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/utsrelease.h>
15 #include <linux/kallsyms.h>
16 #include <linux/seq_file.h>
17 #include <linux/notifier.h>
18 #include <linux/debugfs.h>
19 #include <linux/pagemap.h>
20 #include <linux/hardirq.h>
21 #include <linux/linkage.h>
22 #include <linux/uaccess.h>
23 #include <linux/ftrace.h>
24 #include <linux/module.h>
25 #include <linux/percpu.h>
26 #include <linux/kdebug.h>
27 #include <linux/ctype.h>
28 #include <linux/init.h>
29 #include <linux/poll.h>
30 #include <linux/gfp.h>
32 #include <linux/kprobes.h>
33 #include <linux/writeback.h>
34 #include <linux/splice.h>
36 #include <linux/stacktrace.h>
37 #include <linux/ring_buffer.h>
38 #include <linux/irqflags.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 * We need to change this state when a selftest is running.
50 * A selftest will lurk into the ring-buffer to count the
51 * entries inserted during the selftest although some concurrent
52 * insertions into the ring-buffer such as ftrace_printk could occurred
53 * at the same time, giving false positive or negative results.
55 static bool __read_mostly tracing_selftest_running
;
58 * If a tracer is running, we do not want to run SELFTEST.
60 static bool __read_mostly tracing_selftest_disabled
;
62 /* For tracers that don't implement custom flags */
63 static struct tracer_opt dummy_tracer_opt
[] = {
67 static struct tracer_flags dummy_tracer_flags
= {
69 .opts
= dummy_tracer_opt
72 static int dummy_set_flag(u32 old_flags
, u32 bit
, int set
)
78 * Kill all tracing for good (never come back).
79 * It is initialized to 1 but will turn to zero if the initialization
80 * of the tracer is successful. But that is the only place that sets
83 static int tracing_disabled
= 1;
85 static DEFINE_PER_CPU(local_t
, ftrace_cpu_disabled
);
87 static inline void ftrace_disable_cpu(void)
90 local_inc(&__get_cpu_var(ftrace_cpu_disabled
));
93 static inline void ftrace_enable_cpu(void)
95 local_dec(&__get_cpu_var(ftrace_cpu_disabled
));
99 static cpumask_var_t __read_mostly tracing_buffer_mask
;
101 /* Define which cpu buffers are currently read in trace_pipe */
102 static cpumask_var_t tracing_reader_cpumask
;
104 #define for_each_tracing_cpu(cpu) \
105 for_each_cpu(cpu, tracing_buffer_mask)
108 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
110 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
111 * is set, then ftrace_dump is called. This will output the contents
112 * of the ftrace buffers to the console. This is very useful for
113 * capturing traces that lead to crashes and outputing it to a
116 * It is default off, but you can enable it with either specifying
117 * "ftrace_dump_on_oops" in the kernel command line, or setting
118 * /proc/sys/kernel/ftrace_dump_on_oops to true.
120 int ftrace_dump_on_oops
;
122 static int tracing_set_tracer(const char *buf
);
124 #define BOOTUP_TRACER_SIZE 100
125 static char bootup_tracer_buf
[BOOTUP_TRACER_SIZE
] __initdata
;
126 static char *default_bootup_tracer
;
128 static int __init
set_ftrace(char *str
)
130 strncpy(bootup_tracer_buf
, str
, BOOTUP_TRACER_SIZE
);
131 default_bootup_tracer
= bootup_tracer_buf
;
134 __setup("ftrace=", set_ftrace
);
136 static int __init
set_ftrace_dump_on_oops(char *str
)
138 ftrace_dump_on_oops
= 1;
141 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops
);
144 ns2usecs(cycle_t nsec
)
151 cycle_t
ftrace_now(int cpu
)
153 u64 ts
= ring_buffer_time_stamp(cpu
);
154 ring_buffer_normalize_time_stamp(cpu
, &ts
);
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
);
175 * The max_tr is used to snapshot the global_trace when a maximum
176 * latency is reached. Some tracers will use this to store a maximum
177 * trace while it continues examining live traces.
179 * The buffers for the max_tr are set up the same as the global_trace.
180 * When a snapshot is taken, the link list of the max_tr is swapped
181 * with the link list of the global_trace and the buffers are reset for
182 * the global_trace so the tracing can continue.
184 static struct trace_array max_tr
;
186 static DEFINE_PER_CPU(struct trace_array_cpu
, max_data
);
188 /* tracer_enabled is used to toggle activation of a tracer */
189 static int tracer_enabled
= 1;
192 * tracing_is_enabled - return tracer_enabled status
194 * This function is used by other tracers to know the status
195 * of the tracer_enabled flag. Tracers may use this function
196 * to know if it should enable their features when starting
197 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
199 int tracing_is_enabled(void)
201 return tracer_enabled
;
205 * trace_buf_size is the size in bytes that is allocated
206 * for a buffer. Note, the number of bytes is always rounded
209 * This number is purposely set to a low number of 16384.
210 * If the dump on oops happens, it will be much appreciated
211 * to not have to wait for all that output. Anyway this can be
212 * boot time and run time configurable.
214 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
216 static unsigned long trace_buf_size
= TRACE_BUF_SIZE_DEFAULT
;
218 /* trace_types holds a link list of available tracers. */
219 static struct tracer
*trace_types __read_mostly
;
221 /* current_trace points to the tracer that is currently active */
222 static struct tracer
*current_trace __read_mostly
;
225 * max_tracer_type_len is used to simplify the allocating of
226 * buffers to read userspace tracer names. We keep track of
227 * the longest tracer name registered.
229 static int max_tracer_type_len
;
232 * trace_types_lock is used to protect the trace_types list.
233 * This lock is also used to keep user access serialized.
234 * Accesses from userspace will grab this lock while userspace
235 * activities happen inside the kernel.
237 static DEFINE_MUTEX(trace_types_lock
);
239 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
240 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
242 /* trace_flags holds trace_options default values */
243 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
| TRACE_ITER_PRINTK
|
244 TRACE_ITER_ANNOTATE
| TRACE_ITER_CONTEXT_INFO
;
247 * trace_wake_up - wake up tasks waiting for trace input
249 * Simply wakes up any task that is blocked on the trace_wait
250 * queue. These is used with trace_poll for tasks polling the trace.
252 void trace_wake_up(void)
255 * The runqueue_is_locked() can fail, but this is the best we
258 if (!(trace_flags
& TRACE_ITER_BLOCK
) && !runqueue_is_locked())
259 wake_up(&trace_wait
);
262 static int __init
set_buf_size(char *str
)
264 unsigned long buf_size
;
269 ret
= strict_strtoul(str
, 0, &buf_size
);
270 /* nr_entries can not be zero */
271 if (ret
< 0 || buf_size
== 0)
273 trace_buf_size
= buf_size
;
276 __setup("trace_buf_size=", set_buf_size
);
278 unsigned long nsecs_to_usecs(unsigned long nsecs
)
283 /* These must match the bit postions in trace_iterator_flags */
284 static const char *trace_options
[] = {
307 * ftrace_max_lock is used to protect the swapping of buffers
308 * when taking a max snapshot. The buffers themselves are
309 * protected by per_cpu spinlocks. But the action of the swap
310 * needs its own lock.
312 * This is defined as a raw_spinlock_t in order to help
313 * with performance when lockdep debugging is enabled.
315 static raw_spinlock_t ftrace_max_lock
=
316 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
319 * Copy the new maximum trace into the separate maximum-trace
320 * structure. (this way the maximum trace is permanently saved,
321 * for later retrieval via /debugfs/tracing/latency_trace)
324 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
326 struct trace_array_cpu
*data
= tr
->data
[cpu
];
329 max_tr
.time_start
= data
->preempt_timestamp
;
331 data
= max_tr
.data
[cpu
];
332 data
->saved_latency
= tracing_max_latency
;
334 memcpy(data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
335 data
->pid
= tsk
->pid
;
336 data
->uid
= task_uid(tsk
);
337 data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
338 data
->policy
= tsk
->policy
;
339 data
->rt_priority
= tsk
->rt_priority
;
341 /* record this tasks comm */
342 tracing_record_cmdline(tsk
);
346 trace_seq_reset(struct trace_seq
*s
)
352 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
357 if (s
->len
<= s
->readpos
)
360 len
= s
->len
- s
->readpos
;
363 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
371 ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
376 if (s
->len
<= s
->readpos
)
379 len
= s
->len
- s
->readpos
;
382 ret
= memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
391 trace_print_seq(struct seq_file
*m
, struct trace_seq
*s
)
393 int len
= s
->len
>= PAGE_SIZE
? PAGE_SIZE
- 1 : s
->len
;
396 seq_puts(m
, s
->buffer
);
402 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
404 * @tsk: the task with the latency
405 * @cpu: The cpu that initiated the trace.
407 * Flip the buffers between the @tr and the max_tr and record information
408 * about which task was the cause of this latency.
411 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
413 struct ring_buffer
*buf
= tr
->buffer
;
415 WARN_ON_ONCE(!irqs_disabled());
416 __raw_spin_lock(&ftrace_max_lock
);
418 tr
->buffer
= max_tr
.buffer
;
421 ftrace_disable_cpu();
422 ring_buffer_reset(tr
->buffer
);
425 __update_max_tr(tr
, tsk
, cpu
);
426 __raw_spin_unlock(&ftrace_max_lock
);
430 * update_max_tr_single - only copy one trace over, and reset the rest
432 * @tsk - task with the latency
433 * @cpu - the cpu of the buffer to copy.
435 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
438 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
442 WARN_ON_ONCE(!irqs_disabled());
443 __raw_spin_lock(&ftrace_max_lock
);
445 ftrace_disable_cpu();
447 ring_buffer_reset(max_tr
.buffer
);
448 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
452 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
);
454 __update_max_tr(tr
, tsk
, cpu
);
455 __raw_spin_unlock(&ftrace_max_lock
);
459 * register_tracer - register a tracer with the ftrace system.
460 * @type - the plugin for the tracer
462 * Register a new plugin tracer.
464 int register_tracer(struct tracer
*type
)
465 __releases(kernel_lock
)
466 __acquires(kernel_lock
)
473 pr_info("Tracer must have a name\n");
478 * When this gets called we hold the BKL which means that
479 * preemption is disabled. Various trace selftests however
480 * need to disable and enable preemption for successful tests.
481 * So we drop the BKL here and grab it after the tests again.
484 mutex_lock(&trace_types_lock
);
486 tracing_selftest_running
= true;
488 for (t
= trace_types
; t
; t
= t
->next
) {
489 if (strcmp(type
->name
, t
->name
) == 0) {
491 pr_info("Trace %s already registered\n",
499 type
->set_flag
= &dummy_set_flag
;
501 type
->flags
= &dummy_tracer_flags
;
503 if (!type
->flags
->opts
)
504 type
->flags
->opts
= dummy_tracer_opt
;
505 if (!type
->wait_pipe
)
506 type
->wait_pipe
= default_wait_pipe
;
509 #ifdef CONFIG_FTRACE_STARTUP_TEST
510 if (type
->selftest
&& !tracing_selftest_disabled
) {
511 struct tracer
*saved_tracer
= current_trace
;
512 struct trace_array
*tr
= &global_trace
;
516 * Run a selftest on this tracer.
517 * Here we reset the trace buffer, and set the current
518 * tracer to be this tracer. The tracer can then run some
519 * internal tracing to verify that everything is in order.
520 * If we fail, we do not register this tracer.
522 for_each_tracing_cpu(i
)
523 tracing_reset(tr
, i
);
525 current_trace
= type
;
526 /* the test is responsible for initializing and enabling */
527 pr_info("Testing tracer %s: ", type
->name
);
528 ret
= type
->selftest(type
, tr
);
529 /* the test is responsible for resetting too */
530 current_trace
= saved_tracer
;
532 printk(KERN_CONT
"FAILED!\n");
535 /* Only reset on passing, to avoid touching corrupted buffers */
536 for_each_tracing_cpu(i
)
537 tracing_reset(tr
, i
);
539 printk(KERN_CONT
"PASSED\n");
543 type
->next
= trace_types
;
545 len
= strlen(type
->name
);
546 if (len
> max_tracer_type_len
)
547 max_tracer_type_len
= len
;
550 tracing_selftest_running
= false;
551 mutex_unlock(&trace_types_lock
);
553 if (ret
|| !default_bootup_tracer
)
556 if (strncmp(default_bootup_tracer
, type
->name
, BOOTUP_TRACER_SIZE
))
559 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
560 /* Do we want this tracer to start on bootup? */
561 tracing_set_tracer(type
->name
);
562 default_bootup_tracer
= NULL
;
563 /* disable other selftests, since this will break it. */
564 tracing_selftest_disabled
= 1;
565 #ifdef CONFIG_FTRACE_STARTUP_TEST
566 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
575 void unregister_tracer(struct tracer
*type
)
580 mutex_lock(&trace_types_lock
);
581 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
585 pr_info("Trace %s not registered\n", type
->name
);
591 if (type
== current_trace
&& tracer_enabled
) {
594 if (current_trace
->stop
)
595 current_trace
->stop(&global_trace
);
596 current_trace
= &nop_trace
;
599 if (strlen(type
->name
) != max_tracer_type_len
)
602 max_tracer_type_len
= 0;
603 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
604 len
= strlen((*t
)->name
);
605 if (len
> max_tracer_type_len
)
606 max_tracer_type_len
= len
;
609 mutex_unlock(&trace_types_lock
);
612 void tracing_reset(struct trace_array
*tr
, int cpu
)
614 ftrace_disable_cpu();
615 ring_buffer_reset_cpu(tr
->buffer
, cpu
);
619 void tracing_reset_online_cpus(struct trace_array
*tr
)
623 tr
->time_start
= ftrace_now(tr
->cpu
);
625 for_each_online_cpu(cpu
)
626 tracing_reset(tr
, cpu
);
629 #define SAVED_CMDLINES 128
630 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
631 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
632 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
633 static int cmdline_idx
;
634 static DEFINE_SPINLOCK(trace_cmdline_lock
);
636 /* temporary disable recording */
637 static atomic_t trace_record_cmdline_disabled __read_mostly
;
639 static void trace_init_cmdlines(void)
641 memset(&map_pid_to_cmdline
, -1, sizeof(map_pid_to_cmdline
));
642 memset(&map_cmdline_to_pid
, -1, sizeof(map_cmdline_to_pid
));
646 static int trace_stop_count
;
647 static DEFINE_SPINLOCK(tracing_start_lock
);
650 * ftrace_off_permanent - disable all ftrace code permanently
652 * This should only be called when a serious anomally has
653 * been detected. This will turn off the function tracing,
654 * ring buffers, and other tracing utilites. It takes no
655 * locks and can be called from any context.
657 void ftrace_off_permanent(void)
659 tracing_disabled
= 1;
661 tracing_off_permanent();
665 * tracing_start - quick start of the tracer
667 * If tracing is enabled but was stopped by tracing_stop,
668 * this will start the tracer back up.
670 void tracing_start(void)
672 struct ring_buffer
*buffer
;
675 if (tracing_disabled
)
678 spin_lock_irqsave(&tracing_start_lock
, flags
);
679 if (--trace_stop_count
) {
680 if (trace_stop_count
< 0) {
681 /* Someone screwed up their debugging */
683 trace_stop_count
= 0;
689 buffer
= global_trace
.buffer
;
691 ring_buffer_record_enable(buffer
);
693 buffer
= max_tr
.buffer
;
695 ring_buffer_record_enable(buffer
);
699 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
703 * tracing_stop - quick stop of the tracer
705 * Light weight way to stop tracing. Use in conjunction with
708 void tracing_stop(void)
710 struct ring_buffer
*buffer
;
714 spin_lock_irqsave(&tracing_start_lock
, flags
);
715 if (trace_stop_count
++)
718 buffer
= global_trace
.buffer
;
720 ring_buffer_record_disable(buffer
);
722 buffer
= max_tr
.buffer
;
724 ring_buffer_record_disable(buffer
);
727 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
730 void trace_stop_cmdline_recording(void);
732 static void trace_save_cmdline(struct task_struct
*tsk
)
737 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
741 * It's not the end of the world if we don't get
742 * the lock, but we also don't want to spin
743 * nor do we want to disable interrupts,
744 * so if we miss here, then better luck next time.
746 if (!spin_trylock(&trace_cmdline_lock
))
749 idx
= map_pid_to_cmdline
[tsk
->pid
];
750 if (idx
>= SAVED_CMDLINES
) {
751 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
753 map
= map_cmdline_to_pid
[idx
];
754 if (map
<= PID_MAX_DEFAULT
)
755 map_pid_to_cmdline
[map
] = (unsigned)-1;
757 map_pid_to_cmdline
[tsk
->pid
] = idx
;
762 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
764 spin_unlock(&trace_cmdline_lock
);
767 char *trace_find_cmdline(int pid
)
769 char *cmdline
= "<...>";
775 if (pid
> PID_MAX_DEFAULT
)
778 map
= map_pid_to_cmdline
[pid
];
779 if (map
>= SAVED_CMDLINES
)
782 cmdline
= saved_cmdlines
[map
];
788 void tracing_record_cmdline(struct task_struct
*tsk
)
790 if (atomic_read(&trace_record_cmdline_disabled
))
793 trace_save_cmdline(tsk
);
797 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
800 struct task_struct
*tsk
= current
;
802 entry
->preempt_count
= pc
& 0xff;
803 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
804 entry
->tgid
= (tsk
) ? tsk
->tgid
: 0;
806 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
807 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
809 TRACE_FLAG_IRQS_NOSUPPORT
|
811 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
812 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
813 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
816 struct ring_buffer_event
*trace_buffer_lock_reserve(struct trace_array
*tr
,
819 unsigned long flags
, int pc
)
821 struct ring_buffer_event
*event
;
823 event
= ring_buffer_lock_reserve(tr
->buffer
, len
);
825 struct trace_entry
*ent
= ring_buffer_event_data(event
);
827 tracing_generic_entry_update(ent
, flags
, pc
);
833 static void ftrace_trace_stack(struct trace_array
*tr
,
834 unsigned long flags
, int skip
, int pc
);
835 static void ftrace_trace_userstack(struct trace_array
*tr
,
836 unsigned long flags
, int pc
);
838 void trace_buffer_unlock_commit(struct trace_array
*tr
,
839 struct ring_buffer_event
*event
,
840 unsigned long flags
, int pc
)
842 ring_buffer_unlock_commit(tr
->buffer
, event
);
844 ftrace_trace_stack(tr
, flags
, 6, pc
);
845 ftrace_trace_userstack(tr
, flags
, pc
);
850 trace_function(struct trace_array
*tr
,
851 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
854 struct ring_buffer_event
*event
;
855 struct ftrace_entry
*entry
;
857 /* If we are reading the ring buffer, don't trace */
858 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
861 event
= trace_buffer_lock_reserve(tr
, TRACE_FN
, sizeof(*entry
),
865 entry
= ring_buffer_event_data(event
);
867 entry
->parent_ip
= parent_ip
;
868 ring_buffer_unlock_commit(tr
->buffer
, event
);
871 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
872 static void __trace_graph_entry(struct trace_array
*tr
,
873 struct ftrace_graph_ent
*trace
,
877 struct ring_buffer_event
*event
;
878 struct ftrace_graph_ent_entry
*entry
;
880 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
883 event
= trace_buffer_lock_reserve(&global_trace
, TRACE_GRAPH_ENT
,
884 sizeof(*entry
), flags
, pc
);
887 entry
= ring_buffer_event_data(event
);
888 entry
->graph_ent
= *trace
;
889 ring_buffer_unlock_commit(global_trace
.buffer
, event
);
892 static void __trace_graph_return(struct trace_array
*tr
,
893 struct ftrace_graph_ret
*trace
,
897 struct ring_buffer_event
*event
;
898 struct ftrace_graph_ret_entry
*entry
;
900 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
903 event
= trace_buffer_lock_reserve(&global_trace
, TRACE_GRAPH_RET
,
904 sizeof(*entry
), flags
, pc
);
907 entry
= ring_buffer_event_data(event
);
909 ring_buffer_unlock_commit(global_trace
.buffer
, event
);
914 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
915 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
918 if (likely(!atomic_read(&data
->disabled
)))
919 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
922 static void __ftrace_trace_stack(struct trace_array
*tr
,
926 #ifdef CONFIG_STACKTRACE
927 struct ring_buffer_event
*event
;
928 struct stack_entry
*entry
;
929 struct stack_trace trace
;
931 event
= trace_buffer_lock_reserve(tr
, TRACE_STACK
,
932 sizeof(*entry
), flags
, pc
);
935 entry
= ring_buffer_event_data(event
);
936 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
938 trace
.nr_entries
= 0;
939 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
941 trace
.entries
= entry
->caller
;
943 save_stack_trace(&trace
);
944 ring_buffer_unlock_commit(tr
->buffer
, event
);
948 static void ftrace_trace_stack(struct trace_array
*tr
,
952 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
955 __ftrace_trace_stack(tr
, flags
, skip
, pc
);
958 void __trace_stack(struct trace_array
*tr
,
962 __ftrace_trace_stack(tr
, flags
, skip
, pc
);
965 static void ftrace_trace_userstack(struct trace_array
*tr
,
966 unsigned long flags
, int pc
)
968 #ifdef CONFIG_STACKTRACE
969 struct ring_buffer_event
*event
;
970 struct userstack_entry
*entry
;
971 struct stack_trace trace
;
973 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
976 event
= trace_buffer_lock_reserve(tr
, TRACE_USER_STACK
,
977 sizeof(*entry
), flags
, pc
);
980 entry
= ring_buffer_event_data(event
);
982 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
984 trace
.nr_entries
= 0;
985 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
987 trace
.entries
= entry
->caller
;
989 save_stack_trace_user(&trace
);
990 ring_buffer_unlock_commit(tr
->buffer
, event
);
995 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
997 ftrace_trace_userstack(tr
, flags
, preempt_count());
1002 ftrace_trace_special(void *__tr
,
1003 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
,
1006 struct ring_buffer_event
*event
;
1007 struct trace_array
*tr
= __tr
;
1008 struct special_entry
*entry
;
1010 event
= trace_buffer_lock_reserve(tr
, TRACE_SPECIAL
,
1011 sizeof(*entry
), 0, pc
);
1014 entry
= ring_buffer_event_data(event
);
1018 trace_buffer_unlock_commit(tr
, event
, 0, pc
);
1022 __trace_special(void *__tr
, void *__data
,
1023 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1025 ftrace_trace_special(__tr
, arg1
, arg2
, arg3
, preempt_count());
1029 tracing_sched_switch_trace(struct trace_array
*tr
,
1030 struct task_struct
*prev
,
1031 struct task_struct
*next
,
1032 unsigned long flags
, int pc
)
1034 struct ring_buffer_event
*event
;
1035 struct ctx_switch_entry
*entry
;
1037 event
= trace_buffer_lock_reserve(tr
, TRACE_CTX
,
1038 sizeof(*entry
), flags
, pc
);
1041 entry
= ring_buffer_event_data(event
);
1042 entry
->prev_pid
= prev
->pid
;
1043 entry
->prev_prio
= prev
->prio
;
1044 entry
->prev_state
= prev
->state
;
1045 entry
->next_pid
= next
->pid
;
1046 entry
->next_prio
= next
->prio
;
1047 entry
->next_state
= next
->state
;
1048 entry
->next_cpu
= task_cpu(next
);
1049 trace_buffer_unlock_commit(tr
, event
, flags
, pc
);
1053 tracing_sched_wakeup_trace(struct trace_array
*tr
,
1054 struct task_struct
*wakee
,
1055 struct task_struct
*curr
,
1056 unsigned long flags
, int pc
)
1058 struct ring_buffer_event
*event
;
1059 struct ctx_switch_entry
*entry
;
1061 event
= trace_buffer_lock_reserve(tr
, TRACE_WAKE
,
1062 sizeof(*entry
), flags
, pc
);
1065 entry
= ring_buffer_event_data(event
);
1066 entry
->prev_pid
= curr
->pid
;
1067 entry
->prev_prio
= curr
->prio
;
1068 entry
->prev_state
= curr
->state
;
1069 entry
->next_pid
= wakee
->pid
;
1070 entry
->next_prio
= wakee
->prio
;
1071 entry
->next_state
= wakee
->state
;
1072 entry
->next_cpu
= task_cpu(wakee
);
1074 ring_buffer_unlock_commit(tr
->buffer
, event
);
1075 ftrace_trace_stack(tr
, flags
, 6, pc
);
1076 ftrace_trace_userstack(tr
, flags
, pc
);
1080 ftrace_special(unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1082 struct trace_array
*tr
= &global_trace
;
1083 struct trace_array_cpu
*data
;
1084 unsigned long flags
;
1088 if (tracing_disabled
)
1091 pc
= preempt_count();
1092 local_irq_save(flags
);
1093 cpu
= raw_smp_processor_id();
1094 data
= tr
->data
[cpu
];
1096 if (likely(atomic_inc_return(&data
->disabled
) == 1))
1097 ftrace_trace_special(tr
, arg1
, arg2
, arg3
, pc
);
1099 atomic_dec(&data
->disabled
);
1100 local_irq_restore(flags
);
1103 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1104 int trace_graph_entry(struct ftrace_graph_ent
*trace
)
1106 struct trace_array
*tr
= &global_trace
;
1107 struct trace_array_cpu
*data
;
1108 unsigned long flags
;
1113 if (!ftrace_trace_task(current
))
1116 if (!ftrace_graph_addr(trace
->func
))
1119 local_irq_save(flags
);
1120 cpu
= raw_smp_processor_id();
1121 data
= tr
->data
[cpu
];
1122 disabled
= atomic_inc_return(&data
->disabled
);
1123 if (likely(disabled
== 1)) {
1124 pc
= preempt_count();
1125 __trace_graph_entry(tr
, trace
, flags
, pc
);
1127 /* Only do the atomic if it is not already set */
1128 if (!test_tsk_trace_graph(current
))
1129 set_tsk_trace_graph(current
);
1130 atomic_dec(&data
->disabled
);
1131 local_irq_restore(flags
);
1136 void trace_graph_return(struct ftrace_graph_ret
*trace
)
1138 struct trace_array
*tr
= &global_trace
;
1139 struct trace_array_cpu
*data
;
1140 unsigned long flags
;
1145 local_irq_save(flags
);
1146 cpu
= raw_smp_processor_id();
1147 data
= tr
->data
[cpu
];
1148 disabled
= atomic_inc_return(&data
->disabled
);
1149 if (likely(disabled
== 1)) {
1150 pc
= preempt_count();
1151 __trace_graph_return(tr
, trace
, flags
, pc
);
1154 clear_tsk_trace_graph(current
);
1155 atomic_dec(&data
->disabled
);
1156 local_irq_restore(flags
);
1158 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1160 enum trace_file_type
{
1161 TRACE_FILE_LAT_FMT
= 1,
1162 TRACE_FILE_ANNOTATE
= 2,
1165 static void trace_iterator_increment(struct trace_iterator
*iter
)
1167 /* Don't allow ftrace to trace into the ring buffers */
1168 ftrace_disable_cpu();
1171 if (iter
->buffer_iter
[iter
->cpu
])
1172 ring_buffer_read(iter
->buffer_iter
[iter
->cpu
], NULL
);
1174 ftrace_enable_cpu();
1177 static struct trace_entry
*
1178 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
)
1180 struct ring_buffer_event
*event
;
1181 struct ring_buffer_iter
*buf_iter
= iter
->buffer_iter
[cpu
];
1183 /* Don't allow ftrace to trace into the ring buffers */
1184 ftrace_disable_cpu();
1187 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1189 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
);
1191 ftrace_enable_cpu();
1193 return event
? ring_buffer_event_data(event
) : NULL
;
1196 static struct trace_entry
*
1197 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
, u64
*ent_ts
)
1199 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1200 struct trace_entry
*ent
, *next
= NULL
;
1201 int cpu_file
= iter
->cpu_file
;
1202 u64 next_ts
= 0, ts
;
1207 * If we are in a per_cpu trace file, don't bother by iterating over
1208 * all cpu and peek directly.
1210 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1211 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1213 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
);
1215 *ent_cpu
= cpu_file
;
1220 for_each_tracing_cpu(cpu
) {
1222 if (ring_buffer_empty_cpu(buffer
, cpu
))
1225 ent
= peek_next_entry(iter
, cpu
, &ts
);
1228 * Pick the entry with the smallest timestamp:
1230 if (ent
&& (!next
|| ts
< next_ts
)) {
1238 *ent_cpu
= next_cpu
;
1246 /* Find the next real entry, without updating the iterator itself */
1247 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1248 int *ent_cpu
, u64
*ent_ts
)
1250 return __find_next_entry(iter
, ent_cpu
, ent_ts
);
1253 /* Find the next real entry, and increment the iterator to the next entry */
1254 static void *find_next_entry_inc(struct trace_iterator
*iter
)
1256 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
, &iter
->ts
);
1259 trace_iterator_increment(iter
);
1261 return iter
->ent
? iter
: NULL
;
1264 static void trace_consume(struct trace_iterator
*iter
)
1266 /* Don't allow ftrace to trace into the ring buffers */
1267 ftrace_disable_cpu();
1268 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
);
1269 ftrace_enable_cpu();
1272 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1274 struct trace_iterator
*iter
= m
->private;
1280 /* can't go backwards */
1285 ent
= find_next_entry_inc(iter
);
1289 while (ent
&& iter
->idx
< i
)
1290 ent
= find_next_entry_inc(iter
);
1298 * No necessary locking here. The worst thing which can
1299 * happen is loosing events consumed at the same time
1300 * by a trace_pipe reader.
1301 * Other than that, we don't risk to crash the ring buffer
1302 * because it serializes the readers.
1304 * The current tracer is copied to avoid a global locking
1307 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1309 struct trace_iterator
*iter
= m
->private;
1310 static struct tracer
*old_tracer
;
1311 int cpu_file
= iter
->cpu_file
;
1316 /* copy the tracer to avoid using a global lock all around */
1317 mutex_lock(&trace_types_lock
);
1318 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1319 old_tracer
= current_trace
;
1320 *iter
->trace
= *current_trace
;
1322 mutex_unlock(&trace_types_lock
);
1324 atomic_inc(&trace_record_cmdline_disabled
);
1326 if (*pos
!= iter
->pos
) {
1331 ftrace_disable_cpu();
1333 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1334 for_each_tracing_cpu(cpu
)
1335 ring_buffer_iter_reset(iter
->buffer_iter
[cpu
]);
1337 ring_buffer_iter_reset(iter
->buffer_iter
[cpu_file
]);
1340 ftrace_enable_cpu();
1342 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1347 p
= s_next(m
, p
, &l
);
1353 static void s_stop(struct seq_file
*m
, void *p
)
1355 atomic_dec(&trace_record_cmdline_disabled
);
1358 static void print_lat_help_header(struct seq_file
*m
)
1360 seq_puts(m
, "# _------=> CPU# \n");
1361 seq_puts(m
, "# / _-----=> irqs-off \n");
1362 seq_puts(m
, "# | / _----=> need-resched \n");
1363 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1364 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1365 seq_puts(m
, "# |||| / \n");
1366 seq_puts(m
, "# ||||| delay \n");
1367 seq_puts(m
, "# cmd pid ||||| time | caller \n");
1368 seq_puts(m
, "# \\ / ||||| \\ | / \n");
1371 static void print_func_help_header(struct seq_file
*m
)
1373 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1374 seq_puts(m
, "# | | | | |\n");
1379 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1381 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1382 struct trace_array
*tr
= iter
->tr
;
1383 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1384 struct tracer
*type
= current_trace
;
1385 unsigned long total
;
1386 unsigned long entries
;
1387 const char *name
= "preemption";
1392 entries
= ring_buffer_entries(iter
->tr
->buffer
);
1394 ring_buffer_overruns(iter
->tr
->buffer
);
1396 seq_printf(m
, "%s latency trace v1.1.5 on %s\n",
1398 seq_puts(m
, "-----------------------------------"
1399 "---------------------------------\n");
1400 seq_printf(m
, " latency: %lu us, #%lu/%lu, CPU#%d |"
1401 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1402 nsecs_to_usecs(data
->saved_latency
),
1406 #if defined(CONFIG_PREEMPT_NONE)
1408 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1410 #elif defined(CONFIG_PREEMPT)
1415 /* These are reserved for later use */
1418 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1422 seq_puts(m
, " -----------------\n");
1423 seq_printf(m
, " | task: %.16s-%d "
1424 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1425 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1426 data
->policy
, data
->rt_priority
);
1427 seq_puts(m
, " -----------------\n");
1429 if (data
->critical_start
) {
1430 seq_puts(m
, " => started at: ");
1431 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1432 trace_print_seq(m
, &iter
->seq
);
1433 seq_puts(m
, "\n => ended at: ");
1434 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1435 trace_print_seq(m
, &iter
->seq
);
1442 static void test_cpu_buff_start(struct trace_iterator
*iter
)
1444 struct trace_seq
*s
= &iter
->seq
;
1446 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
1449 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
1452 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
1455 cpumask_set_cpu(iter
->cpu
, iter
->started
);
1456 trace_seq_printf(s
, "##### CPU %u buffer started ####\n", iter
->cpu
);
1459 static enum print_line_t
print_lat_fmt(struct trace_iterator
*iter
)
1461 struct trace_seq
*s
= &iter
->seq
;
1462 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1463 struct trace_event
*event
;
1464 struct trace_entry
*entry
= iter
->ent
;
1466 test_cpu_buff_start(iter
);
1468 event
= ftrace_find_event(entry
->type
);
1470 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1471 if (!trace_print_lat_context(iter
))
1476 return event
->latency_trace(iter
, sym_flags
);
1478 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
1481 return TRACE_TYPE_HANDLED
;
1483 return TRACE_TYPE_PARTIAL_LINE
;
1486 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
1488 struct trace_seq
*s
= &iter
->seq
;
1489 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1490 struct trace_entry
*entry
;
1491 struct trace_event
*event
;
1495 test_cpu_buff_start(iter
);
1497 event
= ftrace_find_event(entry
->type
);
1499 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1500 if (!trace_print_context(iter
))
1505 return event
->trace(iter
, sym_flags
);
1507 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
1510 return TRACE_TYPE_HANDLED
;
1512 return TRACE_TYPE_PARTIAL_LINE
;
1515 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
1517 struct trace_seq
*s
= &iter
->seq
;
1518 struct trace_entry
*entry
;
1519 struct trace_event
*event
;
1523 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1524 if (!trace_seq_printf(s
, "%d %d %llu ",
1525 entry
->pid
, iter
->cpu
, iter
->ts
))
1529 event
= ftrace_find_event(entry
->type
);
1531 return event
->raw(iter
, 0);
1533 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
1536 return TRACE_TYPE_HANDLED
;
1538 return TRACE_TYPE_PARTIAL_LINE
;
1541 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
1543 struct trace_seq
*s
= &iter
->seq
;
1544 unsigned char newline
= '\n';
1545 struct trace_entry
*entry
;
1546 struct trace_event
*event
;
1550 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1551 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1552 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1553 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
1556 event
= ftrace_find_event(entry
->type
);
1558 enum print_line_t ret
= event
->hex(iter
, 0);
1559 if (ret
!= TRACE_TYPE_HANDLED
)
1563 SEQ_PUT_FIELD_RET(s
, newline
);
1565 return TRACE_TYPE_HANDLED
;
1568 static enum print_line_t
print_printk_msg_only(struct trace_iterator
*iter
)
1570 struct trace_seq
*s
= &iter
->seq
;
1571 struct trace_entry
*entry
= iter
->ent
;
1572 struct print_entry
*field
;
1575 trace_assign_type(field
, entry
);
1577 ret
= trace_seq_printf(s
, "%s", field
->buf
);
1579 return TRACE_TYPE_PARTIAL_LINE
;
1581 return TRACE_TYPE_HANDLED
;
1584 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
1586 struct trace_seq
*s
= &iter
->seq
;
1587 struct trace_entry
*entry
;
1588 struct trace_event
*event
;
1592 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1593 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1594 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
1595 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
1598 event
= ftrace_find_event(entry
->type
);
1599 return event
? event
->binary(iter
, 0) : TRACE_TYPE_HANDLED
;
1602 static int trace_empty(struct trace_iterator
*iter
)
1606 for_each_tracing_cpu(cpu
) {
1607 if (iter
->buffer_iter
[cpu
]) {
1608 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1611 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1619 static enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
1621 enum print_line_t ret
;
1623 if (iter
->trace
&& iter
->trace
->print_line
) {
1624 ret
= iter
->trace
->print_line(iter
);
1625 if (ret
!= TRACE_TYPE_UNHANDLED
)
1629 if (iter
->ent
->type
== TRACE_PRINT
&&
1630 trace_flags
& TRACE_ITER_PRINTK
&&
1631 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
1632 return print_printk_msg_only(iter
);
1634 if (trace_flags
& TRACE_ITER_BIN
)
1635 return print_bin_fmt(iter
);
1637 if (trace_flags
& TRACE_ITER_HEX
)
1638 return print_hex_fmt(iter
);
1640 if (trace_flags
& TRACE_ITER_RAW
)
1641 return print_raw_fmt(iter
);
1643 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
1644 return print_lat_fmt(iter
);
1646 return print_trace_fmt(iter
);
1649 static int s_show(struct seq_file
*m
, void *v
)
1651 struct trace_iterator
*iter
= v
;
1653 if (iter
->ent
== NULL
) {
1655 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
1658 if (iter
->trace
&& iter
->trace
->print_header
)
1659 iter
->trace
->print_header(m
);
1660 else if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1661 /* print nothing if the buffers are empty */
1662 if (trace_empty(iter
))
1664 print_trace_header(m
, iter
);
1665 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1666 print_lat_help_header(m
);
1668 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1669 print_func_help_header(m
);
1672 print_trace_line(iter
);
1673 trace_print_seq(m
, &iter
->seq
);
1679 static struct seq_operations tracer_seq_ops
= {
1686 static struct trace_iterator
*
1687 __tracing_open(struct inode
*inode
, struct file
*file
, int *ret
)
1689 long cpu_file
= (long) inode
->i_private
;
1690 struct trace_iterator
*iter
;
1694 if (tracing_disabled
) {
1699 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
1706 * We make a copy of the current tracer to avoid concurrent
1707 * changes on it while we are reading.
1709 mutex_lock(&trace_types_lock
);
1710 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
1716 *iter
->trace
= *current_trace
;
1718 if (current_trace
&& current_trace
->print_max
)
1721 iter
->tr
= &global_trace
;
1723 mutex_init(&iter
->mutex
);
1724 iter
->cpu_file
= cpu_file
;
1726 /* Notify the tracer early; before we stop tracing. */
1727 if (iter
->trace
&& iter
->trace
->open
)
1728 iter
->trace
->open(iter
);
1730 /* Annotate start of buffers if we had overruns */
1731 if (ring_buffer_overruns(iter
->tr
->buffer
))
1732 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
1734 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
1735 for_each_tracing_cpu(cpu
) {
1737 iter
->buffer_iter
[cpu
] =
1738 ring_buffer_read_start(iter
->tr
->buffer
, cpu
);
1740 if (!iter
->buffer_iter
[cpu
])
1744 cpu
= iter
->cpu_file
;
1745 iter
->buffer_iter
[cpu
] =
1746 ring_buffer_read_start(iter
->tr
->buffer
, cpu
);
1748 if (!iter
->buffer_iter
[cpu
])
1752 /* TODO stop tracer */
1753 *ret
= seq_open(file
, &tracer_seq_ops
);
1757 m
= file
->private_data
;
1760 /* stop the trace while dumping */
1763 mutex_unlock(&trace_types_lock
);
1769 for_each_tracing_cpu(cpu
) {
1770 if (iter
->buffer_iter
[cpu
])
1771 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
1774 mutex_unlock(&trace_types_lock
);
1778 return ERR_PTR(-ENOMEM
);
1781 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
1783 if (tracing_disabled
)
1786 filp
->private_data
= inode
->i_private
;
1790 static int tracing_release(struct inode
*inode
, struct file
*file
)
1792 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
1793 struct trace_iterator
*iter
= m
->private;
1796 mutex_lock(&trace_types_lock
);
1797 for_each_tracing_cpu(cpu
) {
1798 if (iter
->buffer_iter
[cpu
])
1799 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
1802 if (iter
->trace
&& iter
->trace
->close
)
1803 iter
->trace
->close(iter
);
1805 /* reenable tracing if it was previously enabled */
1807 mutex_unlock(&trace_types_lock
);
1809 seq_release(inode
, file
);
1810 mutex_destroy(&iter
->mutex
);
1816 static int tracing_open(struct inode
*inode
, struct file
*file
)
1820 __tracing_open(inode
, file
, &ret
);
1825 static int tracing_lt_open(struct inode
*inode
, struct file
*file
)
1827 struct trace_iterator
*iter
;
1830 iter
= __tracing_open(inode
, file
, &ret
);
1833 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
1840 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1842 struct tracer
*t
= m
->private;
1854 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
1856 struct tracer
*t
= m
->private;
1859 mutex_lock(&trace_types_lock
);
1860 for (; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
1866 static void t_stop(struct seq_file
*m
, void *p
)
1868 mutex_unlock(&trace_types_lock
);
1871 static int t_show(struct seq_file
*m
, void *v
)
1873 struct tracer
*t
= v
;
1878 seq_printf(m
, "%s", t
->name
);
1887 static struct seq_operations show_traces_seq_ops
= {
1894 static int show_traces_open(struct inode
*inode
, struct file
*file
)
1898 if (tracing_disabled
)
1901 ret
= seq_open(file
, &show_traces_seq_ops
);
1903 struct seq_file
*m
= file
->private_data
;
1904 m
->private = trace_types
;
1910 static struct file_operations tracing_fops
= {
1911 .open
= tracing_open
,
1913 .llseek
= seq_lseek
,
1914 .release
= tracing_release
,
1917 static struct file_operations tracing_lt_fops
= {
1918 .open
= tracing_lt_open
,
1920 .llseek
= seq_lseek
,
1921 .release
= tracing_release
,
1924 static struct file_operations show_traces_fops
= {
1925 .open
= show_traces_open
,
1927 .release
= seq_release
,
1931 * Only trace on a CPU if the bitmask is set:
1933 static cpumask_var_t tracing_cpumask
;
1936 * The tracer itself will not take this lock, but still we want
1937 * to provide a consistent cpumask to user-space:
1939 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
1942 * Temporary storage for the character representation of the
1943 * CPU bitmask (and one more byte for the newline):
1945 static char mask_str
[NR_CPUS
+ 1];
1948 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
1949 size_t count
, loff_t
*ppos
)
1953 mutex_lock(&tracing_cpumask_update_lock
);
1955 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
1956 if (count
- len
< 2) {
1960 len
+= sprintf(mask_str
+ len
, "\n");
1961 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
1964 mutex_unlock(&tracing_cpumask_update_lock
);
1970 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
1971 size_t count
, loff_t
*ppos
)
1974 cpumask_var_t tracing_cpumask_new
;
1976 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
1979 mutex_lock(&tracing_cpumask_update_lock
);
1980 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
1984 local_irq_disable();
1985 __raw_spin_lock(&ftrace_max_lock
);
1986 for_each_tracing_cpu(cpu
) {
1988 * Increase/decrease the disabled counter if we are
1989 * about to flip a bit in the cpumask:
1991 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
1992 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
1993 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
1995 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
1996 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
1997 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2000 __raw_spin_unlock(&ftrace_max_lock
);
2003 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2005 mutex_unlock(&tracing_cpumask_update_lock
);
2006 free_cpumask_var(tracing_cpumask_new
);
2011 mutex_unlock(&tracing_cpumask_update_lock
);
2012 free_cpumask_var(tracing_cpumask
);
2017 static struct file_operations tracing_cpumask_fops
= {
2018 .open
= tracing_open_generic
,
2019 .read
= tracing_cpumask_read
,
2020 .write
= tracing_cpumask_write
,
2024 tracing_trace_options_read(struct file
*filp
, char __user
*ubuf
,
2025 size_t cnt
, loff_t
*ppos
)
2031 u32 tracer_flags
= current_trace
->flags
->val
;
2032 struct tracer_opt
*trace_opts
= current_trace
->flags
->opts
;
2035 /* calculate max size */
2036 for (i
= 0; trace_options
[i
]; i
++) {
2037 len
+= strlen(trace_options
[i
]);
2038 len
+= 3; /* "no" and space */
2042 * Increase the size with names of options specific
2043 * of the current tracer.
2045 for (i
= 0; trace_opts
[i
].name
; i
++) {
2046 len
+= strlen(trace_opts
[i
].name
);
2047 len
+= 3; /* "no" and space */
2050 /* +2 for \n and \0 */
2051 buf
= kmalloc(len
+ 2, GFP_KERNEL
);
2055 for (i
= 0; trace_options
[i
]; i
++) {
2056 if (trace_flags
& (1 << i
))
2057 r
+= sprintf(buf
+ r
, "%s ", trace_options
[i
]);
2059 r
+= sprintf(buf
+ r
, "no%s ", trace_options
[i
]);
2062 for (i
= 0; trace_opts
[i
].name
; i
++) {
2063 if (tracer_flags
& trace_opts
[i
].bit
)
2064 r
+= sprintf(buf
+ r
, "%s ",
2065 trace_opts
[i
].name
);
2067 r
+= sprintf(buf
+ r
, "no%s ",
2068 trace_opts
[i
].name
);
2071 r
+= sprintf(buf
+ r
, "\n");
2072 WARN_ON(r
>= len
+ 2);
2074 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2081 /* Try to assign a tracer specific option */
2082 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2084 struct tracer_flags
*trace_flags
= trace
->flags
;
2085 struct tracer_opt
*opts
= NULL
;
2089 for (i
= 0; trace_flags
->opts
[i
].name
; i
++) {
2090 opts
= &trace_flags
->opts
[i
];
2091 len
= strlen(opts
->name
);
2093 if (strncmp(cmp
, opts
->name
, len
) == 0) {
2094 ret
= trace
->set_flag(trace_flags
->val
,
2100 if (!trace_flags
->opts
[i
].name
)
2103 /* Refused to handle */
2108 trace_flags
->val
&= ~opts
->bit
;
2110 trace_flags
->val
|= opts
->bit
;
2116 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2117 size_t cnt
, loff_t
*ppos
)
2125 if (cnt
>= sizeof(buf
))
2128 if (copy_from_user(&buf
, ubuf
, cnt
))
2133 if (strncmp(buf
, "no", 2) == 0) {
2138 for (i
= 0; trace_options
[i
]; i
++) {
2139 int len
= strlen(trace_options
[i
]);
2141 if (strncmp(cmp
, trace_options
[i
], len
) == 0) {
2143 trace_flags
&= ~(1 << i
);
2145 trace_flags
|= (1 << i
);
2150 /* If no option could be set, test the specific tracer options */
2151 if (!trace_options
[i
]) {
2152 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2162 static struct file_operations tracing_iter_fops
= {
2163 .open
= tracing_open_generic
,
2164 .read
= tracing_trace_options_read
,
2165 .write
= tracing_trace_options_write
,
2168 static const char readme_msg
[] =
2169 "tracing mini-HOWTO:\n\n"
2171 "# mount -t debugfs nodev /debug\n\n"
2172 "# cat /debug/tracing/available_tracers\n"
2173 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2174 "# cat /debug/tracing/current_tracer\n"
2176 "# echo sched_switch > /debug/tracing/current_tracer\n"
2177 "# cat /debug/tracing/current_tracer\n"
2179 "# cat /debug/tracing/trace_options\n"
2180 "noprint-parent nosym-offset nosym-addr noverbose\n"
2181 "# echo print-parent > /debug/tracing/trace_options\n"
2182 "# echo 1 > /debug/tracing/tracing_enabled\n"
2183 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2184 "echo 0 > /debug/tracing/tracing_enabled\n"
2188 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2189 size_t cnt
, loff_t
*ppos
)
2191 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2192 readme_msg
, strlen(readme_msg
));
2195 static struct file_operations tracing_readme_fops
= {
2196 .open
= tracing_open_generic
,
2197 .read
= tracing_readme_read
,
2201 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2202 size_t cnt
, loff_t
*ppos
)
2207 r
= sprintf(buf
, "%u\n", tracer_enabled
);
2208 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2212 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2213 size_t cnt
, loff_t
*ppos
)
2215 struct trace_array
*tr
= filp
->private_data
;
2220 if (cnt
>= sizeof(buf
))
2223 if (copy_from_user(&buf
, ubuf
, cnt
))
2228 ret
= strict_strtoul(buf
, 10, &val
);
2234 mutex_lock(&trace_types_lock
);
2235 if (tracer_enabled
^ val
) {
2238 if (current_trace
->start
)
2239 current_trace
->start(tr
);
2244 if (current_trace
->stop
)
2245 current_trace
->stop(tr
);
2248 mutex_unlock(&trace_types_lock
);
2256 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2257 size_t cnt
, loff_t
*ppos
)
2259 char buf
[max_tracer_type_len
+2];
2262 mutex_lock(&trace_types_lock
);
2264 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2266 r
= sprintf(buf
, "\n");
2267 mutex_unlock(&trace_types_lock
);
2269 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2272 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
2274 tracing_reset_online_cpus(tr
);
2278 static int tracing_set_tracer(const char *buf
)
2280 struct trace_array
*tr
= &global_trace
;
2284 mutex_lock(&trace_types_lock
);
2285 for (t
= trace_types
; t
; t
= t
->next
) {
2286 if (strcmp(t
->name
, buf
) == 0)
2293 if (t
== current_trace
)
2296 trace_branch_disable();
2297 if (current_trace
&& current_trace
->reset
)
2298 current_trace
->reset(tr
);
2302 ret
= tracer_init(t
, tr
);
2307 trace_branch_enable(tr
);
2309 mutex_unlock(&trace_types_lock
);
2315 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2316 size_t cnt
, loff_t
*ppos
)
2318 char buf
[max_tracer_type_len
+1];
2325 if (cnt
> max_tracer_type_len
)
2326 cnt
= max_tracer_type_len
;
2328 if (copy_from_user(&buf
, ubuf
, cnt
))
2333 /* strip ending whitespace. */
2334 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2337 err
= tracing_set_tracer(buf
);
2347 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2348 size_t cnt
, loff_t
*ppos
)
2350 unsigned long *ptr
= filp
->private_data
;
2354 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
2355 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2356 if (r
> sizeof(buf
))
2358 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2362 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2363 size_t cnt
, loff_t
*ppos
)
2365 unsigned long *ptr
= filp
->private_data
;
2370 if (cnt
>= sizeof(buf
))
2373 if (copy_from_user(&buf
, ubuf
, cnt
))
2378 ret
= strict_strtoul(buf
, 10, &val
);
2387 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2389 long cpu_file
= (long) inode
->i_private
;
2390 struct trace_iterator
*iter
;
2393 if (tracing_disabled
)
2396 mutex_lock(&trace_types_lock
);
2398 /* We only allow one reader per cpu */
2399 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
2400 if (!cpumask_empty(tracing_reader_cpumask
)) {
2404 cpumask_setall(tracing_reader_cpumask
);
2406 if (!cpumask_test_cpu(cpu_file
, tracing_reader_cpumask
))
2407 cpumask_set_cpu(cpu_file
, tracing_reader_cpumask
);
2414 /* create a buffer to store the information to pass to userspace */
2415 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2422 * We make a copy of the current tracer to avoid concurrent
2423 * changes on it while we are reading.
2425 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2431 *iter
->trace
= *current_trace
;
2433 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
2438 /* trace pipe does not show start of buffer */
2439 cpumask_setall(iter
->started
);
2441 iter
->cpu_file
= cpu_file
;
2442 iter
->tr
= &global_trace
;
2443 mutex_init(&iter
->mutex
);
2444 filp
->private_data
= iter
;
2446 if (iter
->trace
->pipe_open
)
2447 iter
->trace
->pipe_open(iter
);
2450 mutex_unlock(&trace_types_lock
);
2456 mutex_unlock(&trace_types_lock
);
2460 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
2462 struct trace_iterator
*iter
= file
->private_data
;
2464 mutex_lock(&trace_types_lock
);
2466 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
)
2467 cpumask_clear(tracing_reader_cpumask
);
2469 cpumask_clear_cpu(iter
->cpu_file
, tracing_reader_cpumask
);
2471 mutex_unlock(&trace_types_lock
);
2473 free_cpumask_var(iter
->started
);
2474 mutex_destroy(&iter
->mutex
);
2482 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
2484 struct trace_iterator
*iter
= filp
->private_data
;
2486 if (trace_flags
& TRACE_ITER_BLOCK
) {
2488 * Always select as readable when in blocking mode
2490 return POLLIN
| POLLRDNORM
;
2492 if (!trace_empty(iter
))
2493 return POLLIN
| POLLRDNORM
;
2494 poll_wait(filp
, &trace_wait
, poll_table
);
2495 if (!trace_empty(iter
))
2496 return POLLIN
| POLLRDNORM
;
2503 void default_wait_pipe(struct trace_iterator
*iter
)
2507 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
2509 if (trace_empty(iter
))
2512 finish_wait(&trace_wait
, &wait
);
2516 * This is a make-shift waitqueue.
2517 * A tracer might use this callback on some rare cases:
2519 * 1) the current tracer might hold the runqueue lock when it wakes up
2520 * a reader, hence a deadlock (sched, function, and function graph tracers)
2521 * 2) the function tracers, trace all functions, we don't want
2522 * the overhead of calling wake_up and friends
2523 * (and tracing them too)
2525 * Anyway, this is really very primitive wakeup.
2527 void poll_wait_pipe(struct trace_iterator
*iter
)
2529 set_current_state(TASK_INTERRUPTIBLE
);
2530 /* sleep for 100 msecs, and try again. */
2531 schedule_timeout(HZ
/ 10);
2534 /* Must be called with trace_types_lock mutex held. */
2535 static int tracing_wait_pipe(struct file
*filp
)
2537 struct trace_iterator
*iter
= filp
->private_data
;
2539 while (trace_empty(iter
)) {
2541 if ((filp
->f_flags
& O_NONBLOCK
)) {
2545 mutex_unlock(&iter
->mutex
);
2547 iter
->trace
->wait_pipe(iter
);
2549 mutex_lock(&iter
->mutex
);
2551 if (signal_pending(current
))
2555 * We block until we read something and tracing is disabled.
2556 * We still block if tracing is disabled, but we have never
2557 * read anything. This allows a user to cat this file, and
2558 * then enable tracing. But after we have read something,
2559 * we give an EOF when tracing is again disabled.
2561 * iter->pos will be 0 if we haven't read anything.
2563 if (!tracer_enabled
&& iter
->pos
)
2574 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
2575 size_t cnt
, loff_t
*ppos
)
2577 struct trace_iterator
*iter
= filp
->private_data
;
2578 static struct tracer
*old_tracer
;
2581 /* return any leftover data */
2582 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
2586 trace_seq_reset(&iter
->seq
);
2588 /* copy the tracer to avoid using a global lock all around */
2589 mutex_lock(&trace_types_lock
);
2590 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
2591 old_tracer
= current_trace
;
2592 *iter
->trace
= *current_trace
;
2594 mutex_unlock(&trace_types_lock
);
2597 * Avoid more than one consumer on a single file descriptor
2598 * This is just a matter of traces coherency, the ring buffer itself
2601 mutex_lock(&iter
->mutex
);
2602 if (iter
->trace
->read
) {
2603 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
2609 sret
= tracing_wait_pipe(filp
);
2613 /* stop when tracing is finished */
2614 if (trace_empty(iter
)) {
2619 if (cnt
>= PAGE_SIZE
)
2620 cnt
= PAGE_SIZE
- 1;
2622 /* reset all but tr, trace, and overruns */
2623 memset(&iter
->seq
, 0,
2624 sizeof(struct trace_iterator
) -
2625 offsetof(struct trace_iterator
, seq
));
2628 while (find_next_entry_inc(iter
) != NULL
) {
2629 enum print_line_t ret
;
2630 int len
= iter
->seq
.len
;
2632 ret
= print_trace_line(iter
);
2633 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
2634 /* don't print partial lines */
2635 iter
->seq
.len
= len
;
2638 if (ret
!= TRACE_TYPE_NO_CONSUME
)
2639 trace_consume(iter
);
2641 if (iter
->seq
.len
>= cnt
)
2645 /* Now copy what we have to the user */
2646 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
2647 if (iter
->seq
.readpos
>= iter
->seq
.len
)
2648 trace_seq_reset(&iter
->seq
);
2651 * If there was nothing to send to user, inspite of consuming trace
2652 * entries, go back to wait for more entries.
2658 mutex_unlock(&iter
->mutex
);
2663 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
2664 struct pipe_buffer
*buf
)
2666 __free_page(buf
->page
);
2669 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
2672 __free_page(spd
->pages
[idx
]);
2675 static struct pipe_buf_operations tracing_pipe_buf_ops
= {
2677 .map
= generic_pipe_buf_map
,
2678 .unmap
= generic_pipe_buf_unmap
,
2679 .confirm
= generic_pipe_buf_confirm
,
2680 .release
= tracing_pipe_buf_release
,
2681 .steal
= generic_pipe_buf_steal
,
2682 .get
= generic_pipe_buf_get
,
2686 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
2691 /* Seq buffer is page-sized, exactly what we need. */
2693 count
= iter
->seq
.len
;
2694 ret
= print_trace_line(iter
);
2695 count
= iter
->seq
.len
- count
;
2698 iter
->seq
.len
-= count
;
2701 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
2702 iter
->seq
.len
-= count
;
2706 trace_consume(iter
);
2708 if (!find_next_entry_inc(iter
)) {
2718 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
2720 struct pipe_inode_info
*pipe
,
2724 struct page
*pages
[PIPE_BUFFERS
];
2725 struct partial_page partial
[PIPE_BUFFERS
];
2726 struct trace_iterator
*iter
= filp
->private_data
;
2727 struct splice_pipe_desc spd
= {
2730 .nr_pages
= 0, /* This gets updated below. */
2732 .ops
= &tracing_pipe_buf_ops
,
2733 .spd_release
= tracing_spd_release_pipe
,
2735 static struct tracer
*old_tracer
;
2740 /* copy the tracer to avoid using a global lock all around */
2741 mutex_lock(&trace_types_lock
);
2742 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
2743 old_tracer
= current_trace
;
2744 *iter
->trace
= *current_trace
;
2746 mutex_unlock(&trace_types_lock
);
2748 mutex_lock(&iter
->mutex
);
2750 if (iter
->trace
->splice_read
) {
2751 ret
= iter
->trace
->splice_read(iter
, filp
,
2752 ppos
, pipe
, len
, flags
);
2757 ret
= tracing_wait_pipe(filp
);
2761 if (!iter
->ent
&& !find_next_entry_inc(iter
)) {
2766 /* Fill as many pages as possible. */
2767 for (i
= 0, rem
= len
; i
< PIPE_BUFFERS
&& rem
; i
++) {
2768 pages
[i
] = alloc_page(GFP_KERNEL
);
2772 rem
= tracing_fill_pipe_page(rem
, iter
);
2774 /* Copy the data into the page, so we can start over. */
2775 ret
= trace_seq_to_buffer(&iter
->seq
,
2776 page_address(pages
[i
]),
2779 __free_page(pages
[i
]);
2782 partial
[i
].offset
= 0;
2783 partial
[i
].len
= iter
->seq
.len
;
2785 trace_seq_reset(&iter
->seq
);
2788 mutex_unlock(&iter
->mutex
);
2792 return splice_to_pipe(pipe
, &spd
);
2795 mutex_unlock(&iter
->mutex
);
2801 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
2802 size_t cnt
, loff_t
*ppos
)
2804 struct trace_array
*tr
= filp
->private_data
;
2808 r
= sprintf(buf
, "%lu\n", tr
->entries
>> 10);
2809 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2813 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
2814 size_t cnt
, loff_t
*ppos
)
2820 if (cnt
>= sizeof(buf
))
2823 if (copy_from_user(&buf
, ubuf
, cnt
))
2828 ret
= strict_strtoul(buf
, 10, &val
);
2832 /* must have at least 1 entry */
2836 mutex_lock(&trace_types_lock
);
2840 /* disable all cpu buffers */
2841 for_each_tracing_cpu(cpu
) {
2842 if (global_trace
.data
[cpu
])
2843 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2844 if (max_tr
.data
[cpu
])
2845 atomic_inc(&max_tr
.data
[cpu
]->disabled
);
2848 /* value is in KB */
2851 if (val
!= global_trace
.entries
) {
2852 ret
= ring_buffer_resize(global_trace
.buffer
, val
);
2858 ret
= ring_buffer_resize(max_tr
.buffer
, val
);
2862 r
= ring_buffer_resize(global_trace
.buffer
,
2863 global_trace
.entries
);
2865 /* AARGH! We are left with different
2866 * size max buffer!!!! */
2868 tracing_disabled
= 1;
2873 global_trace
.entries
= val
;
2878 /* If check pages failed, return ENOMEM */
2879 if (tracing_disabled
)
2882 for_each_tracing_cpu(cpu
) {
2883 if (global_trace
.data
[cpu
])
2884 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2885 if (max_tr
.data
[cpu
])
2886 atomic_dec(&max_tr
.data
[cpu
]->disabled
);
2890 max_tr
.entries
= global_trace
.entries
;
2891 mutex_unlock(&trace_types_lock
);
2896 static int mark_printk(const char *fmt
, ...)
2900 va_start(args
, fmt
);
2901 ret
= trace_vprintk(0, -1, fmt
, args
);
2907 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
2908 size_t cnt
, loff_t
*fpos
)
2913 if (tracing_disabled
)
2916 if (cnt
> TRACE_BUF_SIZE
)
2917 cnt
= TRACE_BUF_SIZE
;
2919 buf
= kmalloc(cnt
+ 1, GFP_KERNEL
);
2923 if (copy_from_user(buf
, ubuf
, cnt
)) {
2928 /* Cut from the first nil or newline. */
2930 end
= strchr(buf
, '\n');
2934 cnt
= mark_printk("%s\n", buf
);
2941 static struct file_operations tracing_max_lat_fops
= {
2942 .open
= tracing_open_generic
,
2943 .read
= tracing_max_lat_read
,
2944 .write
= tracing_max_lat_write
,
2947 static struct file_operations tracing_ctrl_fops
= {
2948 .open
= tracing_open_generic
,
2949 .read
= tracing_ctrl_read
,
2950 .write
= tracing_ctrl_write
,
2953 static struct file_operations set_tracer_fops
= {
2954 .open
= tracing_open_generic
,
2955 .read
= tracing_set_trace_read
,
2956 .write
= tracing_set_trace_write
,
2959 static struct file_operations tracing_pipe_fops
= {
2960 .open
= tracing_open_pipe
,
2961 .poll
= tracing_poll_pipe
,
2962 .read
= tracing_read_pipe
,
2963 .splice_read
= tracing_splice_read_pipe
,
2964 .release
= tracing_release_pipe
,
2967 static struct file_operations tracing_entries_fops
= {
2968 .open
= tracing_open_generic
,
2969 .read
= tracing_entries_read
,
2970 .write
= tracing_entries_write
,
2973 static struct file_operations tracing_mark_fops
= {
2974 .open
= tracing_open_generic
,
2975 .write
= tracing_mark_write
,
2978 #ifdef CONFIG_DYNAMIC_FTRACE
2980 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
2986 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
2987 size_t cnt
, loff_t
*ppos
)
2989 static char ftrace_dyn_info_buffer
[1024];
2990 static DEFINE_MUTEX(dyn_info_mutex
);
2991 unsigned long *p
= filp
->private_data
;
2992 char *buf
= ftrace_dyn_info_buffer
;
2993 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
2996 mutex_lock(&dyn_info_mutex
);
2997 r
= sprintf(buf
, "%ld ", *p
);
2999 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
3002 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3004 mutex_unlock(&dyn_info_mutex
);
3009 static struct file_operations tracing_dyn_info_fops
= {
3010 .open
= tracing_open_generic
,
3011 .read
= tracing_read_dyn_info
,
3015 static struct dentry
*d_tracer
;
3017 struct dentry
*tracing_init_dentry(void)
3024 d_tracer
= debugfs_create_dir("tracing", NULL
);
3026 if (!d_tracer
&& !once
) {
3028 pr_warning("Could not create debugfs directory 'tracing'\n");
3035 static struct dentry
*d_percpu
;
3037 struct dentry
*tracing_dentry_percpu(void)
3040 struct dentry
*d_tracer
;
3045 d_tracer
= tracing_init_dentry();
3050 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
3052 if (!d_percpu
&& !once
) {
3054 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3061 static void tracing_init_debugfs_percpu(long cpu
)
3063 struct dentry
*d_percpu
= tracing_dentry_percpu();
3064 struct dentry
*entry
, *d_cpu
;
3065 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3068 if (cpu
> 999 || cpu
< 0)
3071 sprintf(cpu_dir
, "cpu%ld", cpu
);
3072 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
3074 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
3078 /* per cpu trace_pipe */
3079 entry
= debugfs_create_file("trace_pipe", 0444, d_cpu
,
3080 (void *) cpu
, &tracing_pipe_fops
);
3082 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3085 entry
= debugfs_create_file("trace", 0444, d_cpu
,
3086 (void *) cpu
, &tracing_fops
);
3088 pr_warning("Could not create debugfs 'trace' entry\n");
3091 #ifdef CONFIG_FTRACE_SELFTEST
3092 /* Let selftest have access to static functions in this file */
3093 #include "trace_selftest.c"
3096 static __init
int tracer_init_debugfs(void)
3098 struct dentry
*d_tracer
;
3099 struct dentry
*entry
;
3102 d_tracer
= tracing_init_dentry();
3104 entry
= debugfs_create_file("tracing_enabled", 0644, d_tracer
,
3105 &global_trace
, &tracing_ctrl_fops
);
3107 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3109 entry
= debugfs_create_file("trace_options", 0644, d_tracer
,
3110 NULL
, &tracing_iter_fops
);
3112 pr_warning("Could not create debugfs 'trace_options' entry\n");
3114 entry
= debugfs_create_file("tracing_cpumask", 0644, d_tracer
,
3115 NULL
, &tracing_cpumask_fops
);
3117 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3119 entry
= debugfs_create_file("trace", 0444, d_tracer
,
3120 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
3122 pr_warning("Could not create debugfs 'trace' entry\n");
3124 entry
= debugfs_create_file("available_tracers", 0444, d_tracer
,
3125 &global_trace
, &show_traces_fops
);
3127 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3129 entry
= debugfs_create_file("current_tracer", 0444, d_tracer
,
3130 &global_trace
, &set_tracer_fops
);
3132 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3134 entry
= debugfs_create_file("tracing_max_latency", 0644, d_tracer
,
3135 &tracing_max_latency
,
3136 &tracing_max_lat_fops
);
3138 pr_warning("Could not create debugfs "
3139 "'tracing_max_latency' entry\n");
3141 entry
= debugfs_create_file("tracing_thresh", 0644, d_tracer
,
3142 &tracing_thresh
, &tracing_max_lat_fops
);
3144 pr_warning("Could not create debugfs "
3145 "'tracing_thresh' entry\n");
3146 entry
= debugfs_create_file("README", 0644, d_tracer
,
3147 NULL
, &tracing_readme_fops
);
3149 pr_warning("Could not create debugfs 'README' entry\n");
3151 entry
= debugfs_create_file("trace_pipe", 0444, d_tracer
,
3152 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
3154 pr_warning("Could not create debugfs "
3155 "'trace_pipe' entry\n");
3157 entry
= debugfs_create_file("buffer_size_kb", 0644, d_tracer
,
3158 &global_trace
, &tracing_entries_fops
);
3160 pr_warning("Could not create debugfs "
3161 "'buffer_size_kb' entry\n");
3163 entry
= debugfs_create_file("trace_marker", 0220, d_tracer
,
3164 NULL
, &tracing_mark_fops
);
3166 pr_warning("Could not create debugfs "
3167 "'trace_marker' entry\n");
3169 #ifdef CONFIG_DYNAMIC_FTRACE
3170 entry
= debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
3171 &ftrace_update_tot_cnt
,
3172 &tracing_dyn_info_fops
);
3174 pr_warning("Could not create debugfs "
3175 "'dyn_ftrace_total_info' entry\n");
3177 #ifdef CONFIG_SYSPROF_TRACER
3178 init_tracer_sysprof_debugfs(d_tracer
);
3181 for_each_tracing_cpu(cpu
)
3182 tracing_init_debugfs_percpu(cpu
);
3187 int trace_vprintk(unsigned long ip
, int depth
, const char *fmt
, va_list args
)
3189 static DEFINE_SPINLOCK(trace_buf_lock
);
3190 static char trace_buf
[TRACE_BUF_SIZE
];
3192 struct ring_buffer_event
*event
;
3193 struct trace_array
*tr
= &global_trace
;
3194 struct trace_array_cpu
*data
;
3195 int cpu
, len
= 0, size
, pc
;
3196 struct print_entry
*entry
;
3197 unsigned long irq_flags
;
3199 if (tracing_disabled
|| tracing_selftest_running
)
3202 pc
= preempt_count();
3203 preempt_disable_notrace();
3204 cpu
= raw_smp_processor_id();
3205 data
= tr
->data
[cpu
];
3207 if (unlikely(atomic_read(&data
->disabled
)))
3210 pause_graph_tracing();
3211 spin_lock_irqsave(&trace_buf_lock
, irq_flags
);
3212 len
= vsnprintf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
3214 len
= min(len
, TRACE_BUF_SIZE
-1);
3217 size
= sizeof(*entry
) + len
+ 1;
3218 event
= trace_buffer_lock_reserve(tr
, TRACE_PRINT
, size
, irq_flags
, pc
);
3221 entry
= ring_buffer_event_data(event
);
3223 entry
->depth
= depth
;
3225 memcpy(&entry
->buf
, trace_buf
, len
);
3226 entry
->buf
[len
] = 0;
3227 ring_buffer_unlock_commit(tr
->buffer
, event
);
3230 spin_unlock_irqrestore(&trace_buf_lock
, irq_flags
);
3231 unpause_graph_tracing();
3233 preempt_enable_notrace();
3237 EXPORT_SYMBOL_GPL(trace_vprintk
);
3239 int __ftrace_printk(unsigned long ip
, const char *fmt
, ...)
3244 if (!(trace_flags
& TRACE_ITER_PRINTK
))
3248 ret
= trace_vprintk(ip
, task_curr_ret_stack(current
), fmt
, ap
);
3252 EXPORT_SYMBOL_GPL(__ftrace_printk
);
3254 int __ftrace_vprintk(unsigned long ip
, const char *fmt
, va_list ap
)
3256 if (!(trace_flags
& TRACE_ITER_PRINTK
))
3259 return trace_vprintk(ip
, task_curr_ret_stack(current
), fmt
, ap
);
3261 EXPORT_SYMBOL_GPL(__ftrace_vprintk
);
3263 static int trace_panic_handler(struct notifier_block
*this,
3264 unsigned long event
, void *unused
)
3266 if (ftrace_dump_on_oops
)
3271 static struct notifier_block trace_panic_notifier
= {
3272 .notifier_call
= trace_panic_handler
,
3274 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
3277 static int trace_die_handler(struct notifier_block
*self
,
3283 if (ftrace_dump_on_oops
)
3292 static struct notifier_block trace_die_notifier
= {
3293 .notifier_call
= trace_die_handler
,
3298 * printk is set to max of 1024, we really don't need it that big.
3299 * Nothing should be printing 1000 characters anyway.
3301 #define TRACE_MAX_PRINT 1000
3304 * Define here KERN_TRACE so that we have one place to modify
3305 * it if we decide to change what log level the ftrace dump
3308 #define KERN_TRACE KERN_EMERG
3311 trace_printk_seq(struct trace_seq
*s
)
3313 /* Probably should print a warning here. */
3317 /* should be zero ended, but we are paranoid. */
3318 s
->buffer
[s
->len
] = 0;
3320 printk(KERN_TRACE
"%s", s
->buffer
);
3325 void ftrace_dump(void)
3327 static DEFINE_SPINLOCK(ftrace_dump_lock
);
3328 /* use static because iter can be a bit big for the stack */
3329 static struct trace_iterator iter
;
3330 static int dump_ran
;
3331 unsigned long flags
;
3335 spin_lock_irqsave(&ftrace_dump_lock
, flags
);
3341 /* No turning back! */
3345 for_each_tracing_cpu(cpu
) {
3346 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
3349 /* don't look at user memory in panic mode */
3350 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
3352 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
3354 iter
.tr
= &global_trace
;
3355 iter
.trace
= current_trace
;
3358 * We need to stop all tracing on all CPUS to read the
3359 * the next buffer. This is a bit expensive, but is
3360 * not done often. We fill all what we can read,
3361 * and then release the locks again.
3364 while (!trace_empty(&iter
)) {
3367 printk(KERN_TRACE
"---------------------------------\n");
3371 /* reset all but tr, trace, and overruns */
3372 memset(&iter
.seq
, 0,
3373 sizeof(struct trace_iterator
) -
3374 offsetof(struct trace_iterator
, seq
));
3375 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
3378 if (find_next_entry_inc(&iter
) != NULL
) {
3379 print_trace_line(&iter
);
3380 trace_consume(&iter
);
3383 trace_printk_seq(&iter
.seq
);
3387 printk(KERN_TRACE
" (ftrace buffer empty)\n");
3389 printk(KERN_TRACE
"---------------------------------\n");
3392 spin_unlock_irqrestore(&ftrace_dump_lock
, flags
);
3395 __init
static int tracer_alloc_buffers(void)
3397 struct trace_array_cpu
*data
;
3401 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
3404 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
3405 goto out_free_buffer_mask
;
3407 if (!alloc_cpumask_var(&tracing_reader_cpumask
, GFP_KERNEL
))
3408 goto out_free_tracing_cpumask
;
3410 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
3411 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
3412 cpumask_clear(tracing_reader_cpumask
);
3414 /* TODO: make the number of buffers hot pluggable with CPUS */
3415 global_trace
.buffer
= ring_buffer_alloc(trace_buf_size
,
3416 TRACE_BUFFER_FLAGS
);
3417 if (!global_trace
.buffer
) {
3418 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
3420 goto out_free_cpumask
;
3422 global_trace
.entries
= ring_buffer_size(global_trace
.buffer
);
3425 #ifdef CONFIG_TRACER_MAX_TRACE
3426 max_tr
.buffer
= ring_buffer_alloc(trace_buf_size
,
3427 TRACE_BUFFER_FLAGS
);
3428 if (!max_tr
.buffer
) {
3429 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
3431 ring_buffer_free(global_trace
.buffer
);
3432 goto out_free_cpumask
;
3434 max_tr
.entries
= ring_buffer_size(max_tr
.buffer
);
3435 WARN_ON(max_tr
.entries
!= global_trace
.entries
);
3438 /* Allocate the first page for all buffers */
3439 for_each_tracing_cpu(i
) {
3440 data
= global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
3441 max_tr
.data
[i
] = &per_cpu(max_data
, i
);
3444 trace_init_cmdlines();
3446 register_tracer(&nop_trace
);
3447 current_trace
= &nop_trace
;
3448 #ifdef CONFIG_BOOT_TRACER
3449 register_tracer(&boot_tracer
);
3451 /* All seems OK, enable tracing */
3452 tracing_disabled
= 0;
3454 atomic_notifier_chain_register(&panic_notifier_list
,
3455 &trace_panic_notifier
);
3457 register_die_notifier(&trace_die_notifier
);
3461 free_cpumask_var(tracing_reader_cpumask
);
3462 out_free_tracing_cpumask
:
3463 free_cpumask_var(tracing_cpumask
);
3464 out_free_buffer_mask
:
3465 free_cpumask_var(tracing_buffer_mask
);
3470 __init
static int clear_boot_tracer(void)
3473 * The default tracer at boot buffer is an init section.
3474 * This function is called in lateinit. If we did not
3475 * find the boot tracer, then clear it out, to prevent
3476 * later registration from accessing the buffer that is
3477 * about to be freed.
3479 if (!default_bootup_tracer
)
3482 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
3483 default_bootup_tracer
);
3484 default_bootup_tracer
= NULL
;
3489 early_initcall(tracer_alloc_buffers
);
3490 fs_initcall(tracer_init_debugfs
);
3491 late_initcall(clear_boot_tracer
);