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 Nadia Yvette Chambers
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/irq_work.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/rwsem.h>
36 #include <linux/slab.h>
37 #include <linux/ctype.h>
38 #include <linux/init.h>
39 #include <linux/poll.h>
40 #include <linux/nmi.h>
44 #include "trace_output.h"
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
50 int ring_buffer_expanded
;
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
59 static bool __read_mostly tracing_selftest_running
;
62 * If a tracer is running, we do not want to run SELFTEST.
64 bool __read_mostly tracing_selftest_disabled
;
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt
[] = {
71 static struct tracer_flags dummy_tracer_flags
= {
73 .opts
= dummy_tracer_opt
76 static int dummy_set_flag(u32 old_flags
, u32 bit
, int set
)
82 * To prevent the comm cache from being overwritten when no
83 * tracing is active, only save the comm when a trace event
86 static DEFINE_PER_CPU(bool, trace_cmdline_save
);
89 * When a reader is waiting for data, then this variable is
92 static bool trace_wakeup_needed
;
94 static struct irq_work trace_work_wakeup
;
97 * Kill all tracing for good (never come back).
98 * It is initialized to 1 but will turn to zero if the initialization
99 * of the tracer is successful. But that is the only place that sets
102 static int tracing_disabled
= 1;
104 DEFINE_PER_CPU(int, ftrace_cpu_disabled
);
106 cpumask_var_t __read_mostly tracing_buffer_mask
;
109 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
111 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
112 * is set, then ftrace_dump is called. This will output the contents
113 * of the ftrace buffers to the console. This is very useful for
114 * capturing traces that lead to crashes and outputing it to a
117 * It is default off, but you can enable it with either specifying
118 * "ftrace_dump_on_oops" in the kernel command line, or setting
119 * /proc/sys/kernel/ftrace_dump_on_oops
120 * Set 1 if you want to dump buffers of all CPUs
121 * Set 2 if you want to dump the buffer of the CPU that triggered oops
124 enum ftrace_dump_mode ftrace_dump_on_oops
;
126 static int tracing_set_tracer(const char *buf
);
128 #define MAX_TRACER_SIZE 100
129 static char bootup_tracer_buf
[MAX_TRACER_SIZE
] __initdata
;
130 static char *default_bootup_tracer
;
132 static int __init
set_cmdline_ftrace(char *str
)
134 strncpy(bootup_tracer_buf
, str
, MAX_TRACER_SIZE
);
135 default_bootup_tracer
= bootup_tracer_buf
;
136 /* We are using ftrace early, expand it */
137 ring_buffer_expanded
= 1;
140 __setup("ftrace=", set_cmdline_ftrace
);
142 static int __init
set_ftrace_dump_on_oops(char *str
)
144 if (*str
++ != '=' || !*str
) {
145 ftrace_dump_on_oops
= DUMP_ALL
;
149 if (!strcmp("orig_cpu", str
)) {
150 ftrace_dump_on_oops
= DUMP_ORIG
;
156 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops
);
159 static char trace_boot_options_buf
[MAX_TRACER_SIZE
] __initdata
;
160 static char *trace_boot_options __initdata
;
162 static int __init
set_trace_boot_options(char *str
)
164 strncpy(trace_boot_options_buf
, str
, MAX_TRACER_SIZE
);
165 trace_boot_options
= trace_boot_options_buf
;
168 __setup("trace_options=", set_trace_boot_options
);
170 unsigned long long ns2usecs(cycle_t nsec
)
178 * The global_trace is the descriptor that holds the tracing
179 * buffers for the live tracing. For each CPU, it contains
180 * a link list of pages that will store trace entries. The
181 * page descriptor of the pages in the memory is used to hold
182 * the link list by linking the lru item in the page descriptor
183 * to each of the pages in the buffer per CPU.
185 * For each active CPU there is a data field that holds the
186 * pages for the buffer for that CPU. Each CPU has the same number
187 * of pages allocated for its buffer.
189 static struct trace_array global_trace
;
191 static DEFINE_PER_CPU(struct trace_array_cpu
, global_trace_cpu
);
193 int filter_current_check_discard(struct ring_buffer
*buffer
,
194 struct ftrace_event_call
*call
, void *rec
,
195 struct ring_buffer_event
*event
)
197 return filter_check_discard(call
, rec
, buffer
, event
);
199 EXPORT_SYMBOL_GPL(filter_current_check_discard
);
201 cycle_t
ftrace_now(int cpu
)
205 /* Early boot up does not have a buffer yet */
206 if (!global_trace
.buffer
)
207 return trace_clock_local();
209 ts
= ring_buffer_time_stamp(global_trace
.buffer
, cpu
);
210 ring_buffer_normalize_time_stamp(global_trace
.buffer
, cpu
, &ts
);
216 * The max_tr is used to snapshot the global_trace when a maximum
217 * latency is reached. Some tracers will use this to store a maximum
218 * trace while it continues examining live traces.
220 * The buffers for the max_tr are set up the same as the global_trace.
221 * When a snapshot is taken, the link list of the max_tr is swapped
222 * with the link list of the global_trace and the buffers are reset for
223 * the global_trace so the tracing can continue.
225 static struct trace_array max_tr
;
227 static DEFINE_PER_CPU(struct trace_array_cpu
, max_tr_data
);
229 int tracing_is_enabled(void)
231 return tracing_is_on();
235 * trace_buf_size is the size in bytes that is allocated
236 * for a buffer. Note, the number of bytes is always rounded
239 * This number is purposely set to a low number of 16384.
240 * If the dump on oops happens, it will be much appreciated
241 * to not have to wait for all that output. Anyway this can be
242 * boot time and run time configurable.
244 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
246 static unsigned long trace_buf_size
= TRACE_BUF_SIZE_DEFAULT
;
248 /* trace_types holds a link list of available tracers. */
249 static struct tracer
*trace_types __read_mostly
;
251 /* current_trace points to the tracer that is currently active */
252 static struct tracer
*current_trace __read_mostly
;
255 * trace_types_lock is used to protect the trace_types list.
257 static DEFINE_MUTEX(trace_types_lock
);
260 * serialize the access of the ring buffer
262 * ring buffer serializes readers, but it is low level protection.
263 * The validity of the events (which returns by ring_buffer_peek() ..etc)
264 * are not protected by ring buffer.
266 * The content of events may become garbage if we allow other process consumes
267 * these events concurrently:
268 * A) the page of the consumed events may become a normal page
269 * (not reader page) in ring buffer, and this page will be rewrited
270 * by events producer.
271 * B) The page of the consumed events may become a page for splice_read,
272 * and this page will be returned to system.
274 * These primitives allow multi process access to different cpu ring buffer
277 * These primitives don't distinguish read-only and read-consume access.
278 * Multi read-only access are also serialized.
282 static DECLARE_RWSEM(all_cpu_access_lock
);
283 static DEFINE_PER_CPU(struct mutex
, cpu_access_lock
);
285 static inline void trace_access_lock(int cpu
)
287 if (cpu
== TRACE_PIPE_ALL_CPU
) {
288 /* gain it for accessing the whole ring buffer. */
289 down_write(&all_cpu_access_lock
);
291 /* gain it for accessing a cpu ring buffer. */
293 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
294 down_read(&all_cpu_access_lock
);
296 /* Secondly block other access to this @cpu ring buffer. */
297 mutex_lock(&per_cpu(cpu_access_lock
, cpu
));
301 static inline void trace_access_unlock(int cpu
)
303 if (cpu
== TRACE_PIPE_ALL_CPU
) {
304 up_write(&all_cpu_access_lock
);
306 mutex_unlock(&per_cpu(cpu_access_lock
, cpu
));
307 up_read(&all_cpu_access_lock
);
311 static inline void trace_access_lock_init(void)
315 for_each_possible_cpu(cpu
)
316 mutex_init(&per_cpu(cpu_access_lock
, cpu
));
321 static DEFINE_MUTEX(access_lock
);
323 static inline void trace_access_lock(int cpu
)
326 mutex_lock(&access_lock
);
329 static inline void trace_access_unlock(int cpu
)
332 mutex_unlock(&access_lock
);
335 static inline void trace_access_lock_init(void)
341 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
342 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
344 /* trace_flags holds trace_options default values */
345 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
| TRACE_ITER_PRINTK
|
346 TRACE_ITER_ANNOTATE
| TRACE_ITER_CONTEXT_INFO
| TRACE_ITER_SLEEP_TIME
|
347 TRACE_ITER_GRAPH_TIME
| TRACE_ITER_RECORD_CMD
| TRACE_ITER_OVERWRITE
|
348 TRACE_ITER_IRQ_INFO
| TRACE_ITER_MARKERS
;
350 static int trace_stop_count
;
351 static DEFINE_RAW_SPINLOCK(tracing_start_lock
);
354 * trace_wake_up - wake up tasks waiting for trace input
356 * Schedules a delayed work to wake up any task that is blocked on the
357 * trace_wait queue. These is used with trace_poll for tasks polling the
360 static void trace_wake_up(struct irq_work
*work
)
362 wake_up_all(&trace_wait
);
367 * tracing_on - enable tracing buffers
369 * This function enables tracing buffers that may have been
370 * disabled with tracing_off.
372 void tracing_on(void)
374 if (global_trace
.buffer
)
375 ring_buffer_record_on(global_trace
.buffer
);
377 * This flag is only looked at when buffers haven't been
378 * allocated yet. We don't really care about the race
379 * between setting this flag and actually turning
382 global_trace
.buffer_disabled
= 0;
384 EXPORT_SYMBOL_GPL(tracing_on
);
387 * tracing_off - turn off tracing buffers
389 * This function stops the tracing buffers from recording data.
390 * It does not disable any overhead the tracers themselves may
391 * be causing. This function simply causes all recording to
392 * the ring buffers to fail.
394 void tracing_off(void)
396 if (global_trace
.buffer
)
397 ring_buffer_record_off(global_trace
.buffer
);
399 * This flag is only looked at when buffers haven't been
400 * allocated yet. We don't really care about the race
401 * between setting this flag and actually turning
404 global_trace
.buffer_disabled
= 1;
406 EXPORT_SYMBOL_GPL(tracing_off
);
409 * tracing_is_on - show state of ring buffers enabled
411 int tracing_is_on(void)
413 if (global_trace
.buffer
)
414 return ring_buffer_record_is_on(global_trace
.buffer
);
415 return !global_trace
.buffer_disabled
;
417 EXPORT_SYMBOL_GPL(tracing_is_on
);
419 static int __init
set_buf_size(char *str
)
421 unsigned long buf_size
;
425 buf_size
= memparse(str
, &str
);
426 /* nr_entries can not be zero */
429 trace_buf_size
= buf_size
;
432 __setup("trace_buf_size=", set_buf_size
);
434 static int __init
set_tracing_thresh(char *str
)
436 unsigned long threshold
;
441 ret
= kstrtoul(str
, 0, &threshold
);
444 tracing_thresh
= threshold
* 1000;
447 __setup("tracing_thresh=", set_tracing_thresh
);
449 unsigned long nsecs_to_usecs(unsigned long nsecs
)
454 /* These must match the bit postions in trace_iterator_flags */
455 static const char *trace_options
[] = {
487 int in_ns
; /* is this clock in nanoseconds? */
489 { trace_clock_local
, "local", 1 },
490 { trace_clock_global
, "global", 1 },
491 { trace_clock_counter
, "counter", 0 },
498 * trace_parser_get_init - gets the buffer for trace parser
500 int trace_parser_get_init(struct trace_parser
*parser
, int size
)
502 memset(parser
, 0, sizeof(*parser
));
504 parser
->buffer
= kmalloc(size
, GFP_KERNEL
);
513 * trace_parser_put - frees the buffer for trace parser
515 void trace_parser_put(struct trace_parser
*parser
)
517 kfree(parser
->buffer
);
521 * trace_get_user - reads the user input string separated by space
522 * (matched by isspace(ch))
524 * For each string found the 'struct trace_parser' is updated,
525 * and the function returns.
527 * Returns number of bytes read.
529 * See kernel/trace/trace.h for 'struct trace_parser' details.
531 int trace_get_user(struct trace_parser
*parser
, const char __user
*ubuf
,
532 size_t cnt
, loff_t
*ppos
)
539 trace_parser_clear(parser
);
541 ret
= get_user(ch
, ubuf
++);
549 * The parser is not finished with the last write,
550 * continue reading the user input without skipping spaces.
553 /* skip white space */
554 while (cnt
&& isspace(ch
)) {
555 ret
= get_user(ch
, ubuf
++);
562 /* only spaces were written */
572 /* read the non-space input */
573 while (cnt
&& !isspace(ch
)) {
574 if (parser
->idx
< parser
->size
- 1)
575 parser
->buffer
[parser
->idx
++] = ch
;
580 ret
= get_user(ch
, ubuf
++);
587 /* We either got finished input or we have to wait for another call. */
589 parser
->buffer
[parser
->idx
] = 0;
590 parser
->cont
= false;
593 parser
->buffer
[parser
->idx
++] = ch
;
603 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
611 if (s
->len
<= s
->readpos
)
614 len
= s
->len
- s
->readpos
;
617 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
627 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
631 if (s
->len
<= s
->readpos
)
634 len
= s
->len
- s
->readpos
;
637 memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
644 * ftrace_max_lock is used to protect the swapping of buffers
645 * when taking a max snapshot. The buffers themselves are
646 * protected by per_cpu spinlocks. But the action of the swap
647 * needs its own lock.
649 * This is defined as a arch_spinlock_t in order to help
650 * with performance when lockdep debugging is enabled.
652 * It is also used in other places outside the update_max_tr
653 * so it needs to be defined outside of the
654 * CONFIG_TRACER_MAX_TRACE.
656 static arch_spinlock_t ftrace_max_lock
=
657 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
659 unsigned long __read_mostly tracing_thresh
;
661 #ifdef CONFIG_TRACER_MAX_TRACE
662 unsigned long __read_mostly tracing_max_latency
;
665 * Copy the new maximum trace into the separate maximum-trace
666 * structure. (this way the maximum trace is permanently saved,
667 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
670 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
672 struct trace_array_cpu
*data
= tr
->data
[cpu
];
673 struct trace_array_cpu
*max_data
;
676 max_tr
.time_start
= data
->preempt_timestamp
;
678 max_data
= max_tr
.data
[cpu
];
679 max_data
->saved_latency
= tracing_max_latency
;
680 max_data
->critical_start
= data
->critical_start
;
681 max_data
->critical_end
= data
->critical_end
;
683 memcpy(max_data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
684 max_data
->pid
= tsk
->pid
;
685 max_data
->uid
= task_uid(tsk
);
686 max_data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
687 max_data
->policy
= tsk
->policy
;
688 max_data
->rt_priority
= tsk
->rt_priority
;
690 /* record this tasks comm */
691 tracing_record_cmdline(tsk
);
695 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
697 * @tsk: the task with the latency
698 * @cpu: The cpu that initiated the trace.
700 * Flip the buffers between the @tr and the max_tr and record information
701 * about which task was the cause of this latency.
704 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
706 struct ring_buffer
*buf
= tr
->buffer
;
708 if (trace_stop_count
)
711 WARN_ON_ONCE(!irqs_disabled());
713 if (!current_trace
->allocated_snapshot
) {
714 /* Only the nop tracer should hit this when disabling */
715 WARN_ON_ONCE(current_trace
!= &nop_trace
);
719 arch_spin_lock(&ftrace_max_lock
);
721 tr
->buffer
= max_tr
.buffer
;
724 __update_max_tr(tr
, tsk
, cpu
);
725 arch_spin_unlock(&ftrace_max_lock
);
729 * update_max_tr_single - only copy one trace over, and reset the rest
731 * @tsk - task with the latency
732 * @cpu - the cpu of the buffer to copy.
734 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
737 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
741 if (trace_stop_count
)
744 WARN_ON_ONCE(!irqs_disabled());
745 if (WARN_ON_ONCE(!current_trace
->allocated_snapshot
))
748 arch_spin_lock(&ftrace_max_lock
);
750 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
754 * We failed to swap the buffer due to a commit taking
755 * place on this CPU. We fail to record, but we reset
756 * the max trace buffer (no one writes directly to it)
757 * and flag that it failed.
759 trace_array_printk(&max_tr
, _THIS_IP_
,
760 "Failed to swap buffers due to commit in progress\n");
763 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
&& ret
!= -EBUSY
);
765 __update_max_tr(tr
, tsk
, cpu
);
766 arch_spin_unlock(&ftrace_max_lock
);
768 #endif /* CONFIG_TRACER_MAX_TRACE */
770 static void default_wait_pipe(struct trace_iterator
*iter
)
774 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
777 * The events can happen in critical sections where
778 * checking a work queue can cause deadlocks.
779 * After adding a task to the queue, this flag is set
780 * only to notify events to try to wake up the queue
783 * We don't clear it even if the buffer is no longer
784 * empty. The flag only causes the next event to run
785 * irq_work to do the work queue wake up. The worse
786 * that can happen if we race with !trace_empty() is that
787 * an event will cause an irq_work to try to wake up
790 * There's no reason to protect this flag either, as
791 * the work queue and irq_work logic will do the necessary
792 * synchronization for the wake ups. The only thing
793 * that is necessary is that the wake up happens after
794 * a task has been queued. It's OK for spurious wake ups.
796 trace_wakeup_needed
= true;
798 if (trace_empty(iter
))
801 finish_wait(&trace_wait
, &wait
);
805 * register_tracer - register a tracer with the ftrace system.
806 * @type - the plugin for the tracer
808 * Register a new plugin tracer.
810 int register_tracer(struct tracer
*type
)
816 pr_info("Tracer must have a name\n");
820 if (strlen(type
->name
) >= MAX_TRACER_SIZE
) {
821 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE
);
825 mutex_lock(&trace_types_lock
);
827 tracing_selftest_running
= true;
829 for (t
= trace_types
; t
; t
= t
->next
) {
830 if (strcmp(type
->name
, t
->name
) == 0) {
832 pr_info("Tracer %s already registered\n",
840 type
->set_flag
= &dummy_set_flag
;
842 type
->flags
= &dummy_tracer_flags
;
844 if (!type
->flags
->opts
)
845 type
->flags
->opts
= dummy_tracer_opt
;
846 if (!type
->wait_pipe
)
847 type
->wait_pipe
= default_wait_pipe
;
850 #ifdef CONFIG_FTRACE_STARTUP_TEST
851 if (type
->selftest
&& !tracing_selftest_disabled
) {
852 struct tracer
*saved_tracer
= current_trace
;
853 struct trace_array
*tr
= &global_trace
;
856 * Run a selftest on this tracer.
857 * Here we reset the trace buffer, and set the current
858 * tracer to be this tracer. The tracer can then run some
859 * internal tracing to verify that everything is in order.
860 * If we fail, we do not register this tracer.
862 tracing_reset_online_cpus(tr
);
864 current_trace
= type
;
866 if (type
->use_max_tr
) {
867 /* If we expanded the buffers, make sure the max is expanded too */
868 if (ring_buffer_expanded
)
869 ring_buffer_resize(max_tr
.buffer
, trace_buf_size
,
870 RING_BUFFER_ALL_CPUS
);
871 type
->allocated_snapshot
= true;
874 /* the test is responsible for initializing and enabling */
875 pr_info("Testing tracer %s: ", type
->name
);
876 ret
= type
->selftest(type
, tr
);
877 /* the test is responsible for resetting too */
878 current_trace
= saved_tracer
;
880 printk(KERN_CONT
"FAILED!\n");
881 /* Add the warning after printing 'FAILED' */
885 /* Only reset on passing, to avoid touching corrupted buffers */
886 tracing_reset_online_cpus(tr
);
888 if (type
->use_max_tr
) {
889 type
->allocated_snapshot
= false;
891 /* Shrink the max buffer again */
892 if (ring_buffer_expanded
)
893 ring_buffer_resize(max_tr
.buffer
, 1,
894 RING_BUFFER_ALL_CPUS
);
897 printk(KERN_CONT
"PASSED\n");
901 type
->next
= trace_types
;
905 tracing_selftest_running
= false;
906 mutex_unlock(&trace_types_lock
);
908 if (ret
|| !default_bootup_tracer
)
911 if (strncmp(default_bootup_tracer
, type
->name
, MAX_TRACER_SIZE
))
914 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
915 /* Do we want this tracer to start on bootup? */
916 tracing_set_tracer(type
->name
);
917 default_bootup_tracer
= NULL
;
918 /* disable other selftests, since this will break it. */
919 tracing_selftest_disabled
= 1;
920 #ifdef CONFIG_FTRACE_STARTUP_TEST
921 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
929 void tracing_reset(struct trace_array
*tr
, int cpu
)
931 struct ring_buffer
*buffer
= tr
->buffer
;
936 ring_buffer_record_disable(buffer
);
938 /* Make sure all commits have finished */
940 ring_buffer_reset_cpu(buffer
, cpu
);
942 ring_buffer_record_enable(buffer
);
945 void tracing_reset_online_cpus(struct trace_array
*tr
)
947 struct ring_buffer
*buffer
= tr
->buffer
;
953 ring_buffer_record_disable(buffer
);
955 /* Make sure all commits have finished */
958 tr
->time_start
= ftrace_now(tr
->cpu
);
960 for_each_online_cpu(cpu
)
961 ring_buffer_reset_cpu(buffer
, cpu
);
963 ring_buffer_record_enable(buffer
);
966 void tracing_reset_current(int cpu
)
968 tracing_reset(&global_trace
, cpu
);
971 void tracing_reset_current_online_cpus(void)
973 tracing_reset_online_cpus(&global_trace
);
976 #define SAVED_CMDLINES 128
977 #define NO_CMDLINE_MAP UINT_MAX
978 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
979 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
980 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
981 static int cmdline_idx
;
982 static arch_spinlock_t trace_cmdline_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
984 /* temporary disable recording */
985 static atomic_t trace_record_cmdline_disabled __read_mostly
;
987 static void trace_init_cmdlines(void)
989 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
990 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
994 int is_tracing_stopped(void)
996 return trace_stop_count
;
1000 * ftrace_off_permanent - disable all ftrace code permanently
1002 * This should only be called when a serious anomally has
1003 * been detected. This will turn off the function tracing,
1004 * ring buffers, and other tracing utilites. It takes no
1005 * locks and can be called from any context.
1007 void ftrace_off_permanent(void)
1009 tracing_disabled
= 1;
1011 tracing_off_permanent();
1015 * tracing_start - quick start of the tracer
1017 * If tracing is enabled but was stopped by tracing_stop,
1018 * this will start the tracer back up.
1020 void tracing_start(void)
1022 struct ring_buffer
*buffer
;
1023 unsigned long flags
;
1025 if (tracing_disabled
)
1028 raw_spin_lock_irqsave(&tracing_start_lock
, flags
);
1029 if (--trace_stop_count
) {
1030 if (trace_stop_count
< 0) {
1031 /* Someone screwed up their debugging */
1033 trace_stop_count
= 0;
1038 /* Prevent the buffers from switching */
1039 arch_spin_lock(&ftrace_max_lock
);
1041 buffer
= global_trace
.buffer
;
1043 ring_buffer_record_enable(buffer
);
1045 buffer
= max_tr
.buffer
;
1047 ring_buffer_record_enable(buffer
);
1049 arch_spin_unlock(&ftrace_max_lock
);
1053 raw_spin_unlock_irqrestore(&tracing_start_lock
, flags
);
1057 * tracing_stop - quick stop of the tracer
1059 * Light weight way to stop tracing. Use in conjunction with
1062 void tracing_stop(void)
1064 struct ring_buffer
*buffer
;
1065 unsigned long flags
;
1068 raw_spin_lock_irqsave(&tracing_start_lock
, flags
);
1069 if (trace_stop_count
++)
1072 /* Prevent the buffers from switching */
1073 arch_spin_lock(&ftrace_max_lock
);
1075 buffer
= global_trace
.buffer
;
1077 ring_buffer_record_disable(buffer
);
1079 buffer
= max_tr
.buffer
;
1081 ring_buffer_record_disable(buffer
);
1083 arch_spin_unlock(&ftrace_max_lock
);
1086 raw_spin_unlock_irqrestore(&tracing_start_lock
, flags
);
1089 void trace_stop_cmdline_recording(void);
1091 static void trace_save_cmdline(struct task_struct
*tsk
)
1095 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
1099 * It's not the end of the world if we don't get
1100 * the lock, but we also don't want to spin
1101 * nor do we want to disable interrupts,
1102 * so if we miss here, then better luck next time.
1104 if (!arch_spin_trylock(&trace_cmdline_lock
))
1107 idx
= map_pid_to_cmdline
[tsk
->pid
];
1108 if (idx
== NO_CMDLINE_MAP
) {
1109 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
1112 * Check whether the cmdline buffer at idx has a pid
1113 * mapped. We are going to overwrite that entry so we
1114 * need to clear the map_pid_to_cmdline. Otherwise we
1115 * would read the new comm for the old pid.
1117 pid
= map_cmdline_to_pid
[idx
];
1118 if (pid
!= NO_CMDLINE_MAP
)
1119 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
1121 map_cmdline_to_pid
[idx
] = tsk
->pid
;
1122 map_pid_to_cmdline
[tsk
->pid
] = idx
;
1127 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
1129 arch_spin_unlock(&trace_cmdline_lock
);
1132 void trace_find_cmdline(int pid
, char comm
[])
1137 strcpy(comm
, "<idle>");
1141 if (WARN_ON_ONCE(pid
< 0)) {
1142 strcpy(comm
, "<XXX>");
1146 if (pid
> PID_MAX_DEFAULT
) {
1147 strcpy(comm
, "<...>");
1152 arch_spin_lock(&trace_cmdline_lock
);
1153 map
= map_pid_to_cmdline
[pid
];
1154 if (map
!= NO_CMDLINE_MAP
)
1155 strcpy(comm
, saved_cmdlines
[map
]);
1157 strcpy(comm
, "<...>");
1159 arch_spin_unlock(&trace_cmdline_lock
);
1163 void tracing_record_cmdline(struct task_struct
*tsk
)
1165 if (atomic_read(&trace_record_cmdline_disabled
) || !tracing_is_on())
1168 if (!__this_cpu_read(trace_cmdline_save
))
1171 __this_cpu_write(trace_cmdline_save
, false);
1173 trace_save_cmdline(tsk
);
1177 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
1180 struct task_struct
*tsk
= current
;
1182 entry
->preempt_count
= pc
& 0xff;
1183 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
1185 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1186 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
1188 TRACE_FLAG_IRQS_NOSUPPORT
|
1190 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
1191 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
1192 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
1194 EXPORT_SYMBOL_GPL(tracing_generic_entry_update
);
1196 struct ring_buffer_event
*
1197 trace_buffer_lock_reserve(struct ring_buffer
*buffer
,
1200 unsigned long flags
, int pc
)
1202 struct ring_buffer_event
*event
;
1204 event
= ring_buffer_lock_reserve(buffer
, len
);
1205 if (event
!= NULL
) {
1206 struct trace_entry
*ent
= ring_buffer_event_data(event
);
1208 tracing_generic_entry_update(ent
, flags
, pc
);
1216 __buffer_unlock_commit(struct ring_buffer
*buffer
, struct ring_buffer_event
*event
)
1218 __this_cpu_write(trace_cmdline_save
, true);
1219 if (trace_wakeup_needed
) {
1220 trace_wakeup_needed
= false;
1221 /* irq_work_queue() supplies it's own memory barriers */
1222 irq_work_queue(&trace_work_wakeup
);
1224 ring_buffer_unlock_commit(buffer
, event
);
1228 __trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1229 struct ring_buffer_event
*event
,
1230 unsigned long flags
, int pc
)
1232 __buffer_unlock_commit(buffer
, event
);
1234 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1235 ftrace_trace_userstack(buffer
, flags
, pc
);
1238 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1239 struct ring_buffer_event
*event
,
1240 unsigned long flags
, int pc
)
1242 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
);
1244 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit
);
1246 struct ring_buffer_event
*
1247 trace_current_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1248 int type
, unsigned long len
,
1249 unsigned long flags
, int pc
)
1251 *current_rb
= global_trace
.buffer
;
1252 return trace_buffer_lock_reserve(*current_rb
,
1253 type
, len
, flags
, pc
);
1255 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve
);
1257 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
1258 struct ring_buffer_event
*event
,
1259 unsigned long flags
, int pc
)
1261 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
);
1263 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit
);
1265 void trace_buffer_unlock_commit_regs(struct ring_buffer
*buffer
,
1266 struct ring_buffer_event
*event
,
1267 unsigned long flags
, int pc
,
1268 struct pt_regs
*regs
)
1270 __buffer_unlock_commit(buffer
, event
);
1272 ftrace_trace_stack_regs(buffer
, flags
, 0, pc
, regs
);
1273 ftrace_trace_userstack(buffer
, flags
, pc
);
1275 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs
);
1277 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
1278 struct ring_buffer_event
*event
)
1280 ring_buffer_discard_commit(buffer
, event
);
1282 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit
);
1285 trace_function(struct trace_array
*tr
,
1286 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1289 struct ftrace_event_call
*call
= &event_function
;
1290 struct ring_buffer
*buffer
= tr
->buffer
;
1291 struct ring_buffer_event
*event
;
1292 struct ftrace_entry
*entry
;
1294 /* If we are reading the ring buffer, don't trace */
1295 if (unlikely(__this_cpu_read(ftrace_cpu_disabled
)))
1298 event
= trace_buffer_lock_reserve(buffer
, TRACE_FN
, sizeof(*entry
),
1302 entry
= ring_buffer_event_data(event
);
1304 entry
->parent_ip
= parent_ip
;
1306 if (!filter_check_discard(call
, entry
, buffer
, event
))
1307 __buffer_unlock_commit(buffer
, event
);
1311 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1312 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1315 if (likely(!atomic_read(&data
->disabled
)))
1316 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
1319 #ifdef CONFIG_STACKTRACE
1321 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1322 struct ftrace_stack
{
1323 unsigned long calls
[FTRACE_STACK_MAX_ENTRIES
];
1326 static DEFINE_PER_CPU(struct ftrace_stack
, ftrace_stack
);
1327 static DEFINE_PER_CPU(int, ftrace_stack_reserve
);
1329 static void __ftrace_trace_stack(struct ring_buffer
*buffer
,
1330 unsigned long flags
,
1331 int skip
, int pc
, struct pt_regs
*regs
)
1333 struct ftrace_event_call
*call
= &event_kernel_stack
;
1334 struct ring_buffer_event
*event
;
1335 struct stack_entry
*entry
;
1336 struct stack_trace trace
;
1338 int size
= FTRACE_STACK_ENTRIES
;
1340 trace
.nr_entries
= 0;
1344 * Since events can happen in NMIs there's no safe way to
1345 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1346 * or NMI comes in, it will just have to use the default
1347 * FTRACE_STACK_SIZE.
1349 preempt_disable_notrace();
1351 use_stack
= __this_cpu_inc_return(ftrace_stack_reserve
);
1353 * We don't need any atomic variables, just a barrier.
1354 * If an interrupt comes in, we don't care, because it would
1355 * have exited and put the counter back to what we want.
1356 * We just need a barrier to keep gcc from moving things
1360 if (use_stack
== 1) {
1361 trace
.entries
= &__get_cpu_var(ftrace_stack
).calls
[0];
1362 trace
.max_entries
= FTRACE_STACK_MAX_ENTRIES
;
1365 save_stack_trace_regs(regs
, &trace
);
1367 save_stack_trace(&trace
);
1369 if (trace
.nr_entries
> size
)
1370 size
= trace
.nr_entries
;
1372 /* From now on, use_stack is a boolean */
1375 size
*= sizeof(unsigned long);
1377 event
= trace_buffer_lock_reserve(buffer
, TRACE_STACK
,
1378 sizeof(*entry
) + size
, flags
, pc
);
1381 entry
= ring_buffer_event_data(event
);
1383 memset(&entry
->caller
, 0, size
);
1386 memcpy(&entry
->caller
, trace
.entries
,
1387 trace
.nr_entries
* sizeof(unsigned long));
1389 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1390 trace
.entries
= entry
->caller
;
1392 save_stack_trace_regs(regs
, &trace
);
1394 save_stack_trace(&trace
);
1397 entry
->size
= trace
.nr_entries
;
1399 if (!filter_check_discard(call
, entry
, buffer
, event
))
1400 __buffer_unlock_commit(buffer
, event
);
1403 /* Again, don't let gcc optimize things here */
1405 __this_cpu_dec(ftrace_stack_reserve
);
1406 preempt_enable_notrace();
1410 void ftrace_trace_stack_regs(struct ring_buffer
*buffer
, unsigned long flags
,
1411 int skip
, int pc
, struct pt_regs
*regs
)
1413 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1416 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, regs
);
1419 void ftrace_trace_stack(struct ring_buffer
*buffer
, unsigned long flags
,
1422 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1425 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, NULL
);
1428 void __trace_stack(struct trace_array
*tr
, unsigned long flags
, int skip
,
1431 __ftrace_trace_stack(tr
->buffer
, flags
, skip
, pc
, NULL
);
1435 * trace_dump_stack - record a stack back trace in the trace buffer
1437 void trace_dump_stack(void)
1439 unsigned long flags
;
1441 if (tracing_disabled
|| tracing_selftest_running
)
1444 local_save_flags(flags
);
1446 /* skipping 3 traces, seems to get us at the caller of this function */
1447 __ftrace_trace_stack(global_trace
.buffer
, flags
, 3, preempt_count(), NULL
);
1450 static DEFINE_PER_CPU(int, user_stack_count
);
1453 ftrace_trace_userstack(struct ring_buffer
*buffer
, unsigned long flags
, int pc
)
1455 struct ftrace_event_call
*call
= &event_user_stack
;
1456 struct ring_buffer_event
*event
;
1457 struct userstack_entry
*entry
;
1458 struct stack_trace trace
;
1460 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1464 * NMIs can not handle page faults, even with fix ups.
1465 * The save user stack can (and often does) fault.
1467 if (unlikely(in_nmi()))
1471 * prevent recursion, since the user stack tracing may
1472 * trigger other kernel events.
1475 if (__this_cpu_read(user_stack_count
))
1478 __this_cpu_inc(user_stack_count
);
1480 event
= trace_buffer_lock_reserve(buffer
, TRACE_USER_STACK
,
1481 sizeof(*entry
), flags
, pc
);
1483 goto out_drop_count
;
1484 entry
= ring_buffer_event_data(event
);
1486 entry
->tgid
= current
->tgid
;
1487 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1489 trace
.nr_entries
= 0;
1490 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1492 trace
.entries
= entry
->caller
;
1494 save_stack_trace_user(&trace
);
1495 if (!filter_check_discard(call
, entry
, buffer
, event
))
1496 __buffer_unlock_commit(buffer
, event
);
1499 __this_cpu_dec(user_stack_count
);
1505 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1507 ftrace_trace_userstack(tr
, flags
, preempt_count());
1511 #endif /* CONFIG_STACKTRACE */
1513 /* created for use with alloc_percpu */
1514 struct trace_buffer_struct
{
1515 char buffer
[TRACE_BUF_SIZE
];
1518 static struct trace_buffer_struct
*trace_percpu_buffer
;
1519 static struct trace_buffer_struct
*trace_percpu_sirq_buffer
;
1520 static struct trace_buffer_struct
*trace_percpu_irq_buffer
;
1521 static struct trace_buffer_struct
*trace_percpu_nmi_buffer
;
1524 * The buffer used is dependent on the context. There is a per cpu
1525 * buffer for normal context, softirq contex, hard irq context and
1526 * for NMI context. Thise allows for lockless recording.
1528 * Note, if the buffers failed to be allocated, then this returns NULL
1530 static char *get_trace_buf(void)
1532 struct trace_buffer_struct
*percpu_buffer
;
1535 * If we have allocated per cpu buffers, then we do not
1536 * need to do any locking.
1539 percpu_buffer
= trace_percpu_nmi_buffer
;
1541 percpu_buffer
= trace_percpu_irq_buffer
;
1542 else if (in_softirq())
1543 percpu_buffer
= trace_percpu_sirq_buffer
;
1545 percpu_buffer
= trace_percpu_buffer
;
1550 return this_cpu_ptr(&percpu_buffer
->buffer
[0]);
1553 static int alloc_percpu_trace_buffer(void)
1555 struct trace_buffer_struct
*buffers
;
1556 struct trace_buffer_struct
*sirq_buffers
;
1557 struct trace_buffer_struct
*irq_buffers
;
1558 struct trace_buffer_struct
*nmi_buffers
;
1560 buffers
= alloc_percpu(struct trace_buffer_struct
);
1564 sirq_buffers
= alloc_percpu(struct trace_buffer_struct
);
1568 irq_buffers
= alloc_percpu(struct trace_buffer_struct
);
1572 nmi_buffers
= alloc_percpu(struct trace_buffer_struct
);
1576 trace_percpu_buffer
= buffers
;
1577 trace_percpu_sirq_buffer
= sirq_buffers
;
1578 trace_percpu_irq_buffer
= irq_buffers
;
1579 trace_percpu_nmi_buffer
= nmi_buffers
;
1584 free_percpu(irq_buffers
);
1586 free_percpu(sirq_buffers
);
1588 free_percpu(buffers
);
1590 WARN(1, "Could not allocate percpu trace_printk buffer");
1594 static int buffers_allocated
;
1596 void trace_printk_init_buffers(void)
1598 if (buffers_allocated
)
1601 if (alloc_percpu_trace_buffer())
1604 pr_info("ftrace: Allocated trace_printk buffers\n");
1606 /* Expand the buffers to set size */
1607 tracing_update_buffers();
1609 buffers_allocated
= 1;
1612 * trace_printk_init_buffers() can be called by modules.
1613 * If that happens, then we need to start cmdline recording
1614 * directly here. If the global_trace.buffer is already
1615 * allocated here, then this was called by module code.
1617 if (global_trace
.buffer
)
1618 tracing_start_cmdline_record();
1621 void trace_printk_start_comm(void)
1623 /* Start tracing comms if trace printk is set */
1624 if (!buffers_allocated
)
1626 tracing_start_cmdline_record();
1629 static void trace_printk_start_stop_comm(int enabled
)
1631 if (!buffers_allocated
)
1635 tracing_start_cmdline_record();
1637 tracing_stop_cmdline_record();
1641 * trace_vbprintk - write binary msg to tracing buffer
1644 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1646 struct ftrace_event_call
*call
= &event_bprint
;
1647 struct ring_buffer_event
*event
;
1648 struct ring_buffer
*buffer
;
1649 struct trace_array
*tr
= &global_trace
;
1650 struct bprint_entry
*entry
;
1651 unsigned long flags
;
1653 int len
= 0, size
, pc
;
1655 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1658 /* Don't pollute graph traces with trace_vprintk internals */
1659 pause_graph_tracing();
1661 pc
= preempt_count();
1662 preempt_disable_notrace();
1664 tbuffer
= get_trace_buf();
1670 len
= vbin_printf((u32
*)tbuffer
, TRACE_BUF_SIZE
/sizeof(int), fmt
, args
);
1672 if (len
> TRACE_BUF_SIZE
/sizeof(int) || len
< 0)
1675 local_save_flags(flags
);
1676 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1677 buffer
= tr
->buffer
;
1678 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
1682 entry
= ring_buffer_event_data(event
);
1686 memcpy(entry
->buf
, tbuffer
, sizeof(u32
) * len
);
1687 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1688 __buffer_unlock_commit(buffer
, event
);
1689 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1693 preempt_enable_notrace();
1694 unpause_graph_tracing();
1698 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1700 int trace_array_printk(struct trace_array
*tr
,
1701 unsigned long ip
, const char *fmt
, ...)
1706 if (!(trace_flags
& TRACE_ITER_PRINTK
))
1710 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
1715 int trace_array_vprintk(struct trace_array
*tr
,
1716 unsigned long ip
, const char *fmt
, va_list args
)
1718 struct ftrace_event_call
*call
= &event_print
;
1719 struct ring_buffer_event
*event
;
1720 struct ring_buffer
*buffer
;
1721 int len
= 0, size
, pc
;
1722 struct print_entry
*entry
;
1723 unsigned long flags
;
1726 if (tracing_disabled
|| tracing_selftest_running
)
1729 /* Don't pollute graph traces with trace_vprintk internals */
1730 pause_graph_tracing();
1732 pc
= preempt_count();
1733 preempt_disable_notrace();
1736 tbuffer
= get_trace_buf();
1742 len
= vsnprintf(tbuffer
, TRACE_BUF_SIZE
, fmt
, args
);
1743 if (len
> TRACE_BUF_SIZE
)
1746 local_save_flags(flags
);
1747 size
= sizeof(*entry
) + len
+ 1;
1748 buffer
= tr
->buffer
;
1749 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
1753 entry
= ring_buffer_event_data(event
);
1756 memcpy(&entry
->buf
, tbuffer
, len
);
1757 entry
->buf
[len
] = '\0';
1758 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1759 __buffer_unlock_commit(buffer
, event
);
1760 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1763 preempt_enable_notrace();
1764 unpause_graph_tracing();
1769 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1771 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
1773 EXPORT_SYMBOL_GPL(trace_vprintk
);
1775 static void trace_iterator_increment(struct trace_iterator
*iter
)
1777 struct ring_buffer_iter
*buf_iter
= trace_buffer_iter(iter
, iter
->cpu
);
1781 ring_buffer_read(buf_iter
, NULL
);
1784 static struct trace_entry
*
1785 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
,
1786 unsigned long *lost_events
)
1788 struct ring_buffer_event
*event
;
1789 struct ring_buffer_iter
*buf_iter
= trace_buffer_iter(iter
, cpu
);
1792 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1794 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
,
1798 iter
->ent_size
= ring_buffer_event_length(event
);
1799 return ring_buffer_event_data(event
);
1805 static struct trace_entry
*
1806 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
,
1807 unsigned long *missing_events
, u64
*ent_ts
)
1809 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1810 struct trace_entry
*ent
, *next
= NULL
;
1811 unsigned long lost_events
= 0, next_lost
= 0;
1812 int cpu_file
= iter
->cpu_file
;
1813 u64 next_ts
= 0, ts
;
1819 * If we are in a per_cpu trace file, don't bother by iterating over
1820 * all cpu and peek directly.
1822 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1823 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1825 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
, missing_events
);
1827 *ent_cpu
= cpu_file
;
1832 for_each_tracing_cpu(cpu
) {
1834 if (ring_buffer_empty_cpu(buffer
, cpu
))
1837 ent
= peek_next_entry(iter
, cpu
, &ts
, &lost_events
);
1840 * Pick the entry with the smallest timestamp:
1842 if (ent
&& (!next
|| ts
< next_ts
)) {
1846 next_lost
= lost_events
;
1847 next_size
= iter
->ent_size
;
1851 iter
->ent_size
= next_size
;
1854 *ent_cpu
= next_cpu
;
1860 *missing_events
= next_lost
;
1865 /* Find the next real entry, without updating the iterator itself */
1866 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1867 int *ent_cpu
, u64
*ent_ts
)
1869 return __find_next_entry(iter
, ent_cpu
, NULL
, ent_ts
);
1872 /* Find the next real entry, and increment the iterator to the next entry */
1873 void *trace_find_next_entry_inc(struct trace_iterator
*iter
)
1875 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
,
1876 &iter
->lost_events
, &iter
->ts
);
1879 trace_iterator_increment(iter
);
1881 return iter
->ent
? iter
: NULL
;
1884 static void trace_consume(struct trace_iterator
*iter
)
1886 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
,
1887 &iter
->lost_events
);
1890 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1892 struct trace_iterator
*iter
= m
->private;
1896 WARN_ON_ONCE(iter
->leftover
);
1900 /* can't go backwards */
1905 ent
= trace_find_next_entry_inc(iter
);
1909 while (ent
&& iter
->idx
< i
)
1910 ent
= trace_find_next_entry_inc(iter
);
1917 void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
1919 struct trace_array
*tr
= iter
->tr
;
1920 struct ring_buffer_event
*event
;
1921 struct ring_buffer_iter
*buf_iter
;
1922 unsigned long entries
= 0;
1925 tr
->data
[cpu
]->skipped_entries
= 0;
1927 buf_iter
= trace_buffer_iter(iter
, cpu
);
1931 ring_buffer_iter_reset(buf_iter
);
1934 * We could have the case with the max latency tracers
1935 * that a reset never took place on a cpu. This is evident
1936 * by the timestamp being before the start of the buffer.
1938 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
1939 if (ts
>= iter
->tr
->time_start
)
1942 ring_buffer_read(buf_iter
, NULL
);
1945 tr
->data
[cpu
]->skipped_entries
= entries
;
1949 * The current tracer is copied to avoid a global locking
1952 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1954 struct trace_iterator
*iter
= m
->private;
1955 int cpu_file
= iter
->cpu_file
;
1961 * copy the tracer to avoid using a global lock all around.
1962 * iter->trace is a copy of current_trace, the pointer to the
1963 * name may be used instead of a strcmp(), as iter->trace->name
1964 * will point to the same string as current_trace->name.
1966 mutex_lock(&trace_types_lock
);
1967 if (unlikely(current_trace
&& iter
->trace
->name
!= current_trace
->name
))
1968 *iter
->trace
= *current_trace
;
1969 mutex_unlock(&trace_types_lock
);
1971 if (iter
->snapshot
&& iter
->trace
->use_max_tr
)
1972 return ERR_PTR(-EBUSY
);
1974 if (!iter
->snapshot
)
1975 atomic_inc(&trace_record_cmdline_disabled
);
1977 if (*pos
!= iter
->pos
) {
1982 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1983 for_each_tracing_cpu(cpu
)
1984 tracing_iter_reset(iter
, cpu
);
1986 tracing_iter_reset(iter
, cpu_file
);
1989 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1994 * If we overflowed the seq_file before, then we want
1995 * to just reuse the trace_seq buffer again.
2001 p
= s_next(m
, p
, &l
);
2005 trace_event_read_lock();
2006 trace_access_lock(cpu_file
);
2010 static void s_stop(struct seq_file
*m
, void *p
)
2012 struct trace_iterator
*iter
= m
->private;
2014 if (iter
->snapshot
&& iter
->trace
->use_max_tr
)
2017 if (!iter
->snapshot
)
2018 atomic_dec(&trace_record_cmdline_disabled
);
2019 trace_access_unlock(iter
->cpu_file
);
2020 trace_event_read_unlock();
2024 get_total_entries(struct trace_array
*tr
, unsigned long *total
, unsigned long *entries
)
2026 unsigned long count
;
2032 for_each_tracing_cpu(cpu
) {
2033 count
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
2035 * If this buffer has skipped entries, then we hold all
2036 * entries for the trace and we need to ignore the
2037 * ones before the time stamp.
2039 if (tr
->data
[cpu
]->skipped_entries
) {
2040 count
-= tr
->data
[cpu
]->skipped_entries
;
2041 /* total is the same as the entries */
2045 ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
2050 static void print_lat_help_header(struct seq_file
*m
)
2052 seq_puts(m
, "# _------=> CPU# \n");
2053 seq_puts(m
, "# / _-----=> irqs-off \n");
2054 seq_puts(m
, "# | / _----=> need-resched \n");
2055 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
2056 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
2057 seq_puts(m
, "# |||| / delay \n");
2058 seq_puts(m
, "# cmd pid ||||| time | caller \n");
2059 seq_puts(m
, "# \\ / ||||| \\ | / \n");
2062 static void print_event_info(struct trace_array
*tr
, struct seq_file
*m
)
2064 unsigned long total
;
2065 unsigned long entries
;
2067 get_total_entries(tr
, &total
, &entries
);
2068 seq_printf(m
, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2069 entries
, total
, num_online_cpus());
2073 static void print_func_help_header(struct trace_array
*tr
, struct seq_file
*m
)
2075 print_event_info(tr
, m
);
2076 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2077 seq_puts(m
, "# | | | | |\n");
2080 static void print_func_help_header_irq(struct trace_array
*tr
, struct seq_file
*m
)
2082 print_event_info(tr
, m
);
2083 seq_puts(m
, "# _-----=> irqs-off\n");
2084 seq_puts(m
, "# / _----=> need-resched\n");
2085 seq_puts(m
, "# | / _---=> hardirq/softirq\n");
2086 seq_puts(m
, "# || / _--=> preempt-depth\n");
2087 seq_puts(m
, "# ||| / delay\n");
2088 seq_puts(m
, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2089 seq_puts(m
, "# | | | |||| | |\n");
2093 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
2095 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2096 struct trace_array
*tr
= iter
->tr
;
2097 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
2098 struct tracer
*type
= current_trace
;
2099 unsigned long entries
;
2100 unsigned long total
;
2101 const char *name
= "preemption";
2106 get_total_entries(tr
, &total
, &entries
);
2108 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
2110 seq_puts(m
, "# -----------------------------------"
2111 "---------------------------------\n");
2112 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2113 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2114 nsecs_to_usecs(data
->saved_latency
),
2118 #if defined(CONFIG_PREEMPT_NONE)
2120 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2122 #elif defined(CONFIG_PREEMPT)
2127 /* These are reserved for later use */
2130 seq_printf(m
, " #P:%d)\n", num_online_cpus());
2134 seq_puts(m
, "# -----------------\n");
2135 seq_printf(m
, "# | task: %.16s-%d "
2136 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2137 data
->comm
, data
->pid
,
2138 from_kuid_munged(seq_user_ns(m
), data
->uid
), data
->nice
,
2139 data
->policy
, data
->rt_priority
);
2140 seq_puts(m
, "# -----------------\n");
2142 if (data
->critical_start
) {
2143 seq_puts(m
, "# => started at: ");
2144 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
2145 trace_print_seq(m
, &iter
->seq
);
2146 seq_puts(m
, "\n# => ended at: ");
2147 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
2148 trace_print_seq(m
, &iter
->seq
);
2149 seq_puts(m
, "\n#\n");
2155 static void test_cpu_buff_start(struct trace_iterator
*iter
)
2157 struct trace_seq
*s
= &iter
->seq
;
2159 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
2162 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
2165 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
2168 if (iter
->tr
->data
[iter
->cpu
]->skipped_entries
)
2171 cpumask_set_cpu(iter
->cpu
, iter
->started
);
2173 /* Don't print started cpu buffer for the first entry of the trace */
2175 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
2179 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
2181 struct trace_seq
*s
= &iter
->seq
;
2182 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2183 struct trace_entry
*entry
;
2184 struct trace_event
*event
;
2188 test_cpu_buff_start(iter
);
2190 event
= ftrace_find_event(entry
->type
);
2192 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2193 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2194 if (!trace_print_lat_context(iter
))
2197 if (!trace_print_context(iter
))
2203 return event
->funcs
->trace(iter
, sym_flags
, event
);
2205 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
2208 return TRACE_TYPE_HANDLED
;
2210 return TRACE_TYPE_PARTIAL_LINE
;
2213 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
2215 struct trace_seq
*s
= &iter
->seq
;
2216 struct trace_entry
*entry
;
2217 struct trace_event
*event
;
2221 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2222 if (!trace_seq_printf(s
, "%d %d %llu ",
2223 entry
->pid
, iter
->cpu
, iter
->ts
))
2227 event
= ftrace_find_event(entry
->type
);
2229 return event
->funcs
->raw(iter
, 0, event
);
2231 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
2234 return TRACE_TYPE_HANDLED
;
2236 return TRACE_TYPE_PARTIAL_LINE
;
2239 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
2241 struct trace_seq
*s
= &iter
->seq
;
2242 unsigned char newline
= '\n';
2243 struct trace_entry
*entry
;
2244 struct trace_event
*event
;
2248 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2249 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
2250 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
2251 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
2254 event
= ftrace_find_event(entry
->type
);
2256 enum print_line_t ret
= event
->funcs
->hex(iter
, 0, event
);
2257 if (ret
!= TRACE_TYPE_HANDLED
)
2261 SEQ_PUT_FIELD_RET(s
, newline
);
2263 return TRACE_TYPE_HANDLED
;
2266 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
2268 struct trace_seq
*s
= &iter
->seq
;
2269 struct trace_entry
*entry
;
2270 struct trace_event
*event
;
2274 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2275 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
2276 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
2277 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
2280 event
= ftrace_find_event(entry
->type
);
2281 return event
? event
->funcs
->binary(iter
, 0, event
) :
2285 int trace_empty(struct trace_iterator
*iter
)
2287 struct ring_buffer_iter
*buf_iter
;
2290 /* If we are looking at one CPU buffer, only check that one */
2291 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
2292 cpu
= iter
->cpu_file
;
2293 buf_iter
= trace_buffer_iter(iter
, cpu
);
2295 if (!ring_buffer_iter_empty(buf_iter
))
2298 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2304 for_each_tracing_cpu(cpu
) {
2305 buf_iter
= trace_buffer_iter(iter
, cpu
);
2307 if (!ring_buffer_iter_empty(buf_iter
))
2310 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2318 /* Called with trace_event_read_lock() held. */
2319 enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
2321 enum print_line_t ret
;
2323 if (iter
->lost_events
&&
2324 !trace_seq_printf(&iter
->seq
, "CPU:%d [LOST %lu EVENTS]\n",
2325 iter
->cpu
, iter
->lost_events
))
2326 return TRACE_TYPE_PARTIAL_LINE
;
2328 if (iter
->trace
&& iter
->trace
->print_line
) {
2329 ret
= iter
->trace
->print_line(iter
);
2330 if (ret
!= TRACE_TYPE_UNHANDLED
)
2334 if (iter
->ent
->type
== TRACE_BPRINT
&&
2335 trace_flags
& TRACE_ITER_PRINTK
&&
2336 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2337 return trace_print_bprintk_msg_only(iter
);
2339 if (iter
->ent
->type
== TRACE_PRINT
&&
2340 trace_flags
& TRACE_ITER_PRINTK
&&
2341 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2342 return trace_print_printk_msg_only(iter
);
2344 if (trace_flags
& TRACE_ITER_BIN
)
2345 return print_bin_fmt(iter
);
2347 if (trace_flags
& TRACE_ITER_HEX
)
2348 return print_hex_fmt(iter
);
2350 if (trace_flags
& TRACE_ITER_RAW
)
2351 return print_raw_fmt(iter
);
2353 return print_trace_fmt(iter
);
2356 void trace_latency_header(struct seq_file
*m
)
2358 struct trace_iterator
*iter
= m
->private;
2360 /* print nothing if the buffers are empty */
2361 if (trace_empty(iter
))
2364 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
2365 print_trace_header(m
, iter
);
2367 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2368 print_lat_help_header(m
);
2371 void trace_default_header(struct seq_file
*m
)
2373 struct trace_iterator
*iter
= m
->private;
2375 if (!(trace_flags
& TRACE_ITER_CONTEXT_INFO
))
2378 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2379 /* print nothing if the buffers are empty */
2380 if (trace_empty(iter
))
2382 print_trace_header(m
, iter
);
2383 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2384 print_lat_help_header(m
);
2386 if (!(trace_flags
& TRACE_ITER_VERBOSE
)) {
2387 if (trace_flags
& TRACE_ITER_IRQ_INFO
)
2388 print_func_help_header_irq(iter
->tr
, m
);
2390 print_func_help_header(iter
->tr
, m
);
2395 static void test_ftrace_alive(struct seq_file
*m
)
2397 if (!ftrace_is_dead())
2399 seq_printf(m
, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2400 seq_printf(m
, "# MAY BE MISSING FUNCTION EVENTS\n");
2403 static int s_show(struct seq_file
*m
, void *v
)
2405 struct trace_iterator
*iter
= v
;
2408 if (iter
->ent
== NULL
) {
2410 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
2412 test_ftrace_alive(m
);
2414 if (iter
->trace
&& iter
->trace
->print_header
)
2415 iter
->trace
->print_header(m
);
2417 trace_default_header(m
);
2419 } else if (iter
->leftover
) {
2421 * If we filled the seq_file buffer earlier, we
2422 * want to just show it now.
2424 ret
= trace_print_seq(m
, &iter
->seq
);
2426 /* ret should this time be zero, but you never know */
2427 iter
->leftover
= ret
;
2430 print_trace_line(iter
);
2431 ret
= trace_print_seq(m
, &iter
->seq
);
2433 * If we overflow the seq_file buffer, then it will
2434 * ask us for this data again at start up.
2436 * ret is 0 if seq_file write succeeded.
2439 iter
->leftover
= ret
;
2445 static const struct seq_operations tracer_seq_ops
= {
2452 static struct trace_iterator
*
2453 __tracing_open(struct inode
*inode
, struct file
*file
, bool snapshot
)
2455 long cpu_file
= (long) inode
->i_private
;
2456 struct trace_iterator
*iter
;
2459 if (tracing_disabled
)
2460 return ERR_PTR(-ENODEV
);
2462 iter
= __seq_open_private(file
, &tracer_seq_ops
, sizeof(*iter
));
2464 return ERR_PTR(-ENOMEM
);
2466 iter
->buffer_iter
= kzalloc(sizeof(*iter
->buffer_iter
) * num_possible_cpus(),
2468 if (!iter
->buffer_iter
)
2472 * We make a copy of the current tracer to avoid concurrent
2473 * changes on it while we are reading.
2475 mutex_lock(&trace_types_lock
);
2476 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2481 *iter
->trace
= *current_trace
;
2483 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2486 if ((current_trace
&& current_trace
->print_max
) || snapshot
)
2489 iter
->tr
= &global_trace
;
2490 iter
->snapshot
= snapshot
;
2492 mutex_init(&iter
->mutex
);
2493 iter
->cpu_file
= cpu_file
;
2495 /* Notify the tracer early; before we stop tracing. */
2496 if (iter
->trace
&& iter
->trace
->open
)
2497 iter
->trace
->open(iter
);
2499 /* Annotate start of buffers if we had overruns */
2500 if (ring_buffer_overruns(iter
->tr
->buffer
))
2501 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2503 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2504 if (trace_clocks
[trace_clock_id
].in_ns
)
2505 iter
->iter_flags
|= TRACE_FILE_TIME_IN_NS
;
2507 /* stop the trace while dumping if we are not opening "snapshot" */
2508 if (!iter
->snapshot
)
2511 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
2512 for_each_tracing_cpu(cpu
) {
2513 iter
->buffer_iter
[cpu
] =
2514 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2516 ring_buffer_read_prepare_sync();
2517 for_each_tracing_cpu(cpu
) {
2518 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2519 tracing_iter_reset(iter
, cpu
);
2522 cpu
= iter
->cpu_file
;
2523 iter
->buffer_iter
[cpu
] =
2524 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2525 ring_buffer_read_prepare_sync();
2526 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2527 tracing_iter_reset(iter
, cpu
);
2530 mutex_unlock(&trace_types_lock
);
2535 mutex_unlock(&trace_types_lock
);
2537 kfree(iter
->buffer_iter
);
2539 seq_release_private(inode
, file
);
2540 return ERR_PTR(-ENOMEM
);
2543 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2545 if (tracing_disabled
)
2548 filp
->private_data
= inode
->i_private
;
2552 static int tracing_release(struct inode
*inode
, struct file
*file
)
2554 struct seq_file
*m
= file
->private_data
;
2555 struct trace_iterator
*iter
;
2558 if (!(file
->f_mode
& FMODE_READ
))
2563 mutex_lock(&trace_types_lock
);
2564 for_each_tracing_cpu(cpu
) {
2565 if (iter
->buffer_iter
[cpu
])
2566 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2569 if (iter
->trace
&& iter
->trace
->close
)
2570 iter
->trace
->close(iter
);
2572 if (!iter
->snapshot
)
2573 /* reenable tracing if it was previously enabled */
2575 mutex_unlock(&trace_types_lock
);
2577 mutex_destroy(&iter
->mutex
);
2578 free_cpumask_var(iter
->started
);
2580 kfree(iter
->buffer_iter
);
2581 seq_release_private(inode
, file
);
2585 static int tracing_open(struct inode
*inode
, struct file
*file
)
2587 struct trace_iterator
*iter
;
2590 /* If this file was open for write, then erase contents */
2591 if ((file
->f_mode
& FMODE_WRITE
) &&
2592 (file
->f_flags
& O_TRUNC
)) {
2593 long cpu
= (long) inode
->i_private
;
2595 if (cpu
== TRACE_PIPE_ALL_CPU
)
2596 tracing_reset_online_cpus(&global_trace
);
2598 tracing_reset(&global_trace
, cpu
);
2601 if (file
->f_mode
& FMODE_READ
) {
2602 iter
= __tracing_open(inode
, file
, false);
2604 ret
= PTR_ERR(iter
);
2605 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2606 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2612 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2614 struct tracer
*t
= v
;
2624 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2629 mutex_lock(&trace_types_lock
);
2630 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2636 static void t_stop(struct seq_file
*m
, void *p
)
2638 mutex_unlock(&trace_types_lock
);
2641 static int t_show(struct seq_file
*m
, void *v
)
2643 struct tracer
*t
= v
;
2648 seq_printf(m
, "%s", t
->name
);
2657 static const struct seq_operations show_traces_seq_ops
= {
2664 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2666 if (tracing_disabled
)
2669 return seq_open(file
, &show_traces_seq_ops
);
2673 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2674 size_t count
, loff_t
*ppos
)
2679 static loff_t
tracing_seek(struct file
*file
, loff_t offset
, int origin
)
2681 if (file
->f_mode
& FMODE_READ
)
2682 return seq_lseek(file
, offset
, origin
);
2687 static const struct file_operations tracing_fops
= {
2688 .open
= tracing_open
,
2690 .write
= tracing_write_stub
,
2691 .llseek
= tracing_seek
,
2692 .release
= tracing_release
,
2695 static const struct file_operations show_traces_fops
= {
2696 .open
= show_traces_open
,
2698 .release
= seq_release
,
2699 .llseek
= seq_lseek
,
2703 * Only trace on a CPU if the bitmask is set:
2705 static cpumask_var_t tracing_cpumask
;
2708 * The tracer itself will not take this lock, but still we want
2709 * to provide a consistent cpumask to user-space:
2711 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2714 * Temporary storage for the character representation of the
2715 * CPU bitmask (and one more byte for the newline):
2717 static char mask_str
[NR_CPUS
+ 1];
2720 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2721 size_t count
, loff_t
*ppos
)
2725 mutex_lock(&tracing_cpumask_update_lock
);
2727 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2728 if (count
- len
< 2) {
2732 len
+= sprintf(mask_str
+ len
, "\n");
2733 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2736 mutex_unlock(&tracing_cpumask_update_lock
);
2742 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2743 size_t count
, loff_t
*ppos
)
2746 cpumask_var_t tracing_cpumask_new
;
2748 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2751 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2755 mutex_lock(&tracing_cpumask_update_lock
);
2757 local_irq_disable();
2758 arch_spin_lock(&ftrace_max_lock
);
2759 for_each_tracing_cpu(cpu
) {
2761 * Increase/decrease the disabled counter if we are
2762 * about to flip a bit in the cpumask:
2764 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2765 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2766 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2767 ring_buffer_record_disable_cpu(global_trace
.buffer
, cpu
);
2769 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2770 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2771 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2772 ring_buffer_record_enable_cpu(global_trace
.buffer
, cpu
);
2775 arch_spin_unlock(&ftrace_max_lock
);
2778 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2780 mutex_unlock(&tracing_cpumask_update_lock
);
2781 free_cpumask_var(tracing_cpumask_new
);
2786 free_cpumask_var(tracing_cpumask_new
);
2791 static const struct file_operations tracing_cpumask_fops
= {
2792 .open
= tracing_open_generic
,
2793 .read
= tracing_cpumask_read
,
2794 .write
= tracing_cpumask_write
,
2795 .llseek
= generic_file_llseek
,
2798 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
2800 struct tracer_opt
*trace_opts
;
2804 mutex_lock(&trace_types_lock
);
2805 tracer_flags
= current_trace
->flags
->val
;
2806 trace_opts
= current_trace
->flags
->opts
;
2808 for (i
= 0; trace_options
[i
]; i
++) {
2809 if (trace_flags
& (1 << i
))
2810 seq_printf(m
, "%s\n", trace_options
[i
]);
2812 seq_printf(m
, "no%s\n", trace_options
[i
]);
2815 for (i
= 0; trace_opts
[i
].name
; i
++) {
2816 if (tracer_flags
& trace_opts
[i
].bit
)
2817 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
2819 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
2821 mutex_unlock(&trace_types_lock
);
2826 static int __set_tracer_option(struct tracer
*trace
,
2827 struct tracer_flags
*tracer_flags
,
2828 struct tracer_opt
*opts
, int neg
)
2832 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
2837 tracer_flags
->val
&= ~opts
->bit
;
2839 tracer_flags
->val
|= opts
->bit
;
2843 /* Try to assign a tracer specific option */
2844 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2846 struct tracer_flags
*tracer_flags
= trace
->flags
;
2847 struct tracer_opt
*opts
= NULL
;
2850 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
2851 opts
= &tracer_flags
->opts
[i
];
2853 if (strcmp(cmp
, opts
->name
) == 0)
2854 return __set_tracer_option(trace
, trace
->flags
,
2861 static void set_tracer_flags(unsigned int mask
, int enabled
)
2863 /* do nothing if flag is already set */
2864 if (!!(trace_flags
& mask
) == !!enabled
)
2868 trace_flags
|= mask
;
2870 trace_flags
&= ~mask
;
2872 if (mask
== TRACE_ITER_RECORD_CMD
)
2873 trace_event_enable_cmd_record(enabled
);
2875 if (mask
== TRACE_ITER_OVERWRITE
)
2876 ring_buffer_change_overwrite(global_trace
.buffer
, enabled
);
2878 if (mask
== TRACE_ITER_PRINTK
)
2879 trace_printk_start_stop_comm(enabled
);
2882 static int trace_set_options(char *option
)
2889 cmp
= strstrip(option
);
2891 if (strncmp(cmp
, "no", 2) == 0) {
2896 for (i
= 0; trace_options
[i
]; i
++) {
2897 if (strcmp(cmp
, trace_options
[i
]) == 0) {
2898 set_tracer_flags(1 << i
, !neg
);
2903 /* If no option could be set, test the specific tracer options */
2904 if (!trace_options
[i
]) {
2905 mutex_lock(&trace_types_lock
);
2906 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2907 mutex_unlock(&trace_types_lock
);
2914 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2915 size_t cnt
, loff_t
*ppos
)
2919 if (cnt
>= sizeof(buf
))
2922 if (copy_from_user(&buf
, ubuf
, cnt
))
2927 trace_set_options(buf
);
2934 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
2936 if (tracing_disabled
)
2938 return single_open(file
, tracing_trace_options_show
, NULL
);
2941 static const struct file_operations tracing_iter_fops
= {
2942 .open
= tracing_trace_options_open
,
2944 .llseek
= seq_lseek
,
2945 .release
= single_release
,
2946 .write
= tracing_trace_options_write
,
2949 static const char readme_msg
[] =
2950 "tracing mini-HOWTO:\n\n"
2951 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2952 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2953 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2954 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2956 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2957 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2959 "# cat /sys/kernel/debug/tracing/trace_options\n"
2960 "noprint-parent nosym-offset nosym-addr noverbose\n"
2961 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2962 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2963 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2964 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2968 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2969 size_t cnt
, loff_t
*ppos
)
2971 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2972 readme_msg
, strlen(readme_msg
));
2975 static const struct file_operations tracing_readme_fops
= {
2976 .open
= tracing_open_generic
,
2977 .read
= tracing_readme_read
,
2978 .llseek
= generic_file_llseek
,
2982 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
2983 size_t cnt
, loff_t
*ppos
)
2992 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
2996 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
3004 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
3007 pid
= map_cmdline_to_pid
[i
];
3008 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
3011 trace_find_cmdline(pid
, buf_comm
);
3012 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
3017 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
3026 static const struct file_operations tracing_saved_cmdlines_fops
= {
3027 .open
= tracing_open_generic
,
3028 .read
= tracing_saved_cmdlines_read
,
3029 .llseek
= generic_file_llseek
,
3033 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
3034 size_t cnt
, loff_t
*ppos
)
3036 char buf
[MAX_TRACER_SIZE
+2];
3039 mutex_lock(&trace_types_lock
);
3041 r
= sprintf(buf
, "%s\n", current_trace
->name
);
3043 r
= sprintf(buf
, "\n");
3044 mutex_unlock(&trace_types_lock
);
3046 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3049 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
3051 tracing_reset_online_cpus(tr
);
3055 static void set_buffer_entries(struct trace_array
*tr
, unsigned long val
)
3058 for_each_tracing_cpu(cpu
)
3059 tr
->data
[cpu
]->entries
= val
;
3062 /* resize @tr's buffer to the size of @size_tr's entries */
3063 static int resize_buffer_duplicate_size(struct trace_array
*tr
,
3064 struct trace_array
*size_tr
, int cpu_id
)
3068 if (cpu_id
== RING_BUFFER_ALL_CPUS
) {
3069 for_each_tracing_cpu(cpu
) {
3070 ret
= ring_buffer_resize(tr
->buffer
,
3071 size_tr
->data
[cpu
]->entries
, cpu
);
3074 tr
->data
[cpu
]->entries
= size_tr
->data
[cpu
]->entries
;
3077 ret
= ring_buffer_resize(tr
->buffer
,
3078 size_tr
->data
[cpu_id
]->entries
, cpu_id
);
3080 tr
->data
[cpu_id
]->entries
=
3081 size_tr
->data
[cpu_id
]->entries
;
3087 static int __tracing_resize_ring_buffer(unsigned long size
, int cpu
)
3092 * If kernel or user changes the size of the ring buffer
3093 * we use the size that was given, and we can forget about
3094 * expanding it later.
3096 ring_buffer_expanded
= 1;
3098 /* May be called before buffers are initialized */
3099 if (!global_trace
.buffer
)
3102 ret
= ring_buffer_resize(global_trace
.buffer
, size
, cpu
);
3106 if (!current_trace
->use_max_tr
)
3109 ret
= ring_buffer_resize(max_tr
.buffer
, size
, cpu
);
3111 int r
= resize_buffer_duplicate_size(&global_trace
,
3112 &global_trace
, cpu
);
3115 * AARGH! We are left with different
3116 * size max buffer!!!!
3117 * The max buffer is our "snapshot" buffer.
3118 * When a tracer needs a snapshot (one of the
3119 * latency tracers), it swaps the max buffer
3120 * with the saved snap shot. We succeeded to
3121 * update the size of the main buffer, but failed to
3122 * update the size of the max buffer. But when we tried
3123 * to reset the main buffer to the original size, we
3124 * failed there too. This is very unlikely to
3125 * happen, but if it does, warn and kill all
3129 tracing_disabled
= 1;
3134 if (cpu
== RING_BUFFER_ALL_CPUS
)
3135 set_buffer_entries(&max_tr
, size
);
3137 max_tr
.data
[cpu
]->entries
= size
;
3140 if (cpu
== RING_BUFFER_ALL_CPUS
)
3141 set_buffer_entries(&global_trace
, size
);
3143 global_trace
.data
[cpu
]->entries
= size
;
3148 static ssize_t
tracing_resize_ring_buffer(unsigned long size
, int cpu_id
)
3152 mutex_lock(&trace_types_lock
);
3154 if (cpu_id
!= RING_BUFFER_ALL_CPUS
) {
3155 /* make sure, this cpu is enabled in the mask */
3156 if (!cpumask_test_cpu(cpu_id
, tracing_buffer_mask
)) {
3162 ret
= __tracing_resize_ring_buffer(size
, cpu_id
);
3167 mutex_unlock(&trace_types_lock
);
3174 * tracing_update_buffers - used by tracing facility to expand ring buffers
3176 * To save on memory when the tracing is never used on a system with it
3177 * configured in. The ring buffers are set to a minimum size. But once
3178 * a user starts to use the tracing facility, then they need to grow
3179 * to their default size.
3181 * This function is to be called when a tracer is about to be used.
3183 int tracing_update_buffers(void)
3187 mutex_lock(&trace_types_lock
);
3188 if (!ring_buffer_expanded
)
3189 ret
= __tracing_resize_ring_buffer(trace_buf_size
,
3190 RING_BUFFER_ALL_CPUS
);
3191 mutex_unlock(&trace_types_lock
);
3196 struct trace_option_dentry
;
3198 static struct trace_option_dentry
*
3199 create_trace_option_files(struct tracer
*tracer
);
3202 destroy_trace_option_files(struct trace_option_dentry
*topts
);
3204 static int tracing_set_tracer(const char *buf
)
3206 static struct trace_option_dentry
*topts
;
3207 struct trace_array
*tr
= &global_trace
;
3212 mutex_lock(&trace_types_lock
);
3214 if (!ring_buffer_expanded
) {
3215 ret
= __tracing_resize_ring_buffer(trace_buf_size
,
3216 RING_BUFFER_ALL_CPUS
);
3222 for (t
= trace_types
; t
; t
= t
->next
) {
3223 if (strcmp(t
->name
, buf
) == 0)
3230 if (t
== current_trace
)
3233 trace_branch_disable();
3234 if (current_trace
&& current_trace
->reset
)
3235 current_trace
->reset(tr
);
3237 had_max_tr
= current_trace
&& current_trace
->allocated_snapshot
;
3238 current_trace
= &nop_trace
;
3240 if (had_max_tr
&& !t
->use_max_tr
) {
3242 * We need to make sure that the update_max_tr sees that
3243 * current_trace changed to nop_trace to keep it from
3244 * swapping the buffers after we resize it.
3245 * The update_max_tr is called from interrupts disabled
3246 * so a synchronized_sched() is sufficient.
3248 synchronize_sched();
3250 * We don't free the ring buffer. instead, resize it because
3251 * The max_tr ring buffer has some state (e.g. ring->clock) and
3252 * we want preserve it.
3254 ring_buffer_resize(max_tr
.buffer
, 1, RING_BUFFER_ALL_CPUS
);
3255 set_buffer_entries(&max_tr
, 1);
3256 tracing_reset_online_cpus(&max_tr
);
3257 current_trace
->allocated_snapshot
= false;
3259 destroy_trace_option_files(topts
);
3261 topts
= create_trace_option_files(t
);
3262 if (t
->use_max_tr
&& !had_max_tr
) {
3263 /* we need to make per cpu buffer sizes equivalent */
3264 ret
= resize_buffer_duplicate_size(&max_tr
, &global_trace
,
3265 RING_BUFFER_ALL_CPUS
);
3268 t
->allocated_snapshot
= true;
3272 ret
= tracer_init(t
, tr
);
3278 trace_branch_enable(tr
);
3280 mutex_unlock(&trace_types_lock
);
3286 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
3287 size_t cnt
, loff_t
*ppos
)
3289 char buf
[MAX_TRACER_SIZE
+1];
3296 if (cnt
> MAX_TRACER_SIZE
)
3297 cnt
= MAX_TRACER_SIZE
;
3299 if (copy_from_user(&buf
, ubuf
, cnt
))
3304 /* strip ending whitespace. */
3305 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
3308 err
= tracing_set_tracer(buf
);
3318 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
3319 size_t cnt
, loff_t
*ppos
)
3321 unsigned long *ptr
= filp
->private_data
;
3325 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
3326 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
3327 if (r
> sizeof(buf
))
3329 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3333 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
3334 size_t cnt
, loff_t
*ppos
)
3336 unsigned long *ptr
= filp
->private_data
;
3340 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3349 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
3351 long cpu_file
= (long) inode
->i_private
;
3352 struct trace_iterator
*iter
;
3355 if (tracing_disabled
)
3358 mutex_lock(&trace_types_lock
);
3360 /* create a buffer to store the information to pass to userspace */
3361 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
3368 * We make a copy of the current tracer to avoid concurrent
3369 * changes on it while we are reading.
3371 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
3377 *iter
->trace
= *current_trace
;
3379 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
3384 /* trace pipe does not show start of buffer */
3385 cpumask_setall(iter
->started
);
3387 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
3388 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
3390 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3391 if (trace_clocks
[trace_clock_id
].in_ns
)
3392 iter
->iter_flags
|= TRACE_FILE_TIME_IN_NS
;
3394 iter
->cpu_file
= cpu_file
;
3395 iter
->tr
= &global_trace
;
3396 mutex_init(&iter
->mutex
);
3397 filp
->private_data
= iter
;
3399 if (iter
->trace
->pipe_open
)
3400 iter
->trace
->pipe_open(iter
);
3402 nonseekable_open(inode
, filp
);
3404 mutex_unlock(&trace_types_lock
);
3410 mutex_unlock(&trace_types_lock
);
3414 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
3416 struct trace_iterator
*iter
= file
->private_data
;
3418 mutex_lock(&trace_types_lock
);
3420 if (iter
->trace
->pipe_close
)
3421 iter
->trace
->pipe_close(iter
);
3423 mutex_unlock(&trace_types_lock
);
3425 free_cpumask_var(iter
->started
);
3426 mutex_destroy(&iter
->mutex
);
3434 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
3436 struct trace_iterator
*iter
= filp
->private_data
;
3438 if (trace_flags
& TRACE_ITER_BLOCK
) {
3440 * Always select as readable when in blocking mode
3442 return POLLIN
| POLLRDNORM
;
3444 if (!trace_empty(iter
))
3445 return POLLIN
| POLLRDNORM
;
3446 poll_wait(filp
, &trace_wait
, poll_table
);
3447 if (!trace_empty(iter
))
3448 return POLLIN
| POLLRDNORM
;
3455 * This is a make-shift waitqueue.
3456 * A tracer might use this callback on some rare cases:
3458 * 1) the current tracer might hold the runqueue lock when it wakes up
3459 * a reader, hence a deadlock (sched, function, and function graph tracers)
3460 * 2) the function tracers, trace all functions, we don't want
3461 * the overhead of calling wake_up and friends
3462 * (and tracing them too)
3464 * Anyway, this is really very primitive wakeup.
3466 void poll_wait_pipe(struct trace_iterator
*iter
)
3468 set_current_state(TASK_INTERRUPTIBLE
);
3469 /* sleep for 100 msecs, and try again. */
3470 schedule_timeout(HZ
/ 10);
3473 /* Must be called with trace_types_lock mutex held. */
3474 static int tracing_wait_pipe(struct file
*filp
)
3476 struct trace_iterator
*iter
= filp
->private_data
;
3478 while (trace_empty(iter
)) {
3480 if ((filp
->f_flags
& O_NONBLOCK
)) {
3484 mutex_unlock(&iter
->mutex
);
3486 iter
->trace
->wait_pipe(iter
);
3488 mutex_lock(&iter
->mutex
);
3490 if (signal_pending(current
))
3494 * We block until we read something and tracing is disabled.
3495 * We still block if tracing is disabled, but we have never
3496 * read anything. This allows a user to cat this file, and
3497 * then enable tracing. But after we have read something,
3498 * we give an EOF when tracing is again disabled.
3500 * iter->pos will be 0 if we haven't read anything.
3502 if (!tracing_is_enabled() && iter
->pos
)
3513 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
3514 size_t cnt
, loff_t
*ppos
)
3516 struct trace_iterator
*iter
= filp
->private_data
;
3519 /* return any leftover data */
3520 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3524 trace_seq_init(&iter
->seq
);
3526 /* copy the tracer to avoid using a global lock all around */
3527 mutex_lock(&trace_types_lock
);
3528 if (unlikely(current_trace
&& iter
->trace
->name
!= current_trace
->name
))
3529 *iter
->trace
= *current_trace
;
3530 mutex_unlock(&trace_types_lock
);
3533 * Avoid more than one consumer on a single file descriptor
3534 * This is just a matter of traces coherency, the ring buffer itself
3537 mutex_lock(&iter
->mutex
);
3538 if (iter
->trace
->read
) {
3539 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
3545 sret
= tracing_wait_pipe(filp
);
3549 /* stop when tracing is finished */
3550 if (trace_empty(iter
)) {
3555 if (cnt
>= PAGE_SIZE
)
3556 cnt
= PAGE_SIZE
- 1;
3558 /* reset all but tr, trace, and overruns */
3559 memset(&iter
->seq
, 0,
3560 sizeof(struct trace_iterator
) -
3561 offsetof(struct trace_iterator
, seq
));
3564 trace_event_read_lock();
3565 trace_access_lock(iter
->cpu_file
);
3566 while (trace_find_next_entry_inc(iter
) != NULL
) {
3567 enum print_line_t ret
;
3568 int len
= iter
->seq
.len
;
3570 ret
= print_trace_line(iter
);
3571 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3572 /* don't print partial lines */
3573 iter
->seq
.len
= len
;
3576 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3577 trace_consume(iter
);
3579 if (iter
->seq
.len
>= cnt
)
3583 * Setting the full flag means we reached the trace_seq buffer
3584 * size and we should leave by partial output condition above.
3585 * One of the trace_seq_* functions is not used properly.
3587 WARN_ONCE(iter
->seq
.full
, "full flag set for trace type %d",
3590 trace_access_unlock(iter
->cpu_file
);
3591 trace_event_read_unlock();
3593 /* Now copy what we have to the user */
3594 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3595 if (iter
->seq
.readpos
>= iter
->seq
.len
)
3596 trace_seq_init(&iter
->seq
);
3599 * If there was nothing to send to user, in spite of consuming trace
3600 * entries, go back to wait for more entries.
3606 mutex_unlock(&iter
->mutex
);
3611 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
3612 struct pipe_buffer
*buf
)
3614 __free_page(buf
->page
);
3617 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
3620 __free_page(spd
->pages
[idx
]);
3623 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
3625 .map
= generic_pipe_buf_map
,
3626 .unmap
= generic_pipe_buf_unmap
,
3627 .confirm
= generic_pipe_buf_confirm
,
3628 .release
= tracing_pipe_buf_release
,
3629 .steal
= generic_pipe_buf_steal
,
3630 .get
= generic_pipe_buf_get
,
3634 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
3639 /* Seq buffer is page-sized, exactly what we need. */
3641 count
= iter
->seq
.len
;
3642 ret
= print_trace_line(iter
);
3643 count
= iter
->seq
.len
- count
;
3646 iter
->seq
.len
-= count
;
3649 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3650 iter
->seq
.len
-= count
;
3654 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3655 trace_consume(iter
);
3657 if (!trace_find_next_entry_inc(iter
)) {
3667 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3669 struct pipe_inode_info
*pipe
,
3673 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3674 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3675 struct trace_iterator
*iter
= filp
->private_data
;
3676 struct splice_pipe_desc spd
= {
3678 .partial
= partial_def
,
3679 .nr_pages
= 0, /* This gets updated below. */
3680 .nr_pages_max
= PIPE_DEF_BUFFERS
,
3682 .ops
= &tracing_pipe_buf_ops
,
3683 .spd_release
= tracing_spd_release_pipe
,
3689 if (splice_grow_spd(pipe
, &spd
))
3692 /* copy the tracer to avoid using a global lock all around */
3693 mutex_lock(&trace_types_lock
);
3694 if (unlikely(current_trace
&& iter
->trace
->name
!= current_trace
->name
))
3695 *iter
->trace
= *current_trace
;
3696 mutex_unlock(&trace_types_lock
);
3698 mutex_lock(&iter
->mutex
);
3700 if (iter
->trace
->splice_read
) {
3701 ret
= iter
->trace
->splice_read(iter
, filp
,
3702 ppos
, pipe
, len
, flags
);
3707 ret
= tracing_wait_pipe(filp
);
3711 if (!iter
->ent
&& !trace_find_next_entry_inc(iter
)) {
3716 trace_event_read_lock();
3717 trace_access_lock(iter
->cpu_file
);
3719 /* Fill as many pages as possible. */
3720 for (i
= 0, rem
= len
; i
< pipe
->buffers
&& rem
; i
++) {
3721 spd
.pages
[i
] = alloc_page(GFP_KERNEL
);
3725 rem
= tracing_fill_pipe_page(rem
, iter
);
3727 /* Copy the data into the page, so we can start over. */
3728 ret
= trace_seq_to_buffer(&iter
->seq
,
3729 page_address(spd
.pages
[i
]),
3732 __free_page(spd
.pages
[i
]);
3735 spd
.partial
[i
].offset
= 0;
3736 spd
.partial
[i
].len
= iter
->seq
.len
;
3738 trace_seq_init(&iter
->seq
);
3741 trace_access_unlock(iter
->cpu_file
);
3742 trace_event_read_unlock();
3743 mutex_unlock(&iter
->mutex
);
3747 ret
= splice_to_pipe(pipe
, &spd
);
3749 splice_shrink_spd(&spd
);
3753 mutex_unlock(&iter
->mutex
);
3757 struct ftrace_entries_info
{
3758 struct trace_array
*tr
;
3762 static int tracing_entries_open(struct inode
*inode
, struct file
*filp
)
3764 struct ftrace_entries_info
*info
;
3766 if (tracing_disabled
)
3769 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3773 info
->tr
= &global_trace
;
3774 info
->cpu
= (unsigned long)inode
->i_private
;
3776 filp
->private_data
= info
;
3782 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3783 size_t cnt
, loff_t
*ppos
)
3785 struct ftrace_entries_info
*info
= filp
->private_data
;
3786 struct trace_array
*tr
= info
->tr
;
3791 mutex_lock(&trace_types_lock
);
3793 if (info
->cpu
== RING_BUFFER_ALL_CPUS
) {
3794 int cpu
, buf_size_same
;
3799 /* check if all cpu sizes are same */
3800 for_each_tracing_cpu(cpu
) {
3801 /* fill in the size from first enabled cpu */
3803 size
= tr
->data
[cpu
]->entries
;
3804 if (size
!= tr
->data
[cpu
]->entries
) {
3810 if (buf_size_same
) {
3811 if (!ring_buffer_expanded
)
3812 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3814 trace_buf_size
>> 10);
3816 r
= sprintf(buf
, "%lu\n", size
>> 10);
3818 r
= sprintf(buf
, "X\n");
3820 r
= sprintf(buf
, "%lu\n", tr
->data
[info
->cpu
]->entries
>> 10);
3822 mutex_unlock(&trace_types_lock
);
3824 ret
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3829 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3830 size_t cnt
, loff_t
*ppos
)
3832 struct ftrace_entries_info
*info
= filp
->private_data
;
3836 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3840 /* must have at least 1 entry */
3844 /* value is in KB */
3847 ret
= tracing_resize_ring_buffer(val
, info
->cpu
);
3857 tracing_entries_release(struct inode
*inode
, struct file
*filp
)
3859 struct ftrace_entries_info
*info
= filp
->private_data
;
3867 tracing_total_entries_read(struct file
*filp
, char __user
*ubuf
,
3868 size_t cnt
, loff_t
*ppos
)
3870 struct trace_array
*tr
= filp
->private_data
;
3873 unsigned long size
= 0, expanded_size
= 0;
3875 mutex_lock(&trace_types_lock
);
3876 for_each_tracing_cpu(cpu
) {
3877 size
+= tr
->data
[cpu
]->entries
>> 10;
3878 if (!ring_buffer_expanded
)
3879 expanded_size
+= trace_buf_size
>> 10;
3881 if (ring_buffer_expanded
)
3882 r
= sprintf(buf
, "%lu\n", size
);
3884 r
= sprintf(buf
, "%lu (expanded: %lu)\n", size
, expanded_size
);
3885 mutex_unlock(&trace_types_lock
);
3887 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3891 tracing_free_buffer_write(struct file
*filp
, const char __user
*ubuf
,
3892 size_t cnt
, loff_t
*ppos
)
3895 * There is no need to read what the user has written, this function
3896 * is just to make sure that there is no error when "echo" is used
3905 tracing_free_buffer_release(struct inode
*inode
, struct file
*filp
)
3907 /* disable tracing ? */
3908 if (trace_flags
& TRACE_ITER_STOP_ON_FREE
)
3910 /* resize the ring buffer to 0 */
3911 tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS
);
3917 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3918 size_t cnt
, loff_t
*fpos
)
3920 unsigned long addr
= (unsigned long)ubuf
;
3921 struct ring_buffer_event
*event
;
3922 struct ring_buffer
*buffer
;
3923 struct print_entry
*entry
;
3924 unsigned long irq_flags
;
3925 struct page
*pages
[2];
3935 if (tracing_disabled
)
3938 if (!(trace_flags
& TRACE_ITER_MARKERS
))
3941 if (cnt
> TRACE_BUF_SIZE
)
3942 cnt
= TRACE_BUF_SIZE
;
3945 * Userspace is injecting traces into the kernel trace buffer.
3946 * We want to be as non intrusive as possible.
3947 * To do so, we do not want to allocate any special buffers
3948 * or take any locks, but instead write the userspace data
3949 * straight into the ring buffer.
3951 * First we need to pin the userspace buffer into memory,
3952 * which, most likely it is, because it just referenced it.
3953 * But there's no guarantee that it is. By using get_user_pages_fast()
3954 * and kmap_atomic/kunmap_atomic() we can get access to the
3955 * pages directly. We then write the data directly into the
3958 BUILD_BUG_ON(TRACE_BUF_SIZE
>= PAGE_SIZE
);
3960 /* check if we cross pages */
3961 if ((addr
& PAGE_MASK
) != ((addr
+ cnt
) & PAGE_MASK
))
3964 offset
= addr
& (PAGE_SIZE
- 1);
3967 ret
= get_user_pages_fast(addr
, nr_pages
, 0, pages
);
3968 if (ret
< nr_pages
) {
3970 put_page(pages
[ret
]);
3975 for (i
= 0; i
< nr_pages
; i
++)
3976 map_page
[i
] = kmap_atomic(pages
[i
]);
3978 local_save_flags(irq_flags
);
3979 size
= sizeof(*entry
) + cnt
+ 2; /* possible \n added */
3980 buffer
= global_trace
.buffer
;
3981 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
3982 irq_flags
, preempt_count());
3984 /* Ring buffer disabled, return as if not open for write */
3989 entry
= ring_buffer_event_data(event
);
3990 entry
->ip
= _THIS_IP_
;
3992 if (nr_pages
== 2) {
3993 len
= PAGE_SIZE
- offset
;
3994 memcpy(&entry
->buf
, map_page
[0] + offset
, len
);
3995 memcpy(&entry
->buf
[len
], map_page
[1], cnt
- len
);
3997 memcpy(&entry
->buf
, map_page
[0] + offset
, cnt
);
3999 if (entry
->buf
[cnt
- 1] != '\n') {
4000 entry
->buf
[cnt
] = '\n';
4001 entry
->buf
[cnt
+ 1] = '\0';
4003 entry
->buf
[cnt
] = '\0';
4005 __buffer_unlock_commit(buffer
, event
);
4012 for (i
= 0; i
< nr_pages
; i
++){
4013 kunmap_atomic(map_page
[i
]);
4020 static int tracing_clock_show(struct seq_file
*m
, void *v
)
4024 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
4026 "%s%s%s%s", i
? " " : "",
4027 i
== trace_clock_id
? "[" : "", trace_clocks
[i
].name
,
4028 i
== trace_clock_id
? "]" : "");
4034 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
4035 size_t cnt
, loff_t
*fpos
)
4038 const char *clockstr
;
4041 if (cnt
>= sizeof(buf
))
4044 if (copy_from_user(&buf
, ubuf
, cnt
))
4049 clockstr
= strstrip(buf
);
4051 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
4052 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
4055 if (i
== ARRAY_SIZE(trace_clocks
))
4060 mutex_lock(&trace_types_lock
);
4062 ring_buffer_set_clock(global_trace
.buffer
, trace_clocks
[i
].func
);
4064 ring_buffer_set_clock(max_tr
.buffer
, trace_clocks
[i
].func
);
4067 * New clock may not be consistent with the previous clock.
4068 * Reset the buffer so that it doesn't have incomparable timestamps.
4070 tracing_reset_online_cpus(&global_trace
);
4071 tracing_reset_online_cpus(&max_tr
);
4073 mutex_unlock(&trace_types_lock
);
4080 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
4082 if (tracing_disabled
)
4084 return single_open(file
, tracing_clock_show
, NULL
);
4087 #ifdef CONFIG_TRACER_SNAPSHOT
4088 static int tracing_snapshot_open(struct inode
*inode
, struct file
*file
)
4090 struct trace_iterator
*iter
;
4093 if (file
->f_mode
& FMODE_READ
) {
4094 iter
= __tracing_open(inode
, file
, true);
4096 ret
= PTR_ERR(iter
);
4102 tracing_snapshot_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4108 ret
= tracing_update_buffers();
4112 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4116 mutex_lock(&trace_types_lock
);
4118 if (current_trace
&& current_trace
->use_max_tr
) {
4125 if (current_trace
->allocated_snapshot
) {
4126 /* free spare buffer */
4127 ring_buffer_resize(max_tr
.buffer
, 1,
4128 RING_BUFFER_ALL_CPUS
);
4129 set_buffer_entries(&max_tr
, 1);
4130 tracing_reset_online_cpus(&max_tr
);
4131 current_trace
->allocated_snapshot
= false;
4135 if (!current_trace
->allocated_snapshot
) {
4136 /* allocate spare buffer */
4137 ret
= resize_buffer_duplicate_size(&max_tr
,
4138 &global_trace
, RING_BUFFER_ALL_CPUS
);
4141 current_trace
->allocated_snapshot
= true;
4144 local_irq_disable();
4145 /* Now, we're going to swap */
4146 update_max_tr(&global_trace
, current
, smp_processor_id());
4150 if (current_trace
->allocated_snapshot
)
4151 tracing_reset_online_cpus(&max_tr
);
4162 mutex_unlock(&trace_types_lock
);
4165 #endif /* CONFIG_TRACER_SNAPSHOT */
4168 static const struct file_operations tracing_max_lat_fops
= {
4169 .open
= tracing_open_generic
,
4170 .read
= tracing_max_lat_read
,
4171 .write
= tracing_max_lat_write
,
4172 .llseek
= generic_file_llseek
,
4175 static const struct file_operations set_tracer_fops
= {
4176 .open
= tracing_open_generic
,
4177 .read
= tracing_set_trace_read
,
4178 .write
= tracing_set_trace_write
,
4179 .llseek
= generic_file_llseek
,
4182 static const struct file_operations tracing_pipe_fops
= {
4183 .open
= tracing_open_pipe
,
4184 .poll
= tracing_poll_pipe
,
4185 .read
= tracing_read_pipe
,
4186 .splice_read
= tracing_splice_read_pipe
,
4187 .release
= tracing_release_pipe
,
4188 .llseek
= no_llseek
,
4191 static const struct file_operations tracing_entries_fops
= {
4192 .open
= tracing_entries_open
,
4193 .read
= tracing_entries_read
,
4194 .write
= tracing_entries_write
,
4195 .release
= tracing_entries_release
,
4196 .llseek
= generic_file_llseek
,
4199 static const struct file_operations tracing_total_entries_fops
= {
4200 .open
= tracing_open_generic
,
4201 .read
= tracing_total_entries_read
,
4202 .llseek
= generic_file_llseek
,
4205 static const struct file_operations tracing_free_buffer_fops
= {
4206 .write
= tracing_free_buffer_write
,
4207 .release
= tracing_free_buffer_release
,
4210 static const struct file_operations tracing_mark_fops
= {
4211 .open
= tracing_open_generic
,
4212 .write
= tracing_mark_write
,
4213 .llseek
= generic_file_llseek
,
4216 static const struct file_operations trace_clock_fops
= {
4217 .open
= tracing_clock_open
,
4219 .llseek
= seq_lseek
,
4220 .release
= single_release
,
4221 .write
= tracing_clock_write
,
4224 #ifdef CONFIG_TRACER_SNAPSHOT
4225 static const struct file_operations snapshot_fops
= {
4226 .open
= tracing_snapshot_open
,
4228 .write
= tracing_snapshot_write
,
4229 .llseek
= tracing_seek
,
4230 .release
= tracing_release
,
4232 #endif /* CONFIG_TRACER_SNAPSHOT */
4234 struct ftrace_buffer_info
{
4235 struct trace_array
*tr
;
4241 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
4243 int cpu
= (int)(long)inode
->i_private
;
4244 struct ftrace_buffer_info
*info
;
4246 if (tracing_disabled
)
4249 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
4253 info
->tr
= &global_trace
;
4256 /* Force reading ring buffer for first read */
4257 info
->read
= (unsigned int)-1;
4259 filp
->private_data
= info
;
4261 return nonseekable_open(inode
, filp
);
4265 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
4266 size_t count
, loff_t
*ppos
)
4268 struct ftrace_buffer_info
*info
= filp
->private_data
;
4276 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
, info
->cpu
);
4280 /* Do we have previous read data to read? */
4281 if (info
->read
< PAGE_SIZE
)
4284 trace_access_lock(info
->cpu
);
4285 ret
= ring_buffer_read_page(info
->tr
->buffer
,
4289 trace_access_unlock(info
->cpu
);
4296 size
= PAGE_SIZE
- info
->read
;
4300 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
4311 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
4313 struct ftrace_buffer_info
*info
= file
->private_data
;
4316 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
4323 struct ring_buffer
*buffer
;
4328 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
4329 struct pipe_buffer
*buf
)
4331 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
4336 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4341 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
4342 struct pipe_buffer
*buf
)
4344 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
4349 /* Pipe buffer operations for a buffer. */
4350 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
4352 .map
= generic_pipe_buf_map
,
4353 .unmap
= generic_pipe_buf_unmap
,
4354 .confirm
= generic_pipe_buf_confirm
,
4355 .release
= buffer_pipe_buf_release
,
4356 .steal
= generic_pipe_buf_steal
,
4357 .get
= buffer_pipe_buf_get
,
4361 * Callback from splice_to_pipe(), if we need to release some pages
4362 * at the end of the spd in case we error'ed out in filling the pipe.
4364 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
4366 struct buffer_ref
*ref
=
4367 (struct buffer_ref
*)spd
->partial
[i
].private;
4372 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4374 spd
->partial
[i
].private = 0;
4378 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
4379 struct pipe_inode_info
*pipe
, size_t len
,
4382 struct ftrace_buffer_info
*info
= file
->private_data
;
4383 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
4384 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
4385 struct splice_pipe_desc spd
= {
4387 .partial
= partial_def
,
4388 .nr_pages_max
= PIPE_DEF_BUFFERS
,
4390 .ops
= &buffer_pipe_buf_ops
,
4391 .spd_release
= buffer_spd_release
,
4393 struct buffer_ref
*ref
;
4394 int entries
, size
, i
;
4397 if (splice_grow_spd(pipe
, &spd
))
4400 if (*ppos
& (PAGE_SIZE
- 1)) {
4405 if (len
& (PAGE_SIZE
- 1)) {
4406 if (len
< PAGE_SIZE
) {
4413 trace_access_lock(info
->cpu
);
4414 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4416 for (i
= 0; i
< pipe
->buffers
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
4420 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
4425 ref
->buffer
= info
->tr
->buffer
;
4426 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
, info
->cpu
);
4432 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
4435 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4441 * zero out any left over data, this is going to
4444 size
= ring_buffer_page_len(ref
->page
);
4445 if (size
< PAGE_SIZE
)
4446 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
4448 page
= virt_to_page(ref
->page
);
4450 spd
.pages
[i
] = page
;
4451 spd
.partial
[i
].len
= PAGE_SIZE
;
4452 spd
.partial
[i
].offset
= 0;
4453 spd
.partial
[i
].private = (unsigned long)ref
;
4457 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4460 trace_access_unlock(info
->cpu
);
4463 /* did we read anything? */
4464 if (!spd
.nr_pages
) {
4465 if (flags
& SPLICE_F_NONBLOCK
)
4473 ret
= splice_to_pipe(pipe
, &spd
);
4474 splice_shrink_spd(&spd
);
4479 static const struct file_operations tracing_buffers_fops
= {
4480 .open
= tracing_buffers_open
,
4481 .read
= tracing_buffers_read
,
4482 .release
= tracing_buffers_release
,
4483 .splice_read
= tracing_buffers_splice_read
,
4484 .llseek
= no_llseek
,
4488 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
4489 size_t count
, loff_t
*ppos
)
4491 unsigned long cpu
= (unsigned long)filp
->private_data
;
4492 struct trace_array
*tr
= &global_trace
;
4493 struct trace_seq
*s
;
4495 unsigned long long t
;
4496 unsigned long usec_rem
;
4498 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
4504 cnt
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
4505 trace_seq_printf(s
, "entries: %ld\n", cnt
);
4507 cnt
= ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
4508 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
4510 cnt
= ring_buffer_commit_overrun_cpu(tr
->buffer
, cpu
);
4511 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
4513 cnt
= ring_buffer_bytes_cpu(tr
->buffer
, cpu
);
4514 trace_seq_printf(s
, "bytes: %ld\n", cnt
);
4516 if (trace_clocks
[trace_clock_id
].in_ns
) {
4517 /* local or global for trace_clock */
4518 t
= ns2usecs(ring_buffer_oldest_event_ts(tr
->buffer
, cpu
));
4519 usec_rem
= do_div(t
, USEC_PER_SEC
);
4520 trace_seq_printf(s
, "oldest event ts: %5llu.%06lu\n",
4523 t
= ns2usecs(ring_buffer_time_stamp(tr
->buffer
, cpu
));
4524 usec_rem
= do_div(t
, USEC_PER_SEC
);
4525 trace_seq_printf(s
, "now ts: %5llu.%06lu\n", t
, usec_rem
);
4527 /* counter or tsc mode for trace_clock */
4528 trace_seq_printf(s
, "oldest event ts: %llu\n",
4529 ring_buffer_oldest_event_ts(tr
->buffer
, cpu
));
4531 trace_seq_printf(s
, "now ts: %llu\n",
4532 ring_buffer_time_stamp(tr
->buffer
, cpu
));
4535 cnt
= ring_buffer_dropped_events_cpu(tr
->buffer
, cpu
);
4536 trace_seq_printf(s
, "dropped events: %ld\n", cnt
);
4538 cnt
= ring_buffer_read_events_cpu(tr
->buffer
, cpu
);
4539 trace_seq_printf(s
, "read events: %ld\n", cnt
);
4541 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
4548 static const struct file_operations tracing_stats_fops
= {
4549 .open
= tracing_open_generic
,
4550 .read
= tracing_stats_read
,
4551 .llseek
= generic_file_llseek
,
4554 #ifdef CONFIG_DYNAMIC_FTRACE
4556 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
4562 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
4563 size_t cnt
, loff_t
*ppos
)
4565 static char ftrace_dyn_info_buffer
[1024];
4566 static DEFINE_MUTEX(dyn_info_mutex
);
4567 unsigned long *p
= filp
->private_data
;
4568 char *buf
= ftrace_dyn_info_buffer
;
4569 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
4572 mutex_lock(&dyn_info_mutex
);
4573 r
= sprintf(buf
, "%ld ", *p
);
4575 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
4578 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4580 mutex_unlock(&dyn_info_mutex
);
4585 static const struct file_operations tracing_dyn_info_fops
= {
4586 .open
= tracing_open_generic
,
4587 .read
= tracing_read_dyn_info
,
4588 .llseek
= generic_file_llseek
,
4592 static struct dentry
*d_tracer
;
4594 struct dentry
*tracing_init_dentry(void)
4601 if (!debugfs_initialized())
4604 d_tracer
= debugfs_create_dir("tracing", NULL
);
4606 if (!d_tracer
&& !once
) {
4608 pr_warning("Could not create debugfs directory 'tracing'\n");
4615 static struct dentry
*d_percpu
;
4617 static struct dentry
*tracing_dentry_percpu(void)
4620 struct dentry
*d_tracer
;
4625 d_tracer
= tracing_init_dentry();
4630 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
4632 if (!d_percpu
&& !once
) {
4634 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4641 static void tracing_init_debugfs_percpu(long cpu
)
4643 struct dentry
*d_percpu
= tracing_dentry_percpu();
4644 struct dentry
*d_cpu
;
4645 char cpu_dir
[30]; /* 30 characters should be more than enough */
4650 snprintf(cpu_dir
, 30, "cpu%ld", cpu
);
4651 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
4653 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
4657 /* per cpu trace_pipe */
4658 trace_create_file("trace_pipe", 0444, d_cpu
,
4659 (void *) cpu
, &tracing_pipe_fops
);
4662 trace_create_file("trace", 0644, d_cpu
,
4663 (void *) cpu
, &tracing_fops
);
4665 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
4666 (void *) cpu
, &tracing_buffers_fops
);
4668 trace_create_file("stats", 0444, d_cpu
,
4669 (void *) cpu
, &tracing_stats_fops
);
4671 trace_create_file("buffer_size_kb", 0444, d_cpu
,
4672 (void *) cpu
, &tracing_entries_fops
);
4675 #ifdef CONFIG_FTRACE_SELFTEST
4676 /* Let selftest have access to static functions in this file */
4677 #include "trace_selftest.c"
4680 struct trace_option_dentry
{
4681 struct tracer_opt
*opt
;
4682 struct tracer_flags
*flags
;
4683 struct dentry
*entry
;
4687 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4690 struct trace_option_dentry
*topt
= filp
->private_data
;
4693 if (topt
->flags
->val
& topt
->opt
->bit
)
4698 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4702 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4705 struct trace_option_dentry
*topt
= filp
->private_data
;
4709 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4713 if (val
!= 0 && val
!= 1)
4716 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
4717 mutex_lock(&trace_types_lock
);
4718 ret
= __set_tracer_option(current_trace
, topt
->flags
,
4720 mutex_unlock(&trace_types_lock
);
4731 static const struct file_operations trace_options_fops
= {
4732 .open
= tracing_open_generic
,
4733 .read
= trace_options_read
,
4734 .write
= trace_options_write
,
4735 .llseek
= generic_file_llseek
,
4739 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4742 long index
= (long)filp
->private_data
;
4745 if (trace_flags
& (1 << index
))
4750 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4754 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4757 long index
= (long)filp
->private_data
;
4761 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4765 if (val
!= 0 && val
!= 1)
4767 set_tracer_flags(1 << index
, val
);
4774 static const struct file_operations trace_options_core_fops
= {
4775 .open
= tracing_open_generic
,
4776 .read
= trace_options_core_read
,
4777 .write
= trace_options_core_write
,
4778 .llseek
= generic_file_llseek
,
4781 struct dentry
*trace_create_file(const char *name
,
4783 struct dentry
*parent
,
4785 const struct file_operations
*fops
)
4789 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
4791 pr_warning("Could not create debugfs '%s' entry\n", name
);
4797 static struct dentry
*trace_options_init_dentry(void)
4799 struct dentry
*d_tracer
;
4800 static struct dentry
*t_options
;
4805 d_tracer
= tracing_init_dentry();
4809 t_options
= debugfs_create_dir("options", d_tracer
);
4811 pr_warning("Could not create debugfs directory 'options'\n");
4819 create_trace_option_file(struct trace_option_dentry
*topt
,
4820 struct tracer_flags
*flags
,
4821 struct tracer_opt
*opt
)
4823 struct dentry
*t_options
;
4825 t_options
= trace_options_init_dentry();
4829 topt
->flags
= flags
;
4832 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
4833 &trace_options_fops
);
4837 static struct trace_option_dentry
*
4838 create_trace_option_files(struct tracer
*tracer
)
4840 struct trace_option_dentry
*topts
;
4841 struct tracer_flags
*flags
;
4842 struct tracer_opt
*opts
;
4848 flags
= tracer
->flags
;
4850 if (!flags
|| !flags
->opts
)
4855 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4858 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
4862 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4863 create_trace_option_file(&topts
[cnt
], flags
,
4870 destroy_trace_option_files(struct trace_option_dentry
*topts
)
4877 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
4878 if (topts
[cnt
].entry
)
4879 debugfs_remove(topts
[cnt
].entry
);
4885 static struct dentry
*
4886 create_trace_option_core_file(const char *option
, long index
)
4888 struct dentry
*t_options
;
4890 t_options
= trace_options_init_dentry();
4894 return trace_create_file(option
, 0644, t_options
, (void *)index
,
4895 &trace_options_core_fops
);
4898 static __init
void create_trace_options_dir(void)
4900 struct dentry
*t_options
;
4903 t_options
= trace_options_init_dentry();
4907 for (i
= 0; trace_options
[i
]; i
++)
4908 create_trace_option_core_file(trace_options
[i
], i
);
4912 rb_simple_read(struct file
*filp
, char __user
*ubuf
,
4913 size_t cnt
, loff_t
*ppos
)
4915 struct trace_array
*tr
= filp
->private_data
;
4916 struct ring_buffer
*buffer
= tr
->buffer
;
4921 r
= ring_buffer_record_is_on(buffer
);
4925 r
= sprintf(buf
, "%d\n", r
);
4927 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4931 rb_simple_write(struct file
*filp
, const char __user
*ubuf
,
4932 size_t cnt
, loff_t
*ppos
)
4934 struct trace_array
*tr
= filp
->private_data
;
4935 struct ring_buffer
*buffer
= tr
->buffer
;
4939 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4944 mutex_lock(&trace_types_lock
);
4946 ring_buffer_record_on(buffer
);
4947 if (current_trace
->start
)
4948 current_trace
->start(tr
);
4950 ring_buffer_record_off(buffer
);
4951 if (current_trace
->stop
)
4952 current_trace
->stop(tr
);
4954 mutex_unlock(&trace_types_lock
);
4962 static const struct file_operations rb_simple_fops
= {
4963 .open
= tracing_open_generic
,
4964 .read
= rb_simple_read
,
4965 .write
= rb_simple_write
,
4966 .llseek
= default_llseek
,
4969 static __init
int tracer_init_debugfs(void)
4971 struct dentry
*d_tracer
;
4974 trace_access_lock_init();
4976 d_tracer
= tracing_init_dentry();
4978 trace_create_file("trace_options", 0644, d_tracer
,
4979 NULL
, &tracing_iter_fops
);
4981 trace_create_file("tracing_cpumask", 0644, d_tracer
,
4982 NULL
, &tracing_cpumask_fops
);
4984 trace_create_file("trace", 0644, d_tracer
,
4985 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
4987 trace_create_file("available_tracers", 0444, d_tracer
,
4988 &global_trace
, &show_traces_fops
);
4990 trace_create_file("current_tracer", 0644, d_tracer
,
4991 &global_trace
, &set_tracer_fops
);
4993 #ifdef CONFIG_TRACER_MAX_TRACE
4994 trace_create_file("tracing_max_latency", 0644, d_tracer
,
4995 &tracing_max_latency
, &tracing_max_lat_fops
);
4998 trace_create_file("tracing_thresh", 0644, d_tracer
,
4999 &tracing_thresh
, &tracing_max_lat_fops
);
5001 trace_create_file("README", 0444, d_tracer
,
5002 NULL
, &tracing_readme_fops
);
5004 trace_create_file("trace_pipe", 0444, d_tracer
,
5005 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
5007 trace_create_file("buffer_size_kb", 0644, d_tracer
,
5008 (void *) RING_BUFFER_ALL_CPUS
, &tracing_entries_fops
);
5010 trace_create_file("buffer_total_size_kb", 0444, d_tracer
,
5011 &global_trace
, &tracing_total_entries_fops
);
5013 trace_create_file("free_buffer", 0644, d_tracer
,
5014 &global_trace
, &tracing_free_buffer_fops
);
5016 trace_create_file("trace_marker", 0220, d_tracer
,
5017 NULL
, &tracing_mark_fops
);
5019 trace_create_file("saved_cmdlines", 0444, d_tracer
,
5020 NULL
, &tracing_saved_cmdlines_fops
);
5022 trace_create_file("trace_clock", 0644, d_tracer
, NULL
,
5025 trace_create_file("tracing_on", 0644, d_tracer
,
5026 &global_trace
, &rb_simple_fops
);
5028 #ifdef CONFIG_DYNAMIC_FTRACE
5029 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
5030 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
5033 #ifdef CONFIG_TRACER_SNAPSHOT
5034 trace_create_file("snapshot", 0644, d_tracer
,
5035 (void *) TRACE_PIPE_ALL_CPU
, &snapshot_fops
);
5038 create_trace_options_dir();
5040 for_each_tracing_cpu(cpu
)
5041 tracing_init_debugfs_percpu(cpu
);
5046 static int trace_panic_handler(struct notifier_block
*this,
5047 unsigned long event
, void *unused
)
5049 if (ftrace_dump_on_oops
)
5050 ftrace_dump(ftrace_dump_on_oops
);
5054 static struct notifier_block trace_panic_notifier
= {
5055 .notifier_call
= trace_panic_handler
,
5057 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
5060 static int trace_die_handler(struct notifier_block
*self
,
5066 if (ftrace_dump_on_oops
)
5067 ftrace_dump(ftrace_dump_on_oops
);
5075 static struct notifier_block trace_die_notifier
= {
5076 .notifier_call
= trace_die_handler
,
5081 * printk is set to max of 1024, we really don't need it that big.
5082 * Nothing should be printing 1000 characters anyway.
5084 #define TRACE_MAX_PRINT 1000
5087 * Define here KERN_TRACE so that we have one place to modify
5088 * it if we decide to change what log level the ftrace dump
5091 #define KERN_TRACE KERN_EMERG
5094 trace_printk_seq(struct trace_seq
*s
)
5096 /* Probably should print a warning here. */
5100 /* should be zero ended, but we are paranoid. */
5101 s
->buffer
[s
->len
] = 0;
5103 printk(KERN_TRACE
"%s", s
->buffer
);
5108 void trace_init_global_iter(struct trace_iterator
*iter
)
5110 iter
->tr
= &global_trace
;
5111 iter
->trace
= current_trace
;
5112 iter
->cpu_file
= TRACE_PIPE_ALL_CPU
;
5116 __ftrace_dump(bool disable_tracing
, enum ftrace_dump_mode oops_dump_mode
)
5118 static arch_spinlock_t ftrace_dump_lock
=
5119 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
5120 /* use static because iter can be a bit big for the stack */
5121 static struct trace_iterator iter
;
5122 unsigned int old_userobj
;
5123 static int dump_ran
;
5124 unsigned long flags
;
5128 local_irq_save(flags
);
5129 arch_spin_lock(&ftrace_dump_lock
);
5137 /* Did function tracer already get disabled? */
5138 if (ftrace_is_dead()) {
5139 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
5140 printk("# MAY BE MISSING FUNCTION EVENTS\n");
5143 if (disable_tracing
)
5146 /* Simulate the iterator */
5147 trace_init_global_iter(&iter
);
5149 for_each_tracing_cpu(cpu
) {
5150 atomic_inc(&iter
.tr
->data
[cpu
]->disabled
);
5153 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
5155 /* don't look at user memory in panic mode */
5156 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
5158 switch (oops_dump_mode
) {
5160 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
5163 iter
.cpu_file
= raw_smp_processor_id();
5168 printk(KERN_TRACE
"Bad dumping mode, switching to all CPUs dump\n");
5169 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
5172 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
5175 * We need to stop all tracing on all CPUS to read the
5176 * the next buffer. This is a bit expensive, but is
5177 * not done often. We fill all what we can read,
5178 * and then release the locks again.
5181 while (!trace_empty(&iter
)) {
5184 printk(KERN_TRACE
"---------------------------------\n");
5188 /* reset all but tr, trace, and overruns */
5189 memset(&iter
.seq
, 0,
5190 sizeof(struct trace_iterator
) -
5191 offsetof(struct trace_iterator
, seq
));
5192 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
5195 if (trace_find_next_entry_inc(&iter
) != NULL
) {
5198 ret
= print_trace_line(&iter
);
5199 if (ret
!= TRACE_TYPE_NO_CONSUME
)
5200 trace_consume(&iter
);
5202 touch_nmi_watchdog();
5204 trace_printk_seq(&iter
.seq
);
5208 printk(KERN_TRACE
" (ftrace buffer empty)\n");
5210 printk(KERN_TRACE
"---------------------------------\n");
5213 /* Re-enable tracing if requested */
5214 if (!disable_tracing
) {
5215 trace_flags
|= old_userobj
;
5217 for_each_tracing_cpu(cpu
) {
5218 atomic_dec(&iter
.tr
->data
[cpu
]->disabled
);
5224 arch_spin_unlock(&ftrace_dump_lock
);
5225 local_irq_restore(flags
);
5228 /* By default: disable tracing after the dump */
5229 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode
)
5231 __ftrace_dump(true, oops_dump_mode
);
5233 EXPORT_SYMBOL_GPL(ftrace_dump
);
5235 __init
static int tracer_alloc_buffers(void)
5238 enum ring_buffer_flags rb_flags
;
5243 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
5246 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
5247 goto out_free_buffer_mask
;
5249 /* Only allocate trace_printk buffers if a trace_printk exists */
5250 if (__stop___trace_bprintk_fmt
!= __start___trace_bprintk_fmt
)
5251 /* Must be called before global_trace.buffer is allocated */
5252 trace_printk_init_buffers();
5254 /* To save memory, keep the ring buffer size to its minimum */
5255 if (ring_buffer_expanded
)
5256 ring_buf_size
= trace_buf_size
;
5260 rb_flags
= trace_flags
& TRACE_ITER_OVERWRITE
? RB_FL_OVERWRITE
: 0;
5262 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
5263 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
5265 /* TODO: make the number of buffers hot pluggable with CPUS */
5266 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
, rb_flags
);
5267 if (!global_trace
.buffer
) {
5268 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
5270 goto out_free_cpumask
;
5272 if (global_trace
.buffer_disabled
)
5276 #ifdef CONFIG_TRACER_MAX_TRACE
5277 max_tr
.buffer
= ring_buffer_alloc(1, rb_flags
);
5278 if (!max_tr
.buffer
) {
5279 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
5281 ring_buffer_free(global_trace
.buffer
);
5282 goto out_free_cpumask
;
5286 /* Allocate the first page for all buffers */
5287 for_each_tracing_cpu(i
) {
5288 global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
5289 max_tr
.data
[i
] = &per_cpu(max_tr_data
, i
);
5292 set_buffer_entries(&global_trace
,
5293 ring_buffer_size(global_trace
.buffer
, 0));
5294 #ifdef CONFIG_TRACER_MAX_TRACE
5295 set_buffer_entries(&max_tr
, 1);
5298 trace_init_cmdlines();
5299 init_irq_work(&trace_work_wakeup
, trace_wake_up
);
5301 register_tracer(&nop_trace
);
5302 current_trace
= &nop_trace
;
5303 /* All seems OK, enable tracing */
5304 tracing_disabled
= 0;
5306 atomic_notifier_chain_register(&panic_notifier_list
,
5307 &trace_panic_notifier
);
5309 register_die_notifier(&trace_die_notifier
);
5311 while (trace_boot_options
) {
5314 option
= strsep(&trace_boot_options
, ",");
5315 trace_set_options(option
);
5321 free_cpumask_var(tracing_cpumask
);
5322 out_free_buffer_mask
:
5323 free_cpumask_var(tracing_buffer_mask
);
5328 __init
static int clear_boot_tracer(void)
5331 * The default tracer at boot buffer is an init section.
5332 * This function is called in lateinit. If we did not
5333 * find the boot tracer, then clear it out, to prevent
5334 * later registration from accessing the buffer that is
5335 * about to be freed.
5337 if (!default_bootup_tracer
)
5340 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
5341 default_bootup_tracer
);
5342 default_bootup_tracer
= NULL
;
5347 early_initcall(tracer_alloc_buffers
);
5348 fs_initcall(tracer_init_debugfs
);
5349 late_initcall(clear_boot_tracer
);