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/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
42 #include "trace_output.h"
45 * On boot up, the ring buffer is set to the minimum size, so that
46 * we do not waste memory on systems that are not using tracing.
48 int ring_buffer_expanded
;
51 * We need to change this state when a selftest is running.
52 * A selftest will lurk into the ring-buffer to count the
53 * entries inserted during the selftest although some concurrent
54 * insertions into the ring-buffer such as trace_printk could occurred
55 * at the same time, giving false positive or negative results.
57 static bool __read_mostly tracing_selftest_running
;
60 * If a tracer is running, we do not want to run SELFTEST.
62 bool __read_mostly tracing_selftest_disabled
;
64 /* For tracers that don't implement custom flags */
65 static struct tracer_opt dummy_tracer_opt
[] = {
69 static struct tracer_flags dummy_tracer_flags
= {
71 .opts
= dummy_tracer_opt
74 static int dummy_set_flag(u32 old_flags
, u32 bit
, int set
)
80 * Kill all tracing for good (never come back).
81 * It is initialized to 1 but will turn to zero if the initialization
82 * of the tracer is successful. But that is the only place that sets
85 static int tracing_disabled
= 1;
87 DEFINE_PER_CPU(int, ftrace_cpu_disabled
);
89 static inline void ftrace_disable_cpu(void)
92 __this_cpu_inc(ftrace_cpu_disabled
);
95 static inline void ftrace_enable_cpu(void)
97 __this_cpu_dec(ftrace_cpu_disabled
);
101 cpumask_var_t __read_mostly tracing_buffer_mask
;
104 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
106 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
107 * is set, then ftrace_dump is called. This will output the contents
108 * of the ftrace buffers to the console. This is very useful for
109 * capturing traces that lead to crashes and outputing it to a
112 * It is default off, but you can enable it with either specifying
113 * "ftrace_dump_on_oops" in the kernel command line, or setting
114 * /proc/sys/kernel/ftrace_dump_on_oops
115 * Set 1 if you want to dump buffers of all CPUs
116 * Set 2 if you want to dump the buffer of the CPU that triggered oops
119 enum ftrace_dump_mode ftrace_dump_on_oops
;
121 static int tracing_set_tracer(const char *buf
);
123 #define MAX_TRACER_SIZE 100
124 static char bootup_tracer_buf
[MAX_TRACER_SIZE
] __initdata
;
125 static char *default_bootup_tracer
;
127 static int __init
set_cmdline_ftrace(char *str
)
129 strncpy(bootup_tracer_buf
, str
, MAX_TRACER_SIZE
);
130 default_bootup_tracer
= bootup_tracer_buf
;
131 /* We are using ftrace early, expand it */
132 ring_buffer_expanded
= 1;
135 __setup("ftrace=", set_cmdline_ftrace
);
137 static int __init
set_ftrace_dump_on_oops(char *str
)
139 if (*str
++ != '=' || !*str
) {
140 ftrace_dump_on_oops
= DUMP_ALL
;
144 if (!strcmp("orig_cpu", str
)) {
145 ftrace_dump_on_oops
= DUMP_ORIG
;
151 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops
);
153 unsigned long long ns2usecs(cycle_t nsec
)
161 * The global_trace is the descriptor that holds the tracing
162 * buffers for the live tracing. For each CPU, it contains
163 * a link list of pages that will store trace entries. The
164 * page descriptor of the pages in the memory is used to hold
165 * the link list by linking the lru item in the page descriptor
166 * to each of the pages in the buffer per CPU.
168 * For each active CPU there is a data field that holds the
169 * pages for the buffer for that CPU. Each CPU has the same number
170 * of pages allocated for its buffer.
172 static struct trace_array global_trace
;
174 static DEFINE_PER_CPU(struct trace_array_cpu
, global_trace_cpu
);
176 int filter_current_check_discard(struct ring_buffer
*buffer
,
177 struct ftrace_event_call
*call
, void *rec
,
178 struct ring_buffer_event
*event
)
180 return filter_check_discard(call
, rec
, buffer
, event
);
182 EXPORT_SYMBOL_GPL(filter_current_check_discard
);
184 cycle_t
ftrace_now(int cpu
)
188 /* Early boot up does not have a buffer yet */
189 if (!global_trace
.buffer
)
190 return trace_clock_local();
192 ts
= ring_buffer_time_stamp(global_trace
.buffer
, cpu
);
193 ring_buffer_normalize_time_stamp(global_trace
.buffer
, cpu
, &ts
);
199 * The max_tr is used to snapshot the global_trace when a maximum
200 * latency is reached. Some tracers will use this to store a maximum
201 * trace while it continues examining live traces.
203 * The buffers for the max_tr are set up the same as the global_trace.
204 * When a snapshot is taken, the link list of the max_tr is swapped
205 * with the link list of the global_trace and the buffers are reset for
206 * the global_trace so the tracing can continue.
208 static struct trace_array max_tr
;
210 static DEFINE_PER_CPU(struct trace_array_cpu
, max_tr_data
);
212 /* tracer_enabled is used to toggle activation of a tracer */
213 static int tracer_enabled
= 1;
216 * tracing_is_enabled - return tracer_enabled status
218 * This function is used by other tracers to know the status
219 * of the tracer_enabled flag. Tracers may use this function
220 * to know if it should enable their features when starting
221 * up. See irqsoff tracer for an example (start_irqsoff_tracer).
223 int tracing_is_enabled(void)
225 return tracer_enabled
;
229 * trace_buf_size is the size in bytes that is allocated
230 * for a buffer. Note, the number of bytes is always rounded
233 * This number is purposely set to a low number of 16384.
234 * If the dump on oops happens, it will be much appreciated
235 * to not have to wait for all that output. Anyway this can be
236 * boot time and run time configurable.
238 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
240 static unsigned long trace_buf_size
= TRACE_BUF_SIZE_DEFAULT
;
242 /* trace_types holds a link list of available tracers. */
243 static struct tracer
*trace_types __read_mostly
;
245 /* current_trace points to the tracer that is currently active */
246 static struct tracer
*current_trace __read_mostly
;
249 * trace_types_lock is used to protect the trace_types list.
251 static DEFINE_MUTEX(trace_types_lock
);
254 * serialize the access of the ring buffer
256 * ring buffer serializes readers, but it is low level protection.
257 * The validity of the events (which returns by ring_buffer_peek() ..etc)
258 * are not protected by ring buffer.
260 * The content of events may become garbage if we allow other process consumes
261 * these events concurrently:
262 * A) the page of the consumed events may become a normal page
263 * (not reader page) in ring buffer, and this page will be rewrited
264 * by events producer.
265 * B) The page of the consumed events may become a page for splice_read,
266 * and this page will be returned to system.
268 * These primitives allow multi process access to different cpu ring buffer
271 * These primitives don't distinguish read-only and read-consume access.
272 * Multi read-only access are also serialized.
276 static DECLARE_RWSEM(all_cpu_access_lock
);
277 static DEFINE_PER_CPU(struct mutex
, cpu_access_lock
);
279 static inline void trace_access_lock(int cpu
)
281 if (cpu
== TRACE_PIPE_ALL_CPU
) {
282 /* gain it for accessing the whole ring buffer. */
283 down_write(&all_cpu_access_lock
);
285 /* gain it for accessing a cpu ring buffer. */
287 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
288 down_read(&all_cpu_access_lock
);
290 /* Secondly block other access to this @cpu ring buffer. */
291 mutex_lock(&per_cpu(cpu_access_lock
, cpu
));
295 static inline void trace_access_unlock(int cpu
)
297 if (cpu
== TRACE_PIPE_ALL_CPU
) {
298 up_write(&all_cpu_access_lock
);
300 mutex_unlock(&per_cpu(cpu_access_lock
, cpu
));
301 up_read(&all_cpu_access_lock
);
305 static inline void trace_access_lock_init(void)
309 for_each_possible_cpu(cpu
)
310 mutex_init(&per_cpu(cpu_access_lock
, cpu
));
315 static DEFINE_MUTEX(access_lock
);
317 static inline void trace_access_lock(int cpu
)
320 mutex_lock(&access_lock
);
323 static inline void trace_access_unlock(int cpu
)
326 mutex_unlock(&access_lock
);
329 static inline void trace_access_lock_init(void)
335 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
336 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
338 /* trace_flags holds trace_options default values */
339 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
| TRACE_ITER_PRINTK
|
340 TRACE_ITER_ANNOTATE
| TRACE_ITER_CONTEXT_INFO
| TRACE_ITER_SLEEP_TIME
|
341 TRACE_ITER_GRAPH_TIME
| TRACE_ITER_RECORD_CMD
| TRACE_ITER_OVERWRITE
;
343 static int trace_stop_count
;
344 static DEFINE_SPINLOCK(tracing_start_lock
);
346 static void wakeup_work_handler(struct work_struct
*work
)
348 wake_up(&trace_wait
);
351 static DECLARE_DELAYED_WORK(wakeup_work
, wakeup_work_handler
);
354 * trace_wake_up - wake up tasks waiting for trace input
356 * Schedules a delayed work to wake up any task that is blocked on the
357 * trace_wait queue. These is used with trace_poll for tasks polling the
360 void trace_wake_up(void)
362 const unsigned long delay
= msecs_to_jiffies(2);
364 if (trace_flags
& TRACE_ITER_BLOCK
)
366 schedule_delayed_work(&wakeup_work
, delay
);
369 static int __init
set_buf_size(char *str
)
371 unsigned long buf_size
;
375 buf_size
= memparse(str
, &str
);
376 /* nr_entries can not be zero */
379 trace_buf_size
= buf_size
;
382 __setup("trace_buf_size=", set_buf_size
);
384 static int __init
set_tracing_thresh(char *str
)
386 unsigned long threshhold
;
391 ret
= strict_strtoul(str
, 0, &threshhold
);
394 tracing_thresh
= threshhold
* 1000;
397 __setup("tracing_thresh=", set_tracing_thresh
);
399 unsigned long nsecs_to_usecs(unsigned long nsecs
)
404 /* These must match the bit postions in trace_iterator_flags */
405 static const char *trace_options
[] = {
436 { trace_clock_local
, "local" },
437 { trace_clock_global
, "global" },
438 { trace_clock_counter
, "counter" },
444 * trace_parser_get_init - gets the buffer for trace parser
446 int trace_parser_get_init(struct trace_parser
*parser
, int size
)
448 memset(parser
, 0, sizeof(*parser
));
450 parser
->buffer
= kmalloc(size
, GFP_KERNEL
);
459 * trace_parser_put - frees the buffer for trace parser
461 void trace_parser_put(struct trace_parser
*parser
)
463 kfree(parser
->buffer
);
467 * trace_get_user - reads the user input string separated by space
468 * (matched by isspace(ch))
470 * For each string found the 'struct trace_parser' is updated,
471 * and the function returns.
473 * Returns number of bytes read.
475 * See kernel/trace/trace.h for 'struct trace_parser' details.
477 int trace_get_user(struct trace_parser
*parser
, const char __user
*ubuf
,
478 size_t cnt
, loff_t
*ppos
)
485 trace_parser_clear(parser
);
487 ret
= get_user(ch
, ubuf
++);
495 * The parser is not finished with the last write,
496 * continue reading the user input without skipping spaces.
499 /* skip white space */
500 while (cnt
&& isspace(ch
)) {
501 ret
= get_user(ch
, ubuf
++);
508 /* only spaces were written */
518 /* read the non-space input */
519 while (cnt
&& !isspace(ch
)) {
520 if (parser
->idx
< parser
->size
- 1)
521 parser
->buffer
[parser
->idx
++] = ch
;
526 ret
= get_user(ch
, ubuf
++);
533 /* We either got finished input or we have to wait for another call. */
535 parser
->buffer
[parser
->idx
] = 0;
536 parser
->cont
= false;
539 parser
->buffer
[parser
->idx
++] = ch
;
549 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
557 if (s
->len
<= s
->readpos
)
560 len
= s
->len
- s
->readpos
;
563 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
573 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
578 if (s
->len
<= s
->readpos
)
581 len
= s
->len
- s
->readpos
;
584 ret
= memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
593 * ftrace_max_lock is used to protect the swapping of buffers
594 * when taking a max snapshot. The buffers themselves are
595 * protected by per_cpu spinlocks. But the action of the swap
596 * needs its own lock.
598 * This is defined as a arch_spinlock_t in order to help
599 * with performance when lockdep debugging is enabled.
601 * It is also used in other places outside the update_max_tr
602 * so it needs to be defined outside of the
603 * CONFIG_TRACER_MAX_TRACE.
605 static arch_spinlock_t ftrace_max_lock
=
606 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
608 unsigned long __read_mostly tracing_thresh
;
610 #ifdef CONFIG_TRACER_MAX_TRACE
611 unsigned long __read_mostly tracing_max_latency
;
614 * Copy the new maximum trace into the separate maximum-trace
615 * structure. (this way the maximum trace is permanently saved,
616 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
619 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
621 struct trace_array_cpu
*data
= tr
->data
[cpu
];
622 struct trace_array_cpu
*max_data
;
625 max_tr
.time_start
= data
->preempt_timestamp
;
627 max_data
= max_tr
.data
[cpu
];
628 max_data
->saved_latency
= tracing_max_latency
;
629 max_data
->critical_start
= data
->critical_start
;
630 max_data
->critical_end
= data
->critical_end
;
632 memcpy(max_data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
633 max_data
->pid
= tsk
->pid
;
634 max_data
->uid
= task_uid(tsk
);
635 max_data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
636 max_data
->policy
= tsk
->policy
;
637 max_data
->rt_priority
= tsk
->rt_priority
;
639 /* record this tasks comm */
640 tracing_record_cmdline(tsk
);
644 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
646 * @tsk: the task with the latency
647 * @cpu: The cpu that initiated the trace.
649 * Flip the buffers between the @tr and the max_tr and record information
650 * about which task was the cause of this latency.
653 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
655 struct ring_buffer
*buf
= tr
->buffer
;
657 if (trace_stop_count
)
660 WARN_ON_ONCE(!irqs_disabled());
661 if (!current_trace
->use_max_tr
) {
665 arch_spin_lock(&ftrace_max_lock
);
667 tr
->buffer
= max_tr
.buffer
;
670 __update_max_tr(tr
, tsk
, cpu
);
671 arch_spin_unlock(&ftrace_max_lock
);
675 * update_max_tr_single - only copy one trace over, and reset the rest
677 * @tsk - task with the latency
678 * @cpu - the cpu of the buffer to copy.
680 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
683 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
687 if (trace_stop_count
)
690 WARN_ON_ONCE(!irqs_disabled());
691 if (!current_trace
->use_max_tr
) {
696 arch_spin_lock(&ftrace_max_lock
);
698 ftrace_disable_cpu();
700 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
704 * We failed to swap the buffer due to a commit taking
705 * place on this CPU. We fail to record, but we reset
706 * the max trace buffer (no one writes directly to it)
707 * and flag that it failed.
709 trace_array_printk(&max_tr
, _THIS_IP_
,
710 "Failed to swap buffers due to commit in progress\n");
715 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
&& ret
!= -EBUSY
);
717 __update_max_tr(tr
, tsk
, cpu
);
718 arch_spin_unlock(&ftrace_max_lock
);
720 #endif /* CONFIG_TRACER_MAX_TRACE */
723 * register_tracer - register a tracer with the ftrace system.
724 * @type - the plugin for the tracer
726 * Register a new plugin tracer.
728 int register_tracer(struct tracer
*type
)
729 __releases(kernel_lock
)
730 __acquires(kernel_lock
)
736 pr_info("Tracer must have a name\n");
740 if (strlen(type
->name
) >= MAX_TRACER_SIZE
) {
741 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE
);
745 mutex_lock(&trace_types_lock
);
747 tracing_selftest_running
= true;
749 for (t
= trace_types
; t
; t
= t
->next
) {
750 if (strcmp(type
->name
, t
->name
) == 0) {
752 pr_info("Tracer %s already registered\n",
760 type
->set_flag
= &dummy_set_flag
;
762 type
->flags
= &dummy_tracer_flags
;
764 if (!type
->flags
->opts
)
765 type
->flags
->opts
= dummy_tracer_opt
;
766 if (!type
->wait_pipe
)
767 type
->wait_pipe
= default_wait_pipe
;
770 #ifdef CONFIG_FTRACE_STARTUP_TEST
771 if (type
->selftest
&& !tracing_selftest_disabled
) {
772 struct tracer
*saved_tracer
= current_trace
;
773 struct trace_array
*tr
= &global_trace
;
776 * Run a selftest on this tracer.
777 * Here we reset the trace buffer, and set the current
778 * tracer to be this tracer. The tracer can then run some
779 * internal tracing to verify that everything is in order.
780 * If we fail, we do not register this tracer.
782 tracing_reset_online_cpus(tr
);
784 current_trace
= type
;
786 /* If we expanded the buffers, make sure the max is expanded too */
787 if (ring_buffer_expanded
&& type
->use_max_tr
)
788 ring_buffer_resize(max_tr
.buffer
, trace_buf_size
);
790 /* the test is responsible for initializing and enabling */
791 pr_info("Testing tracer %s: ", type
->name
);
792 ret
= type
->selftest(type
, tr
);
793 /* the test is responsible for resetting too */
794 current_trace
= saved_tracer
;
796 printk(KERN_CONT
"FAILED!\n");
799 /* Only reset on passing, to avoid touching corrupted buffers */
800 tracing_reset_online_cpus(tr
);
802 /* Shrink the max buffer again */
803 if (ring_buffer_expanded
&& type
->use_max_tr
)
804 ring_buffer_resize(max_tr
.buffer
, 1);
806 printk(KERN_CONT
"PASSED\n");
810 type
->next
= trace_types
;
814 tracing_selftest_running
= false;
815 mutex_unlock(&trace_types_lock
);
817 if (ret
|| !default_bootup_tracer
)
820 if (strncmp(default_bootup_tracer
, type
->name
, MAX_TRACER_SIZE
))
823 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
824 /* Do we want this tracer to start on bootup? */
825 tracing_set_tracer(type
->name
);
826 default_bootup_tracer
= NULL
;
827 /* disable other selftests, since this will break it. */
828 tracing_selftest_disabled
= 1;
829 #ifdef CONFIG_FTRACE_STARTUP_TEST
830 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
838 void unregister_tracer(struct tracer
*type
)
842 mutex_lock(&trace_types_lock
);
843 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
847 pr_info("Tracer %s not registered\n", type
->name
);
853 if (type
== current_trace
&& tracer_enabled
) {
856 if (current_trace
->stop
)
857 current_trace
->stop(&global_trace
);
858 current_trace
= &nop_trace
;
861 mutex_unlock(&trace_types_lock
);
864 static void __tracing_reset(struct ring_buffer
*buffer
, int cpu
)
866 ftrace_disable_cpu();
867 ring_buffer_reset_cpu(buffer
, cpu
);
871 void tracing_reset(struct trace_array
*tr
, int cpu
)
873 struct ring_buffer
*buffer
= tr
->buffer
;
875 ring_buffer_record_disable(buffer
);
877 /* Make sure all commits have finished */
879 __tracing_reset(buffer
, cpu
);
881 ring_buffer_record_enable(buffer
);
884 void tracing_reset_online_cpus(struct trace_array
*tr
)
886 struct ring_buffer
*buffer
= tr
->buffer
;
889 ring_buffer_record_disable(buffer
);
891 /* Make sure all commits have finished */
894 tr
->time_start
= ftrace_now(tr
->cpu
);
896 for_each_online_cpu(cpu
)
897 __tracing_reset(buffer
, cpu
);
899 ring_buffer_record_enable(buffer
);
902 void tracing_reset_current(int cpu
)
904 tracing_reset(&global_trace
, cpu
);
907 void tracing_reset_current_online_cpus(void)
909 tracing_reset_online_cpus(&global_trace
);
912 #define SAVED_CMDLINES 128
913 #define NO_CMDLINE_MAP UINT_MAX
914 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
915 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
916 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
917 static int cmdline_idx
;
918 static arch_spinlock_t trace_cmdline_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
920 /* temporary disable recording */
921 static atomic_t trace_record_cmdline_disabled __read_mostly
;
923 static void trace_init_cmdlines(void)
925 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
926 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
930 int is_tracing_stopped(void)
932 return trace_stop_count
;
936 * ftrace_off_permanent - disable all ftrace code permanently
938 * This should only be called when a serious anomally has
939 * been detected. This will turn off the function tracing,
940 * ring buffers, and other tracing utilites. It takes no
941 * locks and can be called from any context.
943 void ftrace_off_permanent(void)
945 tracing_disabled
= 1;
947 tracing_off_permanent();
951 * tracing_start - quick start of the tracer
953 * If tracing is enabled but was stopped by tracing_stop,
954 * this will start the tracer back up.
956 void tracing_start(void)
958 struct ring_buffer
*buffer
;
961 if (tracing_disabled
)
964 spin_lock_irqsave(&tracing_start_lock
, flags
);
965 if (--trace_stop_count
) {
966 if (trace_stop_count
< 0) {
967 /* Someone screwed up their debugging */
969 trace_stop_count
= 0;
974 /* Prevent the buffers from switching */
975 arch_spin_lock(&ftrace_max_lock
);
977 buffer
= global_trace
.buffer
;
979 ring_buffer_record_enable(buffer
);
981 buffer
= max_tr
.buffer
;
983 ring_buffer_record_enable(buffer
);
985 arch_spin_unlock(&ftrace_max_lock
);
989 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
993 * tracing_stop - quick stop of the tracer
995 * Light weight way to stop tracing. Use in conjunction with
998 void tracing_stop(void)
1000 struct ring_buffer
*buffer
;
1001 unsigned long flags
;
1004 spin_lock_irqsave(&tracing_start_lock
, flags
);
1005 if (trace_stop_count
++)
1008 /* Prevent the buffers from switching */
1009 arch_spin_lock(&ftrace_max_lock
);
1011 buffer
= global_trace
.buffer
;
1013 ring_buffer_record_disable(buffer
);
1015 buffer
= max_tr
.buffer
;
1017 ring_buffer_record_disable(buffer
);
1019 arch_spin_unlock(&ftrace_max_lock
);
1022 spin_unlock_irqrestore(&tracing_start_lock
, flags
);
1025 void trace_stop_cmdline_recording(void);
1027 static void trace_save_cmdline(struct task_struct
*tsk
)
1031 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
1035 * It's not the end of the world if we don't get
1036 * the lock, but we also don't want to spin
1037 * nor do we want to disable interrupts,
1038 * so if we miss here, then better luck next time.
1040 if (!arch_spin_trylock(&trace_cmdline_lock
))
1043 idx
= map_pid_to_cmdline
[tsk
->pid
];
1044 if (idx
== NO_CMDLINE_MAP
) {
1045 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
1048 * Check whether the cmdline buffer at idx has a pid
1049 * mapped. We are going to overwrite that entry so we
1050 * need to clear the map_pid_to_cmdline. Otherwise we
1051 * would read the new comm for the old pid.
1053 pid
= map_cmdline_to_pid
[idx
];
1054 if (pid
!= NO_CMDLINE_MAP
)
1055 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
1057 map_cmdline_to_pid
[idx
] = tsk
->pid
;
1058 map_pid_to_cmdline
[tsk
->pid
] = idx
;
1063 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
1065 arch_spin_unlock(&trace_cmdline_lock
);
1068 void trace_find_cmdline(int pid
, char comm
[])
1073 strcpy(comm
, "<idle>");
1077 if (WARN_ON_ONCE(pid
< 0)) {
1078 strcpy(comm
, "<XXX>");
1082 if (pid
> PID_MAX_DEFAULT
) {
1083 strcpy(comm
, "<...>");
1088 arch_spin_lock(&trace_cmdline_lock
);
1089 map
= map_pid_to_cmdline
[pid
];
1090 if (map
!= NO_CMDLINE_MAP
)
1091 strcpy(comm
, saved_cmdlines
[map
]);
1093 strcpy(comm
, "<...>");
1095 arch_spin_unlock(&trace_cmdline_lock
);
1099 void tracing_record_cmdline(struct task_struct
*tsk
)
1101 if (atomic_read(&trace_record_cmdline_disabled
) || !tracer_enabled
||
1105 trace_save_cmdline(tsk
);
1109 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
1112 struct task_struct
*tsk
= current
;
1114 entry
->preempt_count
= pc
& 0xff;
1115 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
1118 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1119 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
1121 TRACE_FLAG_IRQS_NOSUPPORT
|
1123 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
1124 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
1125 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
1127 EXPORT_SYMBOL_GPL(tracing_generic_entry_update
);
1129 struct ring_buffer_event
*
1130 trace_buffer_lock_reserve(struct ring_buffer
*buffer
,
1133 unsigned long flags
, int pc
)
1135 struct ring_buffer_event
*event
;
1137 event
= ring_buffer_lock_reserve(buffer
, len
);
1138 if (event
!= NULL
) {
1139 struct trace_entry
*ent
= ring_buffer_event_data(event
);
1141 tracing_generic_entry_update(ent
, flags
, pc
);
1149 __trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1150 struct ring_buffer_event
*event
,
1151 unsigned long flags
, int pc
,
1154 ring_buffer_unlock_commit(buffer
, event
);
1156 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1157 ftrace_trace_userstack(buffer
, flags
, pc
);
1163 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1164 struct ring_buffer_event
*event
,
1165 unsigned long flags
, int pc
)
1167 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1170 struct ring_buffer_event
*
1171 trace_current_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1172 int type
, unsigned long len
,
1173 unsigned long flags
, int pc
)
1175 *current_rb
= global_trace
.buffer
;
1176 return trace_buffer_lock_reserve(*current_rb
,
1177 type
, len
, flags
, pc
);
1179 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve
);
1181 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
1182 struct ring_buffer_event
*event
,
1183 unsigned long flags
, int pc
)
1185 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1187 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit
);
1189 void trace_nowake_buffer_unlock_commit(struct ring_buffer
*buffer
,
1190 struct ring_buffer_event
*event
,
1191 unsigned long flags
, int pc
)
1193 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 0);
1195 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit
);
1197 void trace_nowake_buffer_unlock_commit_regs(struct ring_buffer
*buffer
,
1198 struct ring_buffer_event
*event
,
1199 unsigned long flags
, int pc
,
1200 struct pt_regs
*regs
)
1202 ring_buffer_unlock_commit(buffer
, event
);
1204 ftrace_trace_stack_regs(buffer
, flags
, 0, pc
, regs
);
1205 ftrace_trace_userstack(buffer
, flags
, pc
);
1207 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit_regs
);
1209 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
1210 struct ring_buffer_event
*event
)
1212 ring_buffer_discard_commit(buffer
, event
);
1214 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit
);
1217 trace_function(struct trace_array
*tr
,
1218 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1221 struct ftrace_event_call
*call
= &event_function
;
1222 struct ring_buffer
*buffer
= tr
->buffer
;
1223 struct ring_buffer_event
*event
;
1224 struct ftrace_entry
*entry
;
1226 /* If we are reading the ring buffer, don't trace */
1227 if (unlikely(__this_cpu_read(ftrace_cpu_disabled
)))
1230 event
= trace_buffer_lock_reserve(buffer
, TRACE_FN
, sizeof(*entry
),
1234 entry
= ring_buffer_event_data(event
);
1236 entry
->parent_ip
= parent_ip
;
1238 if (!filter_check_discard(call
, entry
, buffer
, event
))
1239 ring_buffer_unlock_commit(buffer
, event
);
1243 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1244 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1247 if (likely(!atomic_read(&data
->disabled
)))
1248 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
1251 #ifdef CONFIG_STACKTRACE
1253 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1254 struct ftrace_stack
{
1255 unsigned long calls
[FTRACE_STACK_MAX_ENTRIES
];
1258 static DEFINE_PER_CPU(struct ftrace_stack
, ftrace_stack
);
1259 static DEFINE_PER_CPU(int, ftrace_stack_reserve
);
1261 static void __ftrace_trace_stack(struct ring_buffer
*buffer
,
1262 unsigned long flags
,
1263 int skip
, int pc
, struct pt_regs
*regs
)
1265 struct ftrace_event_call
*call
= &event_kernel_stack
;
1266 struct ring_buffer_event
*event
;
1267 struct stack_entry
*entry
;
1268 struct stack_trace trace
;
1270 int size
= FTRACE_STACK_ENTRIES
;
1272 trace
.nr_entries
= 0;
1276 * Since events can happen in NMIs there's no safe way to
1277 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1278 * or NMI comes in, it will just have to use the default
1279 * FTRACE_STACK_SIZE.
1281 preempt_disable_notrace();
1283 use_stack
= ++__get_cpu_var(ftrace_stack_reserve
);
1285 * We don't need any atomic variables, just a barrier.
1286 * If an interrupt comes in, we don't care, because it would
1287 * have exited and put the counter back to what we want.
1288 * We just need a barrier to keep gcc from moving things
1292 if (use_stack
== 1) {
1293 trace
.entries
= &__get_cpu_var(ftrace_stack
).calls
[0];
1294 trace
.max_entries
= FTRACE_STACK_MAX_ENTRIES
;
1297 save_stack_trace_regs(regs
, &trace
);
1299 save_stack_trace(&trace
);
1301 if (trace
.nr_entries
> size
)
1302 size
= trace
.nr_entries
;
1304 /* From now on, use_stack is a boolean */
1307 size
*= sizeof(unsigned long);
1309 event
= trace_buffer_lock_reserve(buffer
, TRACE_STACK
,
1310 sizeof(*entry
) + size
, flags
, pc
);
1313 entry
= ring_buffer_event_data(event
);
1315 memset(&entry
->caller
, 0, size
);
1318 memcpy(&entry
->caller
, trace
.entries
,
1319 trace
.nr_entries
* sizeof(unsigned long));
1321 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1322 trace
.entries
= entry
->caller
;
1324 save_stack_trace_regs(regs
, &trace
);
1326 save_stack_trace(&trace
);
1329 entry
->size
= trace
.nr_entries
;
1331 if (!filter_check_discard(call
, entry
, buffer
, event
))
1332 ring_buffer_unlock_commit(buffer
, event
);
1335 /* Again, don't let gcc optimize things here */
1337 __get_cpu_var(ftrace_stack_reserve
)--;
1338 preempt_enable_notrace();
1342 void ftrace_trace_stack_regs(struct ring_buffer
*buffer
, unsigned long flags
,
1343 int skip
, int pc
, struct pt_regs
*regs
)
1345 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1348 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, regs
);
1351 void ftrace_trace_stack(struct ring_buffer
*buffer
, unsigned long flags
,
1354 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1357 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, NULL
);
1360 void __trace_stack(struct trace_array
*tr
, unsigned long flags
, int skip
,
1363 __ftrace_trace_stack(tr
->buffer
, flags
, skip
, pc
, NULL
);
1367 * trace_dump_stack - record a stack back trace in the trace buffer
1369 void trace_dump_stack(void)
1371 unsigned long flags
;
1373 if (tracing_disabled
|| tracing_selftest_running
)
1376 local_save_flags(flags
);
1378 /* skipping 3 traces, seems to get us at the caller of this function */
1379 __ftrace_trace_stack(global_trace
.buffer
, flags
, 3, preempt_count(), NULL
);
1382 static DEFINE_PER_CPU(int, user_stack_count
);
1385 ftrace_trace_userstack(struct ring_buffer
*buffer
, unsigned long flags
, int pc
)
1387 struct ftrace_event_call
*call
= &event_user_stack
;
1388 struct ring_buffer_event
*event
;
1389 struct userstack_entry
*entry
;
1390 struct stack_trace trace
;
1392 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1396 * NMIs can not handle page faults, even with fix ups.
1397 * The save user stack can (and often does) fault.
1399 if (unlikely(in_nmi()))
1403 * prevent recursion, since the user stack tracing may
1404 * trigger other kernel events.
1407 if (__this_cpu_read(user_stack_count
))
1410 __this_cpu_inc(user_stack_count
);
1412 event
= trace_buffer_lock_reserve(buffer
, TRACE_USER_STACK
,
1413 sizeof(*entry
), flags
, pc
);
1415 goto out_drop_count
;
1416 entry
= ring_buffer_event_data(event
);
1418 entry
->tgid
= current
->tgid
;
1419 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1421 trace
.nr_entries
= 0;
1422 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1424 trace
.entries
= entry
->caller
;
1426 save_stack_trace_user(&trace
);
1427 if (!filter_check_discard(call
, entry
, buffer
, event
))
1428 ring_buffer_unlock_commit(buffer
, event
);
1431 __this_cpu_dec(user_stack_count
);
1437 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1439 ftrace_trace_userstack(tr
, flags
, preempt_count());
1443 #endif /* CONFIG_STACKTRACE */
1446 * trace_vbprintk - write binary msg to tracing buffer
1449 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1451 static arch_spinlock_t trace_buf_lock
=
1452 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
1453 static u32 trace_buf
[TRACE_BUF_SIZE
];
1455 struct ftrace_event_call
*call
= &event_bprint
;
1456 struct ring_buffer_event
*event
;
1457 struct ring_buffer
*buffer
;
1458 struct trace_array
*tr
= &global_trace
;
1459 struct trace_array_cpu
*data
;
1460 struct bprint_entry
*entry
;
1461 unsigned long flags
;
1463 int cpu
, len
= 0, size
, pc
;
1465 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1468 /* Don't pollute graph traces with trace_vprintk internals */
1469 pause_graph_tracing();
1471 pc
= preempt_count();
1472 preempt_disable_notrace();
1473 cpu
= raw_smp_processor_id();
1474 data
= tr
->data
[cpu
];
1476 disable
= atomic_inc_return(&data
->disabled
);
1477 if (unlikely(disable
!= 1))
1480 /* Lockdep uses trace_printk for lock tracing */
1481 local_irq_save(flags
);
1482 arch_spin_lock(&trace_buf_lock
);
1483 len
= vbin_printf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1485 if (len
> TRACE_BUF_SIZE
|| len
< 0)
1488 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1489 buffer
= tr
->buffer
;
1490 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
1494 entry
= ring_buffer_event_data(event
);
1498 memcpy(entry
->buf
, trace_buf
, sizeof(u32
) * len
);
1499 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1500 ring_buffer_unlock_commit(buffer
, event
);
1501 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1505 arch_spin_unlock(&trace_buf_lock
);
1506 local_irq_restore(flags
);
1509 atomic_dec_return(&data
->disabled
);
1510 preempt_enable_notrace();
1511 unpause_graph_tracing();
1515 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1517 int trace_array_printk(struct trace_array
*tr
,
1518 unsigned long ip
, const char *fmt
, ...)
1523 if (!(trace_flags
& TRACE_ITER_PRINTK
))
1527 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
1532 int trace_array_vprintk(struct trace_array
*tr
,
1533 unsigned long ip
, const char *fmt
, va_list args
)
1535 static arch_spinlock_t trace_buf_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
1536 static char trace_buf
[TRACE_BUF_SIZE
];
1538 struct ftrace_event_call
*call
= &event_print
;
1539 struct ring_buffer_event
*event
;
1540 struct ring_buffer
*buffer
;
1541 struct trace_array_cpu
*data
;
1542 int cpu
, len
= 0, size
, pc
;
1543 struct print_entry
*entry
;
1544 unsigned long irq_flags
;
1547 if (tracing_disabled
|| tracing_selftest_running
)
1550 pc
= preempt_count();
1551 preempt_disable_notrace();
1552 cpu
= raw_smp_processor_id();
1553 data
= tr
->data
[cpu
];
1555 disable
= atomic_inc_return(&data
->disabled
);
1556 if (unlikely(disable
!= 1))
1559 pause_graph_tracing();
1560 raw_local_irq_save(irq_flags
);
1561 arch_spin_lock(&trace_buf_lock
);
1562 len
= vsnprintf(trace_buf
, TRACE_BUF_SIZE
, fmt
, args
);
1564 size
= sizeof(*entry
) + len
+ 1;
1565 buffer
= tr
->buffer
;
1566 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
1570 entry
= ring_buffer_event_data(event
);
1573 memcpy(&entry
->buf
, trace_buf
, len
);
1574 entry
->buf
[len
] = '\0';
1575 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1576 ring_buffer_unlock_commit(buffer
, event
);
1577 ftrace_trace_stack(buffer
, irq_flags
, 6, pc
);
1581 arch_spin_unlock(&trace_buf_lock
);
1582 raw_local_irq_restore(irq_flags
);
1583 unpause_graph_tracing();
1585 atomic_dec_return(&data
->disabled
);
1586 preempt_enable_notrace();
1591 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1593 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
1595 EXPORT_SYMBOL_GPL(trace_vprintk
);
1597 static void trace_iterator_increment(struct trace_iterator
*iter
)
1599 /* Don't allow ftrace to trace into the ring buffers */
1600 ftrace_disable_cpu();
1603 if (iter
->buffer_iter
[iter
->cpu
])
1604 ring_buffer_read(iter
->buffer_iter
[iter
->cpu
], NULL
);
1606 ftrace_enable_cpu();
1609 static struct trace_entry
*
1610 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
,
1611 unsigned long *lost_events
)
1613 struct ring_buffer_event
*event
;
1614 struct ring_buffer_iter
*buf_iter
= iter
->buffer_iter
[cpu
];
1616 /* Don't allow ftrace to trace into the ring buffers */
1617 ftrace_disable_cpu();
1620 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1622 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
,
1625 ftrace_enable_cpu();
1628 iter
->ent_size
= ring_buffer_event_length(event
);
1629 return ring_buffer_event_data(event
);
1635 static struct trace_entry
*
1636 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
,
1637 unsigned long *missing_events
, u64
*ent_ts
)
1639 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1640 struct trace_entry
*ent
, *next
= NULL
;
1641 unsigned long lost_events
= 0, next_lost
= 0;
1642 int cpu_file
= iter
->cpu_file
;
1643 u64 next_ts
= 0, ts
;
1648 * If we are in a per_cpu trace file, don't bother by iterating over
1649 * all cpu and peek directly.
1651 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1652 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1654 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
, missing_events
);
1656 *ent_cpu
= cpu_file
;
1661 for_each_tracing_cpu(cpu
) {
1663 if (ring_buffer_empty_cpu(buffer
, cpu
))
1666 ent
= peek_next_entry(iter
, cpu
, &ts
, &lost_events
);
1669 * Pick the entry with the smallest timestamp:
1671 if (ent
&& (!next
|| ts
< next_ts
)) {
1675 next_lost
= lost_events
;
1680 *ent_cpu
= next_cpu
;
1686 *missing_events
= next_lost
;
1691 /* Find the next real entry, without updating the iterator itself */
1692 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1693 int *ent_cpu
, u64
*ent_ts
)
1695 return __find_next_entry(iter
, ent_cpu
, NULL
, ent_ts
);
1698 /* Find the next real entry, and increment the iterator to the next entry */
1699 void *trace_find_next_entry_inc(struct trace_iterator
*iter
)
1701 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
,
1702 &iter
->lost_events
, &iter
->ts
);
1705 trace_iterator_increment(iter
);
1707 return iter
->ent
? iter
: NULL
;
1710 static void trace_consume(struct trace_iterator
*iter
)
1712 /* Don't allow ftrace to trace into the ring buffers */
1713 ftrace_disable_cpu();
1714 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
,
1715 &iter
->lost_events
);
1716 ftrace_enable_cpu();
1719 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1721 struct trace_iterator
*iter
= m
->private;
1725 WARN_ON_ONCE(iter
->leftover
);
1729 /* can't go backwards */
1734 ent
= trace_find_next_entry_inc(iter
);
1738 while (ent
&& iter
->idx
< i
)
1739 ent
= trace_find_next_entry_inc(iter
);
1746 void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
1748 struct trace_array
*tr
= iter
->tr
;
1749 struct ring_buffer_event
*event
;
1750 struct ring_buffer_iter
*buf_iter
;
1751 unsigned long entries
= 0;
1754 tr
->data
[cpu
]->skipped_entries
= 0;
1756 if (!iter
->buffer_iter
[cpu
])
1759 buf_iter
= iter
->buffer_iter
[cpu
];
1760 ring_buffer_iter_reset(buf_iter
);
1763 * We could have the case with the max latency tracers
1764 * that a reset never took place on a cpu. This is evident
1765 * by the timestamp being before the start of the buffer.
1767 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
1768 if (ts
>= iter
->tr
->time_start
)
1771 ring_buffer_read(buf_iter
, NULL
);
1774 tr
->data
[cpu
]->skipped_entries
= entries
;
1778 * The current tracer is copied to avoid a global locking
1781 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1783 struct trace_iterator
*iter
= m
->private;
1784 static struct tracer
*old_tracer
;
1785 int cpu_file
= iter
->cpu_file
;
1790 /* copy the tracer to avoid using a global lock all around */
1791 mutex_lock(&trace_types_lock
);
1792 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1793 old_tracer
= current_trace
;
1794 *iter
->trace
= *current_trace
;
1796 mutex_unlock(&trace_types_lock
);
1798 atomic_inc(&trace_record_cmdline_disabled
);
1800 if (*pos
!= iter
->pos
) {
1805 ftrace_disable_cpu();
1807 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1808 for_each_tracing_cpu(cpu
)
1809 tracing_iter_reset(iter
, cpu
);
1811 tracing_iter_reset(iter
, cpu_file
);
1813 ftrace_enable_cpu();
1816 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1821 * If we overflowed the seq_file before, then we want
1822 * to just reuse the trace_seq buffer again.
1828 p
= s_next(m
, p
, &l
);
1832 trace_event_read_lock();
1833 trace_access_lock(cpu_file
);
1837 static void s_stop(struct seq_file
*m
, void *p
)
1839 struct trace_iterator
*iter
= m
->private;
1841 atomic_dec(&trace_record_cmdline_disabled
);
1842 trace_access_unlock(iter
->cpu_file
);
1843 trace_event_read_unlock();
1846 static void print_lat_help_header(struct seq_file
*m
)
1848 seq_puts(m
, "# _------=> CPU# \n");
1849 seq_puts(m
, "# / _-----=> irqs-off \n");
1850 seq_puts(m
, "# | / _----=> need-resched \n");
1851 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1852 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1853 seq_puts(m
, "# |||| / delay \n");
1854 seq_puts(m
, "# cmd pid ||||| time | caller \n");
1855 seq_puts(m
, "# \\ / ||||| \\ | / \n");
1858 static void print_func_help_header(struct seq_file
*m
)
1860 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1861 seq_puts(m
, "# | | | | |\n");
1866 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1868 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1869 struct trace_array
*tr
= iter
->tr
;
1870 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1871 struct tracer
*type
= current_trace
;
1872 unsigned long entries
= 0;
1873 unsigned long total
= 0;
1874 unsigned long count
;
1875 const char *name
= "preemption";
1882 for_each_tracing_cpu(cpu
) {
1883 count
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
1885 * If this buffer has skipped entries, then we hold all
1886 * entries for the trace and we need to ignore the
1887 * ones before the time stamp.
1889 if (tr
->data
[cpu
]->skipped_entries
) {
1890 count
-= tr
->data
[cpu
]->skipped_entries
;
1891 /* total is the same as the entries */
1895 ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
1899 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
1901 seq_puts(m
, "# -----------------------------------"
1902 "---------------------------------\n");
1903 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1904 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1905 nsecs_to_usecs(data
->saved_latency
),
1909 #if defined(CONFIG_PREEMPT_NONE)
1911 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1913 #elif defined(CONFIG_PREEMPT)
1918 /* These are reserved for later use */
1921 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1925 seq_puts(m
, "# -----------------\n");
1926 seq_printf(m
, "# | task: %.16s-%d "
1927 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1928 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1929 data
->policy
, data
->rt_priority
);
1930 seq_puts(m
, "# -----------------\n");
1932 if (data
->critical_start
) {
1933 seq_puts(m
, "# => started at: ");
1934 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1935 trace_print_seq(m
, &iter
->seq
);
1936 seq_puts(m
, "\n# => ended at: ");
1937 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1938 trace_print_seq(m
, &iter
->seq
);
1939 seq_puts(m
, "\n#\n");
1945 static void test_cpu_buff_start(struct trace_iterator
*iter
)
1947 struct trace_seq
*s
= &iter
->seq
;
1949 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
1952 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
1955 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
1958 if (iter
->tr
->data
[iter
->cpu
]->skipped_entries
)
1961 cpumask_set_cpu(iter
->cpu
, iter
->started
);
1963 /* Don't print started cpu buffer for the first entry of the trace */
1965 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
1969 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
1971 struct trace_seq
*s
= &iter
->seq
;
1972 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1973 struct trace_entry
*entry
;
1974 struct trace_event
*event
;
1978 test_cpu_buff_start(iter
);
1980 event
= ftrace_find_event(entry
->type
);
1982 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
1983 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1984 if (!trace_print_lat_context(iter
))
1987 if (!trace_print_context(iter
))
1993 return event
->funcs
->trace(iter
, sym_flags
, event
);
1995 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
1998 return TRACE_TYPE_HANDLED
;
2000 return TRACE_TYPE_PARTIAL_LINE
;
2003 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
2005 struct trace_seq
*s
= &iter
->seq
;
2006 struct trace_entry
*entry
;
2007 struct trace_event
*event
;
2011 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2012 if (!trace_seq_printf(s
, "%d %d %llu ",
2013 entry
->pid
, iter
->cpu
, iter
->ts
))
2017 event
= ftrace_find_event(entry
->type
);
2019 return event
->funcs
->raw(iter
, 0, event
);
2021 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
2024 return TRACE_TYPE_HANDLED
;
2026 return TRACE_TYPE_PARTIAL_LINE
;
2029 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
2031 struct trace_seq
*s
= &iter
->seq
;
2032 unsigned char newline
= '\n';
2033 struct trace_entry
*entry
;
2034 struct trace_event
*event
;
2038 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2039 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
2040 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
2041 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
2044 event
= ftrace_find_event(entry
->type
);
2046 enum print_line_t ret
= event
->funcs
->hex(iter
, 0, event
);
2047 if (ret
!= TRACE_TYPE_HANDLED
)
2051 SEQ_PUT_FIELD_RET(s
, newline
);
2053 return TRACE_TYPE_HANDLED
;
2056 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
2058 struct trace_seq
*s
= &iter
->seq
;
2059 struct trace_entry
*entry
;
2060 struct trace_event
*event
;
2064 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2065 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
2066 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
2067 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
2070 event
= ftrace_find_event(entry
->type
);
2071 return event
? event
->funcs
->binary(iter
, 0, event
) :
2075 int trace_empty(struct trace_iterator
*iter
)
2079 /* If we are looking at one CPU buffer, only check that one */
2080 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
2081 cpu
= iter
->cpu_file
;
2082 if (iter
->buffer_iter
[cpu
]) {
2083 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
2086 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2092 for_each_tracing_cpu(cpu
) {
2093 if (iter
->buffer_iter
[cpu
]) {
2094 if (!ring_buffer_iter_empty(iter
->buffer_iter
[cpu
]))
2097 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2105 /* Called with trace_event_read_lock() held. */
2106 enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
2108 enum print_line_t ret
;
2110 if (iter
->lost_events
&&
2111 !trace_seq_printf(&iter
->seq
, "CPU:%d [LOST %lu EVENTS]\n",
2112 iter
->cpu
, iter
->lost_events
))
2113 return TRACE_TYPE_PARTIAL_LINE
;
2115 if (iter
->trace
&& iter
->trace
->print_line
) {
2116 ret
= iter
->trace
->print_line(iter
);
2117 if (ret
!= TRACE_TYPE_UNHANDLED
)
2121 if (iter
->ent
->type
== TRACE_BPRINT
&&
2122 trace_flags
& TRACE_ITER_PRINTK
&&
2123 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2124 return trace_print_bprintk_msg_only(iter
);
2126 if (iter
->ent
->type
== TRACE_PRINT
&&
2127 trace_flags
& TRACE_ITER_PRINTK
&&
2128 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2129 return trace_print_printk_msg_only(iter
);
2131 if (trace_flags
& TRACE_ITER_BIN
)
2132 return print_bin_fmt(iter
);
2134 if (trace_flags
& TRACE_ITER_HEX
)
2135 return print_hex_fmt(iter
);
2137 if (trace_flags
& TRACE_ITER_RAW
)
2138 return print_raw_fmt(iter
);
2140 return print_trace_fmt(iter
);
2143 void trace_latency_header(struct seq_file
*m
)
2145 struct trace_iterator
*iter
= m
->private;
2147 /* print nothing if the buffers are empty */
2148 if (trace_empty(iter
))
2151 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
2152 print_trace_header(m
, iter
);
2154 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2155 print_lat_help_header(m
);
2158 void trace_default_header(struct seq_file
*m
)
2160 struct trace_iterator
*iter
= m
->private;
2162 if (!(trace_flags
& TRACE_ITER_CONTEXT_INFO
))
2165 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2166 /* print nothing if the buffers are empty */
2167 if (trace_empty(iter
))
2169 print_trace_header(m
, iter
);
2170 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2171 print_lat_help_header(m
);
2173 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2174 print_func_help_header(m
);
2178 static void test_ftrace_alive(struct seq_file
*m
)
2180 if (!ftrace_is_dead())
2182 seq_printf(m
, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2183 seq_printf(m
, "# MAY BE MISSING FUNCTION EVENTS\n");
2186 static int s_show(struct seq_file
*m
, void *v
)
2188 struct trace_iterator
*iter
= v
;
2191 if (iter
->ent
== NULL
) {
2193 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
2195 test_ftrace_alive(m
);
2197 if (iter
->trace
&& iter
->trace
->print_header
)
2198 iter
->trace
->print_header(m
);
2200 trace_default_header(m
);
2202 } else if (iter
->leftover
) {
2204 * If we filled the seq_file buffer earlier, we
2205 * want to just show it now.
2207 ret
= trace_print_seq(m
, &iter
->seq
);
2209 /* ret should this time be zero, but you never know */
2210 iter
->leftover
= ret
;
2213 print_trace_line(iter
);
2214 ret
= trace_print_seq(m
, &iter
->seq
);
2216 * If we overflow the seq_file buffer, then it will
2217 * ask us for this data again at start up.
2219 * ret is 0 if seq_file write succeeded.
2222 iter
->leftover
= ret
;
2228 static const struct seq_operations tracer_seq_ops
= {
2235 static struct trace_iterator
*
2236 __tracing_open(struct inode
*inode
, struct file
*file
)
2238 long cpu_file
= (long) inode
->i_private
;
2239 void *fail_ret
= ERR_PTR(-ENOMEM
);
2240 struct trace_iterator
*iter
;
2244 if (tracing_disabled
)
2245 return ERR_PTR(-ENODEV
);
2247 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2249 return ERR_PTR(-ENOMEM
);
2252 * We make a copy of the current tracer to avoid concurrent
2253 * changes on it while we are reading.
2255 mutex_lock(&trace_types_lock
);
2256 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2261 *iter
->trace
= *current_trace
;
2263 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2266 if (current_trace
&& current_trace
->print_max
)
2269 iter
->tr
= &global_trace
;
2271 mutex_init(&iter
->mutex
);
2272 iter
->cpu_file
= cpu_file
;
2274 /* Notify the tracer early; before we stop tracing. */
2275 if (iter
->trace
&& iter
->trace
->open
)
2276 iter
->trace
->open(iter
);
2278 /* Annotate start of buffers if we had overruns */
2279 if (ring_buffer_overruns(iter
->tr
->buffer
))
2280 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2282 /* stop the trace while dumping */
2285 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
2286 for_each_tracing_cpu(cpu
) {
2287 iter
->buffer_iter
[cpu
] =
2288 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2290 ring_buffer_read_prepare_sync();
2291 for_each_tracing_cpu(cpu
) {
2292 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2293 tracing_iter_reset(iter
, cpu
);
2296 cpu
= iter
->cpu_file
;
2297 iter
->buffer_iter
[cpu
] =
2298 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2299 ring_buffer_read_prepare_sync();
2300 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2301 tracing_iter_reset(iter
, cpu
);
2304 ret
= seq_open(file
, &tracer_seq_ops
);
2306 fail_ret
= ERR_PTR(ret
);
2310 m
= file
->private_data
;
2313 mutex_unlock(&trace_types_lock
);
2318 for_each_tracing_cpu(cpu
) {
2319 if (iter
->buffer_iter
[cpu
])
2320 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2322 free_cpumask_var(iter
->started
);
2325 mutex_unlock(&trace_types_lock
);
2332 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2334 if (tracing_disabled
)
2337 filp
->private_data
= inode
->i_private
;
2341 static int tracing_release(struct inode
*inode
, struct file
*file
)
2343 struct seq_file
*m
= file
->private_data
;
2344 struct trace_iterator
*iter
;
2347 if (!(file
->f_mode
& FMODE_READ
))
2352 mutex_lock(&trace_types_lock
);
2353 for_each_tracing_cpu(cpu
) {
2354 if (iter
->buffer_iter
[cpu
])
2355 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2358 if (iter
->trace
&& iter
->trace
->close
)
2359 iter
->trace
->close(iter
);
2361 /* reenable tracing if it was previously enabled */
2363 mutex_unlock(&trace_types_lock
);
2365 seq_release(inode
, file
);
2366 mutex_destroy(&iter
->mutex
);
2367 free_cpumask_var(iter
->started
);
2373 static int tracing_open(struct inode
*inode
, struct file
*file
)
2375 struct trace_iterator
*iter
;
2378 /* If this file was open for write, then erase contents */
2379 if ((file
->f_mode
& FMODE_WRITE
) &&
2380 (file
->f_flags
& O_TRUNC
)) {
2381 long cpu
= (long) inode
->i_private
;
2383 if (cpu
== TRACE_PIPE_ALL_CPU
)
2384 tracing_reset_online_cpus(&global_trace
);
2386 tracing_reset(&global_trace
, cpu
);
2389 if (file
->f_mode
& FMODE_READ
) {
2390 iter
= __tracing_open(inode
, file
);
2392 ret
= PTR_ERR(iter
);
2393 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2394 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2400 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2402 struct tracer
*t
= v
;
2412 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2417 mutex_lock(&trace_types_lock
);
2418 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2424 static void t_stop(struct seq_file
*m
, void *p
)
2426 mutex_unlock(&trace_types_lock
);
2429 static int t_show(struct seq_file
*m
, void *v
)
2431 struct tracer
*t
= v
;
2436 seq_printf(m
, "%s", t
->name
);
2445 static const struct seq_operations show_traces_seq_ops
= {
2452 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2454 if (tracing_disabled
)
2457 return seq_open(file
, &show_traces_seq_ops
);
2461 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2462 size_t count
, loff_t
*ppos
)
2467 static loff_t
tracing_seek(struct file
*file
, loff_t offset
, int origin
)
2469 if (file
->f_mode
& FMODE_READ
)
2470 return seq_lseek(file
, offset
, origin
);
2475 static const struct file_operations tracing_fops
= {
2476 .open
= tracing_open
,
2478 .write
= tracing_write_stub
,
2479 .llseek
= tracing_seek
,
2480 .release
= tracing_release
,
2483 static const struct file_operations show_traces_fops
= {
2484 .open
= show_traces_open
,
2486 .release
= seq_release
,
2487 .llseek
= seq_lseek
,
2491 * Only trace on a CPU if the bitmask is set:
2493 static cpumask_var_t tracing_cpumask
;
2496 * The tracer itself will not take this lock, but still we want
2497 * to provide a consistent cpumask to user-space:
2499 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2502 * Temporary storage for the character representation of the
2503 * CPU bitmask (and one more byte for the newline):
2505 static char mask_str
[NR_CPUS
+ 1];
2508 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2509 size_t count
, loff_t
*ppos
)
2513 mutex_lock(&tracing_cpumask_update_lock
);
2515 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2516 if (count
- len
< 2) {
2520 len
+= sprintf(mask_str
+ len
, "\n");
2521 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2524 mutex_unlock(&tracing_cpumask_update_lock
);
2530 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2531 size_t count
, loff_t
*ppos
)
2534 cpumask_var_t tracing_cpumask_new
;
2536 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2539 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2543 mutex_lock(&tracing_cpumask_update_lock
);
2545 local_irq_disable();
2546 arch_spin_lock(&ftrace_max_lock
);
2547 for_each_tracing_cpu(cpu
) {
2549 * Increase/decrease the disabled counter if we are
2550 * about to flip a bit in the cpumask:
2552 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2553 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2554 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2556 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2557 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2558 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2561 arch_spin_unlock(&ftrace_max_lock
);
2564 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2566 mutex_unlock(&tracing_cpumask_update_lock
);
2567 free_cpumask_var(tracing_cpumask_new
);
2572 free_cpumask_var(tracing_cpumask_new
);
2577 static const struct file_operations tracing_cpumask_fops
= {
2578 .open
= tracing_open_generic
,
2579 .read
= tracing_cpumask_read
,
2580 .write
= tracing_cpumask_write
,
2581 .llseek
= generic_file_llseek
,
2584 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
2586 struct tracer_opt
*trace_opts
;
2590 mutex_lock(&trace_types_lock
);
2591 tracer_flags
= current_trace
->flags
->val
;
2592 trace_opts
= current_trace
->flags
->opts
;
2594 for (i
= 0; trace_options
[i
]; i
++) {
2595 if (trace_flags
& (1 << i
))
2596 seq_printf(m
, "%s\n", trace_options
[i
]);
2598 seq_printf(m
, "no%s\n", trace_options
[i
]);
2601 for (i
= 0; trace_opts
[i
].name
; i
++) {
2602 if (tracer_flags
& trace_opts
[i
].bit
)
2603 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
2605 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
2607 mutex_unlock(&trace_types_lock
);
2612 static int __set_tracer_option(struct tracer
*trace
,
2613 struct tracer_flags
*tracer_flags
,
2614 struct tracer_opt
*opts
, int neg
)
2618 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
2623 tracer_flags
->val
&= ~opts
->bit
;
2625 tracer_flags
->val
|= opts
->bit
;
2629 /* Try to assign a tracer specific option */
2630 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2632 struct tracer_flags
*tracer_flags
= trace
->flags
;
2633 struct tracer_opt
*opts
= NULL
;
2636 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
2637 opts
= &tracer_flags
->opts
[i
];
2639 if (strcmp(cmp
, opts
->name
) == 0)
2640 return __set_tracer_option(trace
, trace
->flags
,
2647 static void set_tracer_flags(unsigned int mask
, int enabled
)
2649 /* do nothing if flag is already set */
2650 if (!!(trace_flags
& mask
) == !!enabled
)
2654 trace_flags
|= mask
;
2656 trace_flags
&= ~mask
;
2658 if (mask
== TRACE_ITER_RECORD_CMD
)
2659 trace_event_enable_cmd_record(enabled
);
2661 if (mask
== TRACE_ITER_OVERWRITE
)
2662 ring_buffer_change_overwrite(global_trace
.buffer
, enabled
);
2666 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2667 size_t cnt
, loff_t
*ppos
)
2675 if (cnt
>= sizeof(buf
))
2678 if (copy_from_user(&buf
, ubuf
, cnt
))
2682 cmp
= strstrip(buf
);
2684 if (strncmp(cmp
, "no", 2) == 0) {
2689 for (i
= 0; trace_options
[i
]; i
++) {
2690 if (strcmp(cmp
, trace_options
[i
]) == 0) {
2691 set_tracer_flags(1 << i
, !neg
);
2696 /* If no option could be set, test the specific tracer options */
2697 if (!trace_options
[i
]) {
2698 mutex_lock(&trace_types_lock
);
2699 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2700 mutex_unlock(&trace_types_lock
);
2710 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
2712 if (tracing_disabled
)
2714 return single_open(file
, tracing_trace_options_show
, NULL
);
2717 static const struct file_operations tracing_iter_fops
= {
2718 .open
= tracing_trace_options_open
,
2720 .llseek
= seq_lseek
,
2721 .release
= single_release
,
2722 .write
= tracing_trace_options_write
,
2725 static const char readme_msg
[] =
2726 "tracing mini-HOWTO:\n\n"
2727 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2728 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2729 "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2730 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2732 "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2733 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2735 "# cat /sys/kernel/debug/tracing/trace_options\n"
2736 "noprint-parent nosym-offset nosym-addr noverbose\n"
2737 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2738 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2739 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2740 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2744 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2745 size_t cnt
, loff_t
*ppos
)
2747 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2748 readme_msg
, strlen(readme_msg
));
2751 static const struct file_operations tracing_readme_fops
= {
2752 .open
= tracing_open_generic
,
2753 .read
= tracing_readme_read
,
2754 .llseek
= generic_file_llseek
,
2758 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
2759 size_t cnt
, loff_t
*ppos
)
2768 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
2772 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
2780 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
2783 pid
= map_cmdline_to_pid
[i
];
2784 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
2787 trace_find_cmdline(pid
, buf_comm
);
2788 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
2793 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
2802 static const struct file_operations tracing_saved_cmdlines_fops
= {
2803 .open
= tracing_open_generic
,
2804 .read
= tracing_saved_cmdlines_read
,
2805 .llseek
= generic_file_llseek
,
2809 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2810 size_t cnt
, loff_t
*ppos
)
2815 r
= sprintf(buf
, "%u\n", tracer_enabled
);
2816 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2820 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2821 size_t cnt
, loff_t
*ppos
)
2823 struct trace_array
*tr
= filp
->private_data
;
2827 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
2833 mutex_lock(&trace_types_lock
);
2834 if (tracer_enabled
^ val
) {
2836 /* Only need to warn if this is used to change the state */
2837 WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on");
2841 if (current_trace
->start
)
2842 current_trace
->start(tr
);
2847 if (current_trace
->stop
)
2848 current_trace
->stop(tr
);
2851 mutex_unlock(&trace_types_lock
);
2859 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2860 size_t cnt
, loff_t
*ppos
)
2862 char buf
[MAX_TRACER_SIZE
+2];
2865 mutex_lock(&trace_types_lock
);
2867 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2869 r
= sprintf(buf
, "\n");
2870 mutex_unlock(&trace_types_lock
);
2872 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2875 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
2877 tracing_reset_online_cpus(tr
);
2881 static int __tracing_resize_ring_buffer(unsigned long size
)
2886 * If kernel or user changes the size of the ring buffer
2887 * we use the size that was given, and we can forget about
2888 * expanding it later.
2890 ring_buffer_expanded
= 1;
2892 ret
= ring_buffer_resize(global_trace
.buffer
, size
);
2896 if (!current_trace
->use_max_tr
)
2899 ret
= ring_buffer_resize(max_tr
.buffer
, size
);
2903 r
= ring_buffer_resize(global_trace
.buffer
,
2904 global_trace
.entries
);
2907 * AARGH! We are left with different
2908 * size max buffer!!!!
2909 * The max buffer is our "snapshot" buffer.
2910 * When a tracer needs a snapshot (one of the
2911 * latency tracers), it swaps the max buffer
2912 * with the saved snap shot. We succeeded to
2913 * update the size of the main buffer, but failed to
2914 * update the size of the max buffer. But when we tried
2915 * to reset the main buffer to the original size, we
2916 * failed there too. This is very unlikely to
2917 * happen, but if it does, warn and kill all
2921 tracing_disabled
= 1;
2926 max_tr
.entries
= size
;
2928 global_trace
.entries
= size
;
2933 static ssize_t
tracing_resize_ring_buffer(unsigned long size
)
2935 int cpu
, ret
= size
;
2937 mutex_lock(&trace_types_lock
);
2941 /* disable all cpu buffers */
2942 for_each_tracing_cpu(cpu
) {
2943 if (global_trace
.data
[cpu
])
2944 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2945 if (max_tr
.data
[cpu
])
2946 atomic_inc(&max_tr
.data
[cpu
]->disabled
);
2949 if (size
!= global_trace
.entries
)
2950 ret
= __tracing_resize_ring_buffer(size
);
2955 for_each_tracing_cpu(cpu
) {
2956 if (global_trace
.data
[cpu
])
2957 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2958 if (max_tr
.data
[cpu
])
2959 atomic_dec(&max_tr
.data
[cpu
]->disabled
);
2963 mutex_unlock(&trace_types_lock
);
2970 * tracing_update_buffers - used by tracing facility to expand ring buffers
2972 * To save on memory when the tracing is never used on a system with it
2973 * configured in. The ring buffers are set to a minimum size. But once
2974 * a user starts to use the tracing facility, then they need to grow
2975 * to their default size.
2977 * This function is to be called when a tracer is about to be used.
2979 int tracing_update_buffers(void)
2983 mutex_lock(&trace_types_lock
);
2984 if (!ring_buffer_expanded
)
2985 ret
= __tracing_resize_ring_buffer(trace_buf_size
);
2986 mutex_unlock(&trace_types_lock
);
2991 struct trace_option_dentry
;
2993 static struct trace_option_dentry
*
2994 create_trace_option_files(struct tracer
*tracer
);
2997 destroy_trace_option_files(struct trace_option_dentry
*topts
);
2999 static int tracing_set_tracer(const char *buf
)
3001 static struct trace_option_dentry
*topts
;
3002 struct trace_array
*tr
= &global_trace
;
3006 mutex_lock(&trace_types_lock
);
3008 if (!ring_buffer_expanded
) {
3009 ret
= __tracing_resize_ring_buffer(trace_buf_size
);
3015 for (t
= trace_types
; t
; t
= t
->next
) {
3016 if (strcmp(t
->name
, buf
) == 0)
3023 if (t
== current_trace
)
3026 trace_branch_disable();
3027 if (current_trace
&& current_trace
->reset
)
3028 current_trace
->reset(tr
);
3029 if (current_trace
&& current_trace
->use_max_tr
) {
3031 * We don't free the ring buffer. instead, resize it because
3032 * The max_tr ring buffer has some state (e.g. ring->clock) and
3033 * we want preserve it.
3035 ring_buffer_resize(max_tr
.buffer
, 1);
3038 destroy_trace_option_files(topts
);
3042 topts
= create_trace_option_files(current_trace
);
3043 if (current_trace
->use_max_tr
) {
3044 ret
= ring_buffer_resize(max_tr
.buffer
, global_trace
.entries
);
3047 max_tr
.entries
= global_trace
.entries
;
3051 ret
= tracer_init(t
, tr
);
3056 trace_branch_enable(tr
);
3058 mutex_unlock(&trace_types_lock
);
3064 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
3065 size_t cnt
, loff_t
*ppos
)
3067 char buf
[MAX_TRACER_SIZE
+1];
3074 if (cnt
> MAX_TRACER_SIZE
)
3075 cnt
= MAX_TRACER_SIZE
;
3077 if (copy_from_user(&buf
, ubuf
, cnt
))
3082 /* strip ending whitespace. */
3083 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
3086 err
= tracing_set_tracer(buf
);
3096 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
3097 size_t cnt
, loff_t
*ppos
)
3099 unsigned long *ptr
= filp
->private_data
;
3103 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
3104 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
3105 if (r
> sizeof(buf
))
3107 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3111 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
3112 size_t cnt
, loff_t
*ppos
)
3114 unsigned long *ptr
= filp
->private_data
;
3118 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3127 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
3129 long cpu_file
= (long) inode
->i_private
;
3130 struct trace_iterator
*iter
;
3133 if (tracing_disabled
)
3136 mutex_lock(&trace_types_lock
);
3138 /* create a buffer to store the information to pass to userspace */
3139 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
3146 * We make a copy of the current tracer to avoid concurrent
3147 * changes on it while we are reading.
3149 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
3155 *iter
->trace
= *current_trace
;
3157 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
3162 /* trace pipe does not show start of buffer */
3163 cpumask_setall(iter
->started
);
3165 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
3166 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
3168 iter
->cpu_file
= cpu_file
;
3169 iter
->tr
= &global_trace
;
3170 mutex_init(&iter
->mutex
);
3171 filp
->private_data
= iter
;
3173 if (iter
->trace
->pipe_open
)
3174 iter
->trace
->pipe_open(iter
);
3176 nonseekable_open(inode
, filp
);
3178 mutex_unlock(&trace_types_lock
);
3184 mutex_unlock(&trace_types_lock
);
3188 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
3190 struct trace_iterator
*iter
= file
->private_data
;
3192 mutex_lock(&trace_types_lock
);
3194 if (iter
->trace
->pipe_close
)
3195 iter
->trace
->pipe_close(iter
);
3197 mutex_unlock(&trace_types_lock
);
3199 free_cpumask_var(iter
->started
);
3200 mutex_destroy(&iter
->mutex
);
3208 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
3210 struct trace_iterator
*iter
= filp
->private_data
;
3212 if (trace_flags
& TRACE_ITER_BLOCK
) {
3214 * Always select as readable when in blocking mode
3216 return POLLIN
| POLLRDNORM
;
3218 if (!trace_empty(iter
))
3219 return POLLIN
| POLLRDNORM
;
3220 poll_wait(filp
, &trace_wait
, poll_table
);
3221 if (!trace_empty(iter
))
3222 return POLLIN
| POLLRDNORM
;
3229 void default_wait_pipe(struct trace_iterator
*iter
)
3233 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
3235 if (trace_empty(iter
))
3238 finish_wait(&trace_wait
, &wait
);
3242 * This is a make-shift waitqueue.
3243 * A tracer might use this callback on some rare cases:
3245 * 1) the current tracer might hold the runqueue lock when it wakes up
3246 * a reader, hence a deadlock (sched, function, and function graph tracers)
3247 * 2) the function tracers, trace all functions, we don't want
3248 * the overhead of calling wake_up and friends
3249 * (and tracing them too)
3251 * Anyway, this is really very primitive wakeup.
3253 void poll_wait_pipe(struct trace_iterator
*iter
)
3255 set_current_state(TASK_INTERRUPTIBLE
);
3256 /* sleep for 100 msecs, and try again. */
3257 schedule_timeout(HZ
/ 10);
3260 /* Must be called with trace_types_lock mutex held. */
3261 static int tracing_wait_pipe(struct file
*filp
)
3263 struct trace_iterator
*iter
= filp
->private_data
;
3265 while (trace_empty(iter
)) {
3267 if ((filp
->f_flags
& O_NONBLOCK
)) {
3271 mutex_unlock(&iter
->mutex
);
3273 iter
->trace
->wait_pipe(iter
);
3275 mutex_lock(&iter
->mutex
);
3277 if (signal_pending(current
))
3281 * We block until we read something and tracing is disabled.
3282 * We still block if tracing is disabled, but we have never
3283 * read anything. This allows a user to cat this file, and
3284 * then enable tracing. But after we have read something,
3285 * we give an EOF when tracing is again disabled.
3287 * iter->pos will be 0 if we haven't read anything.
3289 if (!tracer_enabled
&& iter
->pos
)
3300 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
3301 size_t cnt
, loff_t
*ppos
)
3303 struct trace_iterator
*iter
= filp
->private_data
;
3304 static struct tracer
*old_tracer
;
3307 /* return any leftover data */
3308 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3312 trace_seq_init(&iter
->seq
);
3314 /* copy the tracer to avoid using a global lock all around */
3315 mutex_lock(&trace_types_lock
);
3316 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3317 old_tracer
= current_trace
;
3318 *iter
->trace
= *current_trace
;
3320 mutex_unlock(&trace_types_lock
);
3323 * Avoid more than one consumer on a single file descriptor
3324 * This is just a matter of traces coherency, the ring buffer itself
3327 mutex_lock(&iter
->mutex
);
3328 if (iter
->trace
->read
) {
3329 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
3335 sret
= tracing_wait_pipe(filp
);
3339 /* stop when tracing is finished */
3340 if (trace_empty(iter
)) {
3345 if (cnt
>= PAGE_SIZE
)
3346 cnt
= PAGE_SIZE
- 1;
3348 /* reset all but tr, trace, and overruns */
3349 memset(&iter
->seq
, 0,
3350 sizeof(struct trace_iterator
) -
3351 offsetof(struct trace_iterator
, seq
));
3354 trace_event_read_lock();
3355 trace_access_lock(iter
->cpu_file
);
3356 while (trace_find_next_entry_inc(iter
) != NULL
) {
3357 enum print_line_t ret
;
3358 int len
= iter
->seq
.len
;
3360 ret
= print_trace_line(iter
);
3361 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3362 /* don't print partial lines */
3363 iter
->seq
.len
= len
;
3366 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3367 trace_consume(iter
);
3369 if (iter
->seq
.len
>= cnt
)
3373 * Setting the full flag means we reached the trace_seq buffer
3374 * size and we should leave by partial output condition above.
3375 * One of the trace_seq_* functions is not used properly.
3377 WARN_ONCE(iter
->seq
.full
, "full flag set for trace type %d",
3380 trace_access_unlock(iter
->cpu_file
);
3381 trace_event_read_unlock();
3383 /* Now copy what we have to the user */
3384 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3385 if (iter
->seq
.readpos
>= iter
->seq
.len
)
3386 trace_seq_init(&iter
->seq
);
3389 * If there was nothing to send to user, in spite of consuming trace
3390 * entries, go back to wait for more entries.
3396 mutex_unlock(&iter
->mutex
);
3401 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
3402 struct pipe_buffer
*buf
)
3404 __free_page(buf
->page
);
3407 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
3410 __free_page(spd
->pages
[idx
]);
3413 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
3415 .map
= generic_pipe_buf_map
,
3416 .unmap
= generic_pipe_buf_unmap
,
3417 .confirm
= generic_pipe_buf_confirm
,
3418 .release
= tracing_pipe_buf_release
,
3419 .steal
= generic_pipe_buf_steal
,
3420 .get
= generic_pipe_buf_get
,
3424 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
3429 /* Seq buffer is page-sized, exactly what we need. */
3431 count
= iter
->seq
.len
;
3432 ret
= print_trace_line(iter
);
3433 count
= iter
->seq
.len
- count
;
3436 iter
->seq
.len
-= count
;
3439 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3440 iter
->seq
.len
-= count
;
3444 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3445 trace_consume(iter
);
3447 if (!trace_find_next_entry_inc(iter
)) {
3457 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3459 struct pipe_inode_info
*pipe
,
3463 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3464 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3465 struct trace_iterator
*iter
= filp
->private_data
;
3466 struct splice_pipe_desc spd
= {
3468 .partial
= partial_def
,
3469 .nr_pages
= 0, /* This gets updated below. */
3471 .ops
= &tracing_pipe_buf_ops
,
3472 .spd_release
= tracing_spd_release_pipe
,
3474 static struct tracer
*old_tracer
;
3479 if (splice_grow_spd(pipe
, &spd
))
3482 /* copy the tracer to avoid using a global lock all around */
3483 mutex_lock(&trace_types_lock
);
3484 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3485 old_tracer
= current_trace
;
3486 *iter
->trace
= *current_trace
;
3488 mutex_unlock(&trace_types_lock
);
3490 mutex_lock(&iter
->mutex
);
3492 if (iter
->trace
->splice_read
) {
3493 ret
= iter
->trace
->splice_read(iter
, filp
,
3494 ppos
, pipe
, len
, flags
);
3499 ret
= tracing_wait_pipe(filp
);
3503 if (!iter
->ent
&& !trace_find_next_entry_inc(iter
)) {
3508 trace_event_read_lock();
3509 trace_access_lock(iter
->cpu_file
);
3511 /* Fill as many pages as possible. */
3512 for (i
= 0, rem
= len
; i
< pipe
->buffers
&& rem
; i
++) {
3513 spd
.pages
[i
] = alloc_page(GFP_KERNEL
);
3517 rem
= tracing_fill_pipe_page(rem
, iter
);
3519 /* Copy the data into the page, so we can start over. */
3520 ret
= trace_seq_to_buffer(&iter
->seq
,
3521 page_address(spd
.pages
[i
]),
3524 __free_page(spd
.pages
[i
]);
3527 spd
.partial
[i
].offset
= 0;
3528 spd
.partial
[i
].len
= iter
->seq
.len
;
3530 trace_seq_init(&iter
->seq
);
3533 trace_access_unlock(iter
->cpu_file
);
3534 trace_event_read_unlock();
3535 mutex_unlock(&iter
->mutex
);
3539 ret
= splice_to_pipe(pipe
, &spd
);
3541 splice_shrink_spd(pipe
, &spd
);
3545 mutex_unlock(&iter
->mutex
);
3550 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3551 size_t cnt
, loff_t
*ppos
)
3553 struct trace_array
*tr
= filp
->private_data
;
3557 mutex_lock(&trace_types_lock
);
3558 if (!ring_buffer_expanded
)
3559 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3561 trace_buf_size
>> 10);
3563 r
= sprintf(buf
, "%lu\n", tr
->entries
>> 10);
3564 mutex_unlock(&trace_types_lock
);
3566 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3570 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3571 size_t cnt
, loff_t
*ppos
)
3576 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3580 /* must have at least 1 entry */
3584 /* value is in KB */
3587 ret
= tracing_resize_ring_buffer(val
);
3597 tracing_total_entries_read(struct file
*filp
, char __user
*ubuf
,
3598 size_t cnt
, loff_t
*ppos
)
3600 struct trace_array
*tr
= filp
->private_data
;
3603 unsigned long size
= 0, expanded_size
= 0;
3605 mutex_lock(&trace_types_lock
);
3606 for_each_tracing_cpu(cpu
) {
3607 size
+= tr
->entries
>> 10;
3608 if (!ring_buffer_expanded
)
3609 expanded_size
+= trace_buf_size
>> 10;
3611 if (ring_buffer_expanded
)
3612 r
= sprintf(buf
, "%lu\n", size
);
3614 r
= sprintf(buf
, "%lu (expanded: %lu)\n", size
, expanded_size
);
3615 mutex_unlock(&trace_types_lock
);
3617 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3621 tracing_free_buffer_write(struct file
*filp
, const char __user
*ubuf
,
3622 size_t cnt
, loff_t
*ppos
)
3625 * There is no need to read what the user has written, this function
3626 * is just to make sure that there is no error when "echo" is used
3635 tracing_free_buffer_release(struct inode
*inode
, struct file
*filp
)
3637 /* disable tracing ? */
3638 if (trace_flags
& TRACE_ITER_STOP_ON_FREE
)
3640 /* resize the ring buffer to 0 */
3641 tracing_resize_ring_buffer(0);
3647 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3648 size_t cnt
, loff_t
*fpos
)
3650 unsigned long addr
= (unsigned long)ubuf
;
3651 struct ring_buffer_event
*event
;
3652 struct ring_buffer
*buffer
;
3653 struct print_entry
*entry
;
3654 unsigned long irq_flags
;
3655 struct page
*pages
[2];
3665 if (tracing_disabled
)
3668 if (cnt
> TRACE_BUF_SIZE
)
3669 cnt
= TRACE_BUF_SIZE
;
3672 * Userspace is injecting traces into the kernel trace buffer.
3673 * We want to be as non intrusive as possible.
3674 * To do so, we do not want to allocate any special buffers
3675 * or take any locks, but instead write the userspace data
3676 * straight into the ring buffer.
3678 * First we need to pin the userspace buffer into memory,
3679 * which, most likely it is, because it just referenced it.
3680 * But there's no guarantee that it is. By using get_user_pages_fast()
3681 * and kmap_atomic/kunmap_atomic() we can get access to the
3682 * pages directly. We then write the data directly into the
3685 BUILD_BUG_ON(TRACE_BUF_SIZE
>= PAGE_SIZE
);
3687 /* check if we cross pages */
3688 if ((addr
& PAGE_MASK
) != ((addr
+ cnt
) & PAGE_MASK
))
3691 offset
= addr
& (PAGE_SIZE
- 1);
3694 ret
= get_user_pages_fast(addr
, nr_pages
, 0, pages
);
3695 if (ret
< nr_pages
) {
3697 put_page(pages
[ret
]);
3702 page1
= kmap_atomic(pages
[0]);
3704 page2
= kmap_atomic(pages
[1]);
3706 local_save_flags(irq_flags
);
3707 size
= sizeof(*entry
) + cnt
+ 2; /* possible \n added */
3708 buffer
= global_trace
.buffer
;
3709 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
3710 irq_flags
, preempt_count());
3712 /* Ring buffer disabled, return as if not open for write */
3717 entry
= ring_buffer_event_data(event
);
3718 entry
->ip
= _THIS_IP_
;
3720 if (nr_pages
== 2) {
3721 len
= PAGE_SIZE
- offset
;
3722 memcpy(&entry
->buf
, page1
+ offset
, len
);
3723 memcpy(&entry
->buf
[len
], page2
, cnt
- len
);
3725 memcpy(&entry
->buf
, page1
+ offset
, cnt
);
3727 if (entry
->buf
[cnt
- 1] != '\n') {
3728 entry
->buf
[cnt
] = '\n';
3729 entry
->buf
[cnt
+ 1] = '\0';
3731 entry
->buf
[cnt
] = '\0';
3733 ring_buffer_unlock_commit(buffer
, event
);
3741 kunmap_atomic(page2
);
3742 kunmap_atomic(page1
);
3743 while (nr_pages
> 0)
3744 put_page(pages
[--nr_pages
]);
3749 static int tracing_clock_show(struct seq_file
*m
, void *v
)
3753 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
3755 "%s%s%s%s", i
? " " : "",
3756 i
== trace_clock_id
? "[" : "", trace_clocks
[i
].name
,
3757 i
== trace_clock_id
? "]" : "");
3763 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
3764 size_t cnt
, loff_t
*fpos
)
3767 const char *clockstr
;
3770 if (cnt
>= sizeof(buf
))
3773 if (copy_from_user(&buf
, ubuf
, cnt
))
3778 clockstr
= strstrip(buf
);
3780 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
3781 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
3784 if (i
== ARRAY_SIZE(trace_clocks
))
3789 mutex_lock(&trace_types_lock
);
3791 ring_buffer_set_clock(global_trace
.buffer
, trace_clocks
[i
].func
);
3793 ring_buffer_set_clock(max_tr
.buffer
, trace_clocks
[i
].func
);
3795 mutex_unlock(&trace_types_lock
);
3802 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
3804 if (tracing_disabled
)
3806 return single_open(file
, tracing_clock_show
, NULL
);
3809 static const struct file_operations tracing_max_lat_fops
= {
3810 .open
= tracing_open_generic
,
3811 .read
= tracing_max_lat_read
,
3812 .write
= tracing_max_lat_write
,
3813 .llseek
= generic_file_llseek
,
3816 static const struct file_operations tracing_ctrl_fops
= {
3817 .open
= tracing_open_generic
,
3818 .read
= tracing_ctrl_read
,
3819 .write
= tracing_ctrl_write
,
3820 .llseek
= generic_file_llseek
,
3823 static const struct file_operations set_tracer_fops
= {
3824 .open
= tracing_open_generic
,
3825 .read
= tracing_set_trace_read
,
3826 .write
= tracing_set_trace_write
,
3827 .llseek
= generic_file_llseek
,
3830 static const struct file_operations tracing_pipe_fops
= {
3831 .open
= tracing_open_pipe
,
3832 .poll
= tracing_poll_pipe
,
3833 .read
= tracing_read_pipe
,
3834 .splice_read
= tracing_splice_read_pipe
,
3835 .release
= tracing_release_pipe
,
3836 .llseek
= no_llseek
,
3839 static const struct file_operations tracing_entries_fops
= {
3840 .open
= tracing_open_generic
,
3841 .read
= tracing_entries_read
,
3842 .write
= tracing_entries_write
,
3843 .llseek
= generic_file_llseek
,
3846 static const struct file_operations tracing_total_entries_fops
= {
3847 .open
= tracing_open_generic
,
3848 .read
= tracing_total_entries_read
,
3849 .llseek
= generic_file_llseek
,
3852 static const struct file_operations tracing_free_buffer_fops
= {
3853 .write
= tracing_free_buffer_write
,
3854 .release
= tracing_free_buffer_release
,
3857 static const struct file_operations tracing_mark_fops
= {
3858 .open
= tracing_open_generic
,
3859 .write
= tracing_mark_write
,
3860 .llseek
= generic_file_llseek
,
3863 static const struct file_operations trace_clock_fops
= {
3864 .open
= tracing_clock_open
,
3866 .llseek
= seq_lseek
,
3867 .release
= single_release
,
3868 .write
= tracing_clock_write
,
3871 struct ftrace_buffer_info
{
3872 struct trace_array
*tr
;
3878 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
3880 int cpu
= (int)(long)inode
->i_private
;
3881 struct ftrace_buffer_info
*info
;
3883 if (tracing_disabled
)
3886 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3890 info
->tr
= &global_trace
;
3893 /* Force reading ring buffer for first read */
3894 info
->read
= (unsigned int)-1;
3896 filp
->private_data
= info
;
3898 return nonseekable_open(inode
, filp
);
3902 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
3903 size_t count
, loff_t
*ppos
)
3905 struct ftrace_buffer_info
*info
= filp
->private_data
;
3913 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
, info
->cpu
);
3917 /* Do we have previous read data to read? */
3918 if (info
->read
< PAGE_SIZE
)
3921 trace_access_lock(info
->cpu
);
3922 ret
= ring_buffer_read_page(info
->tr
->buffer
,
3926 trace_access_unlock(info
->cpu
);
3933 size
= PAGE_SIZE
- info
->read
;
3937 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
3948 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
3950 struct ftrace_buffer_info
*info
= file
->private_data
;
3953 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
3960 struct ring_buffer
*buffer
;
3965 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
3966 struct pipe_buffer
*buf
)
3968 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3973 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
3978 static int buffer_pipe_buf_steal(struct pipe_inode_info
*pipe
,
3979 struct pipe_buffer
*buf
)
3984 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
3985 struct pipe_buffer
*buf
)
3987 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
3992 /* Pipe buffer operations for a buffer. */
3993 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
3995 .map
= generic_pipe_buf_map
,
3996 .unmap
= generic_pipe_buf_unmap
,
3997 .confirm
= generic_pipe_buf_confirm
,
3998 .release
= buffer_pipe_buf_release
,
3999 .steal
= buffer_pipe_buf_steal
,
4000 .get
= buffer_pipe_buf_get
,
4004 * Callback from splice_to_pipe(), if we need to release some pages
4005 * at the end of the spd in case we error'ed out in filling the pipe.
4007 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
4009 struct buffer_ref
*ref
=
4010 (struct buffer_ref
*)spd
->partial
[i
].private;
4015 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4017 spd
->partial
[i
].private = 0;
4021 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
4022 struct pipe_inode_info
*pipe
, size_t len
,
4025 struct ftrace_buffer_info
*info
= file
->private_data
;
4026 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
4027 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
4028 struct splice_pipe_desc spd
= {
4030 .partial
= partial_def
,
4032 .ops
= &buffer_pipe_buf_ops
,
4033 .spd_release
= buffer_spd_release
,
4035 struct buffer_ref
*ref
;
4036 int entries
, size
, i
;
4039 if (splice_grow_spd(pipe
, &spd
))
4042 if (*ppos
& (PAGE_SIZE
- 1)) {
4043 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
4048 if (len
& (PAGE_SIZE
- 1)) {
4049 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
4050 if (len
< PAGE_SIZE
) {
4057 trace_access_lock(info
->cpu
);
4058 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4060 for (i
= 0; i
< pipe
->buffers
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
4064 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
4069 ref
->buffer
= info
->tr
->buffer
;
4070 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
, info
->cpu
);
4076 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
4079 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4085 * zero out any left over data, this is going to
4088 size
= ring_buffer_page_len(ref
->page
);
4089 if (size
< PAGE_SIZE
)
4090 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
4092 page
= virt_to_page(ref
->page
);
4094 spd
.pages
[i
] = page
;
4095 spd
.partial
[i
].len
= PAGE_SIZE
;
4096 spd
.partial
[i
].offset
= 0;
4097 spd
.partial
[i
].private = (unsigned long)ref
;
4101 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4104 trace_access_unlock(info
->cpu
);
4107 /* did we read anything? */
4108 if (!spd
.nr_pages
) {
4109 if (flags
& SPLICE_F_NONBLOCK
)
4117 ret
= splice_to_pipe(pipe
, &spd
);
4118 splice_shrink_spd(pipe
, &spd
);
4123 static const struct file_operations tracing_buffers_fops
= {
4124 .open
= tracing_buffers_open
,
4125 .read
= tracing_buffers_read
,
4126 .release
= tracing_buffers_release
,
4127 .splice_read
= tracing_buffers_splice_read
,
4128 .llseek
= no_llseek
,
4132 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
4133 size_t count
, loff_t
*ppos
)
4135 unsigned long cpu
= (unsigned long)filp
->private_data
;
4136 struct trace_array
*tr
= &global_trace
;
4137 struct trace_seq
*s
;
4139 unsigned long long t
;
4140 unsigned long usec_rem
;
4142 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
4148 cnt
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
4149 trace_seq_printf(s
, "entries: %ld\n", cnt
);
4151 cnt
= ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
4152 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
4154 cnt
= ring_buffer_commit_overrun_cpu(tr
->buffer
, cpu
);
4155 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
4157 cnt
= ring_buffer_bytes_cpu(tr
->buffer
, cpu
);
4158 trace_seq_printf(s
, "bytes: %ld\n", cnt
);
4160 t
= ns2usecs(ring_buffer_oldest_event_ts(tr
->buffer
, cpu
));
4161 usec_rem
= do_div(t
, USEC_PER_SEC
);
4162 trace_seq_printf(s
, "oldest event ts: %5llu.%06lu\n", t
, usec_rem
);
4164 t
= ns2usecs(ring_buffer_time_stamp(tr
->buffer
, cpu
));
4165 usec_rem
= do_div(t
, USEC_PER_SEC
);
4166 trace_seq_printf(s
, "now ts: %5llu.%06lu\n", t
, usec_rem
);
4168 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
4175 static const struct file_operations tracing_stats_fops
= {
4176 .open
= tracing_open_generic
,
4177 .read
= tracing_stats_read
,
4178 .llseek
= generic_file_llseek
,
4181 #ifdef CONFIG_DYNAMIC_FTRACE
4183 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
4189 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
4190 size_t cnt
, loff_t
*ppos
)
4192 static char ftrace_dyn_info_buffer
[1024];
4193 static DEFINE_MUTEX(dyn_info_mutex
);
4194 unsigned long *p
= filp
->private_data
;
4195 char *buf
= ftrace_dyn_info_buffer
;
4196 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
4199 mutex_lock(&dyn_info_mutex
);
4200 r
= sprintf(buf
, "%ld ", *p
);
4202 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
4205 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4207 mutex_unlock(&dyn_info_mutex
);
4212 static const struct file_operations tracing_dyn_info_fops
= {
4213 .open
= tracing_open_generic
,
4214 .read
= tracing_read_dyn_info
,
4215 .llseek
= generic_file_llseek
,
4219 static struct dentry
*d_tracer
;
4221 struct dentry
*tracing_init_dentry(void)
4228 if (!debugfs_initialized())
4231 d_tracer
= debugfs_create_dir("tracing", NULL
);
4233 if (!d_tracer
&& !once
) {
4235 pr_warning("Could not create debugfs directory 'tracing'\n");
4242 static struct dentry
*d_percpu
;
4244 struct dentry
*tracing_dentry_percpu(void)
4247 struct dentry
*d_tracer
;
4252 d_tracer
= tracing_init_dentry();
4257 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
4259 if (!d_percpu
&& !once
) {
4261 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4268 static void tracing_init_debugfs_percpu(long cpu
)
4270 struct dentry
*d_percpu
= tracing_dentry_percpu();
4271 struct dentry
*d_cpu
;
4272 char cpu_dir
[30]; /* 30 characters should be more than enough */
4274 snprintf(cpu_dir
, 30, "cpu%ld", cpu
);
4275 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
4277 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
4281 /* per cpu trace_pipe */
4282 trace_create_file("trace_pipe", 0444, d_cpu
,
4283 (void *) cpu
, &tracing_pipe_fops
);
4286 trace_create_file("trace", 0644, d_cpu
,
4287 (void *) cpu
, &tracing_fops
);
4289 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
4290 (void *) cpu
, &tracing_buffers_fops
);
4292 trace_create_file("stats", 0444, d_cpu
,
4293 (void *) cpu
, &tracing_stats_fops
);
4296 #ifdef CONFIG_FTRACE_SELFTEST
4297 /* Let selftest have access to static functions in this file */
4298 #include "trace_selftest.c"
4301 struct trace_option_dentry
{
4302 struct tracer_opt
*opt
;
4303 struct tracer_flags
*flags
;
4304 struct dentry
*entry
;
4308 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4311 struct trace_option_dentry
*topt
= filp
->private_data
;
4314 if (topt
->flags
->val
& topt
->opt
->bit
)
4319 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4323 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4326 struct trace_option_dentry
*topt
= filp
->private_data
;
4330 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4334 if (val
!= 0 && val
!= 1)
4337 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
4338 mutex_lock(&trace_types_lock
);
4339 ret
= __set_tracer_option(current_trace
, topt
->flags
,
4341 mutex_unlock(&trace_types_lock
);
4352 static const struct file_operations trace_options_fops
= {
4353 .open
= tracing_open_generic
,
4354 .read
= trace_options_read
,
4355 .write
= trace_options_write
,
4356 .llseek
= generic_file_llseek
,
4360 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4363 long index
= (long)filp
->private_data
;
4366 if (trace_flags
& (1 << index
))
4371 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4375 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4378 long index
= (long)filp
->private_data
;
4382 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4386 if (val
!= 0 && val
!= 1)
4388 set_tracer_flags(1 << index
, val
);
4395 static const struct file_operations trace_options_core_fops
= {
4396 .open
= tracing_open_generic
,
4397 .read
= trace_options_core_read
,
4398 .write
= trace_options_core_write
,
4399 .llseek
= generic_file_llseek
,
4402 struct dentry
*trace_create_file(const char *name
,
4404 struct dentry
*parent
,
4406 const struct file_operations
*fops
)
4410 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
4412 pr_warning("Could not create debugfs '%s' entry\n", name
);
4418 static struct dentry
*trace_options_init_dentry(void)
4420 struct dentry
*d_tracer
;
4421 static struct dentry
*t_options
;
4426 d_tracer
= tracing_init_dentry();
4430 t_options
= debugfs_create_dir("options", d_tracer
);
4432 pr_warning("Could not create debugfs directory 'options'\n");
4440 create_trace_option_file(struct trace_option_dentry
*topt
,
4441 struct tracer_flags
*flags
,
4442 struct tracer_opt
*opt
)
4444 struct dentry
*t_options
;
4446 t_options
= trace_options_init_dentry();
4450 topt
->flags
= flags
;
4453 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
4454 &trace_options_fops
);
4458 static struct trace_option_dentry
*
4459 create_trace_option_files(struct tracer
*tracer
)
4461 struct trace_option_dentry
*topts
;
4462 struct tracer_flags
*flags
;
4463 struct tracer_opt
*opts
;
4469 flags
= tracer
->flags
;
4471 if (!flags
|| !flags
->opts
)
4476 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4479 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
4483 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4484 create_trace_option_file(&topts
[cnt
], flags
,
4491 destroy_trace_option_files(struct trace_option_dentry
*topts
)
4498 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
4499 if (topts
[cnt
].entry
)
4500 debugfs_remove(topts
[cnt
].entry
);
4506 static struct dentry
*
4507 create_trace_option_core_file(const char *option
, long index
)
4509 struct dentry
*t_options
;
4511 t_options
= trace_options_init_dentry();
4515 return trace_create_file(option
, 0644, t_options
, (void *)index
,
4516 &trace_options_core_fops
);
4519 static __init
void create_trace_options_dir(void)
4521 struct dentry
*t_options
;
4524 t_options
= trace_options_init_dentry();
4528 for (i
= 0; trace_options
[i
]; i
++)
4529 create_trace_option_core_file(trace_options
[i
], i
);
4532 static __init
int tracer_init_debugfs(void)
4534 struct dentry
*d_tracer
;
4537 trace_access_lock_init();
4539 d_tracer
= tracing_init_dentry();
4541 trace_create_file("tracing_enabled", 0644, d_tracer
,
4542 &global_trace
, &tracing_ctrl_fops
);
4544 trace_create_file("trace_options", 0644, d_tracer
,
4545 NULL
, &tracing_iter_fops
);
4547 trace_create_file("tracing_cpumask", 0644, d_tracer
,
4548 NULL
, &tracing_cpumask_fops
);
4550 trace_create_file("trace", 0644, d_tracer
,
4551 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
4553 trace_create_file("available_tracers", 0444, d_tracer
,
4554 &global_trace
, &show_traces_fops
);
4556 trace_create_file("current_tracer", 0644, d_tracer
,
4557 &global_trace
, &set_tracer_fops
);
4559 #ifdef CONFIG_TRACER_MAX_TRACE
4560 trace_create_file("tracing_max_latency", 0644, d_tracer
,
4561 &tracing_max_latency
, &tracing_max_lat_fops
);
4564 trace_create_file("tracing_thresh", 0644, d_tracer
,
4565 &tracing_thresh
, &tracing_max_lat_fops
);
4567 trace_create_file("README", 0444, d_tracer
,
4568 NULL
, &tracing_readme_fops
);
4570 trace_create_file("trace_pipe", 0444, d_tracer
,
4571 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
4573 trace_create_file("buffer_size_kb", 0644, d_tracer
,
4574 &global_trace
, &tracing_entries_fops
);
4576 trace_create_file("buffer_total_size_kb", 0444, d_tracer
,
4577 &global_trace
, &tracing_total_entries_fops
);
4579 trace_create_file("free_buffer", 0644, d_tracer
,
4580 &global_trace
, &tracing_free_buffer_fops
);
4582 trace_create_file("trace_marker", 0220, d_tracer
,
4583 NULL
, &tracing_mark_fops
);
4585 trace_create_file("saved_cmdlines", 0444, d_tracer
,
4586 NULL
, &tracing_saved_cmdlines_fops
);
4588 trace_create_file("trace_clock", 0644, d_tracer
, NULL
,
4591 #ifdef CONFIG_DYNAMIC_FTRACE
4592 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
4593 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
4596 create_trace_options_dir();
4598 for_each_tracing_cpu(cpu
)
4599 tracing_init_debugfs_percpu(cpu
);
4604 static int trace_panic_handler(struct notifier_block
*this,
4605 unsigned long event
, void *unused
)
4607 if (ftrace_dump_on_oops
)
4608 ftrace_dump(ftrace_dump_on_oops
);
4612 static struct notifier_block trace_panic_notifier
= {
4613 .notifier_call
= trace_panic_handler
,
4615 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
4618 static int trace_die_handler(struct notifier_block
*self
,
4624 if (ftrace_dump_on_oops
)
4625 ftrace_dump(ftrace_dump_on_oops
);
4633 static struct notifier_block trace_die_notifier
= {
4634 .notifier_call
= trace_die_handler
,
4639 * printk is set to max of 1024, we really don't need it that big.
4640 * Nothing should be printing 1000 characters anyway.
4642 #define TRACE_MAX_PRINT 1000
4645 * Define here KERN_TRACE so that we have one place to modify
4646 * it if we decide to change what log level the ftrace dump
4649 #define KERN_TRACE KERN_EMERG
4652 trace_printk_seq(struct trace_seq
*s
)
4654 /* Probably should print a warning here. */
4658 /* should be zero ended, but we are paranoid. */
4659 s
->buffer
[s
->len
] = 0;
4661 printk(KERN_TRACE
"%s", s
->buffer
);
4666 void trace_init_global_iter(struct trace_iterator
*iter
)
4668 iter
->tr
= &global_trace
;
4669 iter
->trace
= current_trace
;
4670 iter
->cpu_file
= TRACE_PIPE_ALL_CPU
;
4674 __ftrace_dump(bool disable_tracing
, enum ftrace_dump_mode oops_dump_mode
)
4676 static arch_spinlock_t ftrace_dump_lock
=
4677 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
4678 /* use static because iter can be a bit big for the stack */
4679 static struct trace_iterator iter
;
4680 unsigned int old_userobj
;
4681 static int dump_ran
;
4682 unsigned long flags
;
4686 local_irq_save(flags
);
4687 arch_spin_lock(&ftrace_dump_lock
);
4695 /* Did function tracer already get disabled? */
4696 if (ftrace_is_dead()) {
4697 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
4698 printk("# MAY BE MISSING FUNCTION EVENTS\n");
4701 if (disable_tracing
)
4704 trace_init_global_iter(&iter
);
4706 for_each_tracing_cpu(cpu
) {
4707 atomic_inc(&iter
.tr
->data
[cpu
]->disabled
);
4710 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
4712 /* don't look at user memory in panic mode */
4713 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
4715 /* Simulate the iterator */
4716 iter
.tr
= &global_trace
;
4717 iter
.trace
= current_trace
;
4719 switch (oops_dump_mode
) {
4721 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4724 iter
.cpu_file
= raw_smp_processor_id();
4729 printk(KERN_TRACE
"Bad dumping mode, switching to all CPUs dump\n");
4730 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4733 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
4736 * We need to stop all tracing on all CPUS to read the
4737 * the next buffer. This is a bit expensive, but is
4738 * not done often. We fill all what we can read,
4739 * and then release the locks again.
4742 while (!trace_empty(&iter
)) {
4745 printk(KERN_TRACE
"---------------------------------\n");
4749 /* reset all but tr, trace, and overruns */
4750 memset(&iter
.seq
, 0,
4751 sizeof(struct trace_iterator
) -
4752 offsetof(struct trace_iterator
, seq
));
4753 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
4756 if (trace_find_next_entry_inc(&iter
) != NULL
) {
4759 ret
= print_trace_line(&iter
);
4760 if (ret
!= TRACE_TYPE_NO_CONSUME
)
4761 trace_consume(&iter
);
4764 trace_printk_seq(&iter
.seq
);
4768 printk(KERN_TRACE
" (ftrace buffer empty)\n");
4770 printk(KERN_TRACE
"---------------------------------\n");
4773 /* Re-enable tracing if requested */
4774 if (!disable_tracing
) {
4775 trace_flags
|= old_userobj
;
4777 for_each_tracing_cpu(cpu
) {
4778 atomic_dec(&iter
.tr
->data
[cpu
]->disabled
);
4784 arch_spin_unlock(&ftrace_dump_lock
);
4785 local_irq_restore(flags
);
4788 /* By default: disable tracing after the dump */
4789 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode
)
4791 __ftrace_dump(true, oops_dump_mode
);
4794 __init
static int tracer_alloc_buffers(void)
4797 enum ring_buffer_flags rb_flags
;
4802 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
4805 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
4806 goto out_free_buffer_mask
;
4808 /* To save memory, keep the ring buffer size to its minimum */
4809 if (ring_buffer_expanded
)
4810 ring_buf_size
= trace_buf_size
;
4814 rb_flags
= trace_flags
& TRACE_ITER_OVERWRITE
? RB_FL_OVERWRITE
: 0;
4816 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
4817 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
4819 /* TODO: make the number of buffers hot pluggable with CPUS */
4820 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
, rb_flags
);
4821 if (!global_trace
.buffer
) {
4822 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
4824 goto out_free_cpumask
;
4826 global_trace
.entries
= ring_buffer_size(global_trace
.buffer
);
4829 #ifdef CONFIG_TRACER_MAX_TRACE
4830 max_tr
.buffer
= ring_buffer_alloc(1, rb_flags
);
4831 if (!max_tr
.buffer
) {
4832 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
4834 ring_buffer_free(global_trace
.buffer
);
4835 goto out_free_cpumask
;
4840 /* Allocate the first page for all buffers */
4841 for_each_tracing_cpu(i
) {
4842 global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
4843 max_tr
.data
[i
] = &per_cpu(max_tr_data
, i
);
4846 trace_init_cmdlines();
4848 register_tracer(&nop_trace
);
4849 current_trace
= &nop_trace
;
4850 /* All seems OK, enable tracing */
4851 tracing_disabled
= 0;
4853 atomic_notifier_chain_register(&panic_notifier_list
,
4854 &trace_panic_notifier
);
4856 register_die_notifier(&trace_die_notifier
);
4861 free_cpumask_var(tracing_cpumask
);
4862 out_free_buffer_mask
:
4863 free_cpumask_var(tracing_buffer_mask
);
4868 __init
static int clear_boot_tracer(void)
4871 * The default tracer at boot buffer is an init section.
4872 * This function is called in lateinit. If we did not
4873 * find the boot tracer, then clear it out, to prevent
4874 * later registration from accessing the buffer that is
4875 * about to be freed.
4877 if (!default_bootup_tracer
)
4880 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
4881 default_bootup_tracer
);
4882 default_bootup_tracer
= NULL
;
4887 early_initcall(tracer_alloc_buffers
);
4888 fs_initcall(tracer_init_debugfs
);
4889 late_initcall(clear_boot_tracer
);