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 * trace_vbprintk - write binary msg to tracing buffer
1338 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1340 static arch_spinlock_t trace_buf_lock
=
1341 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
1342 static u32 trace_buf
[TRACE_BUF_SIZE
];
1344 struct ftrace_event_call
*call
= &event_bprint
;
1345 struct ring_buffer_event
*event
;
1346 struct ring_buffer
*buffer
;
1347 struct trace_array
*tr
= &global_trace
;
1348 struct trace_array_cpu
*data
;
1349 struct bprint_entry
*entry
;
1350 unsigned long flags
;
1352 int cpu
, len
= 0, size
, pc
;
1354 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1357 /* Don't pollute graph traces with trace_vprintk internals */
1358 pause_graph_tracing();
1360 pc
= preempt_count();
1361 preempt_disable_notrace();
1362 cpu
= raw_smp_processor_id();
1363 data
= tr
->data
[cpu
];
1365 disable
= atomic_inc_return(&data
->disabled
);
1366 if (unlikely(disable
!= 1))
1369 /* Lockdep uses trace_printk for lock tracing */
1370 local_irq_save(flags
);
1371 arch_spin_lock(&trace_buf_lock
);
1372 len
= vbin_printf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1374 if (len
> TRACE_BUF_SIZE
|| len
< 0)
1377 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1378 buffer
= tr
->buffer
;
1379 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
1383 entry
= ring_buffer_event_data(event
);
1387 memcpy(entry
->buf
, trace_buf
, sizeof(u32
) * len
);
1388 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1389 ring_buffer_unlock_commit(buffer
, event
);
1390 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1394 arch_spin_unlock(&trace_buf_lock
);
1395 local_irq_restore(flags
);
1398 atomic_dec_return(&data
->disabled
);
1399 preempt_enable_notrace();
1400 unpause_graph_tracing();
1404 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1406 int trace_array_printk(struct trace_array
*tr
,
1407 unsigned long ip
, const char *fmt
, ...)
1412 if (!(trace_flags
& TRACE_ITER_PRINTK
))
1416 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
1421 int trace_array_vprintk(struct trace_array
*tr
,
1422 unsigned long ip
, const char *fmt
, va_list args
)
1424 static arch_spinlock_t trace_buf_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
1425 static char trace_buf
[TRACE_BUF_SIZE
];
1427 struct ftrace_event_call
*call
= &event_print
;
1428 struct ring_buffer_event
*event
;
1429 struct ring_buffer
*buffer
;
1430 struct trace_array_cpu
*data
;
1431 int cpu
, len
= 0, size
, pc
;
1432 struct print_entry
*entry
;
1433 unsigned long irq_flags
;
1436 if (tracing_disabled
|| tracing_selftest_running
)
1439 pc
= preempt_count();
1440 preempt_disable_notrace();
1441 cpu
= raw_smp_processor_id();
1442 data
= tr
->data
[cpu
];
1444 disable
= atomic_inc_return(&data
->disabled
);
1445 if (unlikely(disable
!= 1))
1448 pause_graph_tracing();
1449 raw_local_irq_save(irq_flags
);
1450 arch_spin_lock(&trace_buf_lock
);
1451 len
= vsnprintf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1453 size
= sizeof(*entry
) + len
+ 1;
1454 buffer
= tr
->buffer
;
1455 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
1459 entry
= ring_buffer_event_data(event
);
1462 memcpy(&entry
->buf
, trace_buf
, len
);
1463 entry
->buf
[len
] = '\0';
1464 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1465 ring_buffer_unlock_commit(buffer
, event
);
1466 ftrace_trace_stack(buffer
, irq_flags
, 6, pc
);
1470 arch_spin_unlock(&trace_buf_lock
);
1471 raw_local_irq_restore(irq_flags
);
1472 unpause_graph_tracing();
1474 atomic_dec_return(&data
->disabled
);
1475 preempt_enable_notrace();
1480 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1482 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
1484 EXPORT_SYMBOL_GPL(trace_vprintk
);
1486 enum trace_file_type
{
1487 TRACE_FILE_LAT_FMT
= 1,
1488 TRACE_FILE_ANNOTATE
= 2,
1491 static void trace_iterator_increment(struct trace_iterator
*iter
)
1493 /* Don't allow ftrace to trace into the ring buffers */
1494 ftrace_disable_cpu();
1497 if (iter
->buffer_iter
[iter
->cpu
])
1498 ring_buffer_read(iter
->buffer_iter
[iter
->cpu
], NULL
);
1500 ftrace_enable_cpu();
1503 static struct trace_entry
*
1504 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
,
1505 unsigned long *lost_events
)
1507 struct ring_buffer_event
*event
;
1508 struct ring_buffer_iter
*buf_iter
= iter
->buffer_iter
[cpu
];
1510 /* Don't allow ftrace to trace into the ring buffers */
1511 ftrace_disable_cpu();
1514 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1516 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
,
1519 ftrace_enable_cpu();
1521 return event
? ring_buffer_event_data(event
) : NULL
;
1524 static struct trace_entry
*
1525 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
,
1526 unsigned long *missing_events
, u64
*ent_ts
)
1528 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1529 struct trace_entry
*ent
, *next
= NULL
;
1530 unsigned long lost_events
= 0, next_lost
= 0;
1531 int cpu_file
= iter
->cpu_file
;
1532 u64 next_ts
= 0, ts
;
1537 * If we are in a per_cpu trace file, don't bother by iterating over
1538 * all cpu and peek directly.
1540 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1541 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1543 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
, missing_events
);
1545 *ent_cpu
= cpu_file
;
1550 for_each_tracing_cpu(cpu
) {
1552 if (ring_buffer_empty_cpu(buffer
, cpu
))
1555 ent
= peek_next_entry(iter
, cpu
, &ts
, &lost_events
);
1558 * Pick the entry with the smallest timestamp:
1560 if (ent
&& (!next
|| ts
< next_ts
)) {
1564 next_lost
= lost_events
;
1569 *ent_cpu
= next_cpu
;
1575 *missing_events
= next_lost
;
1580 /* Find the next real entry, without updating the iterator itself */
1581 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1582 int *ent_cpu
, u64
*ent_ts
)
1584 return __find_next_entry(iter
, ent_cpu
, NULL
, ent_ts
);
1587 /* Find the next real entry, and increment the iterator to the next entry */
1588 static void *find_next_entry_inc(struct trace_iterator
*iter
)
1590 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
,
1591 &iter
->lost_events
, &iter
->ts
);
1594 trace_iterator_increment(iter
);
1596 return iter
->ent
? iter
: NULL
;
1599 static void trace_consume(struct trace_iterator
*iter
)
1601 /* Don't allow ftrace to trace into the ring buffers */
1602 ftrace_disable_cpu();
1603 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
,
1604 &iter
->lost_events
);
1605 ftrace_enable_cpu();
1608 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1610 struct trace_iterator
*iter
= m
->private;
1614 WARN_ON_ONCE(iter
->leftover
);
1618 /* can't go backwards */
1623 ent
= find_next_entry_inc(iter
);
1627 while (ent
&& iter
->idx
< i
)
1628 ent
= find_next_entry_inc(iter
);
1635 static void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
1637 struct trace_array
*tr
= iter
->tr
;
1638 struct ring_buffer_event
*event
;
1639 struct ring_buffer_iter
*buf_iter
;
1640 unsigned long entries
= 0;
1643 tr
->data
[cpu
]->skipped_entries
= 0;
1645 if (!iter
->buffer_iter
[cpu
])
1648 buf_iter
= iter
->buffer_iter
[cpu
];
1649 ring_buffer_iter_reset(buf_iter
);
1652 * We could have the case with the max latency tracers
1653 * that a reset never took place on a cpu. This is evident
1654 * by the timestamp being before the start of the buffer.
1656 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
1657 if (ts
>= iter
->tr
->time_start
)
1660 ring_buffer_read(buf_iter
, NULL
);
1663 tr
->data
[cpu
]->skipped_entries
= entries
;
1667 * The current tracer is copied to avoid a global locking
1670 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1672 struct trace_iterator
*iter
= m
->private;
1673 static struct tracer
*old_tracer
;
1674 int cpu_file
= iter
->cpu_file
;
1679 /* copy the tracer to avoid using a global lock all around */
1680 mutex_lock(&trace_types_lock
);
1681 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1682 old_tracer
= current_trace
;
1683 *iter
->trace
= *current_trace
;
1685 mutex_unlock(&trace_types_lock
);
1687 atomic_inc(&trace_record_cmdline_disabled
);
1689 if (*pos
!= iter
->pos
) {
1694 ftrace_disable_cpu();
1696 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1697 for_each_tracing_cpu(cpu
)
1698 tracing_iter_reset(iter
, cpu
);
1700 tracing_iter_reset(iter
, cpu_file
);
1702 ftrace_enable_cpu();
1705 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1710 * If we overflowed the seq_file before, then we want
1711 * to just reuse the trace_seq buffer again.
1717 p
= s_next(m
, p
, &l
);
1721 trace_event_read_lock();
1722 trace_access_lock(cpu_file
);
1726 static void s_stop(struct seq_file
*m
, void *p
)
1728 struct trace_iterator
*iter
= m
->private;
1730 atomic_dec(&trace_record_cmdline_disabled
);
1731 trace_access_unlock(iter
->cpu_file
);
1732 trace_event_read_unlock();
1735 static void print_lat_help_header(struct seq_file
*m
)
1737 seq_puts(m
, "# _------=> CPU# \n");
1738 seq_puts(m
, "# / _-----=> irqs-off \n");
1739 seq_puts(m
, "# | / _----=> need-resched \n");
1740 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1741 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1742 seq_puts(m
, "# |||| /_--=> lock-depth \n");
1743 seq_puts(m
, "# |||||/ delay \n");
1744 seq_puts(m
, "# cmd pid |||||| time | caller \n");
1745 seq_puts(m
, "# \\ / |||||| \\ | / \n");
1748 static void print_func_help_header(struct seq_file
*m
)
1750 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1751 seq_puts(m
, "# | | | | |\n");
1756 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1758 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1759 struct trace_array
*tr
= iter
->tr
;
1760 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1761 struct tracer
*type
= current_trace
;
1762 unsigned long entries
= 0;
1763 unsigned long total
= 0;
1764 unsigned long count
;
1765 const char *name
= "preemption";
1772 for_each_tracing_cpu(cpu
) {
1773 count
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
1775 * If this buffer has skipped entries, then we hold all
1776 * entries for the trace and we need to ignore the
1777 * ones before the time stamp.
1779 if (tr
->data
[cpu
]->skipped_entries
) {
1780 count
-= tr
->data
[cpu
]->skipped_entries
;
1781 /* total is the same as the entries */
1785 ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
1789 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
1791 seq_puts(m
, "# -----------------------------------"
1792 "---------------------------------\n");
1793 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1794 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1795 nsecs_to_usecs(data
->saved_latency
),
1799 #if defined(CONFIG_PREEMPT_NONE)
1801 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1803 #elif defined(CONFIG_PREEMPT)
1808 /* These are reserved for later use */
1811 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1815 seq_puts(m
, "# -----------------\n");
1816 seq_printf(m
, "# | task: %.16s-%d "
1817 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1818 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1819 data
->policy
, data
->rt_priority
);
1820 seq_puts(m
, "# -----------------\n");
1822 if (data
->critical_start
) {
1823 seq_puts(m
, "# => started at: ");
1824 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1825 trace_print_seq(m
, &iter
->seq
);
1826 seq_puts(m
, "\n# => ended at: ");
1827 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1828 trace_print_seq(m
, &iter
->seq
);
1829 seq_puts(m
, "\n#\n");
1835 static void test_cpu_buff_start(struct trace_iterator
*iter
)
1837 struct trace_seq
*s
= &iter
->seq
;
1839 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
1842 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
1845 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
1848 if (iter
->tr
->data
[iter
->cpu
]->skipped_entries
)
1851 cpumask_set_cpu(iter
->cpu
, iter
->started
);
1853 /* Don't print started cpu buffer for the first entry of the trace */
1855 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
1859 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
1861 struct trace_seq
*s
= &iter
->seq
;
1862 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1863 struct trace_entry
*entry
;
1864 struct trace_event
*event
;
1868 test_cpu_buff_start(iter
);
1870 event
= ftrace_find_event(entry
->type
);
1872 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1873 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1874 if (!trace_print_lat_context(iter
))
1877 if (!trace_print_context(iter
))
1883 return event
->funcs
->trace(iter
, sym_flags
, event
);
1885 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
1888 return TRACE_TYPE_HANDLED
;
1890 return TRACE_TYPE_PARTIAL_LINE
;
1893 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
1895 struct trace_seq
*s
= &iter
->seq
;
1896 struct trace_entry
*entry
;
1897 struct trace_event
*event
;
1901 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1902 if (!trace_seq_printf(s
, "%d %d %llu ",
1903 entry
->pid
, iter
->cpu
, iter
->ts
))
1907 event
= ftrace_find_event(entry
->type
);
1909 return event
->funcs
->raw(iter
, 0, event
);
1911 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
1914 return TRACE_TYPE_HANDLED
;
1916 return TRACE_TYPE_PARTIAL_LINE
;
1919 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
1921 struct trace_seq
*s
= &iter
->seq
;
1922 unsigned char newline
= '\n';
1923 struct trace_entry
*entry
;
1924 struct trace_event
*event
;
1928 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1929 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1930 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1931 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
1934 event
= ftrace_find_event(entry
->type
);
1936 enum print_line_t ret
= event
->funcs
->hex(iter
, 0, event
);
1937 if (ret
!= TRACE_TYPE_HANDLED
)
1941 SEQ_PUT_FIELD_RET(s
, newline
);
1943 return TRACE_TYPE_HANDLED
;
1946 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
1948 struct trace_seq
*s
= &iter
->seq
;
1949 struct trace_entry
*entry
;
1950 struct trace_event
*event
;
1954 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1955 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1956 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
1957 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
1960 event
= ftrace_find_event(entry
->type
);
1961 return event
? event
->funcs
->binary(iter
, 0, event
) :
1965 int trace_empty(struct trace_iterator
*iter
)
1969 /* If we are looking at one CPU buffer, only check that one */
1970 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
1971 cpu
= iter
->cpu_file
;
1972 if (iter
->buffer_iter
[cpu
]) {
1973 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1976 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1982 for_each_tracing_cpu(cpu
) {
1983 if (iter
->buffer_iter
[cpu
]) {
1984 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
1987 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
1995 /* Called with trace_event_read_lock() held. */
1996 static enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
1998 enum print_line_t ret
;
2000 if (iter
->lost_events
)
2001 trace_seq_printf(&iter
->seq
, "CPU:%d [LOST %lu EVENTS]\n",
2002 iter
->cpu
, iter
->lost_events
);
2004 if (iter
->trace
&& iter
->trace
->print_line
) {
2005 ret
= iter
->trace
->print_line(iter
);
2006 if (ret
!= TRACE_TYPE_UNHANDLED
)
2010 if (iter
->ent
->type
== TRACE_BPRINT
&&
2011 trace_flags
& TRACE_ITER_PRINTK
&&
2012 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2013 return trace_print_bprintk_msg_only(iter
);
2015 if (iter
->ent
->type
== TRACE_PRINT
&&
2016 trace_flags
& TRACE_ITER_PRINTK
&&
2017 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2018 return trace_print_printk_msg_only(iter
);
2020 if (trace_flags
& TRACE_ITER_BIN
)
2021 return print_bin_fmt(iter
);
2023 if (trace_flags
& TRACE_ITER_HEX
)
2024 return print_hex_fmt(iter
);
2026 if (trace_flags
& TRACE_ITER_RAW
)
2027 return print_raw_fmt(iter
);
2029 return print_trace_fmt(iter
);
2032 void trace_default_header(struct seq_file
*m
)
2034 struct trace_iterator
*iter
= m
->private;
2036 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2037 /* print nothing if the buffers are empty */
2038 if (trace_empty(iter
))
2040 print_trace_header(m
, iter
);
2041 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2042 print_lat_help_header(m
);
2044 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2045 print_func_help_header(m
);
2049 static int s_show(struct seq_file
*m
, void *v
)
2051 struct trace_iterator
*iter
= v
;
2054 if (iter
->ent
== NULL
) {
2056 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
2059 if (iter
->trace
&& iter
->trace
->print_header
)
2060 iter
->trace
->print_header(m
);
2062 trace_default_header(m
);
2064 } else if (iter
->leftover
) {
2066 * If we filled the seq_file buffer earlier, we
2067 * want to just show it now.
2069 ret
= trace_print_seq(m
, &iter
->seq
);
2071 /* ret should this time be zero, but you never know */
2072 iter
->leftover
= ret
;
2075 print_trace_line(iter
);
2076 ret
= trace_print_seq(m
, &iter
->seq
);
2078 * If we overflow the seq_file buffer, then it will
2079 * ask us for this data again at start up.
2081 * ret is 0 if seq_file write succeeded.
2084 iter
->leftover
= ret
;
2090 static const struct seq_operations tracer_seq_ops
= {
2097 static struct trace_iterator
*
2098 __tracing_open(struct inode
*inode
, struct file
*file
)
2100 long cpu_file
= (long) inode
->i_private
;
2101 void *fail_ret
= ERR_PTR(-ENOMEM
);
2102 struct trace_iterator
*iter
;
2106 if (tracing_disabled
)
2107 return ERR_PTR(-ENODEV
);
2109 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2111 return ERR_PTR(-ENOMEM
);
2114 * We make a copy of the current tracer to avoid concurrent
2115 * changes on it while we are reading.
2117 mutex_lock(&trace_types_lock
);
2118 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2123 *iter
->trace
= *current_trace
;
2125 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2128 if (current_trace
&& current_trace
->print_max
)
2131 iter
->tr
= &global_trace
;
2133 mutex_init(&iter
->mutex
);
2134 iter
->cpu_file
= cpu_file
;
2136 /* Notify the tracer early; before we stop tracing. */
2137 if (iter
->trace
&& iter
->trace
->open
)
2138 iter
->trace
->open(iter
);
2140 /* Annotate start of buffers if we had overruns */
2141 if (ring_buffer_overruns(iter
->tr
->buffer
))
2142 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2144 /* stop the trace while dumping */
2147 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
2148 for_each_tracing_cpu(cpu
) {
2149 iter
->buffer_iter
[cpu
] =
2150 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2152 ring_buffer_read_prepare_sync();
2153 for_each_tracing_cpu(cpu
) {
2154 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2155 tracing_iter_reset(iter
, cpu
);
2158 cpu
= iter
->cpu_file
;
2159 iter
->buffer_iter
[cpu
] =
2160 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2161 ring_buffer_read_prepare_sync();
2162 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2163 tracing_iter_reset(iter
, cpu
);
2166 ret
= seq_open(file
, &tracer_seq_ops
);
2168 fail_ret
= ERR_PTR(ret
);
2172 m
= file
->private_data
;
2175 mutex_unlock(&trace_types_lock
);
2180 for_each_tracing_cpu(cpu
) {
2181 if (iter
->buffer_iter
[cpu
])
2182 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2184 free_cpumask_var(iter
->started
);
2187 mutex_unlock(&trace_types_lock
);
2194 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2196 if (tracing_disabled
)
2199 filp
->private_data
= inode
->i_private
;
2203 static int tracing_release(struct inode
*inode
, struct file
*file
)
2205 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
2206 struct trace_iterator
*iter
;
2209 if (!(file
->f_mode
& FMODE_READ
))
2214 mutex_lock(&trace_types_lock
);
2215 for_each_tracing_cpu(cpu
) {
2216 if (iter
->buffer_iter
[cpu
])
2217 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2220 if (iter
->trace
&& iter
->trace
->close
)
2221 iter
->trace
->close(iter
);
2223 /* reenable tracing if it was previously enabled */
2225 mutex_unlock(&trace_types_lock
);
2227 seq_release(inode
, file
);
2228 mutex_destroy(&iter
->mutex
);
2229 free_cpumask_var(iter
->started
);
2235 static int tracing_open(struct inode
*inode
, struct file
*file
)
2237 struct trace_iterator
*iter
;
2240 /* If this file was open for write, then erase contents */
2241 if ((file
->f_mode
& FMODE_WRITE
) &&
2242 (file
->f_flags
& O_TRUNC
)) {
2243 long cpu
= (long) inode
->i_private
;
2245 if (cpu
== TRACE_PIPE_ALL_CPU
)
2246 tracing_reset_online_cpus(&global_trace
);
2248 tracing_reset(&global_trace
, cpu
);
2251 if (file
->f_mode
& FMODE_READ
) {
2252 iter
= __tracing_open(inode
, file
);
2254 ret
= PTR_ERR(iter
);
2255 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2256 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2262 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2264 struct tracer
*t
= v
;
2274 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2279 mutex_lock(&trace_types_lock
);
2280 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2286 static void t_stop(struct seq_file
*m
, void *p
)
2288 mutex_unlock(&trace_types_lock
);
2291 static int t_show(struct seq_file
*m
, void *v
)
2293 struct tracer
*t
= v
;
2298 seq_printf(m
, "%s", t
->name
);
2307 static const struct seq_operations show_traces_seq_ops
= {
2314 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2316 if (tracing_disabled
)
2319 return seq_open(file
, &show_traces_seq_ops
);
2323 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2324 size_t count
, loff_t
*ppos
)
2329 static const struct file_operations tracing_fops
= {
2330 .open
= tracing_open
,
2332 .write
= tracing_write_stub
,
2333 .llseek
= seq_lseek
,
2334 .release
= tracing_release
,
2337 static const struct file_operations show_traces_fops
= {
2338 .open
= show_traces_open
,
2340 .release
= seq_release
,
2344 * Only trace on a CPU if the bitmask is set:
2346 static cpumask_var_t tracing_cpumask
;
2349 * The tracer itself will not take this lock, but still we want
2350 * to provide a consistent cpumask to user-space:
2352 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2355 * Temporary storage for the character representation of the
2356 * CPU bitmask (and one more byte for the newline):
2358 static char mask_str
[NR_CPUS
+ 1];
2361 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2362 size_t count
, loff_t
*ppos
)
2366 mutex_lock(&tracing_cpumask_update_lock
);
2368 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2369 if (count
- len
< 2) {
2373 len
+= sprintf(mask_str
+ len
, "\n");
2374 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2377 mutex_unlock(&tracing_cpumask_update_lock
);
2383 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2384 size_t count
, loff_t
*ppos
)
2387 cpumask_var_t tracing_cpumask_new
;
2389 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2392 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2396 mutex_lock(&tracing_cpumask_update_lock
);
2398 local_irq_disable();
2399 arch_spin_lock(&ftrace_max_lock
);
2400 for_each_tracing_cpu(cpu
) {
2402 * Increase/decrease the disabled counter if we are
2403 * about to flip a bit in the cpumask:
2405 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2406 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2407 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2409 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2410 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2411 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2414 arch_spin_unlock(&ftrace_max_lock
);
2417 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2419 mutex_unlock(&tracing_cpumask_update_lock
);
2420 free_cpumask_var(tracing_cpumask_new
);
2425 free_cpumask_var(tracing_cpumask_new
);
2430 static const struct file_operations tracing_cpumask_fops
= {
2431 .open
= tracing_open_generic
,
2432 .read
= tracing_cpumask_read
,
2433 .write
= tracing_cpumask_write
,
2436 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
2438 struct tracer_opt
*trace_opts
;
2442 mutex_lock(&trace_types_lock
);
2443 tracer_flags
= current_trace
->flags
->val
;
2444 trace_opts
= current_trace
->flags
->opts
;
2446 for (i
= 0; trace_options
[i
]; i
++) {
2447 if (trace_flags
& (1 << i
))
2448 seq_printf(m
, "%s\n", trace_options
[i
]);
2450 seq_printf(m
, "no%s\n", trace_options
[i
]);
2453 for (i
= 0; trace_opts
[i
].name
; i
++) {
2454 if (tracer_flags
& trace_opts
[i
].bit
)
2455 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
2457 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
2459 mutex_unlock(&trace_types_lock
);
2464 static int __set_tracer_option(struct tracer
*trace
,
2465 struct tracer_flags
*tracer_flags
,
2466 struct tracer_opt
*opts
, int neg
)
2470 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
2475 tracer_flags
->val
&= ~opts
->bit
;
2477 tracer_flags
->val
|= opts
->bit
;
2481 /* Try to assign a tracer specific option */
2482 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2484 struct tracer_flags
*tracer_flags
= trace
->flags
;
2485 struct tracer_opt
*opts
= NULL
;
2488 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
2489 opts
= &tracer_flags
->opts
[i
];
2491 if (strcmp(cmp
, opts
->name
) == 0)
2492 return __set_tracer_option(trace
, trace
->flags
,
2499 static void set_tracer_flags(unsigned int mask
, int enabled
)
2501 /* do nothing if flag is already set */
2502 if (!!(trace_flags
& mask
) == !!enabled
)
2506 trace_flags
|= mask
;
2508 trace_flags
&= ~mask
;
2512 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2513 size_t cnt
, loff_t
*ppos
)
2521 if (cnt
>= sizeof(buf
))
2524 if (copy_from_user(&buf
, ubuf
, cnt
))
2528 cmp
= strstrip(buf
);
2530 if (strncmp(cmp
, "no", 2) == 0) {
2535 for (i
= 0; trace_options
[i
]; i
++) {
2536 if (strcmp(cmp
, trace_options
[i
]) == 0) {
2537 set_tracer_flags(1 << i
, !neg
);
2542 /* If no option could be set, test the specific tracer options */
2543 if (!trace_options
[i
]) {
2544 mutex_lock(&trace_types_lock
);
2545 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2546 mutex_unlock(&trace_types_lock
);
2556 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
2558 if (tracing_disabled
)
2560 return single_open(file
, tracing_trace_options_show
, NULL
);
2563 static const struct file_operations tracing_iter_fops
= {
2564 .open
= tracing_trace_options_open
,
2566 .llseek
= seq_lseek
,
2567 .release
= single_release
,
2568 .write
= tracing_trace_options_write
,
2571 static const char readme_msg
[] =
2572 "tracing mini-HOWTO:\n\n"
2573 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2574 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2575 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2576 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2578 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2579 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2581 "# cat /sys/kernel/debug/tracing/trace_options\n"
2582 "noprint-parent nosym-offset nosym-addr noverbose\n"
2583 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2584 "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2585 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2586 "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2590 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2591 size_t cnt
, loff_t
*ppos
)
2593 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2594 readme_msg
, strlen(readme_msg
));
2597 static const struct file_operations tracing_readme_fops
= {
2598 .open
= tracing_open_generic
,
2599 .read
= tracing_readme_read
,
2603 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
2604 size_t cnt
, loff_t
*ppos
)
2613 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
2617 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
2625 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
2628 pid
= map_cmdline_to_pid
[i
];
2629 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
2632 trace_find_cmdline(pid
, buf_comm
);
2633 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
2638 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
2647 static const struct file_operations tracing_saved_cmdlines_fops
= {
2648 .open
= tracing_open_generic
,
2649 .read
= tracing_saved_cmdlines_read
,
2653 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2654 size_t cnt
, loff_t
*ppos
)
2659 r
= sprintf(buf
, "%u\n", tracer_enabled
);
2660 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2664 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2665 size_t cnt
, loff_t
*ppos
)
2667 struct trace_array
*tr
= filp
->private_data
;
2672 if (cnt
>= sizeof(buf
))
2675 if (copy_from_user(&buf
, ubuf
, cnt
))
2680 ret
= strict_strtoul(buf
, 10, &val
);
2686 mutex_lock(&trace_types_lock
);
2687 if (tracer_enabled
^ val
) {
2690 if (current_trace
->start
)
2691 current_trace
->start(tr
);
2696 if (current_trace
->stop
)
2697 current_trace
->stop(tr
);
2700 mutex_unlock(&trace_types_lock
);
2708 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2709 size_t cnt
, loff_t
*ppos
)
2711 char buf
[MAX_TRACER_SIZE
+2];
2714 mutex_lock(&trace_types_lock
);
2716 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2718 r
= sprintf(buf
, "\n");
2719 mutex_unlock(&trace_types_lock
);
2721 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2724 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
2726 tracing_reset_online_cpus(tr
);
2730 static int tracing_resize_ring_buffer(unsigned long size
)
2735 * If kernel or user changes the size of the ring buffer
2736 * we use the size that was given, and we can forget about
2737 * expanding it later.
2739 ring_buffer_expanded
= 1;
2741 ret
= ring_buffer_resize(global_trace
.buffer
, size
);
2745 ret
= ring_buffer_resize(max_tr
.buffer
, size
);
2749 r
= ring_buffer_resize(global_trace
.buffer
,
2750 global_trace
.entries
);
2753 * AARGH! We are left with different
2754 * size max buffer!!!!
2755 * The max buffer is our "snapshot" buffer.
2756 * When a tracer needs a snapshot (one of the
2757 * latency tracers), it swaps the max buffer
2758 * with the saved snap shot. We succeeded to
2759 * update the size of the main buffer, but failed to
2760 * update the size of the max buffer. But when we tried
2761 * to reset the main buffer to the original size, we
2762 * failed there too. This is very unlikely to
2763 * happen, but if it does, warn and kill all
2767 tracing_disabled
= 1;
2772 global_trace
.entries
= size
;
2778 * tracing_update_buffers - used by tracing facility to expand ring buffers
2780 * To save on memory when the tracing is never used on a system with it
2781 * configured in. The ring buffers are set to a minimum size. But once
2782 * a user starts to use the tracing facility, then they need to grow
2783 * to their default size.
2785 * This function is to be called when a tracer is about to be used.
2787 int tracing_update_buffers(void)
2791 mutex_lock(&trace_types_lock
);
2792 if (!ring_buffer_expanded
)
2793 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2794 mutex_unlock(&trace_types_lock
);
2799 struct trace_option_dentry
;
2801 static struct trace_option_dentry
*
2802 create_trace_option_files(struct tracer
*tracer
);
2805 destroy_trace_option_files(struct trace_option_dentry
*topts
);
2807 static int tracing_set_tracer(const char *buf
)
2809 static struct trace_option_dentry
*topts
;
2810 struct trace_array
*tr
= &global_trace
;
2814 mutex_lock(&trace_types_lock
);
2816 if (!ring_buffer_expanded
) {
2817 ret
= tracing_resize_ring_buffer(trace_buf_size
);
2823 for (t
= trace_types
; t
; t
= t
->next
) {
2824 if (strcmp(t
->name
, buf
) == 0)
2831 if (t
== current_trace
)
2834 trace_branch_disable();
2835 if (current_trace
&& current_trace
->reset
)
2836 current_trace
->reset(tr
);
2838 destroy_trace_option_files(topts
);
2842 topts
= create_trace_option_files(current_trace
);
2845 ret
= tracer_init(t
, tr
);
2850 trace_branch_enable(tr
);
2852 mutex_unlock(&trace_types_lock
);
2858 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2859 size_t cnt
, loff_t
*ppos
)
2861 char buf
[MAX_TRACER_SIZE
+1];
2868 if (cnt
> MAX_TRACER_SIZE
)
2869 cnt
= MAX_TRACER_SIZE
;
2871 if (copy_from_user(&buf
, ubuf
, cnt
))
2876 /* strip ending whitespace. */
2877 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2880 err
= tracing_set_tracer(buf
);
2890 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2891 size_t cnt
, loff_t
*ppos
)
2893 unsigned long *ptr
= filp
->private_data
;
2897 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
2898 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2899 if (r
> sizeof(buf
))
2901 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2905 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2906 size_t cnt
, loff_t
*ppos
)
2908 unsigned long *ptr
= filp
->private_data
;
2913 if (cnt
>= sizeof(buf
))
2916 if (copy_from_user(&buf
, ubuf
, cnt
))
2921 ret
= strict_strtoul(buf
, 10, &val
);
2930 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2932 long cpu_file
= (long) inode
->i_private
;
2933 struct trace_iterator
*iter
;
2936 if (tracing_disabled
)
2939 mutex_lock(&trace_types_lock
);
2941 /* create a buffer to store the information to pass to userspace */
2942 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2949 * We make a copy of the current tracer to avoid concurrent
2950 * changes on it while we are reading.
2952 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2958 *iter
->trace
= *current_trace
;
2960 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
2965 /* trace pipe does not show start of buffer */
2966 cpumask_setall(iter
->started
);
2968 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2969 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2971 iter
->cpu_file
= cpu_file
;
2972 iter
->tr
= &global_trace
;
2973 mutex_init(&iter
->mutex
);
2974 filp
->private_data
= iter
;
2976 if (iter
->trace
->pipe_open
)
2977 iter
->trace
->pipe_open(iter
);
2980 mutex_unlock(&trace_types_lock
);
2986 mutex_unlock(&trace_types_lock
);
2990 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
2992 struct trace_iterator
*iter
= file
->private_data
;
2994 mutex_lock(&trace_types_lock
);
2996 if (iter
->trace
->pipe_close
)
2997 iter
->trace
->pipe_close(iter
);
2999 mutex_unlock(&trace_types_lock
);
3001 free_cpumask_var(iter
->started
);
3002 mutex_destroy(&iter
->mutex
);
3010 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
3012 struct trace_iterator
*iter
= filp
->private_data
;
3014 if (trace_flags
& TRACE_ITER_BLOCK
) {
3016 * Always select as readable when in blocking mode
3018 return POLLIN
| POLLRDNORM
;
3020 if (!trace_empty(iter
))
3021 return POLLIN
| POLLRDNORM
;
3022 poll_wait(filp
, &trace_wait
, poll_table
);
3023 if (!trace_empty(iter
))
3024 return POLLIN
| POLLRDNORM
;
3031 void default_wait_pipe(struct trace_iterator
*iter
)
3035 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
3037 if (trace_empty(iter
))
3040 finish_wait(&trace_wait
, &wait
);
3044 * This is a make-shift waitqueue.
3045 * A tracer might use this callback on some rare cases:
3047 * 1) the current tracer might hold the runqueue lock when it wakes up
3048 * a reader, hence a deadlock (sched, function, and function graph tracers)
3049 * 2) the function tracers, trace all functions, we don't want
3050 * the overhead of calling wake_up and friends
3051 * (and tracing them too)
3053 * Anyway, this is really very primitive wakeup.
3055 void poll_wait_pipe(struct trace_iterator
*iter
)
3057 set_current_state(TASK_INTERRUPTIBLE
);
3058 /* sleep for 100 msecs, and try again. */
3059 schedule_timeout(HZ
/ 10);
3062 /* Must be called with trace_types_lock mutex held. */
3063 static int tracing_wait_pipe(struct file
*filp
)
3065 struct trace_iterator
*iter
= filp
->private_data
;
3067 while (trace_empty(iter
)) {
3069 if ((filp
->f_flags
& O_NONBLOCK
)) {
3073 mutex_unlock(&iter
->mutex
);
3075 iter
->trace
->wait_pipe(iter
);
3077 mutex_lock(&iter
->mutex
);
3079 if (signal_pending(current
))
3083 * We block until we read something and tracing is disabled.
3084 * We still block if tracing is disabled, but we have never
3085 * read anything. This allows a user to cat this file, and
3086 * then enable tracing. But after we have read something,
3087 * we give an EOF when tracing is again disabled.
3089 * iter->pos will be 0 if we haven't read anything.
3091 if (!tracer_enabled
&& iter
->pos
)
3102 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
3103 size_t cnt
, loff_t
*ppos
)
3105 struct trace_iterator
*iter
= filp
->private_data
;
3106 static struct tracer
*old_tracer
;
3109 /* return any leftover data */
3110 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3114 trace_seq_init(&iter
->seq
);
3116 /* copy the tracer to avoid using a global lock all around */
3117 mutex_lock(&trace_types_lock
);
3118 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3119 old_tracer
= current_trace
;
3120 *iter
->trace
= *current_trace
;
3122 mutex_unlock(&trace_types_lock
);
3125 * Avoid more than one consumer on a single file descriptor
3126 * This is just a matter of traces coherency, the ring buffer itself
3129 mutex_lock(&iter
->mutex
);
3130 if (iter
->trace
->read
) {
3131 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
3137 sret
= tracing_wait_pipe(filp
);
3141 /* stop when tracing is finished */
3142 if (trace_empty(iter
)) {
3147 if (cnt
>= PAGE_SIZE
)
3148 cnt
= PAGE_SIZE
- 1;
3150 /* reset all but tr, trace, and overruns */
3151 memset(&iter
->seq
, 0,
3152 sizeof(struct trace_iterator
) -
3153 offsetof(struct trace_iterator
, seq
));
3156 trace_event_read_lock();
3157 trace_access_lock(iter
->cpu_file
);
3158 while (find_next_entry_inc(iter
) != NULL
) {
3159 enum print_line_t ret
;
3160 int len
= iter
->seq
.len
;
3162 ret
= print_trace_line(iter
);
3163 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3164 /* don't print partial lines */
3165 iter
->seq
.len
= len
;
3168 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3169 trace_consume(iter
);
3171 if (iter
->seq
.len
>= cnt
)
3174 trace_access_unlock(iter
->cpu_file
);
3175 trace_event_read_unlock();
3177 /* Now copy what we have to the user */
3178 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3179 if (iter
->seq
.readpos
>= iter
->seq
.len
)
3180 trace_seq_init(&iter
->seq
);
3183 * If there was nothing to send to user, inspite of consuming trace
3184 * entries, go back to wait for more entries.
3190 mutex_unlock(&iter
->mutex
);
3195 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
3196 struct pipe_buffer
*buf
)
3198 __free_page(buf
->page
);
3201 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
3204 __free_page(spd
->pages
[idx
]);
3207 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
3209 .map
= generic_pipe_buf_map
,
3210 .unmap
= generic_pipe_buf_unmap
,
3211 .confirm
= generic_pipe_buf_confirm
,
3212 .release
= tracing_pipe_buf_release
,
3213 .steal
= generic_pipe_buf_steal
,
3214 .get
= generic_pipe_buf_get
,
3218 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
3223 /* Seq buffer is page-sized, exactly what we need. */
3225 count
= iter
->seq
.len
;
3226 ret
= print_trace_line(iter
);
3227 count
= iter
->seq
.len
- count
;
3230 iter
->seq
.len
-= count
;
3233 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3234 iter
->seq
.len
-= count
;
3238 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3239 trace_consume(iter
);
3241 if (!find_next_entry_inc(iter
)) {
3251 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3253 struct pipe_inode_info
*pipe
,
3257 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3258 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3259 struct trace_iterator
*iter
= filp
->private_data
;
3260 struct splice_pipe_desc spd
= {
3262 .partial
= partial_def
,
3263 .nr_pages
= 0, /* This gets updated below. */
3265 .ops
= &tracing_pipe_buf_ops
,
3266 .spd_release
= tracing_spd_release_pipe
,
3268 static struct tracer
*old_tracer
;
3273 if (splice_grow_spd(pipe
, &spd
))
3276 /* copy the tracer to avoid using a global lock all around */
3277 mutex_lock(&trace_types_lock
);
3278 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3279 old_tracer
= current_trace
;
3280 *iter
->trace
= *current_trace
;
3282 mutex_unlock(&trace_types_lock
);
3284 mutex_lock(&iter
->mutex
);
3286 if (iter
->trace
->splice_read
) {
3287 ret
= iter
->trace
->splice_read(iter
, filp
,
3288 ppos
, pipe
, len
, flags
);
3293 ret
= tracing_wait_pipe(filp
);
3297 if (!iter
->ent
&& !find_next_entry_inc(iter
)) {
3302 trace_event_read_lock();
3303 trace_access_lock(iter
->cpu_file
);
3305 /* Fill as many pages as possible. */
3306 for (i
= 0, rem
= len
; i
< pipe
->buffers
&& rem
; i
++) {
3307 spd
.pages
[i
] = alloc_page(GFP_KERNEL
);
3311 rem
= tracing_fill_pipe_page(rem
, iter
);
3313 /* Copy the data into the page, so we can start over. */
3314 ret
= trace_seq_to_buffer(&iter
->seq
,
3315 page_address(spd
.pages
[i
]),
3318 __free_page(spd
.pages
[i
]);
3321 spd
.partial
[i
].offset
= 0;
3322 spd
.partial
[i
].len
= iter
->seq
.len
;
3324 trace_seq_init(&iter
->seq
);
3327 trace_access_unlock(iter
->cpu_file
);
3328 trace_event_read_unlock();
3329 mutex_unlock(&iter
->mutex
);
3333 ret
= splice_to_pipe(pipe
, &spd
);
3335 splice_shrink_spd(pipe
, &spd
);
3339 mutex_unlock(&iter
->mutex
);
3344 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3345 size_t cnt
, loff_t
*ppos
)
3347 struct trace_array
*tr
= filp
->private_data
;
3351 mutex_lock(&trace_types_lock
);
3352 if (!ring_buffer_expanded
)
3353 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3355 trace_buf_size
>> 10);
3357 r
= sprintf(buf
, "%lu\n", tr
->entries
>> 10);
3358 mutex_unlock(&trace_types_lock
);
3360 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3364 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3365 size_t cnt
, loff_t
*ppos
)
3371 if (cnt
>= sizeof(buf
))
3374 if (copy_from_user(&buf
, ubuf
, cnt
))
3379 ret
= strict_strtoul(buf
, 10, &val
);
3383 /* must have at least 1 entry */
3387 mutex_lock(&trace_types_lock
);
3391 /* disable all cpu buffers */
3392 for_each_tracing_cpu(cpu
) {
3393 if (global_trace
.data
[cpu
])
3394 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
3395 if (max_tr
.data
[cpu
])
3396 atomic_inc(&max_tr
.data
[cpu
]->disabled
);
3399 /* value is in KB */
3402 if (val
!= global_trace
.entries
) {
3403 ret
= tracing_resize_ring_buffer(val
);
3412 /* If check pages failed, return ENOMEM */
3413 if (tracing_disabled
)
3416 for_each_tracing_cpu(cpu
) {
3417 if (global_trace
.data
[cpu
])
3418 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
3419 if (max_tr
.data
[cpu
])
3420 atomic_dec(&max_tr
.data
[cpu
]->disabled
);
3424 max_tr
.entries
= global_trace
.entries
;
3425 mutex_unlock(&trace_types_lock
);
3430 static int mark_printk(const char *fmt
, ...)
3434 va_start(args
, fmt
);
3435 ret
= trace_vprintk(0, fmt
, args
);
3441 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3442 size_t cnt
, loff_t
*fpos
)
3446 if (tracing_disabled
)
3449 if (cnt
> TRACE_BUF_SIZE
)
3450 cnt
= TRACE_BUF_SIZE
;
3452 buf
= kmalloc(cnt
+ 2, GFP_KERNEL
);
3456 if (copy_from_user(buf
, ubuf
, cnt
)) {
3460 if (buf
[cnt
-1] != '\n') {
3466 cnt
= mark_printk("%s", buf
);
3473 static int tracing_clock_show(struct seq_file
*m
, void *v
)
3477 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
3479 "%s%s%s%s", i
? " " : "",
3480 i
== trace_clock_id
? "[" : "", trace_clocks
[i
].name
,
3481 i
== trace_clock_id
? "]" : "");
3487 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
3488 size_t cnt
, loff_t
*fpos
)
3491 const char *clockstr
;
3494 if (cnt
>= sizeof(buf
))
3497 if (copy_from_user(&buf
, ubuf
, cnt
))
3502 clockstr
= strstrip(buf
);
3504 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
3505 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
3508 if (i
== ARRAY_SIZE(trace_clocks
))
3513 mutex_lock(&trace_types_lock
);
3515 ring_buffer_set_clock(global_trace
.buffer
, trace_clocks
[i
].func
);
3517 ring_buffer_set_clock(max_tr
.buffer
, trace_clocks
[i
].func
);
3519 mutex_unlock(&trace_types_lock
);
3526 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
3528 if (tracing_disabled
)
3530 return single_open(file
, tracing_clock_show
, NULL
);
3533 static const struct file_operations tracing_max_lat_fops
= {
3534 .open
= tracing_open_generic
,
3535 .read
= tracing_max_lat_read
,
3536 .write
= tracing_max_lat_write
,
3539 static const struct file_operations tracing_ctrl_fops
= {
3540 .open
= tracing_open_generic
,
3541 .read
= tracing_ctrl_read
,
3542 .write
= tracing_ctrl_write
,
3545 static const struct file_operations set_tracer_fops
= {
3546 .open
= tracing_open_generic
,
3547 .read
= tracing_set_trace_read
,
3548 .write
= tracing_set_trace_write
,
3551 static const struct file_operations tracing_pipe_fops
= {
3552 .open
= tracing_open_pipe
,
3553 .poll
= tracing_poll_pipe
,
3554 .read
= tracing_read_pipe
,
3555 .splice_read
= tracing_splice_read_pipe
,
3556 .release
= tracing_release_pipe
,
3559 static const struct file_operations tracing_entries_fops
= {
3560 .open
= tracing_open_generic
,
3561 .read
= tracing_entries_read
,
3562 .write
= tracing_entries_write
,
3565 static const struct file_operations tracing_mark_fops
= {
3566 .open
= tracing_open_generic
,
3567 .write
= tracing_mark_write
,
3570 static const struct file_operations trace_clock_fops
= {
3571 .open
= tracing_clock_open
,
3573 .llseek
= seq_lseek
,
3574 .release
= single_release
,
3575 .write
= tracing_clock_write
,
3578 struct ftrace_buffer_info
{
3579 struct trace_array
*tr
;
3585 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
3587 int cpu
= (int)(long)inode
->i_private
;
3588 struct ftrace_buffer_info
*info
;
3590 if (tracing_disabled
)
3593 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3597 info
->tr
= &global_trace
;
3600 /* Force reading ring buffer for first read */
3601 info
->read
= (unsigned int)-1;
3603 filp
->private_data
= info
;
3605 return nonseekable_open(inode
, filp
);
3609 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
3610 size_t count
, loff_t
*ppos
)
3612 struct ftrace_buffer_info
*info
= filp
->private_data
;
3620 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
);
3624 /* Do we have previous read data to read? */
3625 if (info
->read
< PAGE_SIZE
)
3630 trace_access_lock(info
->cpu
);
3631 ret
= ring_buffer_read_page(info
->tr
->buffer
,
3635 trace_access_unlock(info
->cpu
);
3640 size
= PAGE_SIZE
- info
->read
;
3644 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
3655 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
3657 struct ftrace_buffer_info
*info
= file
->private_data
;
3660 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
3667 struct ring_buffer
*buffer
;
3672 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
3673 struct pipe_buffer
*buf
)
3675 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3680 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3685 static int buffer_pipe_buf_steal(struct pipe_inode_info
*pipe
,
3686 struct pipe_buffer
*buf
)
3691 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
3692 struct pipe_buffer
*buf
)
3694 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3699 /* Pipe buffer operations for a buffer. */
3700 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
3702 .map
= generic_pipe_buf_map
,
3703 .unmap
= generic_pipe_buf_unmap
,
3704 .confirm
= generic_pipe_buf_confirm
,
3705 .release
= buffer_pipe_buf_release
,
3706 .steal
= buffer_pipe_buf_steal
,
3707 .get
= buffer_pipe_buf_get
,
3711 * Callback from splice_to_pipe(), if we need to release some pages
3712 * at the end of the spd in case we error'ed out in filling the pipe.
3714 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
3716 struct buffer_ref
*ref
=
3717 (struct buffer_ref
*)spd
->partial
[i
].private;
3722 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3724 spd
->partial
[i
].private = 0;
3728 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
3729 struct pipe_inode_info
*pipe
, size_t len
,
3732 struct ftrace_buffer_info
*info
= file
->private_data
;
3733 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3734 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3735 struct splice_pipe_desc spd
= {
3737 .partial
= partial_def
,
3739 .ops
= &buffer_pipe_buf_ops
,
3740 .spd_release
= buffer_spd_release
,
3742 struct buffer_ref
*ref
;
3743 int entries
, size
, i
;
3746 if (splice_grow_spd(pipe
, &spd
))
3749 if (*ppos
& (PAGE_SIZE
- 1)) {
3750 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3755 if (len
& (PAGE_SIZE
- 1)) {
3756 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3757 if (len
< PAGE_SIZE
) {
3764 trace_access_lock(info
->cpu
);
3765 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
3767 for (i
= 0; i
< pipe
->buffers
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
3771 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
3776 ref
->buffer
= info
->tr
->buffer
;
3777 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
);
3783 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
3786 ring_buffer_free_read_page(ref
->buffer
,
3793 * zero out any left over data, this is going to
3796 size
= ring_buffer_page_len(ref
->page
);
3797 if (size
< PAGE_SIZE
)
3798 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
3800 page
= virt_to_page(ref
->page
);
3802 spd
.pages
[i
] = page
;
3803 spd
.partial
[i
].len
= PAGE_SIZE
;
3804 spd
.partial
[i
].offset
= 0;
3805 spd
.partial
[i
].private = (unsigned long)ref
;
3809 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
3812 trace_access_unlock(info
->cpu
);
3815 /* did we read anything? */
3816 if (!spd
.nr_pages
) {
3817 if (flags
& SPLICE_F_NONBLOCK
)
3825 ret
= splice_to_pipe(pipe
, &spd
);
3826 splice_shrink_spd(pipe
, &spd
);
3831 static const struct file_operations tracing_buffers_fops
= {
3832 .open
= tracing_buffers_open
,
3833 .read
= tracing_buffers_read
,
3834 .release
= tracing_buffers_release
,
3835 .splice_read
= tracing_buffers_splice_read
,
3836 .llseek
= no_llseek
,
3840 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
3841 size_t count
, loff_t
*ppos
)
3843 unsigned long cpu
= (unsigned long)filp
->private_data
;
3844 struct trace_array
*tr
= &global_trace
;
3845 struct trace_seq
*s
;
3848 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
3854 cnt
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
3855 trace_seq_printf(s
, "entries: %ld\n", cnt
);
3857 cnt
= ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
3858 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
3860 cnt
= ring_buffer_commit_overrun_cpu(tr
->buffer
, cpu
);
3861 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
3863 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
3870 static const struct file_operations tracing_stats_fops
= {
3871 .open
= tracing_open_generic
,
3872 .read
= tracing_stats_read
,
3875 #ifdef CONFIG_DYNAMIC_FTRACE
3877 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
3883 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
3884 size_t cnt
, loff_t
*ppos
)
3886 static char ftrace_dyn_info_buffer
[1024];
3887 static DEFINE_MUTEX(dyn_info_mutex
);
3888 unsigned long *p
= filp
->private_data
;
3889 char *buf
= ftrace_dyn_info_buffer
;
3890 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
3893 mutex_lock(&dyn_info_mutex
);
3894 r
= sprintf(buf
, "%ld ", *p
);
3896 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
3899 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3901 mutex_unlock(&dyn_info_mutex
);
3906 static const struct file_operations tracing_dyn_info_fops
= {
3907 .open
= tracing_open_generic
,
3908 .read
= tracing_read_dyn_info
,
3912 static struct dentry
*d_tracer
;
3914 struct dentry
*tracing_init_dentry(void)
3921 if (!debugfs_initialized())
3924 d_tracer
= debugfs_create_dir("tracing", NULL
);
3926 if (!d_tracer
&& !once
) {
3928 pr_warning("Could not create debugfs directory 'tracing'\n");
3935 static struct dentry
*d_percpu
;
3937 struct dentry
*tracing_dentry_percpu(void)
3940 struct dentry
*d_tracer
;
3945 d_tracer
= tracing_init_dentry();
3950 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
3952 if (!d_percpu
&& !once
) {
3954 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3961 static void tracing_init_debugfs_percpu(long cpu
)
3963 struct dentry
*d_percpu
= tracing_dentry_percpu();
3964 struct dentry
*d_cpu
;
3965 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3968 if (cpu
> 999 || cpu
< 0)
3971 sprintf(cpu_dir
, "cpu%ld", cpu
);
3972 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
3974 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
3978 /* per cpu trace_pipe */
3979 trace_create_file("trace_pipe", 0444, d_cpu
,
3980 (void *) cpu
, &tracing_pipe_fops
);
3983 trace_create_file("trace", 0644, d_cpu
,
3984 (void *) cpu
, &tracing_fops
);
3986 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
3987 (void *) cpu
, &tracing_buffers_fops
);
3989 trace_create_file("stats", 0444, d_cpu
,
3990 (void *) cpu
, &tracing_stats_fops
);
3993 #ifdef CONFIG_FTRACE_SELFTEST
3994 /* Let selftest have access to static functions in this file */
3995 #include "trace_selftest.c"
3998 struct trace_option_dentry
{
3999 struct tracer_opt
*opt
;
4000 struct tracer_flags
*flags
;
4001 struct dentry
*entry
;
4005 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4008 struct trace_option_dentry
*topt
= filp
->private_data
;
4011 if (topt
->flags
->val
& topt
->opt
->bit
)
4016 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4020 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4023 struct trace_option_dentry
*topt
= filp
->private_data
;
4028 if (cnt
>= sizeof(buf
))
4031 if (copy_from_user(&buf
, ubuf
, cnt
))
4036 ret
= strict_strtoul(buf
, 10, &val
);
4040 if (val
!= 0 && val
!= 1)
4043 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
4044 mutex_lock(&trace_types_lock
);
4045 ret
= __set_tracer_option(current_trace
, topt
->flags
,
4047 mutex_unlock(&trace_types_lock
);
4058 static const struct file_operations trace_options_fops
= {
4059 .open
= tracing_open_generic
,
4060 .read
= trace_options_read
,
4061 .write
= trace_options_write
,
4065 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4068 long index
= (long)filp
->private_data
;
4071 if (trace_flags
& (1 << index
))
4076 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4080 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4083 long index
= (long)filp
->private_data
;
4088 if (cnt
>= sizeof(buf
))
4091 if (copy_from_user(&buf
, ubuf
, cnt
))
4096 ret
= strict_strtoul(buf
, 10, &val
);
4100 if (val
!= 0 && val
!= 1)
4102 set_tracer_flags(1 << index
, val
);
4109 static const struct file_operations trace_options_core_fops
= {
4110 .open
= tracing_open_generic
,
4111 .read
= trace_options_core_read
,
4112 .write
= trace_options_core_write
,
4115 struct dentry
*trace_create_file(const char *name
,
4117 struct dentry
*parent
,
4119 const struct file_operations
*fops
)
4123 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
4125 pr_warning("Could not create debugfs '%s' entry\n", name
);
4131 static struct dentry
*trace_options_init_dentry(void)
4133 struct dentry
*d_tracer
;
4134 static struct dentry
*t_options
;
4139 d_tracer
= tracing_init_dentry();
4143 t_options
= debugfs_create_dir("options", d_tracer
);
4145 pr_warning("Could not create debugfs directory 'options'\n");
4153 create_trace_option_file(struct trace_option_dentry
*topt
,
4154 struct tracer_flags
*flags
,
4155 struct tracer_opt
*opt
)
4157 struct dentry
*t_options
;
4159 t_options
= trace_options_init_dentry();
4163 topt
->flags
= flags
;
4166 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
4167 &trace_options_fops
);
4171 static struct trace_option_dentry
*
4172 create_trace_option_files(struct tracer
*tracer
)
4174 struct trace_option_dentry
*topts
;
4175 struct tracer_flags
*flags
;
4176 struct tracer_opt
*opts
;
4182 flags
= tracer
->flags
;
4184 if (!flags
|| !flags
->opts
)
4189 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4192 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
4196 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4197 create_trace_option_file(&topts
[cnt
], flags
,
4204 destroy_trace_option_files(struct trace_option_dentry
*topts
)
4211 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
4212 if (topts
[cnt
].entry
)
4213 debugfs_remove(topts
[cnt
].entry
);
4219 static struct dentry
*
4220 create_trace_option_core_file(const char *option
, long index
)
4222 struct dentry
*t_options
;
4224 t_options
= trace_options_init_dentry();
4228 return trace_create_file(option
, 0644, t_options
, (void *)index
,
4229 &trace_options_core_fops
);
4232 static __init
void create_trace_options_dir(void)
4234 struct dentry
*t_options
;
4237 t_options
= trace_options_init_dentry();
4241 for (i
= 0; trace_options
[i
]; i
++)
4242 create_trace_option_core_file(trace_options
[i
], i
);
4245 static __init
int tracer_init_debugfs(void)
4247 struct dentry
*d_tracer
;
4250 trace_access_lock_init();
4252 d_tracer
= tracing_init_dentry();
4254 trace_create_file("tracing_enabled", 0644, d_tracer
,
4255 &global_trace
, &tracing_ctrl_fops
);
4257 trace_create_file("trace_options", 0644, d_tracer
,
4258 NULL
, &tracing_iter_fops
);
4260 trace_create_file("tracing_cpumask", 0644, d_tracer
,
4261 NULL
, &tracing_cpumask_fops
);
4263 trace_create_file("trace", 0644, d_tracer
,
4264 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
4266 trace_create_file("available_tracers", 0444, d_tracer
,
4267 &global_trace
, &show_traces_fops
);
4269 trace_create_file("current_tracer", 0644, d_tracer
,
4270 &global_trace
, &set_tracer_fops
);
4272 #ifdef CONFIG_TRACER_MAX_TRACE
4273 trace_create_file("tracing_max_latency", 0644, d_tracer
,
4274 &tracing_max_latency
, &tracing_max_lat_fops
);
4277 trace_create_file("tracing_thresh", 0644, d_tracer
,
4278 &tracing_thresh
, &tracing_max_lat_fops
);
4280 trace_create_file("README", 0444, d_tracer
,
4281 NULL
, &tracing_readme_fops
);
4283 trace_create_file("trace_pipe", 0444, d_tracer
,
4284 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
4286 trace_create_file("buffer_size_kb", 0644, d_tracer
,
4287 &global_trace
, &tracing_entries_fops
);
4289 trace_create_file("trace_marker", 0220, d_tracer
,
4290 NULL
, &tracing_mark_fops
);
4292 trace_create_file("saved_cmdlines", 0444, d_tracer
,
4293 NULL
, &tracing_saved_cmdlines_fops
);
4295 trace_create_file("trace_clock", 0644, d_tracer
, NULL
,
4298 #ifdef CONFIG_DYNAMIC_FTRACE
4299 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
4300 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
4303 create_trace_options_dir();
4305 for_each_tracing_cpu(cpu
)
4306 tracing_init_debugfs_percpu(cpu
);
4311 static int trace_panic_handler(struct notifier_block
*this,
4312 unsigned long event
, void *unused
)
4314 if (ftrace_dump_on_oops
)
4315 ftrace_dump(ftrace_dump_on_oops
);
4319 static struct notifier_block trace_panic_notifier
= {
4320 .notifier_call
= trace_panic_handler
,
4322 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
4325 static int trace_die_handler(struct notifier_block
*self
,
4331 if (ftrace_dump_on_oops
)
4332 ftrace_dump(ftrace_dump_on_oops
);
4340 static struct notifier_block trace_die_notifier
= {
4341 .notifier_call
= trace_die_handler
,
4346 * printk is set to max of 1024, we really don't need it that big.
4347 * Nothing should be printing 1000 characters anyway.
4349 #define TRACE_MAX_PRINT 1000
4352 * Define here KERN_TRACE so that we have one place to modify
4353 * it if we decide to change what log level the ftrace dump
4356 #define KERN_TRACE KERN_EMERG
4359 trace_printk_seq(struct trace_seq
*s
)
4361 /* Probably should print a warning here. */
4365 /* should be zero ended, but we are paranoid. */
4366 s
->buffer
[s
->len
] = 0;
4368 printk(KERN_TRACE
"%s", s
->buffer
);
4374 __ftrace_dump(bool disable_tracing
, enum ftrace_dump_mode oops_dump_mode
)
4376 static arch_spinlock_t ftrace_dump_lock
=
4377 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
4378 /* use static because iter can be a bit big for the stack */
4379 static struct trace_iterator iter
;
4380 unsigned int old_userobj
;
4381 static int dump_ran
;
4382 unsigned long flags
;
4386 local_irq_save(flags
);
4387 arch_spin_lock(&ftrace_dump_lock
);
4395 if (disable_tracing
)
4398 for_each_tracing_cpu(cpu
) {
4399 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
4402 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
4404 /* don't look at user memory in panic mode */
4405 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
4407 /* Simulate the iterator */
4408 iter
.tr
= &global_trace
;
4409 iter
.trace
= current_trace
;
4411 switch (oops_dump_mode
) {
4413 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4416 iter
.cpu_file
= raw_smp_processor_id();
4421 printk(KERN_TRACE
"Bad dumping mode, switching to all CPUs dump\n");
4422 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4425 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
4428 * We need to stop all tracing on all CPUS to read the
4429 * the next buffer. This is a bit expensive, but is
4430 * not done often. We fill all what we can read,
4431 * and then release the locks again.
4434 while (!trace_empty(&iter
)) {
4437 printk(KERN_TRACE
"---------------------------------\n");
4441 /* reset all but tr, trace, and overruns */
4442 memset(&iter
.seq
, 0,
4443 sizeof(struct trace_iterator
) -
4444 offsetof(struct trace_iterator
, seq
));
4445 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
4448 if (find_next_entry_inc(&iter
) != NULL
) {
4451 ret
= print_trace_line(&iter
);
4452 if (ret
!= TRACE_TYPE_NO_CONSUME
)
4453 trace_consume(&iter
);
4456 trace_printk_seq(&iter
.seq
);
4460 printk(KERN_TRACE
" (ftrace buffer empty)\n");
4462 printk(KERN_TRACE
"---------------------------------\n");
4465 /* Re-enable tracing if requested */
4466 if (!disable_tracing
) {
4467 trace_flags
|= old_userobj
;
4469 for_each_tracing_cpu(cpu
) {
4470 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
4476 arch_spin_unlock(&ftrace_dump_lock
);
4477 local_irq_restore(flags
);
4480 /* By default: disable tracing after the dump */
4481 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode
)
4483 __ftrace_dump(true, oops_dump_mode
);
4486 __init
static int tracer_alloc_buffers(void)
4492 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
4495 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
4496 goto out_free_buffer_mask
;
4498 /* To save memory, keep the ring buffer size to its minimum */
4499 if (ring_buffer_expanded
)
4500 ring_buf_size
= trace_buf_size
;
4504 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
4505 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
4507 /* TODO: make the number of buffers hot pluggable with CPUS */
4508 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
,
4509 TRACE_BUFFER_FLAGS
);
4510 if (!global_trace
.buffer
) {
4511 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
4513 goto out_free_cpumask
;
4515 global_trace
.entries
= ring_buffer_size(global_trace
.buffer
);
4518 #ifdef CONFIG_TRACER_MAX_TRACE
4519 max_tr
.buffer
= ring_buffer_alloc(ring_buf_size
,
4520 TRACE_BUFFER_FLAGS
);
4521 if (!max_tr
.buffer
) {
4522 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
4524 ring_buffer_free(global_trace
.buffer
);
4525 goto out_free_cpumask
;
4527 max_tr
.entries
= ring_buffer_size(max_tr
.buffer
);
4528 WARN_ON(max_tr
.entries
!= global_trace
.entries
);
4531 /* Allocate the first page for all buffers */
4532 for_each_tracing_cpu(i
) {
4533 global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
4534 max_tr
.data
[i
] = &per_cpu(max_tr_data
, i
);
4537 trace_init_cmdlines();
4539 register_tracer(&nop_trace
);
4540 current_trace
= &nop_trace
;
4541 /* All seems OK, enable tracing */
4542 tracing_disabled
= 0;
4544 atomic_notifier_chain_register(&panic_notifier_list
,
4545 &trace_panic_notifier
);
4547 register_die_notifier(&trace_die_notifier
);
4552 free_cpumask_var(tracing_cpumask
);
4553 out_free_buffer_mask
:
4554 free_cpumask_var(tracing_buffer_mask
);
4559 __init
static int clear_boot_tracer(void)
4562 * The default tracer at boot buffer is an init section.
4563 * This function is called in lateinit. If we did not
4564 * find the boot tracer, then clear it out, to prevent
4565 * later registration from accessing the buffer that is
4566 * about to be freed.
4568 if (!default_bootup_tracer
)
4571 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
4572 default_bootup_tracer
);
4573 default_bootup_tracer
= NULL
;
4578 early_initcall(tracer_alloc_buffers
);
4579 fs_initcall(tracer_init_debugfs
);
4580 late_initcall(clear_boot_tracer
);