2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 William Lee Irwin III
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/rwsem.h>
36 #include <linux/slab.h>
37 #include <linux/ctype.h>
38 #include <linux/init.h>
39 #include <linux/poll.h>
43 #include "trace_output.h"
45 #define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
48 * On boot up, the ring buffer is set to the minimum size, so that
49 * we do not waste memory on systems that are not using tracing.
51 int ring_buffer_expanded
;
54 * We need to change this state when a selftest is running.
55 * A selftest will lurk into the ring-buffer to count the
56 * entries inserted during the selftest although some concurrent
57 * insertions into the ring-buffer such as trace_printk could occurred
58 * at the same time, giving false positive or negative results.
60 static bool __read_mostly tracing_selftest_running
;
63 * If a tracer is running, we do not want to run SELFTEST.
65 bool __read_mostly tracing_selftest_disabled
;
67 /* For tracers that don't implement custom flags */
68 static struct tracer_opt dummy_tracer_opt
[] = {
72 static struct tracer_flags dummy_tracer_flags
= {
74 .opts
= dummy_tracer_opt
77 static int dummy_set_flag(u32 old_flags
, u32 bit
, int set
)
83 * Kill all tracing for good (never come back).
84 * It is initialized to 1 but will turn to zero if the initialization
85 * of the tracer is successful. But that is the only place that sets
88 static int tracing_disabled
= 1;
90 DEFINE_PER_CPU(int, ftrace_cpu_disabled
);
92 static inline void ftrace_disable_cpu(void)
95 __this_cpu_inc(ftrace_cpu_disabled
);
98 static inline void ftrace_enable_cpu(void)
100 __this_cpu_dec(ftrace_cpu_disabled
);
104 static cpumask_var_t __read_mostly tracing_buffer_mask
;
106 #define for_each_tracing_cpu(cpu) \
107 for_each_cpu(cpu, tracing_buffer_mask)
110 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
112 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
113 * is set, then ftrace_dump is called. This will output the contents
114 * of the ftrace buffers to the console. This is very useful for
115 * capturing traces that lead to crashes and outputing it to a
118 * It is default off, but you can enable it with either specifying
119 * "ftrace_dump_on_oops" in the kernel command line, or setting
120 * /proc/sys/kernel/ftrace_dump_on_oops
121 * Set 1 if you want to dump buffers of all CPUs
122 * Set 2 if you want to dump the buffer of the CPU that triggered oops
125 enum ftrace_dump_mode ftrace_dump_on_oops
;
127 static int tracing_set_tracer(const char *buf
);
129 #define MAX_TRACER_SIZE 100
130 static char bootup_tracer_buf
[MAX_TRACER_SIZE
] __initdata
;
131 static char *default_bootup_tracer
;
133 static int __init
set_cmdline_ftrace(char *str
)
135 strncpy(bootup_tracer_buf
, str
, MAX_TRACER_SIZE
);
136 default_bootup_tracer
= bootup_tracer_buf
;
137 /* We are using ftrace early, expand it */
138 ring_buffer_expanded
= 1;
141 __setup("ftrace=", set_cmdline_ftrace
);
143 static int __init
set_ftrace_dump_on_oops(char *str
)
145 if (*str
++ != '=' || !*str
) {
146 ftrace_dump_on_oops
= DUMP_ALL
;
150 if (!strcmp("orig_cpu", str
)) {
151 ftrace_dump_on_oops
= DUMP_ORIG
;
157 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops
);
159 unsigned long long ns2usecs(cycle_t nsec
)
167 * The global_trace is the descriptor that holds the tracing
168 * buffers for the live tracing. For each CPU, it contains
169 * a link list of pages that will store trace entries. The
170 * page descriptor of the pages in the memory is used to hold
171 * the link list by linking the lru item in the page descriptor
172 * to each of the pages in the buffer per CPU.
174 * For each active CPU there is a data field that holds the
175 * pages for the buffer for that CPU. Each CPU has the same number
176 * of pages allocated for its buffer.
178 static struct trace_array global_trace
;
180 static DEFINE_PER_CPU(struct trace_array_cpu
, global_trace_cpu
);
182 int filter_current_check_discard(struct ring_buffer
*buffer
,
183 struct ftrace_event_call
*call
, void *rec
,
184 struct ring_buffer_event
*event
)
186 return filter_check_discard(call
, rec
, buffer
, event
);
188 EXPORT_SYMBOL_GPL(filter_current_check_discard
);
190 cycle_t
ftrace_now(int cpu
)
194 /* Early boot up does not have a buffer yet */
195 if (!global_trace
.buffer
)
196 return trace_clock_local();
198 ts
= ring_buffer_time_stamp(global_trace
.buffer
, cpu
);
199 ring_buffer_normalize_time_stamp(global_trace
.buffer
, cpu
, &ts
);
205 * The max_tr is used to snapshot the global_trace when a maximum
206 * latency is reached. Some tracers will use this to store a maximum
207 * trace while it continues examining live traces.
209 * The buffers for the max_tr are set up the same as the global_trace.
210 * When a snapshot is taken, the link list of the max_tr is swapped
211 * with the link list of the global_trace and the buffers are reset for
212 * the global_trace so the tracing can continue.
214 static struct trace_array max_tr
;
216 static DEFINE_PER_CPU(struct trace_array_cpu
, max_tr_data
);
218 /* tracer_enabled is used to toggle activation of a tracer */
219 static int tracer_enabled
= 1;
222 * tracing_is_enabled - return tracer_enabled status
224 * This function is used by other tracers to know the status
225 * of the tracer_enabled flag. Tracers may use this function
226 * to know if it should enable their features when starting
227 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
229 int tracing_is_enabled(void)
231 return tracer_enabled
;
235 * trace_buf_size is the size in bytes that is allocated
236 * for a buffer. Note, the number of bytes is always rounded
239 * This number is purposely set to a low number of 16384.
240 * If the dump on oops happens, it will be much appreciated
241 * to not have to wait for all that output. Anyway this can be
242 * boot time and run time configurable.
244 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
246 static unsigned long trace_buf_size
= TRACE_BUF_SIZE_DEFAULT
;
248 /* trace_types holds a link list of available tracers. */
249 static struct tracer
*trace_types __read_mostly
;
251 /* current_trace points to the tracer that is currently active */
252 static struct tracer
*current_trace __read_mostly
;
255 * trace_types_lock is used to protect the trace_types list.
257 static DEFINE_MUTEX(trace_types_lock
);
260 * serialize the access of the ring buffer
262 * ring buffer serializes readers, but it is low level protection.
263 * The validity of the events (which returns by ring_buffer_peek() ..etc)
264 * are not protected by ring buffer.
266 * The content of events may become garbage if we allow other process consumes
267 * these events concurrently:
268 * A) the page of the consumed events may become a normal page
269 * (not reader page) in ring buffer, and this page will be rewrited
270 * by events producer.
271 * B) The page of the consumed events may become a page for splice_read,
272 * and this page will be returned to system.
274 * These primitives allow multi process access to different cpu ring buffer
277 * These primitives don't distinguish read-only and read-consume access.
278 * Multi read-only access are also serialized.
282 static DECLARE_RWSEM(all_cpu_access_lock
);
283 static DEFINE_PER_CPU(struct mutex
, cpu_access_lock
);
285 static inline void trace_access_lock(int cpu
)
287 if (cpu
== TRACE_PIPE_ALL_CPU
) {
288 /* gain it for accessing the whole ring buffer. */
289 down_write(&all_cpu_access_lock
);
291 /* gain it for accessing a cpu ring buffer. */
293 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
294 down_read(&all_cpu_access_lock
);
296 /* Secondly block other access to this @cpu ring buffer. */
297 mutex_lock(&per_cpu(cpu_access_lock
, cpu
));
301 static inline void trace_access_unlock(int cpu
)
303 if (cpu
== TRACE_PIPE_ALL_CPU
) {
304 up_write(&all_cpu_access_lock
);
306 mutex_unlock(&per_cpu(cpu_access_lock
, cpu
));
307 up_read(&all_cpu_access_lock
);
311 static inline void trace_access_lock_init(void)
315 for_each_possible_cpu(cpu
)
316 mutex_init(&per_cpu(cpu_access_lock
, cpu
));
321 static DEFINE_MUTEX(access_lock
);
323 static inline void trace_access_lock(int cpu
)
326 mutex_lock(&access_lock
);
329 static inline void trace_access_unlock(int cpu
)
332 mutex_unlock(&access_lock
);
335 static inline void trace_access_lock_init(void)
341 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
342 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
344 /* trace_flags holds trace_options default values */
345 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
| TRACE_ITER_PRINTK
|
346 TRACE_ITER_ANNOTATE
| TRACE_ITER_CONTEXT_INFO
| TRACE_ITER_SLEEP_TIME
|
347 TRACE_ITER_GRAPH_TIME
;
349 static int trace_stop_count
;
350 static DEFINE_SPINLOCK(tracing_start_lock
);
353 * trace_wake_up - wake up tasks waiting for trace input
355 * Simply wakes up any task that is blocked on the trace_wait
356 * queue. These is used with trace_poll for tasks polling the trace.
358 void trace_wake_up(void)
362 if (trace_flags
& TRACE_ITER_BLOCK
)
365 * The runqueue_is_locked() can fail, but this is the best we
369 if (!runqueue_is_locked(cpu
))
370 wake_up(&trace_wait
);
374 static int __init
set_buf_size(char *str
)
376 unsigned long buf_size
;
380 buf_size
= memparse(str
, &str
);
381 /* nr_entries can not be zero */
384 trace_buf_size
= buf_size
;
387 __setup("trace_buf_size=", set_buf_size
);
389 static int __init
set_tracing_thresh(char *str
)
391 unsigned long threshhold
;
396 ret
= strict_strtoul(str
, 0, &threshhold
);
399 tracing_thresh
= threshhold
* 1000;
402 __setup("tracing_thresh=", set_tracing_thresh
);
404 unsigned long nsecs_to_usecs(unsigned long nsecs
)
409 /* These must match the bit postions in trace_iterator_flags */
410 static const char *trace_options
[] = {
438 { trace_clock_local
, "local" },
439 { trace_clock_global
, "global" },
445 * trace_parser_get_init - gets the buffer for trace parser
447 int trace_parser_get_init(struct trace_parser
*parser
, int size
)
449 memset(parser
, 0, sizeof(*parser
));
451 parser
->buffer
= kmalloc(size
, GFP_KERNEL
);
460 * trace_parser_put - frees the buffer for trace parser
462 void trace_parser_put(struct trace_parser
*parser
)
464 kfree(parser
->buffer
);
468 * trace_get_user - reads the user input string separated by space
469 * (matched by isspace(ch))
471 * For each string found the 'struct trace_parser' is updated,
472 * and the function returns.
474 * Returns number of bytes read.
476 * See kernel/trace/trace.h for 'struct trace_parser' details.
478 int trace_get_user(struct trace_parser
*parser
, const char __user
*ubuf
,
479 size_t cnt
, loff_t
*ppos
)
486 trace_parser_clear(parser
);
488 ret
= get_user(ch
, ubuf
++);
496 * The parser is not finished with the last write,
497 * continue reading the user input without skipping spaces.
500 /* skip white space */
501 while (cnt
&& isspace(ch
)) {
502 ret
= get_user(ch
, ubuf
++);
509 /* only spaces were written */
519 /* read the non-space input */
520 while (cnt
&& !isspace(ch
)) {
521 if (parser
->idx
< parser
->size
- 1)
522 parser
->buffer
[parser
->idx
++] = ch
;
527 ret
= get_user(ch
, ubuf
++);
534 /* We either got finished input or we have to wait for another call. */
536 parser
->buffer
[parser
->idx
] = 0;
537 parser
->cont
= false;
540 parser
->buffer
[parser
->idx
++] = ch
;
550 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
558 if (s
->len
<= s
->readpos
)
561 len
= s
->len
- s
->readpos
;
564 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
574 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
579 if (s
->len
<= s
->readpos
)
582 len
= s
->len
- s
->readpos
;
585 ret
= memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
594 * ftrace_max_lock is used to protect the swapping of buffers
595 * when taking a max snapshot. The buffers themselves are
596 * protected by per_cpu spinlocks. But the action of the swap
597 * needs its own lock.
599 * This is defined as a arch_spinlock_t in order to help
600 * with performance when lockdep debugging is enabled.
602 * It is also used in other places outside the update_max_tr
603 * so it needs to be defined outside of the
604 * CONFIG_TRACER_MAX_TRACE.
606 static arch_spinlock_t ftrace_max_lock
=
607 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
609 unsigned long __read_mostly tracing_thresh
;
611 #ifdef CONFIG_TRACER_MAX_TRACE
612 unsigned long __read_mostly tracing_max_latency
;
615 * Copy the new maximum trace into the separate maximum-trace
616 * structure. (this way the maximum trace is permanently saved,
617 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
620 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
622 struct trace_array_cpu
*data
= tr
->data
[cpu
];
623 struct trace_array_cpu
*max_data
;
626 max_tr
.time_start
= data
->preempt_timestamp
;
628 max_data
= max_tr
.data
[cpu
];
629 max_data
->saved_latency
= tracing_max_latency
;
630 max_data
->critical_start
= data
->critical_start
;
631 max_data
->critical_end
= data
->critical_end
;
633 memcpy(max_data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
634 max_data
->pid
= tsk
->pid
;
635 max_data
->uid
= task_uid(tsk
);
636 max_data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
637 max_data
->policy
= tsk
->policy
;
638 max_data
->rt_priority
= tsk
->rt_priority
;
640 /* record this tasks comm */
641 tracing_record_cmdline(tsk
);
645 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
647 * @tsk: the task with the latency
648 * @cpu: The cpu that initiated the trace.
650 * Flip the buffers between the @tr and the max_tr and record information
651 * about which task was the cause of this latency.
654 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
656 struct ring_buffer
*buf
= tr
->buffer
;
658 if (trace_stop_count
)
661 WARN_ON_ONCE(!irqs_disabled());
662 arch_spin_lock(&ftrace_max_lock
);
664 tr
->buffer
= max_tr
.buffer
;
667 __update_max_tr(tr
, tsk
, cpu
);
668 arch_spin_unlock(&ftrace_max_lock
);
672 * update_max_tr_single - only copy one trace over, and reset the rest
674 * @tsk - task with the latency
675 * @cpu - the cpu of the buffer to copy.
677 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
680 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
684 if (trace_stop_count
)
687 WARN_ON_ONCE(!irqs_disabled());
688 arch_spin_lock(&ftrace_max_lock
);
690 ftrace_disable_cpu();
692 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
696 * We failed to swap the buffer due to a commit taking
697 * place on this CPU. We fail to record, but we reset
698 * the max trace buffer (no one writes directly to it)
699 * and flag that it failed.
701 trace_array_printk(&max_tr
, _THIS_IP_
,
702 "Failed to swap buffers due to commit in progress\n");
707 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
&& ret
!= -EBUSY
);
709 __update_max_tr(tr
, tsk
, cpu
);
710 arch_spin_unlock(&ftrace_max_lock
);
712 #endif /* CONFIG_TRACER_MAX_TRACE */
715 * register_tracer - register a tracer with the ftrace system.
716 * @type - the plugin for the tracer
718 * Register a new plugin tracer.
720 int register_tracer(struct tracer
*type
)
721 __releases(kernel_lock
)
722 __acquires(kernel_lock
)
728 pr_info("Tracer must have a name\n");
732 if (strlen(type
->name
) > MAX_TRACER_SIZE
) {
733 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE
);
738 * When this gets called we hold the BKL which means that
739 * preemption is disabled. Various trace selftests however
740 * need to disable and enable preemption for successful tests.
741 * So we drop the BKL here and grab it after the tests again.
744 mutex_lock(&trace_types_lock
);
746 tracing_selftest_running
= true;
748 for (t
= trace_types
; t
; t
= t
->next
) {
749 if (strcmp(type
->name
, t
->name
) == 0) {
751 pr_info("Tracer %s already registered\n",
759 type
->set_flag
= &dummy_set_flag
;
761 type
->flags
= &dummy_tracer_flags
;
763 if (!type
->flags
->opts
)
764 type
->flags
->opts
= dummy_tracer_opt
;
765 if (!type
->wait_pipe
)
766 type
->wait_pipe
= default_wait_pipe
;
769 #ifdef CONFIG_FTRACE_STARTUP_TEST
770 if (type
->selftest
&& !tracing_selftest_disabled
) {
771 struct tracer
*saved_tracer
= current_trace
;
772 struct trace_array
*tr
= &global_trace
;
775 * Run a selftest on this tracer.
776 * Here we reset the trace buffer, and set the current
777 * tracer to be this tracer. The tracer can then run some
778 * internal tracing to verify that everything is in order.
779 * If we fail, we do not register this tracer.
781 tracing_reset_online_cpus(tr
);
783 current_trace
= type
;
784 /* the test is responsible for initializing and enabling */
785 pr_info("Testing tracer %s: ", type
->name
);
786 ret
= type
->selftest(type
, tr
);
787 /* the test is responsible for resetting too */
788 current_trace
= saved_tracer
;
790 printk(KERN_CONT
"FAILED!\n");
793 /* Only reset on passing, to avoid touching corrupted buffers */
794 tracing_reset_online_cpus(tr
);
796 printk(KERN_CONT
"PASSED\n");
800 type
->next
= trace_types
;
804 tracing_selftest_running
= false;
805 mutex_unlock(&trace_types_lock
);
807 if (ret
|| !default_bootup_tracer
)
810 if (strncmp(default_bootup_tracer
, type
->name
, MAX_TRACER_SIZE
))
813 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
814 /* Do we want this tracer to start on bootup? */
815 tracing_set_tracer(type
->name
);
816 default_bootup_tracer
= NULL
;
817 /* disable other selftests, since this will break it. */
818 tracing_selftest_disabled
= 1;
819 #ifdef CONFIG_FTRACE_STARTUP_TEST
820 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
829 void unregister_tracer(struct tracer
*type
)
833 mutex_lock(&trace_types_lock
);
834 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
838 pr_info("Tracer %s not registered\n", type
->name
);
844 if (type
== current_trace
&& tracer_enabled
) {
847 if (current_trace
->stop
)
848 current_trace
->stop(&global_trace
);
849 current_trace
= &nop_trace
;
852 mutex_unlock(&trace_types_lock
);
855 static void __tracing_reset(struct ring_buffer
*buffer
, int cpu
)
857 ftrace_disable_cpu();
858 ring_buffer_reset_cpu(buffer
, cpu
);
862 void tracing_reset(struct trace_array
*tr
, int cpu
)
864 struct ring_buffer
*buffer
= tr
->buffer
;
866 ring_buffer_record_disable(buffer
);
868 /* Make sure all commits have finished */
870 __tracing_reset(buffer
, cpu
);
872 ring_buffer_record_enable(buffer
);
875 void tracing_reset_online_cpus(struct trace_array
*tr
)
877 struct ring_buffer
*buffer
= tr
->buffer
;
880 ring_buffer_record_disable(buffer
);
882 /* Make sure all commits have finished */
885 tr
->time_start
= ftrace_now(tr
->cpu
);
887 for_each_online_cpu(cpu
)
888 __tracing_reset(buffer
, cpu
);
890 ring_buffer_record_enable(buffer
);
893 void tracing_reset_current(int cpu
)
895 tracing_reset(&global_trace
, cpu
);
898 void tracing_reset_current_online_cpus(void)
900 tracing_reset_online_cpus(&global_trace
);
903 #define SAVED_CMDLINES 128
904 #define NO_CMDLINE_MAP UINT_MAX
905 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
906 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
907 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
908 static int cmdline_idx
;
909 static arch_spinlock_t trace_cmdline_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
911 /* temporary disable recording */
912 static atomic_t trace_record_cmdline_disabled __read_mostly
;
914 static void trace_init_cmdlines(void)
916 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
917 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
921 int is_tracing_stopped(void)
923 return trace_stop_count
;
927 * ftrace_off_permanent - disable all ftrace code permanently
929 * This should only be called when a serious anomally has
930 * been detected. This will turn off the function tracing,
931 * ring buffers, and other tracing utilites. It takes no
932 * locks and can be called from any context.
934 void ftrace_off_permanent(void)
936 tracing_disabled
= 1;
938 tracing_off_permanent();
942 * tracing_start - quick start of the tracer
944 * If tracing is enabled but was stopped by tracing_stop,
945 * this will start the tracer back up.
947 void tracing_start(void)
949 struct ring_buffer
*buffer
;
952 if (tracing_disabled
)
955 spin_lock_irqsave(&tracing_start_lock
, flags
);
956 if (--trace_stop_count
) {
957 if (trace_stop_count
< 0) {
958 /* Someone screwed up their debugging */
960 trace_stop_count
= 0;
965 /* Prevent the buffers from switching */
966 arch_spin_lock(&ftrace_max_lock
);
968 buffer
= global_trace
.buffer
;
970 ring_buffer_record_enable(buffer
);
972 buffer
= max_tr
.buffer
;
974 ring_buffer_record_enable(buffer
);
976 arch_spin_unlock(&ftrace_max_lock
);
980 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
984 * tracing_stop - quick stop of the tracer
986 * Light weight way to stop tracing. Use in conjunction with
989 void tracing_stop(void)
991 struct ring_buffer
*buffer
;
995 spin_lock_irqsave(&tracing_start_lock
, flags
);
996 if (trace_stop_count
++)
999 /* Prevent the buffers from switching */
1000 arch_spin_lock(&ftrace_max_lock
);
1002 buffer
= global_trace
.buffer
;
1004 ring_buffer_record_disable(buffer
);
1006 buffer
= max_tr
.buffer
;
1008 ring_buffer_record_disable(buffer
);
1010 arch_spin_unlock(&ftrace_max_lock
);
1013 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
1016 void trace_stop_cmdline_recording(void);
1018 static void trace_save_cmdline(struct task_struct
*tsk
)
1022 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
1026 * It's not the end of the world if we don't get
1027 * the lock, but we also don't want to spin
1028 * nor do we want to disable interrupts,
1029 * so if we miss here, then better luck next time.
1031 if (!arch_spin_trylock(&trace_cmdline_lock
))
1034 idx
= map_pid_to_cmdline
[tsk
->pid
];
1035 if (idx
== NO_CMDLINE_MAP
) {
1036 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
1039 * Check whether the cmdline buffer at idx has a pid
1040 * mapped. We are going to overwrite that entry so we
1041 * need to clear the map_pid_to_cmdline. Otherwise we
1042 * would read the new comm for the old pid.
1044 pid
= map_cmdline_to_pid
[idx
];
1045 if (pid
!= NO_CMDLINE_MAP
)
1046 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
1048 map_cmdline_to_pid
[idx
] = tsk
->pid
;
1049 map_pid_to_cmdline
[tsk
->pid
] = idx
;
1054 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
1056 arch_spin_unlock(&trace_cmdline_lock
);
1059 void trace_find_cmdline(int pid
, char comm
[])
1064 strcpy(comm
, "<idle>");
1068 if (WARN_ON_ONCE(pid
< 0)) {
1069 strcpy(comm
, "<XXX>");
1073 if (pid
> PID_MAX_DEFAULT
) {
1074 strcpy(comm
, "<...>");
1079 arch_spin_lock(&trace_cmdline_lock
);
1080 map
= map_pid_to_cmdline
[pid
];
1081 if (map
!= NO_CMDLINE_MAP
)
1082 strcpy(comm
, saved_cmdlines
[map
]);
1084 strcpy(comm
, "<...>");
1086 arch_spin_unlock(&trace_cmdline_lock
);
1090 void tracing_record_cmdline(struct task_struct
*tsk
)
1092 if (atomic_read(&trace_record_cmdline_disabled
) || !tracer_enabled
||
1096 trace_save_cmdline(tsk
);
1100 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
1103 struct task_struct
*tsk
= current
;
1105 entry
->preempt_count
= pc
& 0xff;
1106 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
1107 entry
->lock_depth
= (tsk
) ? tsk
->lock_depth
: 0;
1109 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1110 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
1112 TRACE_FLAG_IRQS_NOSUPPORT
|
1114 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
1115 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
1116 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
1118 EXPORT_SYMBOL_GPL(tracing_generic_entry_update
);
1120 struct ring_buffer_event
*
1121 trace_buffer_lock_reserve(struct ring_buffer
*buffer
,
1124 unsigned long flags
, int pc
)
1126 struct ring_buffer_event
*event
;
1128 event
= ring_buffer_lock_reserve(buffer
, len
);
1129 if (event
!= NULL
) {
1130 struct trace_entry
*ent
= ring_buffer_event_data(event
);
1132 tracing_generic_entry_update(ent
, flags
, pc
);
1140 __trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1141 struct ring_buffer_event
*event
,
1142 unsigned long flags
, int pc
,
1145 ring_buffer_unlock_commit(buffer
, event
);
1147 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1148 ftrace_trace_userstack(buffer
, flags
, pc
);
1154 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1155 struct ring_buffer_event
*event
,
1156 unsigned long flags
, int pc
)
1158 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1161 struct ring_buffer_event
*
1162 trace_current_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1163 int type
, unsigned long len
,
1164 unsigned long flags
, int pc
)
1166 *current_rb
= global_trace
.buffer
;
1167 return trace_buffer_lock_reserve(*current_rb
,
1168 type
, len
, flags
, pc
);
1170 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve
);
1172 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
1173 struct ring_buffer_event
*event
,
1174 unsigned long flags
, int pc
)
1176 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1178 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit
);
1180 void trace_nowake_buffer_unlock_commit(struct ring_buffer
*buffer
,
1181 struct ring_buffer_event
*event
,
1182 unsigned long flags
, int pc
)
1184 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 0);
1186 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit
);
1188 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
1189 struct ring_buffer_event
*event
)
1191 ring_buffer_discard_commit(buffer
, event
);
1193 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit
);
1196 trace_function(struct trace_array
*tr
,
1197 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1200 struct ftrace_event_call
*call
= &event_function
;
1201 struct ring_buffer
*buffer
= tr
->buffer
;
1202 struct ring_buffer_event
*event
;
1203 struct ftrace_entry
*entry
;
1205 /* If we are reading the ring buffer, don't trace */
1206 if (unlikely(__this_cpu_read(ftrace_cpu_disabled
)))
1209 event
= trace_buffer_lock_reserve(buffer
, TRACE_FN
, sizeof(*entry
),
1213 entry
= ring_buffer_event_data(event
);
1215 entry
->parent_ip
= parent_ip
;
1217 if (!filter_check_discard(call
, entry
, buffer
, event
))
1218 ring_buffer_unlock_commit(buffer
, event
);
1222 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1223 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1226 if (likely(!atomic_read(&data
->disabled
)))
1227 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
1230 #ifdef CONFIG_STACKTRACE
1231 static void __ftrace_trace_stack(struct ring_buffer
*buffer
,
1232 unsigned long flags
,
1235 struct ftrace_event_call
*call
= &event_kernel_stack
;
1236 struct ring_buffer_event
*event
;
1237 struct stack_entry
*entry
;
1238 struct stack_trace trace
;
1240 event
= trace_buffer_lock_reserve(buffer
, TRACE_STACK
,
1241 sizeof(*entry
), flags
, pc
);
1244 entry
= ring_buffer_event_data(event
);
1245 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1247 trace
.nr_entries
= 0;
1248 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1250 trace
.entries
= entry
->caller
;
1252 save_stack_trace(&trace
);
1253 if (!filter_check_discard(call
, entry
, buffer
, event
))
1254 ring_buffer_unlock_commit(buffer
, event
);
1257 void ftrace_trace_stack(struct ring_buffer
*buffer
, unsigned long flags
,
1260 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1263 __ftrace_trace_stack(buffer
, flags
, skip
, pc
);
1266 void __trace_stack(struct trace_array
*tr
, unsigned long flags
, int skip
,
1269 __ftrace_trace_stack(tr
->buffer
, flags
, skip
, pc
);
1273 * trace_dump_stack - record a stack back trace in the trace buffer
1275 void trace_dump_stack(void)
1277 unsigned long flags
;
1279 if (tracing_disabled
|| tracing_selftest_running
)
1282 local_save_flags(flags
);
1284 /* skipping 3 traces, seems to get us at the caller of this function */
1285 __ftrace_trace_stack(global_trace
.buffer
, flags
, 3, preempt_count());
1289 ftrace_trace_userstack(struct ring_buffer
*buffer
, unsigned long flags
, int pc
)
1291 struct ftrace_event_call
*call
= &event_user_stack
;
1292 struct ring_buffer_event
*event
;
1293 struct userstack_entry
*entry
;
1294 struct stack_trace trace
;
1296 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1300 * NMIs can not handle page faults, even with fix ups.
1301 * The save user stack can (and often does) fault.
1303 if (unlikely(in_nmi()))
1306 event
= trace_buffer_lock_reserve(buffer
, TRACE_USER_STACK
,
1307 sizeof(*entry
), flags
, pc
);
1310 entry
= ring_buffer_event_data(event
);
1312 entry
->tgid
= current
->tgid
;
1313 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1315 trace
.nr_entries
= 0;
1316 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1318 trace
.entries
= entry
->caller
;
1320 save_stack_trace_user(&trace
);
1321 if (!filter_check_discard(call
, entry
, buffer
, event
))
1322 ring_buffer_unlock_commit(buffer
, event
);
1326 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1328 ftrace_trace_userstack(tr
, flags
, preempt_count());
1332 #endif /* CONFIG_STACKTRACE */
1335 ftrace_trace_special(void *__tr
,
1336 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
,
1339 struct ftrace_event_call
*call
= &event_special
;
1340 struct ring_buffer_event
*event
;
1341 struct trace_array
*tr
= __tr
;
1342 struct ring_buffer
*buffer
= tr
->buffer
;
1343 struct special_entry
*entry
;
1345 event
= trace_buffer_lock_reserve(buffer
, TRACE_SPECIAL
,
1346 sizeof(*entry
), 0, pc
);
1349 entry
= ring_buffer_event_data(event
);
1354 if (!filter_check_discard(call
, entry
, buffer
, event
))
1355 trace_buffer_unlock_commit(buffer
, event
, 0, pc
);
1359 __trace_special(void *__tr
, void *__data
,
1360 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1362 ftrace_trace_special(__tr
, arg1
, arg2
, arg3
, preempt_count());
1366 ftrace_special(unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1368 struct trace_array
*tr
= &global_trace
;
1369 struct trace_array_cpu
*data
;
1370 unsigned long flags
;
1374 if (tracing_disabled
)
1377 pc
= preempt_count();
1378 local_irq_save(flags
);
1379 cpu
= raw_smp_processor_id();
1380 data
= tr
->data
[cpu
];
1382 if (likely(atomic_inc_return(&data
->disabled
) == 1))
1383 ftrace_trace_special(tr
, arg1
, arg2
, arg3
, pc
);
1385 atomic_dec(&data
->disabled
);
1386 local_irq_restore(flags
);
1390 * trace_vbprintk - write binary msg to tracing buffer
1393 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1395 static arch_spinlock_t trace_buf_lock
=
1396 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
1397 static u32 trace_buf
[TRACE_BUF_SIZE
];
1399 struct ftrace_event_call
*call
= &event_bprint
;
1400 struct ring_buffer_event
*event
;
1401 struct ring_buffer
*buffer
;
1402 struct trace_array
*tr
= &global_trace
;
1403 struct trace_array_cpu
*data
;
1404 struct bprint_entry
*entry
;
1405 unsigned long flags
;
1408 int cpu
, len
= 0, size
, pc
;
1410 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1413 /* Don't pollute graph traces with trace_vprintk internals */
1414 pause_graph_tracing();
1416 pc
= preempt_count();
1417 resched
= ftrace_preempt_disable();
1418 cpu
= raw_smp_processor_id();
1419 data
= tr
->data
[cpu
];
1421 disable
= atomic_inc_return(&data
->disabled
);
1422 if (unlikely(disable
!= 1))
1425 /* Lockdep uses trace_printk for lock tracing */
1426 local_irq_save(flags
);
1427 arch_spin_lock(&trace_buf_lock
);
1428 len
= vbin_printf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1430 if (len
> TRACE_BUF_SIZE
|| len
< 0)
1433 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1434 buffer
= tr
->buffer
;
1435 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
1439 entry
= ring_buffer_event_data(event
);
1443 memcpy(entry
->buf
, trace_buf
, sizeof(u32
) * len
);
1444 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1445 ring_buffer_unlock_commit(buffer
, event
);
1446 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1450 arch_spin_unlock(&trace_buf_lock
);
1451 local_irq_restore(flags
);
1454 atomic_dec_return(&data
->disabled
);
1455 ftrace_preempt_enable(resched
);
1456 unpause_graph_tracing();
1460 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1462 int trace_array_printk(struct trace_array
*tr
,
1463 unsigned long ip
, const char *fmt
, ...)
1468 if (!(trace_flags
& TRACE_ITER_PRINTK
))
1472 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
1477 int trace_array_vprintk(struct trace_array
*tr
,
1478 unsigned long ip
, const char *fmt
, va_list args
)
1480 static arch_spinlock_t trace_buf_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
1481 static char trace_buf
[TRACE_BUF_SIZE
];
1483 struct ftrace_event_call
*call
= &event_print
;
1484 struct ring_buffer_event
*event
;
1485 struct ring_buffer
*buffer
;
1486 struct trace_array_cpu
*data
;
1487 int cpu
, len
= 0, size
, pc
;
1488 struct print_entry
*entry
;
1489 unsigned long irq_flags
;
1492 if (tracing_disabled
|| tracing_selftest_running
)
1495 pc
= preempt_count();
1496 preempt_disable_notrace();
1497 cpu
= raw_smp_processor_id();
1498 data
= tr
->data
[cpu
];
1500 disable
= atomic_inc_return(&data
->disabled
);
1501 if (unlikely(disable
!= 1))
1504 pause_graph_tracing();
1505 raw_local_irq_save(irq_flags
);
1506 arch_spin_lock(&trace_buf_lock
);
1507 len
= vsnprintf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1509 size
= sizeof(*entry
) + len
+ 1;
1510 buffer
= tr
->buffer
;
1511 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
1515 entry
= ring_buffer_event_data(event
);
1518 memcpy(&entry
->buf
, trace_buf
, len
);
1519 entry
->buf
[len
] = '\0';
1520 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1521 ring_buffer_unlock_commit(buffer
, event
);
1522 ftrace_trace_stack(buffer
, irq_flags
, 6, pc
);
1526 arch_spin_unlock(&trace_buf_lock
);
1527 raw_local_irq_restore(irq_flags
);
1528 unpause_graph_tracing();
1530 atomic_dec_return(&data
->disabled
);
1531 preempt_enable_notrace();
1536 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1538 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
1540 EXPORT_SYMBOL_GPL(trace_vprintk
);
1542 enum trace_file_type
{
1543 TRACE_FILE_LAT_FMT
= 1,
1544 TRACE_FILE_ANNOTATE
= 2,
1547 static void trace_iterator_increment(struct trace_iterator
*iter
)
1549 /* Don't allow ftrace to trace into the ring buffers */
1550 ftrace_disable_cpu();
1553 if (iter
->buffer_iter
[iter
->cpu
])
1554 ring_buffer_read(iter
->buffer_iter
[iter
->cpu
], NULL
);
1556 ftrace_enable_cpu();
1559 static struct trace_entry
*
1560 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
,
1561 unsigned long *lost_events
)
1563 struct ring_buffer_event
*event
;
1564 struct ring_buffer_iter
*buf_iter
= iter
->buffer_iter
[cpu
];
1566 /* Don't allow ftrace to trace into the ring buffers */
1567 ftrace_disable_cpu();
1570 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1572 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
,
1575 ftrace_enable_cpu();
1577 return event
? ring_buffer_event_data(event
) : NULL
;
1580 static struct trace_entry
*
1581 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
,
1582 unsigned long *missing_events
, u64
*ent_ts
)
1584 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1585 struct trace_entry
*ent
, *next
= NULL
;
1586 unsigned long lost_events
= 0, next_lost
= 0;
1587 int cpu_file
= iter
->cpu_file
;
1588 u64 next_ts
= 0, ts
;
1593 * If we are in a per_cpu trace file, don't bother by iterating over
1594 * all cpu and peek directly.
1596 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1597 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1599 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
, missing_events
);
1601 *ent_cpu
= cpu_file
;
1606 for_each_tracing_cpu(cpu
) {
1608 if (ring_buffer_empty_cpu(buffer
, cpu
))
1611 ent
= peek_next_entry(iter
, cpu
, &ts
, &lost_events
);
1614 * Pick the entry with the smallest timestamp:
1616 if (ent
&& (!next
|| ts
< next_ts
)) {
1620 next_lost
= lost_events
;
1625 *ent_cpu
= next_cpu
;
1631 *missing_events
= next_lost
;
1636 /* Find the next real entry, without updating the iterator itself */
1637 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1638 int *ent_cpu
, u64
*ent_ts
)
1640 return __find_next_entry(iter
, ent_cpu
, NULL
, ent_ts
);
1643 /* Find the next real entry, and increment the iterator to the next entry */
1644 static void *find_next_entry_inc(struct trace_iterator
*iter
)
1646 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
,
1647 &iter
->lost_events
, &iter
->ts
);
1650 trace_iterator_increment(iter
);
1652 return iter
->ent
? iter
: NULL
;
1655 static void trace_consume(struct trace_iterator
*iter
)
1657 /* Don't allow ftrace to trace into the ring buffers */
1658 ftrace_disable_cpu();
1659 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
,
1660 &iter
->lost_events
);
1661 ftrace_enable_cpu();
1664 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1666 struct trace_iterator
*iter
= m
->private;
1670 WARN_ON_ONCE(iter
->leftover
);
1674 /* can't go backwards */
1679 ent
= find_next_entry_inc(iter
);
1683 while (ent
&& iter
->idx
< i
)
1684 ent
= find_next_entry_inc(iter
);
1691 static void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
1693 struct trace_array
*tr
= iter
->tr
;
1694 struct ring_buffer_event
*event
;
1695 struct ring_buffer_iter
*buf_iter
;
1696 unsigned long entries
= 0;
1699 tr
->data
[cpu
]->skipped_entries
= 0;
1701 if (!iter
->buffer_iter
[cpu
])
1704 buf_iter
= iter
->buffer_iter
[cpu
];
1705 ring_buffer_iter_reset(buf_iter
);
1708 * We could have the case with the max latency tracers
1709 * that a reset never took place on a cpu. This is evident
1710 * by the timestamp being before the start of the buffer.
1712 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
1713 if (ts
>= iter
->tr
->time_start
)
1716 ring_buffer_read(buf_iter
, NULL
);
1719 tr
->data
[cpu
]->skipped_entries
= entries
;
1723 * The current tracer is copied to avoid a global locking
1726 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1728 struct trace_iterator
*iter
= m
->private;
1729 static struct tracer
*old_tracer
;
1730 int cpu_file
= iter
->cpu_file
;
1735 /* copy the tracer to avoid using a global lock all around */
1736 mutex_lock(&trace_types_lock
);
1737 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1738 old_tracer
= current_trace
;
1739 *iter
->trace
= *current_trace
;
1741 mutex_unlock(&trace_types_lock
);
1743 atomic_inc(&trace_record_cmdline_disabled
);
1745 if (*pos
!= iter
->pos
) {
1750 ftrace_disable_cpu();
1752 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1753 for_each_tracing_cpu(cpu
)
1754 tracing_iter_reset(iter
, cpu
);
1756 tracing_iter_reset(iter
, cpu_file
);
1758 ftrace_enable_cpu();
1761 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1766 * If we overflowed the seq_file before, then we want
1767 * to just reuse the trace_seq buffer again.
1773 p
= s_next(m
, p
, &l
);
1777 trace_event_read_lock();
1778 trace_access_lock(cpu_file
);
1782 static void s_stop(struct seq_file
*m
, void *p
)
1784 struct trace_iterator
*iter
= m
->private;
1786 atomic_dec(&trace_record_cmdline_disabled
);
1787 trace_access_unlock(iter
->cpu_file
);
1788 trace_event_read_unlock();
1791 static void print_lat_help_header(struct seq_file
*m
)
1793 seq_puts(m
, "# _------=> CPU# \n");
1794 seq_puts(m
, "# / _-----=> irqs-off \n");
1795 seq_puts(m
, "# | / _----=> need-resched \n");
1796 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1797 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1798 seq_puts(m
, "# |||| /_--=> lock-depth \n");
1799 seq_puts(m
, "# |||||/ delay \n");
1800 seq_puts(m
, "# cmd pid |||||| time | caller \n");
1801 seq_puts(m
, "# \\ / |||||| \\ | / \n");
1804 static void print_func_help_header(struct seq_file
*m
)
1806 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1807 seq_puts(m
, "# | | | | |\n");
1812 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1814 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1815 struct trace_array
*tr
= iter
->tr
;
1816 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1817 struct tracer
*type
= current_trace
;
1818 unsigned long entries
= 0;
1819 unsigned long total
= 0;
1820 unsigned long count
;
1821 const char *name
= "preemption";
1828 for_each_tracing_cpu(cpu
) {
1829 count
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
1831 * If this buffer has skipped entries, then we hold all
1832 * entries for the trace and we need to ignore the
1833 * ones before the time stamp.
1835 if (tr
->data
[cpu
]->skipped_entries
) {
1836 count
-= tr
->data
[cpu
]->skipped_entries
;
1837 /* total is the same as the entries */
1841 ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
1845 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
1847 seq_puts(m
, "# -----------------------------------"
1848 "---------------------------------\n");
1849 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1850 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1851 nsecs_to_usecs(data
->saved_latency
),
1855 #if defined(CONFIG_PREEMPT_NONE)
1857 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1859 #elif defined(CONFIG_PREEMPT)
1864 /* These are reserved for later use */
1867 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1871 seq_puts(m
, "# -----------------\n");
1872 seq_printf(m
, "# | task: %.16s-%d "
1873 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1874 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1875 data
->policy
, data
->rt_priority
);
1876 seq_puts(m
, "# -----------------\n");
1878 if (data
->critical_start
) {
1879 seq_puts(m
, "# => started at: ");
1880 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1881 trace_print_seq(m
, &iter
->seq
);
1882 seq_puts(m
, "\n# => ended at: ");
1883 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1884 trace_print_seq(m
, &iter
->seq
);
1885 seq_puts(m
, "\n#\n");
1891 static void test_cpu_buff_start(struct trace_iterator
*iter
)
1893 struct trace_seq
*s
= &iter
->seq
;
1895 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
1898 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
1901 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
1904 if (iter
->tr
->data
[iter
->cpu
]->skipped_entries
)
1907 cpumask_set_cpu(iter
->cpu
, iter
->started
);
1909 /* Don't print started cpu buffer for the first entry of the trace */
1911 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
1915 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
1917 struct trace_seq
*s
= &iter
->seq
;
1918 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1919 struct trace_entry
*entry
;
1920 struct trace_event
*event
;
1924 test_cpu_buff_start(iter
);
1926 event
= ftrace_find_event(entry
->type
);
1928 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1929 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1930 if (!trace_print_lat_context(iter
))
1933 if (!trace_print_context(iter
))
1939 return event
->trace(iter
, sym_flags
);
1941 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
1944 return TRACE_TYPE_HANDLED
;
1946 return TRACE_TYPE_PARTIAL_LINE
;
1949 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
1951 struct trace_seq
*s
= &iter
->seq
;
1952 struct trace_entry
*entry
;
1953 struct trace_event
*event
;
1957 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1958 if (!trace_seq_printf(s
, "%d %d %llu ",
1959 entry
->pid
, iter
->cpu
, iter
->ts
))
1963 event
= ftrace_find_event(entry
->type
);
1965 return event
->raw(iter
, 0);
1967 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
1970 return TRACE_TYPE_HANDLED
;
1972 return TRACE_TYPE_PARTIAL_LINE
;
1975 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
1977 struct trace_seq
*s
= &iter
->seq
;
1978 unsigned char newline
= '\n';
1979 struct trace_entry
*entry
;
1980 struct trace_event
*event
;
1984 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1985 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1986 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1987 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
1990 event
= ftrace_find_event(entry
->type
);
1992 enum print_line_t ret
= event
->hex(iter
, 0);
1993 if (ret
!= TRACE_TYPE_HANDLED
)
1997 SEQ_PUT_FIELD_RET(s
, newline
);
1999 return TRACE_TYPE_HANDLED
;
2002 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
2004 struct trace_seq
*s
= &iter
->seq
;
2005 struct trace_entry
*entry
;
2006 struct trace_event
*event
;
2010 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2011 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
2012 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
2013 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
2016 event
= ftrace_find_event(entry
->type
);
2017 return event
? event
->binary(iter
, 0) : TRACE_TYPE_HANDLED
;
2020 int trace_empty(struct trace_iterator
*iter
)
2024 /* If we are looking at one CPU buffer, only check that one */
2025 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
2026 cpu
= iter
->cpu_file
;
2027 if (iter
->buffer_iter
[cpu
]) {
2028 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
2031 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2037 for_each_tracing_cpu(cpu
) {
2038 if (iter
->buffer_iter
[cpu
]) {
2039 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
2042 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2050 /* Called with trace_event_read_lock() held. */
2051 static enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
2053 enum print_line_t ret
;
2055 if (iter
->lost_events
)
2056 trace_seq_printf(&iter
->seq
, "CPU:%d [LOST %lu EVENTS]\n",
2057 iter
->cpu
, iter
->lost_events
);
2059 if (iter
->trace
&& iter
->trace
->print_line
) {
2060 ret
= iter
->trace
->print_line(iter
);
2061 if (ret
!= TRACE_TYPE_UNHANDLED
)
2065 if (iter
->ent
->type
== TRACE_BPRINT
&&
2066 trace_flags
& TRACE_ITER_PRINTK
&&
2067 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2068 return trace_print_bprintk_msg_only(iter
);
2070 if (iter
->ent
->type
== TRACE_PRINT
&&
2071 trace_flags
& TRACE_ITER_PRINTK
&&
2072 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2073 return trace_print_printk_msg_only(iter
);
2075 if (trace_flags
& TRACE_ITER_BIN
)
2076 return print_bin_fmt(iter
);
2078 if (trace_flags
& TRACE_ITER_HEX
)
2079 return print_hex_fmt(iter
);
2081 if (trace_flags
& TRACE_ITER_RAW
)
2082 return print_raw_fmt(iter
);
2084 return print_trace_fmt(iter
);
2087 void trace_default_header(struct seq_file
*m
)
2089 struct trace_iterator
*iter
= m
->private;
2091 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2092 /* print nothing if the buffers are empty */
2093 if (trace_empty(iter
))
2095 print_trace_header(m
, iter
);
2096 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2097 print_lat_help_header(m
);
2099 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2100 print_func_help_header(m
);
2104 static int s_show(struct seq_file
*m
, void *v
)
2106 struct trace_iterator
*iter
= v
;
2109 if (iter
->ent
== NULL
) {
2111 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
2114 if (iter
->trace
&& iter
->trace
->print_header
)
2115 iter
->trace
->print_header(m
);
2117 trace_default_header(m
);
2119 } else if (iter
->leftover
) {
2121 * If we filled the seq_file buffer earlier, we
2122 * want to just show it now.
2124 ret
= trace_print_seq(m
, &iter
->seq
);
2126 /* ret should this time be zero, but you never know */
2127 iter
->leftover
= ret
;
2130 print_trace_line(iter
);
2131 ret
= trace_print_seq(m
, &iter
->seq
);
2133 * If we overflow the seq_file buffer, then it will
2134 * ask us for this data again at start up.
2136 * ret is 0 if seq_file write succeeded.
2139 iter
->leftover
= ret
;
2145 static const struct seq_operations tracer_seq_ops
= {
2152 static struct trace_iterator
*
2153 __tracing_open(struct inode
*inode
, struct file
*file
)
2155 long cpu_file
= (long) inode
->i_private
;
2156 void *fail_ret
= ERR_PTR(-ENOMEM
);
2157 struct trace_iterator
*iter
;
2161 if (tracing_disabled
)
2162 return ERR_PTR(-ENODEV
);
2164 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2166 return ERR_PTR(-ENOMEM
);
2169 * We make a copy of the current tracer to avoid concurrent
2170 * changes on it while we are reading.
2172 mutex_lock(&trace_types_lock
);
2173 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2178 *iter
->trace
= *current_trace
;
2180 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2183 if (current_trace
&& current_trace
->print_max
)
2186 iter
->tr
= &global_trace
;
2188 mutex_init(&iter
->mutex
);
2189 iter
->cpu_file
= cpu_file
;
2191 /* Notify the tracer early; before we stop tracing. */
2192 if (iter
->trace
&& iter
->trace
->open
)
2193 iter
->trace
->open(iter
);
2195 /* Annotate start of buffers if we had overruns */
2196 if (ring_buffer_overruns(iter
->tr
->buffer
))
2197 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2199 /* stop the trace while dumping */
2202 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
2203 for_each_tracing_cpu(cpu
) {
2204 iter
->buffer_iter
[cpu
] =
2205 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2207 ring_buffer_read_prepare_sync();
2208 for_each_tracing_cpu(cpu
) {
2209 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2210 tracing_iter_reset(iter
, cpu
);
2213 cpu
= iter
->cpu_file
;
2214 iter
->buffer_iter
[cpu
] =
2215 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2216 ring_buffer_read_prepare_sync();
2217 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2218 tracing_iter_reset(iter
, cpu
);
2221 ret
= seq_open(file
, &tracer_seq_ops
);
2223 fail_ret
= ERR_PTR(ret
);
2227 m
= file
->private_data
;
2230 mutex_unlock(&trace_types_lock
);
2235 for_each_tracing_cpu(cpu
) {
2236 if (iter
->buffer_iter
[cpu
])
2237 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2239 free_cpumask_var(iter
->started
);
2242 mutex_unlock(&trace_types_lock
);
2249 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2251 if (tracing_disabled
)
2254 filp
->private_data
= inode
->i_private
;
2258 static int tracing_release(struct inode
*inode
, struct file
*file
)
2260 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
2261 struct trace_iterator
*iter
;
2264 if (!(file
->f_mode
& FMODE_READ
))
2269 mutex_lock(&trace_types_lock
);
2270 for_each_tracing_cpu(cpu
) {
2271 if (iter
->buffer_iter
[cpu
])
2272 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2275 if (iter
->trace
&& iter
->trace
->close
)
2276 iter
->trace
->close(iter
);
2278 /* reenable tracing if it was previously enabled */
2280 mutex_unlock(&trace_types_lock
);
2282 seq_release(inode
, file
);
2283 mutex_destroy(&iter
->mutex
);
2284 free_cpumask_var(iter
->started
);
2290 static int tracing_open(struct inode
*inode
, struct file
*file
)
2292 struct trace_iterator
*iter
;
2295 /* If this file was open for write, then erase contents */
2296 if ((file
->f_mode
& FMODE_WRITE
) &&
2297 (file
->f_flags
& O_TRUNC
)) {
2298 long cpu
= (long) inode
->i_private
;
2300 if (cpu
== TRACE_PIPE_ALL_CPU
)
2301 tracing_reset_online_cpus(&global_trace
);
2303 tracing_reset(&global_trace
, cpu
);
2306 if (file
->f_mode
& FMODE_READ
) {
2307 iter
= __tracing_open(inode
, file
);
2309 ret
= PTR_ERR(iter
);
2310 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2311 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2317 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2319 struct tracer
*t
= v
;
2329 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2334 mutex_lock(&trace_types_lock
);
2335 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2341 static void t_stop(struct seq_file
*m
, void *p
)
2343 mutex_unlock(&trace_types_lock
);
2346 static int t_show(struct seq_file
*m
, void *v
)
2348 struct tracer
*t
= v
;
2353 seq_printf(m
, "%s", t
->name
);
2362 static const struct seq_operations show_traces_seq_ops
= {
2369 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2371 if (tracing_disabled
)
2374 return seq_open(file
, &show_traces_seq_ops
);
2378 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2379 size_t count
, loff_t
*ppos
)
2384 static const struct file_operations tracing_fops
= {
2385 .open
= tracing_open
,
2387 .write
= tracing_write_stub
,
2388 .llseek
= seq_lseek
,
2389 .release
= tracing_release
,
2392 static const struct file_operations show_traces_fops
= {
2393 .open
= show_traces_open
,
2395 .release
= seq_release
,
2399 * Only trace on a CPU if the bitmask is set:
2401 static cpumask_var_t tracing_cpumask
;
2404 * The tracer itself will not take this lock, but still we want
2405 * to provide a consistent cpumask to user-space:
2407 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2410 * Temporary storage for the character representation of the
2411 * CPU bitmask (and one more byte for the newline):
2413 static char mask_str
[NR_CPUS
+ 1];
2416 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2417 size_t count
, loff_t
*ppos
)
2421 mutex_lock(&tracing_cpumask_update_lock
);
2423 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2424 if (count
- len
< 2) {
2428 len
+= sprintf(mask_str
+ len
, "\n");
2429 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2432 mutex_unlock(&tracing_cpumask_update_lock
);
2438 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2439 size_t count
, loff_t
*ppos
)
2442 cpumask_var_t tracing_cpumask_new
;
2444 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2447 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2451 mutex_lock(&tracing_cpumask_update_lock
);
2453 local_irq_disable();
2454 arch_spin_lock(&ftrace_max_lock
);
2455 for_each_tracing_cpu(cpu
) {
2457 * Increase/decrease the disabled counter if we are
2458 * about to flip a bit in the cpumask:
2460 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2461 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2462 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2464 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2465 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2466 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2469 arch_spin_unlock(&ftrace_max_lock
);
2472 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2474 mutex_unlock(&tracing_cpumask_update_lock
);
2475 free_cpumask_var(tracing_cpumask_new
);
2480 free_cpumask_var(tracing_cpumask_new
);
2485 static const struct file_operations tracing_cpumask_fops
= {
2486 .open
= tracing_open_generic
,
2487 .read
= tracing_cpumask_read
,
2488 .write
= tracing_cpumask_write
,
2491 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
2493 struct tracer_opt
*trace_opts
;
2497 mutex_lock(&trace_types_lock
);
2498 tracer_flags
= current_trace
->flags
->val
;
2499 trace_opts
= current_trace
->flags
->opts
;
2501 for (i
= 0; trace_options
[i
]; i
++) {
2502 if (trace_flags
& (1 << i
))
2503 seq_printf(m
, "%s\n", trace_options
[i
]);
2505 seq_printf(m
, "no%s\n", trace_options
[i
]);
2508 for (i
= 0; trace_opts
[i
].name
; i
++) {
2509 if (tracer_flags
& trace_opts
[i
].bit
)
2510 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
2512 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
2514 mutex_unlock(&trace_types_lock
);
2519 static int __set_tracer_option(struct tracer
*trace
,
2520 struct tracer_flags
*tracer_flags
,
2521 struct tracer_opt
*opts
, int neg
)
2525 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
2530 tracer_flags
->val
&= ~opts
->bit
;
2532 tracer_flags
->val
|= opts
->bit
;
2536 /* Try to assign a tracer specific option */
2537 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2539 struct tracer_flags
*tracer_flags
= trace
->flags
;
2540 struct tracer_opt
*opts
= NULL
;
2543 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
2544 opts
= &tracer_flags
->opts
[i
];
2546 if (strcmp(cmp
, opts
->name
) == 0)
2547 return __set_tracer_option(trace
, trace
->flags
,
2554 static void set_tracer_flags(unsigned int mask
, int enabled
)
2556 /* do nothing if flag is already set */
2557 if (!!(trace_flags
& mask
) == !!enabled
)
2561 trace_flags
|= mask
;
2563 trace_flags
&= ~mask
;
2567 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2568 size_t cnt
, loff_t
*ppos
)
2576 if (cnt
>= sizeof(buf
))
2579 if (copy_from_user(&buf
, ubuf
, cnt
))
2583 cmp
= strstrip(buf
);
2585 if (strncmp(cmp
, "no", 2) == 0) {
2590 for (i
= 0; trace_options
[i
]; i
++) {
2591 if (strcmp(cmp
, trace_options
[i
]) == 0) {
2592 set_tracer_flags(1 << i
, !neg
);
2597 /* If no option could be set, test the specific tracer options */
2598 if (!trace_options
[i
]) {
2599 mutex_lock(&trace_types_lock
);
2600 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2601 mutex_unlock(&trace_types_lock
);
2611 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
2613 if (tracing_disabled
)
2615 return single_open(file
, tracing_trace_options_show
, NULL
);
2618 static const struct file_operations tracing_iter_fops
= {
2619 .open
= tracing_trace_options_open
,
2621 .llseek
= seq_lseek
,
2622 .release
= single_release
,
2623 .write
= tracing_trace_options_write
,
2626 static const char readme_msg
[] =
2627 "tracing mini-HOWTO:\n\n"
2628 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2629 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2630 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2631 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2633 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2634 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2636 "# cat /sys/kernel/debug/tracing/trace_options\n"
2637 "noprint-parent nosym-offset nosym-addr noverbose\n"
2638 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2639 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2640 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2641 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2645 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2646 size_t cnt
, loff_t
*ppos
)
2648 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2649 readme_msg
, strlen(readme_msg
));
2652 static const struct file_operations tracing_readme_fops
= {
2653 .open
= tracing_open_generic
,
2654 .read
= tracing_readme_read
,
2658 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
2659 size_t cnt
, loff_t
*ppos
)
2668 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
2672 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
2680 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
2683 pid
= map_cmdline_to_pid
[i
];
2684 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
2687 trace_find_cmdline(pid
, buf_comm
);
2688 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
2693 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
2702 static const struct file_operations tracing_saved_cmdlines_fops
= {
2703 .open
= tracing_open_generic
,
2704 .read
= tracing_saved_cmdlines_read
,
2708 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2709 size_t cnt
, loff_t
*ppos
)
2714 r
= sprintf(buf
, "%u\n", tracer_enabled
);
2715 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2719 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2720 size_t cnt
, loff_t
*ppos
)
2722 struct trace_array
*tr
= filp
->private_data
;
2727 if (cnt
>= sizeof(buf
))
2730 if (copy_from_user(&buf
, ubuf
, cnt
))
2735 ret
= strict_strtoul(buf
, 10, &val
);
2741 mutex_lock(&trace_types_lock
);
2742 if (tracer_enabled
^ val
) {
2745 if (current_trace
->start
)
2746 current_trace
->start(tr
);
2751 if (current_trace
->stop
)
2752 current_trace
->stop(tr
);
2755 mutex_unlock(&trace_types_lock
);
2763 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2764 size_t cnt
, loff_t
*ppos
)
2766 char buf
[MAX_TRACER_SIZE
+2];
2769 mutex_lock(&trace_types_lock
);
2771 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2773 r
= sprintf(buf
, "\n");
2774 mutex_unlock(&trace_types_lock
);
2776 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2779 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
2781 tracing_reset_online_cpus(tr
);
2785 static int tracing_resize_ring_buffer(unsigned long size
)
2790 * If kernel or user changes the size of the ring buffer
2791 * we use the size that was given, and we can forget about
2792 * expanding it later.
2794 ring_buffer_expanded
= 1;
2796 ret
= ring_buffer_resize(global_trace
.buffer
, size
);
2800 ret
= ring_buffer_resize(max_tr
.buffer
, size
);
2804 r
= ring_buffer_resize(global_trace
.buffer
,
2805 global_trace
.entries
);
2808 * AARGH! We are left with different
2809 * size max buffer!!!!
2810 * The max buffer is our "snapshot" buffer.
2811 * When a tracer needs a snapshot (one of the
2812 * latency tracers), it swaps the max buffer
2813 * with the saved snap shot. We succeeded to
2814 * update the size of the main buffer, but failed to
2815 * update the size of the max buffer. But when we tried
2816 * to reset the main buffer to the original size, we
2817 * failed there too. This is very unlikely to
2818 * happen, but if it does, warn and kill all
2822 tracing_disabled
= 1;
2827 global_trace
.entries
= size
;
2833 * tracing_update_buffers - used by tracing facility to expand ring buffers
2835 * To save on memory when the tracing is never used on a system with it
2836 * configured in. The ring buffers are set to a minimum size. But once
2837 * a user starts to use the tracing facility, then they need to grow
2838 * to their default size.
2840 * This function is to be called when a tracer is about to be used.
2842 int tracing_update_buffers(void)
2846 mutex_lock(&trace_types_lock
);
2847 if (!ring_buffer_expanded
)
2848 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2849 mutex_unlock(&trace_types_lock
);
2854 struct trace_option_dentry
;
2856 static struct trace_option_dentry
*
2857 create_trace_option_files(struct tracer
*tracer
);
2860 destroy_trace_option_files(struct trace_option_dentry
*topts
);
2862 static int tracing_set_tracer(const char *buf
)
2864 static struct trace_option_dentry
*topts
;
2865 struct trace_array
*tr
= &global_trace
;
2869 mutex_lock(&trace_types_lock
);
2871 if (!ring_buffer_expanded
) {
2872 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2878 for (t
= trace_types
; t
; t
= t
->next
) {
2879 if (strcmp(t
->name
, buf
) == 0)
2886 if (t
== current_trace
)
2889 trace_branch_disable();
2890 if (current_trace
&& current_trace
->reset
)
2891 current_trace
->reset(tr
);
2893 destroy_trace_option_files(topts
);
2897 topts
= create_trace_option_files(current_trace
);
2900 ret
= tracer_init(t
, tr
);
2905 trace_branch_enable(tr
);
2907 mutex_unlock(&trace_types_lock
);
2913 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2914 size_t cnt
, loff_t
*ppos
)
2916 char buf
[MAX_TRACER_SIZE
+1];
2923 if (cnt
> MAX_TRACER_SIZE
)
2924 cnt
= MAX_TRACER_SIZE
;
2926 if (copy_from_user(&buf
, ubuf
, cnt
))
2931 /* strip ending whitespace. */
2932 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2935 err
= tracing_set_tracer(buf
);
2945 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2946 size_t cnt
, loff_t
*ppos
)
2948 unsigned long *ptr
= filp
->private_data
;
2952 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
2953 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2954 if (r
> sizeof(buf
))
2956 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2960 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2961 size_t cnt
, loff_t
*ppos
)
2963 unsigned long *ptr
= filp
->private_data
;
2968 if (cnt
>= sizeof(buf
))
2971 if (copy_from_user(&buf
, ubuf
, cnt
))
2976 ret
= strict_strtoul(buf
, 10, &val
);
2985 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2987 long cpu_file
= (long) inode
->i_private
;
2988 struct trace_iterator
*iter
;
2991 if (tracing_disabled
)
2994 mutex_lock(&trace_types_lock
);
2996 /* create a buffer to store the information to pass to userspace */
2997 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
3004 * We make a copy of the current tracer to avoid concurrent
3005 * changes on it while we are reading.
3007 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
3013 *iter
->trace
= *current_trace
;
3015 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
3020 /* trace pipe does not show start of buffer */
3021 cpumask_setall(iter
->started
);
3023 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
3024 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
3026 iter
->cpu_file
= cpu_file
;
3027 iter
->tr
= &global_trace
;
3028 mutex_init(&iter
->mutex
);
3029 filp
->private_data
= iter
;
3031 if (iter
->trace
->pipe_open
)
3032 iter
->trace
->pipe_open(iter
);
3035 mutex_unlock(&trace_types_lock
);
3041 mutex_unlock(&trace_types_lock
);
3045 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
3047 struct trace_iterator
*iter
= file
->private_data
;
3049 mutex_lock(&trace_types_lock
);
3051 if (iter
->trace
->pipe_close
)
3052 iter
->trace
->pipe_close(iter
);
3054 mutex_unlock(&trace_types_lock
);
3056 free_cpumask_var(iter
->started
);
3057 mutex_destroy(&iter
->mutex
);
3065 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
3067 struct trace_iterator
*iter
= filp
->private_data
;
3069 if (trace_flags
& TRACE_ITER_BLOCK
) {
3071 * Always select as readable when in blocking mode
3073 return POLLIN
| POLLRDNORM
;
3075 if (!trace_empty(iter
))
3076 return POLLIN
| POLLRDNORM
;
3077 poll_wait(filp
, &trace_wait
, poll_table
);
3078 if (!trace_empty(iter
))
3079 return POLLIN
| POLLRDNORM
;
3086 void default_wait_pipe(struct trace_iterator
*iter
)
3090 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
3092 if (trace_empty(iter
))
3095 finish_wait(&trace_wait
, &wait
);
3099 * This is a make-shift waitqueue.
3100 * A tracer might use this callback on some rare cases:
3102 * 1) the current tracer might hold the runqueue lock when it wakes up
3103 * a reader, hence a deadlock (sched, function, and function graph tracers)
3104 * 2) the function tracers, trace all functions, we don't want
3105 * the overhead of calling wake_up and friends
3106 * (and tracing them too)
3108 * Anyway, this is really very primitive wakeup.
3110 void poll_wait_pipe(struct trace_iterator
*iter
)
3112 set_current_state(TASK_INTERRUPTIBLE
);
3113 /* sleep for 100 msecs, and try again. */
3114 schedule_timeout(HZ
/ 10);
3117 /* Must be called with trace_types_lock mutex held. */
3118 static int tracing_wait_pipe(struct file
*filp
)
3120 struct trace_iterator
*iter
= filp
->private_data
;
3122 while (trace_empty(iter
)) {
3124 if ((filp
->f_flags
& O_NONBLOCK
)) {
3128 mutex_unlock(&iter
->mutex
);
3130 iter
->trace
->wait_pipe(iter
);
3132 mutex_lock(&iter
->mutex
);
3134 if (signal_pending(current
))
3138 * We block until we read something and tracing is disabled.
3139 * We still block if tracing is disabled, but we have never
3140 * read anything. This allows a user to cat this file, and
3141 * then enable tracing. But after we have read something,
3142 * we give an EOF when tracing is again disabled.
3144 * iter->pos will be 0 if we haven't read anything.
3146 if (!tracer_enabled
&& iter
->pos
)
3157 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
3158 size_t cnt
, loff_t
*ppos
)
3160 struct trace_iterator
*iter
= filp
->private_data
;
3161 static struct tracer
*old_tracer
;
3164 /* return any leftover data */
3165 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3169 trace_seq_init(&iter
->seq
);
3171 /* copy the tracer to avoid using a global lock all around */
3172 mutex_lock(&trace_types_lock
);
3173 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3174 old_tracer
= current_trace
;
3175 *iter
->trace
= *current_trace
;
3177 mutex_unlock(&trace_types_lock
);
3180 * Avoid more than one consumer on a single file descriptor
3181 * This is just a matter of traces coherency, the ring buffer itself
3184 mutex_lock(&iter
->mutex
);
3185 if (iter
->trace
->read
) {
3186 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
3192 sret
= tracing_wait_pipe(filp
);
3196 /* stop when tracing is finished */
3197 if (trace_empty(iter
)) {
3202 if (cnt
>= PAGE_SIZE
)
3203 cnt
= PAGE_SIZE
- 1;
3205 /* reset all but tr, trace, and overruns */
3206 memset(&iter
->seq
, 0,
3207 sizeof(struct trace_iterator
) -
3208 offsetof(struct trace_iterator
, seq
));
3211 trace_event_read_lock();
3212 trace_access_lock(iter
->cpu_file
);
3213 while (find_next_entry_inc(iter
) != NULL
) {
3214 enum print_line_t ret
;
3215 int len
= iter
->seq
.len
;
3217 ret
= print_trace_line(iter
);
3218 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3219 /* don't print partial lines */
3220 iter
->seq
.len
= len
;
3223 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3224 trace_consume(iter
);
3226 if (iter
->seq
.len
>= cnt
)
3229 trace_access_unlock(iter
->cpu_file
);
3230 trace_event_read_unlock();
3232 /* Now copy what we have to the user */
3233 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3234 if (iter
->seq
.readpos
>= iter
->seq
.len
)
3235 trace_seq_init(&iter
->seq
);
3238 * If there was nothing to send to user, inspite of consuming trace
3239 * entries, go back to wait for more entries.
3245 mutex_unlock(&iter
->mutex
);
3250 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
3251 struct pipe_buffer
*buf
)
3253 __free_page(buf
->page
);
3256 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
3259 __free_page(spd
->pages
[idx
]);
3262 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
3264 .map
= generic_pipe_buf_map
,
3265 .unmap
= generic_pipe_buf_unmap
,
3266 .confirm
= generic_pipe_buf_confirm
,
3267 .release
= tracing_pipe_buf_release
,
3268 .steal
= generic_pipe_buf_steal
,
3269 .get
= generic_pipe_buf_get
,
3273 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
3278 /* Seq buffer is page-sized, exactly what we need. */
3280 count
= iter
->seq
.len
;
3281 ret
= print_trace_line(iter
);
3282 count
= iter
->seq
.len
- count
;
3285 iter
->seq
.len
-= count
;
3288 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3289 iter
->seq
.len
-= count
;
3293 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3294 trace_consume(iter
);
3296 if (!find_next_entry_inc(iter
)) {
3306 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3308 struct pipe_inode_info
*pipe
,
3312 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3313 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3314 struct trace_iterator
*iter
= filp
->private_data
;
3315 struct splice_pipe_desc spd
= {
3317 .partial
= partial_def
,
3318 .nr_pages
= 0, /* This gets updated below. */
3320 .ops
= &tracing_pipe_buf_ops
,
3321 .spd_release
= tracing_spd_release_pipe
,
3323 static struct tracer
*old_tracer
;
3328 if (splice_grow_spd(pipe
, &spd
))
3331 /* copy the tracer to avoid using a global lock all around */
3332 mutex_lock(&trace_types_lock
);
3333 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3334 old_tracer
= current_trace
;
3335 *iter
->trace
= *current_trace
;
3337 mutex_unlock(&trace_types_lock
);
3339 mutex_lock(&iter
->mutex
);
3341 if (iter
->trace
->splice_read
) {
3342 ret
= iter
->trace
->splice_read(iter
, filp
,
3343 ppos
, pipe
, len
, flags
);
3348 ret
= tracing_wait_pipe(filp
);
3352 if (!iter
->ent
&& !find_next_entry_inc(iter
)) {
3357 trace_event_read_lock();
3358 trace_access_lock(iter
->cpu_file
);
3360 /* Fill as many pages as possible. */
3361 for (i
= 0, rem
= len
; i
< pipe
->buffers
&& rem
; i
++) {
3362 spd
.pages
[i
] = alloc_page(GFP_KERNEL
);
3366 rem
= tracing_fill_pipe_page(rem
, iter
);
3368 /* Copy the data into the page, so we can start over. */
3369 ret
= trace_seq_to_buffer(&iter
->seq
,
3370 page_address(spd
.pages
[i
]),
3373 __free_page(spd
.pages
[i
]);
3376 spd
.partial
[i
].offset
= 0;
3377 spd
.partial
[i
].len
= iter
->seq
.len
;
3379 trace_seq_init(&iter
->seq
);
3382 trace_access_unlock(iter
->cpu_file
);
3383 trace_event_read_unlock();
3384 mutex_unlock(&iter
->mutex
);
3388 ret
= splice_to_pipe(pipe
, &spd
);
3390 splice_shrink_spd(pipe
, &spd
);
3394 mutex_unlock(&iter
->mutex
);
3399 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3400 size_t cnt
, loff_t
*ppos
)
3402 struct trace_array
*tr
= filp
->private_data
;
3406 mutex_lock(&trace_types_lock
);
3407 if (!ring_buffer_expanded
)
3408 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3410 trace_buf_size
>> 10);
3412 r
= sprintf(buf
, "%lu\n", tr
->entries
>> 10);
3413 mutex_unlock(&trace_types_lock
);
3415 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3419 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3420 size_t cnt
, loff_t
*ppos
)
3426 if (cnt
>= sizeof(buf
))
3429 if (copy_from_user(&buf
, ubuf
, cnt
))
3434 ret
= strict_strtoul(buf
, 10, &val
);
3438 /* must have at least 1 entry */
3442 mutex_lock(&trace_types_lock
);
3446 /* disable all cpu buffers */
3447 for_each_tracing_cpu(cpu
) {
3448 if (global_trace
.data
[cpu
])
3449 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
3450 if (max_tr
.data
[cpu
])
3451 atomic_inc(&max_tr
.data
[cpu
]->disabled
);
3454 /* value is in KB */
3457 if (val
!= global_trace
.entries
) {
3458 ret
= tracing_resize_ring_buffer(val
);
3467 /* If check pages failed, return ENOMEM */
3468 if (tracing_disabled
)
3471 for_each_tracing_cpu(cpu
) {
3472 if (global_trace
.data
[cpu
])
3473 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
3474 if (max_tr
.data
[cpu
])
3475 atomic_dec(&max_tr
.data
[cpu
]->disabled
);
3479 max_tr
.entries
= global_trace
.entries
;
3480 mutex_unlock(&trace_types_lock
);
3485 static int mark_printk(const char *fmt
, ...)
3489 va_start(args
, fmt
);
3490 ret
= trace_vprintk(0, fmt
, args
);
3496 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3497 size_t cnt
, loff_t
*fpos
)
3501 if (tracing_disabled
)
3504 if (cnt
> TRACE_BUF_SIZE
)
3505 cnt
= TRACE_BUF_SIZE
;
3507 buf
= kmalloc(cnt
+ 2, GFP_KERNEL
);
3511 if (copy_from_user(buf
, ubuf
, cnt
)) {
3515 if (buf
[cnt
-1] != '\n') {
3521 cnt
= mark_printk("%s", buf
);
3528 static int tracing_clock_show(struct seq_file
*m
, void *v
)
3532 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
3534 "%s%s%s%s", i
? " " : "",
3535 i
== trace_clock_id
? "[" : "", trace_clocks
[i
].name
,
3536 i
== trace_clock_id
? "]" : "");
3542 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
3543 size_t cnt
, loff_t
*fpos
)
3546 const char *clockstr
;
3549 if (cnt
>= sizeof(buf
))
3552 if (copy_from_user(&buf
, ubuf
, cnt
))
3557 clockstr
= strstrip(buf
);
3559 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
3560 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
3563 if (i
== ARRAY_SIZE(trace_clocks
))
3568 mutex_lock(&trace_types_lock
);
3570 ring_buffer_set_clock(global_trace
.buffer
, trace_clocks
[i
].func
);
3572 ring_buffer_set_clock(max_tr
.buffer
, trace_clocks
[i
].func
);
3574 mutex_unlock(&trace_types_lock
);
3581 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
3583 if (tracing_disabled
)
3585 return single_open(file
, tracing_clock_show
, NULL
);
3588 static const struct file_operations tracing_max_lat_fops
= {
3589 .open
= tracing_open_generic
,
3590 .read
= tracing_max_lat_read
,
3591 .write
= tracing_max_lat_write
,
3594 static const struct file_operations tracing_ctrl_fops
= {
3595 .open
= tracing_open_generic
,
3596 .read
= tracing_ctrl_read
,
3597 .write
= tracing_ctrl_write
,
3600 static const struct file_operations set_tracer_fops
= {
3601 .open
= tracing_open_generic
,
3602 .read
= tracing_set_trace_read
,
3603 .write
= tracing_set_trace_write
,
3606 static const struct file_operations tracing_pipe_fops
= {
3607 .open
= tracing_open_pipe
,
3608 .poll
= tracing_poll_pipe
,
3609 .read
= tracing_read_pipe
,
3610 .splice_read
= tracing_splice_read_pipe
,
3611 .release
= tracing_release_pipe
,
3614 static const struct file_operations tracing_entries_fops
= {
3615 .open
= tracing_open_generic
,
3616 .read
= tracing_entries_read
,
3617 .write
= tracing_entries_write
,
3620 static const struct file_operations tracing_mark_fops
= {
3621 .open
= tracing_open_generic
,
3622 .write
= tracing_mark_write
,
3625 static const struct file_operations trace_clock_fops
= {
3626 .open
= tracing_clock_open
,
3628 .llseek
= seq_lseek
,
3629 .release
= single_release
,
3630 .write
= tracing_clock_write
,
3633 struct ftrace_buffer_info
{
3634 struct trace_array
*tr
;
3640 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
3642 int cpu
= (int)(long)inode
->i_private
;
3643 struct ftrace_buffer_info
*info
;
3645 if (tracing_disabled
)
3648 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3652 info
->tr
= &global_trace
;
3655 /* Force reading ring buffer for first read */
3656 info
->read
= (unsigned int)-1;
3658 filp
->private_data
= info
;
3660 return nonseekable_open(inode
, filp
);
3664 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
3665 size_t count
, loff_t
*ppos
)
3667 struct ftrace_buffer_info
*info
= filp
->private_data
;
3676 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
);
3680 /* Do we have previous read data to read? */
3681 if (info
->read
< PAGE_SIZE
)
3686 trace_access_lock(info
->cpu
);
3687 ret
= ring_buffer_read_page(info
->tr
->buffer
,
3691 trace_access_unlock(info
->cpu
);
3695 pos
= ring_buffer_page_len(info
->spare
);
3697 if (pos
< PAGE_SIZE
)
3698 memset(info
->spare
+ pos
, 0, PAGE_SIZE
- pos
);
3701 size
= PAGE_SIZE
- info
->read
;
3705 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
3716 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
3718 struct ftrace_buffer_info
*info
= file
->private_data
;
3721 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
3728 struct ring_buffer
*buffer
;
3733 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
3734 struct pipe_buffer
*buf
)
3736 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3741 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3746 static int buffer_pipe_buf_steal(struct pipe_inode_info
*pipe
,
3747 struct pipe_buffer
*buf
)
3752 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
3753 struct pipe_buffer
*buf
)
3755 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3760 /* Pipe buffer operations for a buffer. */
3761 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
3763 .map
= generic_pipe_buf_map
,
3764 .unmap
= generic_pipe_buf_unmap
,
3765 .confirm
= generic_pipe_buf_confirm
,
3766 .release
= buffer_pipe_buf_release
,
3767 .steal
= buffer_pipe_buf_steal
,
3768 .get
= buffer_pipe_buf_get
,
3772 * Callback from splice_to_pipe(), if we need to release some pages
3773 * at the end of the spd in case we error'ed out in filling the pipe.
3775 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
3777 struct buffer_ref
*ref
=
3778 (struct buffer_ref
*)spd
->partial
[i
].private;
3783 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3785 spd
->partial
[i
].private = 0;
3789 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
3790 struct pipe_inode_info
*pipe
, size_t len
,
3793 struct ftrace_buffer_info
*info
= file
->private_data
;
3794 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3795 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3796 struct splice_pipe_desc spd
= {
3798 .partial
= partial_def
,
3800 .ops
= &buffer_pipe_buf_ops
,
3801 .spd_release
= buffer_spd_release
,
3803 struct buffer_ref
*ref
;
3804 int entries
, size
, i
;
3807 if (splice_grow_spd(pipe
, &spd
))
3810 if (*ppos
& (PAGE_SIZE
- 1)) {
3811 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3816 if (len
& (PAGE_SIZE
- 1)) {
3817 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3818 if (len
< PAGE_SIZE
) {
3825 trace_access_lock(info
->cpu
);
3826 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
3828 for (i
= 0; i
< pipe
->buffers
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
3832 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
3837 ref
->buffer
= info
->tr
->buffer
;
3838 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
);
3844 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
3847 ring_buffer_free_read_page(ref
->buffer
,
3854 * zero out any left over data, this is going to
3857 size
= ring_buffer_page_len(ref
->page
);
3858 if (size
< PAGE_SIZE
)
3859 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
3861 page
= virt_to_page(ref
->page
);
3863 spd
.pages
[i
] = page
;
3864 spd
.partial
[i
].len
= PAGE_SIZE
;
3865 spd
.partial
[i
].offset
= 0;
3866 spd
.partial
[i
].private = (unsigned long)ref
;
3870 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
3873 trace_access_unlock(info
->cpu
);
3876 /* did we read anything? */
3877 if (!spd
.nr_pages
) {
3878 if (flags
& SPLICE_F_NONBLOCK
)
3886 ret
= splice_to_pipe(pipe
, &spd
);
3887 splice_shrink_spd(pipe
, &spd
);
3892 static const struct file_operations tracing_buffers_fops
= {
3893 .open
= tracing_buffers_open
,
3894 .read
= tracing_buffers_read
,
3895 .release
= tracing_buffers_release
,
3896 .splice_read
= tracing_buffers_splice_read
,
3897 .llseek
= no_llseek
,
3901 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
3902 size_t count
, loff_t
*ppos
)
3904 unsigned long cpu
= (unsigned long)filp
->private_data
;
3905 struct trace_array
*tr
= &global_trace
;
3906 struct trace_seq
*s
;
3909 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
3915 cnt
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
3916 trace_seq_printf(s
, "entries: %ld\n", cnt
);
3918 cnt
= ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
3919 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
3921 cnt
= ring_buffer_commit_overrun_cpu(tr
->buffer
, cpu
);
3922 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
3924 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
3931 static const struct file_operations tracing_stats_fops
= {
3932 .open
= tracing_open_generic
,
3933 .read
= tracing_stats_read
,
3936 #ifdef CONFIG_DYNAMIC_FTRACE
3938 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
3944 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
3945 size_t cnt
, loff_t
*ppos
)
3947 static char ftrace_dyn_info_buffer
[1024];
3948 static DEFINE_MUTEX(dyn_info_mutex
);
3949 unsigned long *p
= filp
->private_data
;
3950 char *buf
= ftrace_dyn_info_buffer
;
3951 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
3954 mutex_lock(&dyn_info_mutex
);
3955 r
= sprintf(buf
, "%ld ", *p
);
3957 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
3960 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3962 mutex_unlock(&dyn_info_mutex
);
3967 static const struct file_operations tracing_dyn_info_fops
= {
3968 .open
= tracing_open_generic
,
3969 .read
= tracing_read_dyn_info
,
3973 static struct dentry
*d_tracer
;
3975 struct dentry
*tracing_init_dentry(void)
3982 if (!debugfs_initialized())
3985 d_tracer
= debugfs_create_dir("tracing", NULL
);
3987 if (!d_tracer
&& !once
) {
3989 pr_warning("Could not create debugfs directory 'tracing'\n");
3996 static struct dentry
*d_percpu
;
3998 struct dentry
*tracing_dentry_percpu(void)
4001 struct dentry
*d_tracer
;
4006 d_tracer
= tracing_init_dentry();
4011 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
4013 if (!d_percpu
&& !once
) {
4015 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4022 static void tracing_init_debugfs_percpu(long cpu
)
4024 struct dentry
*d_percpu
= tracing_dentry_percpu();
4025 struct dentry
*d_cpu
;
4026 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
4029 if (cpu
> 999 || cpu
< 0)
4032 sprintf(cpu_dir
, "cpu%ld", cpu
);
4033 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
4035 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
4039 /* per cpu trace_pipe */
4040 trace_create_file("trace_pipe", 0444, d_cpu
,
4041 (void *) cpu
, &tracing_pipe_fops
);
4044 trace_create_file("trace", 0644, d_cpu
,
4045 (void *) cpu
, &tracing_fops
);
4047 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
4048 (void *) cpu
, &tracing_buffers_fops
);
4050 trace_create_file("stats", 0444, d_cpu
,
4051 (void *) cpu
, &tracing_stats_fops
);
4054 #ifdef CONFIG_FTRACE_SELFTEST
4055 /* Let selftest have access to static functions in this file */
4056 #include "trace_selftest.c"
4059 struct trace_option_dentry
{
4060 struct tracer_opt
*opt
;
4061 struct tracer_flags
*flags
;
4062 struct dentry
*entry
;
4066 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4069 struct trace_option_dentry
*topt
= filp
->private_data
;
4072 if (topt
->flags
->val
& topt
->opt
->bit
)
4077 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4081 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4084 struct trace_option_dentry
*topt
= filp
->private_data
;
4089 if (cnt
>= sizeof(buf
))
4092 if (copy_from_user(&buf
, ubuf
, cnt
))
4097 ret
= strict_strtoul(buf
, 10, &val
);
4101 if (val
!= 0 && val
!= 1)
4104 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
4105 mutex_lock(&trace_types_lock
);
4106 ret
= __set_tracer_option(current_trace
, topt
->flags
,
4108 mutex_unlock(&trace_types_lock
);
4119 static const struct file_operations trace_options_fops
= {
4120 .open
= tracing_open_generic
,
4121 .read
= trace_options_read
,
4122 .write
= trace_options_write
,
4126 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4129 long index
= (long)filp
->private_data
;
4132 if (trace_flags
& (1 << index
))
4137 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4141 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4144 long index
= (long)filp
->private_data
;
4149 if (cnt
>= sizeof(buf
))
4152 if (copy_from_user(&buf
, ubuf
, cnt
))
4157 ret
= strict_strtoul(buf
, 10, &val
);
4161 if (val
!= 0 && val
!= 1)
4163 set_tracer_flags(1 << index
, val
);
4170 static const struct file_operations trace_options_core_fops
= {
4171 .open
= tracing_open_generic
,
4172 .read
= trace_options_core_read
,
4173 .write
= trace_options_core_write
,
4176 struct dentry
*trace_create_file(const char *name
,
4178 struct dentry
*parent
,
4180 const struct file_operations
*fops
)
4184 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
4186 pr_warning("Could not create debugfs '%s' entry\n", name
);
4192 static struct dentry
*trace_options_init_dentry(void)
4194 struct dentry
*d_tracer
;
4195 static struct dentry
*t_options
;
4200 d_tracer
= tracing_init_dentry();
4204 t_options
= debugfs_create_dir("options", d_tracer
);
4206 pr_warning("Could not create debugfs directory 'options'\n");
4214 create_trace_option_file(struct trace_option_dentry
*topt
,
4215 struct tracer_flags
*flags
,
4216 struct tracer_opt
*opt
)
4218 struct dentry
*t_options
;
4220 t_options
= trace_options_init_dentry();
4224 topt
->flags
= flags
;
4227 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
4228 &trace_options_fops
);
4232 static struct trace_option_dentry
*
4233 create_trace_option_files(struct tracer
*tracer
)
4235 struct trace_option_dentry
*topts
;
4236 struct tracer_flags
*flags
;
4237 struct tracer_opt
*opts
;
4243 flags
= tracer
->flags
;
4245 if (!flags
|| !flags
->opts
)
4250 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4253 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
4257 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4258 create_trace_option_file(&topts
[cnt
], flags
,
4265 destroy_trace_option_files(struct trace_option_dentry
*topts
)
4272 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
4273 if (topts
[cnt
].entry
)
4274 debugfs_remove(topts
[cnt
].entry
);
4280 static struct dentry
*
4281 create_trace_option_core_file(const char *option
, long index
)
4283 struct dentry
*t_options
;
4285 t_options
= trace_options_init_dentry();
4289 return trace_create_file(option
, 0644, t_options
, (void *)index
,
4290 &trace_options_core_fops
);
4293 static __init
void create_trace_options_dir(void)
4295 struct dentry
*t_options
;
4298 t_options
= trace_options_init_dentry();
4302 for (i
= 0; trace_options
[i
]; i
++)
4303 create_trace_option_core_file(trace_options
[i
], i
);
4306 static __init
int tracer_init_debugfs(void)
4308 struct dentry
*d_tracer
;
4311 trace_access_lock_init();
4313 d_tracer
= tracing_init_dentry();
4315 trace_create_file("tracing_enabled", 0644, d_tracer
,
4316 &global_trace
, &tracing_ctrl_fops
);
4318 trace_create_file("trace_options", 0644, d_tracer
,
4319 NULL
, &tracing_iter_fops
);
4321 trace_create_file("tracing_cpumask", 0644, d_tracer
,
4322 NULL
, &tracing_cpumask_fops
);
4324 trace_create_file("trace", 0644, d_tracer
,
4325 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
4327 trace_create_file("available_tracers", 0444, d_tracer
,
4328 &global_trace
, &show_traces_fops
);
4330 trace_create_file("current_tracer", 0644, d_tracer
,
4331 &global_trace
, &set_tracer_fops
);
4333 #ifdef CONFIG_TRACER_MAX_TRACE
4334 trace_create_file("tracing_max_latency", 0644, d_tracer
,
4335 &tracing_max_latency
, &tracing_max_lat_fops
);
4338 trace_create_file("tracing_thresh", 0644, d_tracer
,
4339 &tracing_thresh
, &tracing_max_lat_fops
);
4341 trace_create_file("README", 0444, d_tracer
,
4342 NULL
, &tracing_readme_fops
);
4344 trace_create_file("trace_pipe", 0444, d_tracer
,
4345 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
4347 trace_create_file("buffer_size_kb", 0644, d_tracer
,
4348 &global_trace
, &tracing_entries_fops
);
4350 trace_create_file("trace_marker", 0220, d_tracer
,
4351 NULL
, &tracing_mark_fops
);
4353 trace_create_file("saved_cmdlines", 0444, d_tracer
,
4354 NULL
, &tracing_saved_cmdlines_fops
);
4356 trace_create_file("trace_clock", 0644, d_tracer
, NULL
,
4359 #ifdef CONFIG_DYNAMIC_FTRACE
4360 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
4361 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
4363 #ifdef CONFIG_SYSPROF_TRACER
4364 init_tracer_sysprof_debugfs(d_tracer
);
4367 create_trace_options_dir();
4369 for_each_tracing_cpu(cpu
)
4370 tracing_init_debugfs_percpu(cpu
);
4375 static int trace_panic_handler(struct notifier_block
*this,
4376 unsigned long event
, void *unused
)
4378 if (ftrace_dump_on_oops
)
4379 ftrace_dump(ftrace_dump_on_oops
);
4383 static struct notifier_block trace_panic_notifier
= {
4384 .notifier_call
= trace_panic_handler
,
4386 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
4389 static int trace_die_handler(struct notifier_block
*self
,
4395 if (ftrace_dump_on_oops
)
4396 ftrace_dump(ftrace_dump_on_oops
);
4404 static struct notifier_block trace_die_notifier
= {
4405 .notifier_call
= trace_die_handler
,
4410 * printk is set to max of 1024, we really don't need it that big.
4411 * Nothing should be printing 1000 characters anyway.
4413 #define TRACE_MAX_PRINT 1000
4416 * Define here KERN_TRACE so that we have one place to modify
4417 * it if we decide to change what log level the ftrace dump
4420 #define KERN_TRACE KERN_EMERG
4423 trace_printk_seq(struct trace_seq
*s
)
4425 /* Probably should print a warning here. */
4429 /* should be zero ended, but we are paranoid. */
4430 s
->buffer
[s
->len
] = 0;
4432 printk(KERN_TRACE
"%s", s
->buffer
);
4438 __ftrace_dump(bool disable_tracing
, enum ftrace_dump_mode oops_dump_mode
)
4440 static arch_spinlock_t ftrace_dump_lock
=
4441 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
4442 /* use static because iter can be a bit big for the stack */
4443 static struct trace_iterator iter
;
4444 unsigned int old_userobj
;
4445 static int dump_ran
;
4446 unsigned long flags
;
4450 local_irq_save(flags
);
4451 arch_spin_lock(&ftrace_dump_lock
);
4459 if (disable_tracing
)
4462 for_each_tracing_cpu(cpu
) {
4463 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
4466 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
4468 /* don't look at user memory in panic mode */
4469 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
4471 /* Simulate the iterator */
4472 iter
.tr
= &global_trace
;
4473 iter
.trace
= current_trace
;
4475 switch (oops_dump_mode
) {
4477 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4480 iter
.cpu_file
= raw_smp_processor_id();
4485 printk(KERN_TRACE
"Bad dumping mode, switching to all CPUs dump\n");
4486 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4489 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
4492 * We need to stop all tracing on all CPUS to read the
4493 * the next buffer. This is a bit expensive, but is
4494 * not done often. We fill all what we can read,
4495 * and then release the locks again.
4498 while (!trace_empty(&iter
)) {
4501 printk(KERN_TRACE
"---------------------------------\n");
4505 /* reset all but tr, trace, and overruns */
4506 memset(&iter
.seq
, 0,
4507 sizeof(struct trace_iterator
) -
4508 offsetof(struct trace_iterator
, seq
));
4509 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
4512 if (find_next_entry_inc(&iter
) != NULL
) {
4515 ret
= print_trace_line(&iter
);
4516 if (ret
!= TRACE_TYPE_NO_CONSUME
)
4517 trace_consume(&iter
);
4520 trace_printk_seq(&iter
.seq
);
4524 printk(KERN_TRACE
" (ftrace buffer empty)\n");
4526 printk(KERN_TRACE
"---------------------------------\n");
4529 /* Re-enable tracing if requested */
4530 if (!disable_tracing
) {
4531 trace_flags
|= old_userobj
;
4533 for_each_tracing_cpu(cpu
) {
4534 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
4540 arch_spin_unlock(&ftrace_dump_lock
);
4541 local_irq_restore(flags
);
4544 /* By default: disable tracing after the dump */
4545 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode
)
4547 __ftrace_dump(true, oops_dump_mode
);
4550 __init
static int tracer_alloc_buffers(void)
4556 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
4559 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
4560 goto out_free_buffer_mask
;
4562 /* To save memory, keep the ring buffer size to its minimum */
4563 if (ring_buffer_expanded
)
4564 ring_buf_size
= trace_buf_size
;
4568 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
4569 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
4571 /* TODO: make the number of buffers hot pluggable with CPUS */
4572 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
,
4573 TRACE_BUFFER_FLAGS
);
4574 if (!global_trace
.buffer
) {
4575 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
4577 goto out_free_cpumask
;
4579 global_trace
.entries
= ring_buffer_size(global_trace
.buffer
);
4582 #ifdef CONFIG_TRACER_MAX_TRACE
4583 max_tr
.buffer
= ring_buffer_alloc(ring_buf_size
,
4584 TRACE_BUFFER_FLAGS
);
4585 if (!max_tr
.buffer
) {
4586 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
4588 ring_buffer_free(global_trace
.buffer
);
4589 goto out_free_cpumask
;
4591 max_tr
.entries
= ring_buffer_size(max_tr
.buffer
);
4592 WARN_ON(max_tr
.entries
!= global_trace
.entries
);
4595 /* Allocate the first page for all buffers */
4596 for_each_tracing_cpu(i
) {
4597 global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
4598 max_tr
.data
[i
] = &per_cpu(max_tr_data
, i
);
4601 trace_init_cmdlines();
4603 register_tracer(&nop_trace
);
4604 current_trace
= &nop_trace
;
4605 #ifdef CONFIG_BOOT_TRACER
4606 register_tracer(&boot_tracer
);
4608 /* All seems OK, enable tracing */
4609 tracing_disabled
= 0;
4611 atomic_notifier_chain_register(&panic_notifier_list
,
4612 &trace_panic_notifier
);
4614 register_die_notifier(&trace_die_notifier
);
4619 free_cpumask_var(tracing_cpumask
);
4620 out_free_buffer_mask
:
4621 free_cpumask_var(tracing_buffer_mask
);
4626 __init
static int clear_boot_tracer(void)
4629 * The default tracer at boot buffer is an init section.
4630 * This function is called in lateinit. If we did not
4631 * find the boot tracer, then clear it out, to prevent
4632 * later registration from accessing the buffer that is
4633 * about to be freed.
4635 if (!default_bootup_tracer
)
4638 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
4639 default_bootup_tracer
);
4640 default_bootup_tracer
= NULL
;
4645 early_initcall(tracer_alloc_buffers
);
4646 fs_initcall(tracer_init_debugfs
);
4647 late_initcall(clear_boot_tracer
);