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
;
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
[] = {
323 * ftrace_max_lock is used to protect the swapping of buffers
324 * when taking a max snapshot. The buffers themselves are
325 * protected by per_cpu spinlocks. But the action of the swap
326 * needs its own lock.
328 * This is defined as a raw_spinlock_t in order to help
329 * with performance when lockdep debugging is enabled.
331 static raw_spinlock_t ftrace_max_lock
=
332 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
335 * Copy the new maximum trace into the separate maximum-trace
336 * structure. (this way the maximum trace is permanently saved,
337 * for later retrieval via /debugfs/tracing/latency_trace)
340 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
342 struct trace_array_cpu
*data
= tr
->data
[cpu
];
345 max_tr
.time_start
= data
->preempt_timestamp
;
347 data
= max_tr
.data
[cpu
];
348 data
->saved_latency
= tracing_max_latency
;
350 memcpy(data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
351 data
->pid
= tsk
->pid
;
352 data
->uid
= task_uid(tsk
);
353 data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
354 data
->policy
= tsk
->policy
;
355 data
->rt_priority
= tsk
->rt_priority
;
357 /* record this tasks comm */
358 tracing_record_cmdline(tsk
);
361 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
369 if (s
->len
<= s
->readpos
)
372 len
= s
->len
- s
->readpos
;
375 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
385 ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
390 if (s
->len
<= s
->readpos
)
393 len
= s
->len
- s
->readpos
;
396 ret
= memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
405 trace_print_seq(struct seq_file
*m
, struct trace_seq
*s
)
407 int len
= s
->len
>= PAGE_SIZE
? PAGE_SIZE
- 1 : s
->len
;
410 seq_puts(m
, s
->buffer
);
416 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
418 * @tsk: the task with the latency
419 * @cpu: The cpu that initiated the trace.
421 * Flip the buffers between the @tr and the max_tr and record information
422 * about which task was the cause of this latency.
425 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
427 struct ring_buffer
*buf
= tr
->buffer
;
429 WARN_ON_ONCE(!irqs_disabled());
430 __raw_spin_lock(&ftrace_max_lock
);
432 tr
->buffer
= max_tr
.buffer
;
435 ftrace_disable_cpu();
436 ring_buffer_reset(tr
->buffer
);
439 __update_max_tr(tr
, tsk
, cpu
);
440 __raw_spin_unlock(&ftrace_max_lock
);
444 * update_max_tr_single - only copy one trace over, and reset the rest
446 * @tsk - task with the latency
447 * @cpu - the cpu of the buffer to copy.
449 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
452 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
456 WARN_ON_ONCE(!irqs_disabled());
457 __raw_spin_lock(&ftrace_max_lock
);
459 ftrace_disable_cpu();
461 ring_buffer_reset(max_tr
.buffer
);
462 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
466 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
);
468 __update_max_tr(tr
, tsk
, cpu
);
469 __raw_spin_unlock(&ftrace_max_lock
);
473 * register_tracer - register a tracer with the ftrace system.
474 * @type - the plugin for the tracer
476 * Register a new plugin tracer.
478 int register_tracer(struct tracer
*type
)
479 __releases(kernel_lock
)
480 __acquires(kernel_lock
)
487 pr_info("Tracer must have a name\n");
492 * When this gets called we hold the BKL which means that
493 * preemption is disabled. Various trace selftests however
494 * need to disable and enable preemption for successful tests.
495 * So we drop the BKL here and grab it after the tests again.
498 mutex_lock(&trace_types_lock
);
500 tracing_selftest_running
= true;
502 for (t
= trace_types
; t
; t
= t
->next
) {
503 if (strcmp(type
->name
, t
->name
) == 0) {
505 pr_info("Trace %s already registered\n",
513 type
->set_flag
= &dummy_set_flag
;
515 type
->flags
= &dummy_tracer_flags
;
517 if (!type
->flags
->opts
)
518 type
->flags
->opts
= dummy_tracer_opt
;
519 if (!type
->wait_pipe
)
520 type
->wait_pipe
= default_wait_pipe
;
523 #ifdef CONFIG_FTRACE_STARTUP_TEST
524 if (type
->selftest
&& !tracing_selftest_disabled
) {
525 struct tracer
*saved_tracer
= current_trace
;
526 struct trace_array
*tr
= &global_trace
;
530 * Run a selftest on this tracer.
531 * Here we reset the trace buffer, and set the current
532 * tracer to be this tracer. The tracer can then run some
533 * internal tracing to verify that everything is in order.
534 * If we fail, we do not register this tracer.
536 for_each_tracing_cpu(i
)
537 tracing_reset(tr
, i
);
539 current_trace
= type
;
540 /* the test is responsible for initializing and enabling */
541 pr_info("Testing tracer %s: ", type
->name
);
542 ret
= type
->selftest(type
, tr
);
543 /* the test is responsible for resetting too */
544 current_trace
= saved_tracer
;
546 printk(KERN_CONT
"FAILED!\n");
549 /* Only reset on passing, to avoid touching corrupted buffers */
550 for_each_tracing_cpu(i
)
551 tracing_reset(tr
, i
);
553 printk(KERN_CONT
"PASSED\n");
557 type
->next
= trace_types
;
559 len
= strlen(type
->name
);
560 if (len
> max_tracer_type_len
)
561 max_tracer_type_len
= len
;
564 tracing_selftest_running
= false;
565 mutex_unlock(&trace_types_lock
);
567 if (ret
|| !default_bootup_tracer
)
570 if (strncmp(default_bootup_tracer
, type
->name
, BOOTUP_TRACER_SIZE
))
573 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
574 /* Do we want this tracer to start on bootup? */
575 tracing_set_tracer(type
->name
);
576 default_bootup_tracer
= NULL
;
577 /* disable other selftests, since this will break it. */
578 tracing_selftest_disabled
= 1;
579 #ifdef CONFIG_FTRACE_STARTUP_TEST
580 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
589 void unregister_tracer(struct tracer
*type
)
594 mutex_lock(&trace_types_lock
);
595 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
599 pr_info("Trace %s not registered\n", type
->name
);
605 if (type
== current_trace
&& tracer_enabled
) {
608 if (current_trace
->stop
)
609 current_trace
->stop(&global_trace
);
610 current_trace
= &nop_trace
;
613 if (strlen(type
->name
) != max_tracer_type_len
)
616 max_tracer_type_len
= 0;
617 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
618 len
= strlen((*t
)->name
);
619 if (len
> max_tracer_type_len
)
620 max_tracer_type_len
= len
;
623 mutex_unlock(&trace_types_lock
);
626 void tracing_reset(struct trace_array
*tr
, int cpu
)
628 ftrace_disable_cpu();
629 ring_buffer_reset_cpu(tr
->buffer
, cpu
);
633 void tracing_reset_online_cpus(struct trace_array
*tr
)
637 tr
->time_start
= ftrace_now(tr
->cpu
);
639 for_each_online_cpu(cpu
)
640 tracing_reset(tr
, cpu
);
643 #define SAVED_CMDLINES 128
644 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
645 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
646 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
647 static int cmdline_idx
;
648 static raw_spinlock_t trace_cmdline_lock
= __RAW_SPIN_LOCK_UNLOCKED
;
650 /* temporary disable recording */
651 static atomic_t trace_record_cmdline_disabled __read_mostly
;
653 static void trace_init_cmdlines(void)
655 memset(&map_pid_to_cmdline
, -1, sizeof(map_pid_to_cmdline
));
656 memset(&map_cmdline_to_pid
, -1, sizeof(map_cmdline_to_pid
));
660 static int trace_stop_count
;
661 static DEFINE_SPINLOCK(tracing_start_lock
);
664 * ftrace_off_permanent - disable all ftrace code permanently
666 * This should only be called when a serious anomally has
667 * been detected. This will turn off the function tracing,
668 * ring buffers, and other tracing utilites. It takes no
669 * locks and can be called from any context.
671 void ftrace_off_permanent(void)
673 tracing_disabled
= 1;
675 tracing_off_permanent();
679 * tracing_start - quick start of the tracer
681 * If tracing is enabled but was stopped by tracing_stop,
682 * this will start the tracer back up.
684 void tracing_start(void)
686 struct ring_buffer
*buffer
;
689 if (tracing_disabled
)
692 spin_lock_irqsave(&tracing_start_lock
, flags
);
693 if (--trace_stop_count
) {
694 if (trace_stop_count
< 0) {
695 /* Someone screwed up their debugging */
697 trace_stop_count
= 0;
703 buffer
= global_trace
.buffer
;
705 ring_buffer_record_enable(buffer
);
707 buffer
= max_tr
.buffer
;
709 ring_buffer_record_enable(buffer
);
713 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
717 * tracing_stop - quick stop of the tracer
719 * Light weight way to stop tracing. Use in conjunction with
722 void tracing_stop(void)
724 struct ring_buffer
*buffer
;
728 spin_lock_irqsave(&tracing_start_lock
, flags
);
729 if (trace_stop_count
++)
732 buffer
= global_trace
.buffer
;
734 ring_buffer_record_disable(buffer
);
736 buffer
= max_tr
.buffer
;
738 ring_buffer_record_disable(buffer
);
741 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
744 void trace_stop_cmdline_recording(void);
746 static void trace_save_cmdline(struct task_struct
*tsk
)
751 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
755 * It's not the end of the world if we don't get
756 * the lock, but we also don't want to spin
757 * nor do we want to disable interrupts,
758 * so if we miss here, then better luck next time.
760 if (!__raw_spin_trylock(&trace_cmdline_lock
))
763 idx
= map_pid_to_cmdline
[tsk
->pid
];
764 if (idx
>= SAVED_CMDLINES
) {
765 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
767 map
= map_cmdline_to_pid
[idx
];
768 if (map
<= PID_MAX_DEFAULT
)
769 map_pid_to_cmdline
[map
] = (unsigned)-1;
771 map_pid_to_cmdline
[tsk
->pid
] = idx
;
776 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
778 __raw_spin_unlock(&trace_cmdline_lock
);
781 void trace_find_cmdline(int pid
, char comm
[])
786 strcpy(comm
, "<idle>");
790 if (pid
> PID_MAX_DEFAULT
) {
791 strcpy(comm
, "<...>");
795 __raw_spin_lock(&trace_cmdline_lock
);
796 map
= map_pid_to_cmdline
[pid
];
797 if (map
>= SAVED_CMDLINES
)
800 strcpy(comm
, saved_cmdlines
[map
]);
803 __raw_spin_unlock(&trace_cmdline_lock
);
806 void tracing_record_cmdline(struct task_struct
*tsk
)
808 if (atomic_read(&trace_record_cmdline_disabled
) || !tracing_is_on())
811 trace_save_cmdline(tsk
);
815 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
818 struct task_struct
*tsk
= current
;
820 entry
->preempt_count
= pc
& 0xff;
821 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
822 entry
->tgid
= (tsk
) ? tsk
->tgid
: 0;
824 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
825 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
827 TRACE_FLAG_IRQS_NOSUPPORT
|
829 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
830 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
831 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
834 struct ring_buffer_event
*trace_buffer_lock_reserve(struct trace_array
*tr
,
837 unsigned long flags
, int pc
)
839 struct ring_buffer_event
*event
;
841 event
= ring_buffer_lock_reserve(tr
->buffer
, len
);
843 struct trace_entry
*ent
= ring_buffer_event_data(event
);
845 tracing_generic_entry_update(ent
, flags
, pc
);
851 static void ftrace_trace_stack(struct trace_array
*tr
,
852 unsigned long flags
, int skip
, int pc
);
853 static void ftrace_trace_userstack(struct trace_array
*tr
,
854 unsigned long flags
, int pc
);
856 void trace_buffer_unlock_commit(struct trace_array
*tr
,
857 struct ring_buffer_event
*event
,
858 unsigned long flags
, int pc
)
860 ring_buffer_unlock_commit(tr
->buffer
, event
);
862 ftrace_trace_stack(tr
, flags
, 6, pc
);
863 ftrace_trace_userstack(tr
, flags
, pc
);
867 struct ring_buffer_event
*
868 trace_current_buffer_lock_reserve(unsigned char type
, unsigned long len
,
869 unsigned long flags
, int pc
)
871 return trace_buffer_lock_reserve(&global_trace
,
872 type
, len
, flags
, pc
);
875 void trace_current_buffer_unlock_commit(struct ring_buffer_event
*event
,
876 unsigned long flags
, int pc
)
878 return trace_buffer_unlock_commit(&global_trace
, event
, flags
, pc
);
882 trace_function(struct trace_array
*tr
,
883 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
886 struct ring_buffer_event
*event
;
887 struct ftrace_entry
*entry
;
889 /* If we are reading the ring buffer, don't trace */
890 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
893 event
= trace_buffer_lock_reserve(tr
, TRACE_FN
, sizeof(*entry
),
897 entry
= ring_buffer_event_data(event
);
899 entry
->parent_ip
= parent_ip
;
900 ring_buffer_unlock_commit(tr
->buffer
, event
);
903 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
904 static void __trace_graph_entry(struct trace_array
*tr
,
905 struct ftrace_graph_ent
*trace
,
909 struct ring_buffer_event
*event
;
910 struct ftrace_graph_ent_entry
*entry
;
912 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
915 event
= trace_buffer_lock_reserve(&global_trace
, TRACE_GRAPH_ENT
,
916 sizeof(*entry
), flags
, pc
);
919 entry
= ring_buffer_event_data(event
);
920 entry
->graph_ent
= *trace
;
921 ring_buffer_unlock_commit(global_trace
.buffer
, event
);
924 static void __trace_graph_return(struct trace_array
*tr
,
925 struct ftrace_graph_ret
*trace
,
929 struct ring_buffer_event
*event
;
930 struct ftrace_graph_ret_entry
*entry
;
932 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
935 event
= trace_buffer_lock_reserve(&global_trace
, TRACE_GRAPH_RET
,
936 sizeof(*entry
), flags
, pc
);
939 entry
= ring_buffer_event_data(event
);
941 ring_buffer_unlock_commit(global_trace
.buffer
, event
);
946 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
947 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
950 if (likely(!atomic_read(&data
->disabled
)))
951 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
954 static void __ftrace_trace_stack(struct trace_array
*tr
,
958 #ifdef CONFIG_STACKTRACE
959 struct ring_buffer_event
*event
;
960 struct stack_entry
*entry
;
961 struct stack_trace trace
;
963 event
= trace_buffer_lock_reserve(tr
, TRACE_STACK
,
964 sizeof(*entry
), flags
, pc
);
967 entry
= ring_buffer_event_data(event
);
968 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
970 trace
.nr_entries
= 0;
971 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
973 trace
.entries
= entry
->caller
;
975 save_stack_trace(&trace
);
976 ring_buffer_unlock_commit(tr
->buffer
, event
);
980 static void ftrace_trace_stack(struct trace_array
*tr
,
984 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
987 __ftrace_trace_stack(tr
, flags
, skip
, pc
);
990 void __trace_stack(struct trace_array
*tr
,
994 __ftrace_trace_stack(tr
, flags
, skip
, pc
);
997 static void ftrace_trace_userstack(struct trace_array
*tr
,
998 unsigned long flags
, int pc
)
1000 #ifdef CONFIG_STACKTRACE
1001 struct ring_buffer_event
*event
;
1002 struct userstack_entry
*entry
;
1003 struct stack_trace trace
;
1005 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1008 event
= trace_buffer_lock_reserve(tr
, TRACE_USER_STACK
,
1009 sizeof(*entry
), flags
, pc
);
1012 entry
= ring_buffer_event_data(event
);
1014 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1016 trace
.nr_entries
= 0;
1017 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1019 trace
.entries
= entry
->caller
;
1021 save_stack_trace_user(&trace
);
1022 ring_buffer_unlock_commit(tr
->buffer
, event
);
1027 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1029 ftrace_trace_userstack(tr
, flags
, preempt_count());
1034 ftrace_trace_special(void *__tr
,
1035 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
,
1038 struct ring_buffer_event
*event
;
1039 struct trace_array
*tr
= __tr
;
1040 struct special_entry
*entry
;
1042 event
= trace_buffer_lock_reserve(tr
, TRACE_SPECIAL
,
1043 sizeof(*entry
), 0, pc
);
1046 entry
= ring_buffer_event_data(event
);
1050 trace_buffer_unlock_commit(tr
, event
, 0, pc
);
1054 __trace_special(void *__tr
, void *__data
,
1055 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1057 ftrace_trace_special(__tr
, arg1
, arg2
, arg3
, preempt_count());
1061 tracing_sched_switch_trace(struct trace_array
*tr
,
1062 struct task_struct
*prev
,
1063 struct task_struct
*next
,
1064 unsigned long flags
, int pc
)
1066 struct ring_buffer_event
*event
;
1067 struct ctx_switch_entry
*entry
;
1069 event
= trace_buffer_lock_reserve(tr
, TRACE_CTX
,
1070 sizeof(*entry
), flags
, pc
);
1073 entry
= ring_buffer_event_data(event
);
1074 entry
->prev_pid
= prev
->pid
;
1075 entry
->prev_prio
= prev
->prio
;
1076 entry
->prev_state
= prev
->state
;
1077 entry
->next_pid
= next
->pid
;
1078 entry
->next_prio
= next
->prio
;
1079 entry
->next_state
= next
->state
;
1080 entry
->next_cpu
= task_cpu(next
);
1081 trace_buffer_unlock_commit(tr
, event
, flags
, pc
);
1085 tracing_sched_wakeup_trace(struct trace_array
*tr
,
1086 struct task_struct
*wakee
,
1087 struct task_struct
*curr
,
1088 unsigned long flags
, int pc
)
1090 struct ring_buffer_event
*event
;
1091 struct ctx_switch_entry
*entry
;
1093 event
= trace_buffer_lock_reserve(tr
, TRACE_WAKE
,
1094 sizeof(*entry
), flags
, pc
);
1097 entry
= ring_buffer_event_data(event
);
1098 entry
->prev_pid
= curr
->pid
;
1099 entry
->prev_prio
= curr
->prio
;
1100 entry
->prev_state
= curr
->state
;
1101 entry
->next_pid
= wakee
->pid
;
1102 entry
->next_prio
= wakee
->prio
;
1103 entry
->next_state
= wakee
->state
;
1104 entry
->next_cpu
= task_cpu(wakee
);
1106 ring_buffer_unlock_commit(tr
->buffer
, event
);
1107 ftrace_trace_stack(tr
, flags
, 6, pc
);
1108 ftrace_trace_userstack(tr
, flags
, pc
);
1112 ftrace_special(unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1114 struct trace_array
*tr
= &global_trace
;
1115 struct trace_array_cpu
*data
;
1116 unsigned long flags
;
1120 if (tracing_disabled
)
1123 pc
= preempt_count();
1124 local_irq_save(flags
);
1125 cpu
= raw_smp_processor_id();
1126 data
= tr
->data
[cpu
];
1128 if (likely(atomic_inc_return(&data
->disabled
) == 1))
1129 ftrace_trace_special(tr
, arg1
, arg2
, arg3
, pc
);
1131 atomic_dec(&data
->disabled
);
1132 local_irq_restore(flags
);
1135 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1136 int trace_graph_entry(struct ftrace_graph_ent
*trace
)
1138 struct trace_array
*tr
= &global_trace
;
1139 struct trace_array_cpu
*data
;
1140 unsigned long flags
;
1145 if (!ftrace_trace_task(current
))
1148 if (!ftrace_graph_addr(trace
->func
))
1151 local_irq_save(flags
);
1152 cpu
= raw_smp_processor_id();
1153 data
= tr
->data
[cpu
];
1154 disabled
= atomic_inc_return(&data
->disabled
);
1155 if (likely(disabled
== 1)) {
1156 pc
= preempt_count();
1157 __trace_graph_entry(tr
, trace
, flags
, pc
);
1159 /* Only do the atomic if it is not already set */
1160 if (!test_tsk_trace_graph(current
))
1161 set_tsk_trace_graph(current
);
1162 atomic_dec(&data
->disabled
);
1163 local_irq_restore(flags
);
1168 void trace_graph_return(struct ftrace_graph_ret
*trace
)
1170 struct trace_array
*tr
= &global_trace
;
1171 struct trace_array_cpu
*data
;
1172 unsigned long flags
;
1177 local_irq_save(flags
);
1178 cpu
= raw_smp_processor_id();
1179 data
= tr
->data
[cpu
];
1180 disabled
= atomic_inc_return(&data
->disabled
);
1181 if (likely(disabled
== 1)) {
1182 pc
= preempt_count();
1183 __trace_graph_return(tr
, trace
, flags
, pc
);
1186 clear_tsk_trace_graph(current
);
1187 atomic_dec(&data
->disabled
);
1188 local_irq_restore(flags
);
1190 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1194 * trace_vbprintk - write binary msg to tracing buffer
1197 int trace_vbprintk(unsigned long ip
, int depth
, const char *fmt
, va_list args
)
1199 static raw_spinlock_t trace_buf_lock
=
1200 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
1201 static u32 trace_buf
[TRACE_BUF_SIZE
];
1203 struct ring_buffer_event
*event
;
1204 struct trace_array
*tr
= &global_trace
;
1205 struct trace_array_cpu
*data
;
1206 struct bprint_entry
*entry
;
1207 unsigned long flags
;
1209 int cpu
, len
= 0, size
, pc
;
1211 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1214 /* Don't pollute graph traces with trace_vprintk internals */
1215 pause_graph_tracing();
1217 pc
= preempt_count();
1218 resched
= ftrace_preempt_disable();
1219 cpu
= raw_smp_processor_id();
1220 data
= tr
->data
[cpu
];
1222 if (unlikely(atomic_read(&data
->disabled
)))
1225 /* Lockdep uses trace_printk for lock tracing */
1226 local_irq_save(flags
);
1227 __raw_spin_lock(&trace_buf_lock
);
1228 len
= vbin_printf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1230 if (len
> TRACE_BUF_SIZE
|| len
< 0)
1233 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1234 event
= trace_buffer_lock_reserve(tr
, TRACE_BPRINT
, size
, flags
, pc
);
1237 entry
= ring_buffer_event_data(event
);
1239 entry
->depth
= depth
;
1242 memcpy(entry
->buf
, trace_buf
, sizeof(u32
) * len
);
1243 ring_buffer_unlock_commit(tr
->buffer
, event
);
1246 __raw_spin_unlock(&trace_buf_lock
);
1247 local_irq_restore(flags
);
1250 ftrace_preempt_enable(resched
);
1251 unpause_graph_tracing();
1255 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1257 int trace_vprintk(unsigned long ip
, int depth
, const char *fmt
, va_list args
)
1259 static raw_spinlock_t trace_buf_lock
= __RAW_SPIN_LOCK_UNLOCKED
;
1260 static char trace_buf
[TRACE_BUF_SIZE
];
1262 struct ring_buffer_event
*event
;
1263 struct trace_array
*tr
= &global_trace
;
1264 struct trace_array_cpu
*data
;
1265 int cpu
, len
= 0, size
, pc
;
1266 struct print_entry
*entry
;
1267 unsigned long irq_flags
;
1269 if (tracing_disabled
|| tracing_selftest_running
)
1272 pc
= preempt_count();
1273 preempt_disable_notrace();
1274 cpu
= raw_smp_processor_id();
1275 data
= tr
->data
[cpu
];
1277 if (unlikely(atomic_read(&data
->disabled
)))
1280 pause_graph_tracing();
1281 raw_local_irq_save(irq_flags
);
1282 __raw_spin_lock(&trace_buf_lock
);
1283 len
= vsnprintf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1285 len
= min(len
, TRACE_BUF_SIZE
-1);
1288 size
= sizeof(*entry
) + len
+ 1;
1289 event
= trace_buffer_lock_reserve(tr
, TRACE_PRINT
, size
, irq_flags
, pc
);
1292 entry
= ring_buffer_event_data(event
);
1294 entry
->depth
= depth
;
1296 memcpy(&entry
->buf
, trace_buf
, len
);
1297 entry
->buf
[len
] = 0;
1298 ring_buffer_unlock_commit(tr
->buffer
, event
);
1301 __raw_spin_unlock(&trace_buf_lock
);
1302 raw_local_irq_restore(irq_flags
);
1303 unpause_graph_tracing();
1305 preempt_enable_notrace();
1309 EXPORT_SYMBOL_GPL(trace_vprintk
);
1311 enum trace_file_type
{
1312 TRACE_FILE_LAT_FMT
= 1,
1313 TRACE_FILE_ANNOTATE
= 2,
1316 static void trace_iterator_increment(struct trace_iterator
*iter
)
1318 /* Don't allow ftrace to trace into the ring buffers */
1319 ftrace_disable_cpu();
1322 if (iter
->buffer_iter
[iter
->cpu
])
1323 ring_buffer_read(iter
->buffer_iter
[iter
->cpu
], NULL
);
1325 ftrace_enable_cpu();
1328 static struct trace_entry
*
1329 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
)
1331 struct ring_buffer_event
*event
;
1332 struct ring_buffer_iter
*buf_iter
= iter
->buffer_iter
[cpu
];
1334 /* Don't allow ftrace to trace into the ring buffers */
1335 ftrace_disable_cpu();
1338 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1340 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
);
1342 ftrace_enable_cpu();
1344 return event
? ring_buffer_event_data(event
) : NULL
;
1347 static struct trace_entry
*
1348 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
, u64
*ent_ts
)
1350 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1351 struct trace_entry
*ent
, *next
= NULL
;
1352 int cpu_file
= iter
->cpu_file
;
1353 u64 next_ts
= 0, ts
;
1358 * If we are in a per_cpu trace file, don't bother by iterating over
1359 * all cpu and peek directly.
1361 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1362 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1364 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
);
1366 *ent_cpu
= cpu_file
;
1371 for_each_tracing_cpu(cpu
) {
1373 if (ring_buffer_empty_cpu(buffer
, cpu
))
1376 ent
= peek_next_entry(iter
, cpu
, &ts
);
1379 * Pick the entry with the smallest timestamp:
1381 if (ent
&& (!next
|| ts
< next_ts
)) {
1389 *ent_cpu
= next_cpu
;
1397 /* Find the next real entry, without updating the iterator itself */
1398 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1399 int *ent_cpu
, u64
*ent_ts
)
1401 return __find_next_entry(iter
, ent_cpu
, ent_ts
);
1404 /* Find the next real entry, and increment the iterator to the next entry */
1405 static void *find_next_entry_inc(struct trace_iterator
*iter
)
1407 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
, &iter
->ts
);
1410 trace_iterator_increment(iter
);
1412 return iter
->ent
? iter
: NULL
;
1415 static void trace_consume(struct trace_iterator
*iter
)
1417 /* Don't allow ftrace to trace into the ring buffers */
1418 ftrace_disable_cpu();
1419 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
);
1420 ftrace_enable_cpu();
1423 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1425 struct trace_iterator
*iter
= m
->private;
1431 /* can't go backwards */
1436 ent
= find_next_entry_inc(iter
);
1440 while (ent
&& iter
->idx
< i
)
1441 ent
= find_next_entry_inc(iter
);
1449 * No necessary locking here. The worst thing which can
1450 * happen is loosing events consumed at the same time
1451 * by a trace_pipe reader.
1452 * Other than that, we don't risk to crash the ring buffer
1453 * because it serializes the readers.
1455 * The current tracer is copied to avoid a global locking
1458 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1460 struct trace_iterator
*iter
= m
->private;
1461 static struct tracer
*old_tracer
;
1462 int cpu_file
= iter
->cpu_file
;
1467 /* copy the tracer to avoid using a global lock all around */
1468 mutex_lock(&trace_types_lock
);
1469 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1470 old_tracer
= current_trace
;
1471 *iter
->trace
= *current_trace
;
1473 mutex_unlock(&trace_types_lock
);
1475 atomic_inc(&trace_record_cmdline_disabled
);
1477 if (*pos
!= iter
->pos
) {
1482 ftrace_disable_cpu();
1484 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1485 for_each_tracing_cpu(cpu
)
1486 ring_buffer_iter_reset(iter
->buffer_iter
[cpu
]);
1488 ring_buffer_iter_reset(iter
->buffer_iter
[cpu_file
]);
1491 ftrace_enable_cpu();
1493 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1498 p
= s_next(m
, p
, &l
);
1504 static void s_stop(struct seq_file
*m
, void *p
)
1506 atomic_dec(&trace_record_cmdline_disabled
);
1509 static void print_lat_help_header(struct seq_file
*m
)
1511 seq_puts(m
, "# _------=> CPU# \n");
1512 seq_puts(m
, "# / _-----=> irqs-off \n");
1513 seq_puts(m
, "# | / _----=> need-resched \n");
1514 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1515 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1516 seq_puts(m
, "# |||| / \n");
1517 seq_puts(m
, "# ||||| delay \n");
1518 seq_puts(m
, "# cmd pid ||||| time | caller \n");
1519 seq_puts(m
, "# \\ / ||||| \\ | / \n");
1522 static void print_func_help_header(struct seq_file
*m
)
1524 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1525 seq_puts(m
, "# | | | | |\n");
1530 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1532 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1533 struct trace_array
*tr
= iter
->tr
;
1534 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1535 struct tracer
*type
= current_trace
;
1536 unsigned long total
;
1537 unsigned long entries
;
1538 const char *name
= "preemption";
1543 entries
= ring_buffer_entries(iter
->tr
->buffer
);
1545 ring_buffer_overruns(iter
->tr
->buffer
);
1547 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
1549 seq_puts(m
, "# -----------------------------------"
1550 "---------------------------------\n");
1551 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1552 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1553 nsecs_to_usecs(data
->saved_latency
),
1557 #if defined(CONFIG_PREEMPT_NONE)
1559 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1561 #elif defined(CONFIG_PREEMPT)
1566 /* These are reserved for later use */
1569 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1573 seq_puts(m
, "# -----------------\n");
1574 seq_printf(m
, "# | task: %.16s-%d "
1575 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1576 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1577 data
->policy
, data
->rt_priority
);
1578 seq_puts(m
, "# -----------------\n");
1580 if (data
->critical_start
) {
1581 seq_puts(m
, "# => started at: ");
1582 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1583 trace_print_seq(m
, &iter
->seq
);
1584 seq_puts(m
, "\n# => ended at: ");
1585 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1586 trace_print_seq(m
, &iter
->seq
);
1593 static void test_cpu_buff_start(struct trace_iterator
*iter
)
1595 struct trace_seq
*s
= &iter
->seq
;
1597 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
1600 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
1603 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
1606 cpumask_set_cpu(iter
->cpu
, iter
->started
);
1607 trace_seq_printf(s
, "##### CPU %u buffer started ####\n", iter
->cpu
);
1610 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
1612 struct trace_seq
*s
= &iter
->seq
;
1613 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1614 struct trace_entry
*entry
;
1615 struct trace_event
*event
;
1619 test_cpu_buff_start(iter
);
1621 event
= ftrace_find_event(entry
->type
);
1623 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1624 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1625 if (!trace_print_lat_context(iter
))
1628 if (!trace_print_context(iter
))
1634 return event
->trace(iter
, sym_flags
);
1636 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
1639 return TRACE_TYPE_HANDLED
;
1641 return TRACE_TYPE_PARTIAL_LINE
;
1644 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
1646 struct trace_seq
*s
= &iter
->seq
;
1647 struct trace_entry
*entry
;
1648 struct trace_event
*event
;
1652 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1653 if (!trace_seq_printf(s
, "%d %d %llu ",
1654 entry
->pid
, iter
->cpu
, iter
->ts
))
1658 event
= ftrace_find_event(entry
->type
);
1660 return event
->raw(iter
, 0);
1662 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
1665 return TRACE_TYPE_HANDLED
;
1667 return TRACE_TYPE_PARTIAL_LINE
;
1670 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
1672 struct trace_seq
*s
= &iter
->seq
;
1673 unsigned char newline
= '\n';
1674 struct trace_entry
*entry
;
1675 struct trace_event
*event
;
1679 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1680 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1681 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1682 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
1685 event
= ftrace_find_event(entry
->type
);
1687 enum print_line_t ret
= event
->hex(iter
, 0);
1688 if (ret
!= TRACE_TYPE_HANDLED
)
1692 SEQ_PUT_FIELD_RET(s
, newline
);
1694 return TRACE_TYPE_HANDLED
;
1697 static enum print_line_t
print_bprintk_msg_only(struct trace_iterator
*iter
)
1699 struct trace_seq
*s
= &iter
->seq
;
1700 struct trace_entry
*entry
= iter
->ent
;
1701 struct bprint_entry
*field
;
1704 trace_assign_type(field
, entry
);
1706 ret
= trace_seq_bprintf(s
, field
->fmt
, field
->buf
);
1708 return TRACE_TYPE_PARTIAL_LINE
;
1710 return TRACE_TYPE_HANDLED
;
1713 static enum print_line_t
print_printk_msg_only(struct trace_iterator
*iter
)
1715 struct trace_seq
*s
= &iter
->seq
;
1716 struct trace_entry
*entry
= iter
->ent
;
1717 struct print_entry
*field
;
1720 trace_assign_type(field
, entry
);
1722 ret
= trace_seq_printf(s
, "%s", field
->buf
);
1724 return TRACE_TYPE_PARTIAL_LINE
;
1726 return TRACE_TYPE_HANDLED
;
1729 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
1731 struct trace_seq
*s
= &iter
->seq
;
1732 struct trace_entry
*entry
;
1733 struct trace_event
*event
;
1737 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1738 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1739 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
1740 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
1743 event
= ftrace_find_event(entry
->type
);
1744 return event
? event
->binary(iter
, 0) : TRACE_TYPE_HANDLED
;
1747 static int trace_empty(struct trace_iterator
*iter
)
1751 /* If we are looking at one CPU buffer, only check that one */
1752 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
1753 cpu
= iter
->cpu_file
;
1754 if (iter
->buffer_iter
[cpu
]) {
1755 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1758 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1764 for_each_tracing_cpu(cpu
) {
1765 if (iter
->buffer_iter
[cpu
]) {
1766 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1769 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1777 static enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
1779 enum print_line_t ret
;
1781 if (iter
->trace
&& iter
->trace
->print_line
) {
1782 ret
= iter
->trace
->print_line(iter
);
1783 if (ret
!= TRACE_TYPE_UNHANDLED
)
1787 if (iter
->ent
->type
== TRACE_BPRINT
&&
1788 trace_flags
& TRACE_ITER_PRINTK
&&
1789 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
1790 return print_bprintk_msg_only(iter
);
1792 if (iter
->ent
->type
== TRACE_PRINT
&&
1793 trace_flags
& TRACE_ITER_PRINTK
&&
1794 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
1795 return print_printk_msg_only(iter
);
1797 if (trace_flags
& TRACE_ITER_BIN
)
1798 return print_bin_fmt(iter
);
1800 if (trace_flags
& TRACE_ITER_HEX
)
1801 return print_hex_fmt(iter
);
1803 if (trace_flags
& TRACE_ITER_RAW
)
1804 return print_raw_fmt(iter
);
1806 return print_trace_fmt(iter
);
1809 static int s_show(struct seq_file
*m
, void *v
)
1811 struct trace_iterator
*iter
= v
;
1813 if (iter
->ent
== NULL
) {
1815 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
1818 if (iter
->trace
&& iter
->trace
->print_header
)
1819 iter
->trace
->print_header(m
);
1820 else if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1821 /* print nothing if the buffers are empty */
1822 if (trace_empty(iter
))
1824 print_trace_header(m
, iter
);
1825 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1826 print_lat_help_header(m
);
1828 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1829 print_func_help_header(m
);
1832 print_trace_line(iter
);
1833 trace_print_seq(m
, &iter
->seq
);
1839 static struct seq_operations tracer_seq_ops
= {
1846 static struct trace_iterator
*
1847 __tracing_open(struct inode
*inode
, struct file
*file
)
1849 long cpu_file
= (long) inode
->i_private
;
1850 void *fail_ret
= ERR_PTR(-ENOMEM
);
1851 struct trace_iterator
*iter
;
1855 if (tracing_disabled
)
1856 return ERR_PTR(-ENODEV
);
1858 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
1860 return ERR_PTR(-ENOMEM
);
1863 * We make a copy of the current tracer to avoid concurrent
1864 * changes on it while we are reading.
1866 mutex_lock(&trace_types_lock
);
1867 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
1872 *iter
->trace
= *current_trace
;
1874 if (current_trace
&& current_trace
->print_max
)
1877 iter
->tr
= &global_trace
;
1879 mutex_init(&iter
->mutex
);
1880 iter
->cpu_file
= cpu_file
;
1882 /* Notify the tracer early; before we stop tracing. */
1883 if (iter
->trace
&& iter
->trace
->open
)
1884 iter
->trace
->open(iter
);
1886 /* Annotate start of buffers if we had overruns */
1887 if (ring_buffer_overruns(iter
->tr
->buffer
))
1888 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
1890 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
1891 for_each_tracing_cpu(cpu
) {
1893 iter
->buffer_iter
[cpu
] =
1894 ring_buffer_read_start(iter
->tr
->buffer
, cpu
);
1897 cpu
= iter
->cpu_file
;
1898 iter
->buffer_iter
[cpu
] =
1899 ring_buffer_read_start(iter
->tr
->buffer
, cpu
);
1902 /* TODO stop tracer */
1903 ret
= seq_open(file
, &tracer_seq_ops
);
1905 fail_ret
= ERR_PTR(ret
);
1909 m
= file
->private_data
;
1912 /* stop the trace while dumping */
1915 mutex_unlock(&trace_types_lock
);
1920 for_each_tracing_cpu(cpu
) {
1921 if (iter
->buffer_iter
[cpu
])
1922 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
1925 mutex_unlock(&trace_types_lock
);
1932 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
1934 if (tracing_disabled
)
1937 filp
->private_data
= inode
->i_private
;
1941 static int tracing_release(struct inode
*inode
, struct file
*file
)
1943 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
1944 struct trace_iterator
*iter
;
1947 if (!(file
->f_mode
& FMODE_READ
))
1952 mutex_lock(&trace_types_lock
);
1953 for_each_tracing_cpu(cpu
) {
1954 if (iter
->buffer_iter
[cpu
])
1955 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
1958 if (iter
->trace
&& iter
->trace
->close
)
1959 iter
->trace
->close(iter
);
1961 /* reenable tracing if it was previously enabled */
1963 mutex_unlock(&trace_types_lock
);
1965 seq_release(inode
, file
);
1966 mutex_destroy(&iter
->mutex
);
1972 static int tracing_open(struct inode
*inode
, struct file
*file
)
1974 struct trace_iterator
*iter
;
1977 /* If this file was open for write, then erase contents */
1978 if ((file
->f_mode
& FMODE_WRITE
) &&
1979 !(file
->f_flags
& O_APPEND
)) {
1980 long cpu
= (long) inode
->i_private
;
1982 if (cpu
== TRACE_PIPE_ALL_CPU
)
1983 tracing_reset_online_cpus(&global_trace
);
1985 tracing_reset(&global_trace
, cpu
);
1988 if (file
->f_mode
& FMODE_READ
) {
1989 iter
= __tracing_open(inode
, file
);
1991 ret
= PTR_ERR(iter
);
1992 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
1993 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
1999 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2001 struct tracer
*t
= m
->private;
2013 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2015 struct tracer
*t
= m
->private;
2018 mutex_lock(&trace_types_lock
);
2019 for (; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2025 static void t_stop(struct seq_file
*m
, void *p
)
2027 mutex_unlock(&trace_types_lock
);
2030 static int t_show(struct seq_file
*m
, void *v
)
2032 struct tracer
*t
= v
;
2037 seq_printf(m
, "%s", t
->name
);
2046 static struct seq_operations show_traces_seq_ops
= {
2053 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2057 if (tracing_disabled
)
2060 ret
= seq_open(file
, &show_traces_seq_ops
);
2062 struct seq_file
*m
= file
->private_data
;
2063 m
->private = trace_types
;
2070 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2071 size_t count
, loff_t
*ppos
)
2076 static const struct file_operations tracing_fops
= {
2077 .open
= tracing_open
,
2079 .write
= tracing_write_stub
,
2080 .llseek
= seq_lseek
,
2081 .release
= tracing_release
,
2084 static const struct file_operations show_traces_fops
= {
2085 .open
= show_traces_open
,
2087 .release
= seq_release
,
2091 * Only trace on a CPU if the bitmask is set:
2093 static cpumask_var_t tracing_cpumask
;
2096 * The tracer itself will not take this lock, but still we want
2097 * to provide a consistent cpumask to user-space:
2099 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2102 * Temporary storage for the character representation of the
2103 * CPU bitmask (and one more byte for the newline):
2105 static char mask_str
[NR_CPUS
+ 1];
2108 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2109 size_t count
, loff_t
*ppos
)
2113 mutex_lock(&tracing_cpumask_update_lock
);
2115 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2116 if (count
- len
< 2) {
2120 len
+= sprintf(mask_str
+ len
, "\n");
2121 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2124 mutex_unlock(&tracing_cpumask_update_lock
);
2130 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2131 size_t count
, loff_t
*ppos
)
2134 cpumask_var_t tracing_cpumask_new
;
2136 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2139 mutex_lock(&tracing_cpumask_update_lock
);
2140 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2144 local_irq_disable();
2145 __raw_spin_lock(&ftrace_max_lock
);
2146 for_each_tracing_cpu(cpu
) {
2148 * Increase/decrease the disabled counter if we are
2149 * about to flip a bit in the cpumask:
2151 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2152 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2153 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2155 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2156 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2157 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2160 __raw_spin_unlock(&ftrace_max_lock
);
2163 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2165 mutex_unlock(&tracing_cpumask_update_lock
);
2166 free_cpumask_var(tracing_cpumask_new
);
2171 mutex_unlock(&tracing_cpumask_update_lock
);
2172 free_cpumask_var(tracing_cpumask
);
2177 static const struct file_operations tracing_cpumask_fops
= {
2178 .open
= tracing_open_generic
,
2179 .read
= tracing_cpumask_read
,
2180 .write
= tracing_cpumask_write
,
2184 tracing_trace_options_read(struct file
*filp
, char __user
*ubuf
,
2185 size_t cnt
, loff_t
*ppos
)
2187 struct tracer_opt
*trace_opts
;
2195 /* calculate max size */
2196 for (i
= 0; trace_options
[i
]; i
++) {
2197 len
+= strlen(trace_options
[i
]);
2198 len
+= 3; /* "no" and newline */
2201 mutex_lock(&trace_types_lock
);
2202 tracer_flags
= current_trace
->flags
->val
;
2203 trace_opts
= current_trace
->flags
->opts
;
2206 * Increase the size with names of options specific
2207 * of the current tracer.
2209 for (i
= 0; trace_opts
[i
].name
; i
++) {
2210 len
+= strlen(trace_opts
[i
].name
);
2211 len
+= 3; /* "no" and newline */
2214 /* +2 for \n and \0 */
2215 buf
= kmalloc(len
+ 2, GFP_KERNEL
);
2217 mutex_unlock(&trace_types_lock
);
2221 for (i
= 0; trace_options
[i
]; i
++) {
2222 if (trace_flags
& (1 << i
))
2223 r
+= sprintf(buf
+ r
, "%s\n", trace_options
[i
]);
2225 r
+= sprintf(buf
+ r
, "no%s\n", trace_options
[i
]);
2228 for (i
= 0; trace_opts
[i
].name
; i
++) {
2229 if (tracer_flags
& trace_opts
[i
].bit
)
2230 r
+= sprintf(buf
+ r
, "%s\n",
2231 trace_opts
[i
].name
);
2233 r
+= sprintf(buf
+ r
, "no%s\n",
2234 trace_opts
[i
].name
);
2236 mutex_unlock(&trace_types_lock
);
2238 WARN_ON(r
>= len
+ 2);
2240 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2246 /* Try to assign a tracer specific option */
2247 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2249 struct tracer_flags
*trace_flags
= trace
->flags
;
2250 struct tracer_opt
*opts
= NULL
;
2254 for (i
= 0; trace_flags
->opts
[i
].name
; i
++) {
2255 opts
= &trace_flags
->opts
[i
];
2256 len
= strlen(opts
->name
);
2258 if (strncmp(cmp
, opts
->name
, len
) == 0) {
2259 ret
= trace
->set_flag(trace_flags
->val
,
2265 if (!trace_flags
->opts
[i
].name
)
2268 /* Refused to handle */
2273 trace_flags
->val
&= ~opts
->bit
;
2275 trace_flags
->val
|= opts
->bit
;
2280 static void set_tracer_flags(unsigned int mask
, int enabled
)
2282 /* do nothing if flag is already set */
2283 if (!!(trace_flags
& mask
) == !!enabled
)
2287 trace_flags
|= mask
;
2289 trace_flags
&= ~mask
;
2291 if (mask
== TRACE_ITER_GLOBAL_CLK
) {
2295 func
= trace_clock_global
;
2297 func
= trace_clock_local
;
2299 mutex_lock(&trace_types_lock
);
2300 ring_buffer_set_clock(global_trace
.buffer
, func
);
2303 ring_buffer_set_clock(max_tr
.buffer
, func
);
2304 mutex_unlock(&trace_types_lock
);
2309 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2310 size_t cnt
, loff_t
*ppos
)
2318 if (cnt
>= sizeof(buf
))
2321 if (copy_from_user(&buf
, ubuf
, cnt
))
2326 if (strncmp(buf
, "no", 2) == 0) {
2331 for (i
= 0; trace_options
[i
]; i
++) {
2332 int len
= strlen(trace_options
[i
]);
2334 if (strncmp(cmp
, trace_options
[i
], len
) == 0) {
2335 set_tracer_flags(1 << i
, !neg
);
2340 /* If no option could be set, test the specific tracer options */
2341 if (!trace_options
[i
]) {
2342 mutex_lock(&trace_types_lock
);
2343 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2344 mutex_unlock(&trace_types_lock
);
2354 static const struct file_operations tracing_iter_fops
= {
2355 .open
= tracing_open_generic
,
2356 .read
= tracing_trace_options_read
,
2357 .write
= tracing_trace_options_write
,
2360 static const char readme_msg
[] =
2361 "tracing mini-HOWTO:\n\n"
2363 "# mount -t debugfs nodev /debug\n\n"
2364 "# cat /debug/tracing/available_tracers\n"
2365 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2366 "# cat /debug/tracing/current_tracer\n"
2368 "# echo sched_switch > /debug/tracing/current_tracer\n"
2369 "# cat /debug/tracing/current_tracer\n"
2371 "# cat /debug/tracing/trace_options\n"
2372 "noprint-parent nosym-offset nosym-addr noverbose\n"
2373 "# echo print-parent > /debug/tracing/trace_options\n"
2374 "# echo 1 > /debug/tracing/tracing_enabled\n"
2375 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2376 "echo 0 > /debug/tracing/tracing_enabled\n"
2380 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2381 size_t cnt
, loff_t
*ppos
)
2383 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2384 readme_msg
, strlen(readme_msg
));
2387 static const struct file_operations tracing_readme_fops
= {
2388 .open
= tracing_open_generic
,
2389 .read
= tracing_readme_read
,
2393 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2394 size_t cnt
, loff_t
*ppos
)
2399 r
= sprintf(buf
, "%u\n", tracer_enabled
);
2400 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2404 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2405 size_t cnt
, loff_t
*ppos
)
2407 struct trace_array
*tr
= filp
->private_data
;
2412 if (cnt
>= sizeof(buf
))
2415 if (copy_from_user(&buf
, ubuf
, cnt
))
2420 ret
= strict_strtoul(buf
, 10, &val
);
2426 mutex_lock(&trace_types_lock
);
2427 if (tracer_enabled
^ val
) {
2430 if (current_trace
->start
)
2431 current_trace
->start(tr
);
2436 if (current_trace
->stop
)
2437 current_trace
->stop(tr
);
2440 mutex_unlock(&trace_types_lock
);
2448 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2449 size_t cnt
, loff_t
*ppos
)
2451 char buf
[max_tracer_type_len
+2];
2454 mutex_lock(&trace_types_lock
);
2456 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2458 r
= sprintf(buf
, "\n");
2459 mutex_unlock(&trace_types_lock
);
2461 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2464 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
2466 tracing_reset_online_cpus(tr
);
2470 static int tracing_resize_ring_buffer(unsigned long size
)
2475 * If kernel or user changes the size of the ring buffer
2476 * we use the size that was given, and we can forget about
2477 * expanding it later.
2479 ring_buffer_expanded
= 1;
2481 ret
= ring_buffer_resize(global_trace
.buffer
, size
);
2485 ret
= ring_buffer_resize(max_tr
.buffer
, size
);
2489 r
= ring_buffer_resize(global_trace
.buffer
,
2490 global_trace
.entries
);
2493 * AARGH! We are left with different
2494 * size max buffer!!!!
2495 * The max buffer is our "snapshot" buffer.
2496 * When a tracer needs a snapshot (one of the
2497 * latency tracers), it swaps the max buffer
2498 * with the saved snap shot. We succeeded to
2499 * update the size of the main buffer, but failed to
2500 * update the size of the max buffer. But when we tried
2501 * to reset the main buffer to the original size, we
2502 * failed there too. This is very unlikely to
2503 * happen, but if it does, warn and kill all
2507 tracing_disabled
= 1;
2512 global_trace
.entries
= size
;
2518 * tracing_update_buffers - used by tracing facility to expand ring buffers
2520 * To save on memory when the tracing is never used on a system with it
2521 * configured in. The ring buffers are set to a minimum size. But once
2522 * a user starts to use the tracing facility, then they need to grow
2523 * to their default size.
2525 * This function is to be called when a tracer is about to be used.
2527 int tracing_update_buffers(void)
2531 mutex_lock(&trace_types_lock
);
2532 if (!ring_buffer_expanded
)
2533 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2534 mutex_unlock(&trace_types_lock
);
2539 struct trace_option_dentry
;
2541 static struct trace_option_dentry
*
2542 create_trace_option_files(struct tracer
*tracer
);
2545 destroy_trace_option_files(struct trace_option_dentry
*topts
);
2547 static int tracing_set_tracer(const char *buf
)
2549 static struct trace_option_dentry
*topts
;
2550 struct trace_array
*tr
= &global_trace
;
2554 mutex_lock(&trace_types_lock
);
2556 if (!ring_buffer_expanded
) {
2557 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2563 for (t
= trace_types
; t
; t
= t
->next
) {
2564 if (strcmp(t
->name
, buf
) == 0)
2571 if (t
== current_trace
)
2574 trace_branch_disable();
2575 if (current_trace
&& current_trace
->reset
)
2576 current_trace
->reset(tr
);
2578 destroy_trace_option_files(topts
);
2582 topts
= create_trace_option_files(current_trace
);
2585 ret
= tracer_init(t
, tr
);
2590 trace_branch_enable(tr
);
2592 mutex_unlock(&trace_types_lock
);
2598 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2599 size_t cnt
, loff_t
*ppos
)
2601 char buf
[max_tracer_type_len
+1];
2608 if (cnt
> max_tracer_type_len
)
2609 cnt
= max_tracer_type_len
;
2611 if (copy_from_user(&buf
, ubuf
, cnt
))
2616 /* strip ending whitespace. */
2617 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2620 err
= tracing_set_tracer(buf
);
2630 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2631 size_t cnt
, loff_t
*ppos
)
2633 unsigned long *ptr
= filp
->private_data
;
2637 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
2638 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2639 if (r
> sizeof(buf
))
2641 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2645 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2646 size_t cnt
, loff_t
*ppos
)
2648 unsigned long *ptr
= filp
->private_data
;
2653 if (cnt
>= sizeof(buf
))
2656 if (copy_from_user(&buf
, ubuf
, cnt
))
2661 ret
= strict_strtoul(buf
, 10, &val
);
2670 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2672 long cpu_file
= (long) inode
->i_private
;
2673 struct trace_iterator
*iter
;
2676 if (tracing_disabled
)
2679 mutex_lock(&trace_types_lock
);
2681 /* We only allow one reader per cpu */
2682 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
2683 if (!cpumask_empty(tracing_reader_cpumask
)) {
2687 cpumask_setall(tracing_reader_cpumask
);
2689 if (!cpumask_test_cpu(cpu_file
, tracing_reader_cpumask
))
2690 cpumask_set_cpu(cpu_file
, tracing_reader_cpumask
);
2697 /* create a buffer to store the information to pass to userspace */
2698 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2705 * We make a copy of the current tracer to avoid concurrent
2706 * changes on it while we are reading.
2708 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2714 *iter
->trace
= *current_trace
;
2716 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
2721 /* trace pipe does not show start of buffer */
2722 cpumask_setall(iter
->started
);
2724 iter
->cpu_file
= cpu_file
;
2725 iter
->tr
= &global_trace
;
2726 mutex_init(&iter
->mutex
);
2727 filp
->private_data
= iter
;
2729 if (iter
->trace
->pipe_open
)
2730 iter
->trace
->pipe_open(iter
);
2733 mutex_unlock(&trace_types_lock
);
2739 mutex_unlock(&trace_types_lock
);
2743 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
2745 struct trace_iterator
*iter
= file
->private_data
;
2747 mutex_lock(&trace_types_lock
);
2749 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
)
2750 cpumask_clear(tracing_reader_cpumask
);
2752 cpumask_clear_cpu(iter
->cpu_file
, tracing_reader_cpumask
);
2754 mutex_unlock(&trace_types_lock
);
2756 free_cpumask_var(iter
->started
);
2757 mutex_destroy(&iter
->mutex
);
2765 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
2767 struct trace_iterator
*iter
= filp
->private_data
;
2769 if (trace_flags
& TRACE_ITER_BLOCK
) {
2771 * Always select as readable when in blocking mode
2773 return POLLIN
| POLLRDNORM
;
2775 if (!trace_empty(iter
))
2776 return POLLIN
| POLLRDNORM
;
2777 poll_wait(filp
, &trace_wait
, poll_table
);
2778 if (!trace_empty(iter
))
2779 return POLLIN
| POLLRDNORM
;
2786 void default_wait_pipe(struct trace_iterator
*iter
)
2790 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
2792 if (trace_empty(iter
))
2795 finish_wait(&trace_wait
, &wait
);
2799 * This is a make-shift waitqueue.
2800 * A tracer might use this callback on some rare cases:
2802 * 1) the current tracer might hold the runqueue lock when it wakes up
2803 * a reader, hence a deadlock (sched, function, and function graph tracers)
2804 * 2) the function tracers, trace all functions, we don't want
2805 * the overhead of calling wake_up and friends
2806 * (and tracing them too)
2808 * Anyway, this is really very primitive wakeup.
2810 void poll_wait_pipe(struct trace_iterator
*iter
)
2812 set_current_state(TASK_INTERRUPTIBLE
);
2813 /* sleep for 100 msecs, and try again. */
2814 schedule_timeout(HZ
/ 10);
2817 /* Must be called with trace_types_lock mutex held. */
2818 static int tracing_wait_pipe(struct file
*filp
)
2820 struct trace_iterator
*iter
= filp
->private_data
;
2822 while (trace_empty(iter
)) {
2824 if ((filp
->f_flags
& O_NONBLOCK
)) {
2828 mutex_unlock(&iter
->mutex
);
2830 iter
->trace
->wait_pipe(iter
);
2832 mutex_lock(&iter
->mutex
);
2834 if (signal_pending(current
))
2838 * We block until we read something and tracing is disabled.
2839 * We still block if tracing is disabled, but we have never
2840 * read anything. This allows a user to cat this file, and
2841 * then enable tracing. But after we have read something,
2842 * we give an EOF when tracing is again disabled.
2844 * iter->pos will be 0 if we haven't read anything.
2846 if (!tracer_enabled
&& iter
->pos
)
2857 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
2858 size_t cnt
, loff_t
*ppos
)
2860 struct trace_iterator
*iter
= filp
->private_data
;
2861 static struct tracer
*old_tracer
;
2864 /* return any leftover data */
2865 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
2869 trace_seq_init(&iter
->seq
);
2871 /* copy the tracer to avoid using a global lock all around */
2872 mutex_lock(&trace_types_lock
);
2873 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
2874 old_tracer
= current_trace
;
2875 *iter
->trace
= *current_trace
;
2877 mutex_unlock(&trace_types_lock
);
2880 * Avoid more than one consumer on a single file descriptor
2881 * This is just a matter of traces coherency, the ring buffer itself
2884 mutex_lock(&iter
->mutex
);
2885 if (iter
->trace
->read
) {
2886 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
2892 sret
= tracing_wait_pipe(filp
);
2896 /* stop when tracing is finished */
2897 if (trace_empty(iter
)) {
2902 if (cnt
>= PAGE_SIZE
)
2903 cnt
= PAGE_SIZE
- 1;
2905 /* reset all but tr, trace, and overruns */
2906 memset(&iter
->seq
, 0,
2907 sizeof(struct trace_iterator
) -
2908 offsetof(struct trace_iterator
, seq
));
2911 while (find_next_entry_inc(iter
) != NULL
) {
2912 enum print_line_t ret
;
2913 int len
= iter
->seq
.len
;
2915 ret
= print_trace_line(iter
);
2916 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
2917 /* don't print partial lines */
2918 iter
->seq
.len
= len
;
2921 if (ret
!= TRACE_TYPE_NO_CONSUME
)
2922 trace_consume(iter
);
2924 if (iter
->seq
.len
>= cnt
)
2928 /* Now copy what we have to the user */
2929 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
2930 if (iter
->seq
.readpos
>= iter
->seq
.len
)
2931 trace_seq_init(&iter
->seq
);
2934 * If there was nothing to send to user, inspite of consuming trace
2935 * entries, go back to wait for more entries.
2941 mutex_unlock(&iter
->mutex
);
2946 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
2947 struct pipe_buffer
*buf
)
2949 __free_page(buf
->page
);
2952 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
2955 __free_page(spd
->pages
[idx
]);
2958 static struct pipe_buf_operations tracing_pipe_buf_ops
= {
2960 .map
= generic_pipe_buf_map
,
2961 .unmap
= generic_pipe_buf_unmap
,
2962 .confirm
= generic_pipe_buf_confirm
,
2963 .release
= tracing_pipe_buf_release
,
2964 .steal
= generic_pipe_buf_steal
,
2965 .get
= generic_pipe_buf_get
,
2969 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
2974 /* Seq buffer is page-sized, exactly what we need. */
2976 count
= iter
->seq
.len
;
2977 ret
= print_trace_line(iter
);
2978 count
= iter
->seq
.len
- count
;
2981 iter
->seq
.len
-= count
;
2984 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
2985 iter
->seq
.len
-= count
;
2989 trace_consume(iter
);
2991 if (!find_next_entry_inc(iter
)) {
3001 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3003 struct pipe_inode_info
*pipe
,
3007 struct page
*pages
[PIPE_BUFFERS
];
3008 struct partial_page partial
[PIPE_BUFFERS
];
3009 struct trace_iterator
*iter
= filp
->private_data
;
3010 struct splice_pipe_desc spd
= {
3013 .nr_pages
= 0, /* This gets updated below. */
3015 .ops
= &tracing_pipe_buf_ops
,
3016 .spd_release
= tracing_spd_release_pipe
,
3018 static struct tracer
*old_tracer
;
3023 /* copy the tracer to avoid using a global lock all around */
3024 mutex_lock(&trace_types_lock
);
3025 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3026 old_tracer
= current_trace
;
3027 *iter
->trace
= *current_trace
;
3029 mutex_unlock(&trace_types_lock
);
3031 mutex_lock(&iter
->mutex
);
3033 if (iter
->trace
->splice_read
) {
3034 ret
= iter
->trace
->splice_read(iter
, filp
,
3035 ppos
, pipe
, len
, flags
);
3040 ret
= tracing_wait_pipe(filp
);
3044 if (!iter
->ent
&& !find_next_entry_inc(iter
)) {
3049 /* Fill as many pages as possible. */
3050 for (i
= 0, rem
= len
; i
< PIPE_BUFFERS
&& rem
; i
++) {
3051 pages
[i
] = alloc_page(GFP_KERNEL
);
3055 rem
= tracing_fill_pipe_page(rem
, iter
);
3057 /* Copy the data into the page, so we can start over. */
3058 ret
= trace_seq_to_buffer(&iter
->seq
,
3059 page_address(pages
[i
]),
3062 __free_page(pages
[i
]);
3065 partial
[i
].offset
= 0;
3066 partial
[i
].len
= iter
->seq
.len
;
3068 trace_seq_init(&iter
->seq
);
3071 mutex_unlock(&iter
->mutex
);
3075 return splice_to_pipe(pipe
, &spd
);
3078 mutex_unlock(&iter
->mutex
);
3084 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3085 size_t cnt
, loff_t
*ppos
)
3087 struct trace_array
*tr
= filp
->private_data
;
3091 mutex_lock(&trace_types_lock
);
3092 if (!ring_buffer_expanded
)
3093 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3095 trace_buf_size
>> 10);
3097 r
= sprintf(buf
, "%lu\n", tr
->entries
>> 10);
3098 mutex_unlock(&trace_types_lock
);
3100 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3104 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3105 size_t cnt
, loff_t
*ppos
)
3111 if (cnt
>= sizeof(buf
))
3114 if (copy_from_user(&buf
, ubuf
, cnt
))
3119 ret
= strict_strtoul(buf
, 10, &val
);
3123 /* must have at least 1 entry */
3127 mutex_lock(&trace_types_lock
);
3131 /* disable all cpu buffers */
3132 for_each_tracing_cpu(cpu
) {
3133 if (global_trace
.data
[cpu
])
3134 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
3135 if (max_tr
.data
[cpu
])
3136 atomic_inc(&max_tr
.data
[cpu
]->disabled
);
3139 /* value is in KB */
3142 if (val
!= global_trace
.entries
) {
3143 ret
= tracing_resize_ring_buffer(val
);
3152 /* If check pages failed, return ENOMEM */
3153 if (tracing_disabled
)
3156 for_each_tracing_cpu(cpu
) {
3157 if (global_trace
.data
[cpu
])
3158 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
3159 if (max_tr
.data
[cpu
])
3160 atomic_dec(&max_tr
.data
[cpu
]->disabled
);
3164 max_tr
.entries
= global_trace
.entries
;
3165 mutex_unlock(&trace_types_lock
);
3170 static int mark_printk(const char *fmt
, ...)
3174 va_start(args
, fmt
);
3175 ret
= trace_vprintk(0, -1, fmt
, args
);
3181 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3182 size_t cnt
, loff_t
*fpos
)
3187 if (tracing_disabled
)
3190 if (cnt
> TRACE_BUF_SIZE
)
3191 cnt
= TRACE_BUF_SIZE
;
3193 buf
= kmalloc(cnt
+ 1, GFP_KERNEL
);
3197 if (copy_from_user(buf
, ubuf
, cnt
)) {
3202 /* Cut from the first nil or newline. */
3204 end
= strchr(buf
, '\n');
3208 cnt
= mark_printk("%s\n", buf
);
3215 static const struct file_operations tracing_max_lat_fops
= {
3216 .open
= tracing_open_generic
,
3217 .read
= tracing_max_lat_read
,
3218 .write
= tracing_max_lat_write
,
3221 static const struct file_operations tracing_ctrl_fops
= {
3222 .open
= tracing_open_generic
,
3223 .read
= tracing_ctrl_read
,
3224 .write
= tracing_ctrl_write
,
3227 static const struct file_operations set_tracer_fops
= {
3228 .open
= tracing_open_generic
,
3229 .read
= tracing_set_trace_read
,
3230 .write
= tracing_set_trace_write
,
3233 static const struct file_operations tracing_pipe_fops
= {
3234 .open
= tracing_open_pipe
,
3235 .poll
= tracing_poll_pipe
,
3236 .read
= tracing_read_pipe
,
3237 .splice_read
= tracing_splice_read_pipe
,
3238 .release
= tracing_release_pipe
,
3241 static const struct file_operations tracing_entries_fops
= {
3242 .open
= tracing_open_generic
,
3243 .read
= tracing_entries_read
,
3244 .write
= tracing_entries_write
,
3247 static const struct file_operations tracing_mark_fops
= {
3248 .open
= tracing_open_generic
,
3249 .write
= tracing_mark_write
,
3252 struct ftrace_buffer_info
{
3253 struct trace_array
*tr
;
3259 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
3261 int cpu
= (int)(long)inode
->i_private
;
3262 struct ftrace_buffer_info
*info
;
3264 if (tracing_disabled
)
3267 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3271 info
->tr
= &global_trace
;
3273 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
);
3274 /* Force reading ring buffer for first read */
3275 info
->read
= (unsigned int)-1;
3279 filp
->private_data
= info
;
3289 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
3290 size_t count
, loff_t
*ppos
)
3292 struct ftrace_buffer_info
*info
= filp
->private_data
;
3300 /* Do we have previous read data to read? */
3301 if (info
->read
< PAGE_SIZE
)
3306 ret
= ring_buffer_read_page(info
->tr
->buffer
,
3313 pos
= ring_buffer_page_len(info
->spare
);
3315 if (pos
< PAGE_SIZE
)
3316 memset(info
->spare
+ pos
, 0, PAGE_SIZE
- pos
);
3319 size
= PAGE_SIZE
- info
->read
;
3323 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
3334 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
3336 struct ftrace_buffer_info
*info
= file
->private_data
;
3338 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
3345 struct ring_buffer
*buffer
;
3350 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
3351 struct pipe_buffer
*buf
)
3353 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3358 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3363 static int buffer_pipe_buf_steal(struct pipe_inode_info
*pipe
,
3364 struct pipe_buffer
*buf
)
3369 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
3370 struct pipe_buffer
*buf
)
3372 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3377 /* Pipe buffer operations for a buffer. */
3378 static struct pipe_buf_operations buffer_pipe_buf_ops
= {
3380 .map
= generic_pipe_buf_map
,
3381 .unmap
= generic_pipe_buf_unmap
,
3382 .confirm
= generic_pipe_buf_confirm
,
3383 .release
= buffer_pipe_buf_release
,
3384 .steal
= buffer_pipe_buf_steal
,
3385 .get
= buffer_pipe_buf_get
,
3389 * Callback from splice_to_pipe(), if we need to release some pages
3390 * at the end of the spd in case we error'ed out in filling the pipe.
3392 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
3394 struct buffer_ref
*ref
=
3395 (struct buffer_ref
*)spd
->partial
[i
].private;
3400 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3402 spd
->partial
[i
].private = 0;
3406 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
3407 struct pipe_inode_info
*pipe
, size_t len
,
3410 struct ftrace_buffer_info
*info
= file
->private_data
;
3411 struct partial_page partial
[PIPE_BUFFERS
];
3412 struct page
*pages
[PIPE_BUFFERS
];
3413 struct splice_pipe_desc spd
= {
3417 .ops
= &buffer_pipe_buf_ops
,
3418 .spd_release
= buffer_spd_release
,
3420 struct buffer_ref
*ref
;
3425 * We can't seek on a buffer input
3427 if (unlikely(*ppos
))
3431 for (i
= 0; i
< PIPE_BUFFERS
&& len
; i
++, len
-= size
) {
3435 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
3439 ref
->buffer
= info
->tr
->buffer
;
3440 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
);
3446 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
3449 ring_buffer_free_read_page(ref
->buffer
,
3456 * zero out any left over data, this is going to
3459 size
= ring_buffer_page_len(ref
->page
);
3460 if (size
< PAGE_SIZE
)
3461 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
3463 page
= virt_to_page(ref
->page
);
3465 spd
.pages
[i
] = page
;
3466 spd
.partial
[i
].len
= PAGE_SIZE
;
3467 spd
.partial
[i
].offset
= 0;
3468 spd
.partial
[i
].private = (unsigned long)ref
;
3474 /* did we read anything? */
3475 if (!spd
.nr_pages
) {
3476 if (flags
& SPLICE_F_NONBLOCK
)
3484 ret
= splice_to_pipe(pipe
, &spd
);
3489 static const struct file_operations tracing_buffers_fops
= {
3490 .open
= tracing_buffers_open
,
3491 .read
= tracing_buffers_read
,
3492 .release
= tracing_buffers_release
,
3493 .splice_read
= tracing_buffers_splice_read
,
3494 .llseek
= no_llseek
,
3497 #ifdef CONFIG_DYNAMIC_FTRACE
3499 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
3505 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
3506 size_t cnt
, loff_t
*ppos
)
3508 static char ftrace_dyn_info_buffer
[1024];
3509 static DEFINE_MUTEX(dyn_info_mutex
);
3510 unsigned long *p
= filp
->private_data
;
3511 char *buf
= ftrace_dyn_info_buffer
;
3512 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
3515 mutex_lock(&dyn_info_mutex
);
3516 r
= sprintf(buf
, "%ld ", *p
);
3518 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
3521 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3523 mutex_unlock(&dyn_info_mutex
);
3528 static const struct file_operations tracing_dyn_info_fops
= {
3529 .open
= tracing_open_generic
,
3530 .read
= tracing_read_dyn_info
,
3534 static struct dentry
*d_tracer
;
3536 struct dentry
*tracing_init_dentry(void)
3543 d_tracer
= debugfs_create_dir("tracing", NULL
);
3545 if (!d_tracer
&& !once
) {
3547 pr_warning("Could not create debugfs directory 'tracing'\n");
3554 static struct dentry
*d_percpu
;
3556 struct dentry
*tracing_dentry_percpu(void)
3559 struct dentry
*d_tracer
;
3564 d_tracer
= tracing_init_dentry();
3569 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
3571 if (!d_percpu
&& !once
) {
3573 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3580 static void tracing_init_debugfs_percpu(long cpu
)
3582 struct dentry
*d_percpu
= tracing_dentry_percpu();
3583 struct dentry
*entry
, *d_cpu
;
3584 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3587 if (cpu
> 999 || cpu
< 0)
3590 sprintf(cpu_dir
, "cpu%ld", cpu
);
3591 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
3593 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
3597 /* per cpu trace_pipe */
3598 entry
= debugfs_create_file("trace_pipe", 0444, d_cpu
,
3599 (void *) cpu
, &tracing_pipe_fops
);
3601 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3604 entry
= debugfs_create_file("trace", 0644, d_cpu
,
3605 (void *) cpu
, &tracing_fops
);
3607 pr_warning("Could not create debugfs 'trace' entry\n");
3609 entry
= debugfs_create_file("trace_pipe_raw", 0444, d_cpu
,
3610 (void *) cpu
, &tracing_buffers_fops
);
3612 pr_warning("Could not create debugfs 'trace_pipe_raw' entry\n");
3615 #ifdef CONFIG_FTRACE_SELFTEST
3616 /* Let selftest have access to static functions in this file */
3617 #include "trace_selftest.c"
3620 struct trace_option_dentry
{
3621 struct tracer_opt
*opt
;
3622 struct tracer_flags
*flags
;
3623 struct dentry
*entry
;
3627 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
3630 struct trace_option_dentry
*topt
= filp
->private_data
;
3633 if (topt
->flags
->val
& topt
->opt
->bit
)
3638 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
3642 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
3645 struct trace_option_dentry
*topt
= filp
->private_data
;
3650 if (cnt
>= sizeof(buf
))
3653 if (copy_from_user(&buf
, ubuf
, cnt
))
3658 ret
= strict_strtoul(buf
, 10, &val
);
3665 /* do nothing if already cleared */
3666 if (!(topt
->flags
->val
& topt
->opt
->bit
))
3669 mutex_lock(&trace_types_lock
);
3670 if (current_trace
->set_flag
)
3671 ret
= current_trace
->set_flag(topt
->flags
->val
,
3673 mutex_unlock(&trace_types_lock
);
3676 topt
->flags
->val
&= ~topt
->opt
->bit
;
3679 /* do nothing if already set */
3680 if (topt
->flags
->val
& topt
->opt
->bit
)
3683 mutex_lock(&trace_types_lock
);
3684 if (current_trace
->set_flag
)
3685 ret
= current_trace
->set_flag(topt
->flags
->val
,
3687 mutex_unlock(&trace_types_lock
);
3690 topt
->flags
->val
|= topt
->opt
->bit
;
3703 static const struct file_operations trace_options_fops
= {
3704 .open
= tracing_open_generic
,
3705 .read
= trace_options_read
,
3706 .write
= trace_options_write
,
3710 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
3713 long index
= (long)filp
->private_data
;
3716 if (trace_flags
& (1 << index
))
3721 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
3725 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
3728 long index
= (long)filp
->private_data
;
3733 if (cnt
>= sizeof(buf
))
3736 if (copy_from_user(&buf
, ubuf
, cnt
))
3741 ret
= strict_strtoul(buf
, 10, &val
);
3747 trace_flags
&= ~(1 << index
);
3750 trace_flags
|= 1 << index
;
3762 static const struct file_operations trace_options_core_fops
= {
3763 .open
= tracing_open_generic
,
3764 .read
= trace_options_core_read
,
3765 .write
= trace_options_core_write
,
3768 static struct dentry
*trace_options_init_dentry(void)
3770 struct dentry
*d_tracer
;
3771 static struct dentry
*t_options
;
3776 d_tracer
= tracing_init_dentry();
3780 t_options
= debugfs_create_dir("options", d_tracer
);
3782 pr_warning("Could not create debugfs directory 'options'\n");
3790 create_trace_option_file(struct trace_option_dentry
*topt
,
3791 struct tracer_flags
*flags
,
3792 struct tracer_opt
*opt
)
3794 struct dentry
*t_options
;
3795 struct dentry
*entry
;
3797 t_options
= trace_options_init_dentry();
3801 topt
->flags
= flags
;
3804 entry
= debugfs_create_file(opt
->name
, 0644, t_options
, topt
,
3805 &trace_options_fops
);
3807 topt
->entry
= entry
;
3811 static struct trace_option_dentry
*
3812 create_trace_option_files(struct tracer
*tracer
)
3814 struct trace_option_dentry
*topts
;
3815 struct tracer_flags
*flags
;
3816 struct tracer_opt
*opts
;
3822 flags
= tracer
->flags
;
3824 if (!flags
|| !flags
->opts
)
3829 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
3832 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
3836 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
3837 create_trace_option_file(&topts
[cnt
], flags
,
3844 destroy_trace_option_files(struct trace_option_dentry
*topts
)
3851 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
3852 if (topts
[cnt
].entry
)
3853 debugfs_remove(topts
[cnt
].entry
);
3859 static struct dentry
*
3860 create_trace_option_core_file(const char *option
, long index
)
3862 struct dentry
*t_options
;
3863 struct dentry
*entry
;
3865 t_options
= trace_options_init_dentry();
3869 entry
= debugfs_create_file(option
, 0644, t_options
, (void *)index
,
3870 &trace_options_core_fops
);
3875 static __init
void create_trace_options_dir(void)
3877 struct dentry
*t_options
;
3878 struct dentry
*entry
;
3881 t_options
= trace_options_init_dentry();
3885 for (i
= 0; trace_options
[i
]; i
++) {
3886 entry
= create_trace_option_core_file(trace_options
[i
], i
);
3888 pr_warning("Could not create debugfs %s entry\n",
3893 static __init
int tracer_init_debugfs(void)
3895 struct dentry
*d_tracer
;
3896 struct dentry
*entry
;
3899 d_tracer
= tracing_init_dentry();
3901 entry
= debugfs_create_file("tracing_enabled", 0644, d_tracer
,
3902 &global_trace
, &tracing_ctrl_fops
);
3904 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3906 entry
= debugfs_create_file("trace_options", 0644, d_tracer
,
3907 NULL
, &tracing_iter_fops
);
3909 pr_warning("Could not create debugfs 'trace_options' entry\n");
3911 create_trace_options_dir();
3913 entry
= debugfs_create_file("tracing_cpumask", 0644, d_tracer
,
3914 NULL
, &tracing_cpumask_fops
);
3916 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3918 entry
= debugfs_create_file("trace", 0644, d_tracer
,
3919 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
3921 pr_warning("Could not create debugfs 'trace' entry\n");
3923 entry
= debugfs_create_file("available_tracers", 0444, d_tracer
,
3924 &global_trace
, &show_traces_fops
);
3926 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3928 entry
= debugfs_create_file("current_tracer", 0444, d_tracer
,
3929 &global_trace
, &set_tracer_fops
);
3931 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3933 entry
= debugfs_create_file("tracing_max_latency", 0644, d_tracer
,
3934 &tracing_max_latency
,
3935 &tracing_max_lat_fops
);
3937 pr_warning("Could not create debugfs "
3938 "'tracing_max_latency' entry\n");
3940 entry
= debugfs_create_file("tracing_thresh", 0644, d_tracer
,
3941 &tracing_thresh
, &tracing_max_lat_fops
);
3943 pr_warning("Could not create debugfs "
3944 "'tracing_thresh' entry\n");
3945 entry
= debugfs_create_file("README", 0644, d_tracer
,
3946 NULL
, &tracing_readme_fops
);
3948 pr_warning("Could not create debugfs 'README' entry\n");
3950 entry
= debugfs_create_file("trace_pipe", 0444, d_tracer
,
3951 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
3953 pr_warning("Could not create debugfs "
3954 "'trace_pipe' entry\n");
3956 entry
= debugfs_create_file("buffer_size_kb", 0644, d_tracer
,
3957 &global_trace
, &tracing_entries_fops
);
3959 pr_warning("Could not create debugfs "
3960 "'buffer_size_kb' entry\n");
3962 entry
= debugfs_create_file("trace_marker", 0220, d_tracer
,
3963 NULL
, &tracing_mark_fops
);
3965 pr_warning("Could not create debugfs "
3966 "'trace_marker' entry\n");
3968 #ifdef CONFIG_DYNAMIC_FTRACE
3969 entry
= debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
3970 &ftrace_update_tot_cnt
,
3971 &tracing_dyn_info_fops
);
3973 pr_warning("Could not create debugfs "
3974 "'dyn_ftrace_total_info' entry\n");
3976 #ifdef CONFIG_SYSPROF_TRACER
3977 init_tracer_sysprof_debugfs(d_tracer
);
3980 for_each_tracing_cpu(cpu
)
3981 tracing_init_debugfs_percpu(cpu
);
3986 static int trace_panic_handler(struct notifier_block
*this,
3987 unsigned long event
, void *unused
)
3989 if (ftrace_dump_on_oops
)
3994 static struct notifier_block trace_panic_notifier
= {
3995 .notifier_call
= trace_panic_handler
,
3997 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
4000 static int trace_die_handler(struct notifier_block
*self
,
4006 if (ftrace_dump_on_oops
)
4015 static struct notifier_block trace_die_notifier
= {
4016 .notifier_call
= trace_die_handler
,
4021 * printk is set to max of 1024, we really don't need it that big.
4022 * Nothing should be printing 1000 characters anyway.
4024 #define TRACE_MAX_PRINT 1000
4027 * Define here KERN_TRACE so that we have one place to modify
4028 * it if we decide to change what log level the ftrace dump
4031 #define KERN_TRACE KERN_EMERG
4034 trace_printk_seq(struct trace_seq
*s
)
4036 /* Probably should print a warning here. */
4040 /* should be zero ended, but we are paranoid. */
4041 s
->buffer
[s
->len
] = 0;
4043 printk(KERN_TRACE
"%s", s
->buffer
);
4048 void ftrace_dump(void)
4050 static DEFINE_SPINLOCK(ftrace_dump_lock
);
4051 /* use static because iter can be a bit big for the stack */
4052 static struct trace_iterator iter
;
4053 static int dump_ran
;
4054 unsigned long flags
;
4058 spin_lock_irqsave(&ftrace_dump_lock
, flags
);
4064 /* No turning back! */
4068 for_each_tracing_cpu(cpu
) {
4069 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
4072 /* don't look at user memory in panic mode */
4073 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
4075 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
4077 /* Simulate the iterator */
4078 iter
.tr
= &global_trace
;
4079 iter
.trace
= current_trace
;
4080 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4083 * We need to stop all tracing on all CPUS to read the
4084 * the next buffer. This is a bit expensive, but is
4085 * not done often. We fill all what we can read,
4086 * and then release the locks again.
4089 while (!trace_empty(&iter
)) {
4092 printk(KERN_TRACE
"---------------------------------\n");
4096 /* reset all but tr, trace, and overruns */
4097 memset(&iter
.seq
, 0,
4098 sizeof(struct trace_iterator
) -
4099 offsetof(struct trace_iterator
, seq
));
4100 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
4103 if (find_next_entry_inc(&iter
) != NULL
) {
4104 print_trace_line(&iter
);
4105 trace_consume(&iter
);
4108 trace_printk_seq(&iter
.seq
);
4112 printk(KERN_TRACE
" (ftrace buffer empty)\n");
4114 printk(KERN_TRACE
"---------------------------------\n");
4117 spin_unlock_irqrestore(&ftrace_dump_lock
, flags
);
4120 __init
static int tracer_alloc_buffers(void)
4122 struct trace_array_cpu
*data
;
4127 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
4130 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
4131 goto out_free_buffer_mask
;
4133 if (!alloc_cpumask_var(&tracing_reader_cpumask
, GFP_KERNEL
))
4134 goto out_free_tracing_cpumask
;
4136 /* To save memory, keep the ring buffer size to its minimum */
4137 if (ring_buffer_expanded
)
4138 ring_buf_size
= trace_buf_size
;
4142 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
4143 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
4144 cpumask_clear(tracing_reader_cpumask
);
4146 /* TODO: make the number of buffers hot pluggable with CPUS */
4147 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
,
4148 TRACE_BUFFER_FLAGS
);
4149 if (!global_trace
.buffer
) {
4150 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
4152 goto out_free_cpumask
;
4154 global_trace
.entries
= ring_buffer_size(global_trace
.buffer
);
4157 #ifdef CONFIG_TRACER_MAX_TRACE
4158 max_tr
.buffer
= ring_buffer_alloc(ring_buf_size
,
4159 TRACE_BUFFER_FLAGS
);
4160 if (!max_tr
.buffer
) {
4161 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
4163 ring_buffer_free(global_trace
.buffer
);
4164 goto out_free_cpumask
;
4166 max_tr
.entries
= ring_buffer_size(max_tr
.buffer
);
4167 WARN_ON(max_tr
.entries
!= global_trace
.entries
);
4170 /* Allocate the first page for all buffers */
4171 for_each_tracing_cpu(i
) {
4172 data
= global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
4173 max_tr
.data
[i
] = &per_cpu(max_data
, i
);
4176 trace_init_cmdlines();
4178 register_tracer(&nop_trace
);
4179 current_trace
= &nop_trace
;
4180 #ifdef CONFIG_BOOT_TRACER
4181 register_tracer(&boot_tracer
);
4183 /* All seems OK, enable tracing */
4184 tracing_disabled
= 0;
4186 atomic_notifier_chain_register(&panic_notifier_list
,
4187 &trace_panic_notifier
);
4189 register_die_notifier(&trace_die_notifier
);
4194 free_cpumask_var(tracing_reader_cpumask
);
4195 out_free_tracing_cpumask
:
4196 free_cpumask_var(tracing_cpumask
);
4197 out_free_buffer_mask
:
4198 free_cpumask_var(tracing_buffer_mask
);
4203 __init
static int clear_boot_tracer(void)
4206 * The default tracer at boot buffer is an init section.
4207 * This function is called in lateinit. If we did not
4208 * find the boot tracer, then clear it out, to prevent
4209 * later registration from accessing the buffer that is
4210 * about to be freed.
4212 if (!default_bootup_tracer
)
4215 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
4216 default_bootup_tracer
);
4217 default_bootup_tracer
= NULL
;
4222 early_initcall(tracer_alloc_buffers
);
4223 fs_initcall(tracer_init_debugfs
);
4224 late_initcall(clear_boot_tracer
);