_buf
, sizeof(u32
) * len
);
1272 ring_buffer_unlock_commit(tr
->buffer
, event
);
1275 __raw_spin_unlock(&trace_buf_lock
);
1276 local_irq_restore(flags
);
1279 ftrace_preempt_enable(resched
);
1280 unpause_graph_tracing();
1284 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1286 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1288 static raw_spinlock_t trace_buf_lock
= __RAW_SPIN_LOCK_UNLOCKED
;
1289 static char trace_buf
[TRACE_BUF_SIZE
];
1291 struct ring_buffer_event
*event
;
1292 struct trace_array
*tr
= &global_trace
;
1293 struct trace_array_cpu
*data
;
1294 int cpu
, len
= 0, size
, pc
;
1295 struct print_entry
*entry
;
1296 unsigned long irq_flags
;
1298 if (tracing_disabled
|| tracing_selftest_running
)
1301 pc
= preempt_count();
1302 preempt_disable_notrace();
1303 cpu
= raw_smp_processor_id();
1304 data
= tr
->data
[cpu
];
1306 if (unlikely(atomic_read(&data
->disabled
)))
1309 pause_graph_tracing();
1310 raw_local_irq_save(irq_flags
);
1311 __raw_spin_lock(&trace_buf_lock
);
1312 len
= vsnprintf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1314 len
= min(len
, TRACE_BUF_SIZE
-1);
1317 size
= sizeof(*entry
) + len
+ 1;
1318 event
= trace_buffer_lock_reserve(tr
, TRACE_PRINT
, size
, irq_flags
, pc
);
1321 entry
= ring_buffer_event_data(event
);
1324 memcpy(&entry
->buf
, trace_buf
, len
);
1325 entry
->buf
[len
] = 0;
1326 ring_buffer_unlock_commit(tr
->buffer
, event
);
1329 __raw_spin_unlock(&trace_buf_lock
);
1330 raw_local_irq_restore(irq_flags
);
1331 unpause_graph_tracing();
1333 preempt_enable_notrace();
1337 EXPORT_SYMBOL_GPL(trace_vprintk
);
1339 enum trace_file_type
{
1340 TRACE_FILE_LAT_FMT
= 1,
1341 TRACE_FILE_ANNOTATE
= 2,
1344 static void trace_iterator_increment(struct trace_iterator
*iter
)
1346 /* Don't allow ftrace to trace into the ring buffers */
1347 ftrace_disable_cpu();
1350 if (iter
->buffer_iter
[iter
->cpu
])
1351 ring_buffer_read(iter
->buffer_iter
[iter
->cpu
], NULL
);
1353 ftrace_enable_cpu();
1356 static struct trace_entry
*
1357 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
)
1359 struct ring_buffer_event
*event
;
1360 struct ring_buffer_iter
*buf_iter
= iter
->buffer_iter
[cpu
];
1362 /* Don't allow ftrace to trace into the ring buffers */
1363 ftrace_disable_cpu();
1366 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1368 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
);
1370 ftrace_enable_cpu();
1372 return event
? ring_buffer_event_data(event
) : NULL
;
1375 static struct trace_entry
*
1376 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
, u64
*ent_ts
)
1378 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1379 struct trace_entry
*ent
, *next
= NULL
;
1380 int cpu_file
= iter
->cpu_file
;
1381 u64 next_ts
= 0, ts
;
1386 * If we are in a per_cpu trace file, don't bother by iterating over
1387 * all cpu and peek directly.
1389 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1390 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1392 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
);
1394 *ent_cpu
= cpu_file
;
1399 for_each_tracing_cpu(cpu
) {
1401 if (ring_buffer_empty_cpu(buffer
, cpu
))
1404 ent
= peek_next_entry(iter
, cpu
, &ts
);
1407 * Pick the entry with the smallest timestamp:
1409 if (ent
&& (!next
|| ts
< next_ts
)) {
1417 *ent_cpu
= next_cpu
;
1425 /* Find the next real entry, without updating the iterator itself */
1426 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1427 int *ent_cpu
, u64
*ent_ts
)
1429 return __find_next_entry(iter
, ent_cpu
, ent_ts
);
1432 /* Find the next real entry, and increment the iterator to the next entry */
1433 static void *find_next_entry_inc(struct trace_iterator
*iter
)
1435 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
, &iter
->ts
);
1438 trace_iterator_increment(iter
);
1440 return iter
->ent
? iter
: NULL
;
1443 static void trace_consume(struct trace_iterator
*iter
)
1445 /* Don't allow ftrace to trace into the ring buffers */
1446 ftrace_disable_cpu();
1447 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
);
1448 ftrace_enable_cpu();
1451 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1453 struct trace_iterator
*iter
= m
->private;
1459 /* can't go backwards */
1464 ent
= find_next_entry_inc(iter
);
1468 while (ent
&& iter
->idx
< i
)
1469 ent
= find_next_entry_inc(iter
);
1477 * No necessary locking here. The worst thing which can
1478 * happen is loosing events consumed at the same time
1479 * by a trace_pipe reader.
1480 * Other than that, we don't risk to crash the ring buffer
1481 * because it serializes the readers.
1483 * The current tracer is copied to avoid a global locking
1486 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1488 struct trace_iterator
*iter
= m
->private;
1489 static struct tracer
*old_tracer
;
1490 int cpu_file
= iter
->cpu_file
;
1495 /* copy the tracer to avoid using a global lock all around */
1496 mutex_lock(&trace_types_lock
);
1497 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1498 old_tracer
= current_trace
;
1499 *iter
->trace
= *current_trace
;
1501 mutex_unlock(&trace_types_lock
);
1503 atomic_inc(&trace_record_cmdline_disabled
);
1505 if (*pos
!= iter
->pos
) {
1510 ftrace_disable_cpu();
1512 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1513 for_each_tracing_cpu(cpu
)
1514 ring_buffer_iter_reset(iter
->buffer_iter
[cpu
]);
1516 ring_buffer_iter_reset(iter
->buffer_iter
[cpu_file
]);
1519 ftrace_enable_cpu();
1521 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1526 p
= s_next(m
, p
, &l
);
1532 static void s_stop(struct seq_file
*m
, void *p
)
1534 atomic_dec(&trace_record_cmdline_disabled
);
1537 static void print_lat_help_header(struct seq_file
*m
)
1539 seq_puts(m
, "# _------=> CPU# \n");
1540 seq_puts(m
, "# / _-----=> irqs-off \n");
1541 seq_puts(m
, "# | / _----=> need-resched \n");
1542 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1543 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1544 seq_puts(m
, "# |||| / \n");
1545 seq_puts(m
, "# ||||| delay \n");
1546 seq_puts(m
, "# cmd pid ||||| time | caller \n");
1547 seq_puts(m
, "# \\ / ||||| \\ | / \n");
1550 static void print_func_help_header(struct seq_file
*m
)
1552 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1553 seq_puts(m
, "# | | | | |\n");
1558 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1560 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1561 struct trace_array
*tr
= iter
->tr
;
1562 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1563 struct tracer
*type
= current_trace
;
1564 unsigned long total
;
1565 unsigned long entries
;
1566 const char *name
= "preemption";
1571 entries
= ring_buffer_entries(iter
->tr
->buffer
);
1573 ring_buffer_overruns(iter
->tr
->buffer
);
1575 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
1577 seq_puts(m
, "# -----------------------------------"
1578 "---------------------------------\n");
1579 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1580 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1581 nsecs_to_usecs(data
->saved_latency
),
1585 #if defined(CONFIG_PREEMPT_NONE)
1587 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1589 #elif defined(CONFIG_PREEMPT)
1594 /* These are reserved for later use */
1597 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1601 seq_puts(m
, "# -----------------\n");
1602 seq_printf(m
, "# | task: %.16s-%d "
1603 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1604 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1605 data
->policy
, data
->rt_priority
);
1606 seq_puts(m
, "# -----------------\n");
1608 if (data
->critical_start
) {
1609 seq_puts(m
, "# => started at: ");
1610 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1611 trace_print_seq(m
, &iter
->seq
);
1612 seq_puts(m
, "\n# => ended at: ");
1613 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1614 trace_print_seq(m
, &iter
->seq
);
1621 static void test_cpu_buff_start(struct trace_iterator
*iter
)
1623 struct trace_seq
*s
= &iter
->seq
;
1625 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
1628 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
1631 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
1634 cpumask_set_cpu(iter
->cpu
, iter
->started
);
1635 trace_seq_printf(s
, "##### CPU %u buffer started ####\n", iter
->cpu
);
1638 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
1640 struct trace_seq
*s
= &iter
->seq
;
1641 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1642 struct trace_entry
*entry
;
1643 struct trace_event
*event
;
1647 test_cpu_buff_start(iter
);
1649 event
= ftrace_find_event(entry
->type
);
1651 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1652 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1653 if (!trace_print_lat_context(iter
))
1656 if (!trace_print_context(iter
))
1662 return event
->trace(iter
, sym_flags
);
1664 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
1667 return TRACE_TYPE_HANDLED
;
1669 return TRACE_TYPE_PARTIAL_LINE
;
1672 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
1674 struct trace_seq
*s
= &iter
->seq
;
1675 struct trace_entry
*entry
;
1676 struct trace_event
*event
;
1680 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1681 if (!trace_seq_printf(s
, "%d %d %llu ",
1682 entry
->pid
, iter
->cpu
, iter
->ts
))
1686 event
= ftrace_find_event(entry
->type
);
1688 return event
->raw(iter
, 0);
1690 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
1693 return TRACE_TYPE_HANDLED
;
1695 return TRACE_TYPE_PARTIAL_LINE
;
1698 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
1700 struct trace_seq
*s
= &iter
->seq
;
1701 unsigned char newline
= '\n';
1702 struct trace_entry
*entry
;
1703 struct trace_event
*event
;
1707 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1708 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1709 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1710 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
1713 event
= ftrace_find_event(entry
->type
);
1715 enum print_line_t ret
= event
->hex(iter
, 0);
1716 if (ret
!= TRACE_TYPE_HANDLED
)
1720 SEQ_PUT_FIELD_RET(s
, newline
);
1722 return TRACE_TYPE_HANDLED
;
1725 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
1727 struct trace_seq
*s
= &iter
->seq
;
1728 struct trace_entry
*entry
;
1729 struct trace_event
*event
;
1733 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1734 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1735 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
1736 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
1739 event
= ftrace_find_event(entry
->type
);
1740 return event
? event
->binary(iter
, 0) : TRACE_TYPE_HANDLED
;
1743 static int trace_empty(struct trace_iterator
*iter
)
1747 /* If we are looking at one CPU buffer, only check that one */
1748 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
1749 cpu
= iter
->cpu_file
;
1750 if (iter
->buffer_iter
[cpu
]) {
1751 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1754 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1760 for_each_tracing_cpu(cpu
) {
1761 if (iter
->buffer_iter
[cpu
]) {
1762 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1765 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1773 static enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
1775 enum print_line_t ret
;
1777 if (iter
->trace
&& iter
->trace
->print_line
) {
1778 ret
= iter
->trace
->print_line(iter
);
1779 if (ret
!= TRACE_TYPE_UNHANDLED
)
1783 if (iter
->ent
->type
== TRACE_BPRINT
&&
1784 trace_flags
& TRACE_ITER_PRINTK
&&
1785 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
1786 return trace_print_bprintk_msg_only(iter
);
1788 if (iter
->ent
->type
== TRACE_PRINT
&&
1789 trace_flags
& TRACE_ITER_PRINTK
&&
1790 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
1791 return trace_print_printk_msg_only(iter
);
1793 if (trace_flags
& TRACE_ITER_BIN
)
1794 return print_bin_fmt(iter
);
1796 if (trace_flags
& TRACE_ITER_HEX
)
1797 return print_hex_fmt(iter
);
1799 if (trace_flags
& TRACE_ITER_RAW
)
1800 return print_raw_fmt(iter
);
1802 return print_trace_fmt(iter
);
1805 static int s_show(struct seq_file
*m
, void *v
)
1807 struct trace_iterator
*iter
= v
;
1809 if (iter
->ent
== NULL
) {
1811 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
1814 if (iter
->trace
&& iter
->trace
->print_header
)
1815 iter
->trace
->print_header(m
);
1816 else if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1817 /* print nothing if the buffers are empty */
1818 if (trace_empty(iter
))
1820 print_trace_header(m
, iter
);
1821 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1822 print_lat_help_header(m
);
1824 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1825 print_func_help_header(m
);
1828 print_trace_line(iter
);
1829 trace_print_seq(m
, &iter
->seq
);
1835 static struct seq_operations tracer_seq_ops
= {
1842 static struct trace_iterator
*
1843 __tracing_open(struct inode
*inode
, struct file
*file
)
1845 long cpu_file
= (long) inode
->i_private
;
1846 void *fail_ret
= ERR_PTR(-ENOMEM
);
1847 struct trace_iterator
*iter
;
1851 if (tracing_disabled
)
1852 return ERR_PTR(-ENODEV
);
1854 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
1856 return ERR_PTR(-ENOMEM
);
1859 * We make a copy of the current tracer to avoid concurrent
1860 * changes on it while we are reading.
1862 mutex_lock(&trace_types_lock
);
1863 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
1868 *iter
->trace
= *current_trace
;
1870 if (current_trace
&& current_trace
->print_max
)
1873 iter
->tr
= &global_trace
;
1875 mutex_init(&iter
->mutex
);
1876 iter
->cpu_file
= cpu_file
;
1878 /* Notify the tracer early; before we stop tracing. */
1879 if (iter
->trace
&& iter
->trace
->open
)
1880 iter
->trace
->open(iter
);
1882 /* Annotate start of buffers if we had overruns */
1883 if (ring_buffer_overruns(iter
->tr
->buffer
))
1884 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
1886 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
1887 for_each_tracing_cpu(cpu
) {
1889 iter
->buffer_iter
[cpu
] =
1890 ring_buffer_read_start(iter
->tr
->buffer
, cpu
);
1893 cpu
= iter
->cpu_file
;
1894 iter
->buffer_iter
[cpu
] =
1895 ring_buffer_read_start(iter
->tr
->buffer
, cpu
);
1898 /* TODO stop tracer */
1899 ret
= seq_open(file
, &tracer_seq_ops
);
1901 fail_ret
= ERR_PTR(ret
);
1905 m
= file
->private_data
;
1908 /* stop the trace while dumping */
1911 mutex_unlock(&trace_types_lock
);
1916 for_each_tracing_cpu(cpu
) {
1917 if (iter
->buffer_iter
[cpu
])
1918 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
1921 mutex_unlock(&trace_types_lock
);
1928 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
1930 if (tracing_disabled
)
1933 filp
->private_data
= inode
->i_private
;
1937 static int tracing_release(struct inode
*inode
, struct file
*file
)
1939 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
1940 struct trace_iterator
*iter
;
1943 if (!(file
->f_mode
& FMODE_READ
))
1948 mutex_lock(&trace_types_lock
);
1949 for_each_tracing_cpu(cpu
) {
1950 if (iter
->buffer_iter
[cpu
])
1951 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
1954 if (iter
->trace
&& iter
->trace
->close
)
1955 iter
->trace
->close(iter
);
1957 /* reenable tracing if it was previously enabled */
1959 mutex_unlock(&trace_types_lock
);
1961 seq_release(inode
, file
);
1962 mutex_destroy(&iter
->mutex
);
1968 static int tracing_open(struct inode
*inode
, struct file
*file
)
1970 struct trace_iterator
*iter
;
1973 /* If this file was open for write, then erase contents */
1974 if ((file
->f_mode
& FMODE_WRITE
) &&
1975 !(file
->f_flags
& O_APPEND
)) {
1976 long cpu
= (long) inode
->i_private
;
1978 if (cpu
== TRACE_PIPE_ALL_CPU
)
1979 tracing_reset_online_cpus(&global_trace
);
1981 tracing_reset(&global_trace
, cpu
);
1984 if (file
->f_mode
& FMODE_READ
) {
1985 iter
= __tracing_open(inode
, file
);
1987 ret
= PTR_ERR(iter
);
1988 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
1989 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
1995 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1997 struct tracer
*t
= m
->private;
2009 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2011 struct tracer
*t
= m
->private;
2014 mutex_lock(&trace_types_lock
);
2015 for (; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2021 static void t_stop(struct seq_file
*m
, void *p
)
2023 mutex_unlock(&trace_types_lock
);
2026 static int t_show(struct seq_file
*m
, void *v
)
2028 struct tracer
*t
= v
;
2033 seq_printf(m
, "%s", t
->name
);
2042 static struct seq_operations show_traces_seq_ops
= {
2049 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2053 if (tracing_disabled
)
2056 ret
= seq_open(file
, &show_traces_seq_ops
);
2058 struct seq_file
*m
= file
->private_data
;
2059 m
->private = trace_types
;
2066 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2067 size_t count
, loff_t
*ppos
)
2072 static const struct file_operations tracing_fops
= {
2073 .open
= tracing_open
,
2075 .write
= tracing_write_stub
,
2076 .llseek
= seq_lseek
,
2077 .release
= tracing_release
,
2080 static const struct file_operations show_traces_fops
= {
2081 .open
= show_traces_open
,
2083 .release
= seq_release
,
2087 * Only trace on a CPU if the bitmask is set:
2089 static cpumask_var_t tracing_cpumask
;
2092 * The tracer itself will not take this lock, but still we want
2093 * to provide a consistent cpumask to user-space:
2095 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2098 * Temporary storage for the character representation of the
2099 * CPU bitmask (and one more byte for the newline):
2101 static char mask_str
[NR_CPUS
+ 1];
2104 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2105 size_t count
, loff_t
*ppos
)
2109 mutex_lock(&tracing_cpumask_update_lock
);
2111 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2112 if (count
- len
< 2) {
2116 len
+= sprintf(mask_str
+ len
, "\n");
2117 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2120 mutex_unlock(&tracing_cpumask_update_lock
);
2126 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2127 size_t count
, loff_t
*ppos
)
2130 cpumask_var_t tracing_cpumask_new
;
2132 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2135 mutex_lock(&tracing_cpumask_update_lock
);
2136 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2140 local_irq_disable();
2141 __raw_spin_lock(&ftrace_max_lock
);
2142 for_each_tracing_cpu(cpu
) {
2144 * Increase/decrease the disabled counter if we are
2145 * about to flip a bit in the cpumask:
2147 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2148 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2149 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2151 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2152 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2153 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2156 __raw_spin_unlock(&ftrace_max_lock
);
2159 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2161 mutex_unlock(&tracing_cpumask_update_lock
);
2162 free_cpumask_var(tracing_cpumask_new
);
2167 mutex_unlock(&tracing_cpumask_update_lock
);
2168 free_cpumask_var(tracing_cpumask
);
2173 static const struct file_operations tracing_cpumask_fops
= {
2174 .open
= tracing_open_generic
,
2175 .read
= tracing_cpumask_read
,
2176 .write
= tracing_cpumask_write
,
2180 tracing_trace_options_read(struct file
*filp
, char __user
*ubuf
,
2181 size_t cnt
, loff_t
*ppos
)
2183 struct tracer_opt
*trace_opts
;
2191 /* calculate max size */
2192 for (i
= 0; trace_options
[i
]; i
++) {
2193 len
+= strlen(trace_options
[i
]);
2194 len
+= 3; /* "no" and newline */
2197 mutex_lock(&trace_types_lock
);
2198 tracer_flags
= current_trace
->flags
->val
;
2199 trace_opts
= current_trace
->flags
->opts
;
2202 * Increase the size with names of options specific
2203 * of the current tracer.
2205 for (i
= 0; trace_opts
[i
].name
; i
++) {
2206 len
+= strlen(trace_opts
[i
].name
);
2207 len
+= 3; /* "no" and newline */
2210 /* +2 for \n and \0 */
2211 buf
= kmalloc(len
+ 2, GFP_KERNEL
);
2213 mutex_unlock(&trace_types_lock
);
2217 for (i
= 0; trace_options
[i
]; i
++) {
2218 if (trace_flags
& (1 << i
))
2219 r
+= sprintf(buf
+ r
, "%s\n", trace_options
[i
]);
2221 r
+= sprintf(buf
+ r
, "no%s\n", trace_options
[i
]);
2224 for (i
= 0; trace_opts
[i
].name
; i
++) {
2225 if (tracer_flags
& trace_opts
[i
].bit
)
2226 r
+= sprintf(buf
+ r
, "%s\n",
2227 trace_opts
[i
].name
);
2229 r
+= sprintf(buf
+ r
, "no%s\n",
2230 trace_opts
[i
].name
);
2232 mutex_unlock(&trace_types_lock
);
2234 WARN_ON(r
>= len
+ 2);
2236 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2242 /* Try to assign a tracer specific option */
2243 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2245 struct tracer_flags
*trace_flags
= trace
->flags
;
2246 struct tracer_opt
*opts
= NULL
;
2250 for (i
= 0; trace_flags
->opts
[i
].name
; i
++) {
2251 opts
= &trace_flags
->opts
[i
];
2252 len
= strlen(opts
->name
);
2254 if (strncmp(cmp
, opts
->name
, len
) == 0) {
2255 ret
= trace
->set_flag(trace_flags
->val
,
2261 if (!trace_flags
->opts
[i
].name
)
2264 /* Refused to handle */
2269 trace_flags
->val
&= ~opts
->bit
;
2271 trace_flags
->val
|= opts
->bit
;
2276 static void set_tracer_flags(unsigned int mask
, int enabled
)
2278 /* do nothing if flag is already set */
2279 if (!!(trace_flags
& mask
) == !!enabled
)
2283 trace_flags
|= mask
;
2285 trace_flags
&= ~mask
;
2287 if (mask
== TRACE_ITER_GLOBAL_CLK
) {
2291 func
= trace_clock_global
;
2293 func
= trace_clock_local
;
2295 mutex_lock(&trace_types_lock
);
2296 ring_buffer_set_clock(global_trace
.buffer
, func
);
2299 ring_buffer_set_clock(max_tr
.buffer
, func
);
2300 mutex_unlock(&trace_types_lock
);
2305 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2306 size_t cnt
, loff_t
*ppos
)
2314 if (cnt
>= sizeof(buf
))
2317 if (copy_from_user(&buf
, ubuf
, cnt
))
2322 if (strncmp(buf
, "no", 2) == 0) {
2327 for (i
= 0; trace_options
[i
]; i
++) {
2328 int len
= strlen(trace_options
[i
]);
2330 if (strncmp(cmp
, trace_options
[i
], len
) == 0) {
2331 set_tracer_flags(1 << i
, !neg
);
2336 /* If no option could be set, test the specific tracer options */
2337 if (!trace_options
[i
]) {
2338 mutex_lock(&trace_types_lock
);
2339 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2340 mutex_unlock(&trace_types_lock
);
2350 static const struct file_operations tracing_iter_fops
= {
2351 .open
= tracing_open_generic
,
2352 .read
= tracing_trace_options_read
,
2353 .write
= tracing_trace_options_write
,
2356 static const char readme_msg
[] =
2357 "tracing mini-HOWTO:\n\n"
2359 "# mount -t debugfs nodev /debug\n\n"
2360 "# cat /debug/tracing/available_tracers\n"
2361 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2362 "# cat /debug/tracing/current_tracer\n"
2364 "# echo sched_switch > /debug/tracing/current_tracer\n"
2365 "# cat /debug/tracing/current_tracer\n"
2367 "# cat /debug/tracing/trace_options\n"
2368 "noprint-parent nosym-offset nosym-addr noverbose\n"
2369 "# echo print-parent > /debug/tracing/trace_options\n"
2370 "# echo 1 > /debug/tracing/tracing_enabled\n"
2371 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2372 "echo 0 > /debug/tracing/tracing_enabled\n"
2376 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2377 size_t cnt
, loff_t
*ppos
)
2379 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2380 readme_msg
, strlen(readme_msg
));
2383 static const struct file_operations tracing_readme_fops
= {
2384 .open
= tracing_open_generic
,
2385 .read
= tracing_readme_read
,
2389 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2390 size_t cnt
, loff_t
*ppos
)
2395 r
= sprintf(buf
, "%u\n", tracer_enabled
);
2396 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2400 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2401 size_t cnt
, loff_t
*ppos
)
2403 struct trace_array
*tr
= filp
->private_data
;
2408 if (cnt
>= sizeof(buf
))
2411 if (copy_from_user(&buf
, ubuf
, cnt
))
2416 ret
= strict_strtoul(buf
, 10, &val
);
2422 mutex_lock(&trace_types_lock
);
2423 if (tracer_enabled
^ val
) {
2426 if (current_trace
->start
)
2427 current_trace
->start(tr
);
2432 if (current_trace
->stop
)
2433 current_trace
->stop(tr
);
2436 mutex_unlock(&trace_types_lock
);
2444 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2445 size_t cnt
, loff_t
*ppos
)
2447 char buf
[max_tracer_type_len
+2];
2450 mutex_lock(&trace_types_lock
);
2452 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2454 r
= sprintf(buf
, "\n");
2455 mutex_unlock(&trace_types_lock
);
2457 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2460 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
2462 tracing_reset_online_cpus(tr
);
2466 static int tracing_resize_ring_buffer(unsigned long size
)
2471 * If kernel or user changes the size of the ring buffer
2472 * we use the size that was given, and we can forget about
2473 * expanding it later.
2475 ring_buffer_expanded
= 1;
2477 ret
= ring_buffer_resize(global_trace
.buffer
, size
);
2481 ret
= ring_buffer_resize(max_tr
.buffer
, size
);
2485 r
= ring_buffer_resize(global_trace
.buffer
,
2486 global_trace
.entries
);
2489 * AARGH! We are left with different
2490 * size max buffer!!!!
2491 * The max buffer is our "snapshot" buffer.
2492 * When a tracer needs a snapshot (one of the
2493 * latency tracers), it swaps the max buffer
2494 * with the saved snap shot. We succeeded to
2495 * update the size of the main buffer, but failed to
2496 * update the size of the max buffer. But when we tried
2497 * to reset the main buffer to the original size, we
2498 * failed there too. This is very unlikely to
2499 * happen, but if it does, warn and kill all
2503 tracing_disabled
= 1;
2508 global_trace
.entries
= size
;
2514 * tracing_update_buffers - used by tracing facility to expand ring buffers
2516 * To save on memory when the tracing is never used on a system with it
2517 * configured in. The ring buffers are set to a minimum size. But once
2518 * a user starts to use the tracing facility, then they need to grow
2519 * to their default size.
2521 * This function is to be called when a tracer is about to be used.
2523 int tracing_update_buffers(void)
2527 mutex_lock(&trace_types_lock
);
2528 if (!ring_buffer_expanded
)
2529 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2530 mutex_unlock(&trace_types_lock
);
2535 struct trace_option_dentry
;
2537 static struct trace_option_dentry
*
2538 create_trace_option_files(struct tracer
*tracer
);
2541 destroy_trace_option_files(struct trace_option_dentry
*topts
);
2543 static int tracing_set_tracer(const char *buf
)
2545 static struct trace_option_dentry
*topts
;
2546 struct trace_array
*tr
= &global_trace
;
2550 mutex_lock(&trace_types_lock
);
2552 if (!ring_buffer_expanded
) {
2553 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2559 for (t
= trace_types
; t
; t
= t
->next
) {
2560 if (strcmp(t
->name
, buf
) == 0)
2567 if (t
== current_trace
)
2570 trace_branch_disable();
2571 if (current_trace
&& current_trace
->reset
)
2572 current_trace
->reset(tr
);
2574 destroy_trace_option_files(topts
);
2578 topts
= create_trace_option_files(current_trace
);
2581 ret
= tracer_init(t
, tr
);
2586 trace_branch_enable(tr
);
2588 mutex_unlock(&trace_types_lock
);
2594 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2595 size_t cnt
, loff_t
*ppos
)
2597 char buf
[max_tracer_type_len
+1];
2604 if (cnt
> max_tracer_type_len
)
2605 cnt
= max_tracer_type_len
;
2607 if (copy_from_user(&buf
, ubuf
, cnt
))
2612 /* strip ending whitespace. */
2613 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2616 err
= tracing_set_tracer(buf
);
2626 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2627 size_t cnt
, loff_t
*ppos
)
2629 unsigned long *ptr
= filp
->private_data
;
2633 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
2634 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2635 if (r
> sizeof(buf
))
2637 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2641 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2642 size_t cnt
, loff_t
*ppos
)
2644 unsigned long *ptr
= filp
->private_data
;
2649 if (cnt
>= sizeof(buf
))
2652 if (copy_from_user(&buf
, ubuf
, cnt
))
2657 ret
= strict_strtoul(buf
, 10, &val
);
2666 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2668 long cpu_file
= (long) inode
->i_private
;
2669 struct trace_iterator
*iter
;
2672 if (tracing_disabled
)
2675 mutex_lock(&trace_types_lock
);
2677 /* We only allow one reader per cpu */
2678 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
2679 if (!cpumask_empty(tracing_reader_cpumask
)) {
2683 cpumask_setall(tracing_reader_cpumask
);
2685 if (!cpumask_test_cpu(cpu_file
, tracing_reader_cpumask
))
2686 cpumask_set_cpu(cpu_file
, tracing_reader_cpumask
);
2693 /* create a buffer to store the information to pass to userspace */
2694 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2701 * We make a copy of the current tracer to avoid concurrent
2702 * changes on it while we are reading.
2704 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2710 *iter
->trace
= *current_trace
;
2712 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
2717 /* trace pipe does not show start of buffer */
2718 cpumask_setall(iter
->started
);
2720 iter
->cpu_file
= cpu_file
;
2721 iter
->tr
= &global_trace
;
2722 mutex_init(&iter
->mutex
);
2723 filp
->private_data
= iter
;
2725 if (iter
->trace
->pipe_open
)
2726 iter
->trace
->pipe_open(iter
);
2729 mutex_unlock(&trace_types_lock
);
2735 mutex_unlock(&trace_types_lock
);
2739 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
2741 struct trace_iterator
*iter
= file
->private_data
;
2743 mutex_lock(&trace_types_lock
);
2745 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
)
2746 cpumask_clear(tracing_reader_cpumask
);
2748 cpumask_clear_cpu(iter
->cpu_file
, tracing_reader_cpumask
);
2750 mutex_unlock(&trace_types_lock
);
2752 free_cpumask_var(iter
->started
);
2753 mutex_destroy(&iter
->mutex
);
2761 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
2763 struct trace_iterator
*iter
= filp
->private_data
;
2765 if (trace_flags
& TRACE_ITER_BLOCK
) {
2767 * Always select as readable when in blocking mode
2769 return POLLIN
| POLLRDNORM
;
2771 if (!trace_empty(iter
))
2772 return POLLIN
| POLLRDNORM
;
2773 poll_wait(filp
, &trace_wait
, poll_table
);
2774 if (!trace_empty(iter
))
2775 return POLLIN
| POLLRDNORM
;
2782 void default_wait_pipe(struct trace_iterator
*iter
)
2786 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
2788 if (trace_empty(iter
))
2791 finish_wait(&trace_wait
, &wait
);
2795 * This is a make-shift waitqueue.
2796 * A tracer might use this callback on some rare cases:
2798 * 1) the current tracer might hold the runqueue lock when it wakes up
2799 * a reader, hence a deadlock (sched, function, and function graph tracers)
2800 * 2) the function tracers, trace all functions, we don't want
2801 * the overhead of calling wake_up and friends
2802 * (and tracing them too)
2804 * Anyway, this is really very primitive wakeup.
2806 void poll_wait_pipe(struct trace_iterator
*iter
)
2808 set_current_state(TASK_INTERRUPTIBLE
);
2809 /* sleep for 100 msecs, and try again. */
2810 schedule_timeout(HZ
/ 10);
2813 /* Must be called with trace_types_lock mutex held. */
2814 static int tracing_wait_pipe(struct file
*filp
)
2816 struct trace_iterator
*iter
= filp
->private_data
;
2818 while (trace_empty(iter
)) {
2820 if ((filp
->f_flags
& O_NONBLOCK
)) {
2824 mutex_unlock(&iter
->mutex
);
2826 iter
->trace
->wait_pipe(iter
);
2828 mutex_lock(&iter
->mutex
);
2830 if (signal_pending(current
))
2834 * We block until we read something and tracing is disabled.
2835 * We still block if tracing is disabled, but we have never
2836 * read anything. This allows a user to cat this file, and
2837 * then enable tracing. But after we have read something,
2838 * we give an EOF when tracing is again disabled.
2840 * iter->pos will be 0 if we haven't read anything.
2842 if (!tracer_enabled
&& iter
->pos
)
2853 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
2854 size_t cnt
, loff_t
*ppos
)
2856 struct trace_iterator
*iter
= filp
->private_data
;
2857 static struct tracer
*old_tracer
;
2860 /* return any leftover data */
2861 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
2865 trace_seq_init(&iter
->seq
);
2867 /* copy the tracer to avoid using a global lock all around */
2868 mutex_lock(&trace_types_lock
);
2869 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
2870 old_tracer
= current_trace
;
2871 *iter
->trace
= *current_trace
;
2873 mutex_unlock(&trace_types_lock
);
2876 * Avoid more than one consumer on a single file descriptor
2877 * This is just a matter of traces coherency, the ring buffer itself
2880 mutex_lock(&iter
->mutex
);
2881 if (iter
->trace
->read
) {
2882 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
2888 sret
= tracing_wait_pipe(filp
);
2892 /* stop when tracing is finished */
2893 if (trace_empty(iter
)) {
2898 if (cnt
>= PAGE_SIZE
)
2899 cnt
= PAGE_SIZE
- 1;
2901 /* reset all but tr, trace, and overruns */
2902 memset(&iter
->seq
, 0,
2903 sizeof(struct trace_iterator
) -
2904 offsetof(struct trace_iterator
, seq
));
2907 while (find_next_entry_inc(iter
) != NULL
) {
2908 enum print_line_t ret
;
2909 int len
= iter
->seq
.len
;
2911 ret
= print_trace_line(iter
);
2912 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
2913 /* don't print partial lines */
2914 iter
->seq
.len
= len
;
2917 if (ret
!= TRACE_TYPE_NO_CONSUME
)
2918 trace_consume(iter
);
2920 if (iter
->seq
.len
>= cnt
)
2924 /* Now copy what we have to the user */
2925 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
2926 if (iter
->seq
.readpos
>= iter
->seq
.len
)
2927 trace_seq_init(&iter
->seq
);
2930 * If there was nothing to send to user, inspite of consuming trace
2931 * entries, go back to wait for more entries.
2937 mutex_unlock(&iter
->mutex
);
2942 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
2943 struct pipe_buffer
*buf
)
2945 __free_page(buf
->page
);
2948 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
2951 __free_page(spd
->pages
[idx
]);
2954 static struct pipe_buf_operations tracing_pipe_buf_ops
= {
2956 .map
= generic_pipe_buf_map
,
2957 .unmap
= generic_pipe_buf_unmap
,
2958 .confirm
= generic_pipe_buf_confirm
,
2959 .release
= tracing_pipe_buf_release
,
2960 .steal
= generic_pipe_buf_steal
,
2961 .get
= generic_pipe_buf_get
,
2965 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
2970 /* Seq buffer is page-sized, exactly what we need. */
2972 count
= iter
->seq
.len
;
2973 ret
= print_trace_line(iter
);
2974 count
= iter
->seq
.len
- count
;
2977 iter
->seq
.len
-= count
;
2980 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
2981 iter
->seq
.len
-= count
;
2985 trace_consume(iter
);
2987 if (!find_next_entry_inc(iter
)) {
2997 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
2999 struct pipe_inode_info
*pipe
,
3003 struct page
*pages
[PIPE_BUFFERS
];
3004 struct partial_page partial
[PIPE_BUFFERS
];
3005 struct trace_iterator
*iter
= filp
->private_data
;
3006 struct splice_pipe_desc spd
= {
3009 .nr_pages
= 0, /* This gets updated below. */
3011 .ops
= &tracing_pipe_buf_ops
,
3012 .spd_release
= tracing_spd_release_pipe
,
3014 static struct tracer
*old_tracer
;
3019 /* copy the tracer to avoid using a global lock all around */
3020 mutex_lock(&trace_types_lock
);
3021 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3022 old_tracer
= current_trace
;
3023 *iter
->trace
= *current_trace
;
3025 mutex_unlock(&trace_types_lock
);
3027 mutex_lock(&iter
->mutex
);
3029 if (iter
->trace
->splice_read
) {
3030 ret
= iter
->trace
->splice_read(iter
, filp
,
3031 ppos
, pipe
, len
, flags
);
3036 ret
= tracing_wait_pipe(filp
);
3040 if (!iter
->ent
&& !find_next_entry_inc(iter
)) {
3045 /* Fill as many pages as possible. */
3046 for (i
= 0, rem
= len
; i
< PIPE_BUFFERS
&& rem
; i
++) {
3047 pages
[i
] = alloc_page(GFP_KERNEL
);
3051 rem
= tracing_fill_pipe_page(rem
, iter
);
3053 /* Copy the data into the page, so we can start over. */
3054 ret
= trace_seq_to_buffer(&iter
->seq
,
3055 page_address(pages
[i
]),
3058 __free_page(pages
[i
]);
3061 partial
[i
].offset
= 0;
3062 partial
[i
].len
= iter
->seq
.len
;
3064 trace_seq_init(&iter
->seq
);
3067 mutex_unlock(&iter
->mutex
);
3071 return splice_to_pipe(pipe
, &spd
);
3074 mutex_unlock(&iter
->mutex
);
3080 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3081 size_t cnt
, loff_t
*ppos
)
3083 struct trace_array
*tr
= filp
->private_data
;
3087 mutex_lock(&trace_types_lock
);
3088 if (!ring_buffer_expanded
)
3089 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3091 trace_buf_size
>> 10);
3093 r
= sprintf(buf
, "%lu\n", tr
->entries
>> 10);
3094 mutex_unlock(&trace_types_lock
);
3096 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3100 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3101 size_t cnt
, loff_t
*ppos
)
3107 if (cnt
>= sizeof(buf
))
3110 if (copy_from_user(&buf
, ubuf
, cnt
))
3115 ret
= strict_strtoul(buf
, 10, &val
);
3119 /* must have at least 1 entry */
3123 mutex_lock(&trace_types_lock
);
3127 /* disable all cpu buffers */
3128 for_each_tracing_cpu(cpu
) {
3129 if (global_trace
.data
[cpu
])
3130 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
3131 if (max_tr
.data
[cpu
])
3132 atomic_inc(&max_tr
.data
[cpu
]->disabled
);
3135 /* value is in KB */
3138 if (val
!= global_trace
.entries
) {
3139 ret
= tracing_resize_ring_buffer(val
);
3148 /* If check pages failed, return ENOMEM */
3149 if (tracing_disabled
)
3152 for_each_tracing_cpu(cpu
) {
3153 if (global_trace
.data
[cpu
])
3154 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
3155 if (max_tr
.data
[cpu
])
3156 atomic_dec(&max_tr
.data
[cpu
]->disabled
);
3160 max_tr
.entries
= global_trace
.entries
;
3161 mutex_unlock(&trace_types_lock
);
3166 static int mark_printk(const char *fmt
, ...)
3170 va_start(args
, fmt
);
3171 ret
= trace_vprintk(0, fmt
, args
);
3177 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3178 size_t cnt
, loff_t
*fpos
)
3183 if (tracing_disabled
)
3186 if (cnt
> TRACE_BUF_SIZE
)
3187 cnt
= TRACE_BUF_SIZE
;
3189 buf
= kmalloc(cnt
+ 1, GFP_KERNEL
);
3193 if (copy_from_user(buf
, ubuf
, cnt
)) {
3198 /* Cut from the first nil or newline. */
3200 end
= strchr(buf
, '\n');
3204 cnt
= mark_printk("%s\n", buf
);
3211 static const struct file_operations tracing_max_lat_fops
= {
3212 .open
= tracing_open_generic
,
3213 .read
= tracing_max_lat_read
,
3214 .write
= tracing_max_lat_write
,
3217 static const struct file_operations tracing_ctrl_fops
= {
3218 .open
= tracing_open_generic
,
3219 .read
= tracing_ctrl_read
,
3220 .write
= tracing_ctrl_write
,
3223 static const struct file_operations set_tracer_fops
= {
3224 .open
= tracing_open_generic
,
3225 .read
= tracing_set_trace_read
,
3226 .write
= tracing_set_trace_write
,
3229 static const struct file_operations tracing_pipe_fops
= {
3230 .open
= tracing_open_pipe
,
3231 .poll
= tracing_poll_pipe
,
3232 .read
= tracing_read_pipe
,
3233 .splice_read
= tracing_splice_read_pipe
,
3234 .release
= tracing_release_pipe
,
3237 static const struct file_operations tracing_entries_fops
= {
3238 .open
= tracing_open_generic
,
3239 .read
= tracing_entries_read
,
3240 .write
= tracing_entries_write
,
3243 static const struct file_operations tracing_mark_fops
= {
3244 .open
= tracing_open_generic
,
3245 .write
= tracing_mark_write
,
3248 struct ftrace_buffer_info
{
3249 struct trace_array
*tr
;
3255 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
3257 int cpu
= (int)(long)inode
->i_private
;
3258 struct ftrace_buffer_info
*info
;
3260 if (tracing_disabled
)
3263 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3267 info
->tr
= &global_trace
;
3269 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
);
3270 /* Force reading ring buffer for first read */
3271 info
->read
= (unsigned int)-1;
3275 filp
->private_data
= info
;
3285 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
3286 size_t count
, loff_t
*ppos
)
3288 struct ftrace_buffer_info
*info
= filp
->private_data
;
3296 /* Do we have previous read data to read? */
3297 if (info
->read
< PAGE_SIZE
)
3302 ret
= ring_buffer_read_page(info
->tr
->buffer
,
3309 pos
= ring_buffer_page_len(info
->spare
);
3311 if (pos
< PAGE_SIZE
)
3312 memset(info
->spare
+ pos
, 0, PAGE_SIZE
- pos
);
3315 size
= PAGE_SIZE
- info
->read
;
3319 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
3330 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
3332 struct ftrace_buffer_info
*info
= file
->private_data
;
3334 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
3341 struct ring_buffer
*buffer
;
3346 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
3347 struct pipe_buffer
*buf
)
3349 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3354 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3359 static int buffer_pipe_buf_steal(struct pipe_inode_info
*pipe
,
3360 struct pipe_buffer
*buf
)
3365 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
3366 struct pipe_buffer
*buf
)
3368 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3373 /* Pipe buffer operations for a buffer. */
3374 static struct pipe_buf_operations buffer_pipe_buf_ops
= {
3376 .map
= generic_pipe_buf_map
,
3377 .unmap
= generic_pipe_buf_unmap
,
3378 .confirm
= generic_pipe_buf_confirm
,
3379 .release
= buffer_pipe_buf_release
,
3380 .steal
= buffer_pipe_buf_steal
,
3381 .get
= buffer_pipe_buf_get
,
3385 * Callback from splice_to_pipe(), if we need to release some pages
3386 * at the end of the spd in case we error'ed out in filling the pipe.
3388 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
3390 struct buffer_ref
*ref
=
3391 (struct buffer_ref
*)spd
->partial
[i
].private;
3396 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3398 spd
->partial
[i
].private = 0;
3402 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
3403 struct pipe_inode_info
*pipe
, size_t len
,
3406 struct ftrace_buffer_info
*info
= file
->private_data
;
3407 struct partial_page partial
[PIPE_BUFFERS
];
3408 struct page
*pages
[PIPE_BUFFERS
];
3409 struct splice_pipe_desc spd
= {
3413 .ops
= &buffer_pipe_buf_ops
,
3414 .spd_release
= buffer_spd_release
,
3416 struct buffer_ref
*ref
;
3421 * We can't seek on a buffer input
3423 if (unlikely(*ppos
))
3427 for (i
= 0; i
< PIPE_BUFFERS
&& len
; i
++, len
-= size
) {
3431 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
3435 ref
->buffer
= info
->tr
->buffer
;
3436 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
);
3442 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
3445 ring_buffer_free_read_page(ref
->buffer
,
3452 * zero out any left over data, this is going to
3455 size
= ring_buffer_page_len(ref
->page
);
3456 if (size
< PAGE_SIZE
)
3457 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
3459 page
= virt_to_page(ref
->page
);
3461 spd
.pages
[i
] = page
;
3462 spd
.partial
[i
].len
= PAGE_SIZE
;
3463 spd
.partial
[i
].offset
= 0;
3464 spd
.partial
[i
].private = (unsigned long)ref
;
3470 /* did we read anything? */
3471 if (!spd
.nr_pages
) {
3472 if (flags
& SPLICE_F_NONBLOCK
)
3480 ret
= splice_to_pipe(pipe
, &spd
);
3485 static const struct file_operations tracing_buffers_fops
= {
3486 .open
= tracing_buffers_open
,
3487 .read
= tracing_buffers_read
,
3488 .release
= tracing_buffers_release
,
3489 .splice_read
= tracing_buffers_splice_read
,
3490 .llseek
= no_llseek
,
3493 #ifdef CONFIG_DYNAMIC_FTRACE
3495 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
3501 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
3502 size_t cnt
, loff_t
*ppos
)
3504 static char ftrace_dyn_info_buffer
[1024];
3505 static DEFINE_MUTEX(dyn_info_mutex
);
3506 unsigned long *p
= filp
->private_data
;
3507 char *buf
= ftrace_dyn_info_buffer
;
3508 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
3511 mutex_lock(&dyn_info_mutex
);
3512 r
= sprintf(buf
, "%ld ", *p
);
3514 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
3517 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3519 mutex_unlock(&dyn_info_mutex
);
3524 static const struct file_operations tracing_dyn_info_fops
= {
3525 .open
= tracing_open_generic
,
3526 .read
= tracing_read_dyn_info
,
3530 static struct dentry
*d_tracer
;
3532 struct dentry
*tracing_init_dentry(void)
3539 if (!debugfs_initialized())
3542 d_tracer
= debugfs_create_dir("tracing", NULL
);
3544 if (!d_tracer
&& !once
) {
3546 pr_warning("Could not create debugfs directory 'tracing'\n");
3553 static struct dentry
*d_percpu
;
3555 struct dentry
*tracing_dentry_percpu(void)
3558 struct dentry
*d_tracer
;
3563 d_tracer
= tracing_init_dentry();
3568 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
3570 if (!d_percpu
&& !once
) {
3572 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3579 static void tracing_init_debugfs_percpu(long cpu
)
3581 struct dentry
*d_percpu
= tracing_dentry_percpu();
3582 struct dentry
*entry
, *d_cpu
;
3583 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3586 if (cpu
> 999 || cpu
< 0)
3589 sprintf(cpu_dir
, "cpu%ld", cpu
);
3590 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
3592 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
3596 /* per cpu trace_pipe */
3597 entry
= debugfs_create_file("trace_pipe", 0444, d_cpu
,
3598 (void *) cpu
, &tracing_pipe_fops
);
3600 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3603 entry
= debugfs_create_file("trace", 0644, d_cpu
,
3604 (void *) cpu
, &tracing_fops
);
3606 pr_warning("Could not create debugfs 'trace' entry\n");
3608 entry
= debugfs_create_file("trace_pipe_raw", 0444, d_cpu
,
3609 (void *) cpu
, &tracing_buffers_fops
);
3611 pr_warning("Could not create debugfs 'trace_pipe_raw' entry\n");
3614 #ifdef CONFIG_FTRACE_SELFTEST
3615 /* Let selftest have access to static functions in this file */
3616 #include "trace_selftest.c"
3619 struct trace_option_dentry
{
3620 struct tracer_opt
*opt
;
3621 struct tracer_flags
*flags
;
3622 struct dentry
*entry
;
3626 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
3629 struct trace_option_dentry
*topt
= filp
->private_data
;
3632 if (topt
->flags
->val
& topt
->opt
->bit
)
3637 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
3641 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
3644 struct trace_option_dentry
*topt
= filp
->private_data
;
3649 if (cnt
>= sizeof(buf
))
3652 if (copy_from_user(&buf
, ubuf
, cnt
))
3657 ret
= strict_strtoul(buf
, 10, &val
);
3664 /* do nothing if already cleared */
3665 if (!(topt
->flags
->val
& topt
->opt
->bit
))
3668 mutex_lock(&trace_types_lock
);
3669 if (current_trace
->set_flag
)
3670 ret
= current_trace
->set_flag(topt
->flags
->val
,
3672 mutex_unlock(&trace_types_lock
);
3675 topt
->flags
->val
&= ~topt
->opt
->bit
;
3678 /* do nothing if already set */
3679 if (topt
->flags
->val
& topt
->opt
->bit
)
3682 mutex_lock(&trace_types_lock
);
3683 if (current_trace
->set_flag
)
3684 ret
= current_trace
->set_flag(topt
->flags
->val
,
3686 mutex_unlock(&trace_types_lock
);
3689 topt
->flags
->val
|= topt
->opt
->bit
;
3702 static const struct file_operations trace_options_fops
= {
3703 .open
= tracing_open_generic
,
3704 .read
= trace_options_read
,
3705 .write
= trace_options_write
,
3709 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
3712 long index
= (long)filp
->private_data
;
3715 if (trace_flags
& (1 << index
))
3720 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
3724 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
3727 long index
= (long)filp
->private_data
;
3732 if (cnt
>= sizeof(buf
))
3735 if (copy_from_user(&buf
, ubuf
, cnt
))
3740 ret
= strict_strtoul(buf
, 10, &val
);
3746 trace_flags
&= ~(1 << index
);
3749 trace_flags
|= 1 << index
;
3761 static const struct file_operations trace_options_core_fops
= {
3762 .open
= tracing_open_generic
,
3763 .read
= trace_options_core_read
,
3764 .write
= trace_options_core_write
,
3767 static struct dentry
*trace_options_init_dentry(void)
3769 struct dentry
*d_tracer
;
3770 static struct dentry
*t_options
;
3775 d_tracer
= tracing_init_dentry();
3779 t_options
= debugfs_create_dir("options", d_tracer
);
3781 pr_warning("Could not create debugfs directory 'options'\n");
3789 create_trace_option_file(struct trace_option_dentry
*topt
,
3790 struct tracer_flags
*flags
,
3791 struct tracer_opt
*opt
)
3793 struct dentry
*t_options
;
3794 struct dentry
*entry
;
3796 t_options
= trace_options_init_dentry();
3800 topt
->flags
= flags
;
3803 entry
= debugfs_create_file(opt
->name
, 0644, t_options
, topt
,
3804 &trace_options_fops
);
3806 topt
->entry
= entry
;
3810 static struct trace_option_dentry
*
3811 create_trace_option_files(struct tracer
*tracer
)
3813 struct trace_option_dentry
*topts
;
3814 struct tracer_flags
*flags
;
3815 struct tracer_opt
*opts
;
3821 flags
= tracer
->flags
;
3823 if (!flags
|| !flags
->opts
)
3828 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
3831 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
3835 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
3836 create_trace_option_file(&topts
[cnt
], flags
,
3843 destroy_trace_option_files(struct trace_option_dentry
*topts
)
3850 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
3851 if (topts
[cnt
].entry
)
3852 debugfs_remove(topts
[cnt
].entry
);
3858 static struct dentry
*
3859 create_trace_option_core_file(const char *option
, long index
)
3861 struct dentry
*t_options
;
3862 struct dentry
*entry
;
3864 t_options
= trace_options_init_dentry();
3868 entry
= debugfs_create_file(option
, 0644, t_options
, (void *)index
,
3869 &trace_options_core_fops
);
3874 static __init
void create_trace_options_dir(void)
3876 struct dentry
*t_options
;
3877 struct dentry
*entry
;
3880 t_options
= trace_options_init_dentry();
3884 for (i
= 0; trace_options
[i
]; i
++) {
3885 entry
= create_trace_option_core_file(trace_options
[i
], i
);
3887 pr_warning("Could not create debugfs %s entry\n",
3892 static __init
int tracer_init_debugfs(void)
3894 struct dentry
*d_tracer
;
3895 struct dentry
*entry
;
3898 d_tracer
= tracing_init_dentry();
3900 entry
= debugfs_create_file("tracing_enabled", 0644, d_tracer
,
3901 &global_trace
, &tracing_ctrl_fops
);
3903 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3905 entry
= debugfs_create_file("trace_options", 0644, d_tracer
,
3906 NULL
, &tracing_iter_fops
);
3908 pr_warning("Could not create debugfs 'trace_options' entry\n");
3910 create_trace_options_dir();
3912 entry
= debugfs_create_file("tracing_cpumask", 0644, d_tracer
,
3913 NULL
, &tracing_cpumask_fops
);
3915 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3917 entry
= debugfs_create_file("trace", 0644, d_tracer
,
3918 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
3920 pr_warning("Could not create debugfs 'trace' entry\n");
3922 entry
= debugfs_create_file("available_tracers", 0444, d_tracer
,
3923 &global_trace
, &show_traces_fops
);
3925 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3927 entry
= debugfs_create_file("current_tracer", 0444, d_tracer
,
3928 &global_trace
, &set_tracer_fops
);
3930 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3932 entry
= debugfs_create_file("tracing_max_latency", 0644, d_tracer
,
3933 &tracing_max_latency
,
3934 &tracing_max_lat_fops
);
3936 pr_warning("Could not create debugfs "
3937 "'tracing_max_latency' entry\n");
3939 entry
= debugfs_create_file("tracing_thresh", 0644, d_tracer
,
3940 &tracing_thresh
, &tracing_max_lat_fops
);
3942 pr_warning("Could not create debugfs "
3943 "'tracing_thresh' entry\n");
3944 entry
= debugfs_create_file("README", 0644, d_tracer
,
3945 NULL
, &tracing_readme_fops
);
3947 pr_warning("Could not create debugfs 'README' entry\n");
3949 entry
= debugfs_create_file("trace_pipe", 0444, d_tracer
,
3950 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
3952 pr_warning("Could not create debugfs "
3953 "'trace_pipe' entry\n");
3955 entry
= debugfs_create_file("buffer_size_kb", 0644, d_tracer
,
3956 &global_trace
, &tracing_entries_fops
);
3958 pr_warning("Could not create debugfs "
3959 "'buffer_size_kb' entry\n");
3961 entry
= debugfs_create_file("trace_marker", 0220, d_tracer
,
3962 NULL
, &tracing_mark_fops
);
3964 pr_warning("Could not create debugfs "
3965 "'trace_marker' entry\n");
3967 #ifdef CONFIG_DYNAMIC_FTRACE
3968 entry
= debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
3969 &ftrace_update_tot_cnt
,
3970 &tracing_dyn_info_fops
);
3972 pr_warning("Could not create debugfs "
3973 "'dyn_ftrace_total_info' entry\n");
3975 #ifdef CONFIG_SYSPROF_TRACER
3976 init_tracer_sysprof_debugfs(d_tracer
);
3979 for_each_tracing_cpu(cpu
)
3980 tracing_init_debugfs_percpu(cpu
);
3985 static int trace_panic_handler(struct notifier_block
*this,
3986 unsigned long event
, void *unused
)
3988 if (ftrace_dump_on_oops
)
3993 static struct notifier_block trace_panic_notifier
= {
3994 .notifier_call
= trace_panic_handler
,
3996 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
3999 static int trace_die_handler(struct notifier_block
*self
,
4005 if (ftrace_dump_on_oops
)
4014 static struct notifier_block trace_die_notifier
= {
4015 .notifier_call
= trace_die_handler
,
4020 * printk is set to max of 1024, we really don't need it that big.
4021 * Nothing should be printing 1000 characters anyway.
4023 #define TRACE_MAX_PRINT 1000
4026 * Define here KERN_TRACE so that we have one place to modify
4027 * it if we decide to change what log level the ftrace dump
4030 #define KERN_TRACE KERN_EMERG
4033 trace_printk_seq(struct trace_seq
*s
)
4035 /* Probably should print a warning here. */
4039 /* should be zero ended, but we are paranoid. */
4040 s
->buffer
[s
->len
] = 0;
4042 printk(KERN_TRACE
"%s", s
->buffer
);
4047 static void __ftrace_dump(bool disable_tracing
)
4049 static DEFINE_SPINLOCK(ftrace_dump_lock
);
4050 /* use static because iter can be a bit big for the stack */
4051 static struct trace_iterator iter
;
4052 unsigned int old_userobj
;
4053 static int dump_ran
;
4054 unsigned long flags
;
4058 spin_lock_irqsave(&ftrace_dump_lock
, flags
);
4066 if (disable_tracing
)
4069 for_each_tracing_cpu(cpu
) {
4070 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
4073 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
4075 /* don't look at user memory in panic mode */
4076 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
4078 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
4080 /* Simulate the iterator */
4081 iter
.tr
= &global_trace
;
4082 iter
.trace
= current_trace
;
4083 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4086 * We need to stop all tracing on all CPUS to read the
4087 * the next buffer. This is a bit expensive, but is
4088 * not done often. We fill all what we can read,
4089 * and then release the locks again.
4092 while (!trace_empty(&iter
)) {
4095 printk(KERN_TRACE
"---------------------------------\n");
4099 /* reset all but tr, trace, and overruns */
4100 memset(&iter
.seq
, 0,
4101 sizeof(struct trace_iterator
) -
4102 offsetof(struct trace_iterator
, seq
));
4103 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
4106 if (find_next_entry_inc(&iter
) != NULL
) {
4107 print_trace_line(&iter
);
4108 trace_consume(&iter
);
4111 trace_printk_seq(&iter
.seq
);
4115 printk(KERN_TRACE
" (ftrace buffer empty)\n");
4117 printk(KERN_TRACE
"---------------------------------\n");
4119 /* Re-enable tracing if requested */
4120 if (!disable_tracing
) {
4121 trace_flags
|= old_userobj
;
4123 for_each_tracing_cpu(cpu
) {
4124 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
4130 spin_unlock_irqrestore(&ftrace_dump_lock
, flags
);
4133 /* By default: disable tracing after the dump */
4134 void ftrace_dump(void)
4136 __ftrace_dump(true);
4139 __init
static int tracer_alloc_buffers(void)
4141 struct trace_array_cpu
*data
;
4146 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
4149 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
4150 goto out_free_buffer_mask
;
4152 if (!alloc_cpumask_var(&tracing_reader_cpumask
, GFP_KERNEL
))
4153 goto out_free_tracing_cpumask
;
4155 /* To save memory, keep the ring buffer size to its minimum */
4156 if (ring_buffer_expanded
)
4157 ring_buf_size
= trace_buf_size
;
4161 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
4162 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
4163 cpumask_clear(tracing_reader_cpumask
);
4165 /* TODO: make the number of buffers hot pluggable with CPUS */
4166 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
,
4167 TRACE_BUFFER_FLAGS
);
4168 if (!global_trace
.buffer
) {
4169 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
4171 goto out_free_cpumask
;
4173 global_trace
.entries
= ring_buffer_size(global_trace
.buffer
);
4176 #ifdef CONFIG_TRACER_MAX_TRACE
4177 max_tr
.buffer
= ring_buffer_alloc(ring_buf_size
,
4178 TRACE_BUFFER_FLAGS
);
4179 if (!max_tr
.buffer
) {
4180 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
4182 ring_buffer_free(global_trace
.buffer
);
4183 goto out_free_cpumask
;
4185 max_tr
.entries
= ring_buffer_size(max_tr
.buffer
);
4186 WARN_ON(max_tr
.entries
!= global_trace
.entries
);
4189 /* Allocate the first page for all buffers */
4190 for_each_tracing_cpu(i
) {
4191 data
= global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
4192 max_tr
.data
[i
] = &per_cpu(max_data
, i
);
4195 trace_init_cmdlines();
4197 register_tracer(&nop_trace
);
4198 current_trace
= &nop_trace
;
4199 #ifdef CONFIG_BOOT_TRACER
4200 register_tracer(&boot_tracer
);
4202 /* All seems OK, enable tracing */
4203 tracing_disabled
= 0;
4205 atomic_notifier_chain_register(&panic_notifier_list
,
4206 &trace_panic_notifier
);
4208 register_die_notifier(&trace_die_notifier
);
4213 free_cpumask_var(tracing_reader_cpumask
);
4214 out_free_tracing_cpumask
:
4215 free_cpumask_var(tracing_cpumask
);
4216 out_free_buffer_mask
:
4217 free_cpumask_var(tracing_buffer_mask
);
4222 __init
static int clear_boot_tracer(void)
4225 * The default tracer at boot buffer is an init section.
4226 * This function is called in lateinit. If we did not
4227 * find the boot tracer, then clear it out, to prevent
4228 * later registration from accessing the buffer that is
4229 * about to be freed.
4231 if (!default_bootup_tracer
)
4234 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
4235 default_bootup_tracer
);
4236 default_bootup_tracer
= NULL
;
4241 early_initcall(tracer_alloc_buffers
);
4242 fs_initcall(tracer_init_debugfs
);
4243 late_initcall(clear_boot_tracer
);