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 we disabled the tracer, stop now */
714 if (current_trace
== &nop_trace
)
717 if (WARN_ON_ONCE(!current_trace
->use_max_tr
))
720 arch_spin_lock(&ftrace_max_lock
);
722 tr
->buffer
= max_tr
.buffer
;
725 __update_max_tr(tr
, tsk
, cpu
);
726 arch_spin_unlock(&ftrace_max_lock
);
730 * update_max_tr_single - only copy one trace over, and reset the rest
732 * @tsk - task with the latency
733 * @cpu - the cpu of the buffer to copy.
735 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
738 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
742 if (trace_stop_count
)
745 WARN_ON_ONCE(!irqs_disabled());
746 if (!current_trace
->use_max_tr
) {
751 arch_spin_lock(&ftrace_max_lock
);
753 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
757 * We failed to swap the buffer due to a commit taking
758 * place on this CPU. We fail to record, but we reset
759 * the max trace buffer (no one writes directly to it)
760 * and flag that it failed.
762 trace_array_printk(&max_tr
, _THIS_IP_
,
763 "Failed to swap buffers due to commit in progress\n");
766 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
&& ret
!= -EBUSY
);
768 __update_max_tr(tr
, tsk
, cpu
);
769 arch_spin_unlock(&ftrace_max_lock
);
771 #endif /* CONFIG_TRACER_MAX_TRACE */
773 static void default_wait_pipe(struct trace_iterator
*iter
)
777 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
780 * The events can happen in critical sections where
781 * checking a work queue can cause deadlocks.
782 * After adding a task to the queue, this flag is set
783 * only to notify events to try to wake up the queue
786 * We don't clear it even if the buffer is no longer
787 * empty. The flag only causes the next event to run
788 * irq_work to do the work queue wake up. The worse
789 * that can happen if we race with !trace_empty() is that
790 * an event will cause an irq_work to try to wake up
793 * There's no reason to protect this flag either, as
794 * the work queue and irq_work logic will do the necessary
795 * synchronization for the wake ups. The only thing
796 * that is necessary is that the wake up happens after
797 * a task has been queued. It's OK for spurious wake ups.
799 trace_wakeup_needed
= true;
801 if (trace_empty(iter
))
804 finish_wait(&trace_wait
, &wait
);
808 * register_tracer - register a tracer with the ftrace system.
809 * @type - the plugin for the tracer
811 * Register a new plugin tracer.
813 int register_tracer(struct tracer
*type
)
819 pr_info("Tracer must have a name\n");
823 if (strlen(type
->name
) >= MAX_TRACER_SIZE
) {
824 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE
);
828 mutex_lock(&trace_types_lock
);
830 tracing_selftest_running
= true;
832 for (t
= trace_types
; t
; t
= t
->next
) {
833 if (strcmp(type
->name
, t
->name
) == 0) {
835 pr_info("Tracer %s already registered\n",
843 type
->set_flag
= &dummy_set_flag
;
845 type
->flags
= &dummy_tracer_flags
;
847 if (!type
->flags
->opts
)
848 type
->flags
->opts
= dummy_tracer_opt
;
849 if (!type
->wait_pipe
)
850 type
->wait_pipe
= default_wait_pipe
;
853 #ifdef CONFIG_FTRACE_STARTUP_TEST
854 if (type
->selftest
&& !tracing_selftest_disabled
) {
855 struct tracer
*saved_tracer
= current_trace
;
856 struct trace_array
*tr
= &global_trace
;
859 * Run a selftest on this tracer.
860 * Here we reset the trace buffer, and set the current
861 * tracer to be this tracer. The tracer can then run some
862 * internal tracing to verify that everything is in order.
863 * If we fail, we do not register this tracer.
865 tracing_reset_online_cpus(tr
);
867 current_trace
= type
;
869 /* If we expanded the buffers, make sure the max is expanded too */
870 if (ring_buffer_expanded
&& type
->use_max_tr
)
871 ring_buffer_resize(max_tr
.buffer
, trace_buf_size
,
872 RING_BUFFER_ALL_CPUS
);
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 /* Shrink the max buffer again */
889 if (ring_buffer_expanded
&& type
->use_max_tr
)
890 ring_buffer_resize(max_tr
.buffer
, 1,
891 RING_BUFFER_ALL_CPUS
);
893 printk(KERN_CONT
"PASSED\n");
897 type
->next
= trace_types
;
901 tracing_selftest_running
= false;
902 mutex_unlock(&trace_types_lock
);
904 if (ret
|| !default_bootup_tracer
)
907 if (strncmp(default_bootup_tracer
, type
->name
, MAX_TRACER_SIZE
))
910 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
911 /* Do we want this tracer to start on bootup? */
912 tracing_set_tracer(type
->name
);
913 default_bootup_tracer
= NULL
;
914 /* disable other selftests, since this will break it. */
915 tracing_selftest_disabled
= 1;
916 #ifdef CONFIG_FTRACE_STARTUP_TEST
917 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
925 void tracing_reset(struct trace_array
*tr
, int cpu
)
927 struct ring_buffer
*buffer
= tr
->buffer
;
932 ring_buffer_record_disable(buffer
);
934 /* Make sure all commits have finished */
936 ring_buffer_reset_cpu(buffer
, cpu
);
938 ring_buffer_record_enable(buffer
);
941 void tracing_reset_online_cpus(struct trace_array
*tr
)
943 struct ring_buffer
*buffer
= tr
->buffer
;
949 ring_buffer_record_disable(buffer
);
951 /* Make sure all commits have finished */
954 tr
->time_start
= ftrace_now(tr
->cpu
);
956 for_each_online_cpu(cpu
)
957 ring_buffer_reset_cpu(buffer
, cpu
);
959 ring_buffer_record_enable(buffer
);
962 void tracing_reset_current(int cpu
)
964 tracing_reset(&global_trace
, cpu
);
967 void tracing_reset_current_online_cpus(void)
969 tracing_reset_online_cpus(&global_trace
);
972 #define SAVED_CMDLINES 128
973 #define NO_CMDLINE_MAP UINT_MAX
974 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
975 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
976 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
977 static int cmdline_idx
;
978 static arch_spinlock_t trace_cmdline_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
980 /* temporary disable recording */
981 static atomic_t trace_record_cmdline_disabled __read_mostly
;
983 static void trace_init_cmdlines(void)
985 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
986 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
990 int is_tracing_stopped(void)
992 return trace_stop_count
;
996 * ftrace_off_permanent - disable all ftrace code permanently
998 * This should only be called when a serious anomally has
999 * been detected. This will turn off the function tracing,
1000 * ring buffers, and other tracing utilites. It takes no
1001 * locks and can be called from any context.
1003 void ftrace_off_permanent(void)
1005 tracing_disabled
= 1;
1007 tracing_off_permanent();
1011 * tracing_start - quick start of the tracer
1013 * If tracing is enabled but was stopped by tracing_stop,
1014 * this will start the tracer back up.
1016 void tracing_start(void)
1018 struct ring_buffer
*buffer
;
1019 unsigned long flags
;
1021 if (tracing_disabled
)
1024 raw_spin_lock_irqsave(&tracing_start_lock
, flags
);
1025 if (--trace_stop_count
) {
1026 if (trace_stop_count
< 0) {
1027 /* Someone screwed up their debugging */
1029 trace_stop_count
= 0;
1034 /* Prevent the buffers from switching */
1035 arch_spin_lock(&ftrace_max_lock
);
1037 buffer
= global_trace
.buffer
;
1039 ring_buffer_record_enable(buffer
);
1041 buffer
= max_tr
.buffer
;
1043 ring_buffer_record_enable(buffer
);
1045 arch_spin_unlock(&ftrace_max_lock
);
1049 raw_spin_unlock_irqrestore(&tracing_start_lock
, flags
);
1053 * tracing_stop - quick stop of the tracer
1055 * Light weight way to stop tracing. Use in conjunction with
1058 void tracing_stop(void)
1060 struct ring_buffer
*buffer
;
1061 unsigned long flags
;
1064 raw_spin_lock_irqsave(&tracing_start_lock
, flags
);
1065 if (trace_stop_count
++)
1068 /* Prevent the buffers from switching */
1069 arch_spin_lock(&ftrace_max_lock
);
1071 buffer
= global_trace
.buffer
;
1073 ring_buffer_record_disable(buffer
);
1075 buffer
= max_tr
.buffer
;
1077 ring_buffer_record_disable(buffer
);
1079 arch_spin_unlock(&ftrace_max_lock
);
1082 raw_spin_unlock_irqrestore(&tracing_start_lock
, flags
);
1085 void trace_stop_cmdline_recording(void);
1087 static void trace_save_cmdline(struct task_struct
*tsk
)
1091 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
1095 * It's not the end of the world if we don't get
1096 * the lock, but we also don't want to spin
1097 * nor do we want to disable interrupts,
1098 * so if we miss here, then better luck next time.
1100 if (!arch_spin_trylock(&trace_cmdline_lock
))
1103 idx
= map_pid_to_cmdline
[tsk
->pid
];
1104 if (idx
== NO_CMDLINE_MAP
) {
1105 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
1108 * Check whether the cmdline buffer at idx has a pid
1109 * mapped. We are going to overwrite that entry so we
1110 * need to clear the map_pid_to_cmdline. Otherwise we
1111 * would read the new comm for the old pid.
1113 pid
= map_cmdline_to_pid
[idx
];
1114 if (pid
!= NO_CMDLINE_MAP
)
1115 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
1117 map_cmdline_to_pid
[idx
] = tsk
->pid
;
1118 map_pid_to_cmdline
[tsk
->pid
] = idx
;
1123 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
1125 arch_spin_unlock(&trace_cmdline_lock
);
1128 void trace_find_cmdline(int pid
, char comm
[])
1133 strcpy(comm
, "<idle>");
1137 if (WARN_ON_ONCE(pid
< 0)) {
1138 strcpy(comm
, "<XXX>");
1142 if (pid
> PID_MAX_DEFAULT
) {
1143 strcpy(comm
, "<...>");
1148 arch_spin_lock(&trace_cmdline_lock
);
1149 map
= map_pid_to_cmdline
[pid
];
1150 if (map
!= NO_CMDLINE_MAP
)
1151 strcpy(comm
, saved_cmdlines
[map
]);
1153 strcpy(comm
, "<...>");
1155 arch_spin_unlock(&trace_cmdline_lock
);
1159 void tracing_record_cmdline(struct task_struct
*tsk
)
1161 if (atomic_read(&trace_record_cmdline_disabled
) || !tracing_is_on())
1164 if (!__this_cpu_read(trace_cmdline_save
))
1167 __this_cpu_write(trace_cmdline_save
, false);
1169 trace_save_cmdline(tsk
);
1173 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
1176 struct task_struct
*tsk
= current
;
1178 entry
->preempt_count
= pc
& 0xff;
1179 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
1181 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1182 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
1184 TRACE_FLAG_IRQS_NOSUPPORT
|
1186 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
1187 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
1188 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
1190 EXPORT_SYMBOL_GPL(tracing_generic_entry_update
);
1192 struct ring_buffer_event
*
1193 trace_buffer_lock_reserve(struct ring_buffer
*buffer
,
1196 unsigned long flags
, int pc
)
1198 struct ring_buffer_event
*event
;
1200 event
= ring_buffer_lock_reserve(buffer
, len
);
1201 if (event
!= NULL
) {
1202 struct trace_entry
*ent
= ring_buffer_event_data(event
);
1204 tracing_generic_entry_update(ent
, flags
, pc
);
1212 __buffer_unlock_commit(struct ring_buffer
*buffer
, struct ring_buffer_event
*event
)
1214 __this_cpu_write(trace_cmdline_save
, true);
1215 if (trace_wakeup_needed
) {
1216 trace_wakeup_needed
= false;
1217 /* irq_work_queue() supplies it's own memory barriers */
1218 irq_work_queue(&trace_work_wakeup
);
1220 ring_buffer_unlock_commit(buffer
, event
);
1224 __trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1225 struct ring_buffer_event
*event
,
1226 unsigned long flags
, int pc
)
1228 __buffer_unlock_commit(buffer
, event
);
1230 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1231 ftrace_trace_userstack(buffer
, flags
, pc
);
1234 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1235 struct ring_buffer_event
*event
,
1236 unsigned long flags
, int pc
)
1238 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
);
1240 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit
);
1242 struct ring_buffer_event
*
1243 trace_current_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1244 int type
, unsigned long len
,
1245 unsigned long flags
, int pc
)
1247 *current_rb
= global_trace
.buffer
;
1248 return trace_buffer_lock_reserve(*current_rb
,
1249 type
, len
, flags
, pc
);
1251 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve
);
1253 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
1254 struct ring_buffer_event
*event
,
1255 unsigned long flags
, int pc
)
1257 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
);
1259 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit
);
1261 void trace_buffer_unlock_commit_regs(struct ring_buffer
*buffer
,
1262 struct ring_buffer_event
*event
,
1263 unsigned long flags
, int pc
,
1264 struct pt_regs
*regs
)
1266 __buffer_unlock_commit(buffer
, event
);
1268 ftrace_trace_stack_regs(buffer
, flags
, 0, pc
, regs
);
1269 ftrace_trace_userstack(buffer
, flags
, pc
);
1271 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs
);
1273 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
1274 struct ring_buffer_event
*event
)
1276 ring_buffer_discard_commit(buffer
, event
);
1278 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit
);
1281 trace_function(struct trace_array
*tr
,
1282 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1285 struct ftrace_event_call
*call
= &event_function
;
1286 struct ring_buffer
*buffer
= tr
->buffer
;
1287 struct ring_buffer_event
*event
;
1288 struct ftrace_entry
*entry
;
1290 /* If we are reading the ring buffer, don't trace */
1291 if (unlikely(__this_cpu_read(ftrace_cpu_disabled
)))
1294 event
= trace_buffer_lock_reserve(buffer
, TRACE_FN
, sizeof(*entry
),
1298 entry
= ring_buffer_event_data(event
);
1300 entry
->parent_ip
= parent_ip
;
1302 if (!filter_check_discard(call
, entry
, buffer
, event
))
1303 __buffer_unlock_commit(buffer
, event
);
1307 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1308 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1311 if (likely(!atomic_read(&data
->disabled
)))
1312 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
1315 #ifdef CONFIG_STACKTRACE
1317 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1318 struct ftrace_stack
{
1319 unsigned long calls
[FTRACE_STACK_MAX_ENTRIES
];
1322 static DEFINE_PER_CPU(struct ftrace_stack
, ftrace_stack
);
1323 static DEFINE_PER_CPU(int, ftrace_stack_reserve
);
1325 static void __ftrace_trace_stack(struct ring_buffer
*buffer
,
1326 unsigned long flags
,
1327 int skip
, int pc
, struct pt_regs
*regs
)
1329 struct ftrace_event_call
*call
= &event_kernel_stack
;
1330 struct ring_buffer_event
*event
;
1331 struct stack_entry
*entry
;
1332 struct stack_trace trace
;
1334 int size
= FTRACE_STACK_ENTRIES
;
1336 trace
.nr_entries
= 0;
1340 * Since events can happen in NMIs there's no safe way to
1341 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1342 * or NMI comes in, it will just have to use the default
1343 * FTRACE_STACK_SIZE.
1345 preempt_disable_notrace();
1347 use_stack
= __this_cpu_inc_return(ftrace_stack_reserve
);
1349 * We don't need any atomic variables, just a barrier.
1350 * If an interrupt comes in, we don't care, because it would
1351 * have exited and put the counter back to what we want.
1352 * We just need a barrier to keep gcc from moving things
1356 if (use_stack
== 1) {
1357 trace
.entries
= &__get_cpu_var(ftrace_stack
).calls
[0];
1358 trace
.max_entries
= FTRACE_STACK_MAX_ENTRIES
;
1361 save_stack_trace_regs(regs
, &trace
);
1363 save_stack_trace(&trace
);
1365 if (trace
.nr_entries
> size
)
1366 size
= trace
.nr_entries
;
1368 /* From now on, use_stack is a boolean */
1371 size
*= sizeof(unsigned long);
1373 event
= trace_buffer_lock_reserve(buffer
, TRACE_STACK
,
1374 sizeof(*entry
) + size
, flags
, pc
);
1377 entry
= ring_buffer_event_data(event
);
1379 memset(&entry
->caller
, 0, size
);
1382 memcpy(&entry
->caller
, trace
.entries
,
1383 trace
.nr_entries
* sizeof(unsigned long));
1385 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1386 trace
.entries
= entry
->caller
;
1388 save_stack_trace_regs(regs
, &trace
);
1390 save_stack_trace(&trace
);
1393 entry
->size
= trace
.nr_entries
;
1395 if (!filter_check_discard(call
, entry
, buffer
, event
))
1396 __buffer_unlock_commit(buffer
, event
);
1399 /* Again, don't let gcc optimize things here */
1401 __this_cpu_dec(ftrace_stack_reserve
);
1402 preempt_enable_notrace();
1406 void ftrace_trace_stack_regs(struct ring_buffer
*buffer
, unsigned long flags
,
1407 int skip
, int pc
, struct pt_regs
*regs
)
1409 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1412 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, regs
);
1415 void ftrace_trace_stack(struct ring_buffer
*buffer
, unsigned long flags
,
1418 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1421 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, NULL
);
1424 void __trace_stack(struct trace_array
*tr
, unsigned long flags
, int skip
,
1427 __ftrace_trace_stack(tr
->buffer
, flags
, skip
, pc
, NULL
);
1431 * trace_dump_stack - record a stack back trace in the trace buffer
1433 void trace_dump_stack(void)
1435 unsigned long flags
;
1437 if (tracing_disabled
|| tracing_selftest_running
)
1440 local_save_flags(flags
);
1442 /* skipping 3 traces, seems to get us at the caller of this function */
1443 __ftrace_trace_stack(global_trace
.buffer
, flags
, 3, preempt_count(), NULL
);
1446 static DEFINE_PER_CPU(int, user_stack_count
);
1449 ftrace_trace_userstack(struct ring_buffer
*buffer
, unsigned long flags
, int pc
)
1451 struct ftrace_event_call
*call
= &event_user_stack
;
1452 struct ring_buffer_event
*event
;
1453 struct userstack_entry
*entry
;
1454 struct stack_trace trace
;
1456 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1460 * NMIs can not handle page faults, even with fix ups.
1461 * The save user stack can (and often does) fault.
1463 if (unlikely(in_nmi()))
1467 * prevent recursion, since the user stack tracing may
1468 * trigger other kernel events.
1471 if (__this_cpu_read(user_stack_count
))
1474 __this_cpu_inc(user_stack_count
);
1476 event
= trace_buffer_lock_reserve(buffer
, TRACE_USER_STACK
,
1477 sizeof(*entry
), flags
, pc
);
1479 goto out_drop_count
;
1480 entry
= ring_buffer_event_data(event
);
1482 entry
->tgid
= current
->tgid
;
1483 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1485 trace
.nr_entries
= 0;
1486 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1488 trace
.entries
= entry
->caller
;
1490 save_stack_trace_user(&trace
);
1491 if (!filter_check_discard(call
, entry
, buffer
, event
))
1492 __buffer_unlock_commit(buffer
, event
);
1495 __this_cpu_dec(user_stack_count
);
1501 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1503 ftrace_trace_userstack(tr
, flags
, preempt_count());
1507 #endif /* CONFIG_STACKTRACE */
1509 /* created for use with alloc_percpu */
1510 struct trace_buffer_struct
{
1511 char buffer
[TRACE_BUF_SIZE
];
1514 static struct trace_buffer_struct
*trace_percpu_buffer
;
1515 static struct trace_buffer_struct
*trace_percpu_sirq_buffer
;
1516 static struct trace_buffer_struct
*trace_percpu_irq_buffer
;
1517 static struct trace_buffer_struct
*trace_percpu_nmi_buffer
;
1520 * The buffer used is dependent on the context. There is a per cpu
1521 * buffer for normal context, softirq contex, hard irq context and
1522 * for NMI context. Thise allows for lockless recording.
1524 * Note, if the buffers failed to be allocated, then this returns NULL
1526 static char *get_trace_buf(void)
1528 struct trace_buffer_struct
*percpu_buffer
;
1531 * If we have allocated per cpu buffers, then we do not
1532 * need to do any locking.
1535 percpu_buffer
= trace_percpu_nmi_buffer
;
1537 percpu_buffer
= trace_percpu_irq_buffer
;
1538 else if (in_softirq())
1539 percpu_buffer
= trace_percpu_sirq_buffer
;
1541 percpu_buffer
= trace_percpu_buffer
;
1546 return this_cpu_ptr(&percpu_buffer
->buffer
[0]);
1549 static int alloc_percpu_trace_buffer(void)
1551 struct trace_buffer_struct
*buffers
;
1552 struct trace_buffer_struct
*sirq_buffers
;
1553 struct trace_buffer_struct
*irq_buffers
;
1554 struct trace_buffer_struct
*nmi_buffers
;
1556 buffers
= alloc_percpu(struct trace_buffer_struct
);
1560 sirq_buffers
= alloc_percpu(struct trace_buffer_struct
);
1564 irq_buffers
= alloc_percpu(struct trace_buffer_struct
);
1568 nmi_buffers
= alloc_percpu(struct trace_buffer_struct
);
1572 trace_percpu_buffer
= buffers
;
1573 trace_percpu_sirq_buffer
= sirq_buffers
;
1574 trace_percpu_irq_buffer
= irq_buffers
;
1575 trace_percpu_nmi_buffer
= nmi_buffers
;
1580 free_percpu(irq_buffers
);
1582 free_percpu(sirq_buffers
);
1584 free_percpu(buffers
);
1586 WARN(1, "Could not allocate percpu trace_printk buffer");
1590 static int buffers_allocated
;
1592 void trace_printk_init_buffers(void)
1594 if (buffers_allocated
)
1597 if (alloc_percpu_trace_buffer())
1600 pr_info("ftrace: Allocated trace_printk buffers\n");
1602 /* Expand the buffers to set size */
1603 tracing_update_buffers();
1605 buffers_allocated
= 1;
1608 * trace_printk_init_buffers() can be called by modules.
1609 * If that happens, then we need to start cmdline recording
1610 * directly here. If the global_trace.buffer is already
1611 * allocated here, then this was called by module code.
1613 if (global_trace
.buffer
)
1614 tracing_start_cmdline_record();
1617 void trace_printk_start_comm(void)
1619 /* Start tracing comms if trace printk is set */
1620 if (!buffers_allocated
)
1622 tracing_start_cmdline_record();
1625 static void trace_printk_start_stop_comm(int enabled
)
1627 if (!buffers_allocated
)
1631 tracing_start_cmdline_record();
1633 tracing_stop_cmdline_record();
1637 * trace_vbprintk - write binary msg to tracing buffer
1640 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1642 struct ftrace_event_call
*call
= &event_bprint
;
1643 struct ring_buffer_event
*event
;
1644 struct ring_buffer
*buffer
;
1645 struct trace_array
*tr
= &global_trace
;
1646 struct bprint_entry
*entry
;
1647 unsigned long flags
;
1649 int len
= 0, size
, pc
;
1651 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1654 /* Don't pollute graph traces with trace_vprintk internals */
1655 pause_graph_tracing();
1657 pc
= preempt_count();
1658 preempt_disable_notrace();
1660 tbuffer
= get_trace_buf();
1666 len
= vbin_printf((u32
*)tbuffer
, TRACE_BUF_SIZE
/sizeof(int), fmt
, args
);
1668 if (len
> TRACE_BUF_SIZE
/sizeof(int) || len
< 0)
1671 local_save_flags(flags
);
1672 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1673 buffer
= tr
->buffer
;
1674 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
1678 entry
= ring_buffer_event_data(event
);
1682 memcpy(entry
->buf
, tbuffer
, sizeof(u32
) * len
);
1683 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1684 __buffer_unlock_commit(buffer
, event
);
1685 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1689 preempt_enable_notrace();
1690 unpause_graph_tracing();
1694 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1696 int trace_array_printk(struct trace_array
*tr
,
1697 unsigned long ip
, const char *fmt
, ...)
1702 if (!(trace_flags
& TRACE_ITER_PRINTK
))
1706 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
1711 int trace_array_vprintk(struct trace_array
*tr
,
1712 unsigned long ip
, const char *fmt
, va_list args
)
1714 struct ftrace_event_call
*call
= &event_print
;
1715 struct ring_buffer_event
*event
;
1716 struct ring_buffer
*buffer
;
1717 int len
= 0, size
, pc
;
1718 struct print_entry
*entry
;
1719 unsigned long flags
;
1722 if (tracing_disabled
|| tracing_selftest_running
)
1725 /* Don't pollute graph traces with trace_vprintk internals */
1726 pause_graph_tracing();
1728 pc
= preempt_count();
1729 preempt_disable_notrace();
1732 tbuffer
= get_trace_buf();
1738 len
= vsnprintf(tbuffer
, TRACE_BUF_SIZE
, fmt
, args
);
1739 if (len
> TRACE_BUF_SIZE
)
1742 local_save_flags(flags
);
1743 size
= sizeof(*entry
) + len
+ 1;
1744 buffer
= tr
->buffer
;
1745 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
1749 entry
= ring_buffer_event_data(event
);
1752 memcpy(&entry
->buf
, tbuffer
, len
);
1753 entry
->buf
[len
] = '\0';
1754 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1755 __buffer_unlock_commit(buffer
, event
);
1756 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1759 preempt_enable_notrace();
1760 unpause_graph_tracing();
1765 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1767 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
1769 EXPORT_SYMBOL_GPL(trace_vprintk
);
1771 static void trace_iterator_increment(struct trace_iterator
*iter
)
1773 struct ring_buffer_iter
*buf_iter
= trace_buffer_iter(iter
, iter
->cpu
);
1777 ring_buffer_read(buf_iter
, NULL
);
1780 static struct trace_entry
*
1781 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
,
1782 unsigned long *lost_events
)
1784 struct ring_buffer_event
*event
;
1785 struct ring_buffer_iter
*buf_iter
= trace_buffer_iter(iter
, cpu
);
1788 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1790 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
,
1794 iter
->ent_size
= ring_buffer_event_length(event
);
1795 return ring_buffer_event_data(event
);
1801 static struct trace_entry
*
1802 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
,
1803 unsigned long *missing_events
, u64
*ent_ts
)
1805 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1806 struct trace_entry
*ent
, *next
= NULL
;
1807 unsigned long lost_events
= 0, next_lost
= 0;
1808 int cpu_file
= iter
->cpu_file
;
1809 u64 next_ts
= 0, ts
;
1815 * If we are in a per_cpu trace file, don't bother by iterating over
1816 * all cpu and peek directly.
1818 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1819 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1821 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
, missing_events
);
1823 *ent_cpu
= cpu_file
;
1828 for_each_tracing_cpu(cpu
) {
1830 if (ring_buffer_empty_cpu(buffer
, cpu
))
1833 ent
= peek_next_entry(iter
, cpu
, &ts
, &lost_events
);
1836 * Pick the entry with the smallest timestamp:
1838 if (ent
&& (!next
|| ts
< next_ts
)) {
1842 next_lost
= lost_events
;
1843 next_size
= iter
->ent_size
;
1847 iter
->ent_size
= next_size
;
1850 *ent_cpu
= next_cpu
;
1856 *missing_events
= next_lost
;
1861 /* Find the next real entry, without updating the iterator itself */
1862 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1863 int *ent_cpu
, u64
*ent_ts
)
1865 return __find_next_entry(iter
, ent_cpu
, NULL
, ent_ts
);
1868 /* Find the next real entry, and increment the iterator to the next entry */
1869 void *trace_find_next_entry_inc(struct trace_iterator
*iter
)
1871 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
,
1872 &iter
->lost_events
, &iter
->ts
);
1875 trace_iterator_increment(iter
);
1877 return iter
->ent
? iter
: NULL
;
1880 static void trace_consume(struct trace_iterator
*iter
)
1882 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
,
1883 &iter
->lost_events
);
1886 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1888 struct trace_iterator
*iter
= m
->private;
1892 WARN_ON_ONCE(iter
->leftover
);
1896 /* can't go backwards */
1901 ent
= trace_find_next_entry_inc(iter
);
1905 while (ent
&& iter
->idx
< i
)
1906 ent
= trace_find_next_entry_inc(iter
);
1913 void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
1915 struct trace_array
*tr
= iter
->tr
;
1916 struct ring_buffer_event
*event
;
1917 struct ring_buffer_iter
*buf_iter
;
1918 unsigned long entries
= 0;
1921 tr
->data
[cpu
]->skipped_entries
= 0;
1923 buf_iter
= trace_buffer_iter(iter
, cpu
);
1927 ring_buffer_iter_reset(buf_iter
);
1930 * We could have the case with the max latency tracers
1931 * that a reset never took place on a cpu. This is evident
1932 * by the timestamp being before the start of the buffer.
1934 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
1935 if (ts
>= iter
->tr
->time_start
)
1938 ring_buffer_read(buf_iter
, NULL
);
1941 tr
->data
[cpu
]->skipped_entries
= entries
;
1945 * The current tracer is copied to avoid a global locking
1948 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1950 struct trace_iterator
*iter
= m
->private;
1951 int cpu_file
= iter
->cpu_file
;
1957 * copy the tracer to avoid using a global lock all around.
1958 * iter->trace is a copy of current_trace, the pointer to the
1959 * name may be used instead of a strcmp(), as iter->trace->name
1960 * will point to the same string as current_trace->name.
1962 mutex_lock(&trace_types_lock
);
1963 if (unlikely(current_trace
&& iter
->trace
->name
!= current_trace
->name
))
1964 *iter
->trace
= *current_trace
;
1965 mutex_unlock(&trace_types_lock
);
1967 atomic_inc(&trace_record_cmdline_disabled
);
1969 if (*pos
!= iter
->pos
) {
1974 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1975 for_each_tracing_cpu(cpu
)
1976 tracing_iter_reset(iter
, cpu
);
1978 tracing_iter_reset(iter
, cpu_file
);
1981 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1986 * If we overflowed the seq_file before, then we want
1987 * to just reuse the trace_seq buffer again.
1993 p
= s_next(m
, p
, &l
);
1997 trace_event_read_lock();
1998 trace_access_lock(cpu_file
);
2002 static void s_stop(struct seq_file
*m
, void *p
)
2004 struct trace_iterator
*iter
= m
->private;
2006 atomic_dec(&trace_record_cmdline_disabled
);
2007 trace_access_unlock(iter
->cpu_file
);
2008 trace_event_read_unlock();
2012 get_total_entries(struct trace_array
*tr
, unsigned long *total
, unsigned long *entries
)
2014 unsigned long count
;
2020 for_each_tracing_cpu(cpu
) {
2021 count
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
2023 * If this buffer has skipped entries, then we hold all
2024 * entries for the trace and we need to ignore the
2025 * ones before the time stamp.
2027 if (tr
->data
[cpu
]->skipped_entries
) {
2028 count
-= tr
->data
[cpu
]->skipped_entries
;
2029 /* total is the same as the entries */
2033 ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
2038 static void print_lat_help_header(struct seq_file
*m
)
2040 seq_puts(m
, "# _------=> CPU# \n");
2041 seq_puts(m
, "# / _-----=> irqs-off \n");
2042 seq_puts(m
, "# | / _----=> need-resched \n");
2043 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
2044 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
2045 seq_puts(m
, "# |||| / delay \n");
2046 seq_puts(m
, "# cmd pid ||||| time | caller \n");
2047 seq_puts(m
, "# \\ / ||||| \\ | / \n");
2050 static void print_event_info(struct trace_array
*tr
, struct seq_file
*m
)
2052 unsigned long total
;
2053 unsigned long entries
;
2055 get_total_entries(tr
, &total
, &entries
);
2056 seq_printf(m
, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2057 entries
, total
, num_online_cpus());
2061 static void print_func_help_header(struct trace_array
*tr
, struct seq_file
*m
)
2063 print_event_info(tr
, m
);
2064 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2065 seq_puts(m
, "# | | | | |\n");
2068 static void print_func_help_header_irq(struct trace_array
*tr
, struct seq_file
*m
)
2070 print_event_info(tr
, m
);
2071 seq_puts(m
, "# _-----=> irqs-off\n");
2072 seq_puts(m
, "# / _----=> need-resched\n");
2073 seq_puts(m
, "# | / _---=> hardirq/softirq\n");
2074 seq_puts(m
, "# || / _--=> preempt-depth\n");
2075 seq_puts(m
, "# ||| / delay\n");
2076 seq_puts(m
, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2077 seq_puts(m
, "# | | | |||| | |\n");
2081 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
2083 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2084 struct trace_array
*tr
= iter
->tr
;
2085 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
2086 struct tracer
*type
= current_trace
;
2087 unsigned long entries
;
2088 unsigned long total
;
2089 const char *name
= "preemption";
2094 get_total_entries(tr
, &total
, &entries
);
2096 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
2098 seq_puts(m
, "# -----------------------------------"
2099 "---------------------------------\n");
2100 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2101 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2102 nsecs_to_usecs(data
->saved_latency
),
2106 #if defined(CONFIG_PREEMPT_NONE)
2108 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2110 #elif defined(CONFIG_PREEMPT)
2115 /* These are reserved for later use */
2118 seq_printf(m
, " #P:%d)\n", num_online_cpus());
2122 seq_puts(m
, "# -----------------\n");
2123 seq_printf(m
, "# | task: %.16s-%d "
2124 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2125 data
->comm
, data
->pid
,
2126 from_kuid_munged(seq_user_ns(m
), data
->uid
), data
->nice
,
2127 data
->policy
, data
->rt_priority
);
2128 seq_puts(m
, "# -----------------\n");
2130 if (data
->critical_start
) {
2131 seq_puts(m
, "# => started at: ");
2132 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
2133 trace_print_seq(m
, &iter
->seq
);
2134 seq_puts(m
, "\n# => ended at: ");
2135 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
2136 trace_print_seq(m
, &iter
->seq
);
2137 seq_puts(m
, "\n#\n");
2143 static void test_cpu_buff_start(struct trace_iterator
*iter
)
2145 struct trace_seq
*s
= &iter
->seq
;
2147 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
2150 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
2153 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
2156 if (iter
->tr
->data
[iter
->cpu
]->skipped_entries
)
2159 cpumask_set_cpu(iter
->cpu
, iter
->started
);
2161 /* Don't print started cpu buffer for the first entry of the trace */
2163 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
2167 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
2169 struct trace_seq
*s
= &iter
->seq
;
2170 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2171 struct trace_entry
*entry
;
2172 struct trace_event
*event
;
2176 test_cpu_buff_start(iter
);
2178 event
= ftrace_find_event(entry
->type
);
2180 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2181 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2182 if (!trace_print_lat_context(iter
))
2185 if (!trace_print_context(iter
))
2191 return event
->funcs
->trace(iter
, sym_flags
, event
);
2193 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
2196 return TRACE_TYPE_HANDLED
;
2198 return TRACE_TYPE_PARTIAL_LINE
;
2201 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
2203 struct trace_seq
*s
= &iter
->seq
;
2204 struct trace_entry
*entry
;
2205 struct trace_event
*event
;
2209 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2210 if (!trace_seq_printf(s
, "%d %d %llu ",
2211 entry
->pid
, iter
->cpu
, iter
->ts
))
2215 event
= ftrace_find_event(entry
->type
);
2217 return event
->funcs
->raw(iter
, 0, event
);
2219 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
2222 return TRACE_TYPE_HANDLED
;
2224 return TRACE_TYPE_PARTIAL_LINE
;
2227 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
2229 struct trace_seq
*s
= &iter
->seq
;
2230 unsigned char newline
= '\n';
2231 struct trace_entry
*entry
;
2232 struct trace_event
*event
;
2236 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2237 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
2238 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
2239 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
2242 event
= ftrace_find_event(entry
->type
);
2244 enum print_line_t ret
= event
->funcs
->hex(iter
, 0, event
);
2245 if (ret
!= TRACE_TYPE_HANDLED
)
2249 SEQ_PUT_FIELD_RET(s
, newline
);
2251 return TRACE_TYPE_HANDLED
;
2254 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
2256 struct trace_seq
*s
= &iter
->seq
;
2257 struct trace_entry
*entry
;
2258 struct trace_event
*event
;
2262 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2263 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
2264 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
2265 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
2268 event
= ftrace_find_event(entry
->type
);
2269 return event
? event
->funcs
->binary(iter
, 0, event
) :
2273 int trace_empty(struct trace_iterator
*iter
)
2275 struct ring_buffer_iter
*buf_iter
;
2278 /* If we are looking at one CPU buffer, only check that one */
2279 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
2280 cpu
= iter
->cpu_file
;
2281 buf_iter
= trace_buffer_iter(iter
, cpu
);
2283 if (!ring_buffer_iter_empty(buf_iter
))
2286 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2292 for_each_tracing_cpu(cpu
) {
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
))
2306 /* Called with trace_event_read_lock() held. */
2307 enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
2309 enum print_line_t ret
;
2311 if (iter
->lost_events
&&
2312 !trace_seq_printf(&iter
->seq
, "CPU:%d [LOST %lu EVENTS]\n",
2313 iter
->cpu
, iter
->lost_events
))
2314 return TRACE_TYPE_PARTIAL_LINE
;
2316 if (iter
->trace
&& iter
->trace
->print_line
) {
2317 ret
= iter
->trace
->print_line(iter
);
2318 if (ret
!= TRACE_TYPE_UNHANDLED
)
2322 if (iter
->ent
->type
== TRACE_BPRINT
&&
2323 trace_flags
& TRACE_ITER_PRINTK
&&
2324 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2325 return trace_print_bprintk_msg_only(iter
);
2327 if (iter
->ent
->type
== TRACE_PRINT
&&
2328 trace_flags
& TRACE_ITER_PRINTK
&&
2329 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2330 return trace_print_printk_msg_only(iter
);
2332 if (trace_flags
& TRACE_ITER_BIN
)
2333 return print_bin_fmt(iter
);
2335 if (trace_flags
& TRACE_ITER_HEX
)
2336 return print_hex_fmt(iter
);
2338 if (trace_flags
& TRACE_ITER_RAW
)
2339 return print_raw_fmt(iter
);
2341 return print_trace_fmt(iter
);
2344 void trace_latency_header(struct seq_file
*m
)
2346 struct trace_iterator
*iter
= m
->private;
2348 /* print nothing if the buffers are empty */
2349 if (trace_empty(iter
))
2352 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
2353 print_trace_header(m
, iter
);
2355 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2356 print_lat_help_header(m
);
2359 void trace_default_header(struct seq_file
*m
)
2361 struct trace_iterator
*iter
= m
->private;
2363 if (!(trace_flags
& TRACE_ITER_CONTEXT_INFO
))
2366 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2367 /* print nothing if the buffers are empty */
2368 if (trace_empty(iter
))
2370 print_trace_header(m
, iter
);
2371 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2372 print_lat_help_header(m
);
2374 if (!(trace_flags
& TRACE_ITER_VERBOSE
)) {
2375 if (trace_flags
& TRACE_ITER_IRQ_INFO
)
2376 print_func_help_header_irq(iter
->tr
, m
);
2378 print_func_help_header(iter
->tr
, m
);
2383 static void test_ftrace_alive(struct seq_file
*m
)
2385 if (!ftrace_is_dead())
2387 seq_printf(m
, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2388 seq_printf(m
, "# MAY BE MISSING FUNCTION EVENTS\n");
2391 static int s_show(struct seq_file
*m
, void *v
)
2393 struct trace_iterator
*iter
= v
;
2396 if (iter
->ent
== NULL
) {
2398 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
2400 test_ftrace_alive(m
);
2402 if (iter
->trace
&& iter
->trace
->print_header
)
2403 iter
->trace
->print_header(m
);
2405 trace_default_header(m
);
2407 } else if (iter
->leftover
) {
2409 * If we filled the seq_file buffer earlier, we
2410 * want to just show it now.
2412 ret
= trace_print_seq(m
, &iter
->seq
);
2414 /* ret should this time be zero, but you never know */
2415 iter
->leftover
= ret
;
2418 print_trace_line(iter
);
2419 ret
= trace_print_seq(m
, &iter
->seq
);
2421 * If we overflow the seq_file buffer, then it will
2422 * ask us for this data again at start up.
2424 * ret is 0 if seq_file write succeeded.
2427 iter
->leftover
= ret
;
2433 static const struct seq_operations tracer_seq_ops
= {
2440 static struct trace_iterator
*
2441 __tracing_open(struct inode
*inode
, struct file
*file
)
2443 long cpu_file
= (long) inode
->i_private
;
2444 struct trace_iterator
*iter
;
2447 if (tracing_disabled
)
2448 return ERR_PTR(-ENODEV
);
2450 iter
= __seq_open_private(file
, &tracer_seq_ops
, sizeof(*iter
));
2452 return ERR_PTR(-ENOMEM
);
2454 iter
->buffer_iter
= kzalloc(sizeof(*iter
->buffer_iter
) * num_possible_cpus(),
2456 if (!iter
->buffer_iter
)
2460 * We make a copy of the current tracer to avoid concurrent
2461 * changes on it while we are reading.
2463 mutex_lock(&trace_types_lock
);
2464 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2469 *iter
->trace
= *current_trace
;
2471 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2474 if (current_trace
&& current_trace
->print_max
)
2477 iter
->tr
= &global_trace
;
2479 mutex_init(&iter
->mutex
);
2480 iter
->cpu_file
= cpu_file
;
2482 /* Notify the tracer early; before we stop tracing. */
2483 if (iter
->trace
&& iter
->trace
->open
)
2484 iter
->trace
->open(iter
);
2486 /* Annotate start of buffers if we had overruns */
2487 if (ring_buffer_overruns(iter
->tr
->buffer
))
2488 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2490 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2491 if (trace_clocks
[trace_clock_id
].in_ns
)
2492 iter
->iter_flags
|= TRACE_FILE_TIME_IN_NS
;
2494 /* stop the trace while dumping */
2497 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
2498 for_each_tracing_cpu(cpu
) {
2499 iter
->buffer_iter
[cpu
] =
2500 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2502 ring_buffer_read_prepare_sync();
2503 for_each_tracing_cpu(cpu
) {
2504 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2505 tracing_iter_reset(iter
, cpu
);
2508 cpu
= iter
->cpu_file
;
2509 iter
->buffer_iter
[cpu
] =
2510 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2511 ring_buffer_read_prepare_sync();
2512 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2513 tracing_iter_reset(iter
, cpu
);
2516 mutex_unlock(&trace_types_lock
);
2521 mutex_unlock(&trace_types_lock
);
2523 kfree(iter
->buffer_iter
);
2525 seq_release_private(inode
, file
);
2526 return ERR_PTR(-ENOMEM
);
2529 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2531 if (tracing_disabled
)
2534 filp
->private_data
= inode
->i_private
;
2538 static int tracing_release(struct inode
*inode
, struct file
*file
)
2540 struct seq_file
*m
= file
->private_data
;
2541 struct trace_iterator
*iter
;
2544 if (!(file
->f_mode
& FMODE_READ
))
2549 mutex_lock(&trace_types_lock
);
2550 for_each_tracing_cpu(cpu
) {
2551 if (iter
->buffer_iter
[cpu
])
2552 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2555 if (iter
->trace
&& iter
->trace
->close
)
2556 iter
->trace
->close(iter
);
2558 /* reenable tracing if it was previously enabled */
2560 mutex_unlock(&trace_types_lock
);
2562 mutex_destroy(&iter
->mutex
);
2563 free_cpumask_var(iter
->started
);
2565 kfree(iter
->buffer_iter
);
2566 seq_release_private(inode
, file
);
2570 static int tracing_open(struct inode
*inode
, struct file
*file
)
2572 struct trace_iterator
*iter
;
2575 /* If this file was open for write, then erase contents */
2576 if ((file
->f_mode
& FMODE_WRITE
) &&
2577 (file
->f_flags
& O_TRUNC
)) {
2578 long cpu
= (long) inode
->i_private
;
2580 if (cpu
== TRACE_PIPE_ALL_CPU
)
2581 tracing_reset_online_cpus(&global_trace
);
2583 tracing_reset(&global_trace
, cpu
);
2586 if (file
->f_mode
& FMODE_READ
) {
2587 iter
= __tracing_open(inode
, file
);
2589 ret
= PTR_ERR(iter
);
2590 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2591 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2597 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2599 struct tracer
*t
= v
;
2609 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2614 mutex_lock(&trace_types_lock
);
2615 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2621 static void t_stop(struct seq_file
*m
, void *p
)
2623 mutex_unlock(&trace_types_lock
);
2626 static int t_show(struct seq_file
*m
, void *v
)
2628 struct tracer
*t
= v
;
2633 seq_printf(m
, "%s", t
->name
);
2642 static const struct seq_operations show_traces_seq_ops
= {
2649 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2651 if (tracing_disabled
)
2654 return seq_open(file
, &show_traces_seq_ops
);
2658 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2659 size_t count
, loff_t
*ppos
)
2664 static loff_t
tracing_seek(struct file
*file
, loff_t offset
, int origin
)
2666 if (file
->f_mode
& FMODE_READ
)
2667 return seq_lseek(file
, offset
, origin
);
2672 static const struct file_operations tracing_fops
= {
2673 .open
= tracing_open
,
2675 .write
= tracing_write_stub
,
2676 .llseek
= tracing_seek
,
2677 .release
= tracing_release
,
2680 static const struct file_operations show_traces_fops
= {
2681 .open
= show_traces_open
,
2683 .release
= seq_release
,
2684 .llseek
= seq_lseek
,
2688 * Only trace on a CPU if the bitmask is set:
2690 static cpumask_var_t tracing_cpumask
;
2693 * The tracer itself will not take this lock, but still we want
2694 * to provide a consistent cpumask to user-space:
2696 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2699 * Temporary storage for the character representation of the
2700 * CPU bitmask (and one more byte for the newline):
2702 static char mask_str
[NR_CPUS
+ 1];
2705 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2706 size_t count
, loff_t
*ppos
)
2710 mutex_lock(&tracing_cpumask_update_lock
);
2712 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2713 if (count
- len
< 2) {
2717 len
+= sprintf(mask_str
+ len
, "\n");
2718 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2721 mutex_unlock(&tracing_cpumask_update_lock
);
2727 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2728 size_t count
, loff_t
*ppos
)
2731 cpumask_var_t tracing_cpumask_new
;
2733 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2736 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2740 mutex_lock(&tracing_cpumask_update_lock
);
2742 local_irq_disable();
2743 arch_spin_lock(&ftrace_max_lock
);
2744 for_each_tracing_cpu(cpu
) {
2746 * Increase/decrease the disabled counter if we are
2747 * about to flip a bit in the cpumask:
2749 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2750 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2751 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2752 ring_buffer_record_disable_cpu(global_trace
.buffer
, cpu
);
2754 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2755 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2756 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2757 ring_buffer_record_enable_cpu(global_trace
.buffer
, cpu
);
2760 arch_spin_unlock(&ftrace_max_lock
);
2763 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2765 mutex_unlock(&tracing_cpumask_update_lock
);
2766 free_cpumask_var(tracing_cpumask_new
);
2771 free_cpumask_var(tracing_cpumask_new
);
2776 static const struct file_operations tracing_cpumask_fops
= {
2777 .open
= tracing_open_generic
,
2778 .read
= tracing_cpumask_read
,
2779 .write
= tracing_cpumask_write
,
2780 .llseek
= generic_file_llseek
,
2783 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
2785 struct tracer_opt
*trace_opts
;
2789 mutex_lock(&trace_types_lock
);
2790 tracer_flags
= current_trace
->flags
->val
;
2791 trace_opts
= current_trace
->flags
->opts
;
2793 for (i
= 0; trace_options
[i
]; i
++) {
2794 if (trace_flags
& (1 << i
))
2795 seq_printf(m
, "%s\n", trace_options
[i
]);
2797 seq_printf(m
, "no%s\n", trace_options
[i
]);
2800 for (i
= 0; trace_opts
[i
].name
; i
++) {
2801 if (tracer_flags
& trace_opts
[i
].bit
)
2802 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
2804 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
2806 mutex_unlock(&trace_types_lock
);
2811 static int __set_tracer_option(struct tracer
*trace
,
2812 struct tracer_flags
*tracer_flags
,
2813 struct tracer_opt
*opts
, int neg
)
2817 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
2822 tracer_flags
->val
&= ~opts
->bit
;
2824 tracer_flags
->val
|= opts
->bit
;
2828 /* Try to assign a tracer specific option */
2829 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2831 struct tracer_flags
*tracer_flags
= trace
->flags
;
2832 struct tracer_opt
*opts
= NULL
;
2835 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
2836 opts
= &tracer_flags
->opts
[i
];
2838 if (strcmp(cmp
, opts
->name
) == 0)
2839 return __set_tracer_option(trace
, trace
->flags
,
2846 static void set_tracer_flags(unsigned int mask
, int enabled
)
2848 /* do nothing if flag is already set */
2849 if (!!(trace_flags
& mask
) == !!enabled
)
2853 trace_flags
|= mask
;
2855 trace_flags
&= ~mask
;
2857 if (mask
== TRACE_ITER_RECORD_CMD
)
2858 trace_event_enable_cmd_record(enabled
);
2860 if (mask
== TRACE_ITER_OVERWRITE
)
2861 ring_buffer_change_overwrite(global_trace
.buffer
, enabled
);
2863 if (mask
== TRACE_ITER_PRINTK
)
2864 trace_printk_start_stop_comm(enabled
);
2867 static int trace_set_options(char *option
)
2874 cmp
= strstrip(option
);
2876 if (strncmp(cmp
, "no", 2) == 0) {
2881 for (i
= 0; trace_options
[i
]; i
++) {
2882 if (strcmp(cmp
, trace_options
[i
]) == 0) {
2883 set_tracer_flags(1 << i
, !neg
);
2888 /* If no option could be set, test the specific tracer options */
2889 if (!trace_options
[i
]) {
2890 mutex_lock(&trace_types_lock
);
2891 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2892 mutex_unlock(&trace_types_lock
);
2899 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2900 size_t cnt
, loff_t
*ppos
)
2904 if (cnt
>= sizeof(buf
))
2907 if (copy_from_user(&buf
, ubuf
, cnt
))
2912 trace_set_options(buf
);
2919 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
2921 if (tracing_disabled
)
2923 return single_open(file
, tracing_trace_options_show
, NULL
);
2926 static const struct file_operations tracing_iter_fops
= {
2927 .open
= tracing_trace_options_open
,
2929 .llseek
= seq_lseek
,
2930 .release
= single_release
,
2931 .write
= tracing_trace_options_write
,
2934 static const char readme_msg
[] =
2935 "tracing mini-HOWTO:\n\n"
2936 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2937 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2938 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2939 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2941 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2942 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2944 "# cat /sys/kernel/debug/tracing/trace_options\n"
2945 "noprint-parent nosym-offset nosym-addr noverbose\n"
2946 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2947 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2948 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2949 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2953 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2954 size_t cnt
, loff_t
*ppos
)
2956 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2957 readme_msg
, strlen(readme_msg
));
2960 static const struct file_operations tracing_readme_fops
= {
2961 .open
= tracing_open_generic
,
2962 .read
= tracing_readme_read
,
2963 .llseek
= generic_file_llseek
,
2967 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
2968 size_t cnt
, loff_t
*ppos
)
2977 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
2981 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
2989 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
2992 pid
= map_cmdline_to_pid
[i
];
2993 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
2996 trace_find_cmdline(pid
, buf_comm
);
2997 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
3002 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
3011 static const struct file_operations tracing_saved_cmdlines_fops
= {
3012 .open
= tracing_open_generic
,
3013 .read
= tracing_saved_cmdlines_read
,
3014 .llseek
= generic_file_llseek
,
3018 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
3019 size_t cnt
, loff_t
*ppos
)
3021 char buf
[MAX_TRACER_SIZE
+2];
3024 mutex_lock(&trace_types_lock
);
3026 r
= sprintf(buf
, "%s\n", current_trace
->name
);
3028 r
= sprintf(buf
, "\n");
3029 mutex_unlock(&trace_types_lock
);
3031 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3034 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
3036 tracing_reset_online_cpus(tr
);
3040 static void set_buffer_entries(struct trace_array
*tr
, unsigned long val
)
3043 for_each_tracing_cpu(cpu
)
3044 tr
->data
[cpu
]->entries
= val
;
3047 /* resize @tr's buffer to the size of @size_tr's entries */
3048 static int resize_buffer_duplicate_size(struct trace_array
*tr
,
3049 struct trace_array
*size_tr
, int cpu_id
)
3053 if (cpu_id
== RING_BUFFER_ALL_CPUS
) {
3054 for_each_tracing_cpu(cpu
) {
3055 ret
= ring_buffer_resize(tr
->buffer
,
3056 size_tr
->data
[cpu
]->entries
, cpu
);
3059 tr
->data
[cpu
]->entries
= size_tr
->data
[cpu
]->entries
;
3062 ret
= ring_buffer_resize(tr
->buffer
,
3063 size_tr
->data
[cpu_id
]->entries
, cpu_id
);
3065 tr
->data
[cpu_id
]->entries
=
3066 size_tr
->data
[cpu_id
]->entries
;
3072 static int __tracing_resize_ring_buffer(unsigned long size
, int cpu
)
3077 * If kernel or user changes the size of the ring buffer
3078 * we use the size that was given, and we can forget about
3079 * expanding it later.
3081 ring_buffer_expanded
= 1;
3083 /* May be called before buffers are initialized */
3084 if (!global_trace
.buffer
)
3087 ret
= ring_buffer_resize(global_trace
.buffer
, size
, cpu
);
3091 if (!current_trace
->use_max_tr
)
3094 ret
= ring_buffer_resize(max_tr
.buffer
, size
, cpu
);
3096 int r
= resize_buffer_duplicate_size(&global_trace
,
3097 &global_trace
, cpu
);
3100 * AARGH! We are left with different
3101 * size max buffer!!!!
3102 * The max buffer is our "snapshot" buffer.
3103 * When a tracer needs a snapshot (one of the
3104 * latency tracers), it swaps the max buffer
3105 * with the saved snap shot. We succeeded to
3106 * update the size of the main buffer, but failed to
3107 * update the size of the max buffer. But when we tried
3108 * to reset the main buffer to the original size, we
3109 * failed there too. This is very unlikely to
3110 * happen, but if it does, warn and kill all
3114 tracing_disabled
= 1;
3119 if (cpu
== RING_BUFFER_ALL_CPUS
)
3120 set_buffer_entries(&max_tr
, size
);
3122 max_tr
.data
[cpu
]->entries
= size
;
3125 if (cpu
== RING_BUFFER_ALL_CPUS
)
3126 set_buffer_entries(&global_trace
, size
);
3128 global_trace
.data
[cpu
]->entries
= size
;
3133 static ssize_t
tracing_resize_ring_buffer(unsigned long size
, int cpu_id
)
3137 mutex_lock(&trace_types_lock
);
3139 if (cpu_id
!= RING_BUFFER_ALL_CPUS
) {
3140 /* make sure, this cpu is enabled in the mask */
3141 if (!cpumask_test_cpu(cpu_id
, tracing_buffer_mask
)) {
3147 ret
= __tracing_resize_ring_buffer(size
, cpu_id
);
3152 mutex_unlock(&trace_types_lock
);
3159 * tracing_update_buffers - used by tracing facility to expand ring buffers
3161 * To save on memory when the tracing is never used on a system with it
3162 * configured in. The ring buffers are set to a minimum size. But once
3163 * a user starts to use the tracing facility, then they need to grow
3164 * to their default size.
3166 * This function is to be called when a tracer is about to be used.
3168 int tracing_update_buffers(void)
3172 mutex_lock(&trace_types_lock
);
3173 if (!ring_buffer_expanded
)
3174 ret
= __tracing_resize_ring_buffer(trace_buf_size
,
3175 RING_BUFFER_ALL_CPUS
);
3176 mutex_unlock(&trace_types_lock
);
3181 struct trace_option_dentry
;
3183 static struct trace_option_dentry
*
3184 create_trace_option_files(struct tracer
*tracer
);
3187 destroy_trace_option_files(struct trace_option_dentry
*topts
);
3189 static int tracing_set_tracer(const char *buf
)
3191 static struct trace_option_dentry
*topts
;
3192 struct trace_array
*tr
= &global_trace
;
3197 mutex_lock(&trace_types_lock
);
3199 if (!ring_buffer_expanded
) {
3200 ret
= __tracing_resize_ring_buffer(trace_buf_size
,
3201 RING_BUFFER_ALL_CPUS
);
3207 for (t
= trace_types
; t
; t
= t
->next
) {
3208 if (strcmp(t
->name
, buf
) == 0)
3215 if (t
== current_trace
)
3218 trace_branch_disable();
3219 if (current_trace
&& current_trace
->reset
)
3220 current_trace
->reset(tr
);
3222 had_max_tr
= current_trace
&& current_trace
->use_max_tr
;
3223 current_trace
= &nop_trace
;
3225 if (had_max_tr
&& !t
->use_max_tr
) {
3227 * We need to make sure that the update_max_tr sees that
3228 * current_trace changed to nop_trace to keep it from
3229 * swapping the buffers after we resize it.
3230 * The update_max_tr is called from interrupts disabled
3231 * so a synchronized_sched() is sufficient.
3233 synchronize_sched();
3235 * We don't free the ring buffer. instead, resize it because
3236 * The max_tr ring buffer has some state (e.g. ring->clock) and
3237 * we want preserve it.
3239 ring_buffer_resize(max_tr
.buffer
, 1, RING_BUFFER_ALL_CPUS
);
3240 set_buffer_entries(&max_tr
, 1);
3242 destroy_trace_option_files(topts
);
3244 topts
= create_trace_option_files(t
);
3245 if (t
->use_max_tr
&& !had_max_tr
) {
3246 /* we need to make per cpu buffer sizes equivalent */
3247 ret
= resize_buffer_duplicate_size(&max_tr
, &global_trace
,
3248 RING_BUFFER_ALL_CPUS
);
3254 ret
= tracer_init(t
, tr
);
3260 trace_branch_enable(tr
);
3262 mutex_unlock(&trace_types_lock
);
3268 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
3269 size_t cnt
, loff_t
*ppos
)
3271 char buf
[MAX_TRACER_SIZE
+1];
3278 if (cnt
> MAX_TRACER_SIZE
)
3279 cnt
= MAX_TRACER_SIZE
;
3281 if (copy_from_user(&buf
, ubuf
, cnt
))
3286 /* strip ending whitespace. */
3287 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
3290 err
= tracing_set_tracer(buf
);
3300 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
3301 size_t cnt
, loff_t
*ppos
)
3303 unsigned long *ptr
= filp
->private_data
;
3307 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
3308 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
3309 if (r
> sizeof(buf
))
3311 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3315 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
3316 size_t cnt
, loff_t
*ppos
)
3318 unsigned long *ptr
= filp
->private_data
;
3322 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3331 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
3333 long cpu_file
= (long) inode
->i_private
;
3334 struct trace_iterator
*iter
;
3337 if (tracing_disabled
)
3340 mutex_lock(&trace_types_lock
);
3342 /* create a buffer to store the information to pass to userspace */
3343 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
3350 * We make a copy of the current tracer to avoid concurrent
3351 * changes on it while we are reading.
3353 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
3359 *iter
->trace
= *current_trace
;
3361 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
3366 /* trace pipe does not show start of buffer */
3367 cpumask_setall(iter
->started
);
3369 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
3370 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
3372 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3373 if (trace_clocks
[trace_clock_id
].in_ns
)
3374 iter
->iter_flags
|= TRACE_FILE_TIME_IN_NS
;
3376 iter
->cpu_file
= cpu_file
;
3377 iter
->tr
= &global_trace
;
3378 mutex_init(&iter
->mutex
);
3379 filp
->private_data
= iter
;
3381 if (iter
->trace
->pipe_open
)
3382 iter
->trace
->pipe_open(iter
);
3384 nonseekable_open(inode
, filp
);
3386 mutex_unlock(&trace_types_lock
);
3392 mutex_unlock(&trace_types_lock
);
3396 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
3398 struct trace_iterator
*iter
= file
->private_data
;
3400 mutex_lock(&trace_types_lock
);
3402 if (iter
->trace
->pipe_close
)
3403 iter
->trace
->pipe_close(iter
);
3405 mutex_unlock(&trace_types_lock
);
3407 free_cpumask_var(iter
->started
);
3408 mutex_destroy(&iter
->mutex
);
3416 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
3418 struct trace_iterator
*iter
= filp
->private_data
;
3420 if (trace_flags
& TRACE_ITER_BLOCK
) {
3422 * Always select as readable when in blocking mode
3424 return POLLIN
| POLLRDNORM
;
3426 if (!trace_empty(iter
))
3427 return POLLIN
| POLLRDNORM
;
3428 poll_wait(filp
, &trace_wait
, poll_table
);
3429 if (!trace_empty(iter
))
3430 return POLLIN
| POLLRDNORM
;
3437 * This is a make-shift waitqueue.
3438 * A tracer might use this callback on some rare cases:
3440 * 1) the current tracer might hold the runqueue lock when it wakes up
3441 * a reader, hence a deadlock (sched, function, and function graph tracers)
3442 * 2) the function tracers, trace all functions, we don't want
3443 * the overhead of calling wake_up and friends
3444 * (and tracing them too)
3446 * Anyway, this is really very primitive wakeup.
3448 void poll_wait_pipe(struct trace_iterator
*iter
)
3450 set_current_state(TASK_INTERRUPTIBLE
);
3451 /* sleep for 100 msecs, and try again. */
3452 schedule_timeout(HZ
/ 10);
3455 /* Must be called with trace_types_lock mutex held. */
3456 static int tracing_wait_pipe(struct file
*filp
)
3458 struct trace_iterator
*iter
= filp
->private_data
;
3460 while (trace_empty(iter
)) {
3462 if ((filp
->f_flags
& O_NONBLOCK
)) {
3466 mutex_unlock(&iter
->mutex
);
3468 iter
->trace
->wait_pipe(iter
);
3470 mutex_lock(&iter
->mutex
);
3472 if (signal_pending(current
))
3476 * We block until we read something and tracing is disabled.
3477 * We still block if tracing is disabled, but we have never
3478 * read anything. This allows a user to cat this file, and
3479 * then enable tracing. But after we have read something,
3480 * we give an EOF when tracing is again disabled.
3482 * iter->pos will be 0 if we haven't read anything.
3484 if (!tracing_is_enabled() && iter
->pos
)
3495 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
3496 size_t cnt
, loff_t
*ppos
)
3498 struct trace_iterator
*iter
= filp
->private_data
;
3501 /* return any leftover data */
3502 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3506 trace_seq_init(&iter
->seq
);
3508 /* copy the tracer to avoid using a global lock all around */
3509 mutex_lock(&trace_types_lock
);
3510 if (unlikely(current_trace
&& iter
->trace
->name
!= current_trace
->name
))
3511 *iter
->trace
= *current_trace
;
3512 mutex_unlock(&trace_types_lock
);
3515 * Avoid more than one consumer on a single file descriptor
3516 * This is just a matter of traces coherency, the ring buffer itself
3519 mutex_lock(&iter
->mutex
);
3520 if (iter
->trace
->read
) {
3521 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
3527 sret
= tracing_wait_pipe(filp
);
3531 /* stop when tracing is finished */
3532 if (trace_empty(iter
)) {
3537 if (cnt
>= PAGE_SIZE
)
3538 cnt
= PAGE_SIZE
- 1;
3540 /* reset all but tr, trace, and overruns */
3541 memset(&iter
->seq
, 0,
3542 sizeof(struct trace_iterator
) -
3543 offsetof(struct trace_iterator
, seq
));
3546 trace_event_read_lock();
3547 trace_access_lock(iter
->cpu_file
);
3548 while (trace_find_next_entry_inc(iter
) != NULL
) {
3549 enum print_line_t ret
;
3550 int len
= iter
->seq
.len
;
3552 ret
= print_trace_line(iter
);
3553 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3554 /* don't print partial lines */
3555 iter
->seq
.len
= len
;
3558 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3559 trace_consume(iter
);
3561 if (iter
->seq
.len
>= cnt
)
3565 * Setting the full flag means we reached the trace_seq buffer
3566 * size and we should leave by partial output condition above.
3567 * One of the trace_seq_* functions is not used properly.
3569 WARN_ONCE(iter
->seq
.full
, "full flag set for trace type %d",
3572 trace_access_unlock(iter
->cpu_file
);
3573 trace_event_read_unlock();
3575 /* Now copy what we have to the user */
3576 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3577 if (iter
->seq
.readpos
>= iter
->seq
.len
)
3578 trace_seq_init(&iter
->seq
);
3581 * If there was nothing to send to user, in spite of consuming trace
3582 * entries, go back to wait for more entries.
3588 mutex_unlock(&iter
->mutex
);
3593 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
3594 struct pipe_buffer
*buf
)
3596 __free_page(buf
->page
);
3599 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
3602 __free_page(spd
->pages
[idx
]);
3605 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
3607 .map
= generic_pipe_buf_map
,
3608 .unmap
= generic_pipe_buf_unmap
,
3609 .confirm
= generic_pipe_buf_confirm
,
3610 .release
= tracing_pipe_buf_release
,
3611 .steal
= generic_pipe_buf_steal
,
3612 .get
= generic_pipe_buf_get
,
3616 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
3621 /* Seq buffer is page-sized, exactly what we need. */
3623 count
= iter
->seq
.len
;
3624 ret
= print_trace_line(iter
);
3625 count
= iter
->seq
.len
- count
;
3628 iter
->seq
.len
-= count
;
3631 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3632 iter
->seq
.len
-= count
;
3636 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3637 trace_consume(iter
);
3639 if (!trace_find_next_entry_inc(iter
)) {
3649 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3651 struct pipe_inode_info
*pipe
,
3655 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3656 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3657 struct trace_iterator
*iter
= filp
->private_data
;
3658 struct splice_pipe_desc spd
= {
3660 .partial
= partial_def
,
3661 .nr_pages
= 0, /* This gets updated below. */
3662 .nr_pages_max
= PIPE_DEF_BUFFERS
,
3664 .ops
= &tracing_pipe_buf_ops
,
3665 .spd_release
= tracing_spd_release_pipe
,
3671 if (splice_grow_spd(pipe
, &spd
))
3674 /* copy the tracer to avoid using a global lock all around */
3675 mutex_lock(&trace_types_lock
);
3676 if (unlikely(current_trace
&& iter
->trace
->name
!= current_trace
->name
))
3677 *iter
->trace
= *current_trace
;
3678 mutex_unlock(&trace_types_lock
);
3680 mutex_lock(&iter
->mutex
);
3682 if (iter
->trace
->splice_read
) {
3683 ret
= iter
->trace
->splice_read(iter
, filp
,
3684 ppos
, pipe
, len
, flags
);
3689 ret
= tracing_wait_pipe(filp
);
3693 if (!iter
->ent
&& !trace_find_next_entry_inc(iter
)) {
3698 trace_event_read_lock();
3699 trace_access_lock(iter
->cpu_file
);
3701 /* Fill as many pages as possible. */
3702 for (i
= 0, rem
= len
; i
< pipe
->buffers
&& rem
; i
++) {
3703 spd
.pages
[i
] = alloc_page(GFP_KERNEL
);
3707 rem
= tracing_fill_pipe_page(rem
, iter
);
3709 /* Copy the data into the page, so we can start over. */
3710 ret
= trace_seq_to_buffer(&iter
->seq
,
3711 page_address(spd
.pages
[i
]),
3714 __free_page(spd
.pages
[i
]);
3717 spd
.partial
[i
].offset
= 0;
3718 spd
.partial
[i
].len
= iter
->seq
.len
;
3720 trace_seq_init(&iter
->seq
);
3723 trace_access_unlock(iter
->cpu_file
);
3724 trace_event_read_unlock();
3725 mutex_unlock(&iter
->mutex
);
3729 ret
= splice_to_pipe(pipe
, &spd
);
3731 splice_shrink_spd(&spd
);
3735 mutex_unlock(&iter
->mutex
);
3739 struct ftrace_entries_info
{
3740 struct trace_array
*tr
;
3744 static int tracing_entries_open(struct inode
*inode
, struct file
*filp
)
3746 struct ftrace_entries_info
*info
;
3748 if (tracing_disabled
)
3751 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3755 info
->tr
= &global_trace
;
3756 info
->cpu
= (unsigned long)inode
->i_private
;
3758 filp
->private_data
= info
;
3764 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3765 size_t cnt
, loff_t
*ppos
)
3767 struct ftrace_entries_info
*info
= filp
->private_data
;
3768 struct trace_array
*tr
= info
->tr
;
3773 mutex_lock(&trace_types_lock
);
3775 if (info
->cpu
== RING_BUFFER_ALL_CPUS
) {
3776 int cpu
, buf_size_same
;
3781 /* check if all cpu sizes are same */
3782 for_each_tracing_cpu(cpu
) {
3783 /* fill in the size from first enabled cpu */
3785 size
= tr
->data
[cpu
]->entries
;
3786 if (size
!= tr
->data
[cpu
]->entries
) {
3792 if (buf_size_same
) {
3793 if (!ring_buffer_expanded
)
3794 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3796 trace_buf_size
>> 10);
3798 r
= sprintf(buf
, "%lu\n", size
>> 10);
3800 r
= sprintf(buf
, "X\n");
3802 r
= sprintf(buf
, "%lu\n", tr
->data
[info
->cpu
]->entries
>> 10);
3804 mutex_unlock(&trace_types_lock
);
3806 ret
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3811 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3812 size_t cnt
, loff_t
*ppos
)
3814 struct ftrace_entries_info
*info
= filp
->private_data
;
3818 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3822 /* must have at least 1 entry */
3826 /* value is in KB */
3829 ret
= tracing_resize_ring_buffer(val
, info
->cpu
);
3839 tracing_entries_release(struct inode
*inode
, struct file
*filp
)
3841 struct ftrace_entries_info
*info
= filp
->private_data
;
3849 tracing_total_entries_read(struct file
*filp
, char __user
*ubuf
,
3850 size_t cnt
, loff_t
*ppos
)
3852 struct trace_array
*tr
= filp
->private_data
;
3855 unsigned long size
= 0, expanded_size
= 0;
3857 mutex_lock(&trace_types_lock
);
3858 for_each_tracing_cpu(cpu
) {
3859 size
+= tr
->data
[cpu
]->entries
>> 10;
3860 if (!ring_buffer_expanded
)
3861 expanded_size
+= trace_buf_size
>> 10;
3863 if (ring_buffer_expanded
)
3864 r
= sprintf(buf
, "%lu\n", size
);
3866 r
= sprintf(buf
, "%lu (expanded: %lu)\n", size
, expanded_size
);
3867 mutex_unlock(&trace_types_lock
);
3869 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3873 tracing_free_buffer_write(struct file
*filp
, const char __user
*ubuf
,
3874 size_t cnt
, loff_t
*ppos
)
3877 * There is no need to read what the user has written, this function
3878 * is just to make sure that there is no error when "echo" is used
3887 tracing_free_buffer_release(struct inode
*inode
, struct file
*filp
)
3889 /* disable tracing ? */
3890 if (trace_flags
& TRACE_ITER_STOP_ON_FREE
)
3892 /* resize the ring buffer to 0 */
3893 tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS
);
3899 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3900 size_t cnt
, loff_t
*fpos
)
3902 unsigned long addr
= (unsigned long)ubuf
;
3903 struct ring_buffer_event
*event
;
3904 struct ring_buffer
*buffer
;
3905 struct print_entry
*entry
;
3906 unsigned long irq_flags
;
3907 struct page
*pages
[2];
3917 if (tracing_disabled
)
3920 if (!(trace_flags
& TRACE_ITER_MARKERS
))
3923 if (cnt
> TRACE_BUF_SIZE
)
3924 cnt
= TRACE_BUF_SIZE
;
3927 * Userspace is injecting traces into the kernel trace buffer.
3928 * We want to be as non intrusive as possible.
3929 * To do so, we do not want to allocate any special buffers
3930 * or take any locks, but instead write the userspace data
3931 * straight into the ring buffer.
3933 * First we need to pin the userspace buffer into memory,
3934 * which, most likely it is, because it just referenced it.
3935 * But there's no guarantee that it is. By using get_user_pages_fast()
3936 * and kmap_atomic/kunmap_atomic() we can get access to the
3937 * pages directly. We then write the data directly into the
3940 BUILD_BUG_ON(TRACE_BUF_SIZE
>= PAGE_SIZE
);
3942 /* check if we cross pages */
3943 if ((addr
& PAGE_MASK
) != ((addr
+ cnt
) & PAGE_MASK
))
3946 offset
= addr
& (PAGE_SIZE
- 1);
3949 ret
= get_user_pages_fast(addr
, nr_pages
, 0, pages
);
3950 if (ret
< nr_pages
) {
3952 put_page(pages
[ret
]);
3957 for (i
= 0; i
< nr_pages
; i
++)
3958 map_page
[i
] = kmap_atomic(pages
[i
]);
3960 local_save_flags(irq_flags
);
3961 size
= sizeof(*entry
) + cnt
+ 2; /* possible \n added */
3962 buffer
= global_trace
.buffer
;
3963 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
3964 irq_flags
, preempt_count());
3966 /* Ring buffer disabled, return as if not open for write */
3971 entry
= ring_buffer_event_data(event
);
3972 entry
->ip
= _THIS_IP_
;
3974 if (nr_pages
== 2) {
3975 len
= PAGE_SIZE
- offset
;
3976 memcpy(&entry
->buf
, map_page
[0] + offset
, len
);
3977 memcpy(&entry
->buf
[len
], map_page
[1], cnt
- len
);
3979 memcpy(&entry
->buf
, map_page
[0] + offset
, cnt
);
3981 if (entry
->buf
[cnt
- 1] != '\n') {
3982 entry
->buf
[cnt
] = '\n';
3983 entry
->buf
[cnt
+ 1] = '\0';
3985 entry
->buf
[cnt
] = '\0';
3987 __buffer_unlock_commit(buffer
, event
);
3994 for (i
= 0; i
< nr_pages
; i
++){
3995 kunmap_atomic(map_page
[i
]);
4002 static int tracing_clock_show(struct seq_file
*m
, void *v
)
4006 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
4008 "%s%s%s%s", i
? " " : "",
4009 i
== trace_clock_id
? "[" : "", trace_clocks
[i
].name
,
4010 i
== trace_clock_id
? "]" : "");
4016 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
4017 size_t cnt
, loff_t
*fpos
)
4020 const char *clockstr
;
4023 if (cnt
>= sizeof(buf
))
4026 if (copy_from_user(&buf
, ubuf
, cnt
))
4031 clockstr
= strstrip(buf
);
4033 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
4034 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
4037 if (i
== ARRAY_SIZE(trace_clocks
))
4042 mutex_lock(&trace_types_lock
);
4044 ring_buffer_set_clock(global_trace
.buffer
, trace_clocks
[i
].func
);
4046 ring_buffer_set_clock(max_tr
.buffer
, trace_clocks
[i
].func
);
4049 * New clock may not be consistent with the previous clock.
4050 * Reset the buffer so that it doesn't have incomparable timestamps.
4052 tracing_reset_online_cpus(&global_trace
);
4053 tracing_reset_online_cpus(&max_tr
);
4055 mutex_unlock(&trace_types_lock
);
4062 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
4064 if (tracing_disabled
)
4066 return single_open(file
, tracing_clock_show
, NULL
);
4069 static const struct file_operations tracing_max_lat_fops
= {
4070 .open
= tracing_open_generic
,
4071 .read
= tracing_max_lat_read
,
4072 .write
= tracing_max_lat_write
,
4073 .llseek
= generic_file_llseek
,
4076 static const struct file_operations set_tracer_fops
= {
4077 .open
= tracing_open_generic
,
4078 .read
= tracing_set_trace_read
,
4079 .write
= tracing_set_trace_write
,
4080 .llseek
= generic_file_llseek
,
4083 static const struct file_operations tracing_pipe_fops
= {
4084 .open
= tracing_open_pipe
,
4085 .poll
= tracing_poll_pipe
,
4086 .read
= tracing_read_pipe
,
4087 .splice_read
= tracing_splice_read_pipe
,
4088 .release
= tracing_release_pipe
,
4089 .llseek
= no_llseek
,
4092 static const struct file_operations tracing_entries_fops
= {
4093 .open
= tracing_entries_open
,
4094 .read
= tracing_entries_read
,
4095 .write
= tracing_entries_write
,
4096 .release
= tracing_entries_release
,
4097 .llseek
= generic_file_llseek
,
4100 static const struct file_operations tracing_total_entries_fops
= {
4101 .open
= tracing_open_generic
,
4102 .read
= tracing_total_entries_read
,
4103 .llseek
= generic_file_llseek
,
4106 static const struct file_operations tracing_free_buffer_fops
= {
4107 .write
= tracing_free_buffer_write
,
4108 .release
= tracing_free_buffer_release
,
4111 static const struct file_operations tracing_mark_fops
= {
4112 .open
= tracing_open_generic
,
4113 .write
= tracing_mark_write
,
4114 .llseek
= generic_file_llseek
,
4117 static const struct file_operations trace_clock_fops
= {
4118 .open
= tracing_clock_open
,
4120 .llseek
= seq_lseek
,
4121 .release
= single_release
,
4122 .write
= tracing_clock_write
,
4125 struct ftrace_buffer_info
{
4126 struct trace_array
*tr
;
4132 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
4134 int cpu
= (int)(long)inode
->i_private
;
4135 struct ftrace_buffer_info
*info
;
4137 if (tracing_disabled
)
4140 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
4144 info
->tr
= &global_trace
;
4147 /* Force reading ring buffer for first read */
4148 info
->read
= (unsigned int)-1;
4150 filp
->private_data
= info
;
4152 return nonseekable_open(inode
, filp
);
4156 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
4157 size_t count
, loff_t
*ppos
)
4159 struct ftrace_buffer_info
*info
= filp
->private_data
;
4167 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
, info
->cpu
);
4171 /* Do we have previous read data to read? */
4172 if (info
->read
< PAGE_SIZE
)
4175 trace_access_lock(info
->cpu
);
4176 ret
= ring_buffer_read_page(info
->tr
->buffer
,
4180 trace_access_unlock(info
->cpu
);
4187 size
= PAGE_SIZE
- info
->read
;
4191 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
4202 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
4204 struct ftrace_buffer_info
*info
= file
->private_data
;
4207 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
4214 struct ring_buffer
*buffer
;
4219 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
4220 struct pipe_buffer
*buf
)
4222 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
4227 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4232 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
4233 struct pipe_buffer
*buf
)
4235 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
4240 /* Pipe buffer operations for a buffer. */
4241 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
4243 .map
= generic_pipe_buf_map
,
4244 .unmap
= generic_pipe_buf_unmap
,
4245 .confirm
= generic_pipe_buf_confirm
,
4246 .release
= buffer_pipe_buf_release
,
4247 .steal
= generic_pipe_buf_steal
,
4248 .get
= buffer_pipe_buf_get
,
4252 * Callback from splice_to_pipe(), if we need to release some pages
4253 * at the end of the spd in case we error'ed out in filling the pipe.
4255 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
4257 struct buffer_ref
*ref
=
4258 (struct buffer_ref
*)spd
->partial
[i
].private;
4263 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4265 spd
->partial
[i
].private = 0;
4269 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
4270 struct pipe_inode_info
*pipe
, size_t len
,
4273 struct ftrace_buffer_info
*info
= file
->private_data
;
4274 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
4275 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
4276 struct splice_pipe_desc spd
= {
4278 .partial
= partial_def
,
4279 .nr_pages_max
= PIPE_DEF_BUFFERS
,
4281 .ops
= &buffer_pipe_buf_ops
,
4282 .spd_release
= buffer_spd_release
,
4284 struct buffer_ref
*ref
;
4285 int entries
, size
, i
;
4288 if (splice_grow_spd(pipe
, &spd
))
4291 if (*ppos
& (PAGE_SIZE
- 1)) {
4296 if (len
& (PAGE_SIZE
- 1)) {
4297 if (len
< PAGE_SIZE
) {
4304 trace_access_lock(info
->cpu
);
4305 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4307 for (i
= 0; i
< pipe
->buffers
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
4311 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
4316 ref
->buffer
= info
->tr
->buffer
;
4317 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
, info
->cpu
);
4323 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
4326 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4332 * zero out any left over data, this is going to
4335 size
= ring_buffer_page_len(ref
->page
);
4336 if (size
< PAGE_SIZE
)
4337 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
4339 page
= virt_to_page(ref
->page
);
4341 spd
.pages
[i
] = page
;
4342 spd
.partial
[i
].len
= PAGE_SIZE
;
4343 spd
.partial
[i
].offset
= 0;
4344 spd
.partial
[i
].private = (unsigned long)ref
;
4348 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4351 trace_access_unlock(info
->cpu
);
4354 /* did we read anything? */
4355 if (!spd
.nr_pages
) {
4356 if (flags
& SPLICE_F_NONBLOCK
)
4364 ret
= splice_to_pipe(pipe
, &spd
);
4365 splice_shrink_spd(&spd
);
4370 static const struct file_operations tracing_buffers_fops
= {
4371 .open
= tracing_buffers_open
,
4372 .read
= tracing_buffers_read
,
4373 .release
= tracing_buffers_release
,
4374 .splice_read
= tracing_buffers_splice_read
,
4375 .llseek
= no_llseek
,
4379 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
4380 size_t count
, loff_t
*ppos
)
4382 unsigned long cpu
= (unsigned long)filp
->private_data
;
4383 struct trace_array
*tr
= &global_trace
;
4384 struct trace_seq
*s
;
4386 unsigned long long t
;
4387 unsigned long usec_rem
;
4389 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
4395 cnt
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
4396 trace_seq_printf(s
, "entries: %ld\n", cnt
);
4398 cnt
= ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
4399 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
4401 cnt
= ring_buffer_commit_overrun_cpu(tr
->buffer
, cpu
);
4402 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
4404 cnt
= ring_buffer_bytes_cpu(tr
->buffer
, cpu
);
4405 trace_seq_printf(s
, "bytes: %ld\n", cnt
);
4407 if (trace_clocks
[trace_clock_id
].in_ns
) {
4408 /* local or global for trace_clock */
4409 t
= ns2usecs(ring_buffer_oldest_event_ts(tr
->buffer
, cpu
));
4410 usec_rem
= do_div(t
, USEC_PER_SEC
);
4411 trace_seq_printf(s
, "oldest event ts: %5llu.%06lu\n",
4414 t
= ns2usecs(ring_buffer_time_stamp(tr
->buffer
, cpu
));
4415 usec_rem
= do_div(t
, USEC_PER_SEC
);
4416 trace_seq_printf(s
, "now ts: %5llu.%06lu\n", t
, usec_rem
);
4418 /* counter or tsc mode for trace_clock */
4419 trace_seq_printf(s
, "oldest event ts: %llu\n",
4420 ring_buffer_oldest_event_ts(tr
->buffer
, cpu
));
4422 trace_seq_printf(s
, "now ts: %llu\n",
4423 ring_buffer_time_stamp(tr
->buffer
, cpu
));
4426 cnt
= ring_buffer_dropped_events_cpu(tr
->buffer
, cpu
);
4427 trace_seq_printf(s
, "dropped events: %ld\n", cnt
);
4429 cnt
= ring_buffer_read_events_cpu(tr
->buffer
, cpu
);
4430 trace_seq_printf(s
, "read events: %ld\n", cnt
);
4432 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
4439 static const struct file_operations tracing_stats_fops
= {
4440 .open
= tracing_open_generic
,
4441 .read
= tracing_stats_read
,
4442 .llseek
= generic_file_llseek
,
4445 #ifdef CONFIG_DYNAMIC_FTRACE
4447 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
4453 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
4454 size_t cnt
, loff_t
*ppos
)
4456 static char ftrace_dyn_info_buffer
[1024];
4457 static DEFINE_MUTEX(dyn_info_mutex
);
4458 unsigned long *p
= filp
->private_data
;
4459 char *buf
= ftrace_dyn_info_buffer
;
4460 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
4463 mutex_lock(&dyn_info_mutex
);
4464 r
= sprintf(buf
, "%ld ", *p
);
4466 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
4469 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4471 mutex_unlock(&dyn_info_mutex
);
4476 static const struct file_operations tracing_dyn_info_fops
= {
4477 .open
= tracing_open_generic
,
4478 .read
= tracing_read_dyn_info
,
4479 .llseek
= generic_file_llseek
,
4483 static struct dentry
*d_tracer
;
4485 struct dentry
*tracing_init_dentry(void)
4492 if (!debugfs_initialized())
4495 d_tracer
= debugfs_create_dir("tracing", NULL
);
4497 if (!d_tracer
&& !once
) {
4499 pr_warning("Could not create debugfs directory 'tracing'\n");
4506 static struct dentry
*d_percpu
;
4508 static struct dentry
*tracing_dentry_percpu(void)
4511 struct dentry
*d_tracer
;
4516 d_tracer
= tracing_init_dentry();
4521 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
4523 if (!d_percpu
&& !once
) {
4525 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4532 static void tracing_init_debugfs_percpu(long cpu
)
4534 struct dentry
*d_percpu
= tracing_dentry_percpu();
4535 struct dentry
*d_cpu
;
4536 char cpu_dir
[30]; /* 30 characters should be more than enough */
4541 snprintf(cpu_dir
, 30, "cpu%ld", cpu
);
4542 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
4544 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
4548 /* per cpu trace_pipe */
4549 trace_create_file("trace_pipe", 0444, d_cpu
,
4550 (void *) cpu
, &tracing_pipe_fops
);
4553 trace_create_file("trace", 0644, d_cpu
,
4554 (void *) cpu
, &tracing_fops
);
4556 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
4557 (void *) cpu
, &tracing_buffers_fops
);
4559 trace_create_file("stats", 0444, d_cpu
,
4560 (void *) cpu
, &tracing_stats_fops
);
4562 trace_create_file("buffer_size_kb", 0444, d_cpu
,
4563 (void *) cpu
, &tracing_entries_fops
);
4566 #ifdef CONFIG_FTRACE_SELFTEST
4567 /* Let selftest have access to static functions in this file */
4568 #include "trace_selftest.c"
4571 struct trace_option_dentry
{
4572 struct tracer_opt
*opt
;
4573 struct tracer_flags
*flags
;
4574 struct dentry
*entry
;
4578 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4581 struct trace_option_dentry
*topt
= filp
->private_data
;
4584 if (topt
->flags
->val
& topt
->opt
->bit
)
4589 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4593 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4596 struct trace_option_dentry
*topt
= filp
->private_data
;
4600 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4604 if (val
!= 0 && val
!= 1)
4607 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
4608 mutex_lock(&trace_types_lock
);
4609 ret
= __set_tracer_option(current_trace
, topt
->flags
,
4611 mutex_unlock(&trace_types_lock
);
4622 static const struct file_operations trace_options_fops
= {
4623 .open
= tracing_open_generic
,
4624 .read
= trace_options_read
,
4625 .write
= trace_options_write
,
4626 .llseek
= generic_file_llseek
,
4630 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4633 long index
= (long)filp
->private_data
;
4636 if (trace_flags
& (1 << index
))
4641 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4645 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4648 long index
= (long)filp
->private_data
;
4652 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4656 if (val
!= 0 && val
!= 1)
4658 set_tracer_flags(1 << index
, val
);
4665 static const struct file_operations trace_options_core_fops
= {
4666 .open
= tracing_open_generic
,
4667 .read
= trace_options_core_read
,
4668 .write
= trace_options_core_write
,
4669 .llseek
= generic_file_llseek
,
4672 struct dentry
*trace_create_file(const char *name
,
4674 struct dentry
*parent
,
4676 const struct file_operations
*fops
)
4680 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
4682 pr_warning("Could not create debugfs '%s' entry\n", name
);
4688 static struct dentry
*trace_options_init_dentry(void)
4690 struct dentry
*d_tracer
;
4691 static struct dentry
*t_options
;
4696 d_tracer
= tracing_init_dentry();
4700 t_options
= debugfs_create_dir("options", d_tracer
);
4702 pr_warning("Could not create debugfs directory 'options'\n");
4710 create_trace_option_file(struct trace_option_dentry
*topt
,
4711 struct tracer_flags
*flags
,
4712 struct tracer_opt
*opt
)
4714 struct dentry
*t_options
;
4716 t_options
= trace_options_init_dentry();
4720 topt
->flags
= flags
;
4723 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
4724 &trace_options_fops
);
4728 static struct trace_option_dentry
*
4729 create_trace_option_files(struct tracer
*tracer
)
4731 struct trace_option_dentry
*topts
;
4732 struct tracer_flags
*flags
;
4733 struct tracer_opt
*opts
;
4739 flags
= tracer
->flags
;
4741 if (!flags
|| !flags
->opts
)
4746 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4749 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
4753 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4754 create_trace_option_file(&topts
[cnt
], flags
,
4761 destroy_trace_option_files(struct trace_option_dentry
*topts
)
4768 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
4769 if (topts
[cnt
].entry
)
4770 debugfs_remove(topts
[cnt
].entry
);
4776 static struct dentry
*
4777 create_trace_option_core_file(const char *option
, long index
)
4779 struct dentry
*t_options
;
4781 t_options
= trace_options_init_dentry();
4785 return trace_create_file(option
, 0644, t_options
, (void *)index
,
4786 &trace_options_core_fops
);
4789 static __init
void create_trace_options_dir(void)
4791 struct dentry
*t_options
;
4794 t_options
= trace_options_init_dentry();
4798 for (i
= 0; trace_options
[i
]; i
++)
4799 create_trace_option_core_file(trace_options
[i
], i
);
4803 rb_simple_read(struct file
*filp
, char __user
*ubuf
,
4804 size_t cnt
, loff_t
*ppos
)
4806 struct trace_array
*tr
= filp
->private_data
;
4807 struct ring_buffer
*buffer
= tr
->buffer
;
4812 r
= ring_buffer_record_is_on(buffer
);
4816 r
= sprintf(buf
, "%d\n", r
);
4818 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4822 rb_simple_write(struct file
*filp
, const char __user
*ubuf
,
4823 size_t cnt
, loff_t
*ppos
)
4825 struct trace_array
*tr
= filp
->private_data
;
4826 struct ring_buffer
*buffer
= tr
->buffer
;
4830 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4835 mutex_lock(&trace_types_lock
);
4837 ring_buffer_record_on(buffer
);
4838 if (current_trace
->start
)
4839 current_trace
->start(tr
);
4841 ring_buffer_record_off(buffer
);
4842 if (current_trace
->stop
)
4843 current_trace
->stop(tr
);
4845 mutex_unlock(&trace_types_lock
);
4853 static const struct file_operations rb_simple_fops
= {
4854 .open
= tracing_open_generic
,
4855 .read
= rb_simple_read
,
4856 .write
= rb_simple_write
,
4857 .llseek
= default_llseek
,
4860 static __init
int tracer_init_debugfs(void)
4862 struct dentry
*d_tracer
;
4865 trace_access_lock_init();
4867 d_tracer
= tracing_init_dentry();
4869 trace_create_file("trace_options", 0644, d_tracer
,
4870 NULL
, &tracing_iter_fops
);
4872 trace_create_file("tracing_cpumask", 0644, d_tracer
,
4873 NULL
, &tracing_cpumask_fops
);
4875 trace_create_file("trace", 0644, d_tracer
,
4876 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
4878 trace_create_file("available_tracers", 0444, d_tracer
,
4879 &global_trace
, &show_traces_fops
);
4881 trace_create_file("current_tracer", 0644, d_tracer
,
4882 &global_trace
, &set_tracer_fops
);
4884 #ifdef CONFIG_TRACER_MAX_TRACE
4885 trace_create_file("tracing_max_latency", 0644, d_tracer
,
4886 &tracing_max_latency
, &tracing_max_lat_fops
);
4889 trace_create_file("tracing_thresh", 0644, d_tracer
,
4890 &tracing_thresh
, &tracing_max_lat_fops
);
4892 trace_create_file("README", 0444, d_tracer
,
4893 NULL
, &tracing_readme_fops
);
4895 trace_create_file("trace_pipe", 0444, d_tracer
,
4896 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
4898 trace_create_file("buffer_size_kb", 0644, d_tracer
,
4899 (void *) RING_BUFFER_ALL_CPUS
, &tracing_entries_fops
);
4901 trace_create_file("buffer_total_size_kb", 0444, d_tracer
,
4902 &global_trace
, &tracing_total_entries_fops
);
4904 trace_create_file("free_buffer", 0644, d_tracer
,
4905 &global_trace
, &tracing_free_buffer_fops
);
4907 trace_create_file("trace_marker", 0220, d_tracer
,
4908 NULL
, &tracing_mark_fops
);
4910 trace_create_file("saved_cmdlines", 0444, d_tracer
,
4911 NULL
, &tracing_saved_cmdlines_fops
);
4913 trace_create_file("trace_clock", 0644, d_tracer
, NULL
,
4916 trace_create_file("tracing_on", 0644, d_tracer
,
4917 &global_trace
, &rb_simple_fops
);
4919 #ifdef CONFIG_DYNAMIC_FTRACE
4920 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
4921 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
4924 create_trace_options_dir();
4926 for_each_tracing_cpu(cpu
)
4927 tracing_init_debugfs_percpu(cpu
);
4932 static int trace_panic_handler(struct notifier_block
*this,
4933 unsigned long event
, void *unused
)
4935 if (ftrace_dump_on_oops
)
4936 ftrace_dump(ftrace_dump_on_oops
);
4940 static struct notifier_block trace_panic_notifier
= {
4941 .notifier_call
= trace_panic_handler
,
4943 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
4946 static int trace_die_handler(struct notifier_block
*self
,
4952 if (ftrace_dump_on_oops
)
4953 ftrace_dump(ftrace_dump_on_oops
);
4961 static struct notifier_block trace_die_notifier
= {
4962 .notifier_call
= trace_die_handler
,
4967 * printk is set to max of 1024, we really don't need it that big.
4968 * Nothing should be printing 1000 characters anyway.
4970 #define TRACE_MAX_PRINT 1000
4973 * Define here KERN_TRACE so that we have one place to modify
4974 * it if we decide to change what log level the ftrace dump
4977 #define KERN_TRACE KERN_EMERG
4980 trace_printk_seq(struct trace_seq
*s
)
4982 /* Probably should print a warning here. */
4986 /* should be zero ended, but we are paranoid. */
4987 s
->buffer
[s
->len
] = 0;
4989 printk(KERN_TRACE
"%s", s
->buffer
);
4994 void trace_init_global_iter(struct trace_iterator
*iter
)
4996 iter
->tr
= &global_trace
;
4997 iter
->trace
= current_trace
;
4998 iter
->cpu_file
= TRACE_PIPE_ALL_CPU
;
5002 __ftrace_dump(bool disable_tracing
, enum ftrace_dump_mode oops_dump_mode
)
5004 static arch_spinlock_t ftrace_dump_lock
=
5005 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
5006 /* use static because iter can be a bit big for the stack */
5007 static struct trace_iterator iter
;
5008 unsigned int old_userobj
;
5009 static int dump_ran
;
5010 unsigned long flags
;
5014 local_irq_save(flags
);
5015 arch_spin_lock(&ftrace_dump_lock
);
5023 /* Did function tracer already get disabled? */
5024 if (ftrace_is_dead()) {
5025 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
5026 printk("# MAY BE MISSING FUNCTION EVENTS\n");
5029 if (disable_tracing
)
5032 /* Simulate the iterator */
5033 trace_init_global_iter(&iter
);
5035 for_each_tracing_cpu(cpu
) {
5036 atomic_inc(&iter
.tr
->data
[cpu
]->disabled
);
5039 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
5041 /* don't look at user memory in panic mode */
5042 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
5044 switch (oops_dump_mode
) {
5046 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
5049 iter
.cpu_file
= raw_smp_processor_id();
5054 printk(KERN_TRACE
"Bad dumping mode, switching to all CPUs dump\n");
5055 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
5058 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
5061 * We need to stop all tracing on all CPUS to read the
5062 * the next buffer. This is a bit expensive, but is
5063 * not done often. We fill all what we can read,
5064 * and then release the locks again.
5067 while (!trace_empty(&iter
)) {
5070 printk(KERN_TRACE
"---------------------------------\n");
5074 /* reset all but tr, trace, and overruns */
5075 memset(&iter
.seq
, 0,
5076 sizeof(struct trace_iterator
) -
5077 offsetof(struct trace_iterator
, seq
));
5078 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
5081 if (trace_find_next_entry_inc(&iter
) != NULL
) {
5084 ret
= print_trace_line(&iter
);
5085 if (ret
!= TRACE_TYPE_NO_CONSUME
)
5086 trace_consume(&iter
);
5088 touch_nmi_watchdog();
5090 trace_printk_seq(&iter
.seq
);
5094 printk(KERN_TRACE
" (ftrace buffer empty)\n");
5096 printk(KERN_TRACE
"---------------------------------\n");
5099 /* Re-enable tracing if requested */
5100 if (!disable_tracing
) {
5101 trace_flags
|= old_userobj
;
5103 for_each_tracing_cpu(cpu
) {
5104 atomic_dec(&iter
.tr
->data
[cpu
]->disabled
);
5110 arch_spin_unlock(&ftrace_dump_lock
);
5111 local_irq_restore(flags
);
5114 /* By default: disable tracing after the dump */
5115 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode
)
5117 __ftrace_dump(true, oops_dump_mode
);
5119 EXPORT_SYMBOL_GPL(ftrace_dump
);
5121 __init
static int tracer_alloc_buffers(void)
5124 enum ring_buffer_flags rb_flags
;
5129 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
5132 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
5133 goto out_free_buffer_mask
;
5135 /* Only allocate trace_printk buffers if a trace_printk exists */
5136 if (__stop___trace_bprintk_fmt
!= __start___trace_bprintk_fmt
)
5137 /* Must be called before global_trace.buffer is allocated */
5138 trace_printk_init_buffers();
5140 /* To save memory, keep the ring buffer size to its minimum */
5141 if (ring_buffer_expanded
)
5142 ring_buf_size
= trace_buf_size
;
5146 rb_flags
= trace_flags
& TRACE_ITER_OVERWRITE
? RB_FL_OVERWRITE
: 0;
5148 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
5149 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
5151 /* TODO: make the number of buffers hot pluggable with CPUS */
5152 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
, rb_flags
);
5153 if (!global_trace
.buffer
) {
5154 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
5156 goto out_free_cpumask
;
5158 if (global_trace
.buffer_disabled
)
5162 #ifdef CONFIG_TRACER_MAX_TRACE
5163 max_tr
.buffer
= ring_buffer_alloc(1, rb_flags
);
5164 if (!max_tr
.buffer
) {
5165 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
5167 ring_buffer_free(global_trace
.buffer
);
5168 goto out_free_cpumask
;
5172 /* Allocate the first page for all buffers */
5173 for_each_tracing_cpu(i
) {
5174 global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
5175 max_tr
.data
[i
] = &per_cpu(max_tr_data
, i
);
5178 set_buffer_entries(&global_trace
,
5179 ring_buffer_size(global_trace
.buffer
, 0));
5180 #ifdef CONFIG_TRACER_MAX_TRACE
5181 set_buffer_entries(&max_tr
, 1);
5184 trace_init_cmdlines();
5185 init_irq_work(&trace_work_wakeup
, trace_wake_up
);
5187 register_tracer(&nop_trace
);
5188 current_trace
= &nop_trace
;
5189 /* All seems OK, enable tracing */
5190 tracing_disabled
= 0;
5192 atomic_notifier_chain_register(&panic_notifier_list
,
5193 &trace_panic_notifier
);
5195 register_die_notifier(&trace_die_notifier
);
5197 while (trace_boot_options
) {
5200 option
= strsep(&trace_boot_options
, ",");
5201 trace_set_options(option
);
5207 free_cpumask_var(tracing_cpumask
);
5208 out_free_buffer_mask
:
5209 free_cpumask_var(tracing_buffer_mask
);
5214 __init
static int clear_boot_tracer(void)
5217 * The default tracer at boot buffer is an init section.
5218 * This function is called in lateinit. If we did not
5219 * find the boot tracer, then clear it out, to prevent
5220 * later registration from accessing the buffer that is
5221 * about to be freed.
5223 if (!default_bootup_tracer
)
5226 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
5227 default_bootup_tracer
);
5228 default_bootup_tracer
= NULL
;
5233 early_initcall(tracer_alloc_buffers
);
5234 fs_initcall(tracer_init_debugfs
);
5235 late_initcall(clear_boot_tracer
);