2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 William Lee Irwin III
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/ctype.h>
36 #include <linux/init.h>
37 #include <linux/poll.h>
38 #include <linux/gfp.h>
42 #include "trace_output.h"
44 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
50 int ring_buffer_expanded
;
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
59 static bool __read_mostly tracing_selftest_running
;
62 * If a tracer is running, we do not want to run SELFTEST.
64 bool __read_mostly tracing_selftest_disabled
;
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt
[] = {
71 static struct tracer_flags dummy_tracer_flags
= {
73 .opts
= dummy_tracer_opt
76 static int dummy_set_flag(u32 old_flags
, u32 bit
, int set
)
82 * Kill all tracing for good (never come back).
83 * It is initialized to 1 but will turn to zero if the initialization
84 * of the tracer is successful. But that is the only place that sets
87 static int tracing_disabled
= 1;
89 DEFINE_PER_CPU(int, ftrace_cpu_disabled
);
91 static inline void ftrace_disable_cpu(void)
94 __this_cpu_inc(per_cpu_var(ftrace_cpu_disabled
));
97 static inline void ftrace_enable_cpu(void)
99 __this_cpu_dec(per_cpu_var(ftrace_cpu_disabled
));
103 static cpumask_var_t __read_mostly tracing_buffer_mask
;
105 /* Define which cpu buffers are currently read in trace_pipe */
106 static cpumask_var_t tracing_reader_cpumask
;
108 #define for_each_tracing_cpu(cpu) \
109 for_each_cpu(cpu, tracing_buffer_mask)
112 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
114 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
115 * is set, then ftrace_dump is called. This will output the contents
116 * of the ftrace buffers to the console. This is very useful for
117 * capturing traces that lead to crashes and outputing it to a
120 * It is default off, but you can enable it with either specifying
121 * "ftrace_dump_on_oops" in the kernel command line, or setting
122 * /proc/sys/kernel/ftrace_dump_on_oops to true.
124 int ftrace_dump_on_oops
;
126 static int tracing_set_tracer(const char *buf
);
128 #define MAX_TRACER_SIZE 100
129 static char bootup_tracer_buf
[MAX_TRACER_SIZE
] __initdata
;
130 static char *default_bootup_tracer
;
132 static int __init
set_cmdline_ftrace(char *str
)
134 strncpy(bootup_tracer_buf
, str
, MAX_TRACER_SIZE
);
135 default_bootup_tracer
= bootup_tracer_buf
;
136 /* We are using ftrace early, expand it */
137 ring_buffer_expanded
= 1;
140 __setup("ftrace=", set_cmdline_ftrace
);
142 static int __init
set_ftrace_dump_on_oops(char *str
)
144 ftrace_dump_on_oops
= 1;
147 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops
);
149 unsigned long long ns2usecs(cycle_t nsec
)
157 * The global_trace is the descriptor that holds the tracing
158 * buffers for the live tracing. For each CPU, it contains
159 * a link list of pages that will store trace entries. The
160 * page descriptor of the pages in the memory is used to hold
161 * the link list by linking the lru item in the page descriptor
162 * to each of the pages in the buffer per CPU.
164 * For each active CPU there is a data field that holds the
165 * pages for the buffer for that CPU. Each CPU has the same number
166 * of pages allocated for its buffer.
168 static struct trace_array global_trace
;
170 static DEFINE_PER_CPU(struct trace_array_cpu
, global_trace_cpu
);
172 int filter_current_check_discard(struct ring_buffer
*buffer
,
173 struct ftrace_event_call
*call
, void *rec
,
174 struct ring_buffer_event
*event
)
176 return filter_check_discard(call
, rec
, buffer
, event
);
178 EXPORT_SYMBOL_GPL(filter_current_check_discard
);
180 cycle_t
ftrace_now(int cpu
)
184 /* Early boot up does not have a buffer yet */
185 if (!global_trace
.buffer
)
186 return trace_clock_local();
188 ts
= ring_buffer_time_stamp(global_trace
.buffer
, cpu
);
189 ring_buffer_normalize_time_stamp(global_trace
.buffer
, cpu
, &ts
);
195 * The max_tr is used to snapshot the global_trace when a maximum
196 * latency is reached. Some tracers will use this to store a maximum
197 * trace while it continues examining live traces.
199 * The buffers for the max_tr are set up the same as the global_trace.
200 * When a snapshot is taken, the link list of the max_tr is swapped
201 * with the link list of the global_trace and the buffers are reset for
202 * the global_trace so the tracing can continue.
204 static struct trace_array max_tr
;
206 static DEFINE_PER_CPU(struct trace_array_cpu
, max_tr_data
);
208 /* tracer_enabled is used to toggle activation of a tracer */
209 static int tracer_enabled
= 1;
212 * tracing_is_enabled - return tracer_enabled status
214 * This function is used by other tracers to know the status
215 * of the tracer_enabled flag. Tracers may use this function
216 * to know if it should enable their features when starting
217 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
219 int tracing_is_enabled(void)
221 return tracer_enabled
;
225 * trace_buf_size is the size in bytes that is allocated
226 * for a buffer. Note, the number of bytes is always rounded
229 * This number is purposely set to a low number of 16384.
230 * If the dump on oops happens, it will be much appreciated
231 * to not have to wait for all that output. Anyway this can be
232 * boot time and run time configurable.
234 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
236 static unsigned long trace_buf_size
= TRACE_BUF_SIZE_DEFAULT
;
238 /* trace_types holds a link list of available tracers. */
239 static struct tracer
*trace_types __read_mostly
;
241 /* current_trace points to the tracer that is currently active */
242 static struct tracer
*current_trace __read_mostly
;
245 * trace_types_lock is used to protect the trace_types list.
246 * This lock is also used to keep user access serialized.
247 * Accesses from userspace will grab this lock while userspace
248 * activities happen inside the kernel.
250 static DEFINE_MUTEX(trace_types_lock
);
252 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
253 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
255 /* trace_flags holds trace_options default values */
256 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
| TRACE_ITER_PRINTK
|
257 TRACE_ITER_ANNOTATE
| TRACE_ITER_CONTEXT_INFO
| TRACE_ITER_SLEEP_TIME
|
258 TRACE_ITER_GRAPH_TIME
;
260 static int trace_stop_count
;
261 static DEFINE_SPINLOCK(tracing_start_lock
);
264 * trace_wake_up - wake up tasks waiting for trace input
266 * Simply wakes up any task that is blocked on the trace_wait
267 * queue. These is used with trace_poll for tasks polling the trace.
269 void trace_wake_up(void)
273 if (trace_flags
& TRACE_ITER_BLOCK
)
276 * The runqueue_is_locked() can fail, but this is the best we
280 if (!runqueue_is_locked(cpu
))
281 wake_up(&trace_wait
);
285 static int __init
set_buf_size(char *str
)
287 unsigned long buf_size
;
291 buf_size
= memparse(str
, &str
);
292 /* nr_entries can not be zero */
295 trace_buf_size
= buf_size
;
298 __setup("trace_buf_size=", set_buf_size
);
300 unsigned long nsecs_to_usecs(unsigned long nsecs
)
305 /* These must match the bit postions in trace_iterator_flags */
306 static const char *trace_options
[] = {
334 { trace_clock_local
, "local" },
335 { trace_clock_global
, "global" },
341 * trace_parser_get_init - gets the buffer for trace parser
343 int trace_parser_get_init(struct trace_parser
*parser
, int size
)
345 memset(parser
, 0, sizeof(*parser
));
347 parser
->buffer
= kmalloc(size
, GFP_KERNEL
);
356 * trace_parser_put - frees the buffer for trace parser
358 void trace_parser_put(struct trace_parser
*parser
)
360 kfree(parser
->buffer
);
364 * trace_get_user - reads the user input string separated by space
365 * (matched by isspace(ch))
367 * For each string found the 'struct trace_parser' is updated,
368 * and the function returns.
370 * Returns number of bytes read.
372 * See kernel/trace/trace.h for 'struct trace_parser' details.
374 int trace_get_user(struct trace_parser
*parser
, const char __user
*ubuf
,
375 size_t cnt
, loff_t
*ppos
)
382 trace_parser_clear(parser
);
384 ret
= get_user(ch
, ubuf
++);
392 * The parser is not finished with the last write,
393 * continue reading the user input without skipping spaces.
396 /* skip white space */
397 while (cnt
&& isspace(ch
)) {
398 ret
= get_user(ch
, ubuf
++);
405 /* only spaces were written */
415 /* read the non-space input */
416 while (cnt
&& !isspace(ch
)) {
417 if (parser
->idx
< parser
->size
- 1)
418 parser
->buffer
[parser
->idx
++] = ch
;
423 ret
= get_user(ch
, ubuf
++);
430 /* We either got finished input or we have to wait for another call. */
432 parser
->buffer
[parser
->idx
] = 0;
433 parser
->cont
= false;
436 parser
->buffer
[parser
->idx
++] = ch
;
446 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
454 if (s
->len
<= s
->readpos
)
457 len
= s
->len
- s
->readpos
;
460 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
470 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
475 if (s
->len
<= s
->readpos
)
478 len
= s
->len
- s
->readpos
;
481 ret
= memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
490 * ftrace_max_lock is used to protect the swapping of buffers
491 * when taking a max snapshot. The buffers themselves are
492 * protected by per_cpu spinlocks. But the action of the swap
493 * needs its own lock.
495 * This is defined as a arch_spinlock_t in order to help
496 * with performance when lockdep debugging is enabled.
498 * It is also used in other places outside the update_max_tr
499 * so it needs to be defined outside of the
500 * CONFIG_TRACER_MAX_TRACE.
502 static arch_spinlock_t ftrace_max_lock
=
503 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
505 #ifdef CONFIG_TRACER_MAX_TRACE
506 unsigned long __read_mostly tracing_max_latency
;
507 unsigned long __read_mostly tracing_thresh
;
510 * Copy the new maximum trace into the separate maximum-trace
511 * structure. (this way the maximum trace is permanently saved,
512 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
515 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
517 struct trace_array_cpu
*data
= tr
->data
[cpu
];
518 struct trace_array_cpu
*max_data
= tr
->data
[cpu
];
521 max_tr
.time_start
= data
->preempt_timestamp
;
523 max_data
= max_tr
.data
[cpu
];
524 max_data
->saved_latency
= tracing_max_latency
;
525 max_data
->critical_start
= data
->critical_start
;
526 max_data
->critical_end
= data
->critical_end
;
528 memcpy(data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
529 max_data
->pid
= tsk
->pid
;
530 max_data
->uid
= task_uid(tsk
);
531 max_data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
532 max_data
->policy
= tsk
->policy
;
533 max_data
->rt_priority
= tsk
->rt_priority
;
535 /* record this tasks comm */
536 tracing_record_cmdline(tsk
);
540 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
542 * @tsk: the task with the latency
543 * @cpu: The cpu that initiated the trace.
545 * Flip the buffers between the @tr and the max_tr and record information
546 * about which task was the cause of this latency.
549 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
551 struct ring_buffer
*buf
= tr
->buffer
;
553 if (trace_stop_count
)
556 WARN_ON_ONCE(!irqs_disabled());
557 arch_spin_lock(&ftrace_max_lock
);
559 tr
->buffer
= max_tr
.buffer
;
562 __update_max_tr(tr
, tsk
, cpu
);
563 arch_spin_unlock(&ftrace_max_lock
);
567 * update_max_tr_single - only copy one trace over, and reset the rest
569 * @tsk - task with the latency
570 * @cpu - the cpu of the buffer to copy.
572 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
575 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
579 if (trace_stop_count
)
582 WARN_ON_ONCE(!irqs_disabled());
583 arch_spin_lock(&ftrace_max_lock
);
585 ftrace_disable_cpu();
587 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
591 * We failed to swap the buffer due to a commit taking
592 * place on this CPU. We fail to record, but we reset
593 * the max trace buffer (no one writes directly to it)
594 * and flag that it failed.
596 trace_array_printk(&max_tr
, _THIS_IP_
,
597 "Failed to swap buffers due to commit in progress\n");
602 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
&& ret
!= -EBUSY
);
604 __update_max_tr(tr
, tsk
, cpu
);
605 arch_spin_unlock(&ftrace_max_lock
);
607 #endif /* CONFIG_TRACER_MAX_TRACE */
610 * register_tracer - register a tracer with the ftrace system.
611 * @type - the plugin for the tracer
613 * Register a new plugin tracer.
615 int register_tracer(struct tracer
*type
)
616 __releases(kernel_lock
)
617 __acquires(kernel_lock
)
623 pr_info("Tracer must have a name\n");
627 if (strlen(type
->name
) > MAX_TRACER_SIZE
) {
628 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE
);
633 * When this gets called we hold the BKL which means that
634 * preemption is disabled. Various trace selftests however
635 * need to disable and enable preemption for successful tests.
636 * So we drop the BKL here and grab it after the tests again.
639 mutex_lock(&trace_types_lock
);
641 tracing_selftest_running
= true;
643 for (t
= trace_types
; t
; t
= t
->next
) {
644 if (strcmp(type
->name
, t
->name
) == 0) {
646 pr_info("Tracer %s already registered\n",
654 type
->set_flag
= &dummy_set_flag
;
656 type
->flags
= &dummy_tracer_flags
;
658 if (!type
->flags
->opts
)
659 type
->flags
->opts
= dummy_tracer_opt
;
660 if (!type
->wait_pipe
)
661 type
->wait_pipe
= default_wait_pipe
;
664 #ifdef CONFIG_FTRACE_STARTUP_TEST
665 if (type
->selftest
&& !tracing_selftest_disabled
) {
666 struct tracer
*saved_tracer
= current_trace
;
667 struct trace_array
*tr
= &global_trace
;
670 * Run a selftest on this tracer.
671 * Here we reset the trace buffer, and set the current
672 * tracer to be this tracer. The tracer can then run some
673 * internal tracing to verify that everything is in order.
674 * If we fail, we do not register this tracer.
676 tracing_reset_online_cpus(tr
);
678 current_trace
= type
;
679 /* the test is responsible for initializing and enabling */
680 pr_info("Testing tracer %s: ", type
->name
);
681 ret
= type
->selftest(type
, tr
);
682 /* the test is responsible for resetting too */
683 current_trace
= saved_tracer
;
685 printk(KERN_CONT
"FAILED!\n");
688 /* Only reset on passing, to avoid touching corrupted buffers */
689 tracing_reset_online_cpus(tr
);
691 printk(KERN_CONT
"PASSED\n");
695 type
->next
= trace_types
;
699 tracing_selftest_running
= false;
700 mutex_unlock(&trace_types_lock
);
702 if (ret
|| !default_bootup_tracer
)
705 if (strncmp(default_bootup_tracer
, type
->name
, MAX_TRACER_SIZE
))
708 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
709 /* Do we want this tracer to start on bootup? */
710 tracing_set_tracer(type
->name
);
711 default_bootup_tracer
= NULL
;
712 /* disable other selftests, since this will break it. */
713 tracing_selftest_disabled
= 1;
714 #ifdef CONFIG_FTRACE_STARTUP_TEST
715 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
724 void unregister_tracer(struct tracer
*type
)
728 mutex_lock(&trace_types_lock
);
729 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
733 pr_info("Tracer %s not registered\n", type
->name
);
739 if (type
== current_trace
&& tracer_enabled
) {
742 if (current_trace
->stop
)
743 current_trace
->stop(&global_trace
);
744 current_trace
= &nop_trace
;
747 mutex_unlock(&trace_types_lock
);
750 static void __tracing_reset(struct trace_array
*tr
, int cpu
)
752 ftrace_disable_cpu();
753 ring_buffer_reset_cpu(tr
->buffer
, cpu
);
757 void tracing_reset(struct trace_array
*tr
, int cpu
)
759 struct ring_buffer
*buffer
= tr
->buffer
;
761 ring_buffer_record_disable(buffer
);
763 /* Make sure all commits have finished */
765 __tracing_reset(tr
, cpu
);
767 ring_buffer_record_enable(buffer
);
770 void tracing_reset_online_cpus(struct trace_array
*tr
)
772 struct ring_buffer
*buffer
= tr
->buffer
;
775 ring_buffer_record_disable(buffer
);
777 /* Make sure all commits have finished */
780 tr
->time_start
= ftrace_now(tr
->cpu
);
782 for_each_online_cpu(cpu
)
783 __tracing_reset(tr
, cpu
);
785 ring_buffer_record_enable(buffer
);
788 void tracing_reset_current(int cpu
)
790 tracing_reset(&global_trace
, cpu
);
793 void tracing_reset_current_online_cpus(void)
795 tracing_reset_online_cpus(&global_trace
);
798 #define SAVED_CMDLINES 128
799 #define NO_CMDLINE_MAP UINT_MAX
800 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
801 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
802 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
803 static int cmdline_idx
;
804 static arch_spinlock_t trace_cmdline_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
806 /* temporary disable recording */
807 static atomic_t trace_record_cmdline_disabled __read_mostly
;
809 static void trace_init_cmdlines(void)
811 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
812 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
816 int is_tracing_stopped(void)
818 return trace_stop_count
;
822 * ftrace_off_permanent - disable all ftrace code permanently
824 * This should only be called when a serious anomally has
825 * been detected. This will turn off the function tracing,
826 * ring buffers, and other tracing utilites. It takes no
827 * locks and can be called from any context.
829 void ftrace_off_permanent(void)
831 tracing_disabled
= 1;
833 tracing_off_permanent();
837 * tracing_start - quick start of the tracer
839 * If tracing is enabled but was stopped by tracing_stop,
840 * this will start the tracer back up.
842 void tracing_start(void)
844 struct ring_buffer
*buffer
;
847 if (tracing_disabled
)
850 spin_lock_irqsave(&tracing_start_lock
, flags
);
851 if (--trace_stop_count
) {
852 if (trace_stop_count
< 0) {
853 /* Someone screwed up their debugging */
855 trace_stop_count
= 0;
861 buffer
= global_trace
.buffer
;
863 ring_buffer_record_enable(buffer
);
865 buffer
= max_tr
.buffer
;
867 ring_buffer_record_enable(buffer
);
871 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
875 * tracing_stop - quick stop of the tracer
877 * Light weight way to stop tracing. Use in conjunction with
880 void tracing_stop(void)
882 struct ring_buffer
*buffer
;
886 spin_lock_irqsave(&tracing_start_lock
, flags
);
887 if (trace_stop_count
++)
890 buffer
= global_trace
.buffer
;
892 ring_buffer_record_disable(buffer
);
894 buffer
= max_tr
.buffer
;
896 ring_buffer_record_disable(buffer
);
899 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
902 void trace_stop_cmdline_recording(void);
904 static void trace_save_cmdline(struct task_struct
*tsk
)
908 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
912 * It's not the end of the world if we don't get
913 * the lock, but we also don't want to spin
914 * nor do we want to disable interrupts,
915 * so if we miss here, then better luck next time.
917 if (!arch_spin_trylock(&trace_cmdline_lock
))
920 idx
= map_pid_to_cmdline
[tsk
->pid
];
921 if (idx
== NO_CMDLINE_MAP
) {
922 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
925 * Check whether the cmdline buffer at idx has a pid
926 * mapped. We are going to overwrite that entry so we
927 * need to clear the map_pid_to_cmdline. Otherwise we
928 * would read the new comm for the old pid.
930 pid
= map_cmdline_to_pid
[idx
];
931 if (pid
!= NO_CMDLINE_MAP
)
932 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
934 map_cmdline_to_pid
[idx
] = tsk
->pid
;
935 map_pid_to_cmdline
[tsk
->pid
] = idx
;
940 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
942 arch_spin_unlock(&trace_cmdline_lock
);
945 void trace_find_cmdline(int pid
, char comm
[])
950 strcpy(comm
, "<idle>");
954 if (pid
> PID_MAX_DEFAULT
) {
955 strcpy(comm
, "<...>");
960 arch_spin_lock(&trace_cmdline_lock
);
961 map
= map_pid_to_cmdline
[pid
];
962 if (map
!= NO_CMDLINE_MAP
)
963 strcpy(comm
, saved_cmdlines
[map
]);
965 strcpy(comm
, "<...>");
967 arch_spin_unlock(&trace_cmdline_lock
);
971 void tracing_record_cmdline(struct task_struct
*tsk
)
973 if (atomic_read(&trace_record_cmdline_disabled
) || !tracer_enabled
||
977 trace_save_cmdline(tsk
);
981 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
984 struct task_struct
*tsk
= current
;
986 entry
->preempt_count
= pc
& 0xff;
987 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
988 entry
->lock_depth
= (tsk
) ? tsk
->lock_depth
: 0;
990 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
991 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
993 TRACE_FLAG_IRQS_NOSUPPORT
|
995 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
996 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
997 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
999 EXPORT_SYMBOL_GPL(tracing_generic_entry_update
);
1001 struct ring_buffer_event
*
1002 trace_buffer_lock_reserve(struct ring_buffer
*buffer
,
1005 unsigned long flags
, int pc
)
1007 struct ring_buffer_event
*event
;
1009 event
= ring_buffer_lock_reserve(buffer
, len
);
1010 if (event
!= NULL
) {
1011 struct trace_entry
*ent
= ring_buffer_event_data(event
);
1013 tracing_generic_entry_update(ent
, flags
, pc
);
1021 __trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1022 struct ring_buffer_event
*event
,
1023 unsigned long flags
, int pc
,
1026 ring_buffer_unlock_commit(buffer
, event
);
1028 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1029 ftrace_trace_userstack(buffer
, flags
, pc
);
1035 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1036 struct ring_buffer_event
*event
,
1037 unsigned long flags
, int pc
)
1039 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1042 struct ring_buffer_event
*
1043 trace_current_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1044 int type
, unsigned long len
,
1045 unsigned long flags
, int pc
)
1047 *current_rb
= global_trace
.buffer
;
1048 return trace_buffer_lock_reserve(*current_rb
,
1049 type
, len
, flags
, pc
);
1051 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve
);
1053 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
1054 struct ring_buffer_event
*event
,
1055 unsigned long flags
, int pc
)
1057 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1059 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit
);
1061 void trace_nowake_buffer_unlock_commit(struct ring_buffer
*buffer
,
1062 struct ring_buffer_event
*event
,
1063 unsigned long flags
, int pc
)
1065 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 0);
1067 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit
);
1069 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
1070 struct ring_buffer_event
*event
)
1072 ring_buffer_discard_commit(buffer
, event
);
1074 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit
);
1077 trace_function(struct trace_array
*tr
,
1078 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1081 struct ftrace_event_call
*call
= &event_function
;
1082 struct ring_buffer
*buffer
= tr
->buffer
;
1083 struct ring_buffer_event
*event
;
1084 struct ftrace_entry
*entry
;
1086 /* If we are reading the ring buffer, don't trace */
1087 if (unlikely(__this_cpu_read(per_cpu_var(ftrace_cpu_disabled
))))
1090 event
= trace_buffer_lock_reserve(buffer
, TRACE_FN
, sizeof(*entry
),
1094 entry
= ring_buffer_event_data(event
);
1096 entry
->parent_ip
= parent_ip
;
1098 if (!filter_check_discard(call
, entry
, buffer
, event
))
1099 ring_buffer_unlock_commit(buffer
, event
);
1103 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1104 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1107 if (likely(!atomic_read(&data
->disabled
)))
1108 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
1111 #ifdef CONFIG_STACKTRACE
1112 static void __ftrace_trace_stack(struct ring_buffer
*buffer
,
1113 unsigned long flags
,
1116 struct ftrace_event_call
*call
= &event_kernel_stack
;
1117 struct ring_buffer_event
*event
;
1118 struct stack_entry
*entry
;
1119 struct stack_trace trace
;
1121 event
= trace_buffer_lock_reserve(buffer
, TRACE_STACK
,
1122 sizeof(*entry
), flags
, pc
);
1125 entry
= ring_buffer_event_data(event
);
1126 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1128 trace
.nr_entries
= 0;
1129 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1131 trace
.entries
= entry
->caller
;
1133 save_stack_trace(&trace
);
1134 if (!filter_check_discard(call
, entry
, buffer
, event
))
1135 ring_buffer_unlock_commit(buffer
, event
);
1138 void ftrace_trace_stack(struct ring_buffer
*buffer
, unsigned long flags
,
1141 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1144 __ftrace_trace_stack(buffer
, flags
, skip
, pc
);
1147 void __trace_stack(struct trace_array
*tr
, unsigned long flags
, int skip
,
1150 __ftrace_trace_stack(tr
->buffer
, flags
, skip
, pc
);
1154 * trace_dump_stack - record a stack back trace in the trace buffer
1156 void trace_dump_stack(void)
1158 unsigned long flags
;
1160 if (tracing_disabled
|| tracing_selftest_running
)
1163 local_save_flags(flags
);
1165 /* skipping 3 traces, seems to get us at the caller of this function */
1166 __ftrace_trace_stack(global_trace
.buffer
, flags
, 3, preempt_count());
1170 ftrace_trace_userstack(struct ring_buffer
*buffer
, unsigned long flags
, int pc
)
1172 struct ftrace_event_call
*call
= &event_user_stack
;
1173 struct ring_buffer_event
*event
;
1174 struct userstack_entry
*entry
;
1175 struct stack_trace trace
;
1177 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1180 event
= trace_buffer_lock_reserve(buffer
, TRACE_USER_STACK
,
1181 sizeof(*entry
), flags
, pc
);
1184 entry
= ring_buffer_event_data(event
);
1186 entry
->tgid
= current
->tgid
;
1187 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1189 trace
.nr_entries
= 0;
1190 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1192 trace
.entries
= entry
->caller
;
1194 save_stack_trace_user(&trace
);
1195 if (!filter_check_discard(call
, entry
, buffer
, event
))
1196 ring_buffer_unlock_commit(buffer
, event
);
1200 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1202 ftrace_trace_userstack(tr
, flags
, preempt_count());
1206 #endif /* CONFIG_STACKTRACE */
1209 ftrace_trace_special(void *__tr
,
1210 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
,
1213 struct ftrace_event_call
*call
= &event_special
;
1214 struct ring_buffer_event
*event
;
1215 struct trace_array
*tr
= __tr
;
1216 struct ring_buffer
*buffer
= tr
->buffer
;
1217 struct special_entry
*entry
;
1219 event
= trace_buffer_lock_reserve(buffer
, TRACE_SPECIAL
,
1220 sizeof(*entry
), 0, pc
);
1223 entry
= ring_buffer_event_data(event
);
1228 if (!filter_check_discard(call
, entry
, buffer
, event
))
1229 trace_buffer_unlock_commit(buffer
, event
, 0, pc
);
1233 __trace_special(void *__tr
, void *__data
,
1234 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1236 ftrace_trace_special(__tr
, arg1
, arg2
, arg3
, preempt_count());
1240 ftrace_special(unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1242 struct trace_array
*tr
= &global_trace
;
1243 struct trace_array_cpu
*data
;
1244 unsigned long flags
;
1248 if (tracing_disabled
)
1251 pc
= preempt_count();
1252 local_irq_save(flags
);
1253 cpu
= raw_smp_processor_id();
1254 data
= tr
->data
[cpu
];
1256 if (likely(atomic_inc_return(&data
->disabled
) == 1))
1257 ftrace_trace_special(tr
, arg1
, arg2
, arg3
, pc
);
1259 atomic_dec(&data
->disabled
);
1260 local_irq_restore(flags
);
1264 * trace_vbprintk - write binary msg to tracing buffer
1267 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1269 static arch_spinlock_t trace_buf_lock
=
1270 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
1271 static u32 trace_buf
[TRACE_BUF_SIZE
];
1273 struct ftrace_event_call
*call
= &event_bprint
;
1274 struct ring_buffer_event
*event
;
1275 struct ring_buffer
*buffer
;
1276 struct trace_array
*tr
= &global_trace
;
1277 struct trace_array_cpu
*data
;
1278 struct bprint_entry
*entry
;
1279 unsigned long flags
;
1282 int cpu
, len
= 0, size
, pc
;
1284 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1287 /* Don't pollute graph traces with trace_vprintk internals */
1288 pause_graph_tracing();
1290 pc
= preempt_count();
1291 resched
= ftrace_preempt_disable();
1292 cpu
= raw_smp_processor_id();
1293 data
= tr
->data
[cpu
];
1295 disable
= atomic_inc_return(&data
->disabled
);
1296 if (unlikely(disable
!= 1))
1299 /* Lockdep uses trace_printk for lock tracing */
1300 local_irq_save(flags
);
1301 arch_spin_lock(&trace_buf_lock
);
1302 len
= vbin_printf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1304 if (len
> TRACE_BUF_SIZE
|| len
< 0)
1307 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1308 buffer
= tr
->buffer
;
1309 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
1313 entry
= ring_buffer_event_data(event
);
1317 memcpy(entry
->buf
, trace_buf
, sizeof(u32
) * len
);
1318 if (!filter_check_discard(call
, entry
, buffer
, event
))
1319 ring_buffer_unlock_commit(buffer
, event
);
1322 arch_spin_unlock(&trace_buf_lock
);
1323 local_irq_restore(flags
);
1326 atomic_dec_return(&data
->disabled
);
1327 ftrace_preempt_enable(resched
);
1328 unpause_graph_tracing();
1332 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1334 int trace_array_printk(struct trace_array
*tr
,
1335 unsigned long ip
, const char *fmt
, ...)
1340 if (!(trace_flags
& TRACE_ITER_PRINTK
))
1344 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
1349 int trace_array_vprintk(struct trace_array
*tr
,
1350 unsigned long ip
, const char *fmt
, va_list args
)
1352 static arch_spinlock_t trace_buf_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
1353 static char trace_buf
[TRACE_BUF_SIZE
];
1355 struct ftrace_event_call
*call
= &event_print
;
1356 struct ring_buffer_event
*event
;
1357 struct ring_buffer
*buffer
;
1358 struct trace_array_cpu
*data
;
1359 int cpu
, len
= 0, size
, pc
;
1360 struct print_entry
*entry
;
1361 unsigned long irq_flags
;
1364 if (tracing_disabled
|| tracing_selftest_running
)
1367 pc
= preempt_count();
1368 preempt_disable_notrace();
1369 cpu
= raw_smp_processor_id();
1370 data
= tr
->data
[cpu
];
1372 disable
= atomic_inc_return(&data
->disabled
);
1373 if (unlikely(disable
!= 1))
1376 pause_graph_tracing();
1377 raw_local_irq_save(irq_flags
);
1378 arch_spin_lock(&trace_buf_lock
);
1379 len
= vsnprintf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1381 size
= sizeof(*entry
) + len
+ 1;
1382 buffer
= tr
->buffer
;
1383 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
1387 entry
= ring_buffer_event_data(event
);
1390 memcpy(&entry
->buf
, trace_buf
, len
);
1391 entry
->buf
[len
] = '\0';
1392 if (!filter_check_discard(call
, entry
, buffer
, event
))
1393 ring_buffer_unlock_commit(buffer
, event
);
1396 arch_spin_unlock(&trace_buf_lock
);
1397 raw_local_irq_restore(irq_flags
);
1398 unpause_graph_tracing();
1400 atomic_dec_return(&data
->disabled
);
1401 preempt_enable_notrace();
1406 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1408 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
1410 EXPORT_SYMBOL_GPL(trace_vprintk
);
1412 enum trace_file_type
{
1413 TRACE_FILE_LAT_FMT
= 1,
1414 TRACE_FILE_ANNOTATE
= 2,
1417 static void trace_iterator_increment(struct trace_iterator
*iter
)
1419 /* Don't allow ftrace to trace into the ring buffers */
1420 ftrace_disable_cpu();
1423 if (iter
->buffer_iter
[iter
->cpu
])
1424 ring_buffer_read(iter
->buffer_iter
[iter
->cpu
], NULL
);
1426 ftrace_enable_cpu();
1429 static struct trace_entry
*
1430 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
)
1432 struct ring_buffer_event
*event
;
1433 struct ring_buffer_iter
*buf_iter
= iter
->buffer_iter
[cpu
];
1435 /* Don't allow ftrace to trace into the ring buffers */
1436 ftrace_disable_cpu();
1439 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1441 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
);
1443 ftrace_enable_cpu();
1445 return event
? ring_buffer_event_data(event
) : NULL
;
1448 static struct trace_entry
*
1449 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
, u64
*ent_ts
)
1451 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1452 struct trace_entry
*ent
, *next
= NULL
;
1453 int cpu_file
= iter
->cpu_file
;
1454 u64 next_ts
= 0, ts
;
1459 * If we are in a per_cpu trace file, don't bother by iterating over
1460 * all cpu and peek directly.
1462 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1463 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1465 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
);
1467 *ent_cpu
= cpu_file
;
1472 for_each_tracing_cpu(cpu
) {
1474 if (ring_buffer_empty_cpu(buffer
, cpu
))
1477 ent
= peek_next_entry(iter
, cpu
, &ts
);
1480 * Pick the entry with the smallest timestamp:
1482 if (ent
&& (!next
|| ts
< next_ts
)) {
1490 *ent_cpu
= next_cpu
;
1498 /* Find the next real entry, without updating the iterator itself */
1499 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1500 int *ent_cpu
, u64
*ent_ts
)
1502 return __find_next_entry(iter
, ent_cpu
, ent_ts
);
1505 /* Find the next real entry, and increment the iterator to the next entry */
1506 static void *find_next_entry_inc(struct trace_iterator
*iter
)
1508 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
, &iter
->ts
);
1511 trace_iterator_increment(iter
);
1513 return iter
->ent
? iter
: NULL
;
1516 static void trace_consume(struct trace_iterator
*iter
)
1518 /* Don't allow ftrace to trace into the ring buffers */
1519 ftrace_disable_cpu();
1520 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
);
1521 ftrace_enable_cpu();
1524 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1526 struct trace_iterator
*iter
= m
->private;
1530 WARN_ON_ONCE(iter
->leftover
);
1534 /* can't go backwards */
1539 ent
= find_next_entry_inc(iter
);
1543 while (ent
&& iter
->idx
< i
)
1544 ent
= find_next_entry_inc(iter
);
1551 static void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
1553 struct trace_array
*tr
= iter
->tr
;
1554 struct ring_buffer_event
*event
;
1555 struct ring_buffer_iter
*buf_iter
;
1556 unsigned long entries
= 0;
1559 tr
->data
[cpu
]->skipped_entries
= 0;
1561 if (!iter
->buffer_iter
[cpu
])
1564 buf_iter
= iter
->buffer_iter
[cpu
];
1565 ring_buffer_iter_reset(buf_iter
);
1568 * We could have the case with the max latency tracers
1569 * that a reset never took place on a cpu. This is evident
1570 * by the timestamp being before the start of the buffer.
1572 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
1573 if (ts
>= iter
->tr
->time_start
)
1576 ring_buffer_read(buf_iter
, NULL
);
1579 tr
->data
[cpu
]->skipped_entries
= entries
;
1583 * No necessary locking here. The worst thing which can
1584 * happen is loosing events consumed at the same time
1585 * by a trace_pipe reader.
1586 * Other than that, we don't risk to crash the ring buffer
1587 * because it serializes the readers.
1589 * The current tracer is copied to avoid a global locking
1592 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1594 struct trace_iterator
*iter
= m
->private;
1595 static struct tracer
*old_tracer
;
1596 int cpu_file
= iter
->cpu_file
;
1601 /* copy the tracer to avoid using a global lock all around */
1602 mutex_lock(&trace_types_lock
);
1603 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1604 old_tracer
= current_trace
;
1605 *iter
->trace
= *current_trace
;
1607 mutex_unlock(&trace_types_lock
);
1609 atomic_inc(&trace_record_cmdline_disabled
);
1611 if (*pos
!= iter
->pos
) {
1616 ftrace_disable_cpu();
1618 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1619 for_each_tracing_cpu(cpu
)
1620 tracing_iter_reset(iter
, cpu
);
1622 tracing_iter_reset(iter
, cpu_file
);
1624 ftrace_enable_cpu();
1626 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1631 * If we overflowed the seq_file before, then we want
1632 * to just reuse the trace_seq buffer again.
1638 p
= s_next(m
, p
, &l
);
1642 trace_event_read_lock();
1646 static void s_stop(struct seq_file
*m
, void *p
)
1648 atomic_dec(&trace_record_cmdline_disabled
);
1649 trace_event_read_unlock();
1652 static void print_lat_help_header(struct seq_file
*m
)
1654 seq_puts(m
, "# _------=> CPU# \n");
1655 seq_puts(m
, "# / _-----=> irqs-off \n");
1656 seq_puts(m
, "# | / _----=> need-resched \n");
1657 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1658 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1659 seq_puts(m
, "# |||| /_--=> lock-depth \n");
1660 seq_puts(m
, "# |||||/ delay \n");
1661 seq_puts(m
, "# cmd pid |||||| time | caller \n");
1662 seq_puts(m
, "# \\ / |||||| \\ | / \n");
1665 static void print_func_help_header(struct seq_file
*m
)
1667 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1668 seq_puts(m
, "# | | | | |\n");
1673 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1675 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1676 struct trace_array
*tr
= iter
->tr
;
1677 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1678 struct tracer
*type
= current_trace
;
1679 unsigned long entries
= 0;
1680 unsigned long total
= 0;
1681 unsigned long count
;
1682 const char *name
= "preemption";
1689 for_each_tracing_cpu(cpu
) {
1690 count
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
1692 * If this buffer has skipped entries, then we hold all
1693 * entries for the trace and we need to ignore the
1694 * ones before the time stamp.
1696 if (tr
->data
[cpu
]->skipped_entries
) {
1697 count
-= tr
->data
[cpu
]->skipped_entries
;
1698 /* total is the same as the entries */
1702 ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
1706 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
1708 seq_puts(m
, "# -----------------------------------"
1709 "---------------------------------\n");
1710 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1711 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1712 nsecs_to_usecs(data
->saved_latency
),
1716 #if defined(CONFIG_PREEMPT_NONE)
1718 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1720 #elif defined(CONFIG_PREEMPT)
1725 /* These are reserved for later use */
1728 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1732 seq_puts(m
, "# -----------------\n");
1733 seq_printf(m
, "# | task: %.16s-%d "
1734 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1735 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1736 data
->policy
, data
->rt_priority
);
1737 seq_puts(m
, "# -----------------\n");
1739 if (data
->critical_start
) {
1740 seq_puts(m
, "# => started at: ");
1741 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1742 trace_print_seq(m
, &iter
->seq
);
1743 seq_puts(m
, "\n# => ended at: ");
1744 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1745 trace_print_seq(m
, &iter
->seq
);
1746 seq_puts(m
, "\n#\n");
1752 static void test_cpu_buff_start(struct trace_iterator
*iter
)
1754 struct trace_seq
*s
= &iter
->seq
;
1756 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
1759 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
1762 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
1765 if (iter
->tr
->data
[iter
->cpu
]->skipped_entries
)
1768 cpumask_set_cpu(iter
->cpu
, iter
->started
);
1770 /* Don't print started cpu buffer for the first entry of the trace */
1772 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
1776 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
1778 struct trace_seq
*s
= &iter
->seq
;
1779 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1780 struct trace_entry
*entry
;
1781 struct trace_event
*event
;
1785 test_cpu_buff_start(iter
);
1787 event
= ftrace_find_event(entry
->type
);
1789 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1790 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1791 if (!trace_print_lat_context(iter
))
1794 if (!trace_print_context(iter
))
1800 return event
->trace(iter
, sym_flags
);
1802 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
1805 return TRACE_TYPE_HANDLED
;
1807 return TRACE_TYPE_PARTIAL_LINE
;
1810 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
1812 struct trace_seq
*s
= &iter
->seq
;
1813 struct trace_entry
*entry
;
1814 struct trace_event
*event
;
1818 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1819 if (!trace_seq_printf(s
, "%d %d %llu ",
1820 entry
->pid
, iter
->cpu
, iter
->ts
))
1824 event
= ftrace_find_event(entry
->type
);
1826 return event
->raw(iter
, 0);
1828 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
1831 return TRACE_TYPE_HANDLED
;
1833 return TRACE_TYPE_PARTIAL_LINE
;
1836 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
1838 struct trace_seq
*s
= &iter
->seq
;
1839 unsigned char newline
= '\n';
1840 struct trace_entry
*entry
;
1841 struct trace_event
*event
;
1845 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1846 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1847 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1848 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
1851 event
= ftrace_find_event(entry
->type
);
1853 enum print_line_t ret
= event
->hex(iter
, 0);
1854 if (ret
!= TRACE_TYPE_HANDLED
)
1858 SEQ_PUT_FIELD_RET(s
, newline
);
1860 return TRACE_TYPE_HANDLED
;
1863 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
1865 struct trace_seq
*s
= &iter
->seq
;
1866 struct trace_entry
*entry
;
1867 struct trace_event
*event
;
1871 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1872 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1873 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
1874 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
1877 event
= ftrace_find_event(entry
->type
);
1878 return event
? event
->binary(iter
, 0) : TRACE_TYPE_HANDLED
;
1881 static int trace_empty(struct trace_iterator
*iter
)
1885 /* If we are looking at one CPU buffer, only check that one */
1886 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
1887 cpu
= iter
->cpu_file
;
1888 if (iter
->buffer_iter
[cpu
]) {
1889 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1892 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1898 for_each_tracing_cpu(cpu
) {
1899 if (iter
->buffer_iter
[cpu
]) {
1900 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1903 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1911 /* Called with trace_event_read_lock() held. */
1912 static enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
1914 enum print_line_t ret
;
1916 if (iter
->trace
&& iter
->trace
->print_line
) {
1917 ret
= iter
->trace
->print_line(iter
);
1918 if (ret
!= TRACE_TYPE_UNHANDLED
)
1922 if (iter
->ent
->type
== TRACE_BPRINT
&&
1923 trace_flags
& TRACE_ITER_PRINTK
&&
1924 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
1925 return trace_print_bprintk_msg_only(iter
);
1927 if (iter
->ent
->type
== TRACE_PRINT
&&
1928 trace_flags
& TRACE_ITER_PRINTK
&&
1929 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
1930 return trace_print_printk_msg_only(iter
);
1932 if (trace_flags
& TRACE_ITER_BIN
)
1933 return print_bin_fmt(iter
);
1935 if (trace_flags
& TRACE_ITER_HEX
)
1936 return print_hex_fmt(iter
);
1938 if (trace_flags
& TRACE_ITER_RAW
)
1939 return print_raw_fmt(iter
);
1941 return print_trace_fmt(iter
);
1944 static int s_show(struct seq_file
*m
, void *v
)
1946 struct trace_iterator
*iter
= v
;
1949 if (iter
->ent
== NULL
) {
1951 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
1954 if (iter
->trace
&& iter
->trace
->print_header
)
1955 iter
->trace
->print_header(m
);
1956 else if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1957 /* print nothing if the buffers are empty */
1958 if (trace_empty(iter
))
1960 print_trace_header(m
, iter
);
1961 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1962 print_lat_help_header(m
);
1964 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1965 print_func_help_header(m
);
1967 } else if (iter
->leftover
) {
1969 * If we filled the seq_file buffer earlier, we
1970 * want to just show it now.
1972 ret
= trace_print_seq(m
, &iter
->seq
);
1974 /* ret should this time be zero, but you never know */
1975 iter
->leftover
= ret
;
1978 print_trace_line(iter
);
1979 ret
= trace_print_seq(m
, &iter
->seq
);
1981 * If we overflow the seq_file buffer, then it will
1982 * ask us for this data again at start up.
1984 * ret is 0 if seq_file write succeeded.
1987 iter
->leftover
= ret
;
1993 static const struct seq_operations tracer_seq_ops
= {
2000 static struct trace_iterator
*
2001 __tracing_open(struct inode
*inode
, struct file
*file
)
2003 long cpu_file
= (long) inode
->i_private
;
2004 void *fail_ret
= ERR_PTR(-ENOMEM
);
2005 struct trace_iterator
*iter
;
2009 if (tracing_disabled
)
2010 return ERR_PTR(-ENODEV
);
2012 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2014 return ERR_PTR(-ENOMEM
);
2017 * We make a copy of the current tracer to avoid concurrent
2018 * changes on it while we are reading.
2020 mutex_lock(&trace_types_lock
);
2021 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2026 *iter
->trace
= *current_trace
;
2028 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2031 if (current_trace
&& current_trace
->print_max
)
2034 iter
->tr
= &global_trace
;
2036 mutex_init(&iter
->mutex
);
2037 iter
->cpu_file
= cpu_file
;
2039 /* Notify the tracer early; before we stop tracing. */
2040 if (iter
->trace
&& iter
->trace
->open
)
2041 iter
->trace
->open(iter
);
2043 /* Annotate start of buffers if we had overruns */
2044 if (ring_buffer_overruns(iter
->tr
->buffer
))
2045 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2047 /* stop the trace while dumping */
2050 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
2051 for_each_tracing_cpu(cpu
) {
2053 iter
->buffer_iter
[cpu
] =
2054 ring_buffer_read_start(iter
->tr
->buffer
, cpu
);
2055 tracing_iter_reset(iter
, cpu
);
2058 cpu
= iter
->cpu_file
;
2059 iter
->buffer_iter
[cpu
] =
2060 ring_buffer_read_start(iter
->tr
->buffer
, cpu
);
2061 tracing_iter_reset(iter
, cpu
);
2064 ret
= seq_open(file
, &tracer_seq_ops
);
2066 fail_ret
= ERR_PTR(ret
);
2070 m
= file
->private_data
;
2073 mutex_unlock(&trace_types_lock
);
2078 for_each_tracing_cpu(cpu
) {
2079 if (iter
->buffer_iter
[cpu
])
2080 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2082 free_cpumask_var(iter
->started
);
2085 mutex_unlock(&trace_types_lock
);
2092 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2094 if (tracing_disabled
)
2097 filp
->private_data
= inode
->i_private
;
2101 static int tracing_release(struct inode
*inode
, struct file
*file
)
2103 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
2104 struct trace_iterator
*iter
;
2107 if (!(file
->f_mode
& FMODE_READ
))
2112 mutex_lock(&trace_types_lock
);
2113 for_each_tracing_cpu(cpu
) {
2114 if (iter
->buffer_iter
[cpu
])
2115 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2118 if (iter
->trace
&& iter
->trace
->close
)
2119 iter
->trace
->close(iter
);
2121 /* reenable tracing if it was previously enabled */
2123 mutex_unlock(&trace_types_lock
);
2125 seq_release(inode
, file
);
2126 mutex_destroy(&iter
->mutex
);
2127 free_cpumask_var(iter
->started
);
2133 static int tracing_open(struct inode
*inode
, struct file
*file
)
2135 struct trace_iterator
*iter
;
2138 /* If this file was open for write, then erase contents */
2139 if ((file
->f_mode
& FMODE_WRITE
) &&
2140 (file
->f_flags
& O_TRUNC
)) {
2141 long cpu
= (long) inode
->i_private
;
2143 if (cpu
== TRACE_PIPE_ALL_CPU
)
2144 tracing_reset_online_cpus(&global_trace
);
2146 tracing_reset(&global_trace
, cpu
);
2149 if (file
->f_mode
& FMODE_READ
) {
2150 iter
= __tracing_open(inode
, file
);
2152 ret
= PTR_ERR(iter
);
2153 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2154 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2160 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2162 struct tracer
*t
= v
;
2172 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2177 mutex_lock(&trace_types_lock
);
2178 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2184 static void t_stop(struct seq_file
*m
, void *p
)
2186 mutex_unlock(&trace_types_lock
);
2189 static int t_show(struct seq_file
*m
, void *v
)
2191 struct tracer
*t
= v
;
2196 seq_printf(m
, "%s", t
->name
);
2205 static const struct seq_operations show_traces_seq_ops
= {
2212 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2214 if (tracing_disabled
)
2217 return seq_open(file
, &show_traces_seq_ops
);
2221 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2222 size_t count
, loff_t
*ppos
)
2227 static const struct file_operations tracing_fops
= {
2228 .open
= tracing_open
,
2230 .write
= tracing_write_stub
,
2231 .llseek
= seq_lseek
,
2232 .release
= tracing_release
,
2235 static const struct file_operations show_traces_fops
= {
2236 .open
= show_traces_open
,
2238 .release
= seq_release
,
2242 * Only trace on a CPU if the bitmask is set:
2244 static cpumask_var_t tracing_cpumask
;
2247 * The tracer itself will not take this lock, but still we want
2248 * to provide a consistent cpumask to user-space:
2250 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2253 * Temporary storage for the character representation of the
2254 * CPU bitmask (and one more byte for the newline):
2256 static char mask_str
[NR_CPUS
+ 1];
2259 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2260 size_t count
, loff_t
*ppos
)
2264 mutex_lock(&tracing_cpumask_update_lock
);
2266 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2267 if (count
- len
< 2) {
2271 len
+= sprintf(mask_str
+ len
, "\n");
2272 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2275 mutex_unlock(&tracing_cpumask_update_lock
);
2281 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2282 size_t count
, loff_t
*ppos
)
2285 cpumask_var_t tracing_cpumask_new
;
2287 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2290 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2294 mutex_lock(&tracing_cpumask_update_lock
);
2296 local_irq_disable();
2297 arch_spin_lock(&ftrace_max_lock
);
2298 for_each_tracing_cpu(cpu
) {
2300 * Increase/decrease the disabled counter if we are
2301 * about to flip a bit in the cpumask:
2303 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2304 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2305 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2307 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2308 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2309 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2312 arch_spin_unlock(&ftrace_max_lock
);
2315 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2317 mutex_unlock(&tracing_cpumask_update_lock
);
2318 free_cpumask_var(tracing_cpumask_new
);
2323 free_cpumask_var(tracing_cpumask_new
);
2328 static const struct file_operations tracing_cpumask_fops
= {
2329 .open
= tracing_open_generic
,
2330 .read
= tracing_cpumask_read
,
2331 .write
= tracing_cpumask_write
,
2334 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
2336 struct tracer_opt
*trace_opts
;
2340 mutex_lock(&trace_types_lock
);
2341 tracer_flags
= current_trace
->flags
->val
;
2342 trace_opts
= current_trace
->flags
->opts
;
2344 for (i
= 0; trace_options
[i
]; i
++) {
2345 if (trace_flags
& (1 << i
))
2346 seq_printf(m
, "%s\n", trace_options
[i
]);
2348 seq_printf(m
, "no%s\n", trace_options
[i
]);
2351 for (i
= 0; trace_opts
[i
].name
; i
++) {
2352 if (tracer_flags
& trace_opts
[i
].bit
)
2353 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
2355 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
2357 mutex_unlock(&trace_types_lock
);
2362 static int __set_tracer_option(struct tracer
*trace
,
2363 struct tracer_flags
*tracer_flags
,
2364 struct tracer_opt
*opts
, int neg
)
2368 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
2373 tracer_flags
->val
&= ~opts
->bit
;
2375 tracer_flags
->val
|= opts
->bit
;
2379 /* Try to assign a tracer specific option */
2380 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2382 struct tracer_flags
*tracer_flags
= trace
->flags
;
2383 struct tracer_opt
*opts
= NULL
;
2386 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
2387 opts
= &tracer_flags
->opts
[i
];
2389 if (strcmp(cmp
, opts
->name
) == 0)
2390 return __set_tracer_option(trace
, trace
->flags
,
2397 static void set_tracer_flags(unsigned int mask
, int enabled
)
2399 /* do nothing if flag is already set */
2400 if (!!(trace_flags
& mask
) == !!enabled
)
2404 trace_flags
|= mask
;
2406 trace_flags
&= ~mask
;
2410 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2411 size_t cnt
, loff_t
*ppos
)
2419 if (cnt
>= sizeof(buf
))
2422 if (copy_from_user(&buf
, ubuf
, cnt
))
2426 cmp
= strstrip(buf
);
2428 if (strncmp(cmp
, "no", 2) == 0) {
2433 for (i
= 0; trace_options
[i
]; i
++) {
2434 if (strcmp(cmp
, trace_options
[i
]) == 0) {
2435 set_tracer_flags(1 << i
, !neg
);
2440 /* If no option could be set, test the specific tracer options */
2441 if (!trace_options
[i
]) {
2442 mutex_lock(&trace_types_lock
);
2443 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2444 mutex_unlock(&trace_types_lock
);
2454 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
2456 if (tracing_disabled
)
2458 return single_open(file
, tracing_trace_options_show
, NULL
);
2461 static const struct file_operations tracing_iter_fops
= {
2462 .open
= tracing_trace_options_open
,
2464 .llseek
= seq_lseek
,
2465 .release
= single_release
,
2466 .write
= tracing_trace_options_write
,
2469 static const char readme_msg
[] =
2470 "tracing mini-HOWTO:\n\n"
2471 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2472 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2473 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2474 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2476 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2477 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2479 "# cat /sys/kernel/debug/tracing/trace_options\n"
2480 "noprint-parent nosym-offset nosym-addr noverbose\n"
2481 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2482 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2483 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2484 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2488 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2489 size_t cnt
, loff_t
*ppos
)
2491 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2492 readme_msg
, strlen(readme_msg
));
2495 static const struct file_operations tracing_readme_fops
= {
2496 .open
= tracing_open_generic
,
2497 .read
= tracing_readme_read
,
2501 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
2502 size_t cnt
, loff_t
*ppos
)
2511 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
2515 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
2523 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
2526 pid
= map_cmdline_to_pid
[i
];
2527 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
2530 trace_find_cmdline(pid
, buf_comm
);
2531 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
2536 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
2545 static const struct file_operations tracing_saved_cmdlines_fops
= {
2546 .open
= tracing_open_generic
,
2547 .read
= tracing_saved_cmdlines_read
,
2551 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2552 size_t cnt
, loff_t
*ppos
)
2557 r
= sprintf(buf
, "%u\n", tracer_enabled
);
2558 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2562 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2563 size_t cnt
, loff_t
*ppos
)
2565 struct trace_array
*tr
= filp
->private_data
;
2570 if (cnt
>= sizeof(buf
))
2573 if (copy_from_user(&buf
, ubuf
, cnt
))
2578 ret
= strict_strtoul(buf
, 10, &val
);
2584 mutex_lock(&trace_types_lock
);
2585 if (tracer_enabled
^ val
) {
2588 if (current_trace
->start
)
2589 current_trace
->start(tr
);
2594 if (current_trace
->stop
)
2595 current_trace
->stop(tr
);
2598 mutex_unlock(&trace_types_lock
);
2606 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2607 size_t cnt
, loff_t
*ppos
)
2609 char buf
[MAX_TRACER_SIZE
+2];
2612 mutex_lock(&trace_types_lock
);
2614 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2616 r
= sprintf(buf
, "\n");
2617 mutex_unlock(&trace_types_lock
);
2619 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2622 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
2624 tracing_reset_online_cpus(tr
);
2628 static int tracing_resize_ring_buffer(unsigned long size
)
2633 * If kernel or user changes the size of the ring buffer
2634 * we use the size that was given, and we can forget about
2635 * expanding it later.
2637 ring_buffer_expanded
= 1;
2639 ret
= ring_buffer_resize(global_trace
.buffer
, size
);
2643 ret
= ring_buffer_resize(max_tr
.buffer
, size
);
2647 r
= ring_buffer_resize(global_trace
.buffer
,
2648 global_trace
.entries
);
2651 * AARGH! We are left with different
2652 * size max buffer!!!!
2653 * The max buffer is our "snapshot" buffer.
2654 * When a tracer needs a snapshot (one of the
2655 * latency tracers), it swaps the max buffer
2656 * with the saved snap shot. We succeeded to
2657 * update the size of the main buffer, but failed to
2658 * update the size of the max buffer. But when we tried
2659 * to reset the main buffer to the original size, we
2660 * failed there too. This is very unlikely to
2661 * happen, but if it does, warn and kill all
2665 tracing_disabled
= 1;
2670 global_trace
.entries
= size
;
2676 * tracing_update_buffers - used by tracing facility to expand ring buffers
2678 * To save on memory when the tracing is never used on a system with it
2679 * configured in. The ring buffers are set to a minimum size. But once
2680 * a user starts to use the tracing facility, then they need to grow
2681 * to their default size.
2683 * This function is to be called when a tracer is about to be used.
2685 int tracing_update_buffers(void)
2689 mutex_lock(&trace_types_lock
);
2690 if (!ring_buffer_expanded
)
2691 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2692 mutex_unlock(&trace_types_lock
);
2697 struct trace_option_dentry
;
2699 static struct trace_option_dentry
*
2700 create_trace_option_files(struct tracer
*tracer
);
2703 destroy_trace_option_files(struct trace_option_dentry
*topts
);
2705 static int tracing_set_tracer(const char *buf
)
2707 static struct trace_option_dentry
*topts
;
2708 struct trace_array
*tr
= &global_trace
;
2712 mutex_lock(&trace_types_lock
);
2714 if (!ring_buffer_expanded
) {
2715 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2721 for (t
= trace_types
; t
; t
= t
->next
) {
2722 if (strcmp(t
->name
, buf
) == 0)
2729 if (t
== current_trace
)
2732 trace_branch_disable();
2733 if (current_trace
&& current_trace
->reset
)
2734 current_trace
->reset(tr
);
2736 destroy_trace_option_files(topts
);
2740 topts
= create_trace_option_files(current_trace
);
2743 ret
= tracer_init(t
, tr
);
2748 trace_branch_enable(tr
);
2750 mutex_unlock(&trace_types_lock
);
2756 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2757 size_t cnt
, loff_t
*ppos
)
2759 char buf
[MAX_TRACER_SIZE
+1];
2766 if (cnt
> MAX_TRACER_SIZE
)
2767 cnt
= MAX_TRACER_SIZE
;
2769 if (copy_from_user(&buf
, ubuf
, cnt
))
2774 /* strip ending whitespace. */
2775 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2778 err
= tracing_set_tracer(buf
);
2788 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2789 size_t cnt
, loff_t
*ppos
)
2791 unsigned long *ptr
= filp
->private_data
;
2795 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
2796 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2797 if (r
> sizeof(buf
))
2799 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2803 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2804 size_t cnt
, loff_t
*ppos
)
2806 unsigned long *ptr
= filp
->private_data
;
2811 if (cnt
>= sizeof(buf
))
2814 if (copy_from_user(&buf
, ubuf
, cnt
))
2819 ret
= strict_strtoul(buf
, 10, &val
);
2828 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2830 long cpu_file
= (long) inode
->i_private
;
2831 struct trace_iterator
*iter
;
2834 if (tracing_disabled
)
2837 mutex_lock(&trace_types_lock
);
2839 /* We only allow one reader per cpu */
2840 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
2841 if (!cpumask_empty(tracing_reader_cpumask
)) {
2845 cpumask_setall(tracing_reader_cpumask
);
2847 if (!cpumask_test_cpu(cpu_file
, tracing_reader_cpumask
))
2848 cpumask_set_cpu(cpu_file
, tracing_reader_cpumask
);
2855 /* create a buffer to store the information to pass to userspace */
2856 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2863 * We make a copy of the current tracer to avoid concurrent
2864 * changes on it while we are reading.
2866 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2872 *iter
->trace
= *current_trace
;
2874 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
2879 /* trace pipe does not show start of buffer */
2880 cpumask_setall(iter
->started
);
2882 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2883 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2885 iter
->cpu_file
= cpu_file
;
2886 iter
->tr
= &global_trace
;
2887 mutex_init(&iter
->mutex
);
2888 filp
->private_data
= iter
;
2890 if (iter
->trace
->pipe_open
)
2891 iter
->trace
->pipe_open(iter
);
2894 mutex_unlock(&trace_types_lock
);
2900 mutex_unlock(&trace_types_lock
);
2904 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
2906 struct trace_iterator
*iter
= file
->private_data
;
2908 mutex_lock(&trace_types_lock
);
2910 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
)
2911 cpumask_clear(tracing_reader_cpumask
);
2913 cpumask_clear_cpu(iter
->cpu_file
, tracing_reader_cpumask
);
2916 if (iter
->trace
->pipe_close
)
2917 iter
->trace
->pipe_close(iter
);
2919 mutex_unlock(&trace_types_lock
);
2921 free_cpumask_var(iter
->started
);
2922 mutex_destroy(&iter
->mutex
);
2930 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
2932 struct trace_iterator
*iter
= filp
->private_data
;
2934 if (trace_flags
& TRACE_ITER_BLOCK
) {
2936 * Always select as readable when in blocking mode
2938 return POLLIN
| POLLRDNORM
;
2940 if (!trace_empty(iter
))
2941 return POLLIN
| POLLRDNORM
;
2942 poll_wait(filp
, &trace_wait
, poll_table
);
2943 if (!trace_empty(iter
))
2944 return POLLIN
| POLLRDNORM
;
2951 void default_wait_pipe(struct trace_iterator
*iter
)
2955 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
2957 if (trace_empty(iter
))
2960 finish_wait(&trace_wait
, &wait
);
2964 * This is a make-shift waitqueue.
2965 * A tracer might use this callback on some rare cases:
2967 * 1) the current tracer might hold the runqueue lock when it wakes up
2968 * a reader, hence a deadlock (sched, function, and function graph tracers)
2969 * 2) the function tracers, trace all functions, we don't want
2970 * the overhead of calling wake_up and friends
2971 * (and tracing them too)
2973 * Anyway, this is really very primitive wakeup.
2975 void poll_wait_pipe(struct trace_iterator
*iter
)
2977 set_current_state(TASK_INTERRUPTIBLE
);
2978 /* sleep for 100 msecs, and try again. */
2979 schedule_timeout(HZ
/ 10);
2982 /* Must be called with trace_types_lock mutex held. */
2983 static int tracing_wait_pipe(struct file
*filp
)
2985 struct trace_iterator
*iter
= filp
->private_data
;
2987 while (trace_empty(iter
)) {
2989 if ((filp
->f_flags
& O_NONBLOCK
)) {
2993 mutex_unlock(&iter
->mutex
);
2995 iter
->trace
->wait_pipe(iter
);
2997 mutex_lock(&iter
->mutex
);
2999 if (signal_pending(current
))
3003 * We block until we read something and tracing is disabled.
3004 * We still block if tracing is disabled, but we have never
3005 * read anything. This allows a user to cat this file, and
3006 * then enable tracing. But after we have read something,
3007 * we give an EOF when tracing is again disabled.
3009 * iter->pos will be 0 if we haven't read anything.
3011 if (!tracer_enabled
&& iter
->pos
)
3022 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
3023 size_t cnt
, loff_t
*ppos
)
3025 struct trace_iterator
*iter
= filp
->private_data
;
3026 static struct tracer
*old_tracer
;
3029 /* return any leftover data */
3030 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3034 trace_seq_init(&iter
->seq
);
3036 /* copy the tracer to avoid using a global lock all around */
3037 mutex_lock(&trace_types_lock
);
3038 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3039 old_tracer
= current_trace
;
3040 *iter
->trace
= *current_trace
;
3042 mutex_unlock(&trace_types_lock
);
3045 * Avoid more than one consumer on a single file descriptor
3046 * This is just a matter of traces coherency, the ring buffer itself
3049 mutex_lock(&iter
->mutex
);
3050 if (iter
->trace
->read
) {
3051 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
3057 sret
= tracing_wait_pipe(filp
);
3061 /* stop when tracing is finished */
3062 if (trace_empty(iter
)) {
3067 if (cnt
>= PAGE_SIZE
)
3068 cnt
= PAGE_SIZE
- 1;
3070 /* reset all but tr, trace, and overruns */
3071 memset(&iter
->seq
, 0,
3072 sizeof(struct trace_iterator
) -
3073 offsetof(struct trace_iterator
, seq
));
3076 trace_event_read_lock();
3077 while (find_next_entry_inc(iter
) != NULL
) {
3078 enum print_line_t ret
;
3079 int len
= iter
->seq
.len
;
3081 ret
= print_trace_line(iter
);
3082 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3083 /* don't print partial lines */
3084 iter
->seq
.len
= len
;
3087 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3088 trace_consume(iter
);
3090 if (iter
->seq
.len
>= cnt
)
3093 trace_event_read_unlock();
3095 /* Now copy what we have to the user */
3096 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3097 if (iter
->seq
.readpos
>= iter
->seq
.len
)
3098 trace_seq_init(&iter
->seq
);
3101 * If there was nothing to send to user, inspite of consuming trace
3102 * entries, go back to wait for more entries.
3108 mutex_unlock(&iter
->mutex
);
3113 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
3114 struct pipe_buffer
*buf
)
3116 __free_page(buf
->page
);
3119 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
3122 __free_page(spd
->pages
[idx
]);
3125 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
3127 .map
= generic_pipe_buf_map
,
3128 .unmap
= generic_pipe_buf_unmap
,
3129 .confirm
= generic_pipe_buf_confirm
,
3130 .release
= tracing_pipe_buf_release
,
3131 .steal
= generic_pipe_buf_steal
,
3132 .get
= generic_pipe_buf_get
,
3136 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
3141 /* Seq buffer is page-sized, exactly what we need. */
3143 count
= iter
->seq
.len
;
3144 ret
= print_trace_line(iter
);
3145 count
= iter
->seq
.len
- count
;
3148 iter
->seq
.len
-= count
;
3151 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3152 iter
->seq
.len
-= count
;
3156 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3157 trace_consume(iter
);
3159 if (!find_next_entry_inc(iter
)) {
3169 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3171 struct pipe_inode_info
*pipe
,
3175 struct page
*pages
[PIPE_BUFFERS
];
3176 struct partial_page partial
[PIPE_BUFFERS
];
3177 struct trace_iterator
*iter
= filp
->private_data
;
3178 struct splice_pipe_desc spd
= {
3181 .nr_pages
= 0, /* This gets updated below. */
3183 .ops
= &tracing_pipe_buf_ops
,
3184 .spd_release
= tracing_spd_release_pipe
,
3186 static struct tracer
*old_tracer
;
3191 /* copy the tracer to avoid using a global lock all around */
3192 mutex_lock(&trace_types_lock
);
3193 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3194 old_tracer
= current_trace
;
3195 *iter
->trace
= *current_trace
;
3197 mutex_unlock(&trace_types_lock
);
3199 mutex_lock(&iter
->mutex
);
3201 if (iter
->trace
->splice_read
) {
3202 ret
= iter
->trace
->splice_read(iter
, filp
,
3203 ppos
, pipe
, len
, flags
);
3208 ret
= tracing_wait_pipe(filp
);
3212 if (!iter
->ent
&& !find_next_entry_inc(iter
)) {
3217 trace_event_read_lock();
3219 /* Fill as many pages as possible. */
3220 for (i
= 0, rem
= len
; i
< PIPE_BUFFERS
&& rem
; i
++) {
3221 pages
[i
] = alloc_page(GFP_KERNEL
);
3225 rem
= tracing_fill_pipe_page(rem
, iter
);
3227 /* Copy the data into the page, so we can start over. */
3228 ret
= trace_seq_to_buffer(&iter
->seq
,
3229 page_address(pages
[i
]),
3232 __free_page(pages
[i
]);
3235 partial
[i
].offset
= 0;
3236 partial
[i
].len
= iter
->seq
.len
;
3238 trace_seq_init(&iter
->seq
);
3241 trace_event_read_unlock();
3242 mutex_unlock(&iter
->mutex
);
3246 return splice_to_pipe(pipe
, &spd
);
3249 mutex_unlock(&iter
->mutex
);
3255 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3256 size_t cnt
, loff_t
*ppos
)
3258 struct trace_array
*tr
= filp
->private_data
;
3262 mutex_lock(&trace_types_lock
);
3263 if (!ring_buffer_expanded
)
3264 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3266 trace_buf_size
>> 10);
3268 r
= sprintf(buf
, "%lu\n", tr
->entries
>> 10);
3269 mutex_unlock(&trace_types_lock
);
3271 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3275 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3276 size_t cnt
, loff_t
*ppos
)
3282 if (cnt
>= sizeof(buf
))
3285 if (copy_from_user(&buf
, ubuf
, cnt
))
3290 ret
= strict_strtoul(buf
, 10, &val
);
3294 /* must have at least 1 entry */
3298 mutex_lock(&trace_types_lock
);
3302 /* disable all cpu buffers */
3303 for_each_tracing_cpu(cpu
) {
3304 if (global_trace
.data
[cpu
])
3305 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
3306 if (max_tr
.data
[cpu
])
3307 atomic_inc(&max_tr
.data
[cpu
]->disabled
);
3310 /* value is in KB */
3313 if (val
!= global_trace
.entries
) {
3314 ret
= tracing_resize_ring_buffer(val
);
3323 /* If check pages failed, return ENOMEM */
3324 if (tracing_disabled
)
3327 for_each_tracing_cpu(cpu
) {
3328 if (global_trace
.data
[cpu
])
3329 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
3330 if (max_tr
.data
[cpu
])
3331 atomic_dec(&max_tr
.data
[cpu
]->disabled
);
3335 max_tr
.entries
= global_trace
.entries
;
3336 mutex_unlock(&trace_types_lock
);
3341 static int mark_printk(const char *fmt
, ...)
3345 va_start(args
, fmt
);
3346 ret
= trace_vprintk(0, fmt
, args
);
3352 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3353 size_t cnt
, loff_t
*fpos
)
3357 if (tracing_disabled
)
3360 if (cnt
> TRACE_BUF_SIZE
)
3361 cnt
= TRACE_BUF_SIZE
;
3363 buf
= kmalloc(cnt
+ 2, GFP_KERNEL
);
3367 if (copy_from_user(buf
, ubuf
, cnt
)) {
3371 if (buf
[cnt
-1] != '\n') {
3377 cnt
= mark_printk("%s", buf
);
3384 static int tracing_clock_show(struct seq_file
*m
, void *v
)
3388 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
3390 "%s%s%s%s", i
? " " : "",
3391 i
== trace_clock_id
? "[" : "", trace_clocks
[i
].name
,
3392 i
== trace_clock_id
? "]" : "");
3398 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
3399 size_t cnt
, loff_t
*fpos
)
3402 const char *clockstr
;
3405 if (cnt
>= sizeof(buf
))
3408 if (copy_from_user(&buf
, ubuf
, cnt
))
3413 clockstr
= strstrip(buf
);
3415 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
3416 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
3419 if (i
== ARRAY_SIZE(trace_clocks
))
3424 mutex_lock(&trace_types_lock
);
3426 ring_buffer_set_clock(global_trace
.buffer
, trace_clocks
[i
].func
);
3428 ring_buffer_set_clock(max_tr
.buffer
, trace_clocks
[i
].func
);
3430 mutex_unlock(&trace_types_lock
);
3437 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
3439 if (tracing_disabled
)
3441 return single_open(file
, tracing_clock_show
, NULL
);
3444 static const struct file_operations tracing_max_lat_fops
= {
3445 .open
= tracing_open_generic
,
3446 .read
= tracing_max_lat_read
,
3447 .write
= tracing_max_lat_write
,
3450 static const struct file_operations tracing_ctrl_fops
= {
3451 .open
= tracing_open_generic
,
3452 .read
= tracing_ctrl_read
,
3453 .write
= tracing_ctrl_write
,
3456 static const struct file_operations set_tracer_fops
= {
3457 .open
= tracing_open_generic
,
3458 .read
= tracing_set_trace_read
,
3459 .write
= tracing_set_trace_write
,
3462 static const struct file_operations tracing_pipe_fops
= {
3463 .open
= tracing_open_pipe
,
3464 .poll
= tracing_poll_pipe
,
3465 .read
= tracing_read_pipe
,
3466 .splice_read
= tracing_splice_read_pipe
,
3467 .release
= tracing_release_pipe
,
3470 static const struct file_operations tracing_entries_fops
= {
3471 .open
= tracing_open_generic
,
3472 .read
= tracing_entries_read
,
3473 .write
= tracing_entries_write
,
3476 static const struct file_operations tracing_mark_fops
= {
3477 .open
= tracing_open_generic
,
3478 .write
= tracing_mark_write
,
3481 static const struct file_operations trace_clock_fops
= {
3482 .open
= tracing_clock_open
,
3484 .llseek
= seq_lseek
,
3485 .release
= single_release
,
3486 .write
= tracing_clock_write
,
3489 struct ftrace_buffer_info
{
3490 struct trace_array
*tr
;
3496 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
3498 int cpu
= (int)(long)inode
->i_private
;
3499 struct ftrace_buffer_info
*info
;
3501 if (tracing_disabled
)
3504 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3508 info
->tr
= &global_trace
;
3511 /* Force reading ring buffer for first read */
3512 info
->read
= (unsigned int)-1;
3514 filp
->private_data
= info
;
3516 return nonseekable_open(inode
, filp
);
3520 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
3521 size_t count
, loff_t
*ppos
)
3523 struct ftrace_buffer_info
*info
= filp
->private_data
;
3532 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
);
3536 /* Do we have previous read data to read? */
3537 if (info
->read
< PAGE_SIZE
)
3542 ret
= ring_buffer_read_page(info
->tr
->buffer
,
3549 pos
= ring_buffer_page_len(info
->spare
);
3551 if (pos
< PAGE_SIZE
)
3552 memset(info
->spare
+ pos
, 0, PAGE_SIZE
- pos
);
3555 size
= PAGE_SIZE
- info
->read
;
3559 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
3570 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
3572 struct ftrace_buffer_info
*info
= file
->private_data
;
3575 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
3582 struct ring_buffer
*buffer
;
3587 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
3588 struct pipe_buffer
*buf
)
3590 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3595 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3600 static int buffer_pipe_buf_steal(struct pipe_inode_info
*pipe
,
3601 struct pipe_buffer
*buf
)
3606 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
3607 struct pipe_buffer
*buf
)
3609 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3614 /* Pipe buffer operations for a buffer. */
3615 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
3617 .map
= generic_pipe_buf_map
,
3618 .unmap
= generic_pipe_buf_unmap
,
3619 .confirm
= generic_pipe_buf_confirm
,
3620 .release
= buffer_pipe_buf_release
,
3621 .steal
= buffer_pipe_buf_steal
,
3622 .get
= buffer_pipe_buf_get
,
3626 * Callback from splice_to_pipe(), if we need to release some pages
3627 * at the end of the spd in case we error'ed out in filling the pipe.
3629 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
3631 struct buffer_ref
*ref
=
3632 (struct buffer_ref
*)spd
->partial
[i
].private;
3637 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3639 spd
->partial
[i
].private = 0;
3643 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
3644 struct pipe_inode_info
*pipe
, size_t len
,
3647 struct ftrace_buffer_info
*info
= file
->private_data
;
3648 struct partial_page partial
[PIPE_BUFFERS
];
3649 struct page
*pages
[PIPE_BUFFERS
];
3650 struct splice_pipe_desc spd
= {
3654 .ops
= &buffer_pipe_buf_ops
,
3655 .spd_release
= buffer_spd_release
,
3657 struct buffer_ref
*ref
;
3658 int entries
, size
, i
;
3661 if (*ppos
& (PAGE_SIZE
- 1)) {
3662 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3666 if (len
& (PAGE_SIZE
- 1)) {
3667 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3668 if (len
< PAGE_SIZE
)
3673 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
3675 for (i
= 0; i
< PIPE_BUFFERS
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
3679 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
3684 ref
->buffer
= info
->tr
->buffer
;
3685 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
);
3691 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
3694 ring_buffer_free_read_page(ref
->buffer
,
3701 * zero out any left over data, this is going to
3704 size
= ring_buffer_page_len(ref
->page
);
3705 if (size
< PAGE_SIZE
)
3706 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
3708 page
= virt_to_page(ref
->page
);
3710 spd
.pages
[i
] = page
;
3711 spd
.partial
[i
].len
= PAGE_SIZE
;
3712 spd
.partial
[i
].offset
= 0;
3713 spd
.partial
[i
].private = (unsigned long)ref
;
3717 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
3722 /* did we read anything? */
3723 if (!spd
.nr_pages
) {
3724 if (flags
& SPLICE_F_NONBLOCK
)
3732 ret
= splice_to_pipe(pipe
, &spd
);
3737 static const struct file_operations tracing_buffers_fops
= {
3738 .open
= tracing_buffers_open
,
3739 .read
= tracing_buffers_read
,
3740 .release
= tracing_buffers_release
,
3741 .splice_read
= tracing_buffers_splice_read
,
3742 .llseek
= no_llseek
,
3746 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
3747 size_t count
, loff_t
*ppos
)
3749 unsigned long cpu
= (unsigned long)filp
->private_data
;
3750 struct trace_array
*tr
= &global_trace
;
3751 struct trace_seq
*s
;
3754 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
3760 cnt
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
3761 trace_seq_printf(s
, "entries: %ld\n", cnt
);
3763 cnt
= ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
3764 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
3766 cnt
= ring_buffer_commit_overrun_cpu(tr
->buffer
, cpu
);
3767 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
3769 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
3776 static const struct file_operations tracing_stats_fops
= {
3777 .open
= tracing_open_generic
,
3778 .read
= tracing_stats_read
,
3781 #ifdef CONFIG_DYNAMIC_FTRACE
3783 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
3789 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
3790 size_t cnt
, loff_t
*ppos
)
3792 static char ftrace_dyn_info_buffer
[1024];
3793 static DEFINE_MUTEX(dyn_info_mutex
);
3794 unsigned long *p
= filp
->private_data
;
3795 char *buf
= ftrace_dyn_info_buffer
;
3796 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
3799 mutex_lock(&dyn_info_mutex
);
3800 r
= sprintf(buf
, "%ld ", *p
);
3802 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
3805 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3807 mutex_unlock(&dyn_info_mutex
);
3812 static const struct file_operations tracing_dyn_info_fops
= {
3813 .open
= tracing_open_generic
,
3814 .read
= tracing_read_dyn_info
,
3818 static struct dentry
*d_tracer
;
3820 struct dentry
*tracing_init_dentry(void)
3827 if (!debugfs_initialized())
3830 d_tracer
= debugfs_create_dir("tracing", NULL
);
3832 if (!d_tracer
&& !once
) {
3834 pr_warning("Could not create debugfs directory 'tracing'\n");
3841 static struct dentry
*d_percpu
;
3843 struct dentry
*tracing_dentry_percpu(void)
3846 struct dentry
*d_tracer
;
3851 d_tracer
= tracing_init_dentry();
3856 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
3858 if (!d_percpu
&& !once
) {
3860 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3867 static void tracing_init_debugfs_percpu(long cpu
)
3869 struct dentry
*d_percpu
= tracing_dentry_percpu();
3870 struct dentry
*d_cpu
;
3871 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3874 if (cpu
> 999 || cpu
< 0)
3877 sprintf(cpu_dir
, "cpu%ld", cpu
);
3878 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
3880 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
3884 /* per cpu trace_pipe */
3885 trace_create_file("trace_pipe", 0444, d_cpu
,
3886 (void *) cpu
, &tracing_pipe_fops
);
3889 trace_create_file("trace", 0644, d_cpu
,
3890 (void *) cpu
, &tracing_fops
);
3892 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
3893 (void *) cpu
, &tracing_buffers_fops
);
3895 trace_create_file("stats", 0444, d_cpu
,
3896 (void *) cpu
, &tracing_stats_fops
);
3899 #ifdef CONFIG_FTRACE_SELFTEST
3900 /* Let selftest have access to static functions in this file */
3901 #include "trace_selftest.c"
3904 struct trace_option_dentry
{
3905 struct tracer_opt
*opt
;
3906 struct tracer_flags
*flags
;
3907 struct dentry
*entry
;
3911 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
3914 struct trace_option_dentry
*topt
= filp
->private_data
;
3917 if (topt
->flags
->val
& topt
->opt
->bit
)
3922 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
3926 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
3929 struct trace_option_dentry
*topt
= filp
->private_data
;
3934 if (cnt
>= sizeof(buf
))
3937 if (copy_from_user(&buf
, ubuf
, cnt
))
3942 ret
= strict_strtoul(buf
, 10, &val
);
3946 if (val
!= 0 && val
!= 1)
3949 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
3950 mutex_lock(&trace_types_lock
);
3951 ret
= __set_tracer_option(current_trace
, topt
->flags
,
3953 mutex_unlock(&trace_types_lock
);
3964 static const struct file_operations trace_options_fops
= {
3965 .open
= tracing_open_generic
,
3966 .read
= trace_options_read
,
3967 .write
= trace_options_write
,
3971 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
3974 long index
= (long)filp
->private_data
;
3977 if (trace_flags
& (1 << index
))
3982 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
3986 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
3989 long index
= (long)filp
->private_data
;
3994 if (cnt
>= sizeof(buf
))
3997 if (copy_from_user(&buf
, ubuf
, cnt
))
4002 ret
= strict_strtoul(buf
, 10, &val
);
4006 if (val
!= 0 && val
!= 1)
4008 set_tracer_flags(1 << index
, val
);
4015 static const struct file_operations trace_options_core_fops
= {
4016 .open
= tracing_open_generic
,
4017 .read
= trace_options_core_read
,
4018 .write
= trace_options_core_write
,
4021 struct dentry
*trace_create_file(const char *name
,
4023 struct dentry
*parent
,
4025 const struct file_operations
*fops
)
4029 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
4031 pr_warning("Could not create debugfs '%s' entry\n", name
);
4037 static struct dentry
*trace_options_init_dentry(void)
4039 struct dentry
*d_tracer
;
4040 static struct dentry
*t_options
;
4045 d_tracer
= tracing_init_dentry();
4049 t_options
= debugfs_create_dir("options", d_tracer
);
4051 pr_warning("Could not create debugfs directory 'options'\n");
4059 create_trace_option_file(struct trace_option_dentry
*topt
,
4060 struct tracer_flags
*flags
,
4061 struct tracer_opt
*opt
)
4063 struct dentry
*t_options
;
4065 t_options
= trace_options_init_dentry();
4069 topt
->flags
= flags
;
4072 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
4073 &trace_options_fops
);
4077 static struct trace_option_dentry
*
4078 create_trace_option_files(struct tracer
*tracer
)
4080 struct trace_option_dentry
*topts
;
4081 struct tracer_flags
*flags
;
4082 struct tracer_opt
*opts
;
4088 flags
= tracer
->flags
;
4090 if (!flags
|| !flags
->opts
)
4095 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4098 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
4102 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4103 create_trace_option_file(&topts
[cnt
], flags
,
4110 destroy_trace_option_files(struct trace_option_dentry
*topts
)
4117 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
4118 if (topts
[cnt
].entry
)
4119 debugfs_remove(topts
[cnt
].entry
);
4125 static struct dentry
*
4126 create_trace_option_core_file(const char *option
, long index
)
4128 struct dentry
*t_options
;
4130 t_options
= trace_options_init_dentry();
4134 return trace_create_file(option
, 0644, t_options
, (void *)index
,
4135 &trace_options_core_fops
);
4138 static __init
void create_trace_options_dir(void)
4140 struct dentry
*t_options
;
4143 t_options
= trace_options_init_dentry();
4147 for (i
= 0; trace_options
[i
]; i
++)
4148 create_trace_option_core_file(trace_options
[i
], i
);
4151 static __init
int tracer_init_debugfs(void)
4153 struct dentry
*d_tracer
;
4156 d_tracer
= tracing_init_dentry();
4158 trace_create_file("tracing_enabled", 0644, d_tracer
,
4159 &global_trace
, &tracing_ctrl_fops
);
4161 trace_create_file("trace_options", 0644, d_tracer
,
4162 NULL
, &tracing_iter_fops
);
4164 trace_create_file("tracing_cpumask", 0644, d_tracer
,
4165 NULL
, &tracing_cpumask_fops
);
4167 trace_create_file("trace", 0644, d_tracer
,
4168 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
4170 trace_create_file("available_tracers", 0444, d_tracer
,
4171 &global_trace
, &show_traces_fops
);
4173 trace_create_file("current_tracer", 0644, d_tracer
,
4174 &global_trace
, &set_tracer_fops
);
4176 #ifdef CONFIG_TRACER_MAX_TRACE
4177 trace_create_file("tracing_max_latency", 0644, d_tracer
,
4178 &tracing_max_latency
, &tracing_max_lat_fops
);
4180 trace_create_file("tracing_thresh", 0644, d_tracer
,
4181 &tracing_thresh
, &tracing_max_lat_fops
);
4184 trace_create_file("README", 0444, d_tracer
,
4185 NULL
, &tracing_readme_fops
);
4187 trace_create_file("trace_pipe", 0444, d_tracer
,
4188 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
4190 trace_create_file("buffer_size_kb", 0644, d_tracer
,
4191 &global_trace
, &tracing_entries_fops
);
4193 trace_create_file("trace_marker", 0220, d_tracer
,
4194 NULL
, &tracing_mark_fops
);
4196 trace_create_file("saved_cmdlines", 0444, d_tracer
,
4197 NULL
, &tracing_saved_cmdlines_fops
);
4199 trace_create_file("trace_clock", 0644, d_tracer
, NULL
,
4202 #ifdef CONFIG_DYNAMIC_FTRACE
4203 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
4204 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
4206 #ifdef CONFIG_SYSPROF_TRACER
4207 init_tracer_sysprof_debugfs(d_tracer
);
4210 create_trace_options_dir();
4212 for_each_tracing_cpu(cpu
)
4213 tracing_init_debugfs_percpu(cpu
);
4218 static int trace_panic_handler(struct notifier_block
*this,
4219 unsigned long event
, void *unused
)
4221 if (ftrace_dump_on_oops
)
4226 static struct notifier_block trace_panic_notifier
= {
4227 .notifier_call
= trace_panic_handler
,
4229 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
4232 static int trace_die_handler(struct notifier_block
*self
,
4238 if (ftrace_dump_on_oops
)
4247 static struct notifier_block trace_die_notifier
= {
4248 .notifier_call
= trace_die_handler
,
4253 * printk is set to max of 1024, we really don't need it that big.
4254 * Nothing should be printing 1000 characters anyway.
4256 #define TRACE_MAX_PRINT 1000
4259 * Define here KERN_TRACE so that we have one place to modify
4260 * it if we decide to change what log level the ftrace dump
4263 #define KERN_TRACE KERN_EMERG
4266 trace_printk_seq(struct trace_seq
*s
)
4268 /* Probably should print a warning here. */
4272 /* should be zero ended, but we are paranoid. */
4273 s
->buffer
[s
->len
] = 0;
4275 printk(KERN_TRACE
"%s", s
->buffer
);
4280 static void __ftrace_dump(bool disable_tracing
)
4282 static arch_spinlock_t ftrace_dump_lock
=
4283 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
4284 /* use static because iter can be a bit big for the stack */
4285 static struct trace_iterator iter
;
4286 unsigned int old_userobj
;
4287 static int dump_ran
;
4288 unsigned long flags
;
4292 local_irq_save(flags
);
4293 arch_spin_lock(&ftrace_dump_lock
);
4301 if (disable_tracing
)
4304 for_each_tracing_cpu(cpu
) {
4305 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
4308 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
4310 /* don't look at user memory in panic mode */
4311 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
4313 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
4315 /* Simulate the iterator */
4316 iter
.tr
= &global_trace
;
4317 iter
.trace
= current_trace
;
4318 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4321 * We need to stop all tracing on all CPUS to read the
4322 * the next buffer. This is a bit expensive, but is
4323 * not done often. We fill all what we can read,
4324 * and then release the locks again.
4327 while (!trace_empty(&iter
)) {
4330 printk(KERN_TRACE
"---------------------------------\n");
4334 /* reset all but tr, trace, and overruns */
4335 memset(&iter
.seq
, 0,
4336 sizeof(struct trace_iterator
) -
4337 offsetof(struct trace_iterator
, seq
));
4338 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
4341 if (find_next_entry_inc(&iter
) != NULL
) {
4344 ret
= print_trace_line(&iter
);
4345 if (ret
!= TRACE_TYPE_NO_CONSUME
)
4346 trace_consume(&iter
);
4349 trace_printk_seq(&iter
.seq
);
4353 printk(KERN_TRACE
" (ftrace buffer empty)\n");
4355 printk(KERN_TRACE
"---------------------------------\n");
4357 /* Re-enable tracing if requested */
4358 if (!disable_tracing
) {
4359 trace_flags
|= old_userobj
;
4361 for_each_tracing_cpu(cpu
) {
4362 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
4368 arch_spin_unlock(&ftrace_dump_lock
);
4369 local_irq_restore(flags
);
4372 /* By default: disable tracing after the dump */
4373 void ftrace_dump(void)
4375 __ftrace_dump(true);
4378 __init
static int tracer_alloc_buffers(void)
4384 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
4387 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
4388 goto out_free_buffer_mask
;
4390 if (!zalloc_cpumask_var(&tracing_reader_cpumask
, GFP_KERNEL
))
4391 goto out_free_tracing_cpumask
;
4393 /* To save memory, keep the ring buffer size to its minimum */
4394 if (ring_buffer_expanded
)
4395 ring_buf_size
= trace_buf_size
;
4399 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
4400 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
4402 /* TODO: make the number of buffers hot pluggable with CPUS */
4403 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
,
4404 TRACE_BUFFER_FLAGS
);
4405 if (!global_trace
.buffer
) {
4406 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
4408 goto out_free_cpumask
;
4410 global_trace
.entries
= ring_buffer_size(global_trace
.buffer
);
4413 #ifdef CONFIG_TRACER_MAX_TRACE
4414 max_tr
.buffer
= ring_buffer_alloc(ring_buf_size
,
4415 TRACE_BUFFER_FLAGS
);
4416 if (!max_tr
.buffer
) {
4417 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
4419 ring_buffer_free(global_trace
.buffer
);
4420 goto out_free_cpumask
;
4422 max_tr
.entries
= ring_buffer_size(max_tr
.buffer
);
4423 WARN_ON(max_tr
.entries
!= global_trace
.entries
);
4426 /* Allocate the first page for all buffers */
4427 for_each_tracing_cpu(i
) {
4428 global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
4429 max_tr
.data
[i
] = &per_cpu(max_tr_data
, i
);
4432 trace_init_cmdlines();
4434 register_tracer(&nop_trace
);
4435 current_trace
= &nop_trace
;
4436 #ifdef CONFIG_BOOT_TRACER
4437 register_tracer(&boot_tracer
);
4439 /* All seems OK, enable tracing */
4440 tracing_disabled
= 0;
4442 atomic_notifier_chain_register(&panic_notifier_list
,
4443 &trace_panic_notifier
);
4445 register_die_notifier(&trace_die_notifier
);
4450 free_cpumask_var(tracing_reader_cpumask
);
4451 out_free_tracing_cpumask
:
4452 free_cpumask_var(tracing_cpumask
);
4453 out_free_buffer_mask
:
4454 free_cpumask_var(tracing_buffer_mask
);
4459 __init
static int clear_boot_tracer(void)
4462 * The default tracer at boot buffer is an init section.
4463 * This function is called in lateinit. If we did not
4464 * find the boot tracer, then clear it out, to prevent
4465 * later registration from accessing the buffer that is
4466 * about to be freed.
4468 if (!default_bootup_tracer
)
4471 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
4472 default_bootup_tracer
);
4473 default_bootup_tracer
= NULL
;
4478 early_initcall(tracer_alloc_buffers
);
4479 fs_initcall(tracer_init_debugfs
);
4480 late_initcall(clear_boot_tracer
);