2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 William Lee Irwin III
14 #include <linux/ring_buffer.h>
15 #include <linux/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/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(local_t
, ftrace_cpu_disabled
);
91 static inline void ftrace_disable_cpu(void)
94 local_inc(&__get_cpu_var(ftrace_cpu_disabled
));
97 static inline void ftrace_enable_cpu(void)
99 local_dec(&__get_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_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
[] = {
335 { trace_clock_local
, "local" },
336 { trace_clock_global
, "global" },
342 * trace_parser_get_init - gets the buffer for trace parser
344 int trace_parser_get_init(struct trace_parser
*parser
, int size
)
346 memset(parser
, 0, sizeof(*parser
));
348 parser
->buffer
= kmalloc(size
, GFP_KERNEL
);
357 * trace_parser_put - frees the buffer for trace parser
359 void trace_parser_put(struct trace_parser
*parser
)
361 kfree(parser
->buffer
);
365 * trace_get_user - reads the user input string separated by space
366 * (matched by isspace(ch))
368 * For each string found the 'struct trace_parser' is updated,
369 * and the function returns.
371 * Returns number of bytes read.
373 * See kernel/trace/trace.h for 'struct trace_parser' details.
375 int trace_get_user(struct trace_parser
*parser
, const char __user
*ubuf
,
376 size_t cnt
, loff_t
*ppos
)
383 trace_parser_clear(parser
);
385 ret
= get_user(ch
, ubuf
++);
393 * The parser is not finished with the last write,
394 * continue reading the user input without skipping spaces.
397 /* skip white space */
398 while (cnt
&& isspace(ch
)) {
399 ret
= get_user(ch
, ubuf
++);
406 /* only spaces were written */
416 /* read the non-space input */
417 while (cnt
&& !isspace(ch
)) {
418 if (parser
->idx
< parser
->size
- 1)
419 parser
->buffer
[parser
->idx
++] = ch
;
424 ret
= get_user(ch
, ubuf
++);
431 /* We either got finished input or we have to wait for another call. */
433 parser
->buffer
[parser
->idx
] = 0;
434 parser
->cont
= false;
437 parser
->buffer
[parser
->idx
++] = ch
;
447 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
455 if (s
->len
<= s
->readpos
)
458 len
= s
->len
- s
->readpos
;
461 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
471 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
476 if (s
->len
<= s
->readpos
)
479 len
= s
->len
- s
->readpos
;
482 ret
= memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
491 * ftrace_max_lock is used to protect the swapping of buffers
492 * when taking a max snapshot. The buffers themselves are
493 * protected by per_cpu spinlocks. But the action of the swap
494 * needs its own lock.
496 * This is defined as a raw_spinlock_t in order to help
497 * with performance when lockdep debugging is enabled.
499 * It is also used in other places outside the update_max_tr
500 * so it needs to be defined outside of the
501 * CONFIG_TRACER_MAX_TRACE.
503 static raw_spinlock_t ftrace_max_lock
=
504 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
506 #ifdef CONFIG_TRACER_MAX_TRACE
507 unsigned long __read_mostly tracing_max_latency
;
508 unsigned long __read_mostly tracing_thresh
;
511 * Copy the new maximum trace into the separate maximum-trace
512 * structure. (this way the maximum trace is permanently saved,
513 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
516 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
518 struct trace_array_cpu
*data
= tr
->data
[cpu
];
519 struct trace_array_cpu
*max_data
= tr
->data
[cpu
];
522 max_tr
.time_start
= data
->preempt_timestamp
;
524 max_data
= max_tr
.data
[cpu
];
525 max_data
->saved_latency
= tracing_max_latency
;
526 max_data
->critical_start
= data
->critical_start
;
527 max_data
->critical_end
= data
->critical_end
;
529 memcpy(data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
530 max_data
->pid
= tsk
->pid
;
531 max_data
->uid
= task_uid(tsk
);
532 max_data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
533 max_data
->policy
= tsk
->policy
;
534 max_data
->rt_priority
= tsk
->rt_priority
;
536 /* record this tasks comm */
537 tracing_record_cmdline(tsk
);
541 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
543 * @tsk: the task with the latency
544 * @cpu: The cpu that initiated the trace.
546 * Flip the buffers between the @tr and the max_tr and record information
547 * about which task was the cause of this latency.
550 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
552 struct ring_buffer
*buf
= tr
->buffer
;
554 if (trace_stop_count
)
557 WARN_ON_ONCE(!irqs_disabled());
558 __raw_spin_lock(&ftrace_max_lock
);
560 tr
->buffer
= max_tr
.buffer
;
563 __update_max_tr(tr
, tsk
, cpu
);
564 __raw_spin_unlock(&ftrace_max_lock
);
568 * update_max_tr_single - only copy one trace over, and reset the rest
570 * @tsk - task with the latency
571 * @cpu - the cpu of the buffer to copy.
573 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
576 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
580 if (trace_stop_count
)
583 WARN_ON_ONCE(!irqs_disabled());
584 __raw_spin_lock(&ftrace_max_lock
);
586 ftrace_disable_cpu();
588 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
592 * We failed to swap the buffer due to a commit taking
593 * place on this CPU. We fail to record, but we reset
594 * the max trace buffer (no one writes directly to it)
595 * and flag that it failed.
597 trace_array_printk(&max_tr
, _THIS_IP_
,
598 "Failed to swap buffers due to commit in progress\n");
603 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
&& ret
!= -EBUSY
);
605 __update_max_tr(tr
, tsk
, cpu
);
606 __raw_spin_unlock(&ftrace_max_lock
);
608 #endif /* CONFIG_TRACER_MAX_TRACE */
611 * register_tracer - register a tracer with the ftrace system.
612 * @type - the plugin for the tracer
614 * Register a new plugin tracer.
616 int register_tracer(struct tracer
*type
)
617 __releases(kernel_lock
)
618 __acquires(kernel_lock
)
624 pr_info("Tracer must have a name\n");
628 if (strlen(type
->name
) > MAX_TRACER_SIZE
) {
629 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE
);
634 * When this gets called we hold the BKL which means that
635 * preemption is disabled. Various trace selftests however
636 * need to disable and enable preemption for successful tests.
637 * So we drop the BKL here and grab it after the tests again.
640 mutex_lock(&trace_types_lock
);
642 tracing_selftest_running
= true;
644 for (t
= trace_types
; t
; t
= t
->next
) {
645 if (strcmp(type
->name
, t
->name
) == 0) {
647 pr_info("Tracer %s already registered\n",
655 type
->set_flag
= &dummy_set_flag
;
657 type
->flags
= &dummy_tracer_flags
;
659 if (!type
->flags
->opts
)
660 type
->flags
->opts
= dummy_tracer_opt
;
661 if (!type
->wait_pipe
)
662 type
->wait_pipe
= default_wait_pipe
;
665 #ifdef CONFIG_FTRACE_STARTUP_TEST
666 if (type
->selftest
&& !tracing_selftest_disabled
) {
667 struct tracer
*saved_tracer
= current_trace
;
668 struct trace_array
*tr
= &global_trace
;
671 * Run a selftest on this tracer.
672 * Here we reset the trace buffer, and set the current
673 * tracer to be this tracer. The tracer can then run some
674 * internal tracing to verify that everything is in order.
675 * If we fail, we do not register this tracer.
677 tracing_reset_online_cpus(tr
);
679 current_trace
= type
;
680 /* the test is responsible for initializing and enabling */
681 pr_info("Testing tracer %s: ", type
->name
);
682 ret
= type
->selftest(type
, tr
);
683 /* the test is responsible for resetting too */
684 current_trace
= saved_tracer
;
686 printk(KERN_CONT
"FAILED!\n");
689 /* Only reset on passing, to avoid touching corrupted buffers */
690 tracing_reset_online_cpus(tr
);
692 printk(KERN_CONT
"PASSED\n");
696 type
->next
= trace_types
;
700 tracing_selftest_running
= false;
701 mutex_unlock(&trace_types_lock
);
703 if (ret
|| !default_bootup_tracer
)
706 if (strncmp(default_bootup_tracer
, type
->name
, MAX_TRACER_SIZE
))
709 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
710 /* Do we want this tracer to start on bootup? */
711 tracing_set_tracer(type
->name
);
712 default_bootup_tracer
= NULL
;
713 /* disable other selftests, since this will break it. */
714 tracing_selftest_disabled
= 1;
715 #ifdef CONFIG_FTRACE_STARTUP_TEST
716 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
725 void unregister_tracer(struct tracer
*type
)
729 mutex_lock(&trace_types_lock
);
730 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
734 pr_info("Tracer %s not registered\n", type
->name
);
740 if (type
== current_trace
&& tracer_enabled
) {
743 if (current_trace
->stop
)
744 current_trace
->stop(&global_trace
);
745 current_trace
= &nop_trace
;
748 mutex_unlock(&trace_types_lock
);
751 static void __tracing_reset(struct trace_array
*tr
, int cpu
)
753 ftrace_disable_cpu();
754 ring_buffer_reset_cpu(tr
->buffer
, cpu
);
758 void tracing_reset(struct trace_array
*tr
, int cpu
)
760 struct ring_buffer
*buffer
= tr
->buffer
;
762 ring_buffer_record_disable(buffer
);
764 /* Make sure all commits have finished */
766 __tracing_reset(tr
, cpu
);
768 ring_buffer_record_enable(buffer
);
771 void tracing_reset_online_cpus(struct trace_array
*tr
)
773 struct ring_buffer
*buffer
= tr
->buffer
;
776 ring_buffer_record_disable(buffer
);
778 /* Make sure all commits have finished */
781 tr
->time_start
= ftrace_now(tr
->cpu
);
783 for_each_online_cpu(cpu
)
784 __tracing_reset(tr
, cpu
);
786 ring_buffer_record_enable(buffer
);
789 void tracing_reset_current(int cpu
)
791 tracing_reset(&global_trace
, cpu
);
794 void tracing_reset_current_online_cpus(void)
796 tracing_reset_online_cpus(&global_trace
);
799 #define SAVED_CMDLINES 128
800 #define NO_CMDLINE_MAP UINT_MAX
801 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
802 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
803 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
804 static int cmdline_idx
;
805 static raw_spinlock_t trace_cmdline_lock
= __RAW_SPIN_LOCK_UNLOCKED
;
807 /* temporary disable recording */
808 static atomic_t trace_record_cmdline_disabled __read_mostly
;
810 static void trace_init_cmdlines(void)
812 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
813 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
817 int is_tracing_stopped(void)
819 return trace_stop_count
;
823 * ftrace_off_permanent - disable all ftrace code permanently
825 * This should only be called when a serious anomally has
826 * been detected. This will turn off the function tracing,
827 * ring buffers, and other tracing utilites. It takes no
828 * locks and can be called from any context.
830 void ftrace_off_permanent(void)
832 tracing_disabled
= 1;
834 tracing_off_permanent();
838 * tracing_start - quick start of the tracer
840 * If tracing is enabled but was stopped by tracing_stop,
841 * this will start the tracer back up.
843 void tracing_start(void)
845 struct ring_buffer
*buffer
;
848 if (tracing_disabled
)
851 spin_lock_irqsave(&tracing_start_lock
, flags
);
852 if (--trace_stop_count
) {
853 if (trace_stop_count
< 0) {
854 /* Someone screwed up their debugging */
856 trace_stop_count
= 0;
862 buffer
= global_trace
.buffer
;
864 ring_buffer_record_enable(buffer
);
866 buffer
= max_tr
.buffer
;
868 ring_buffer_record_enable(buffer
);
872 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
876 * tracing_stop - quick stop of the tracer
878 * Light weight way to stop tracing. Use in conjunction with
881 void tracing_stop(void)
883 struct ring_buffer
*buffer
;
887 spin_lock_irqsave(&tracing_start_lock
, flags
);
888 if (trace_stop_count
++)
891 buffer
= global_trace
.buffer
;
893 ring_buffer_record_disable(buffer
);
895 buffer
= max_tr
.buffer
;
897 ring_buffer_record_disable(buffer
);
900 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
903 void trace_stop_cmdline_recording(void);
905 static void trace_save_cmdline(struct task_struct
*tsk
)
909 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
913 * It's not the end of the world if we don't get
914 * the lock, but we also don't want to spin
915 * nor do we want to disable interrupts,
916 * so if we miss here, then better luck next time.
918 if (!__raw_spin_trylock(&trace_cmdline_lock
))
921 idx
= map_pid_to_cmdline
[tsk
->pid
];
922 if (idx
== NO_CMDLINE_MAP
) {
923 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
926 * Check whether the cmdline buffer at idx has a pid
927 * mapped. We are going to overwrite that entry so we
928 * need to clear the map_pid_to_cmdline. Otherwise we
929 * would read the new comm for the old pid.
931 pid
= map_cmdline_to_pid
[idx
];
932 if (pid
!= NO_CMDLINE_MAP
)
933 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
935 map_cmdline_to_pid
[idx
] = tsk
->pid
;
936 map_pid_to_cmdline
[tsk
->pid
] = idx
;
941 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
943 __raw_spin_unlock(&trace_cmdline_lock
);
946 void trace_find_cmdline(int pid
, char comm
[])
951 strcpy(comm
, "<idle>");
955 if (pid
> PID_MAX_DEFAULT
) {
956 strcpy(comm
, "<...>");
961 __raw_spin_lock(&trace_cmdline_lock
);
962 map
= map_pid_to_cmdline
[pid
];
963 if (map
!= NO_CMDLINE_MAP
)
964 strcpy(comm
, saved_cmdlines
[map
]);
966 strcpy(comm
, "<...>");
968 __raw_spin_unlock(&trace_cmdline_lock
);
972 void tracing_record_cmdline(struct task_struct
*tsk
)
974 if (atomic_read(&trace_record_cmdline_disabled
) || !tracer_enabled
||
978 trace_save_cmdline(tsk
);
982 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
985 struct task_struct
*tsk
= current
;
987 entry
->preempt_count
= pc
& 0xff;
988 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
989 entry
->lock_depth
= (tsk
) ? tsk
->lock_depth
: 0;
991 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
992 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
994 TRACE_FLAG_IRQS_NOSUPPORT
|
996 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
997 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
998 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
1000 EXPORT_SYMBOL_GPL(tracing_generic_entry_update
);
1002 struct ring_buffer_event
*
1003 trace_buffer_lock_reserve(struct ring_buffer
*buffer
,
1006 unsigned long flags
, int pc
)
1008 struct ring_buffer_event
*event
;
1010 event
= ring_buffer_lock_reserve(buffer
, len
);
1011 if (event
!= NULL
) {
1012 struct trace_entry
*ent
= ring_buffer_event_data(event
);
1014 tracing_generic_entry_update(ent
, flags
, pc
);
1022 __trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1023 struct ring_buffer_event
*event
,
1024 unsigned long flags
, int pc
,
1027 ring_buffer_unlock_commit(buffer
, event
);
1029 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1030 ftrace_trace_userstack(buffer
, flags
, pc
);
1036 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1037 struct ring_buffer_event
*event
,
1038 unsigned long flags
, int pc
)
1040 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1043 struct ring_buffer_event
*
1044 trace_current_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1045 int type
, unsigned long len
,
1046 unsigned long flags
, int pc
)
1048 *current_rb
= global_trace
.buffer
;
1049 return trace_buffer_lock_reserve(*current_rb
,
1050 type
, len
, flags
, pc
);
1052 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve
);
1054 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
1055 struct ring_buffer_event
*event
,
1056 unsigned long flags
, int pc
)
1058 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1060 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit
);
1062 void trace_nowake_buffer_unlock_commit(struct ring_buffer
*buffer
,
1063 struct ring_buffer_event
*event
,
1064 unsigned long flags
, int pc
)
1066 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 0);
1068 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit
);
1070 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
1071 struct ring_buffer_event
*event
)
1073 ring_buffer_discard_commit(buffer
, event
);
1075 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit
);
1078 trace_function(struct trace_array
*tr
,
1079 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1082 struct ftrace_event_call
*call
= &event_function
;
1083 struct ring_buffer
*buffer
= tr
->buffer
;
1084 struct ring_buffer_event
*event
;
1085 struct ftrace_entry
*entry
;
1087 /* If we are reading the ring buffer, don't trace */
1088 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled
))))
1091 event
= trace_buffer_lock_reserve(buffer
, TRACE_FN
, sizeof(*entry
),
1095 entry
= ring_buffer_event_data(event
);
1097 entry
->parent_ip
= parent_ip
;
1099 if (!filter_check_discard(call
, entry
, buffer
, event
))
1100 ring_buffer_unlock_commit(buffer
, event
);
1104 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1105 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1108 if (likely(!atomic_read(&data
->disabled
)))
1109 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
1112 #ifdef CONFIG_STACKTRACE
1113 static void __ftrace_trace_stack(struct ring_buffer
*buffer
,
1114 unsigned long flags
,
1117 struct ftrace_event_call
*call
= &event_kernel_stack
;
1118 struct ring_buffer_event
*event
;
1119 struct stack_entry
*entry
;
1120 struct stack_trace trace
;
1122 event
= trace_buffer_lock_reserve(buffer
, TRACE_STACK
,
1123 sizeof(*entry
), flags
, pc
);
1126 entry
= ring_buffer_event_data(event
);
1127 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1129 trace
.nr_entries
= 0;
1130 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1132 trace
.entries
= entry
->caller
;
1134 save_stack_trace(&trace
);
1135 if (!filter_check_discard(call
, entry
, buffer
, event
))
1136 ring_buffer_unlock_commit(buffer
, event
);
1139 void ftrace_trace_stack(struct ring_buffer
*buffer
, unsigned long flags
,
1142 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1145 __ftrace_trace_stack(buffer
, flags
, skip
, pc
);
1148 void __trace_stack(struct trace_array
*tr
, unsigned long flags
, int skip
,
1151 __ftrace_trace_stack(tr
->buffer
, flags
, skip
, pc
);
1155 ftrace_trace_userstack(struct ring_buffer
*buffer
, unsigned long flags
, int pc
)
1157 struct ftrace_event_call
*call
= &event_user_stack
;
1158 struct ring_buffer_event
*event
;
1159 struct userstack_entry
*entry
;
1160 struct stack_trace trace
;
1162 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1165 event
= trace_buffer_lock_reserve(buffer
, TRACE_USER_STACK
,
1166 sizeof(*entry
), flags
, pc
);
1169 entry
= ring_buffer_event_data(event
);
1171 entry
->tgid
= current
->tgid
;
1172 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1174 trace
.nr_entries
= 0;
1175 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1177 trace
.entries
= entry
->caller
;
1179 save_stack_trace_user(&trace
);
1180 if (!filter_check_discard(call
, entry
, buffer
, event
))
1181 ring_buffer_unlock_commit(buffer
, event
);
1185 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1187 ftrace_trace_userstack(tr
, flags
, preempt_count());
1191 #endif /* CONFIG_STACKTRACE */
1194 ftrace_trace_special(void *__tr
,
1195 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
,
1198 struct ftrace_event_call
*call
= &event_special
;
1199 struct ring_buffer_event
*event
;
1200 struct trace_array
*tr
= __tr
;
1201 struct ring_buffer
*buffer
= tr
->buffer
;
1202 struct special_entry
*entry
;
1204 event
= trace_buffer_lock_reserve(buffer
, TRACE_SPECIAL
,
1205 sizeof(*entry
), 0, pc
);
1208 entry
= ring_buffer_event_data(event
);
1213 if (!filter_check_discard(call
, entry
, buffer
, event
))
1214 trace_buffer_unlock_commit(buffer
, event
, 0, pc
);
1218 __trace_special(void *__tr
, void *__data
,
1219 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1221 ftrace_trace_special(__tr
, arg1
, arg2
, arg3
, preempt_count());
1225 ftrace_special(unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1227 struct trace_array
*tr
= &global_trace
;
1228 struct trace_array_cpu
*data
;
1229 unsigned long flags
;
1233 if (tracing_disabled
)
1236 pc
= preempt_count();
1237 local_irq_save(flags
);
1238 cpu
= raw_smp_processor_id();
1239 data
= tr
->data
[cpu
];
1241 if (likely(atomic_inc_return(&data
->disabled
) == 1))
1242 ftrace_trace_special(tr
, arg1
, arg2
, arg3
, pc
);
1244 atomic_dec(&data
->disabled
);
1245 local_irq_restore(flags
);
1249 * trace_vbprintk - write binary msg to tracing buffer
1252 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1254 static raw_spinlock_t trace_buf_lock
=
1255 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
1256 static u32 trace_buf
[TRACE_BUF_SIZE
];
1258 struct ftrace_event_call
*call
= &event_bprint
;
1259 struct ring_buffer_event
*event
;
1260 struct ring_buffer
*buffer
;
1261 struct trace_array
*tr
= &global_trace
;
1262 struct trace_array_cpu
*data
;
1263 struct bprint_entry
*entry
;
1264 unsigned long flags
;
1267 int cpu
, len
= 0, size
, pc
;
1269 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1272 /* Don't pollute graph traces with trace_vprintk internals */
1273 pause_graph_tracing();
1275 pc
= preempt_count();
1276 resched
= ftrace_preempt_disable();
1277 cpu
= raw_smp_processor_id();
1278 data
= tr
->data
[cpu
];
1280 disable
= atomic_inc_return(&data
->disabled
);
1281 if (unlikely(disable
!= 1))
1284 /* Lockdep uses trace_printk for lock tracing */
1285 local_irq_save(flags
);
1286 __raw_spin_lock(&trace_buf_lock
);
1287 len
= vbin_printf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1289 if (len
> TRACE_BUF_SIZE
|| len
< 0)
1292 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1293 buffer
= tr
->buffer
;
1294 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
1298 entry
= ring_buffer_event_data(event
);
1302 memcpy(entry
->buf
, trace_buf
, sizeof(u32
) * len
);
1303 if (!filter_check_discard(call
, entry
, buffer
, event
))
1304 ring_buffer_unlock_commit(buffer
, event
);
1307 __raw_spin_unlock(&trace_buf_lock
);
1308 local_irq_restore(flags
);
1311 atomic_dec_return(&data
->disabled
);
1312 ftrace_preempt_enable(resched
);
1313 unpause_graph_tracing();
1317 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1319 int trace_array_printk(struct trace_array
*tr
,
1320 unsigned long ip
, const char *fmt
, ...)
1325 if (!(trace_flags
& TRACE_ITER_PRINTK
))
1329 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
1334 int trace_array_vprintk(struct trace_array
*tr
,
1335 unsigned long ip
, const char *fmt
, va_list args
)
1337 static raw_spinlock_t trace_buf_lock
= __RAW_SPIN_LOCK_UNLOCKED
;
1338 static char trace_buf
[TRACE_BUF_SIZE
];
1340 struct ftrace_event_call
*call
= &event_print
;
1341 struct ring_buffer_event
*event
;
1342 struct ring_buffer
*buffer
;
1343 struct trace_array_cpu
*data
;
1344 int cpu
, len
= 0, size
, pc
;
1345 struct print_entry
*entry
;
1346 unsigned long irq_flags
;
1349 if (tracing_disabled
|| tracing_selftest_running
)
1352 pc
= preempt_count();
1353 preempt_disable_notrace();
1354 cpu
= raw_smp_processor_id();
1355 data
= tr
->data
[cpu
];
1357 disable
= atomic_inc_return(&data
->disabled
);
1358 if (unlikely(disable
!= 1))
1361 pause_graph_tracing();
1362 raw_local_irq_save(irq_flags
);
1363 __raw_spin_lock(&trace_buf_lock
);
1365 strncpy(trace_buf
, fmt
, TRACE_BUF_SIZE
);
1366 len
= strlen(trace_buf
);
1368 len
= vsnprintf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1370 size
= sizeof(*entry
) + len
+ 1;
1371 buffer
= tr
->buffer
;
1372 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
1376 entry
= ring_buffer_event_data(event
);
1379 memcpy(&entry
->buf
, trace_buf
, len
);
1380 entry
->buf
[len
] = '\0';
1381 if (!filter_check_discard(call
, entry
, buffer
, event
))
1382 ring_buffer_unlock_commit(buffer
, event
);
1385 __raw_spin_unlock(&trace_buf_lock
);
1386 raw_local_irq_restore(irq_flags
);
1387 unpause_graph_tracing();
1389 atomic_dec_return(&data
->disabled
);
1390 preempt_enable_notrace();
1395 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1397 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
1399 EXPORT_SYMBOL_GPL(trace_vprintk
);
1401 enum trace_file_type
{
1402 TRACE_FILE_LAT_FMT
= 1,
1403 TRACE_FILE_ANNOTATE
= 2,
1406 static void trace_iterator_increment(struct trace_iterator
*iter
)
1408 /* Don't allow ftrace to trace into the ring buffers */
1409 ftrace_disable_cpu();
1412 if (iter
->buffer_iter
[iter
->cpu
])
1413 ring_buffer_read(iter
->buffer_iter
[iter
->cpu
], NULL
);
1415 ftrace_enable_cpu();
1418 static struct trace_entry
*
1419 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
)
1421 struct ring_buffer_event
*event
;
1422 struct ring_buffer_iter
*buf_iter
= iter
->buffer_iter
[cpu
];
1424 /* Don't allow ftrace to trace into the ring buffers */
1425 ftrace_disable_cpu();
1428 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1430 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
);
1432 ftrace_enable_cpu();
1434 return event
? ring_buffer_event_data(event
) : NULL
;
1437 static struct trace_entry
*
1438 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
, u64
*ent_ts
)
1440 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1441 struct trace_entry
*ent
, *next
= NULL
;
1442 int cpu_file
= iter
->cpu_file
;
1443 u64 next_ts
= 0, ts
;
1448 * If we are in a per_cpu trace file, don't bother by iterating over
1449 * all cpu and peek directly.
1451 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1452 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1454 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
);
1456 *ent_cpu
= cpu_file
;
1461 for_each_tracing_cpu(cpu
) {
1463 if (ring_buffer_empty_cpu(buffer
, cpu
))
1466 ent
= peek_next_entry(iter
, cpu
, &ts
);
1469 * Pick the entry with the smallest timestamp:
1471 if (ent
&& (!next
|| ts
< next_ts
)) {
1479 *ent_cpu
= next_cpu
;
1487 /* Find the next real entry, without updating the iterator itself */
1488 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1489 int *ent_cpu
, u64
*ent_ts
)
1491 return __find_next_entry(iter
, ent_cpu
, ent_ts
);
1494 /* Find the next real entry, and increment the iterator to the next entry */
1495 static void *find_next_entry_inc(struct trace_iterator
*iter
)
1497 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
, &iter
->ts
);
1500 trace_iterator_increment(iter
);
1502 return iter
->ent
? iter
: NULL
;
1505 static void trace_consume(struct trace_iterator
*iter
)
1507 /* Don't allow ftrace to trace into the ring buffers */
1508 ftrace_disable_cpu();
1509 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
);
1510 ftrace_enable_cpu();
1513 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1515 struct trace_iterator
*iter
= m
->private;
1519 WARN_ON_ONCE(iter
->leftover
);
1523 /* can't go backwards */
1528 ent
= find_next_entry_inc(iter
);
1532 while (ent
&& iter
->idx
< i
)
1533 ent
= find_next_entry_inc(iter
);
1540 static void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
1542 struct trace_array
*tr
= iter
->tr
;
1543 struct ring_buffer_event
*event
;
1544 struct ring_buffer_iter
*buf_iter
;
1545 unsigned long entries
= 0;
1548 tr
->data
[cpu
]->skipped_entries
= 0;
1550 if (!iter
->buffer_iter
[cpu
])
1553 buf_iter
= iter
->buffer_iter
[cpu
];
1554 ring_buffer_iter_reset(buf_iter
);
1557 * We could have the case with the max latency tracers
1558 * that a reset never took place on a cpu. This is evident
1559 * by the timestamp being before the start of the buffer.
1561 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
1562 if (ts
>= iter
->tr
->time_start
)
1565 ring_buffer_read(buf_iter
, NULL
);
1568 tr
->data
[cpu
]->skipped_entries
= entries
;
1572 * No necessary locking here. The worst thing which can
1573 * happen is loosing events consumed at the same time
1574 * by a trace_pipe reader.
1575 * Other than that, we don't risk to crash the ring buffer
1576 * because it serializes the readers.
1578 * The current tracer is copied to avoid a global locking
1581 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1583 struct trace_iterator
*iter
= m
->private;
1584 static struct tracer
*old_tracer
;
1585 int cpu_file
= iter
->cpu_file
;
1590 /* copy the tracer to avoid using a global lock all around */
1591 mutex_lock(&trace_types_lock
);
1592 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1593 old_tracer
= current_trace
;
1594 *iter
->trace
= *current_trace
;
1596 mutex_unlock(&trace_types_lock
);
1598 atomic_inc(&trace_record_cmdline_disabled
);
1600 if (*pos
!= iter
->pos
) {
1605 ftrace_disable_cpu();
1607 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1608 for_each_tracing_cpu(cpu
)
1609 tracing_iter_reset(iter
, cpu
);
1611 tracing_iter_reset(iter
, cpu_file
);
1613 ftrace_enable_cpu();
1615 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1620 * If we overflowed the seq_file before, then we want
1621 * to just reuse the trace_seq buffer again.
1627 p
= s_next(m
, p
, &l
);
1631 trace_event_read_lock();
1635 static void s_stop(struct seq_file
*m
, void *p
)
1637 atomic_dec(&trace_record_cmdline_disabled
);
1638 trace_event_read_unlock();
1641 static void print_lat_help_header(struct seq_file
*m
)
1643 seq_puts(m
, "# _------=> CPU# \n");
1644 seq_puts(m
, "# / _-----=> irqs-off \n");
1645 seq_puts(m
, "# | / _----=> need-resched \n");
1646 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1647 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1648 seq_puts(m
, "# |||| /_--=> lock-depth \n");
1649 seq_puts(m
, "# |||||/ delay \n");
1650 seq_puts(m
, "# cmd pid |||||| time | caller \n");
1651 seq_puts(m
, "# \\ / |||||| \\ | / \n");
1654 static void print_func_help_header(struct seq_file
*m
)
1656 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1657 seq_puts(m
, "# | | | | |\n");
1662 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1664 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1665 struct trace_array
*tr
= iter
->tr
;
1666 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1667 struct tracer
*type
= current_trace
;
1668 unsigned long entries
= 0;
1669 unsigned long total
= 0;
1670 unsigned long count
;
1671 const char *name
= "preemption";
1678 for_each_tracing_cpu(cpu
) {
1679 count
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
1681 * If this buffer has skipped entries, then we hold all
1682 * entries for the trace and we need to ignore the
1683 * ones before the time stamp.
1685 if (tr
->data
[cpu
]->skipped_entries
) {
1686 count
-= tr
->data
[cpu
]->skipped_entries
;
1687 /* total is the same as the entries */
1691 ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
1695 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
1697 seq_puts(m
, "# -----------------------------------"
1698 "---------------------------------\n");
1699 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1700 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1701 nsecs_to_usecs(data
->saved_latency
),
1705 #if defined(CONFIG_PREEMPT_NONE)
1707 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1709 #elif defined(CONFIG_PREEMPT)
1714 /* These are reserved for later use */
1717 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1721 seq_puts(m
, "# -----------------\n");
1722 seq_printf(m
, "# | task: %.16s-%d "
1723 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1724 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1725 data
->policy
, data
->rt_priority
);
1726 seq_puts(m
, "# -----------------\n");
1728 if (data
->critical_start
) {
1729 seq_puts(m
, "# => started at: ");
1730 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1731 trace_print_seq(m
, &iter
->seq
);
1732 seq_puts(m
, "\n# => ended at: ");
1733 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1734 trace_print_seq(m
, &iter
->seq
);
1735 seq_puts(m
, "\n#\n");
1741 static void test_cpu_buff_start(struct trace_iterator
*iter
)
1743 struct trace_seq
*s
= &iter
->seq
;
1745 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
1748 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
1751 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
1754 if (iter
->tr
->data
[iter
->cpu
]->skipped_entries
)
1757 cpumask_set_cpu(iter
->cpu
, iter
->started
);
1759 /* Don't print started cpu buffer for the first entry of the trace */
1761 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
1765 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
1767 struct trace_seq
*s
= &iter
->seq
;
1768 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1769 struct trace_entry
*entry
;
1770 struct trace_event
*event
;
1774 test_cpu_buff_start(iter
);
1776 event
= ftrace_find_event(entry
->type
);
1778 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1779 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1780 if (!trace_print_lat_context(iter
))
1783 if (!trace_print_context(iter
))
1789 return event
->trace(iter
, sym_flags
);
1791 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
1794 return TRACE_TYPE_HANDLED
;
1796 return TRACE_TYPE_PARTIAL_LINE
;
1799 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
1801 struct trace_seq
*s
= &iter
->seq
;
1802 struct trace_entry
*entry
;
1803 struct trace_event
*event
;
1807 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1808 if (!trace_seq_printf(s
, "%d %d %llu ",
1809 entry
->pid
, iter
->cpu
, iter
->ts
))
1813 event
= ftrace_find_event(entry
->type
);
1815 return event
->raw(iter
, 0);
1817 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
1820 return TRACE_TYPE_HANDLED
;
1822 return TRACE_TYPE_PARTIAL_LINE
;
1825 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
1827 struct trace_seq
*s
= &iter
->seq
;
1828 unsigned char newline
= '\n';
1829 struct trace_entry
*entry
;
1830 struct trace_event
*event
;
1834 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1835 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1836 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1837 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
1840 event
= ftrace_find_event(entry
->type
);
1842 enum print_line_t ret
= event
->hex(iter
, 0);
1843 if (ret
!= TRACE_TYPE_HANDLED
)
1847 SEQ_PUT_FIELD_RET(s
, newline
);
1849 return TRACE_TYPE_HANDLED
;
1852 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
1854 struct trace_seq
*s
= &iter
->seq
;
1855 struct trace_entry
*entry
;
1856 struct trace_event
*event
;
1860 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1861 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1862 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
1863 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
1866 event
= ftrace_find_event(entry
->type
);
1867 return event
? event
->binary(iter
, 0) : TRACE_TYPE_HANDLED
;
1870 static int trace_empty(struct trace_iterator
*iter
)
1874 /* If we are looking at one CPU buffer, only check that one */
1875 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
1876 cpu
= iter
->cpu_file
;
1877 if (iter
->buffer_iter
[cpu
]) {
1878 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1881 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1887 for_each_tracing_cpu(cpu
) {
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
))
1900 /* Called with trace_event_read_lock() held. */
1901 static enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
1903 enum print_line_t ret
;
1905 if (iter
->trace
&& iter
->trace
->print_line
) {
1906 ret
= iter
->trace
->print_line(iter
);
1907 if (ret
!= TRACE_TYPE_UNHANDLED
)
1911 if (iter
->ent
->type
== TRACE_BPRINT
&&
1912 trace_flags
& TRACE_ITER_PRINTK
&&
1913 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
1914 return trace_print_bprintk_msg_only(iter
);
1916 if (iter
->ent
->type
== TRACE_PRINT
&&
1917 trace_flags
& TRACE_ITER_PRINTK
&&
1918 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
1919 return trace_print_printk_msg_only(iter
);
1921 if (trace_flags
& TRACE_ITER_BIN
)
1922 return print_bin_fmt(iter
);
1924 if (trace_flags
& TRACE_ITER_HEX
)
1925 return print_hex_fmt(iter
);
1927 if (trace_flags
& TRACE_ITER_RAW
)
1928 return print_raw_fmt(iter
);
1930 return print_trace_fmt(iter
);
1933 static int s_show(struct seq_file
*m
, void *v
)
1935 struct trace_iterator
*iter
= v
;
1938 if (iter
->ent
== NULL
) {
1940 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
1943 if (iter
->trace
&& iter
->trace
->print_header
)
1944 iter
->trace
->print_header(m
);
1945 else if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1946 /* print nothing if the buffers are empty */
1947 if (trace_empty(iter
))
1949 print_trace_header(m
, iter
);
1950 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1951 print_lat_help_header(m
);
1953 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1954 print_func_help_header(m
);
1956 } else if (iter
->leftover
) {
1958 * If we filled the seq_file buffer earlier, we
1959 * want to just show it now.
1961 ret
= trace_print_seq(m
, &iter
->seq
);
1963 /* ret should this time be zero, but you never know */
1964 iter
->leftover
= ret
;
1967 print_trace_line(iter
);
1968 ret
= trace_print_seq(m
, &iter
->seq
);
1970 * If we overflow the seq_file buffer, then it will
1971 * ask us for this data again at start up.
1973 * ret is 0 if seq_file write succeeded.
1976 iter
->leftover
= ret
;
1982 static const struct seq_operations tracer_seq_ops
= {
1989 static struct trace_iterator
*
1990 __tracing_open(struct inode
*inode
, struct file
*file
)
1992 long cpu_file
= (long) inode
->i_private
;
1993 void *fail_ret
= ERR_PTR(-ENOMEM
);
1994 struct trace_iterator
*iter
;
1998 if (tracing_disabled
)
1999 return ERR_PTR(-ENODEV
);
2001 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2003 return ERR_PTR(-ENOMEM
);
2006 * We make a copy of the current tracer to avoid concurrent
2007 * changes on it while we are reading.
2009 mutex_lock(&trace_types_lock
);
2010 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2015 *iter
->trace
= *current_trace
;
2017 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2020 if (current_trace
&& current_trace
->print_max
)
2023 iter
->tr
= &global_trace
;
2025 mutex_init(&iter
->mutex
);
2026 iter
->cpu_file
= cpu_file
;
2028 /* Notify the tracer early; before we stop tracing. */
2029 if (iter
->trace
&& iter
->trace
->open
)
2030 iter
->trace
->open(iter
);
2032 /* Annotate start of buffers if we had overruns */
2033 if (ring_buffer_overruns(iter
->tr
->buffer
))
2034 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2036 /* stop the trace while dumping */
2039 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
2040 for_each_tracing_cpu(cpu
) {
2042 iter
->buffer_iter
[cpu
] =
2043 ring_buffer_read_start(iter
->tr
->buffer
, cpu
);
2044 tracing_iter_reset(iter
, cpu
);
2047 cpu
= iter
->cpu_file
;
2048 iter
->buffer_iter
[cpu
] =
2049 ring_buffer_read_start(iter
->tr
->buffer
, cpu
);
2050 tracing_iter_reset(iter
, cpu
);
2053 ret
= seq_open(file
, &tracer_seq_ops
);
2055 fail_ret
= ERR_PTR(ret
);
2059 m
= file
->private_data
;
2062 mutex_unlock(&trace_types_lock
);
2067 for_each_tracing_cpu(cpu
) {
2068 if (iter
->buffer_iter
[cpu
])
2069 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2071 free_cpumask_var(iter
->started
);
2074 mutex_unlock(&trace_types_lock
);
2081 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2083 if (tracing_disabled
)
2086 filp
->private_data
= inode
->i_private
;
2090 static int tracing_release(struct inode
*inode
, struct file
*file
)
2092 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
2093 struct trace_iterator
*iter
;
2096 if (!(file
->f_mode
& FMODE_READ
))
2101 mutex_lock(&trace_types_lock
);
2102 for_each_tracing_cpu(cpu
) {
2103 if (iter
->buffer_iter
[cpu
])
2104 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2107 if (iter
->trace
&& iter
->trace
->close
)
2108 iter
->trace
->close(iter
);
2110 /* reenable tracing if it was previously enabled */
2112 mutex_unlock(&trace_types_lock
);
2114 seq_release(inode
, file
);
2115 mutex_destroy(&iter
->mutex
);
2116 free_cpumask_var(iter
->started
);
2122 static int tracing_open(struct inode
*inode
, struct file
*file
)
2124 struct trace_iterator
*iter
;
2127 /* If this file was open for write, then erase contents */
2128 if ((file
->f_mode
& FMODE_WRITE
) &&
2129 (file
->f_flags
& O_TRUNC
)) {
2130 long cpu
= (long) inode
->i_private
;
2132 if (cpu
== TRACE_PIPE_ALL_CPU
)
2133 tracing_reset_online_cpus(&global_trace
);
2135 tracing_reset(&global_trace
, cpu
);
2138 if (file
->f_mode
& FMODE_READ
) {
2139 iter
= __tracing_open(inode
, file
);
2141 ret
= PTR_ERR(iter
);
2142 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2143 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2149 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2151 struct tracer
*t
= v
;
2161 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2166 mutex_lock(&trace_types_lock
);
2167 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2173 static void t_stop(struct seq_file
*m
, void *p
)
2175 mutex_unlock(&trace_types_lock
);
2178 static int t_show(struct seq_file
*m
, void *v
)
2180 struct tracer
*t
= v
;
2185 seq_printf(m
, "%s", t
->name
);
2194 static const struct seq_operations show_traces_seq_ops
= {
2201 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2203 if (tracing_disabled
)
2206 return seq_open(file
, &show_traces_seq_ops
);
2210 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2211 size_t count
, loff_t
*ppos
)
2216 static const struct file_operations tracing_fops
= {
2217 .open
= tracing_open
,
2219 .write
= tracing_write_stub
,
2220 .llseek
= seq_lseek
,
2221 .release
= tracing_release
,
2224 static const struct file_operations show_traces_fops
= {
2225 .open
= show_traces_open
,
2227 .release
= seq_release
,
2231 * Only trace on a CPU if the bitmask is set:
2233 static cpumask_var_t tracing_cpumask
;
2236 * The tracer itself will not take this lock, but still we want
2237 * to provide a consistent cpumask to user-space:
2239 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2242 * Temporary storage for the character representation of the
2243 * CPU bitmask (and one more byte for the newline):
2245 static char mask_str
[NR_CPUS
+ 1];
2248 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2249 size_t count
, loff_t
*ppos
)
2253 mutex_lock(&tracing_cpumask_update_lock
);
2255 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2256 if (count
- len
< 2) {
2260 len
+= sprintf(mask_str
+ len
, "\n");
2261 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2264 mutex_unlock(&tracing_cpumask_update_lock
);
2270 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2271 size_t count
, loff_t
*ppos
)
2274 cpumask_var_t tracing_cpumask_new
;
2276 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2279 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2283 mutex_lock(&tracing_cpumask_update_lock
);
2285 local_irq_disable();
2286 __raw_spin_lock(&ftrace_max_lock
);
2287 for_each_tracing_cpu(cpu
) {
2289 * Increase/decrease the disabled counter if we are
2290 * about to flip a bit in the cpumask:
2292 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2293 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2294 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2296 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2297 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2298 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2301 __raw_spin_unlock(&ftrace_max_lock
);
2304 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2306 mutex_unlock(&tracing_cpumask_update_lock
);
2307 free_cpumask_var(tracing_cpumask_new
);
2312 free_cpumask_var(tracing_cpumask_new
);
2317 static const struct file_operations tracing_cpumask_fops
= {
2318 .open
= tracing_open_generic
,
2319 .read
= tracing_cpumask_read
,
2320 .write
= tracing_cpumask_write
,
2324 tracing_trace_options_read(struct file
*filp
, char __user
*ubuf
,
2325 size_t cnt
, loff_t
*ppos
)
2327 struct tracer_opt
*trace_opts
;
2335 /* calculate max size */
2336 for (i
= 0; trace_options
[i
]; i
++) {
2337 len
+= strlen(trace_options
[i
]);
2338 len
+= 3; /* "no" and newline */
2341 mutex_lock(&trace_types_lock
);
2342 tracer_flags
= current_trace
->flags
->val
;
2343 trace_opts
= current_trace
->flags
->opts
;
2346 * Increase the size with names of options specific
2347 * of the current tracer.
2349 for (i
= 0; trace_opts
[i
].name
; i
++) {
2350 len
+= strlen(trace_opts
[i
].name
);
2351 len
+= 3; /* "no" and newline */
2355 buf
= kmalloc(len
+ 1, GFP_KERNEL
);
2357 mutex_unlock(&trace_types_lock
);
2361 for (i
= 0; trace_options
[i
]; i
++) {
2362 if (trace_flags
& (1 << i
))
2363 r
+= sprintf(buf
+ r
, "%s\n", trace_options
[i
]);
2365 r
+= sprintf(buf
+ r
, "no%s\n", trace_options
[i
]);
2368 for (i
= 0; trace_opts
[i
].name
; i
++) {
2369 if (tracer_flags
& trace_opts
[i
].bit
)
2370 r
+= sprintf(buf
+ r
, "%s\n",
2371 trace_opts
[i
].name
);
2373 r
+= sprintf(buf
+ r
, "no%s\n",
2374 trace_opts
[i
].name
);
2376 mutex_unlock(&trace_types_lock
);
2378 WARN_ON(r
>= len
+ 1);
2380 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2386 /* Try to assign a tracer specific option */
2387 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2389 struct tracer_flags
*tracer_flags
= trace
->flags
;
2390 struct tracer_opt
*opts
= NULL
;
2394 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
2395 opts
= &tracer_flags
->opts
[i
];
2396 len
= strlen(opts
->name
);
2398 if (strncmp(cmp
, opts
->name
, len
) == 0) {
2399 ret
= trace
->set_flag(tracer_flags
->val
,
2405 if (!tracer_flags
->opts
[i
].name
)
2408 /* Refused to handle */
2413 tracer_flags
->val
&= ~opts
->bit
;
2415 tracer_flags
->val
|= opts
->bit
;
2420 static void set_tracer_flags(unsigned int mask
, int enabled
)
2422 /* do nothing if flag is already set */
2423 if (!!(trace_flags
& mask
) == !!enabled
)
2427 trace_flags
|= mask
;
2429 trace_flags
&= ~mask
;
2433 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2434 size_t cnt
, loff_t
*ppos
)
2442 if (cnt
>= sizeof(buf
))
2445 if (copy_from_user(&buf
, ubuf
, cnt
))
2450 if (strncmp(buf
, "no", 2) == 0) {
2455 for (i
= 0; trace_options
[i
]; i
++) {
2456 int len
= strlen(trace_options
[i
]);
2458 if (strncmp(cmp
, trace_options
[i
], len
) == 0) {
2459 set_tracer_flags(1 << i
, !neg
);
2464 /* If no option could be set, test the specific tracer options */
2465 if (!trace_options
[i
]) {
2466 mutex_lock(&trace_types_lock
);
2467 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2468 mutex_unlock(&trace_types_lock
);
2478 static const struct file_operations tracing_iter_fops
= {
2479 .open
= tracing_open_generic
,
2480 .read
= tracing_trace_options_read
,
2481 .write
= tracing_trace_options_write
,
2484 static const char readme_msg
[] =
2485 "tracing mini-HOWTO:\n\n"
2486 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2487 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2488 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2489 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2491 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2492 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2494 "# cat /sys/kernel/debug/tracing/trace_options\n"
2495 "noprint-parent nosym-offset nosym-addr noverbose\n"
2496 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2497 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2498 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2499 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2503 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2504 size_t cnt
, loff_t
*ppos
)
2506 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2507 readme_msg
, strlen(readme_msg
));
2510 static const struct file_operations tracing_readme_fops
= {
2511 .open
= tracing_open_generic
,
2512 .read
= tracing_readme_read
,
2516 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
2517 size_t cnt
, loff_t
*ppos
)
2526 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
2530 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
2538 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
2541 pid
= map_cmdline_to_pid
[i
];
2542 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
2545 trace_find_cmdline(pid
, buf_comm
);
2546 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
2551 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
2560 static const struct file_operations tracing_saved_cmdlines_fops
= {
2561 .open
= tracing_open_generic
,
2562 .read
= tracing_saved_cmdlines_read
,
2566 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2567 size_t cnt
, loff_t
*ppos
)
2572 r
= sprintf(buf
, "%u\n", tracer_enabled
);
2573 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2577 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2578 size_t cnt
, loff_t
*ppos
)
2580 struct trace_array
*tr
= filp
->private_data
;
2585 if (cnt
>= sizeof(buf
))
2588 if (copy_from_user(&buf
, ubuf
, cnt
))
2593 ret
= strict_strtoul(buf
, 10, &val
);
2599 mutex_lock(&trace_types_lock
);
2600 if (tracer_enabled
^ val
) {
2603 if (current_trace
->start
)
2604 current_trace
->start(tr
);
2609 if (current_trace
->stop
)
2610 current_trace
->stop(tr
);
2613 mutex_unlock(&trace_types_lock
);
2621 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2622 size_t cnt
, loff_t
*ppos
)
2624 char buf
[MAX_TRACER_SIZE
+2];
2627 mutex_lock(&trace_types_lock
);
2629 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2631 r
= sprintf(buf
, "\n");
2632 mutex_unlock(&trace_types_lock
);
2634 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2637 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
2639 tracing_reset_online_cpus(tr
);
2643 static int tracing_resize_ring_buffer(unsigned long size
)
2648 * If kernel or user changes the size of the ring buffer
2649 * we use the size that was given, and we can forget about
2650 * expanding it later.
2652 ring_buffer_expanded
= 1;
2654 ret
= ring_buffer_resize(global_trace
.buffer
, size
);
2658 ret
= ring_buffer_resize(max_tr
.buffer
, size
);
2662 r
= ring_buffer_resize(global_trace
.buffer
,
2663 global_trace
.entries
);
2666 * AARGH! We are left with different
2667 * size max buffer!!!!
2668 * The max buffer is our "snapshot" buffer.
2669 * When a tracer needs a snapshot (one of the
2670 * latency tracers), it swaps the max buffer
2671 * with the saved snap shot. We succeeded to
2672 * update the size of the main buffer, but failed to
2673 * update the size of the max buffer. But when we tried
2674 * to reset the main buffer to the original size, we
2675 * failed there too. This is very unlikely to
2676 * happen, but if it does, warn and kill all
2680 tracing_disabled
= 1;
2685 global_trace
.entries
= size
;
2691 * tracing_update_buffers - used by tracing facility to expand ring buffers
2693 * To save on memory when the tracing is never used on a system with it
2694 * configured in. The ring buffers are set to a minimum size. But once
2695 * a user starts to use the tracing facility, then they need to grow
2696 * to their default size.
2698 * This function is to be called when a tracer is about to be used.
2700 int tracing_update_buffers(void)
2704 mutex_lock(&trace_types_lock
);
2705 if (!ring_buffer_expanded
)
2706 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2707 mutex_unlock(&trace_types_lock
);
2712 struct trace_option_dentry
;
2714 static struct trace_option_dentry
*
2715 create_trace_option_files(struct tracer
*tracer
);
2718 destroy_trace_option_files(struct trace_option_dentry
*topts
);
2720 static int tracing_set_tracer(const char *buf
)
2722 static struct trace_option_dentry
*topts
;
2723 struct trace_array
*tr
= &global_trace
;
2727 mutex_lock(&trace_types_lock
);
2729 if (!ring_buffer_expanded
) {
2730 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2736 for (t
= trace_types
; t
; t
= t
->next
) {
2737 if (strcmp(t
->name
, buf
) == 0)
2744 if (t
== current_trace
)
2747 trace_branch_disable();
2748 if (current_trace
&& current_trace
->reset
)
2749 current_trace
->reset(tr
);
2751 destroy_trace_option_files(topts
);
2755 topts
= create_trace_option_files(current_trace
);
2758 ret
= tracer_init(t
, tr
);
2763 trace_branch_enable(tr
);
2765 mutex_unlock(&trace_types_lock
);
2771 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2772 size_t cnt
, loff_t
*ppos
)
2774 char buf
[MAX_TRACER_SIZE
+1];
2781 if (cnt
> MAX_TRACER_SIZE
)
2782 cnt
= MAX_TRACER_SIZE
;
2784 if (copy_from_user(&buf
, ubuf
, cnt
))
2789 /* strip ending whitespace. */
2790 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2793 err
= tracing_set_tracer(buf
);
2803 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2804 size_t cnt
, loff_t
*ppos
)
2806 unsigned long *ptr
= filp
->private_data
;
2810 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
2811 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2812 if (r
> sizeof(buf
))
2814 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2818 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2819 size_t cnt
, loff_t
*ppos
)
2821 unsigned long *ptr
= filp
->private_data
;
2826 if (cnt
>= sizeof(buf
))
2829 if (copy_from_user(&buf
, ubuf
, cnt
))
2834 ret
= strict_strtoul(buf
, 10, &val
);
2843 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2845 long cpu_file
= (long) inode
->i_private
;
2846 struct trace_iterator
*iter
;
2849 if (tracing_disabled
)
2852 mutex_lock(&trace_types_lock
);
2854 /* We only allow one reader per cpu */
2855 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
2856 if (!cpumask_empty(tracing_reader_cpumask
)) {
2860 cpumask_setall(tracing_reader_cpumask
);
2862 if (!cpumask_test_cpu(cpu_file
, tracing_reader_cpumask
))
2863 cpumask_set_cpu(cpu_file
, tracing_reader_cpumask
);
2870 /* create a buffer to store the information to pass to userspace */
2871 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2878 * We make a copy of the current tracer to avoid concurrent
2879 * changes on it while we are reading.
2881 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2887 *iter
->trace
= *current_trace
;
2889 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
2894 /* trace pipe does not show start of buffer */
2895 cpumask_setall(iter
->started
);
2897 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2898 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2900 iter
->cpu_file
= cpu_file
;
2901 iter
->tr
= &global_trace
;
2902 mutex_init(&iter
->mutex
);
2903 filp
->private_data
= iter
;
2905 if (iter
->trace
->pipe_open
)
2906 iter
->trace
->pipe_open(iter
);
2909 mutex_unlock(&trace_types_lock
);
2915 mutex_unlock(&trace_types_lock
);
2919 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
2921 struct trace_iterator
*iter
= file
->private_data
;
2923 mutex_lock(&trace_types_lock
);
2925 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
)
2926 cpumask_clear(tracing_reader_cpumask
);
2928 cpumask_clear_cpu(iter
->cpu_file
, tracing_reader_cpumask
);
2931 if (iter
->trace
->pipe_close
)
2932 iter
->trace
->pipe_close(iter
);
2934 mutex_unlock(&trace_types_lock
);
2936 free_cpumask_var(iter
->started
);
2937 mutex_destroy(&iter
->mutex
);
2945 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
2947 struct trace_iterator
*iter
= filp
->private_data
;
2949 if (trace_flags
& TRACE_ITER_BLOCK
) {
2951 * Always select as readable when in blocking mode
2953 return POLLIN
| POLLRDNORM
;
2955 if (!trace_empty(iter
))
2956 return POLLIN
| POLLRDNORM
;
2957 poll_wait(filp
, &trace_wait
, poll_table
);
2958 if (!trace_empty(iter
))
2959 return POLLIN
| POLLRDNORM
;
2966 void default_wait_pipe(struct trace_iterator
*iter
)
2970 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
2972 if (trace_empty(iter
))
2975 finish_wait(&trace_wait
, &wait
);
2979 * This is a make-shift waitqueue.
2980 * A tracer might use this callback on some rare cases:
2982 * 1) the current tracer might hold the runqueue lock when it wakes up
2983 * a reader, hence a deadlock (sched, function, and function graph tracers)
2984 * 2) the function tracers, trace all functions, we don't want
2985 * the overhead of calling wake_up and friends
2986 * (and tracing them too)
2988 * Anyway, this is really very primitive wakeup.
2990 void poll_wait_pipe(struct trace_iterator
*iter
)
2992 set_current_state(TASK_INTERRUPTIBLE
);
2993 /* sleep for 100 msecs, and try again. */
2994 schedule_timeout(HZ
/ 10);
2997 /* Must be called with trace_types_lock mutex held. */
2998 static int tracing_wait_pipe(struct file
*filp
)
3000 struct trace_iterator
*iter
= filp
->private_data
;
3002 while (trace_empty(iter
)) {
3004 if ((filp
->f_flags
& O_NONBLOCK
)) {
3008 mutex_unlock(&iter
->mutex
);
3010 iter
->trace
->wait_pipe(iter
);
3012 mutex_lock(&iter
->mutex
);
3014 if (signal_pending(current
))
3018 * We block until we read something and tracing is disabled.
3019 * We still block if tracing is disabled, but we have never
3020 * read anything. This allows a user to cat this file, and
3021 * then enable tracing. But after we have read something,
3022 * we give an EOF when tracing is again disabled.
3024 * iter->pos will be 0 if we haven't read anything.
3026 if (!tracer_enabled
&& iter
->pos
)
3037 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
3038 size_t cnt
, loff_t
*ppos
)
3040 struct trace_iterator
*iter
= filp
->private_data
;
3041 static struct tracer
*old_tracer
;
3044 /* return any leftover data */
3045 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3049 trace_seq_init(&iter
->seq
);
3051 /* copy the tracer to avoid using a global lock all around */
3052 mutex_lock(&trace_types_lock
);
3053 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3054 old_tracer
= current_trace
;
3055 *iter
->trace
= *current_trace
;
3057 mutex_unlock(&trace_types_lock
);
3060 * Avoid more than one consumer on a single file descriptor
3061 * This is just a matter of traces coherency, the ring buffer itself
3064 mutex_lock(&iter
->mutex
);
3065 if (iter
->trace
->read
) {
3066 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
3072 sret
= tracing_wait_pipe(filp
);
3076 /* stop when tracing is finished */
3077 if (trace_empty(iter
)) {
3082 if (cnt
>= PAGE_SIZE
)
3083 cnt
= PAGE_SIZE
- 1;
3085 /* reset all but tr, trace, and overruns */
3086 memset(&iter
->seq
, 0,
3087 sizeof(struct trace_iterator
) -
3088 offsetof(struct trace_iterator
, seq
));
3091 trace_event_read_lock();
3092 while (find_next_entry_inc(iter
) != NULL
) {
3093 enum print_line_t ret
;
3094 int len
= iter
->seq
.len
;
3096 ret
= print_trace_line(iter
);
3097 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3098 /* don't print partial lines */
3099 iter
->seq
.len
= len
;
3102 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3103 trace_consume(iter
);
3105 if (iter
->seq
.len
>= cnt
)
3108 trace_event_read_unlock();
3110 /* Now copy what we have to the user */
3111 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3112 if (iter
->seq
.readpos
>= iter
->seq
.len
)
3113 trace_seq_init(&iter
->seq
);
3116 * If there was nothing to send to user, inspite of consuming trace
3117 * entries, go back to wait for more entries.
3123 mutex_unlock(&iter
->mutex
);
3128 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
3129 struct pipe_buffer
*buf
)
3131 __free_page(buf
->page
);
3134 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
3137 __free_page(spd
->pages
[idx
]);
3140 static struct pipe_buf_operations tracing_pipe_buf_ops
= {
3142 .map
= generic_pipe_buf_map
,
3143 .unmap
= generic_pipe_buf_unmap
,
3144 .confirm
= generic_pipe_buf_confirm
,
3145 .release
= tracing_pipe_buf_release
,
3146 .steal
= generic_pipe_buf_steal
,
3147 .get
= generic_pipe_buf_get
,
3151 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
3156 /* Seq buffer is page-sized, exactly what we need. */
3158 count
= iter
->seq
.len
;
3159 ret
= print_trace_line(iter
);
3160 count
= iter
->seq
.len
- count
;
3163 iter
->seq
.len
-= count
;
3166 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3167 iter
->seq
.len
-= count
;
3171 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3172 trace_consume(iter
);
3174 if (!find_next_entry_inc(iter
)) {
3184 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3186 struct pipe_inode_info
*pipe
,
3190 struct page
*pages
[PIPE_BUFFERS
];
3191 struct partial_page partial
[PIPE_BUFFERS
];
3192 struct trace_iterator
*iter
= filp
->private_data
;
3193 struct splice_pipe_desc spd
= {
3196 .nr_pages
= 0, /* This gets updated below. */
3198 .ops
= &tracing_pipe_buf_ops
,
3199 .spd_release
= tracing_spd_release_pipe
,
3201 static struct tracer
*old_tracer
;
3206 /* copy the tracer to avoid using a global lock all around */
3207 mutex_lock(&trace_types_lock
);
3208 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3209 old_tracer
= current_trace
;
3210 *iter
->trace
= *current_trace
;
3212 mutex_unlock(&trace_types_lock
);
3214 mutex_lock(&iter
->mutex
);
3216 if (iter
->trace
->splice_read
) {
3217 ret
= iter
->trace
->splice_read(iter
, filp
,
3218 ppos
, pipe
, len
, flags
);
3223 ret
= tracing_wait_pipe(filp
);
3227 if (!iter
->ent
&& !find_next_entry_inc(iter
)) {
3232 trace_event_read_lock();
3234 /* Fill as many pages as possible. */
3235 for (i
= 0, rem
= len
; i
< PIPE_BUFFERS
&& rem
; i
++) {
3236 pages
[i
] = alloc_page(GFP_KERNEL
);
3240 rem
= tracing_fill_pipe_page(rem
, iter
);
3242 /* Copy the data into the page, so we can start over. */
3243 ret
= trace_seq_to_buffer(&iter
->seq
,
3244 page_address(pages
[i
]),
3247 __free_page(pages
[i
]);
3250 partial
[i
].offset
= 0;
3251 partial
[i
].len
= iter
->seq
.len
;
3253 trace_seq_init(&iter
->seq
);
3256 trace_event_read_unlock();
3257 mutex_unlock(&iter
->mutex
);
3261 return splice_to_pipe(pipe
, &spd
);
3264 mutex_unlock(&iter
->mutex
);
3270 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3271 size_t cnt
, loff_t
*ppos
)
3273 struct trace_array
*tr
= filp
->private_data
;
3277 mutex_lock(&trace_types_lock
);
3278 if (!ring_buffer_expanded
)
3279 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3281 trace_buf_size
>> 10);
3283 r
= sprintf(buf
, "%lu\n", tr
->entries
>> 10);
3284 mutex_unlock(&trace_types_lock
);
3286 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3290 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3291 size_t cnt
, loff_t
*ppos
)
3297 if (cnt
>= sizeof(buf
))
3300 if (copy_from_user(&buf
, ubuf
, cnt
))
3305 ret
= strict_strtoul(buf
, 10, &val
);
3309 /* must have at least 1 entry */
3313 mutex_lock(&trace_types_lock
);
3317 /* disable all cpu buffers */
3318 for_each_tracing_cpu(cpu
) {
3319 if (global_trace
.data
[cpu
])
3320 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
3321 if (max_tr
.data
[cpu
])
3322 atomic_inc(&max_tr
.data
[cpu
]->disabled
);
3325 /* value is in KB */
3328 if (val
!= global_trace
.entries
) {
3329 ret
= tracing_resize_ring_buffer(val
);
3338 /* If check pages failed, return ENOMEM */
3339 if (tracing_disabled
)
3342 for_each_tracing_cpu(cpu
) {
3343 if (global_trace
.data
[cpu
])
3344 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
3345 if (max_tr
.data
[cpu
])
3346 atomic_dec(&max_tr
.data
[cpu
]->disabled
);
3350 max_tr
.entries
= global_trace
.entries
;
3351 mutex_unlock(&trace_types_lock
);
3357 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3358 size_t cnt
, loff_t
*fpos
)
3362 if (tracing_disabled
)
3365 if (cnt
> TRACE_BUF_SIZE
)
3366 cnt
= TRACE_BUF_SIZE
;
3368 buf
= kmalloc(cnt
+ 2, GFP_KERNEL
);
3372 if (copy_from_user(buf
, ubuf
, cnt
)) {
3376 if (buf
[cnt
-1] != '\n') {
3382 cnt
= trace_vprintk(0, buf
, NULL
);
3389 static ssize_t
tracing_clock_read(struct file
*filp
, char __user
*ubuf
,
3390 size_t cnt
, loff_t
*ppos
)
3396 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
3397 bufiter
+= snprintf(buf
+ bufiter
, sizeof(buf
) - bufiter
,
3398 "%s%s%s%s", i
? " " : "",
3399 i
== trace_clock_id
? "[" : "", trace_clocks
[i
].name
,
3400 i
== trace_clock_id
? "]" : "");
3401 bufiter
+= snprintf(buf
+ bufiter
, sizeof(buf
) - bufiter
, "\n");
3403 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, bufiter
);
3406 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
3407 size_t cnt
, loff_t
*fpos
)
3410 const char *clockstr
;
3413 if (cnt
>= sizeof(buf
))
3416 if (copy_from_user(&buf
, ubuf
, cnt
))
3421 clockstr
= strstrip(buf
);
3423 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
3424 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
3427 if (i
== ARRAY_SIZE(trace_clocks
))
3432 mutex_lock(&trace_types_lock
);
3434 ring_buffer_set_clock(global_trace
.buffer
, trace_clocks
[i
].func
);
3436 ring_buffer_set_clock(max_tr
.buffer
, trace_clocks
[i
].func
);
3438 mutex_unlock(&trace_types_lock
);
3445 static const struct file_operations tracing_max_lat_fops
= {
3446 .open
= tracing_open_generic
,
3447 .read
= tracing_max_lat_read
,
3448 .write
= tracing_max_lat_write
,
3451 static const struct file_operations tracing_ctrl_fops
= {
3452 .open
= tracing_open_generic
,
3453 .read
= tracing_ctrl_read
,
3454 .write
= tracing_ctrl_write
,
3457 static const struct file_operations set_tracer_fops
= {
3458 .open
= tracing_open_generic
,
3459 .read
= tracing_set_trace_read
,
3460 .write
= tracing_set_trace_write
,
3463 static const struct file_operations tracing_pipe_fops
= {
3464 .open
= tracing_open_pipe
,
3465 .poll
= tracing_poll_pipe
,
3466 .read
= tracing_read_pipe
,
3467 .splice_read
= tracing_splice_read_pipe
,
3468 .release
= tracing_release_pipe
,
3471 static const struct file_operations tracing_entries_fops
= {
3472 .open
= tracing_open_generic
,
3473 .read
= tracing_entries_read
,
3474 .write
= tracing_entries_write
,
3477 static const struct file_operations tracing_mark_fops
= {
3478 .open
= tracing_open_generic
,
3479 .write
= tracing_mark_write
,
3482 static const struct file_operations trace_clock_fops
= {
3483 .open
= tracing_open_generic
,
3484 .read
= tracing_clock_read
,
3485 .write
= tracing_clock_write
,
3488 struct ftrace_buffer_info
{
3489 struct trace_array
*tr
;
3495 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
3497 int cpu
= (int)(long)inode
->i_private
;
3498 struct ftrace_buffer_info
*info
;
3500 if (tracing_disabled
)
3503 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3507 info
->tr
= &global_trace
;
3510 /* Force reading ring buffer for first read */
3511 info
->read
= (unsigned int)-1;
3513 filp
->private_data
= info
;
3515 return nonseekable_open(inode
, filp
);
3519 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
3520 size_t count
, loff_t
*ppos
)
3522 struct ftrace_buffer_info
*info
= filp
->private_data
;
3531 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
);
3535 /* Do we have previous read data to read? */
3536 if (info
->read
< PAGE_SIZE
)
3541 ret
= ring_buffer_read_page(info
->tr
->buffer
,
3548 pos
= ring_buffer_page_len(info
->spare
);
3550 if (pos
< PAGE_SIZE
)
3551 memset(info
->spare
+ pos
, 0, PAGE_SIZE
- pos
);
3554 size
= PAGE_SIZE
- info
->read
;
3558 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
3569 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
3571 struct ftrace_buffer_info
*info
= file
->private_data
;
3574 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
3581 struct ring_buffer
*buffer
;
3586 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
3587 struct pipe_buffer
*buf
)
3589 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3594 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3599 static int buffer_pipe_buf_steal(struct pipe_inode_info
*pipe
,
3600 struct pipe_buffer
*buf
)
3605 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
3606 struct pipe_buffer
*buf
)
3608 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3613 /* Pipe buffer operations for a buffer. */
3614 static struct pipe_buf_operations buffer_pipe_buf_ops
= {
3616 .map
= generic_pipe_buf_map
,
3617 .unmap
= generic_pipe_buf_unmap
,
3618 .confirm
= generic_pipe_buf_confirm
,
3619 .release
= buffer_pipe_buf_release
,
3620 .steal
= buffer_pipe_buf_steal
,
3621 .get
= buffer_pipe_buf_get
,
3625 * Callback from splice_to_pipe(), if we need to release some pages
3626 * at the end of the spd in case we error'ed out in filling the pipe.
3628 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
3630 struct buffer_ref
*ref
=
3631 (struct buffer_ref
*)spd
->partial
[i
].private;
3636 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3638 spd
->partial
[i
].private = 0;
3642 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
3643 struct pipe_inode_info
*pipe
, size_t len
,
3646 struct ftrace_buffer_info
*info
= file
->private_data
;
3647 struct partial_page partial
[PIPE_BUFFERS
];
3648 struct page
*pages
[PIPE_BUFFERS
];
3649 struct splice_pipe_desc spd
= {
3653 .ops
= &buffer_pipe_buf_ops
,
3654 .spd_release
= buffer_spd_release
,
3656 struct buffer_ref
*ref
;
3657 int entries
, size
, i
;
3660 if (*ppos
& (PAGE_SIZE
- 1)) {
3661 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3665 if (len
& (PAGE_SIZE
- 1)) {
3666 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3667 if (len
< PAGE_SIZE
)
3672 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
3674 for (i
= 0; i
< PIPE_BUFFERS
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
3678 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
3683 ref
->buffer
= info
->tr
->buffer
;
3684 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
);
3690 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
3693 ring_buffer_free_read_page(ref
->buffer
,
3700 * zero out any left over data, this is going to
3703 size
= ring_buffer_page_len(ref
->page
);
3704 if (size
< PAGE_SIZE
)
3705 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
3707 page
= virt_to_page(ref
->page
);
3709 spd
.pages
[i
] = page
;
3710 spd
.partial
[i
].len
= PAGE_SIZE
;
3711 spd
.partial
[i
].offset
= 0;
3712 spd
.partial
[i
].private = (unsigned long)ref
;
3716 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
3721 /* did we read anything? */
3722 if (!spd
.nr_pages
) {
3723 if (flags
& SPLICE_F_NONBLOCK
)
3731 ret
= splice_to_pipe(pipe
, &spd
);
3736 static const struct file_operations tracing_buffers_fops
= {
3737 .open
= tracing_buffers_open
,
3738 .read
= tracing_buffers_read
,
3739 .release
= tracing_buffers_release
,
3740 .splice_read
= tracing_buffers_splice_read
,
3741 .llseek
= no_llseek
,
3745 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
3746 size_t count
, loff_t
*ppos
)
3748 unsigned long cpu
= (unsigned long)filp
->private_data
;
3749 struct trace_array
*tr
= &global_trace
;
3750 struct trace_seq
*s
;
3753 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
3759 cnt
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
3760 trace_seq_printf(s
, "entries: %ld\n", cnt
);
3762 cnt
= ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
3763 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
3765 cnt
= ring_buffer_commit_overrun_cpu(tr
->buffer
, cpu
);
3766 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
3768 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
3775 static const struct file_operations tracing_stats_fops
= {
3776 .open
= tracing_open_generic
,
3777 .read
= tracing_stats_read
,
3780 #ifdef CONFIG_DYNAMIC_FTRACE
3782 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
3788 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
3789 size_t cnt
, loff_t
*ppos
)
3791 static char ftrace_dyn_info_buffer
[1024];
3792 static DEFINE_MUTEX(dyn_info_mutex
);
3793 unsigned long *p
= filp
->private_data
;
3794 char *buf
= ftrace_dyn_info_buffer
;
3795 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
3798 mutex_lock(&dyn_info_mutex
);
3799 r
= sprintf(buf
, "%ld ", *p
);
3801 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
3804 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3806 mutex_unlock(&dyn_info_mutex
);
3811 static const struct file_operations tracing_dyn_info_fops
= {
3812 .open
= tracing_open_generic
,
3813 .read
= tracing_read_dyn_info
,
3817 static struct dentry
*d_tracer
;
3819 struct dentry
*tracing_init_dentry(void)
3826 if (!debugfs_initialized())
3829 d_tracer
= debugfs_create_dir("tracing", NULL
);
3831 if (!d_tracer
&& !once
) {
3833 pr_warning("Could not create debugfs directory 'tracing'\n");
3840 static struct dentry
*d_percpu
;
3842 struct dentry
*tracing_dentry_percpu(void)
3845 struct dentry
*d_tracer
;
3850 d_tracer
= tracing_init_dentry();
3855 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
3857 if (!d_percpu
&& !once
) {
3859 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3866 static void tracing_init_debugfs_percpu(long cpu
)
3868 struct dentry
*d_percpu
= tracing_dentry_percpu();
3869 struct dentry
*d_cpu
;
3870 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3873 if (cpu
> 999 || cpu
< 0)
3876 sprintf(cpu_dir
, "cpu%ld", cpu
);
3877 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
3879 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
3883 /* per cpu trace_pipe */
3884 trace_create_file("trace_pipe", 0444, d_cpu
,
3885 (void *) cpu
, &tracing_pipe_fops
);
3888 trace_create_file("trace", 0644, d_cpu
,
3889 (void *) cpu
, &tracing_fops
);
3891 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
3892 (void *) cpu
, &tracing_buffers_fops
);
3894 trace_create_file("stats", 0444, d_cpu
,
3895 (void *) cpu
, &tracing_stats_fops
);
3898 #ifdef CONFIG_FTRACE_SELFTEST
3899 /* Let selftest have access to static functions in this file */
3900 #include "trace_selftest.c"
3903 struct trace_option_dentry
{
3904 struct tracer_opt
*opt
;
3905 struct tracer_flags
*flags
;
3906 struct dentry
*entry
;
3910 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
3913 struct trace_option_dentry
*topt
= filp
->private_data
;
3916 if (topt
->flags
->val
& topt
->opt
->bit
)
3921 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
3925 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
3928 struct trace_option_dentry
*topt
= filp
->private_data
;
3933 if (cnt
>= sizeof(buf
))
3936 if (copy_from_user(&buf
, ubuf
, cnt
))
3941 ret
= strict_strtoul(buf
, 10, &val
);
3948 /* do nothing if already cleared */
3949 if (!(topt
->flags
->val
& topt
->opt
->bit
))
3952 mutex_lock(&trace_types_lock
);
3953 if (current_trace
->set_flag
)
3954 ret
= current_trace
->set_flag(topt
->flags
->val
,
3956 mutex_unlock(&trace_types_lock
);
3959 topt
->flags
->val
&= ~topt
->opt
->bit
;
3962 /* do nothing if already set */
3963 if (topt
->flags
->val
& topt
->opt
->bit
)
3966 mutex_lock(&trace_types_lock
);
3967 if (current_trace
->set_flag
)
3968 ret
= current_trace
->set_flag(topt
->flags
->val
,
3970 mutex_unlock(&trace_types_lock
);
3973 topt
->flags
->val
|= topt
->opt
->bit
;
3986 static const struct file_operations trace_options_fops
= {
3987 .open
= tracing_open_generic
,
3988 .read
= trace_options_read
,
3989 .write
= trace_options_write
,
3993 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
3996 long index
= (long)filp
->private_data
;
3999 if (trace_flags
& (1 << index
))
4004 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4008 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4011 long index
= (long)filp
->private_data
;
4016 if (cnt
>= sizeof(buf
))
4019 if (copy_from_user(&buf
, ubuf
, cnt
))
4024 ret
= strict_strtoul(buf
, 10, &val
);
4028 if (val
!= 0 && val
!= 1)
4030 set_tracer_flags(1 << index
, val
);
4037 static const struct file_operations trace_options_core_fops
= {
4038 .open
= tracing_open_generic
,
4039 .read
= trace_options_core_read
,
4040 .write
= trace_options_core_write
,
4043 struct dentry
*trace_create_file(const char *name
,
4045 struct dentry
*parent
,
4047 const struct file_operations
*fops
)
4051 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
4053 pr_warning("Could not create debugfs '%s' entry\n", name
);
4059 static struct dentry
*trace_options_init_dentry(void)
4061 struct dentry
*d_tracer
;
4062 static struct dentry
*t_options
;
4067 d_tracer
= tracing_init_dentry();
4071 t_options
= debugfs_create_dir("options", d_tracer
);
4073 pr_warning("Could not create debugfs directory 'options'\n");
4081 create_trace_option_file(struct trace_option_dentry
*topt
,
4082 struct tracer_flags
*flags
,
4083 struct tracer_opt
*opt
)
4085 struct dentry
*t_options
;
4087 t_options
= trace_options_init_dentry();
4091 topt
->flags
= flags
;
4094 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
4095 &trace_options_fops
);
4099 static struct trace_option_dentry
*
4100 create_trace_option_files(struct tracer
*tracer
)
4102 struct trace_option_dentry
*topts
;
4103 struct tracer_flags
*flags
;
4104 struct tracer_opt
*opts
;
4110 flags
= tracer
->flags
;
4112 if (!flags
|| !flags
->opts
)
4117 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4120 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
4124 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4125 create_trace_option_file(&topts
[cnt
], flags
,
4132 destroy_trace_option_files(struct trace_option_dentry
*topts
)
4139 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
4140 if (topts
[cnt
].entry
)
4141 debugfs_remove(topts
[cnt
].entry
);
4147 static struct dentry
*
4148 create_trace_option_core_file(const char *option
, long index
)
4150 struct dentry
*t_options
;
4152 t_options
= trace_options_init_dentry();
4156 return trace_create_file(option
, 0644, t_options
, (void *)index
,
4157 &trace_options_core_fops
);
4160 static __init
void create_trace_options_dir(void)
4162 struct dentry
*t_options
;
4165 t_options
= trace_options_init_dentry();
4169 for (i
= 0; trace_options
[i
]; i
++)
4170 create_trace_option_core_file(trace_options
[i
], i
);
4173 static __init
int tracer_init_debugfs(void)
4175 struct dentry
*d_tracer
;
4178 d_tracer
= tracing_init_dentry();
4180 trace_create_file("tracing_enabled", 0644, d_tracer
,
4181 &global_trace
, &tracing_ctrl_fops
);
4183 trace_create_file("trace_options", 0644, d_tracer
,
4184 NULL
, &tracing_iter_fops
);
4186 trace_create_file("tracing_cpumask", 0644, d_tracer
,
4187 NULL
, &tracing_cpumask_fops
);
4189 trace_create_file("trace", 0644, d_tracer
,
4190 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
4192 trace_create_file("available_tracers", 0444, d_tracer
,
4193 &global_trace
, &show_traces_fops
);
4195 trace_create_file("current_tracer", 0644, d_tracer
,
4196 &global_trace
, &set_tracer_fops
);
4198 #ifdef CONFIG_TRACER_MAX_TRACE
4199 trace_create_file("tracing_max_latency", 0644, d_tracer
,
4200 &tracing_max_latency
, &tracing_max_lat_fops
);
4202 trace_create_file("tracing_thresh", 0644, d_tracer
,
4203 &tracing_thresh
, &tracing_max_lat_fops
);
4206 trace_create_file("README", 0444, d_tracer
,
4207 NULL
, &tracing_readme_fops
);
4209 trace_create_file("trace_pipe", 0444, d_tracer
,
4210 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
4212 trace_create_file("buffer_size_kb", 0644, d_tracer
,
4213 &global_trace
, &tracing_entries_fops
);
4215 trace_create_file("trace_marker", 0220, d_tracer
,
4216 NULL
, &tracing_mark_fops
);
4218 trace_create_file("saved_cmdlines", 0444, d_tracer
,
4219 NULL
, &tracing_saved_cmdlines_fops
);
4221 trace_create_file("trace_clock", 0644, d_tracer
, NULL
,
4224 #ifdef CONFIG_DYNAMIC_FTRACE
4225 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
4226 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
4228 #ifdef CONFIG_SYSPROF_TRACER
4229 init_tracer_sysprof_debugfs(d_tracer
);
4232 create_trace_options_dir();
4234 for_each_tracing_cpu(cpu
)
4235 tracing_init_debugfs_percpu(cpu
);
4240 static int trace_panic_handler(struct notifier_block
*this,
4241 unsigned long event
, void *unused
)
4243 if (ftrace_dump_on_oops
)
4248 static struct notifier_block trace_panic_notifier
= {
4249 .notifier_call
= trace_panic_handler
,
4251 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
4254 static int trace_die_handler(struct notifier_block
*self
,
4260 if (ftrace_dump_on_oops
)
4269 static struct notifier_block trace_die_notifier
= {
4270 .notifier_call
= trace_die_handler
,
4275 * printk is set to max of 1024, we really don't need it that big.
4276 * Nothing should be printing 1000 characters anyway.
4278 #define TRACE_MAX_PRINT 1000
4281 * Define here KERN_TRACE so that we have one place to modify
4282 * it if we decide to change what log level the ftrace dump
4285 #define KERN_TRACE KERN_EMERG
4288 trace_printk_seq(struct trace_seq
*s
)
4290 /* Probably should print a warning here. */
4294 /* should be zero ended, but we are paranoid. */
4295 s
->buffer
[s
->len
] = 0;
4297 printk(KERN_TRACE
"%s", s
->buffer
);
4302 static void __ftrace_dump(bool disable_tracing
)
4304 static raw_spinlock_t ftrace_dump_lock
=
4305 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
4306 /* use static because iter can be a bit big for the stack */
4307 static struct trace_iterator iter
;
4308 unsigned int old_userobj
;
4309 static int dump_ran
;
4310 unsigned long flags
;
4314 local_irq_save(flags
);
4315 __raw_spin_lock(&ftrace_dump_lock
);
4323 if (disable_tracing
)
4326 for_each_tracing_cpu(cpu
) {
4327 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
4330 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
4332 /* don't look at user memory in panic mode */
4333 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
4335 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
4337 /* Simulate the iterator */
4338 iter
.tr
= &global_trace
;
4339 iter
.trace
= current_trace
;
4340 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4343 * We need to stop all tracing on all CPUS to read the
4344 * the next buffer. This is a bit expensive, but is
4345 * not done often. We fill all what we can read,
4346 * and then release the locks again.
4349 while (!trace_empty(&iter
)) {
4352 printk(KERN_TRACE
"---------------------------------\n");
4356 /* reset all but tr, trace, and overruns */
4357 memset(&iter
.seq
, 0,
4358 sizeof(struct trace_iterator
) -
4359 offsetof(struct trace_iterator
, seq
));
4360 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
4363 if (find_next_entry_inc(&iter
) != NULL
) {
4366 ret
= print_trace_line(&iter
);
4367 if (ret
!= TRACE_TYPE_NO_CONSUME
)
4368 trace_consume(&iter
);
4371 trace_printk_seq(&iter
.seq
);
4375 printk(KERN_TRACE
" (ftrace buffer empty)\n");
4377 printk(KERN_TRACE
"---------------------------------\n");
4379 /* Re-enable tracing if requested */
4380 if (!disable_tracing
) {
4381 trace_flags
|= old_userobj
;
4383 for_each_tracing_cpu(cpu
) {
4384 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
4390 __raw_spin_unlock(&ftrace_dump_lock
);
4391 local_irq_restore(flags
);
4394 /* By default: disable tracing after the dump */
4395 void ftrace_dump(void)
4397 __ftrace_dump(true);
4400 __init
static int tracer_alloc_buffers(void)
4406 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
4409 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
4410 goto out_free_buffer_mask
;
4412 if (!zalloc_cpumask_var(&tracing_reader_cpumask
, GFP_KERNEL
))
4413 goto out_free_tracing_cpumask
;
4415 /* To save memory, keep the ring buffer size to its minimum */
4416 if (ring_buffer_expanded
)
4417 ring_buf_size
= trace_buf_size
;
4421 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
4422 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
4424 /* TODO: make the number of buffers hot pluggable with CPUS */
4425 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
,
4426 TRACE_BUFFER_FLAGS
);
4427 if (!global_trace
.buffer
) {
4428 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
4430 goto out_free_cpumask
;
4432 global_trace
.entries
= ring_buffer_size(global_trace
.buffer
);
4435 #ifdef CONFIG_TRACER_MAX_TRACE
4436 max_tr
.buffer
= ring_buffer_alloc(ring_buf_size
,
4437 TRACE_BUFFER_FLAGS
);
4438 if (!max_tr
.buffer
) {
4439 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
4441 ring_buffer_free(global_trace
.buffer
);
4442 goto out_free_cpumask
;
4444 max_tr
.entries
= ring_buffer_size(max_tr
.buffer
);
4445 WARN_ON(max_tr
.entries
!= global_trace
.entries
);
4448 /* Allocate the first page for all buffers */
4449 for_each_tracing_cpu(i
) {
4450 global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
4451 max_tr
.data
[i
] = &per_cpu(max_data
, i
);
4454 trace_init_cmdlines();
4456 register_tracer(&nop_trace
);
4457 current_trace
= &nop_trace
;
4458 #ifdef CONFIG_BOOT_TRACER
4459 register_tracer(&boot_tracer
);
4461 /* All seems OK, enable tracing */
4462 tracing_disabled
= 0;
4464 atomic_notifier_chain_register(&panic_notifier_list
,
4465 &trace_panic_notifier
);
4467 register_die_notifier(&trace_die_notifier
);
4472 free_cpumask_var(tracing_reader_cpumask
);
4473 out_free_tracing_cpumask
:
4474 free_cpumask_var(tracing_cpumask
);
4475 out_free_buffer_mask
:
4476 free_cpumask_var(tracing_buffer_mask
);
4481 __init
static int clear_boot_tracer(void)
4484 * The default tracer at boot buffer is an init section.
4485 * This function is called in lateinit. If we did not
4486 * find the boot tracer, then clear it out, to prevent
4487 * later registration from accessing the buffer that is
4488 * about to be freed.
4490 if (!default_bootup_tracer
)
4493 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
4494 default_bootup_tracer
);
4495 default_bootup_tracer
= NULL
;
4500 early_initcall(tracer_alloc_buffers
);
4501 fs_initcall(tracer_init_debugfs
);
4502 late_initcall(clear_boot_tracer
);