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>
39 #include <linux/nmi.h>
43 #include "trace_output.h"
46 * On boot up, the ring buffer is set to the minimum size, so that
47 * we do not waste memory on systems that are not using tracing.
49 int ring_buffer_expanded
;
52 * We need to change this state when a selftest is running.
53 * A selftest will lurk into the ring-buffer to count the
54 * entries inserted during the selftest although some concurrent
55 * insertions into the ring-buffer such as trace_printk could occurred
56 * at the same time, giving false positive or negative results.
58 static bool __read_mostly tracing_selftest_running
;
61 * If a tracer is running, we do not want to run SELFTEST.
63 bool __read_mostly tracing_selftest_disabled
;
65 /* For tracers that don't implement custom flags */
66 static struct tracer_opt dummy_tracer_opt
[] = {
70 static struct tracer_flags dummy_tracer_flags
= {
72 .opts
= dummy_tracer_opt
75 static int dummy_set_flag(u32 old_flags
, u32 bit
, int set
)
81 * To prevent the comm cache from being overwritten when no
82 * tracing is active, only save the comm when a trace event
85 static DEFINE_PER_CPU(bool, trace_cmdline_save
);
88 * Kill all tracing for good (never come back).
89 * It is initialized to 1 but will turn to zero if the initialization
90 * of the tracer is successful. But that is the only place that sets
93 static int tracing_disabled
= 1;
95 DEFINE_PER_CPU(int, ftrace_cpu_disabled
);
97 cpumask_var_t __read_mostly tracing_buffer_mask
;
100 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
102 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
103 * is set, then ftrace_dump is called. This will output the contents
104 * of the ftrace buffers to the console. This is very useful for
105 * capturing traces that lead to crashes and outputing it to a
108 * It is default off, but you can enable it with either specifying
109 * "ftrace_dump_on_oops" in the kernel command line, or setting
110 * /proc/sys/kernel/ftrace_dump_on_oops
111 * Set 1 if you want to dump buffers of all CPUs
112 * Set 2 if you want to dump the buffer of the CPU that triggered oops
115 enum ftrace_dump_mode ftrace_dump_on_oops
;
117 static int tracing_set_tracer(const char *buf
);
119 #define MAX_TRACER_SIZE 100
120 static char bootup_tracer_buf
[MAX_TRACER_SIZE
] __initdata
;
121 static char *default_bootup_tracer
;
123 static int __init
set_cmdline_ftrace(char *str
)
125 strncpy(bootup_tracer_buf
, str
, MAX_TRACER_SIZE
);
126 default_bootup_tracer
= bootup_tracer_buf
;
127 /* We are using ftrace early, expand it */
128 ring_buffer_expanded
= 1;
131 __setup("ftrace=", set_cmdline_ftrace
);
133 static int __init
set_ftrace_dump_on_oops(char *str
)
135 if (*str
++ != '=' || !*str
) {
136 ftrace_dump_on_oops
= DUMP_ALL
;
140 if (!strcmp("orig_cpu", str
)) {
141 ftrace_dump_on_oops
= DUMP_ORIG
;
147 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops
);
149 unsigned long long ns2usecs(cycle_t nsec
)
157 * The global_trace is the descriptor that holds the tracing
158 * buffers for the live tracing. For each CPU, it contains
159 * a link list of pages that will store trace entries. The
160 * page descriptor of the pages in the memory is used to hold
161 * the link list by linking the lru item in the page descriptor
162 * to each of the pages in the buffer per CPU.
164 * For each active CPU there is a data field that holds the
165 * pages for the buffer for that CPU. Each CPU has the same number
166 * of pages allocated for its buffer.
168 static struct trace_array global_trace
;
170 static DEFINE_PER_CPU(struct trace_array_cpu
, global_trace_cpu
);
172 int filter_current_check_discard(struct ring_buffer
*buffer
,
173 struct ftrace_event_call
*call
, void *rec
,
174 struct ring_buffer_event
*event
)
176 return filter_check_discard(call
, rec
, buffer
, event
);
178 EXPORT_SYMBOL_GPL(filter_current_check_discard
);
180 cycle_t
ftrace_now(int cpu
)
184 /* Early boot up does not have a buffer yet */
185 if (!global_trace
.buffer
)
186 return trace_clock_local();
188 ts
= ring_buffer_time_stamp(global_trace
.buffer
, cpu
);
189 ring_buffer_normalize_time_stamp(global_trace
.buffer
, cpu
, &ts
);
195 * The max_tr is used to snapshot the global_trace when a maximum
196 * latency is reached. Some tracers will use this to store a maximum
197 * trace while it continues examining live traces.
199 * The buffers for the max_tr are set up the same as the global_trace.
200 * When a snapshot is taken, the link list of the max_tr is swapped
201 * with the link list of the global_trace and the buffers are reset for
202 * the global_trace so the tracing can continue.
204 static struct trace_array max_tr
;
206 static DEFINE_PER_CPU(struct trace_array_cpu
, max_tr_data
);
208 int tracing_is_enabled(void)
210 return tracing_is_on();
214 * trace_buf_size is the size in bytes that is allocated
215 * for a buffer. Note, the number of bytes is always rounded
218 * This number is purposely set to a low number of 16384.
219 * If the dump on oops happens, it will be much appreciated
220 * to not have to wait for all that output. Anyway this can be
221 * boot time and run time configurable.
223 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
225 static unsigned long trace_buf_size
= TRACE_BUF_SIZE_DEFAULT
;
227 /* trace_types holds a link list of available tracers. */
228 static struct tracer
*trace_types __read_mostly
;
230 /* current_trace points to the tracer that is currently active */
231 static struct tracer
*current_trace __read_mostly
;
234 * trace_types_lock is used to protect the trace_types list.
236 static DEFINE_MUTEX(trace_types_lock
);
239 * serialize the access of the ring buffer
241 * ring buffer serializes readers, but it is low level protection.
242 * The validity of the events (which returns by ring_buffer_peek() ..etc)
243 * are not protected by ring buffer.
245 * The content of events may become garbage if we allow other process consumes
246 * these events concurrently:
247 * A) the page of the consumed events may become a normal page
248 * (not reader page) in ring buffer, and this page will be rewrited
249 * by events producer.
250 * B) The page of the consumed events may become a page for splice_read,
251 * and this page will be returned to system.
253 * These primitives allow multi process access to different cpu ring buffer
256 * These primitives don't distinguish read-only and read-consume access.
257 * Multi read-only access are also serialized.
261 static DECLARE_RWSEM(all_cpu_access_lock
);
262 static DEFINE_PER_CPU(struct mutex
, cpu_access_lock
);
264 static inline void trace_access_lock(int cpu
)
266 if (cpu
== TRACE_PIPE_ALL_CPU
) {
267 /* gain it for accessing the whole ring buffer. */
268 down_write(&all_cpu_access_lock
);
270 /* gain it for accessing a cpu ring buffer. */
272 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
273 down_read(&all_cpu_access_lock
);
275 /* Secondly block other access to this @cpu ring buffer. */
276 mutex_lock(&per_cpu(cpu_access_lock
, cpu
));
280 static inline void trace_access_unlock(int cpu
)
282 if (cpu
== TRACE_PIPE_ALL_CPU
) {
283 up_write(&all_cpu_access_lock
);
285 mutex_unlock(&per_cpu(cpu_access_lock
, cpu
));
286 up_read(&all_cpu_access_lock
);
290 static inline void trace_access_lock_init(void)
294 for_each_possible_cpu(cpu
)
295 mutex_init(&per_cpu(cpu_access_lock
, cpu
));
300 static DEFINE_MUTEX(access_lock
);
302 static inline void trace_access_lock(int cpu
)
305 mutex_lock(&access_lock
);
308 static inline void trace_access_unlock(int cpu
)
311 mutex_unlock(&access_lock
);
314 static inline void trace_access_lock_init(void)
320 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
321 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
323 /* trace_flags holds trace_options default values */
324 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
| TRACE_ITER_PRINTK
|
325 TRACE_ITER_ANNOTATE
| TRACE_ITER_CONTEXT_INFO
| TRACE_ITER_SLEEP_TIME
|
326 TRACE_ITER_GRAPH_TIME
| TRACE_ITER_RECORD_CMD
| TRACE_ITER_OVERWRITE
|
327 TRACE_ITER_IRQ_INFO
| TRACE_ITER_MARKERS
;
329 static int trace_stop_count
;
330 static DEFINE_RAW_SPINLOCK(tracing_start_lock
);
332 static void wakeup_work_handler(struct work_struct
*work
)
334 wake_up(&trace_wait
);
337 static DECLARE_DELAYED_WORK(wakeup_work
, wakeup_work_handler
);
340 * tracing_on - enable tracing buffers
342 * This function enables tracing buffers that may have been
343 * disabled with tracing_off.
345 void tracing_on(void)
347 if (global_trace
.buffer
)
348 ring_buffer_record_on(global_trace
.buffer
);
350 * This flag is only looked at when buffers haven't been
351 * allocated yet. We don't really care about the race
352 * between setting this flag and actually turning
355 global_trace
.buffer_disabled
= 0;
357 EXPORT_SYMBOL_GPL(tracing_on
);
360 * tracing_off - turn off tracing buffers
362 * This function stops the tracing buffers from recording data.
363 * It does not disable any overhead the tracers themselves may
364 * be causing. This function simply causes all recording to
365 * the ring buffers to fail.
367 void tracing_off(void)
369 if (global_trace
.buffer
)
370 ring_buffer_record_off(global_trace
.buffer
);
372 * This flag is only looked at when buffers haven't been
373 * allocated yet. We don't really care about the race
374 * between setting this flag and actually turning
377 global_trace
.buffer_disabled
= 1;
379 EXPORT_SYMBOL_GPL(tracing_off
);
382 * tracing_is_on - show state of ring buffers enabled
384 int tracing_is_on(void)
386 if (global_trace
.buffer
)
387 return ring_buffer_record_is_on(global_trace
.buffer
);
388 return !global_trace
.buffer_disabled
;
390 EXPORT_SYMBOL_GPL(tracing_is_on
);
393 * trace_wake_up - wake up tasks waiting for trace input
395 * Schedules a delayed work to wake up any task that is blocked on the
396 * trace_wait queue. These is used with trace_poll for tasks polling the
399 void trace_wake_up(void)
401 const unsigned long delay
= msecs_to_jiffies(2);
403 if (trace_flags
& TRACE_ITER_BLOCK
)
405 schedule_delayed_work(&wakeup_work
, delay
);
408 static int __init
set_buf_size(char *str
)
410 unsigned long buf_size
;
414 buf_size
= memparse(str
, &str
);
415 /* nr_entries can not be zero */
418 trace_buf_size
= buf_size
;
421 __setup("trace_buf_size=", set_buf_size
);
423 static int __init
set_tracing_thresh(char *str
)
425 unsigned long threshold
;
430 ret
= kstrtoul(str
, 0, &threshold
);
433 tracing_thresh
= threshold
* 1000;
436 __setup("tracing_thresh=", set_tracing_thresh
);
438 unsigned long nsecs_to_usecs(unsigned long nsecs
)
443 /* These must match the bit postions in trace_iterator_flags */
444 static const char *trace_options
[] = {
477 { trace_clock_local
, "local" },
478 { trace_clock_global
, "global" },
479 { trace_clock_counter
, "counter" },
485 * trace_parser_get_init - gets the buffer for trace parser
487 int trace_parser_get_init(struct trace_parser
*parser
, int size
)
489 memset(parser
, 0, sizeof(*parser
));
491 parser
->buffer
= kmalloc(size
, GFP_KERNEL
);
500 * trace_parser_put - frees the buffer for trace parser
502 void trace_parser_put(struct trace_parser
*parser
)
504 kfree(parser
->buffer
);
508 * trace_get_user - reads the user input string separated by space
509 * (matched by isspace(ch))
511 * For each string found the 'struct trace_parser' is updated,
512 * and the function returns.
514 * Returns number of bytes read.
516 * See kernel/trace/trace.h for 'struct trace_parser' details.
518 int trace_get_user(struct trace_parser
*parser
, const char __user
*ubuf
,
519 size_t cnt
, loff_t
*ppos
)
526 trace_parser_clear(parser
);
528 ret
= get_user(ch
, ubuf
++);
536 * The parser is not finished with the last write,
537 * continue reading the user input without skipping spaces.
540 /* skip white space */
541 while (cnt
&& isspace(ch
)) {
542 ret
= get_user(ch
, ubuf
++);
549 /* only spaces were written */
559 /* read the non-space input */
560 while (cnt
&& !isspace(ch
)) {
561 if (parser
->idx
< parser
->size
- 1)
562 parser
->buffer
[parser
->idx
++] = ch
;
567 ret
= get_user(ch
, ubuf
++);
574 /* We either got finished input or we have to wait for another call. */
576 parser
->buffer
[parser
->idx
] = 0;
577 parser
->cont
= false;
580 parser
->buffer
[parser
->idx
++] = ch
;
590 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
598 if (s
->len
<= s
->readpos
)
601 len
= s
->len
- s
->readpos
;
604 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
614 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
618 if (s
->len
<= s
->readpos
)
621 len
= s
->len
- s
->readpos
;
624 memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
631 * ftrace_max_lock is used to protect the swapping of buffers
632 * when taking a max snapshot. The buffers themselves are
633 * protected by per_cpu spinlocks. But the action of the swap
634 * needs its own lock.
636 * This is defined as a arch_spinlock_t in order to help
637 * with performance when lockdep debugging is enabled.
639 * It is also used in other places outside the update_max_tr
640 * so it needs to be defined outside of the
641 * CONFIG_TRACER_MAX_TRACE.
643 static arch_spinlock_t ftrace_max_lock
=
644 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
646 unsigned long __read_mostly tracing_thresh
;
648 #ifdef CONFIG_TRACER_MAX_TRACE
649 unsigned long __read_mostly tracing_max_latency
;
652 * Copy the new maximum trace into the separate maximum-trace
653 * structure. (this way the maximum trace is permanently saved,
654 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
657 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
659 struct trace_array_cpu
*data
= tr
->data
[cpu
];
660 struct trace_array_cpu
*max_data
;
663 max_tr
.time_start
= data
->preempt_timestamp
;
665 max_data
= max_tr
.data
[cpu
];
666 max_data
->saved_latency
= tracing_max_latency
;
667 max_data
->critical_start
= data
->critical_start
;
668 max_data
->critical_end
= data
->critical_end
;
670 memcpy(max_data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
671 max_data
->pid
= tsk
->pid
;
672 max_data
->uid
= task_uid(tsk
);
673 max_data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
674 max_data
->policy
= tsk
->policy
;
675 max_data
->rt_priority
= tsk
->rt_priority
;
677 /* record this tasks comm */
678 tracing_record_cmdline(tsk
);
682 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
684 * @tsk: the task with the latency
685 * @cpu: The cpu that initiated the trace.
687 * Flip the buffers between the @tr and the max_tr and record information
688 * about which task was the cause of this latency.
691 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
693 struct ring_buffer
*buf
= tr
->buffer
;
695 if (trace_stop_count
)
698 WARN_ON_ONCE(!irqs_disabled());
699 if (!current_trace
->use_max_tr
) {
703 arch_spin_lock(&ftrace_max_lock
);
705 tr
->buffer
= max_tr
.buffer
;
708 __update_max_tr(tr
, tsk
, cpu
);
709 arch_spin_unlock(&ftrace_max_lock
);
713 * update_max_tr_single - only copy one trace over, and reset the rest
715 * @tsk - task with the latency
716 * @cpu - the cpu of the buffer to copy.
718 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
721 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
725 if (trace_stop_count
)
728 WARN_ON_ONCE(!irqs_disabled());
729 if (!current_trace
->use_max_tr
) {
734 arch_spin_lock(&ftrace_max_lock
);
736 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
740 * We failed to swap the buffer due to a commit taking
741 * place on this CPU. We fail to record, but we reset
742 * the max trace buffer (no one writes directly to it)
743 * and flag that it failed.
745 trace_array_printk(&max_tr
, _THIS_IP_
,
746 "Failed to swap buffers due to commit in progress\n");
749 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
&& ret
!= -EBUSY
);
751 __update_max_tr(tr
, tsk
, cpu
);
752 arch_spin_unlock(&ftrace_max_lock
);
754 #endif /* CONFIG_TRACER_MAX_TRACE */
757 * register_tracer - register a tracer with the ftrace system.
758 * @type - the plugin for the tracer
760 * Register a new plugin tracer.
762 int register_tracer(struct tracer
*type
)
768 pr_info("Tracer must have a name\n");
772 if (strlen(type
->name
) >= MAX_TRACER_SIZE
) {
773 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE
);
777 mutex_lock(&trace_types_lock
);
779 tracing_selftest_running
= true;
781 for (t
= trace_types
; t
; t
= t
->next
) {
782 if (strcmp(type
->name
, t
->name
) == 0) {
784 pr_info("Tracer %s already registered\n",
792 type
->set_flag
= &dummy_set_flag
;
794 type
->flags
= &dummy_tracer_flags
;
796 if (!type
->flags
->opts
)
797 type
->flags
->opts
= dummy_tracer_opt
;
798 if (!type
->wait_pipe
)
799 type
->wait_pipe
= default_wait_pipe
;
802 #ifdef CONFIG_FTRACE_STARTUP_TEST
803 if (type
->selftest
&& !tracing_selftest_disabled
) {
804 struct tracer
*saved_tracer
= current_trace
;
805 struct trace_array
*tr
= &global_trace
;
808 * Run a selftest on this tracer.
809 * Here we reset the trace buffer, and set the current
810 * tracer to be this tracer. The tracer can then run some
811 * internal tracing to verify that everything is in order.
812 * If we fail, we do not register this tracer.
814 tracing_reset_online_cpus(tr
);
816 current_trace
= type
;
818 /* If we expanded the buffers, make sure the max is expanded too */
819 if (ring_buffer_expanded
&& type
->use_max_tr
)
820 ring_buffer_resize(max_tr
.buffer
, trace_buf_size
,
821 RING_BUFFER_ALL_CPUS
);
823 /* the test is responsible for initializing and enabling */
824 pr_info("Testing tracer %s: ", type
->name
);
825 ret
= type
->selftest(type
, tr
);
826 /* the test is responsible for resetting too */
827 current_trace
= saved_tracer
;
829 printk(KERN_CONT
"FAILED!\n");
830 /* Add the warning after printing 'FAILED' */
834 /* Only reset on passing, to avoid touching corrupted buffers */
835 tracing_reset_online_cpus(tr
);
837 /* Shrink the max buffer again */
838 if (ring_buffer_expanded
&& type
->use_max_tr
)
839 ring_buffer_resize(max_tr
.buffer
, 1,
840 RING_BUFFER_ALL_CPUS
);
842 printk(KERN_CONT
"PASSED\n");
846 type
->next
= trace_types
;
850 tracing_selftest_running
= false;
851 mutex_unlock(&trace_types_lock
);
853 if (ret
|| !default_bootup_tracer
)
856 if (strncmp(default_bootup_tracer
, type
->name
, MAX_TRACER_SIZE
))
859 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
860 /* Do we want this tracer to start on bootup? */
861 tracing_set_tracer(type
->name
);
862 default_bootup_tracer
= NULL
;
863 /* disable other selftests, since this will break it. */
864 tracing_selftest_disabled
= 1;
865 #ifdef CONFIG_FTRACE_STARTUP_TEST
866 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
874 void tracing_reset(struct trace_array
*tr
, int cpu
)
876 struct ring_buffer
*buffer
= tr
->buffer
;
878 ring_buffer_record_disable(buffer
);
880 /* Make sure all commits have finished */
882 ring_buffer_reset_cpu(buffer
, cpu
);
884 ring_buffer_record_enable(buffer
);
887 void tracing_reset_online_cpus(struct trace_array
*tr
)
889 struct ring_buffer
*buffer
= tr
->buffer
;
892 ring_buffer_record_disable(buffer
);
894 /* Make sure all commits have finished */
897 tr
->time_start
= ftrace_now(tr
->cpu
);
899 for_each_online_cpu(cpu
)
900 ring_buffer_reset_cpu(buffer
, cpu
);
902 ring_buffer_record_enable(buffer
);
905 void tracing_reset_current(int cpu
)
907 tracing_reset(&global_trace
, cpu
);
910 void tracing_reset_current_online_cpus(void)
912 tracing_reset_online_cpus(&global_trace
);
915 #define SAVED_CMDLINES 128
916 #define NO_CMDLINE_MAP UINT_MAX
917 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
918 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
919 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
920 static int cmdline_idx
;
921 static arch_spinlock_t trace_cmdline_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
923 /* temporary disable recording */
924 static atomic_t trace_record_cmdline_disabled __read_mostly
;
926 static void trace_init_cmdlines(void)
928 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
929 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
933 int is_tracing_stopped(void)
935 return trace_stop_count
;
939 * ftrace_off_permanent - disable all ftrace code permanently
941 * This should only be called when a serious anomally has
942 * been detected. This will turn off the function tracing,
943 * ring buffers, and other tracing utilites. It takes no
944 * locks and can be called from any context.
946 void ftrace_off_permanent(void)
948 tracing_disabled
= 1;
950 tracing_off_permanent();
954 * tracing_start - quick start of the tracer
956 * If tracing is enabled but was stopped by tracing_stop,
957 * this will start the tracer back up.
959 void tracing_start(void)
961 struct ring_buffer
*buffer
;
964 if (tracing_disabled
)
967 raw_spin_lock_irqsave(&tracing_start_lock
, flags
);
968 if (--trace_stop_count
) {
969 if (trace_stop_count
< 0) {
970 /* Someone screwed up their debugging */
972 trace_stop_count
= 0;
977 /* Prevent the buffers from switching */
978 arch_spin_lock(&ftrace_max_lock
);
980 buffer
= global_trace
.buffer
;
982 ring_buffer_record_enable(buffer
);
984 buffer
= max_tr
.buffer
;
986 ring_buffer_record_enable(buffer
);
988 arch_spin_unlock(&ftrace_max_lock
);
992 raw_spin_unlock_irqrestore(&tracing_start_lock
, flags
);
996 * tracing_stop - quick stop of the tracer
998 * Light weight way to stop tracing. Use in conjunction with
1001 void tracing_stop(void)
1003 struct ring_buffer
*buffer
;
1004 unsigned long flags
;
1007 raw_spin_lock_irqsave(&tracing_start_lock
, flags
);
1008 if (trace_stop_count
++)
1011 /* Prevent the buffers from switching */
1012 arch_spin_lock(&ftrace_max_lock
);
1014 buffer
= global_trace
.buffer
;
1016 ring_buffer_record_disable(buffer
);
1018 buffer
= max_tr
.buffer
;
1020 ring_buffer_record_disable(buffer
);
1022 arch_spin_unlock(&ftrace_max_lock
);
1025 raw_spin_unlock_irqrestore(&tracing_start_lock
, flags
);
1028 void trace_stop_cmdline_recording(void);
1030 static void trace_save_cmdline(struct task_struct
*tsk
)
1034 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
1038 * It's not the end of the world if we don't get
1039 * the lock, but we also don't want to spin
1040 * nor do we want to disable interrupts,
1041 * so if we miss here, then better luck next time.
1043 if (!arch_spin_trylock(&trace_cmdline_lock
))
1046 idx
= map_pid_to_cmdline
[tsk
->pid
];
1047 if (idx
== NO_CMDLINE_MAP
) {
1048 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
1051 * Check whether the cmdline buffer at idx has a pid
1052 * mapped. We are going to overwrite that entry so we
1053 * need to clear the map_pid_to_cmdline. Otherwise we
1054 * would read the new comm for the old pid.
1056 pid
= map_cmdline_to_pid
[idx
];
1057 if (pid
!= NO_CMDLINE_MAP
)
1058 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
1060 map_cmdline_to_pid
[idx
] = tsk
->pid
;
1061 map_pid_to_cmdline
[tsk
->pid
] = idx
;
1066 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
1068 arch_spin_unlock(&trace_cmdline_lock
);
1071 void trace_find_cmdline(int pid
, char comm
[])
1076 strcpy(comm
, "<idle>");
1080 if (WARN_ON_ONCE(pid
< 0)) {
1081 strcpy(comm
, "<XXX>");
1085 if (pid
> PID_MAX_DEFAULT
) {
1086 strcpy(comm
, "<...>");
1091 arch_spin_lock(&trace_cmdline_lock
);
1092 map
= map_pid_to_cmdline
[pid
];
1093 if (map
!= NO_CMDLINE_MAP
)
1094 strcpy(comm
, saved_cmdlines
[map
]);
1096 strcpy(comm
, "<...>");
1098 arch_spin_unlock(&trace_cmdline_lock
);
1102 void tracing_record_cmdline(struct task_struct
*tsk
)
1104 if (atomic_read(&trace_record_cmdline_disabled
) || !tracing_is_on())
1107 if (!__this_cpu_read(trace_cmdline_save
))
1110 __this_cpu_write(trace_cmdline_save
, false);
1112 trace_save_cmdline(tsk
);
1116 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
1119 struct task_struct
*tsk
= current
;
1121 entry
->preempt_count
= pc
& 0xff;
1122 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
1125 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1126 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
1128 TRACE_FLAG_IRQS_NOSUPPORT
|
1130 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
1131 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
1132 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
1134 EXPORT_SYMBOL_GPL(tracing_generic_entry_update
);
1136 struct ring_buffer_event
*
1137 trace_buffer_lock_reserve(struct ring_buffer
*buffer
,
1140 unsigned long flags
, int pc
)
1142 struct ring_buffer_event
*event
;
1144 event
= ring_buffer_lock_reserve(buffer
, len
);
1145 if (event
!= NULL
) {
1146 struct trace_entry
*ent
= ring_buffer_event_data(event
);
1148 tracing_generic_entry_update(ent
, flags
, pc
);
1156 __buffer_unlock_commit(struct ring_buffer
*buffer
, struct ring_buffer_event
*event
)
1158 __this_cpu_write(trace_cmdline_save
, true);
1159 ring_buffer_unlock_commit(buffer
, event
);
1163 __trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1164 struct ring_buffer_event
*event
,
1165 unsigned long flags
, int pc
,
1168 __buffer_unlock_commit(buffer
, event
);
1170 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1171 ftrace_trace_userstack(buffer
, flags
, pc
);
1177 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1178 struct ring_buffer_event
*event
,
1179 unsigned long flags
, int pc
)
1181 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1184 struct ring_buffer_event
*
1185 trace_current_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1186 int type
, unsigned long len
,
1187 unsigned long flags
, int pc
)
1189 *current_rb
= global_trace
.buffer
;
1190 return trace_buffer_lock_reserve(*current_rb
,
1191 type
, len
, flags
, pc
);
1193 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve
);
1195 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
1196 struct ring_buffer_event
*event
,
1197 unsigned long flags
, int pc
)
1199 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 1);
1201 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit
);
1203 void trace_nowake_buffer_unlock_commit(struct ring_buffer
*buffer
,
1204 struct ring_buffer_event
*event
,
1205 unsigned long flags
, int pc
)
1207 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
, 0);
1209 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit
);
1211 void trace_nowake_buffer_unlock_commit_regs(struct ring_buffer
*buffer
,
1212 struct ring_buffer_event
*event
,
1213 unsigned long flags
, int pc
,
1214 struct pt_regs
*regs
)
1216 __buffer_unlock_commit(buffer
, event
);
1218 ftrace_trace_stack_regs(buffer
, flags
, 0, pc
, regs
);
1219 ftrace_trace_userstack(buffer
, flags
, pc
);
1221 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit_regs
);
1223 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
1224 struct ring_buffer_event
*event
)
1226 ring_buffer_discard_commit(buffer
, event
);
1228 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit
);
1231 trace_function(struct trace_array
*tr
,
1232 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1235 struct ftrace_event_call
*call
= &event_function
;
1236 struct ring_buffer
*buffer
= tr
->buffer
;
1237 struct ring_buffer_event
*event
;
1238 struct ftrace_entry
*entry
;
1240 /* If we are reading the ring buffer, don't trace */
1241 if (unlikely(__this_cpu_read(ftrace_cpu_disabled
)))
1244 event
= trace_buffer_lock_reserve(buffer
, TRACE_FN
, sizeof(*entry
),
1248 entry
= ring_buffer_event_data(event
);
1250 entry
->parent_ip
= parent_ip
;
1252 if (!filter_check_discard(call
, entry
, buffer
, event
))
1253 __buffer_unlock_commit(buffer
, event
);
1257 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1258 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1261 if (likely(!atomic_read(&data
->disabled
)))
1262 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
1265 #ifdef CONFIG_STACKTRACE
1267 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1268 struct ftrace_stack
{
1269 unsigned long calls
[FTRACE_STACK_MAX_ENTRIES
];
1272 static DEFINE_PER_CPU(struct ftrace_stack
, ftrace_stack
);
1273 static DEFINE_PER_CPU(int, ftrace_stack_reserve
);
1275 static void __ftrace_trace_stack(struct ring_buffer
*buffer
,
1276 unsigned long flags
,
1277 int skip
, int pc
, struct pt_regs
*regs
)
1279 struct ftrace_event_call
*call
= &event_kernel_stack
;
1280 struct ring_buffer_event
*event
;
1281 struct stack_entry
*entry
;
1282 struct stack_trace trace
;
1284 int size
= FTRACE_STACK_ENTRIES
;
1286 trace
.nr_entries
= 0;
1290 * Since events can happen in NMIs there's no safe way to
1291 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1292 * or NMI comes in, it will just have to use the default
1293 * FTRACE_STACK_SIZE.
1295 preempt_disable_notrace();
1297 use_stack
= ++__get_cpu_var(ftrace_stack_reserve
);
1299 * We don't need any atomic variables, just a barrier.
1300 * If an interrupt comes in, we don't care, because it would
1301 * have exited and put the counter back to what we want.
1302 * We just need a barrier to keep gcc from moving things
1306 if (use_stack
== 1) {
1307 trace
.entries
= &__get_cpu_var(ftrace_stack
).calls
[0];
1308 trace
.max_entries
= FTRACE_STACK_MAX_ENTRIES
;
1311 save_stack_trace_regs(regs
, &trace
);
1313 save_stack_trace(&trace
);
1315 if (trace
.nr_entries
> size
)
1316 size
= trace
.nr_entries
;
1318 /* From now on, use_stack is a boolean */
1321 size
*= sizeof(unsigned long);
1323 event
= trace_buffer_lock_reserve(buffer
, TRACE_STACK
,
1324 sizeof(*entry
) + size
, flags
, pc
);
1327 entry
= ring_buffer_event_data(event
);
1329 memset(&entry
->caller
, 0, size
);
1332 memcpy(&entry
->caller
, trace
.entries
,
1333 trace
.nr_entries
* sizeof(unsigned long));
1335 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1336 trace
.entries
= entry
->caller
;
1338 save_stack_trace_regs(regs
, &trace
);
1340 save_stack_trace(&trace
);
1343 entry
->size
= trace
.nr_entries
;
1345 if (!filter_check_discard(call
, entry
, buffer
, event
))
1346 __buffer_unlock_commit(buffer
, event
);
1349 /* Again, don't let gcc optimize things here */
1351 __get_cpu_var(ftrace_stack_reserve
)--;
1352 preempt_enable_notrace();
1356 void ftrace_trace_stack_regs(struct ring_buffer
*buffer
, unsigned long flags
,
1357 int skip
, int pc
, struct pt_regs
*regs
)
1359 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1362 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, regs
);
1365 void ftrace_trace_stack(struct ring_buffer
*buffer
, unsigned long flags
,
1368 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1371 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, NULL
);
1374 void __trace_stack(struct trace_array
*tr
, unsigned long flags
, int skip
,
1377 __ftrace_trace_stack(tr
->buffer
, flags
, skip
, pc
, NULL
);
1381 * trace_dump_stack - record a stack back trace in the trace buffer
1383 void trace_dump_stack(void)
1385 unsigned long flags
;
1387 if (tracing_disabled
|| tracing_selftest_running
)
1390 local_save_flags(flags
);
1392 /* skipping 3 traces, seems to get us at the caller of this function */
1393 __ftrace_trace_stack(global_trace
.buffer
, flags
, 3, preempt_count(), NULL
);
1396 static DEFINE_PER_CPU(int, user_stack_count
);
1399 ftrace_trace_userstack(struct ring_buffer
*buffer
, unsigned long flags
, int pc
)
1401 struct ftrace_event_call
*call
= &event_user_stack
;
1402 struct ring_buffer_event
*event
;
1403 struct userstack_entry
*entry
;
1404 struct stack_trace trace
;
1406 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1410 * NMIs can not handle page faults, even with fix ups.
1411 * The save user stack can (and often does) fault.
1413 if (unlikely(in_nmi()))
1417 * prevent recursion, since the user stack tracing may
1418 * trigger other kernel events.
1421 if (__this_cpu_read(user_stack_count
))
1424 __this_cpu_inc(user_stack_count
);
1426 event
= trace_buffer_lock_reserve(buffer
, TRACE_USER_STACK
,
1427 sizeof(*entry
), flags
, pc
);
1429 goto out_drop_count
;
1430 entry
= ring_buffer_event_data(event
);
1432 entry
->tgid
= current
->tgid
;
1433 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1435 trace
.nr_entries
= 0;
1436 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1438 trace
.entries
= entry
->caller
;
1440 save_stack_trace_user(&trace
);
1441 if (!filter_check_discard(call
, entry
, buffer
, event
))
1442 __buffer_unlock_commit(buffer
, event
);
1445 __this_cpu_dec(user_stack_count
);
1451 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1453 ftrace_trace_userstack(tr
, flags
, preempt_count());
1457 #endif /* CONFIG_STACKTRACE */
1459 /* created for use with alloc_percpu */
1460 struct trace_buffer_struct
{
1461 char buffer
[TRACE_BUF_SIZE
];
1464 static struct trace_buffer_struct
*trace_percpu_buffer
;
1465 static struct trace_buffer_struct
*trace_percpu_sirq_buffer
;
1466 static struct trace_buffer_struct
*trace_percpu_irq_buffer
;
1467 static struct trace_buffer_struct
*trace_percpu_nmi_buffer
;
1470 * The buffer used is dependent on the context. There is a per cpu
1471 * buffer for normal context, softirq contex, hard irq context and
1472 * for NMI context. Thise allows for lockless recording.
1474 * Note, if the buffers failed to be allocated, then this returns NULL
1476 static char *get_trace_buf(void)
1478 struct trace_buffer_struct
*percpu_buffer
;
1479 struct trace_buffer_struct
*buffer
;
1482 * If we have allocated per cpu buffers, then we do not
1483 * need to do any locking.
1486 percpu_buffer
= trace_percpu_nmi_buffer
;
1488 percpu_buffer
= trace_percpu_irq_buffer
;
1489 else if (in_softirq())
1490 percpu_buffer
= trace_percpu_sirq_buffer
;
1492 percpu_buffer
= trace_percpu_buffer
;
1497 buffer
= per_cpu_ptr(percpu_buffer
, smp_processor_id());
1499 return buffer
->buffer
;
1502 static int alloc_percpu_trace_buffer(void)
1504 struct trace_buffer_struct
*buffers
;
1505 struct trace_buffer_struct
*sirq_buffers
;
1506 struct trace_buffer_struct
*irq_buffers
;
1507 struct trace_buffer_struct
*nmi_buffers
;
1509 buffers
= alloc_percpu(struct trace_buffer_struct
);
1513 sirq_buffers
= alloc_percpu(struct trace_buffer_struct
);
1517 irq_buffers
= alloc_percpu(struct trace_buffer_struct
);
1521 nmi_buffers
= alloc_percpu(struct trace_buffer_struct
);
1525 trace_percpu_buffer
= buffers
;
1526 trace_percpu_sirq_buffer
= sirq_buffers
;
1527 trace_percpu_irq_buffer
= irq_buffers
;
1528 trace_percpu_nmi_buffer
= nmi_buffers
;
1533 free_percpu(irq_buffers
);
1535 free_percpu(sirq_buffers
);
1537 free_percpu(buffers
);
1539 WARN(1, "Could not allocate percpu trace_printk buffer");
1543 static int buffers_allocated
;
1545 void trace_printk_init_buffers(void)
1547 if (buffers_allocated
)
1550 if (alloc_percpu_trace_buffer())
1553 pr_info("ftrace: Allocated trace_printk buffers\n");
1555 /* Expand the buffers to set size */
1556 tracing_update_buffers();
1558 buffers_allocated
= 1;
1561 * trace_printk_init_buffers() can be called by modules.
1562 * If that happens, then we need to start cmdline recording
1563 * directly here. If the global_trace.buffer is already
1564 * allocated here, then this was called by module code.
1566 if (global_trace
.buffer
)
1567 tracing_start_cmdline_record();
1570 void trace_printk_start_comm(void)
1572 /* Start tracing comms if trace printk is set */
1573 if (!buffers_allocated
)
1575 tracing_start_cmdline_record();
1578 static void trace_printk_start_stop_comm(int enabled
)
1580 if (!buffers_allocated
)
1584 tracing_start_cmdline_record();
1586 tracing_stop_cmdline_record();
1590 * trace_vbprintk - write binary msg to tracing buffer
1593 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1595 struct ftrace_event_call
*call
= &event_bprint
;
1596 struct ring_buffer_event
*event
;
1597 struct ring_buffer
*buffer
;
1598 struct trace_array
*tr
= &global_trace
;
1599 struct bprint_entry
*entry
;
1600 unsigned long flags
;
1602 int len
= 0, size
, pc
;
1604 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1607 /* Don't pollute graph traces with trace_vprintk internals */
1608 pause_graph_tracing();
1610 pc
= preempt_count();
1611 preempt_disable_notrace();
1613 tbuffer
= get_trace_buf();
1619 len
= vbin_printf((u32
*)tbuffer
, TRACE_BUF_SIZE
/sizeof(int), fmt
, args
);
1621 if (len
> TRACE_BUF_SIZE
/sizeof(int) || len
< 0)
1624 local_save_flags(flags
);
1625 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1626 buffer
= tr
->buffer
;
1627 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
1631 entry
= ring_buffer_event_data(event
);
1635 memcpy(entry
->buf
, tbuffer
, sizeof(u32
) * len
);
1636 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1637 __buffer_unlock_commit(buffer
, event
);
1638 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1642 preempt_enable_notrace();
1643 unpause_graph_tracing();
1647 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1649 int trace_array_printk(struct trace_array
*tr
,
1650 unsigned long ip
, const char *fmt
, ...)
1655 if (!(trace_flags
& TRACE_ITER_PRINTK
))
1659 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
1664 int trace_array_vprintk(struct trace_array
*tr
,
1665 unsigned long ip
, const char *fmt
, va_list args
)
1667 struct ftrace_event_call
*call
= &event_print
;
1668 struct ring_buffer_event
*event
;
1669 struct ring_buffer
*buffer
;
1670 int len
= 0, size
, pc
;
1671 struct print_entry
*entry
;
1672 unsigned long flags
;
1675 if (tracing_disabled
|| tracing_selftest_running
)
1678 /* Don't pollute graph traces with trace_vprintk internals */
1679 pause_graph_tracing();
1681 pc
= preempt_count();
1682 preempt_disable_notrace();
1685 tbuffer
= get_trace_buf();
1691 len
= vsnprintf(tbuffer
, TRACE_BUF_SIZE
, fmt
, args
);
1692 if (len
> TRACE_BUF_SIZE
)
1695 local_save_flags(flags
);
1696 size
= sizeof(*entry
) + len
+ 1;
1697 buffer
= tr
->buffer
;
1698 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
1702 entry
= ring_buffer_event_data(event
);
1705 memcpy(&entry
->buf
, tbuffer
, len
);
1706 entry
->buf
[len
] = '\0';
1707 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1708 __buffer_unlock_commit(buffer
, event
);
1709 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1712 preempt_enable_notrace();
1713 unpause_graph_tracing();
1718 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1720 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
1722 EXPORT_SYMBOL_GPL(trace_vprintk
);
1724 static void trace_iterator_increment(struct trace_iterator
*iter
)
1726 struct ring_buffer_iter
*buf_iter
= trace_buffer_iter(iter
, iter
->cpu
);
1730 ring_buffer_read(buf_iter
, NULL
);
1733 static struct trace_entry
*
1734 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
,
1735 unsigned long *lost_events
)
1737 struct ring_buffer_event
*event
;
1738 struct ring_buffer_iter
*buf_iter
= trace_buffer_iter(iter
, cpu
);
1741 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1743 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
,
1747 iter
->ent_size
= ring_buffer_event_length(event
);
1748 return ring_buffer_event_data(event
);
1754 static struct trace_entry
*
1755 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
,
1756 unsigned long *missing_events
, u64
*ent_ts
)
1758 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1759 struct trace_entry
*ent
, *next
= NULL
;
1760 unsigned long lost_events
= 0, next_lost
= 0;
1761 int cpu_file
= iter
->cpu_file
;
1762 u64 next_ts
= 0, ts
;
1768 * If we are in a per_cpu trace file, don't bother by iterating over
1769 * all cpu and peek directly.
1771 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1772 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1774 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
, missing_events
);
1776 *ent_cpu
= cpu_file
;
1781 for_each_tracing_cpu(cpu
) {
1783 if (ring_buffer_empty_cpu(buffer
, cpu
))
1786 ent
= peek_next_entry(iter
, cpu
, &ts
, &lost_events
);
1789 * Pick the entry with the smallest timestamp:
1791 if (ent
&& (!next
|| ts
< next_ts
)) {
1795 next_lost
= lost_events
;
1796 next_size
= iter
->ent_size
;
1800 iter
->ent_size
= next_size
;
1803 *ent_cpu
= next_cpu
;
1809 *missing_events
= next_lost
;
1814 /* Find the next real entry, without updating the iterator itself */
1815 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1816 int *ent_cpu
, u64
*ent_ts
)
1818 return __find_next_entry(iter
, ent_cpu
, NULL
, ent_ts
);
1821 /* Find the next real entry, and increment the iterator to the next entry */
1822 void *trace_find_next_entry_inc(struct trace_iterator
*iter
)
1824 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
,
1825 &iter
->lost_events
, &iter
->ts
);
1828 trace_iterator_increment(iter
);
1830 return iter
->ent
? iter
: NULL
;
1833 static void trace_consume(struct trace_iterator
*iter
)
1835 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
,
1836 &iter
->lost_events
);
1839 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1841 struct trace_iterator
*iter
= m
->private;
1845 WARN_ON_ONCE(iter
->leftover
);
1849 /* can't go backwards */
1854 ent
= trace_find_next_entry_inc(iter
);
1858 while (ent
&& iter
->idx
< i
)
1859 ent
= trace_find_next_entry_inc(iter
);
1866 void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
1868 struct trace_array
*tr
= iter
->tr
;
1869 struct ring_buffer_event
*event
;
1870 struct ring_buffer_iter
*buf_iter
;
1871 unsigned long entries
= 0;
1874 tr
->data
[cpu
]->skipped_entries
= 0;
1876 buf_iter
= trace_buffer_iter(iter
, cpu
);
1880 ring_buffer_iter_reset(buf_iter
);
1883 * We could have the case with the max latency tracers
1884 * that a reset never took place on a cpu. This is evident
1885 * by the timestamp being before the start of the buffer.
1887 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
1888 if (ts
>= iter
->tr
->time_start
)
1891 ring_buffer_read(buf_iter
, NULL
);
1894 tr
->data
[cpu
]->skipped_entries
= entries
;
1898 * The current tracer is copied to avoid a global locking
1901 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1903 struct trace_iterator
*iter
= m
->private;
1904 static struct tracer
*old_tracer
;
1905 int cpu_file
= iter
->cpu_file
;
1910 /* copy the tracer to avoid using a global lock all around */
1911 mutex_lock(&trace_types_lock
);
1912 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1913 old_tracer
= current_trace
;
1914 *iter
->trace
= *current_trace
;
1916 mutex_unlock(&trace_types_lock
);
1918 atomic_inc(&trace_record_cmdline_disabled
);
1920 if (*pos
!= iter
->pos
) {
1925 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1926 for_each_tracing_cpu(cpu
)
1927 tracing_iter_reset(iter
, cpu
);
1929 tracing_iter_reset(iter
, cpu_file
);
1932 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1937 * If we overflowed the seq_file before, then we want
1938 * to just reuse the trace_seq buffer again.
1944 p
= s_next(m
, p
, &l
);
1948 trace_event_read_lock();
1949 trace_access_lock(cpu_file
);
1953 static void s_stop(struct seq_file
*m
, void *p
)
1955 struct trace_iterator
*iter
= m
->private;
1957 atomic_dec(&trace_record_cmdline_disabled
);
1958 trace_access_unlock(iter
->cpu_file
);
1959 trace_event_read_unlock();
1963 get_total_entries(struct trace_array
*tr
, unsigned long *total
, unsigned long *entries
)
1965 unsigned long count
;
1971 for_each_tracing_cpu(cpu
) {
1972 count
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
1974 * If this buffer has skipped entries, then we hold all
1975 * entries for the trace and we need to ignore the
1976 * ones before the time stamp.
1978 if (tr
->data
[cpu
]->skipped_entries
) {
1979 count
-= tr
->data
[cpu
]->skipped_entries
;
1980 /* total is the same as the entries */
1984 ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
1989 static void print_lat_help_header(struct seq_file
*m
)
1991 seq_puts(m
, "# _------=> CPU# \n");
1992 seq_puts(m
, "# / _-----=> irqs-off \n");
1993 seq_puts(m
, "# | / _----=> need-resched \n");
1994 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1995 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1996 seq_puts(m
, "# |||| / delay \n");
1997 seq_puts(m
, "# cmd pid ||||| time | caller \n");
1998 seq_puts(m
, "# \\ / ||||| \\ | / \n");
2001 static void print_event_info(struct trace_array
*tr
, struct seq_file
*m
)
2003 unsigned long total
;
2004 unsigned long entries
;
2006 get_total_entries(tr
, &total
, &entries
);
2007 seq_printf(m
, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2008 entries
, total
, num_online_cpus());
2012 static void print_func_help_header(struct trace_array
*tr
, struct seq_file
*m
)
2014 print_event_info(tr
, m
);
2015 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2016 seq_puts(m
, "# | | | | |\n");
2019 static void print_func_help_header_irq(struct trace_array
*tr
, struct seq_file
*m
)
2021 print_event_info(tr
, m
);
2022 seq_puts(m
, "# _-----=> irqs-off\n");
2023 seq_puts(m
, "# / _----=> need-resched\n");
2024 seq_puts(m
, "# | / _---=> hardirq/softirq\n");
2025 seq_puts(m
, "# || / _--=> preempt-depth\n");
2026 seq_puts(m
, "# ||| / delay\n");
2027 seq_puts(m
, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2028 seq_puts(m
, "# | | | |||| | |\n");
2032 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
2034 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2035 struct trace_array
*tr
= iter
->tr
;
2036 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
2037 struct tracer
*type
= current_trace
;
2038 unsigned long entries
;
2039 unsigned long total
;
2040 const char *name
= "preemption";
2045 get_total_entries(tr
, &total
, &entries
);
2047 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
2049 seq_puts(m
, "# -----------------------------------"
2050 "---------------------------------\n");
2051 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2052 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2053 nsecs_to_usecs(data
->saved_latency
),
2057 #if defined(CONFIG_PREEMPT_NONE)
2059 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2061 #elif defined(CONFIG_PREEMPT)
2066 /* These are reserved for later use */
2069 seq_printf(m
, " #P:%d)\n", num_online_cpus());
2073 seq_puts(m
, "# -----------------\n");
2074 seq_printf(m
, "# | task: %.16s-%d "
2075 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2076 data
->comm
, data
->pid
,
2077 from_kuid_munged(seq_user_ns(m
), data
->uid
), data
->nice
,
2078 data
->policy
, data
->rt_priority
);
2079 seq_puts(m
, "# -----------------\n");
2081 if (data
->critical_start
) {
2082 seq_puts(m
, "# => started at: ");
2083 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
2084 trace_print_seq(m
, &iter
->seq
);
2085 seq_puts(m
, "\n# => ended at: ");
2086 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
2087 trace_print_seq(m
, &iter
->seq
);
2088 seq_puts(m
, "\n#\n");
2094 static void test_cpu_buff_start(struct trace_iterator
*iter
)
2096 struct trace_seq
*s
= &iter
->seq
;
2098 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
2101 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
2104 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
2107 if (iter
->tr
->data
[iter
->cpu
]->skipped_entries
)
2110 cpumask_set_cpu(iter
->cpu
, iter
->started
);
2112 /* Don't print started cpu buffer for the first entry of the trace */
2114 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
2118 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
2120 struct trace_seq
*s
= &iter
->seq
;
2121 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2122 struct trace_entry
*entry
;
2123 struct trace_event
*event
;
2127 test_cpu_buff_start(iter
);
2129 event
= ftrace_find_event(entry
->type
);
2131 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2132 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2133 if (!trace_print_lat_context(iter
))
2136 if (!trace_print_context(iter
))
2142 return event
->funcs
->trace(iter
, sym_flags
, event
);
2144 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
2147 return TRACE_TYPE_HANDLED
;
2149 return TRACE_TYPE_PARTIAL_LINE
;
2152 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
2154 struct trace_seq
*s
= &iter
->seq
;
2155 struct trace_entry
*entry
;
2156 struct trace_event
*event
;
2160 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2161 if (!trace_seq_printf(s
, "%d %d %llu ",
2162 entry
->pid
, iter
->cpu
, iter
->ts
))
2166 event
= ftrace_find_event(entry
->type
);
2168 return event
->funcs
->raw(iter
, 0, event
);
2170 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
2173 return TRACE_TYPE_HANDLED
;
2175 return TRACE_TYPE_PARTIAL_LINE
;
2178 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
2180 struct trace_seq
*s
= &iter
->seq
;
2181 unsigned char newline
= '\n';
2182 struct trace_entry
*entry
;
2183 struct trace_event
*event
;
2187 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2188 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
2189 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
2190 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
2193 event
= ftrace_find_event(entry
->type
);
2195 enum print_line_t ret
= event
->funcs
->hex(iter
, 0, event
);
2196 if (ret
!= TRACE_TYPE_HANDLED
)
2200 SEQ_PUT_FIELD_RET(s
, newline
);
2202 return TRACE_TYPE_HANDLED
;
2205 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
2207 struct trace_seq
*s
= &iter
->seq
;
2208 struct trace_entry
*entry
;
2209 struct trace_event
*event
;
2213 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2214 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
2215 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
2216 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
2219 event
= ftrace_find_event(entry
->type
);
2220 return event
? event
->funcs
->binary(iter
, 0, event
) :
2224 int trace_empty(struct trace_iterator
*iter
)
2226 struct ring_buffer_iter
*buf_iter
;
2229 /* If we are looking at one CPU buffer, only check that one */
2230 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
2231 cpu
= iter
->cpu_file
;
2232 buf_iter
= trace_buffer_iter(iter
, cpu
);
2234 if (!ring_buffer_iter_empty(buf_iter
))
2237 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2243 for_each_tracing_cpu(cpu
) {
2244 buf_iter
= trace_buffer_iter(iter
, cpu
);
2246 if (!ring_buffer_iter_empty(buf_iter
))
2249 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2257 /* Called with trace_event_read_lock() held. */
2258 enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
2260 enum print_line_t ret
;
2262 if (iter
->lost_events
&&
2263 !trace_seq_printf(&iter
->seq
, "CPU:%d [LOST %lu EVENTS]\n",
2264 iter
->cpu
, iter
->lost_events
))
2265 return TRACE_TYPE_PARTIAL_LINE
;
2267 if (iter
->trace
&& iter
->trace
->print_line
) {
2268 ret
= iter
->trace
->print_line(iter
);
2269 if (ret
!= TRACE_TYPE_UNHANDLED
)
2273 if (iter
->ent
->type
== TRACE_BPRINT
&&
2274 trace_flags
& TRACE_ITER_PRINTK
&&
2275 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2276 return trace_print_bprintk_msg_only(iter
);
2278 if (iter
->ent
->type
== TRACE_PRINT
&&
2279 trace_flags
& TRACE_ITER_PRINTK
&&
2280 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2281 return trace_print_printk_msg_only(iter
);
2283 if (trace_flags
& TRACE_ITER_BIN
)
2284 return print_bin_fmt(iter
);
2286 if (trace_flags
& TRACE_ITER_HEX
)
2287 return print_hex_fmt(iter
);
2289 if (trace_flags
& TRACE_ITER_RAW
)
2290 return print_raw_fmt(iter
);
2292 return print_trace_fmt(iter
);
2295 void trace_latency_header(struct seq_file
*m
)
2297 struct trace_iterator
*iter
= m
->private;
2299 /* print nothing if the buffers are empty */
2300 if (trace_empty(iter
))
2303 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
2304 print_trace_header(m
, iter
);
2306 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2307 print_lat_help_header(m
);
2310 void trace_default_header(struct seq_file
*m
)
2312 struct trace_iterator
*iter
= m
->private;
2314 if (!(trace_flags
& TRACE_ITER_CONTEXT_INFO
))
2317 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2318 /* print nothing if the buffers are empty */
2319 if (trace_empty(iter
))
2321 print_trace_header(m
, iter
);
2322 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2323 print_lat_help_header(m
);
2325 if (!(trace_flags
& TRACE_ITER_VERBOSE
)) {
2326 if (trace_flags
& TRACE_ITER_IRQ_INFO
)
2327 print_func_help_header_irq(iter
->tr
, m
);
2329 print_func_help_header(iter
->tr
, m
);
2334 static void test_ftrace_alive(struct seq_file
*m
)
2336 if (!ftrace_is_dead())
2338 seq_printf(m
, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2339 seq_printf(m
, "# MAY BE MISSING FUNCTION EVENTS\n");
2342 static int s_show(struct seq_file
*m
, void *v
)
2344 struct trace_iterator
*iter
= v
;
2347 if (iter
->ent
== NULL
) {
2349 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
2351 test_ftrace_alive(m
);
2353 if (iter
->trace
&& iter
->trace
->print_header
)
2354 iter
->trace
->print_header(m
);
2356 trace_default_header(m
);
2358 } else if (iter
->leftover
) {
2360 * If we filled the seq_file buffer earlier, we
2361 * want to just show it now.
2363 ret
= trace_print_seq(m
, &iter
->seq
);
2365 /* ret should this time be zero, but you never know */
2366 iter
->leftover
= ret
;
2369 print_trace_line(iter
);
2370 ret
= trace_print_seq(m
, &iter
->seq
);
2372 * If we overflow the seq_file buffer, then it will
2373 * ask us for this data again at start up.
2375 * ret is 0 if seq_file write succeeded.
2378 iter
->leftover
= ret
;
2384 static const struct seq_operations tracer_seq_ops
= {
2391 static struct trace_iterator
*
2392 __tracing_open(struct inode
*inode
, struct file
*file
)
2394 long cpu_file
= (long) inode
->i_private
;
2395 struct trace_iterator
*iter
;
2398 if (tracing_disabled
)
2399 return ERR_PTR(-ENODEV
);
2401 iter
= __seq_open_private(file
, &tracer_seq_ops
, sizeof(*iter
));
2403 return ERR_PTR(-ENOMEM
);
2405 iter
->buffer_iter
= kzalloc(sizeof(*iter
->buffer_iter
) * num_possible_cpus(),
2407 if (!iter
->buffer_iter
)
2411 * We make a copy of the current tracer to avoid concurrent
2412 * changes on it while we are reading.
2414 mutex_lock(&trace_types_lock
);
2415 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2420 *iter
->trace
= *current_trace
;
2422 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2425 if (current_trace
&& current_trace
->print_max
)
2428 iter
->tr
= &global_trace
;
2430 mutex_init(&iter
->mutex
);
2431 iter
->cpu_file
= cpu_file
;
2433 /* Notify the tracer early; before we stop tracing. */
2434 if (iter
->trace
&& iter
->trace
->open
)
2435 iter
->trace
->open(iter
);
2437 /* Annotate start of buffers if we had overruns */
2438 if (ring_buffer_overruns(iter
->tr
->buffer
))
2439 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2441 /* stop the trace while dumping */
2444 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
2445 for_each_tracing_cpu(cpu
) {
2446 iter
->buffer_iter
[cpu
] =
2447 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2449 ring_buffer_read_prepare_sync();
2450 for_each_tracing_cpu(cpu
) {
2451 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2452 tracing_iter_reset(iter
, cpu
);
2455 cpu
= iter
->cpu_file
;
2456 iter
->buffer_iter
[cpu
] =
2457 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2458 ring_buffer_read_prepare_sync();
2459 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2460 tracing_iter_reset(iter
, cpu
);
2463 mutex_unlock(&trace_types_lock
);
2468 mutex_unlock(&trace_types_lock
);
2470 kfree(iter
->buffer_iter
);
2472 seq_release_private(inode
, file
);
2473 return ERR_PTR(-ENOMEM
);
2476 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2478 if (tracing_disabled
)
2481 filp
->private_data
= inode
->i_private
;
2485 static int tracing_release(struct inode
*inode
, struct file
*file
)
2487 struct seq_file
*m
= file
->private_data
;
2488 struct trace_iterator
*iter
;
2491 if (!(file
->f_mode
& FMODE_READ
))
2496 mutex_lock(&trace_types_lock
);
2497 for_each_tracing_cpu(cpu
) {
2498 if (iter
->buffer_iter
[cpu
])
2499 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2502 if (iter
->trace
&& iter
->trace
->close
)
2503 iter
->trace
->close(iter
);
2505 /* reenable tracing if it was previously enabled */
2507 mutex_unlock(&trace_types_lock
);
2509 mutex_destroy(&iter
->mutex
);
2510 free_cpumask_var(iter
->started
);
2512 kfree(iter
->buffer_iter
);
2513 seq_release_private(inode
, file
);
2517 static int tracing_open(struct inode
*inode
, struct file
*file
)
2519 struct trace_iterator
*iter
;
2522 /* If this file was open for write, then erase contents */
2523 if ((file
->f_mode
& FMODE_WRITE
) &&
2524 (file
->f_flags
& O_TRUNC
)) {
2525 long cpu
= (long) inode
->i_private
;
2527 if (cpu
== TRACE_PIPE_ALL_CPU
)
2528 tracing_reset_online_cpus(&global_trace
);
2530 tracing_reset(&global_trace
, cpu
);
2533 if (file
->f_mode
& FMODE_READ
) {
2534 iter
= __tracing_open(inode
, file
);
2536 ret
= PTR_ERR(iter
);
2537 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2538 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2544 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2546 struct tracer
*t
= v
;
2556 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2561 mutex_lock(&trace_types_lock
);
2562 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2568 static void t_stop(struct seq_file
*m
, void *p
)
2570 mutex_unlock(&trace_types_lock
);
2573 static int t_show(struct seq_file
*m
, void *v
)
2575 struct tracer
*t
= v
;
2580 seq_printf(m
, "%s", t
->name
);
2589 static const struct seq_operations show_traces_seq_ops
= {
2596 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2598 if (tracing_disabled
)
2601 return seq_open(file
, &show_traces_seq_ops
);
2605 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2606 size_t count
, loff_t
*ppos
)
2611 static loff_t
tracing_seek(struct file
*file
, loff_t offset
, int origin
)
2613 if (file
->f_mode
& FMODE_READ
)
2614 return seq_lseek(file
, offset
, origin
);
2619 static const struct file_operations tracing_fops
= {
2620 .open
= tracing_open
,
2622 .write
= tracing_write_stub
,
2623 .llseek
= tracing_seek
,
2624 .release
= tracing_release
,
2627 static const struct file_operations show_traces_fops
= {
2628 .open
= show_traces_open
,
2630 .release
= seq_release
,
2631 .llseek
= seq_lseek
,
2635 * Only trace on a CPU if the bitmask is set:
2637 static cpumask_var_t tracing_cpumask
;
2640 * The tracer itself will not take this lock, but still we want
2641 * to provide a consistent cpumask to user-space:
2643 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2646 * Temporary storage for the character representation of the
2647 * CPU bitmask (and one more byte for the newline):
2649 static char mask_str
[NR_CPUS
+ 1];
2652 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2653 size_t count
, loff_t
*ppos
)
2657 mutex_lock(&tracing_cpumask_update_lock
);
2659 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2660 if (count
- len
< 2) {
2664 len
+= sprintf(mask_str
+ len
, "\n");
2665 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2668 mutex_unlock(&tracing_cpumask_update_lock
);
2674 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2675 size_t count
, loff_t
*ppos
)
2678 cpumask_var_t tracing_cpumask_new
;
2680 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2683 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2687 mutex_lock(&tracing_cpumask_update_lock
);
2689 local_irq_disable();
2690 arch_spin_lock(&ftrace_max_lock
);
2691 for_each_tracing_cpu(cpu
) {
2693 * Increase/decrease the disabled counter if we are
2694 * about to flip a bit in the cpumask:
2696 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2697 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2698 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2699 ring_buffer_record_disable_cpu(global_trace
.buffer
, cpu
);
2701 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2702 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2703 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2704 ring_buffer_record_enable_cpu(global_trace
.buffer
, cpu
);
2707 arch_spin_unlock(&ftrace_max_lock
);
2710 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2712 mutex_unlock(&tracing_cpumask_update_lock
);
2713 free_cpumask_var(tracing_cpumask_new
);
2718 free_cpumask_var(tracing_cpumask_new
);
2723 static const struct file_operations tracing_cpumask_fops
= {
2724 .open
= tracing_open_generic
,
2725 .read
= tracing_cpumask_read
,
2726 .write
= tracing_cpumask_write
,
2727 .llseek
= generic_file_llseek
,
2730 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
2732 struct tracer_opt
*trace_opts
;
2736 mutex_lock(&trace_types_lock
);
2737 tracer_flags
= current_trace
->flags
->val
;
2738 trace_opts
= current_trace
->flags
->opts
;
2740 for (i
= 0; trace_options
[i
]; i
++) {
2741 if (trace_flags
& (1 << i
))
2742 seq_printf(m
, "%s\n", trace_options
[i
]);
2744 seq_printf(m
, "no%s\n", trace_options
[i
]);
2747 for (i
= 0; trace_opts
[i
].name
; i
++) {
2748 if (tracer_flags
& trace_opts
[i
].bit
)
2749 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
2751 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
2753 mutex_unlock(&trace_types_lock
);
2758 static int __set_tracer_option(struct tracer
*trace
,
2759 struct tracer_flags
*tracer_flags
,
2760 struct tracer_opt
*opts
, int neg
)
2764 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
2769 tracer_flags
->val
&= ~opts
->bit
;
2771 tracer_flags
->val
|= opts
->bit
;
2775 /* Try to assign a tracer specific option */
2776 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2778 struct tracer_flags
*tracer_flags
= trace
->flags
;
2779 struct tracer_opt
*opts
= NULL
;
2782 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
2783 opts
= &tracer_flags
->opts
[i
];
2785 if (strcmp(cmp
, opts
->name
) == 0)
2786 return __set_tracer_option(trace
, trace
->flags
,
2793 static void set_tracer_flags(unsigned int mask
, int enabled
)
2795 /* do nothing if flag is already set */
2796 if (!!(trace_flags
& mask
) == !!enabled
)
2800 trace_flags
|= mask
;
2802 trace_flags
&= ~mask
;
2804 if (mask
== TRACE_ITER_RECORD_CMD
)
2805 trace_event_enable_cmd_record(enabled
);
2807 if (mask
== TRACE_ITER_OVERWRITE
)
2808 ring_buffer_change_overwrite(global_trace
.buffer
, enabled
);
2810 if (mask
== TRACE_ITER_PRINTK
)
2811 trace_printk_start_stop_comm(enabled
);
2815 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2816 size_t cnt
, loff_t
*ppos
)
2824 if (cnt
>= sizeof(buf
))
2827 if (copy_from_user(&buf
, ubuf
, cnt
))
2831 cmp
= strstrip(buf
);
2833 if (strncmp(cmp
, "no", 2) == 0) {
2838 for (i
= 0; trace_options
[i
]; i
++) {
2839 if (strcmp(cmp
, trace_options
[i
]) == 0) {
2840 set_tracer_flags(1 << i
, !neg
);
2845 /* If no option could be set, test the specific tracer options */
2846 if (!trace_options
[i
]) {
2847 mutex_lock(&trace_types_lock
);
2848 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2849 mutex_unlock(&trace_types_lock
);
2859 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
2861 if (tracing_disabled
)
2863 return single_open(file
, tracing_trace_options_show
, NULL
);
2866 static const struct file_operations tracing_iter_fops
= {
2867 .open
= tracing_trace_options_open
,
2869 .llseek
= seq_lseek
,
2870 .release
= single_release
,
2871 .write
= tracing_trace_options_write
,
2874 static const char readme_msg
[] =
2875 "tracing mini-HOWTO:\n\n"
2876 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2877 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2878 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2879 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2881 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2882 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2884 "# cat /sys/kernel/debug/tracing/trace_options\n"
2885 "noprint-parent nosym-offset nosym-addr noverbose\n"
2886 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2887 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2888 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2889 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2893 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2894 size_t cnt
, loff_t
*ppos
)
2896 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2897 readme_msg
, strlen(readme_msg
));
2900 static const struct file_operations tracing_readme_fops
= {
2901 .open
= tracing_open_generic
,
2902 .read
= tracing_readme_read
,
2903 .llseek
= generic_file_llseek
,
2907 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
2908 size_t cnt
, loff_t
*ppos
)
2917 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
2921 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
2929 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
2932 pid
= map_cmdline_to_pid
[i
];
2933 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
2936 trace_find_cmdline(pid
, buf_comm
);
2937 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
2942 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
2951 static const struct file_operations tracing_saved_cmdlines_fops
= {
2952 .open
= tracing_open_generic
,
2953 .read
= tracing_saved_cmdlines_read
,
2954 .llseek
= generic_file_llseek
,
2958 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2959 size_t cnt
, loff_t
*ppos
)
2961 char buf
[MAX_TRACER_SIZE
+2];
2964 mutex_lock(&trace_types_lock
);
2966 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2968 r
= sprintf(buf
, "\n");
2969 mutex_unlock(&trace_types_lock
);
2971 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2974 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
2976 tracing_reset_online_cpus(tr
);
2980 static void set_buffer_entries(struct trace_array
*tr
, unsigned long val
)
2983 for_each_tracing_cpu(cpu
)
2984 tr
->data
[cpu
]->entries
= val
;
2987 static int __tracing_resize_ring_buffer(unsigned long size
, int cpu
)
2992 * If kernel or user changes the size of the ring buffer
2993 * we use the size that was given, and we can forget about
2994 * expanding it later.
2996 ring_buffer_expanded
= 1;
2998 /* May be called before buffers are initialized */
2999 if (!global_trace
.buffer
)
3002 ret
= ring_buffer_resize(global_trace
.buffer
, size
, cpu
);
3006 if (!current_trace
->use_max_tr
)
3009 ret
= ring_buffer_resize(max_tr
.buffer
, size
, cpu
);
3013 if (cpu
== RING_BUFFER_ALL_CPUS
) {
3015 for_each_tracing_cpu(i
) {
3016 r
= ring_buffer_resize(global_trace
.buffer
,
3017 global_trace
.data
[i
]->entries
,
3023 r
= ring_buffer_resize(global_trace
.buffer
,
3024 global_trace
.data
[cpu
]->entries
,
3030 * AARGH! We are left with different
3031 * size max buffer!!!!
3032 * The max buffer is our "snapshot" buffer.
3033 * When a tracer needs a snapshot (one of the
3034 * latency tracers), it swaps the max buffer
3035 * with the saved snap shot. We succeeded to
3036 * update the size of the main buffer, but failed to
3037 * update the size of the max buffer. But when we tried
3038 * to reset the main buffer to the original size, we
3039 * failed there too. This is very unlikely to
3040 * happen, but if it does, warn and kill all
3044 tracing_disabled
= 1;
3049 if (cpu
== RING_BUFFER_ALL_CPUS
)
3050 set_buffer_entries(&max_tr
, size
);
3052 max_tr
.data
[cpu
]->entries
= size
;
3055 if (cpu
== RING_BUFFER_ALL_CPUS
)
3056 set_buffer_entries(&global_trace
, size
);
3058 global_trace
.data
[cpu
]->entries
= size
;
3063 static ssize_t
tracing_resize_ring_buffer(unsigned long size
, int cpu_id
)
3067 mutex_lock(&trace_types_lock
);
3069 if (cpu_id
!= RING_BUFFER_ALL_CPUS
) {
3070 /* make sure, this cpu is enabled in the mask */
3071 if (!cpumask_test_cpu(cpu_id
, tracing_buffer_mask
)) {
3077 ret
= __tracing_resize_ring_buffer(size
, cpu_id
);
3082 mutex_unlock(&trace_types_lock
);
3089 * tracing_update_buffers - used by tracing facility to expand ring buffers
3091 * To save on memory when the tracing is never used on a system with it
3092 * configured in. The ring buffers are set to a minimum size. But once
3093 * a user starts to use the tracing facility, then they need to grow
3094 * to their default size.
3096 * This function is to be called when a tracer is about to be used.
3098 int tracing_update_buffers(void)
3102 mutex_lock(&trace_types_lock
);
3103 if (!ring_buffer_expanded
)
3104 ret
= __tracing_resize_ring_buffer(trace_buf_size
,
3105 RING_BUFFER_ALL_CPUS
);
3106 mutex_unlock(&trace_types_lock
);
3111 struct trace_option_dentry
;
3113 static struct trace_option_dentry
*
3114 create_trace_option_files(struct tracer
*tracer
);
3117 destroy_trace_option_files(struct trace_option_dentry
*topts
);
3119 static int tracing_set_tracer(const char *buf
)
3121 static struct trace_option_dentry
*topts
;
3122 struct trace_array
*tr
= &global_trace
;
3126 mutex_lock(&trace_types_lock
);
3128 if (!ring_buffer_expanded
) {
3129 ret
= __tracing_resize_ring_buffer(trace_buf_size
,
3130 RING_BUFFER_ALL_CPUS
);
3136 for (t
= trace_types
; t
; t
= t
->next
) {
3137 if (strcmp(t
->name
, buf
) == 0)
3144 if (t
== current_trace
)
3147 trace_branch_disable();
3148 if (current_trace
&& current_trace
->reset
)
3149 current_trace
->reset(tr
);
3150 if (current_trace
&& current_trace
->use_max_tr
) {
3152 * We don't free the ring buffer. instead, resize it because
3153 * The max_tr ring buffer has some state (e.g. ring->clock) and
3154 * we want preserve it.
3156 ring_buffer_resize(max_tr
.buffer
, 1, RING_BUFFER_ALL_CPUS
);
3157 set_buffer_entries(&max_tr
, 1);
3159 destroy_trace_option_files(topts
);
3161 current_trace
= &nop_trace
;
3163 topts
= create_trace_option_files(t
);
3164 if (t
->use_max_tr
) {
3166 /* we need to make per cpu buffer sizes equivalent */
3167 for_each_tracing_cpu(cpu
) {
3168 ret
= ring_buffer_resize(max_tr
.buffer
,
3169 global_trace
.data
[cpu
]->entries
,
3173 max_tr
.data
[cpu
]->entries
=
3174 global_trace
.data
[cpu
]->entries
;
3179 ret
= tracer_init(t
, tr
);
3185 trace_branch_enable(tr
);
3187 mutex_unlock(&trace_types_lock
);
3193 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
3194 size_t cnt
, loff_t
*ppos
)
3196 char buf
[MAX_TRACER_SIZE
+1];
3203 if (cnt
> MAX_TRACER_SIZE
)
3204 cnt
= MAX_TRACER_SIZE
;
3206 if (copy_from_user(&buf
, ubuf
, cnt
))
3211 /* strip ending whitespace. */
3212 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
3215 err
= tracing_set_tracer(buf
);
3225 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
3226 size_t cnt
, loff_t
*ppos
)
3228 unsigned long *ptr
= filp
->private_data
;
3232 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
3233 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
3234 if (r
> sizeof(buf
))
3236 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3240 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
3241 size_t cnt
, loff_t
*ppos
)
3243 unsigned long *ptr
= filp
->private_data
;
3247 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3256 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
3258 long cpu_file
= (long) inode
->i_private
;
3259 struct trace_iterator
*iter
;
3262 if (tracing_disabled
)
3265 mutex_lock(&trace_types_lock
);
3267 /* create a buffer to store the information to pass to userspace */
3268 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
3275 * We make a copy of the current tracer to avoid concurrent
3276 * changes on it while we are reading.
3278 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
3284 *iter
->trace
= *current_trace
;
3286 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
3291 /* trace pipe does not show start of buffer */
3292 cpumask_setall(iter
->started
);
3294 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
3295 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
3297 iter
->cpu_file
= cpu_file
;
3298 iter
->tr
= &global_trace
;
3299 mutex_init(&iter
->mutex
);
3300 filp
->private_data
= iter
;
3302 if (iter
->trace
->pipe_open
)
3303 iter
->trace
->pipe_open(iter
);
3305 nonseekable_open(inode
, filp
);
3307 mutex_unlock(&trace_types_lock
);
3313 mutex_unlock(&trace_types_lock
);
3317 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
3319 struct trace_iterator
*iter
= file
->private_data
;
3321 mutex_lock(&trace_types_lock
);
3323 if (iter
->trace
->pipe_close
)
3324 iter
->trace
->pipe_close(iter
);
3326 mutex_unlock(&trace_types_lock
);
3328 free_cpumask_var(iter
->started
);
3329 mutex_destroy(&iter
->mutex
);
3337 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
3339 struct trace_iterator
*iter
= filp
->private_data
;
3341 if (trace_flags
& TRACE_ITER_BLOCK
) {
3343 * Always select as readable when in blocking mode
3345 return POLLIN
| POLLRDNORM
;
3347 if (!trace_empty(iter
))
3348 return POLLIN
| POLLRDNORM
;
3349 poll_wait(filp
, &trace_wait
, poll_table
);
3350 if (!trace_empty(iter
))
3351 return POLLIN
| POLLRDNORM
;
3358 void default_wait_pipe(struct trace_iterator
*iter
)
3362 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
3364 if (trace_empty(iter
))
3367 finish_wait(&trace_wait
, &wait
);
3371 * This is a make-shift waitqueue.
3372 * A tracer might use this callback on some rare cases:
3374 * 1) the current tracer might hold the runqueue lock when it wakes up
3375 * a reader, hence a deadlock (sched, function, and function graph tracers)
3376 * 2) the function tracers, trace all functions, we don't want
3377 * the overhead of calling wake_up and friends
3378 * (and tracing them too)
3380 * Anyway, this is really very primitive wakeup.
3382 void poll_wait_pipe(struct trace_iterator
*iter
)
3384 set_current_state(TASK_INTERRUPTIBLE
);
3385 /* sleep for 100 msecs, and try again. */
3386 schedule_timeout(HZ
/ 10);
3389 /* Must be called with trace_types_lock mutex held. */
3390 static int tracing_wait_pipe(struct file
*filp
)
3392 struct trace_iterator
*iter
= filp
->private_data
;
3394 while (trace_empty(iter
)) {
3396 if ((filp
->f_flags
& O_NONBLOCK
)) {
3400 mutex_unlock(&iter
->mutex
);
3402 iter
->trace
->wait_pipe(iter
);
3404 mutex_lock(&iter
->mutex
);
3406 if (signal_pending(current
))
3410 * We block until we read something and tracing is enabled.
3411 * We still block if tracing is disabled, but we have never
3412 * read anything. This allows a user to cat this file, and
3413 * then enable tracing. But after we have read something,
3414 * we give an EOF when tracing is again disabled.
3416 * iter->pos will be 0 if we haven't read anything.
3418 if (tracing_is_enabled() && iter
->pos
)
3429 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
3430 size_t cnt
, loff_t
*ppos
)
3432 struct trace_iterator
*iter
= filp
->private_data
;
3433 static struct tracer
*old_tracer
;
3436 /* return any leftover data */
3437 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3441 trace_seq_init(&iter
->seq
);
3443 /* copy the tracer to avoid using a global lock all around */
3444 mutex_lock(&trace_types_lock
);
3445 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3446 old_tracer
= current_trace
;
3447 *iter
->trace
= *current_trace
;
3449 mutex_unlock(&trace_types_lock
);
3452 * Avoid more than one consumer on a single file descriptor
3453 * This is just a matter of traces coherency, the ring buffer itself
3456 mutex_lock(&iter
->mutex
);
3457 if (iter
->trace
->read
) {
3458 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
3464 sret
= tracing_wait_pipe(filp
);
3468 /* stop when tracing is finished */
3469 if (trace_empty(iter
)) {
3474 if (cnt
>= PAGE_SIZE
)
3475 cnt
= PAGE_SIZE
- 1;
3477 /* reset all but tr, trace, and overruns */
3478 memset(&iter
->seq
, 0,
3479 sizeof(struct trace_iterator
) -
3480 offsetof(struct trace_iterator
, seq
));
3483 trace_event_read_lock();
3484 trace_access_lock(iter
->cpu_file
);
3485 while (trace_find_next_entry_inc(iter
) != NULL
) {
3486 enum print_line_t ret
;
3487 int len
= iter
->seq
.len
;
3489 ret
= print_trace_line(iter
);
3490 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3491 /* don't print partial lines */
3492 iter
->seq
.len
= len
;
3495 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3496 trace_consume(iter
);
3498 if (iter
->seq
.len
>= cnt
)
3502 * Setting the full flag means we reached the trace_seq buffer
3503 * size and we should leave by partial output condition above.
3504 * One of the trace_seq_* functions is not used properly.
3506 WARN_ONCE(iter
->seq
.full
, "full flag set for trace type %d",
3509 trace_access_unlock(iter
->cpu_file
);
3510 trace_event_read_unlock();
3512 /* Now copy what we have to the user */
3513 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3514 if (iter
->seq
.readpos
>= iter
->seq
.len
)
3515 trace_seq_init(&iter
->seq
);
3518 * If there was nothing to send to user, in spite of consuming trace
3519 * entries, go back to wait for more entries.
3525 mutex_unlock(&iter
->mutex
);
3530 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
3531 struct pipe_buffer
*buf
)
3533 __free_page(buf
->page
);
3536 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
3539 __free_page(spd
->pages
[idx
]);
3542 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
3544 .map
= generic_pipe_buf_map
,
3545 .unmap
= generic_pipe_buf_unmap
,
3546 .confirm
= generic_pipe_buf_confirm
,
3547 .release
= tracing_pipe_buf_release
,
3548 .steal
= generic_pipe_buf_steal
,
3549 .get
= generic_pipe_buf_get
,
3553 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
3558 /* Seq buffer is page-sized, exactly what we need. */
3560 count
= iter
->seq
.len
;
3561 ret
= print_trace_line(iter
);
3562 count
= iter
->seq
.len
- count
;
3565 iter
->seq
.len
-= count
;
3568 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3569 iter
->seq
.len
-= count
;
3573 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3574 trace_consume(iter
);
3576 if (!trace_find_next_entry_inc(iter
)) {
3586 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3588 struct pipe_inode_info
*pipe
,
3592 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3593 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3594 struct trace_iterator
*iter
= filp
->private_data
;
3595 struct splice_pipe_desc spd
= {
3597 .partial
= partial_def
,
3598 .nr_pages
= 0, /* This gets updated below. */
3599 .nr_pages_max
= PIPE_DEF_BUFFERS
,
3601 .ops
= &tracing_pipe_buf_ops
,
3602 .spd_release
= tracing_spd_release_pipe
,
3604 static struct tracer
*old_tracer
;
3609 if (splice_grow_spd(pipe
, &spd
))
3612 /* copy the tracer to avoid using a global lock all around */
3613 mutex_lock(&trace_types_lock
);
3614 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3615 old_tracer
= current_trace
;
3616 *iter
->trace
= *current_trace
;
3618 mutex_unlock(&trace_types_lock
);
3620 mutex_lock(&iter
->mutex
);
3622 if (iter
->trace
->splice_read
) {
3623 ret
= iter
->trace
->splice_read(iter
, filp
,
3624 ppos
, pipe
, len
, flags
);
3629 ret
= tracing_wait_pipe(filp
);
3633 if (!iter
->ent
&& !trace_find_next_entry_inc(iter
)) {
3638 trace_event_read_lock();
3639 trace_access_lock(iter
->cpu_file
);
3641 /* Fill as many pages as possible. */
3642 for (i
= 0, rem
= len
; i
< pipe
->buffers
&& rem
; i
++) {
3643 spd
.pages
[i
] = alloc_page(GFP_KERNEL
);
3647 rem
= tracing_fill_pipe_page(rem
, iter
);
3649 /* Copy the data into the page, so we can start over. */
3650 ret
= trace_seq_to_buffer(&iter
->seq
,
3651 page_address(spd
.pages
[i
]),
3654 __free_page(spd
.pages
[i
]);
3657 spd
.partial
[i
].offset
= 0;
3658 spd
.partial
[i
].len
= iter
->seq
.len
;
3660 trace_seq_init(&iter
->seq
);
3663 trace_access_unlock(iter
->cpu_file
);
3664 trace_event_read_unlock();
3665 mutex_unlock(&iter
->mutex
);
3669 ret
= splice_to_pipe(pipe
, &spd
);
3671 splice_shrink_spd(&spd
);
3675 mutex_unlock(&iter
->mutex
);
3679 struct ftrace_entries_info
{
3680 struct trace_array
*tr
;
3684 static int tracing_entries_open(struct inode
*inode
, struct file
*filp
)
3686 struct ftrace_entries_info
*info
;
3688 if (tracing_disabled
)
3691 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3695 info
->tr
= &global_trace
;
3696 info
->cpu
= (unsigned long)inode
->i_private
;
3698 filp
->private_data
= info
;
3704 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3705 size_t cnt
, loff_t
*ppos
)
3707 struct ftrace_entries_info
*info
= filp
->private_data
;
3708 struct trace_array
*tr
= info
->tr
;
3713 mutex_lock(&trace_types_lock
);
3715 if (info
->cpu
== RING_BUFFER_ALL_CPUS
) {
3716 int cpu
, buf_size_same
;
3721 /* check if all cpu sizes are same */
3722 for_each_tracing_cpu(cpu
) {
3723 /* fill in the size from first enabled cpu */
3725 size
= tr
->data
[cpu
]->entries
;
3726 if (size
!= tr
->data
[cpu
]->entries
) {
3732 if (buf_size_same
) {
3733 if (!ring_buffer_expanded
)
3734 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3736 trace_buf_size
>> 10);
3738 r
= sprintf(buf
, "%lu\n", size
>> 10);
3740 r
= sprintf(buf
, "X\n");
3742 r
= sprintf(buf
, "%lu\n", tr
->data
[info
->cpu
]->entries
>> 10);
3744 mutex_unlock(&trace_types_lock
);
3746 ret
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3751 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3752 size_t cnt
, loff_t
*ppos
)
3754 struct ftrace_entries_info
*info
= filp
->private_data
;
3758 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3762 /* must have at least 1 entry */
3766 /* value is in KB */
3769 ret
= tracing_resize_ring_buffer(val
, info
->cpu
);
3779 tracing_entries_release(struct inode
*inode
, struct file
*filp
)
3781 struct ftrace_entries_info
*info
= filp
->private_data
;
3789 tracing_total_entries_read(struct file
*filp
, char __user
*ubuf
,
3790 size_t cnt
, loff_t
*ppos
)
3792 struct trace_array
*tr
= filp
->private_data
;
3795 unsigned long size
= 0, expanded_size
= 0;
3797 mutex_lock(&trace_types_lock
);
3798 for_each_tracing_cpu(cpu
) {
3799 size
+= tr
->data
[cpu
]->entries
>> 10;
3800 if (!ring_buffer_expanded
)
3801 expanded_size
+= trace_buf_size
>> 10;
3803 if (ring_buffer_expanded
)
3804 r
= sprintf(buf
, "%lu\n", size
);
3806 r
= sprintf(buf
, "%lu (expanded: %lu)\n", size
, expanded_size
);
3807 mutex_unlock(&trace_types_lock
);
3809 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3813 tracing_free_buffer_write(struct file
*filp
, const char __user
*ubuf
,
3814 size_t cnt
, loff_t
*ppos
)
3817 * There is no need to read what the user has written, this function
3818 * is just to make sure that there is no error when "echo" is used
3827 tracing_free_buffer_release(struct inode
*inode
, struct file
*filp
)
3829 /* disable tracing ? */
3830 if (trace_flags
& TRACE_ITER_STOP_ON_FREE
)
3832 /* resize the ring buffer to 0 */
3833 tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS
);
3839 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3840 size_t cnt
, loff_t
*fpos
)
3842 unsigned long addr
= (unsigned long)ubuf
;
3843 struct ring_buffer_event
*event
;
3844 struct ring_buffer
*buffer
;
3845 struct print_entry
*entry
;
3846 unsigned long irq_flags
;
3847 struct page
*pages
[2];
3857 if (tracing_disabled
)
3860 if (!(trace_flags
& TRACE_ITER_MARKERS
))
3863 if (cnt
> TRACE_BUF_SIZE
)
3864 cnt
= TRACE_BUF_SIZE
;
3867 * Userspace is injecting traces into the kernel trace buffer.
3868 * We want to be as non intrusive as possible.
3869 * To do so, we do not want to allocate any special buffers
3870 * or take any locks, but instead write the userspace data
3871 * straight into the ring buffer.
3873 * First we need to pin the userspace buffer into memory,
3874 * which, most likely it is, because it just referenced it.
3875 * But there's no guarantee that it is. By using get_user_pages_fast()
3876 * and kmap_atomic/kunmap_atomic() we can get access to the
3877 * pages directly. We then write the data directly into the
3880 BUILD_BUG_ON(TRACE_BUF_SIZE
>= PAGE_SIZE
);
3882 /* check if we cross pages */
3883 if ((addr
& PAGE_MASK
) != ((addr
+ cnt
) & PAGE_MASK
))
3886 offset
= addr
& (PAGE_SIZE
- 1);
3889 ret
= get_user_pages_fast(addr
, nr_pages
, 0, pages
);
3890 if (ret
< nr_pages
) {
3892 put_page(pages
[ret
]);
3897 for (i
= 0; i
< nr_pages
; i
++)
3898 map_page
[i
] = kmap_atomic(pages
[i
]);
3900 local_save_flags(irq_flags
);
3901 size
= sizeof(*entry
) + cnt
+ 2; /* possible \n added */
3902 buffer
= global_trace
.buffer
;
3903 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
3904 irq_flags
, preempt_count());
3906 /* Ring buffer disabled, return as if not open for write */
3911 entry
= ring_buffer_event_data(event
);
3912 entry
->ip
= _THIS_IP_
;
3914 if (nr_pages
== 2) {
3915 len
= PAGE_SIZE
- offset
;
3916 memcpy(&entry
->buf
, map_page
[0] + offset
, len
);
3917 memcpy(&entry
->buf
[len
], map_page
[1], cnt
- len
);
3919 memcpy(&entry
->buf
, map_page
[0] + offset
, cnt
);
3921 if (entry
->buf
[cnt
- 1] != '\n') {
3922 entry
->buf
[cnt
] = '\n';
3923 entry
->buf
[cnt
+ 1] = '\0';
3925 entry
->buf
[cnt
] = '\0';
3927 __buffer_unlock_commit(buffer
, event
);
3934 for (i
= 0; i
< nr_pages
; i
++){
3935 kunmap_atomic(map_page
[i
]);
3942 static int tracing_clock_show(struct seq_file
*m
, void *v
)
3946 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
3948 "%s%s%s%s", i
? " " : "",
3949 i
== trace_clock_id
? "[" : "", trace_clocks
[i
].name
,
3950 i
== trace_clock_id
? "]" : "");
3956 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
3957 size_t cnt
, loff_t
*fpos
)
3960 const char *clockstr
;
3963 if (cnt
>= sizeof(buf
))
3966 if (copy_from_user(&buf
, ubuf
, cnt
))
3971 clockstr
= strstrip(buf
);
3973 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
3974 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
3977 if (i
== ARRAY_SIZE(trace_clocks
))
3982 mutex_lock(&trace_types_lock
);
3984 ring_buffer_set_clock(global_trace
.buffer
, trace_clocks
[i
].func
);
3986 ring_buffer_set_clock(max_tr
.buffer
, trace_clocks
[i
].func
);
3989 * New clock may not be consistent with the previous clock.
3990 * Reset the buffer so that it doesn't have incomparable timestamps.
3992 tracing_reset_online_cpus(&global_trace
);
3994 tracing_reset_online_cpus(&max_tr
);
3996 mutex_unlock(&trace_types_lock
);
4003 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
4005 if (tracing_disabled
)
4007 return single_open(file
, tracing_clock_show
, NULL
);
4010 static const struct file_operations tracing_max_lat_fops
= {
4011 .open
= tracing_open_generic
,
4012 .read
= tracing_max_lat_read
,
4013 .write
= tracing_max_lat_write
,
4014 .llseek
= generic_file_llseek
,
4017 static const struct file_operations set_tracer_fops
= {
4018 .open
= tracing_open_generic
,
4019 .read
= tracing_set_trace_read
,
4020 .write
= tracing_set_trace_write
,
4021 .llseek
= generic_file_llseek
,
4024 static const struct file_operations tracing_pipe_fops
= {
4025 .open
= tracing_open_pipe
,
4026 .poll
= tracing_poll_pipe
,
4027 .read
= tracing_read_pipe
,
4028 .splice_read
= tracing_splice_read_pipe
,
4029 .release
= tracing_release_pipe
,
4030 .llseek
= no_llseek
,
4033 static const struct file_operations tracing_entries_fops
= {
4034 .open
= tracing_entries_open
,
4035 .read
= tracing_entries_read
,
4036 .write
= tracing_entries_write
,
4037 .release
= tracing_entries_release
,
4038 .llseek
= generic_file_llseek
,
4041 static const struct file_operations tracing_total_entries_fops
= {
4042 .open
= tracing_open_generic
,
4043 .read
= tracing_total_entries_read
,
4044 .llseek
= generic_file_llseek
,
4047 static const struct file_operations tracing_free_buffer_fops
= {
4048 .write
= tracing_free_buffer_write
,
4049 .release
= tracing_free_buffer_release
,
4052 static const struct file_operations tracing_mark_fops
= {
4053 .open
= tracing_open_generic
,
4054 .write
= tracing_mark_write
,
4055 .llseek
= generic_file_llseek
,
4058 static const struct file_operations trace_clock_fops
= {
4059 .open
= tracing_clock_open
,
4061 .llseek
= seq_lseek
,
4062 .release
= single_release
,
4063 .write
= tracing_clock_write
,
4066 struct ftrace_buffer_info
{
4067 struct trace_array
*tr
;
4073 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
4075 int cpu
= (int)(long)inode
->i_private
;
4076 struct ftrace_buffer_info
*info
;
4078 if (tracing_disabled
)
4081 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
4085 info
->tr
= &global_trace
;
4088 /* Force reading ring buffer for first read */
4089 info
->read
= (unsigned int)-1;
4091 filp
->private_data
= info
;
4093 return nonseekable_open(inode
, filp
);
4097 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
4098 size_t count
, loff_t
*ppos
)
4100 struct ftrace_buffer_info
*info
= filp
->private_data
;
4108 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
, info
->cpu
);
4112 /* Do we have previous read data to read? */
4113 if (info
->read
< PAGE_SIZE
)
4116 trace_access_lock(info
->cpu
);
4117 ret
= ring_buffer_read_page(info
->tr
->buffer
,
4121 trace_access_unlock(info
->cpu
);
4128 size
= PAGE_SIZE
- info
->read
;
4132 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
4143 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
4145 struct ftrace_buffer_info
*info
= file
->private_data
;
4148 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
4155 struct ring_buffer
*buffer
;
4160 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
4161 struct pipe_buffer
*buf
)
4163 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
4168 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4173 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
4174 struct pipe_buffer
*buf
)
4176 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
4181 /* Pipe buffer operations for a buffer. */
4182 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
4184 .map
= generic_pipe_buf_map
,
4185 .unmap
= generic_pipe_buf_unmap
,
4186 .confirm
= generic_pipe_buf_confirm
,
4187 .release
= buffer_pipe_buf_release
,
4188 .steal
= generic_pipe_buf_steal
,
4189 .get
= buffer_pipe_buf_get
,
4193 * Callback from splice_to_pipe(), if we need to release some pages
4194 * at the end of the spd in case we error'ed out in filling the pipe.
4196 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
4198 struct buffer_ref
*ref
=
4199 (struct buffer_ref
*)spd
->partial
[i
].private;
4204 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4206 spd
->partial
[i
].private = 0;
4210 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
4211 struct pipe_inode_info
*pipe
, size_t len
,
4214 struct ftrace_buffer_info
*info
= file
->private_data
;
4215 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
4216 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
4217 struct splice_pipe_desc spd
= {
4219 .partial
= partial_def
,
4220 .nr_pages_max
= PIPE_DEF_BUFFERS
,
4222 .ops
= &buffer_pipe_buf_ops
,
4223 .spd_release
= buffer_spd_release
,
4225 struct buffer_ref
*ref
;
4226 int entries
, size
, i
;
4229 if (splice_grow_spd(pipe
, &spd
))
4232 if (*ppos
& (PAGE_SIZE
- 1)) {
4233 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
4238 if (len
& (PAGE_SIZE
- 1)) {
4239 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
4240 if (len
< PAGE_SIZE
) {
4247 trace_access_lock(info
->cpu
);
4248 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4250 for (i
= 0; i
< pipe
->buffers
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
4254 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
4259 ref
->buffer
= info
->tr
->buffer
;
4260 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
, info
->cpu
);
4266 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
4269 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4275 * zero out any left over data, this is going to
4278 size
= ring_buffer_page_len(ref
->page
);
4279 if (size
< PAGE_SIZE
)
4280 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
4282 page
= virt_to_page(ref
->page
);
4284 spd
.pages
[i
] = page
;
4285 spd
.partial
[i
].len
= PAGE_SIZE
;
4286 spd
.partial
[i
].offset
= 0;
4287 spd
.partial
[i
].private = (unsigned long)ref
;
4291 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4294 trace_access_unlock(info
->cpu
);
4297 /* did we read anything? */
4298 if (!spd
.nr_pages
) {
4299 if (flags
& SPLICE_F_NONBLOCK
)
4307 ret
= splice_to_pipe(pipe
, &spd
);
4308 splice_shrink_spd(&spd
);
4313 static const struct file_operations tracing_buffers_fops
= {
4314 .open
= tracing_buffers_open
,
4315 .read
= tracing_buffers_read
,
4316 .release
= tracing_buffers_release
,
4317 .splice_read
= tracing_buffers_splice_read
,
4318 .llseek
= no_llseek
,
4322 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
4323 size_t count
, loff_t
*ppos
)
4325 unsigned long cpu
= (unsigned long)filp
->private_data
;
4326 struct trace_array
*tr
= &global_trace
;
4327 struct trace_seq
*s
;
4329 unsigned long long t
;
4330 unsigned long usec_rem
;
4332 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
4338 cnt
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
4339 trace_seq_printf(s
, "entries: %ld\n", cnt
);
4341 cnt
= ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
4342 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
4344 cnt
= ring_buffer_commit_overrun_cpu(tr
->buffer
, cpu
);
4345 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
4347 cnt
= ring_buffer_bytes_cpu(tr
->buffer
, cpu
);
4348 trace_seq_printf(s
, "bytes: %ld\n", cnt
);
4350 t
= ns2usecs(ring_buffer_oldest_event_ts(tr
->buffer
, cpu
));
4351 usec_rem
= do_div(t
, USEC_PER_SEC
);
4352 trace_seq_printf(s
, "oldest event ts: %5llu.%06lu\n", t
, usec_rem
);
4354 t
= ns2usecs(ring_buffer_time_stamp(tr
->buffer
, cpu
));
4355 usec_rem
= do_div(t
, USEC_PER_SEC
);
4356 trace_seq_printf(s
, "now ts: %5llu.%06lu\n", t
, usec_rem
);
4358 cnt
= ring_buffer_dropped_events_cpu(tr
->buffer
, cpu
);
4359 trace_seq_printf(s
, "dropped events: %ld\n", cnt
);
4361 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
4368 static const struct file_operations tracing_stats_fops
= {
4369 .open
= tracing_open_generic
,
4370 .read
= tracing_stats_read
,
4371 .llseek
= generic_file_llseek
,
4374 #ifdef CONFIG_DYNAMIC_FTRACE
4376 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
4382 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
4383 size_t cnt
, loff_t
*ppos
)
4385 static char ftrace_dyn_info_buffer
[1024];
4386 static DEFINE_MUTEX(dyn_info_mutex
);
4387 unsigned long *p
= filp
->private_data
;
4388 char *buf
= ftrace_dyn_info_buffer
;
4389 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
4392 mutex_lock(&dyn_info_mutex
);
4393 r
= sprintf(buf
, "%ld ", *p
);
4395 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
4398 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4400 mutex_unlock(&dyn_info_mutex
);
4405 static const struct file_operations tracing_dyn_info_fops
= {
4406 .open
= tracing_open_generic
,
4407 .read
= tracing_read_dyn_info
,
4408 .llseek
= generic_file_llseek
,
4412 static struct dentry
*d_tracer
;
4414 struct dentry
*tracing_init_dentry(void)
4421 if (!debugfs_initialized())
4424 d_tracer
= debugfs_create_dir("tracing", NULL
);
4426 if (!d_tracer
&& !once
) {
4428 pr_warning("Could not create debugfs directory 'tracing'\n");
4435 static struct dentry
*d_percpu
;
4437 struct dentry
*tracing_dentry_percpu(void)
4440 struct dentry
*d_tracer
;
4445 d_tracer
= tracing_init_dentry();
4450 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
4452 if (!d_percpu
&& !once
) {
4454 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4461 static void tracing_init_debugfs_percpu(long cpu
)
4463 struct dentry
*d_percpu
= tracing_dentry_percpu();
4464 struct dentry
*d_cpu
;
4465 char cpu_dir
[30]; /* 30 characters should be more than enough */
4470 snprintf(cpu_dir
, 30, "cpu%ld", cpu
);
4471 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
4473 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
4477 /* per cpu trace_pipe */
4478 trace_create_file("trace_pipe", 0444, d_cpu
,
4479 (void *) cpu
, &tracing_pipe_fops
);
4482 trace_create_file("trace", 0644, d_cpu
,
4483 (void *) cpu
, &tracing_fops
);
4485 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
4486 (void *) cpu
, &tracing_buffers_fops
);
4488 trace_create_file("stats", 0444, d_cpu
,
4489 (void *) cpu
, &tracing_stats_fops
);
4491 trace_create_file("buffer_size_kb", 0444, d_cpu
,
4492 (void *) cpu
, &tracing_entries_fops
);
4495 #ifdef CONFIG_FTRACE_SELFTEST
4496 /* Let selftest have access to static functions in this file */
4497 #include "trace_selftest.c"
4500 struct trace_option_dentry
{
4501 struct tracer_opt
*opt
;
4502 struct tracer_flags
*flags
;
4503 struct dentry
*entry
;
4507 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4510 struct trace_option_dentry
*topt
= filp
->private_data
;
4513 if (topt
->flags
->val
& topt
->opt
->bit
)
4518 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4522 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4525 struct trace_option_dentry
*topt
= filp
->private_data
;
4529 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4533 if (val
!= 0 && val
!= 1)
4536 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
4537 mutex_lock(&trace_types_lock
);
4538 ret
= __set_tracer_option(current_trace
, topt
->flags
,
4540 mutex_unlock(&trace_types_lock
);
4551 static const struct file_operations trace_options_fops
= {
4552 .open
= tracing_open_generic
,
4553 .read
= trace_options_read
,
4554 .write
= trace_options_write
,
4555 .llseek
= generic_file_llseek
,
4559 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4562 long index
= (long)filp
->private_data
;
4565 if (trace_flags
& (1 << index
))
4570 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4574 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4577 long index
= (long)filp
->private_data
;
4581 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4585 if (val
!= 0 && val
!= 1)
4587 set_tracer_flags(1 << index
, val
);
4594 static const struct file_operations trace_options_core_fops
= {
4595 .open
= tracing_open_generic
,
4596 .read
= trace_options_core_read
,
4597 .write
= trace_options_core_write
,
4598 .llseek
= generic_file_llseek
,
4601 struct dentry
*trace_create_file(const char *name
,
4603 struct dentry
*parent
,
4605 const struct file_operations
*fops
)
4609 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
4611 pr_warning("Could not create debugfs '%s' entry\n", name
);
4617 static struct dentry
*trace_options_init_dentry(void)
4619 struct dentry
*d_tracer
;
4620 static struct dentry
*t_options
;
4625 d_tracer
= tracing_init_dentry();
4629 t_options
= debugfs_create_dir("options", d_tracer
);
4631 pr_warning("Could not create debugfs directory 'options'\n");
4639 create_trace_option_file(struct trace_option_dentry
*topt
,
4640 struct tracer_flags
*flags
,
4641 struct tracer_opt
*opt
)
4643 struct dentry
*t_options
;
4645 t_options
= trace_options_init_dentry();
4649 topt
->flags
= flags
;
4652 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
4653 &trace_options_fops
);
4657 static struct trace_option_dentry
*
4658 create_trace_option_files(struct tracer
*tracer
)
4660 struct trace_option_dentry
*topts
;
4661 struct tracer_flags
*flags
;
4662 struct tracer_opt
*opts
;
4668 flags
= tracer
->flags
;
4670 if (!flags
|| !flags
->opts
)
4675 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4678 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
4682 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4683 create_trace_option_file(&topts
[cnt
], flags
,
4690 destroy_trace_option_files(struct trace_option_dentry
*topts
)
4697 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
4698 if (topts
[cnt
].entry
)
4699 debugfs_remove(topts
[cnt
].entry
);
4705 static struct dentry
*
4706 create_trace_option_core_file(const char *option
, long index
)
4708 struct dentry
*t_options
;
4710 t_options
= trace_options_init_dentry();
4714 return trace_create_file(option
, 0644, t_options
, (void *)index
,
4715 &trace_options_core_fops
);
4718 static __init
void create_trace_options_dir(void)
4720 struct dentry
*t_options
;
4723 t_options
= trace_options_init_dentry();
4727 for (i
= 0; trace_options
[i
]; i
++)
4728 create_trace_option_core_file(trace_options
[i
], i
);
4732 rb_simple_read(struct file
*filp
, char __user
*ubuf
,
4733 size_t cnt
, loff_t
*ppos
)
4735 struct trace_array
*tr
= filp
->private_data
;
4736 struct ring_buffer
*buffer
= tr
->buffer
;
4741 r
= ring_buffer_record_is_on(buffer
);
4745 r
= sprintf(buf
, "%d\n", r
);
4747 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4751 rb_simple_write(struct file
*filp
, const char __user
*ubuf
,
4752 size_t cnt
, loff_t
*ppos
)
4754 struct trace_array
*tr
= filp
->private_data
;
4755 struct ring_buffer
*buffer
= tr
->buffer
;
4759 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4765 ring_buffer_record_on(buffer
);
4767 ring_buffer_record_off(buffer
);
4775 static const struct file_operations rb_simple_fops
= {
4776 .open
= tracing_open_generic
,
4777 .read
= rb_simple_read
,
4778 .write
= rb_simple_write
,
4779 .llseek
= default_llseek
,
4782 static __init
int tracer_init_debugfs(void)
4784 struct dentry
*d_tracer
;
4787 trace_access_lock_init();
4789 d_tracer
= tracing_init_dentry();
4791 trace_create_file("tracing_enabled", 0644, d_tracer
,
4792 &global_trace
, &rb_simple_fops
);
4794 trace_create_file("trace_options", 0644, d_tracer
,
4795 NULL
, &tracing_iter_fops
);
4797 trace_create_file("tracing_cpumask", 0644, d_tracer
,
4798 NULL
, &tracing_cpumask_fops
);
4800 trace_create_file("trace", 0644, d_tracer
,
4801 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
4803 trace_create_file("available_tracers", 0444, d_tracer
,
4804 &global_trace
, &show_traces_fops
);
4806 trace_create_file("current_tracer", 0644, d_tracer
,
4807 &global_trace
, &set_tracer_fops
);
4809 #ifdef CONFIG_TRACER_MAX_TRACE
4810 trace_create_file("tracing_max_latency", 0644, d_tracer
,
4811 &tracing_max_latency
, &tracing_max_lat_fops
);
4814 trace_create_file("tracing_thresh", 0644, d_tracer
,
4815 &tracing_thresh
, &tracing_max_lat_fops
);
4817 trace_create_file("README", 0444, d_tracer
,
4818 NULL
, &tracing_readme_fops
);
4820 trace_create_file("trace_pipe", 0444, d_tracer
,
4821 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
4823 trace_create_file("buffer_size_kb", 0644, d_tracer
,
4824 (void *) RING_BUFFER_ALL_CPUS
, &tracing_entries_fops
);
4826 trace_create_file("buffer_total_size_kb", 0444, d_tracer
,
4827 &global_trace
, &tracing_total_entries_fops
);
4829 trace_create_file("free_buffer", 0644, d_tracer
,
4830 &global_trace
, &tracing_free_buffer_fops
);
4832 trace_create_file("trace_marker", 0220, d_tracer
,
4833 NULL
, &tracing_mark_fops
);
4835 trace_create_file("saved_cmdlines", 0444, d_tracer
,
4836 NULL
, &tracing_saved_cmdlines_fops
);
4838 trace_create_file("trace_clock", 0644, d_tracer
, NULL
,
4841 trace_create_file("tracing_on", 0644, d_tracer
,
4842 &global_trace
, &rb_simple_fops
);
4844 #ifdef CONFIG_DYNAMIC_FTRACE
4845 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
4846 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
4849 create_trace_options_dir();
4851 for_each_tracing_cpu(cpu
)
4852 tracing_init_debugfs_percpu(cpu
);
4857 static int trace_panic_handler(struct notifier_block
*this,
4858 unsigned long event
, void *unused
)
4860 if (ftrace_dump_on_oops
)
4861 ftrace_dump(ftrace_dump_on_oops
);
4865 static struct notifier_block trace_panic_notifier
= {
4866 .notifier_call
= trace_panic_handler
,
4868 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
4871 static int trace_die_handler(struct notifier_block
*self
,
4877 if (ftrace_dump_on_oops
)
4878 ftrace_dump(ftrace_dump_on_oops
);
4886 static struct notifier_block trace_die_notifier
= {
4887 .notifier_call
= trace_die_handler
,
4892 * printk is set to max of 1024, we really don't need it that big.
4893 * Nothing should be printing 1000 characters anyway.
4895 #define TRACE_MAX_PRINT 1000
4898 * Define here KERN_TRACE so that we have one place to modify
4899 * it if we decide to change what log level the ftrace dump
4902 #define KERN_TRACE KERN_EMERG
4905 trace_printk_seq(struct trace_seq
*s
)
4907 /* Probably should print a warning here. */
4911 /* should be zero ended, but we are paranoid. */
4912 s
->buffer
[s
->len
] = 0;
4914 printk(KERN_TRACE
"%s", s
->buffer
);
4919 void trace_init_global_iter(struct trace_iterator
*iter
)
4921 iter
->tr
= &global_trace
;
4922 iter
->trace
= current_trace
;
4923 iter
->cpu_file
= TRACE_PIPE_ALL_CPU
;
4927 __ftrace_dump(bool disable_tracing
, enum ftrace_dump_mode oops_dump_mode
)
4929 static arch_spinlock_t ftrace_dump_lock
=
4930 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
4931 /* use static because iter can be a bit big for the stack */
4932 static struct trace_iterator iter
;
4933 unsigned int old_userobj
;
4934 static int dump_ran
;
4935 unsigned long flags
;
4939 local_irq_save(flags
);
4940 arch_spin_lock(&ftrace_dump_lock
);
4948 /* Did function tracer already get disabled? */
4949 if (ftrace_is_dead()) {
4950 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
4951 printk("# MAY BE MISSING FUNCTION EVENTS\n");
4954 if (disable_tracing
)
4957 trace_init_global_iter(&iter
);
4959 for_each_tracing_cpu(cpu
) {
4960 atomic_inc(&iter
.tr
->data
[cpu
]->disabled
);
4963 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
4965 /* don't look at user memory in panic mode */
4966 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
4968 /* Simulate the iterator */
4969 iter
.tr
= &global_trace
;
4970 iter
.trace
= current_trace
;
4972 switch (oops_dump_mode
) {
4974 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4977 iter
.cpu_file
= raw_smp_processor_id();
4982 printk(KERN_TRACE
"Bad dumping mode, switching to all CPUs dump\n");
4983 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
4986 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
4989 * We need to stop all tracing on all CPUS to read the
4990 * the next buffer. This is a bit expensive, but is
4991 * not done often. We fill all what we can read,
4992 * and then release the locks again.
4995 while (!trace_empty(&iter
)) {
4998 printk(KERN_TRACE
"---------------------------------\n");
5002 /* reset all but tr, trace, and overruns */
5003 memset(&iter
.seq
, 0,
5004 sizeof(struct trace_iterator
) -
5005 offsetof(struct trace_iterator
, seq
));
5006 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
5009 if (trace_find_next_entry_inc(&iter
) != NULL
) {
5012 ret
= print_trace_line(&iter
);
5013 if (ret
!= TRACE_TYPE_NO_CONSUME
)
5014 trace_consume(&iter
);
5016 touch_nmi_watchdog();
5018 trace_printk_seq(&iter
.seq
);
5022 printk(KERN_TRACE
" (ftrace buffer empty)\n");
5024 printk(KERN_TRACE
"---------------------------------\n");
5027 /* Re-enable tracing if requested */
5028 if (!disable_tracing
) {
5029 trace_flags
|= old_userobj
;
5031 for_each_tracing_cpu(cpu
) {
5032 atomic_dec(&iter
.tr
->data
[cpu
]->disabled
);
5038 arch_spin_unlock(&ftrace_dump_lock
);
5039 local_irq_restore(flags
);
5042 /* By default: disable tracing after the dump */
5043 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode
)
5045 __ftrace_dump(true, oops_dump_mode
);
5047 EXPORT_SYMBOL_GPL(ftrace_dump
);
5049 __init
static int tracer_alloc_buffers(void)
5052 enum ring_buffer_flags rb_flags
;
5057 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
5060 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
5061 goto out_free_buffer_mask
;
5063 /* Only allocate trace_printk buffers if a trace_printk exists */
5064 if (__stop___trace_bprintk_fmt
!= __start___trace_bprintk_fmt
)
5065 /* Must be called before global_trace.buffer is allocated */
5066 trace_printk_init_buffers();
5068 /* To save memory, keep the ring buffer size to its minimum */
5069 if (ring_buffer_expanded
)
5070 ring_buf_size
= trace_buf_size
;
5074 rb_flags
= trace_flags
& TRACE_ITER_OVERWRITE
? RB_FL_OVERWRITE
: 0;
5076 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
5077 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
5079 /* TODO: make the number of buffers hot pluggable with CPUS */
5080 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
, rb_flags
);
5081 if (!global_trace
.buffer
) {
5082 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
5084 goto out_free_cpumask
;
5086 if (global_trace
.buffer_disabled
)
5090 #ifdef CONFIG_TRACER_MAX_TRACE
5091 max_tr
.buffer
= ring_buffer_alloc(1, rb_flags
);
5092 if (!max_tr
.buffer
) {
5093 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
5095 ring_buffer_free(global_trace
.buffer
);
5096 goto out_free_cpumask
;
5100 /* Allocate the first page for all buffers */
5101 for_each_tracing_cpu(i
) {
5102 global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
5103 max_tr
.data
[i
] = &per_cpu(max_tr_data
, i
);
5106 set_buffer_entries(&global_trace
,
5107 ring_buffer_size(global_trace
.buffer
, 0));
5108 #ifdef CONFIG_TRACER_MAX_TRACE
5109 set_buffer_entries(&max_tr
, 1);
5112 trace_init_cmdlines();
5114 register_tracer(&nop_trace
);
5115 current_trace
= &nop_trace
;
5116 /* All seems OK, enable tracing */
5117 tracing_disabled
= 0;
5119 atomic_notifier_chain_register(&panic_notifier_list
,
5120 &trace_panic_notifier
);
5122 register_die_notifier(&trace_die_notifier
);
5127 free_cpumask_var(tracing_cpumask
);
5128 out_free_buffer_mask
:
5129 free_cpumask_var(tracing_buffer_mask
);
5134 __init
static int clear_boot_tracer(void)
5137 * The default tracer at boot buffer is an init section.
5138 * This function is called in lateinit. If we did not
5139 * find the boot tracer, then clear it out, to prevent
5140 * later registration from accessing the buffer that is
5141 * about to be freed.
5143 if (!default_bootup_tracer
)
5146 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
5147 default_bootup_tracer
);
5148 default_bootup_tracer
= NULL
;
5153 early_initcall(tracer_alloc_buffers
);
5154 fs_initcall(tracer_init_debugfs
);
5155 late_initcall(clear_boot_tracer
);