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
= ++__get_cpu_var(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 __get_cpu_var(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 static struct tracer
*old_tracer
;
1952 int cpu_file
= iter
->cpu_file
;
1957 /* copy the tracer to avoid using a global lock all around */
1958 mutex_lock(&trace_types_lock
);
1959 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1960 old_tracer
= current_trace
;
1961 *iter
->trace
= *current_trace
;
1963 mutex_unlock(&trace_types_lock
);
1965 atomic_inc(&trace_record_cmdline_disabled
);
1967 if (*pos
!= iter
->pos
) {
1972 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1973 for_each_tracing_cpu(cpu
)
1974 tracing_iter_reset(iter
, cpu
);
1976 tracing_iter_reset(iter
, cpu_file
);
1979 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1984 * If we overflowed the seq_file before, then we want
1985 * to just reuse the trace_seq buffer again.
1991 p
= s_next(m
, p
, &l
);
1995 trace_event_read_lock();
1996 trace_access_lock(cpu_file
);
2000 static void s_stop(struct seq_file
*m
, void *p
)
2002 struct trace_iterator
*iter
= m
->private;
2004 atomic_dec(&trace_record_cmdline_disabled
);
2005 trace_access_unlock(iter
->cpu_file
);
2006 trace_event_read_unlock();
2010 get_total_entries(struct trace_array
*tr
, unsigned long *total
, unsigned long *entries
)
2012 unsigned long count
;
2018 for_each_tracing_cpu(cpu
) {
2019 count
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
2021 * If this buffer has skipped entries, then we hold all
2022 * entries for the trace and we need to ignore the
2023 * ones before the time stamp.
2025 if (tr
->data
[cpu
]->skipped_entries
) {
2026 count
-= tr
->data
[cpu
]->skipped_entries
;
2027 /* total is the same as the entries */
2031 ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
2036 static void print_lat_help_header(struct seq_file
*m
)
2038 seq_puts(m
, "# _------=> CPU# \n");
2039 seq_puts(m
, "# / _-----=> irqs-off \n");
2040 seq_puts(m
, "# | / _----=> need-resched \n");
2041 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
2042 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
2043 seq_puts(m
, "# |||| / delay \n");
2044 seq_puts(m
, "# cmd pid ||||| time | caller \n");
2045 seq_puts(m
, "# \\ / ||||| \\ | / \n");
2048 static void print_event_info(struct trace_array
*tr
, struct seq_file
*m
)
2050 unsigned long total
;
2051 unsigned long entries
;
2053 get_total_entries(tr
, &total
, &entries
);
2054 seq_printf(m
, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2055 entries
, total
, num_online_cpus());
2059 static void print_func_help_header(struct trace_array
*tr
, struct seq_file
*m
)
2061 print_event_info(tr
, m
);
2062 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2063 seq_puts(m
, "# | | | | |\n");
2066 static void print_func_help_header_irq(struct trace_array
*tr
, struct seq_file
*m
)
2068 print_event_info(tr
, m
);
2069 seq_puts(m
, "# _-----=> irqs-off\n");
2070 seq_puts(m
, "# / _----=> need-resched\n");
2071 seq_puts(m
, "# | / _---=> hardirq/softirq\n");
2072 seq_puts(m
, "# || / _--=> preempt-depth\n");
2073 seq_puts(m
, "# ||| / delay\n");
2074 seq_puts(m
, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2075 seq_puts(m
, "# | | | |||| | |\n");
2079 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
2081 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2082 struct trace_array
*tr
= iter
->tr
;
2083 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
2084 struct tracer
*type
= current_trace
;
2085 unsigned long entries
;
2086 unsigned long total
;
2087 const char *name
= "preemption";
2092 get_total_entries(tr
, &total
, &entries
);
2094 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
2096 seq_puts(m
, "# -----------------------------------"
2097 "---------------------------------\n");
2098 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2099 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2100 nsecs_to_usecs(data
->saved_latency
),
2104 #if defined(CONFIG_PREEMPT_NONE)
2106 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2108 #elif defined(CONFIG_PREEMPT)
2113 /* These are reserved for later use */
2116 seq_printf(m
, " #P:%d)\n", num_online_cpus());
2120 seq_puts(m
, "# -----------------\n");
2121 seq_printf(m
, "# | task: %.16s-%d "
2122 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2123 data
->comm
, data
->pid
,
2124 from_kuid_munged(seq_user_ns(m
), data
->uid
), data
->nice
,
2125 data
->policy
, data
->rt_priority
);
2126 seq_puts(m
, "# -----------------\n");
2128 if (data
->critical_start
) {
2129 seq_puts(m
, "# => started at: ");
2130 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
2131 trace_print_seq(m
, &iter
->seq
);
2132 seq_puts(m
, "\n# => ended at: ");
2133 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
2134 trace_print_seq(m
, &iter
->seq
);
2135 seq_puts(m
, "\n#\n");
2141 static void test_cpu_buff_start(struct trace_iterator
*iter
)
2143 struct trace_seq
*s
= &iter
->seq
;
2145 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
2148 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
2151 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
2154 if (iter
->tr
->data
[iter
->cpu
]->skipped_entries
)
2157 cpumask_set_cpu(iter
->cpu
, iter
->started
);
2159 /* Don't print started cpu buffer for the first entry of the trace */
2161 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
2165 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
2167 struct trace_seq
*s
= &iter
->seq
;
2168 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2169 struct trace_entry
*entry
;
2170 struct trace_event
*event
;
2174 test_cpu_buff_start(iter
);
2176 event
= ftrace_find_event(entry
->type
);
2178 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2179 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2180 if (!trace_print_lat_context(iter
))
2183 if (!trace_print_context(iter
))
2189 return event
->funcs
->trace(iter
, sym_flags
, event
);
2191 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
2194 return TRACE_TYPE_HANDLED
;
2196 return TRACE_TYPE_PARTIAL_LINE
;
2199 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
2201 struct trace_seq
*s
= &iter
->seq
;
2202 struct trace_entry
*entry
;
2203 struct trace_event
*event
;
2207 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2208 if (!trace_seq_printf(s
, "%d %d %llu ",
2209 entry
->pid
, iter
->cpu
, iter
->ts
))
2213 event
= ftrace_find_event(entry
->type
);
2215 return event
->funcs
->raw(iter
, 0, event
);
2217 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
2220 return TRACE_TYPE_HANDLED
;
2222 return TRACE_TYPE_PARTIAL_LINE
;
2225 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
2227 struct trace_seq
*s
= &iter
->seq
;
2228 unsigned char newline
= '\n';
2229 struct trace_entry
*entry
;
2230 struct trace_event
*event
;
2234 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2235 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
2236 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
2237 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
2240 event
= ftrace_find_event(entry
->type
);
2242 enum print_line_t ret
= event
->funcs
->hex(iter
, 0, event
);
2243 if (ret
!= TRACE_TYPE_HANDLED
)
2247 SEQ_PUT_FIELD_RET(s
, newline
);
2249 return TRACE_TYPE_HANDLED
;
2252 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
2254 struct trace_seq
*s
= &iter
->seq
;
2255 struct trace_entry
*entry
;
2256 struct trace_event
*event
;
2260 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2261 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
2262 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
2263 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
2266 event
= ftrace_find_event(entry
->type
);
2267 return event
? event
->funcs
->binary(iter
, 0, event
) :
2271 int trace_empty(struct trace_iterator
*iter
)
2273 struct ring_buffer_iter
*buf_iter
;
2276 /* If we are looking at one CPU buffer, only check that one */
2277 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
2278 cpu
= iter
->cpu_file
;
2279 buf_iter
= trace_buffer_iter(iter
, cpu
);
2281 if (!ring_buffer_iter_empty(buf_iter
))
2284 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2290 for_each_tracing_cpu(cpu
) {
2291 buf_iter
= trace_buffer_iter(iter
, cpu
);
2293 if (!ring_buffer_iter_empty(buf_iter
))
2296 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2304 /* Called with trace_event_read_lock() held. */
2305 enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
2307 enum print_line_t ret
;
2309 if (iter
->lost_events
&&
2310 !trace_seq_printf(&iter
->seq
, "CPU:%d [LOST %lu EVENTS]\n",
2311 iter
->cpu
, iter
->lost_events
))
2312 return TRACE_TYPE_PARTIAL_LINE
;
2314 if (iter
->trace
&& iter
->trace
->print_line
) {
2315 ret
= iter
->trace
->print_line(iter
);
2316 if (ret
!= TRACE_TYPE_UNHANDLED
)
2320 if (iter
->ent
->type
== TRACE_BPRINT
&&
2321 trace_flags
& TRACE_ITER_PRINTK
&&
2322 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2323 return trace_print_bprintk_msg_only(iter
);
2325 if (iter
->ent
->type
== TRACE_PRINT
&&
2326 trace_flags
& TRACE_ITER_PRINTK
&&
2327 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2328 return trace_print_printk_msg_only(iter
);
2330 if (trace_flags
& TRACE_ITER_BIN
)
2331 return print_bin_fmt(iter
);
2333 if (trace_flags
& TRACE_ITER_HEX
)
2334 return print_hex_fmt(iter
);
2336 if (trace_flags
& TRACE_ITER_RAW
)
2337 return print_raw_fmt(iter
);
2339 return print_trace_fmt(iter
);
2342 void trace_latency_header(struct seq_file
*m
)
2344 struct trace_iterator
*iter
= m
->private;
2346 /* print nothing if the buffers are empty */
2347 if (trace_empty(iter
))
2350 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
2351 print_trace_header(m
, iter
);
2353 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2354 print_lat_help_header(m
);
2357 void trace_default_header(struct seq_file
*m
)
2359 struct trace_iterator
*iter
= m
->private;
2361 if (!(trace_flags
& TRACE_ITER_CONTEXT_INFO
))
2364 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2365 /* print nothing if the buffers are empty */
2366 if (trace_empty(iter
))
2368 print_trace_header(m
, iter
);
2369 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2370 print_lat_help_header(m
);
2372 if (!(trace_flags
& TRACE_ITER_VERBOSE
)) {
2373 if (trace_flags
& TRACE_ITER_IRQ_INFO
)
2374 print_func_help_header_irq(iter
->tr
, m
);
2376 print_func_help_header(iter
->tr
, m
);
2381 static void test_ftrace_alive(struct seq_file
*m
)
2383 if (!ftrace_is_dead())
2385 seq_printf(m
, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2386 seq_printf(m
, "# MAY BE MISSING FUNCTION EVENTS\n");
2389 static int s_show(struct seq_file
*m
, void *v
)
2391 struct trace_iterator
*iter
= v
;
2394 if (iter
->ent
== NULL
) {
2396 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
2398 test_ftrace_alive(m
);
2400 if (iter
->trace
&& iter
->trace
->print_header
)
2401 iter
->trace
->print_header(m
);
2403 trace_default_header(m
);
2405 } else if (iter
->leftover
) {
2407 * If we filled the seq_file buffer earlier, we
2408 * want to just show it now.
2410 ret
= trace_print_seq(m
, &iter
->seq
);
2412 /* ret should this time be zero, but you never know */
2413 iter
->leftover
= ret
;
2416 print_trace_line(iter
);
2417 ret
= trace_print_seq(m
, &iter
->seq
);
2419 * If we overflow the seq_file buffer, then it will
2420 * ask us for this data again at start up.
2422 * ret is 0 if seq_file write succeeded.
2425 iter
->leftover
= ret
;
2431 static const struct seq_operations tracer_seq_ops
= {
2438 static struct trace_iterator
*
2439 __tracing_open(struct inode
*inode
, struct file
*file
)
2441 long cpu_file
= (long) inode
->i_private
;
2442 struct trace_iterator
*iter
;
2445 if (tracing_disabled
)
2446 return ERR_PTR(-ENODEV
);
2448 iter
= __seq_open_private(file
, &tracer_seq_ops
, sizeof(*iter
));
2450 return ERR_PTR(-ENOMEM
);
2452 iter
->buffer_iter
= kzalloc(sizeof(*iter
->buffer_iter
) * num_possible_cpus(),
2454 if (!iter
->buffer_iter
)
2458 * We make a copy of the current tracer to avoid concurrent
2459 * changes on it while we are reading.
2461 mutex_lock(&trace_types_lock
);
2462 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2467 *iter
->trace
= *current_trace
;
2469 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2472 if (current_trace
&& current_trace
->print_max
)
2475 iter
->tr
= &global_trace
;
2477 mutex_init(&iter
->mutex
);
2478 iter
->cpu_file
= cpu_file
;
2480 /* Notify the tracer early; before we stop tracing. */
2481 if (iter
->trace
&& iter
->trace
->open
)
2482 iter
->trace
->open(iter
);
2484 /* Annotate start of buffers if we had overruns */
2485 if (ring_buffer_overruns(iter
->tr
->buffer
))
2486 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2488 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2489 if (trace_clocks
[trace_clock_id
].in_ns
)
2490 iter
->iter_flags
|= TRACE_FILE_TIME_IN_NS
;
2492 /* stop the trace while dumping */
2495 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
2496 for_each_tracing_cpu(cpu
) {
2497 iter
->buffer_iter
[cpu
] =
2498 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2500 ring_buffer_read_prepare_sync();
2501 for_each_tracing_cpu(cpu
) {
2502 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2503 tracing_iter_reset(iter
, cpu
);
2506 cpu
= iter
->cpu_file
;
2507 iter
->buffer_iter
[cpu
] =
2508 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2509 ring_buffer_read_prepare_sync();
2510 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2511 tracing_iter_reset(iter
, cpu
);
2514 mutex_unlock(&trace_types_lock
);
2519 mutex_unlock(&trace_types_lock
);
2521 kfree(iter
->buffer_iter
);
2523 seq_release_private(inode
, file
);
2524 return ERR_PTR(-ENOMEM
);
2527 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2529 if (tracing_disabled
)
2532 filp
->private_data
= inode
->i_private
;
2536 static int tracing_release(struct inode
*inode
, struct file
*file
)
2538 struct seq_file
*m
= file
->private_data
;
2539 struct trace_iterator
*iter
;
2542 if (!(file
->f_mode
& FMODE_READ
))
2547 mutex_lock(&trace_types_lock
);
2548 for_each_tracing_cpu(cpu
) {
2549 if (iter
->buffer_iter
[cpu
])
2550 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2553 if (iter
->trace
&& iter
->trace
->close
)
2554 iter
->trace
->close(iter
);
2556 /* reenable tracing if it was previously enabled */
2558 mutex_unlock(&trace_types_lock
);
2560 mutex_destroy(&iter
->mutex
);
2561 free_cpumask_var(iter
->started
);
2563 kfree(iter
->buffer_iter
);
2564 seq_release_private(inode
, file
);
2568 static int tracing_open(struct inode
*inode
, struct file
*file
)
2570 struct trace_iterator
*iter
;
2573 /* If this file was open for write, then erase contents */
2574 if ((file
->f_mode
& FMODE_WRITE
) &&
2575 (file
->f_flags
& O_TRUNC
)) {
2576 long cpu
= (long) inode
->i_private
;
2578 if (cpu
== TRACE_PIPE_ALL_CPU
)
2579 tracing_reset_online_cpus(&global_trace
);
2581 tracing_reset(&global_trace
, cpu
);
2584 if (file
->f_mode
& FMODE_READ
) {
2585 iter
= __tracing_open(inode
, file
);
2587 ret
= PTR_ERR(iter
);
2588 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2589 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2595 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2597 struct tracer
*t
= v
;
2607 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2612 mutex_lock(&trace_types_lock
);
2613 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2619 static void t_stop(struct seq_file
*m
, void *p
)
2621 mutex_unlock(&trace_types_lock
);
2624 static int t_show(struct seq_file
*m
, void *v
)
2626 struct tracer
*t
= v
;
2631 seq_printf(m
, "%s", t
->name
);
2640 static const struct seq_operations show_traces_seq_ops
= {
2647 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2649 if (tracing_disabled
)
2652 return seq_open(file
, &show_traces_seq_ops
);
2656 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2657 size_t count
, loff_t
*ppos
)
2662 static loff_t
tracing_seek(struct file
*file
, loff_t offset
, int origin
)
2664 if (file
->f_mode
& FMODE_READ
)
2665 return seq_lseek(file
, offset
, origin
);
2670 static const struct file_operations tracing_fops
= {
2671 .open
= tracing_open
,
2673 .write
= tracing_write_stub
,
2674 .llseek
= tracing_seek
,
2675 .release
= tracing_release
,
2678 static const struct file_operations show_traces_fops
= {
2679 .open
= show_traces_open
,
2681 .release
= seq_release
,
2682 .llseek
= seq_lseek
,
2686 * Only trace on a CPU if the bitmask is set:
2688 static cpumask_var_t tracing_cpumask
;
2691 * The tracer itself will not take this lock, but still we want
2692 * to provide a consistent cpumask to user-space:
2694 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2697 * Temporary storage for the character representation of the
2698 * CPU bitmask (and one more byte for the newline):
2700 static char mask_str
[NR_CPUS
+ 1];
2703 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2704 size_t count
, loff_t
*ppos
)
2708 mutex_lock(&tracing_cpumask_update_lock
);
2710 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2711 if (count
- len
< 2) {
2715 len
+= sprintf(mask_str
+ len
, "\n");
2716 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2719 mutex_unlock(&tracing_cpumask_update_lock
);
2725 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2726 size_t count
, loff_t
*ppos
)
2729 cpumask_var_t tracing_cpumask_new
;
2731 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2734 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2738 mutex_lock(&tracing_cpumask_update_lock
);
2740 local_irq_disable();
2741 arch_spin_lock(&ftrace_max_lock
);
2742 for_each_tracing_cpu(cpu
) {
2744 * Increase/decrease the disabled counter if we are
2745 * about to flip a bit in the cpumask:
2747 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2748 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2749 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2750 ring_buffer_record_disable_cpu(global_trace
.buffer
, cpu
);
2752 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2753 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2754 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2755 ring_buffer_record_enable_cpu(global_trace
.buffer
, cpu
);
2758 arch_spin_unlock(&ftrace_max_lock
);
2761 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2763 mutex_unlock(&tracing_cpumask_update_lock
);
2764 free_cpumask_var(tracing_cpumask_new
);
2769 free_cpumask_var(tracing_cpumask_new
);
2774 static const struct file_operations tracing_cpumask_fops
= {
2775 .open
= tracing_open_generic
,
2776 .read
= tracing_cpumask_read
,
2777 .write
= tracing_cpumask_write
,
2778 .llseek
= generic_file_llseek
,
2781 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
2783 struct tracer_opt
*trace_opts
;
2787 mutex_lock(&trace_types_lock
);
2788 tracer_flags
= current_trace
->flags
->val
;
2789 trace_opts
= current_trace
->flags
->opts
;
2791 for (i
= 0; trace_options
[i
]; i
++) {
2792 if (trace_flags
& (1 << i
))
2793 seq_printf(m
, "%s\n", trace_options
[i
]);
2795 seq_printf(m
, "no%s\n", trace_options
[i
]);
2798 for (i
= 0; trace_opts
[i
].name
; i
++) {
2799 if (tracer_flags
& trace_opts
[i
].bit
)
2800 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
2802 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
2804 mutex_unlock(&trace_types_lock
);
2809 static int __set_tracer_option(struct tracer
*trace
,
2810 struct tracer_flags
*tracer_flags
,
2811 struct tracer_opt
*opts
, int neg
)
2815 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
2820 tracer_flags
->val
&= ~opts
->bit
;
2822 tracer_flags
->val
|= opts
->bit
;
2826 /* Try to assign a tracer specific option */
2827 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2829 struct tracer_flags
*tracer_flags
= trace
->flags
;
2830 struct tracer_opt
*opts
= NULL
;
2833 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
2834 opts
= &tracer_flags
->opts
[i
];
2836 if (strcmp(cmp
, opts
->name
) == 0)
2837 return __set_tracer_option(trace
, trace
->flags
,
2844 static void set_tracer_flags(unsigned int mask
, int enabled
)
2846 /* do nothing if flag is already set */
2847 if (!!(trace_flags
& mask
) == !!enabled
)
2851 trace_flags
|= mask
;
2853 trace_flags
&= ~mask
;
2855 if (mask
== TRACE_ITER_RECORD_CMD
)
2856 trace_event_enable_cmd_record(enabled
);
2858 if (mask
== TRACE_ITER_OVERWRITE
)
2859 ring_buffer_change_overwrite(global_trace
.buffer
, enabled
);
2861 if (mask
== TRACE_ITER_PRINTK
)
2862 trace_printk_start_stop_comm(enabled
);
2865 static int trace_set_options(char *option
)
2872 cmp
= strstrip(option
);
2874 if (strncmp(cmp
, "no", 2) == 0) {
2879 for (i
= 0; trace_options
[i
]; i
++) {
2880 if (strcmp(cmp
, trace_options
[i
]) == 0) {
2881 set_tracer_flags(1 << i
, !neg
);
2886 /* If no option could be set, test the specific tracer options */
2887 if (!trace_options
[i
]) {
2888 mutex_lock(&trace_types_lock
);
2889 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2890 mutex_unlock(&trace_types_lock
);
2897 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2898 size_t cnt
, loff_t
*ppos
)
2902 if (cnt
>= sizeof(buf
))
2905 if (copy_from_user(&buf
, ubuf
, cnt
))
2910 trace_set_options(buf
);
2917 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
2919 if (tracing_disabled
)
2921 return single_open(file
, tracing_trace_options_show
, NULL
);
2924 static const struct file_operations tracing_iter_fops
= {
2925 .open
= tracing_trace_options_open
,
2927 .llseek
= seq_lseek
,
2928 .release
= single_release
,
2929 .write
= tracing_trace_options_write
,
2932 static const char readme_msg
[] =
2933 "tracing mini-HOWTO:\n\n"
2934 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2935 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2936 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2937 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2939 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2940 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2942 "# cat /sys/kernel/debug/tracing/trace_options\n"
2943 "noprint-parent nosym-offset nosym-addr noverbose\n"
2944 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2945 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2946 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2947 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2951 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2952 size_t cnt
, loff_t
*ppos
)
2954 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2955 readme_msg
, strlen(readme_msg
));
2958 static const struct file_operations tracing_readme_fops
= {
2959 .open
= tracing_open_generic
,
2960 .read
= tracing_readme_read
,
2961 .llseek
= generic_file_llseek
,
2965 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
2966 size_t cnt
, loff_t
*ppos
)
2975 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
2979 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
2987 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
2990 pid
= map_cmdline_to_pid
[i
];
2991 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
2994 trace_find_cmdline(pid
, buf_comm
);
2995 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
3000 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
3009 static const struct file_operations tracing_saved_cmdlines_fops
= {
3010 .open
= tracing_open_generic
,
3011 .read
= tracing_saved_cmdlines_read
,
3012 .llseek
= generic_file_llseek
,
3016 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
3017 size_t cnt
, loff_t
*ppos
)
3019 char buf
[MAX_TRACER_SIZE
+2];
3022 mutex_lock(&trace_types_lock
);
3024 r
= sprintf(buf
, "%s\n", current_trace
->name
);
3026 r
= sprintf(buf
, "\n");
3027 mutex_unlock(&trace_types_lock
);
3029 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3032 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
3034 tracing_reset_online_cpus(tr
);
3038 static void set_buffer_entries(struct trace_array
*tr
, unsigned long val
)
3041 for_each_tracing_cpu(cpu
)
3042 tr
->data
[cpu
]->entries
= val
;
3045 /* resize @tr's buffer to the size of @size_tr's entries */
3046 static int resize_buffer_duplicate_size(struct trace_array
*tr
,
3047 struct trace_array
*size_tr
, int cpu_id
)
3051 if (cpu_id
== RING_BUFFER_ALL_CPUS
) {
3052 for_each_tracing_cpu(cpu
) {
3053 ret
= ring_buffer_resize(tr
->buffer
,
3054 size_tr
->data
[cpu
]->entries
, cpu
);
3057 tr
->data
[cpu
]->entries
= size_tr
->data
[cpu
]->entries
;
3060 ret
= ring_buffer_resize(tr
->buffer
,
3061 size_tr
->data
[cpu_id
]->entries
, cpu_id
);
3063 tr
->data
[cpu_id
]->entries
=
3064 size_tr
->data
[cpu_id
]->entries
;
3070 static int __tracing_resize_ring_buffer(unsigned long size
, int cpu
)
3075 * If kernel or user changes the size of the ring buffer
3076 * we use the size that was given, and we can forget about
3077 * expanding it later.
3079 ring_buffer_expanded
= 1;
3081 /* May be called before buffers are initialized */
3082 if (!global_trace
.buffer
)
3085 ret
= ring_buffer_resize(global_trace
.buffer
, size
, cpu
);
3089 if (!current_trace
->use_max_tr
)
3092 ret
= ring_buffer_resize(max_tr
.buffer
, size
, cpu
);
3094 int r
= resize_buffer_duplicate_size(&global_trace
,
3095 &global_trace
, cpu
);
3098 * AARGH! We are left with different
3099 * size max buffer!!!!
3100 * The max buffer is our "snapshot" buffer.
3101 * When a tracer needs a snapshot (one of the
3102 * latency tracers), it swaps the max buffer
3103 * with the saved snap shot. We succeeded to
3104 * update the size of the main buffer, but failed to
3105 * update the size of the max buffer. But when we tried
3106 * to reset the main buffer to the original size, we
3107 * failed there too. This is very unlikely to
3108 * happen, but if it does, warn and kill all
3112 tracing_disabled
= 1;
3117 if (cpu
== RING_BUFFER_ALL_CPUS
)
3118 set_buffer_entries(&max_tr
, size
);
3120 max_tr
.data
[cpu
]->entries
= size
;
3123 if (cpu
== RING_BUFFER_ALL_CPUS
)
3124 set_buffer_entries(&global_trace
, size
);
3126 global_trace
.data
[cpu
]->entries
= size
;
3131 static ssize_t
tracing_resize_ring_buffer(unsigned long size
, int cpu_id
)
3135 mutex_lock(&trace_types_lock
);
3137 if (cpu_id
!= RING_BUFFER_ALL_CPUS
) {
3138 /* make sure, this cpu is enabled in the mask */
3139 if (!cpumask_test_cpu(cpu_id
, tracing_buffer_mask
)) {
3145 ret
= __tracing_resize_ring_buffer(size
, cpu_id
);
3150 mutex_unlock(&trace_types_lock
);
3157 * tracing_update_buffers - used by tracing facility to expand ring buffers
3159 * To save on memory when the tracing is never used on a system with it
3160 * configured in. The ring buffers are set to a minimum size. But once
3161 * a user starts to use the tracing facility, then they need to grow
3162 * to their default size.
3164 * This function is to be called when a tracer is about to be used.
3166 int tracing_update_buffers(void)
3170 mutex_lock(&trace_types_lock
);
3171 if (!ring_buffer_expanded
)
3172 ret
= __tracing_resize_ring_buffer(trace_buf_size
,
3173 RING_BUFFER_ALL_CPUS
);
3174 mutex_unlock(&trace_types_lock
);
3179 struct trace_option_dentry
;
3181 static struct trace_option_dentry
*
3182 create_trace_option_files(struct tracer
*tracer
);
3185 destroy_trace_option_files(struct trace_option_dentry
*topts
);
3187 static int tracing_set_tracer(const char *buf
)
3189 static struct trace_option_dentry
*topts
;
3190 struct trace_array
*tr
= &global_trace
;
3195 mutex_lock(&trace_types_lock
);
3197 if (!ring_buffer_expanded
) {
3198 ret
= __tracing_resize_ring_buffer(trace_buf_size
,
3199 RING_BUFFER_ALL_CPUS
);
3205 for (t
= trace_types
; t
; t
= t
->next
) {
3206 if (strcmp(t
->name
, buf
) == 0)
3213 if (t
== current_trace
)
3216 trace_branch_disable();
3217 if (current_trace
&& current_trace
->reset
)
3218 current_trace
->reset(tr
);
3220 had_max_tr
= current_trace
&& current_trace
->use_max_tr
;
3221 current_trace
= &nop_trace
;
3223 if (had_max_tr
&& !t
->use_max_tr
) {
3225 * We need to make sure that the update_max_tr sees that
3226 * current_trace changed to nop_trace to keep it from
3227 * swapping the buffers after we resize it.
3228 * The update_max_tr is called from interrupts disabled
3229 * so a synchronized_sched() is sufficient.
3231 synchronize_sched();
3233 * We don't free the ring buffer. instead, resize it because
3234 * The max_tr ring buffer has some state (e.g. ring->clock) and
3235 * we want preserve it.
3237 ring_buffer_resize(max_tr
.buffer
, 1, RING_BUFFER_ALL_CPUS
);
3238 set_buffer_entries(&max_tr
, 1);
3240 destroy_trace_option_files(topts
);
3242 topts
= create_trace_option_files(t
);
3243 if (t
->use_max_tr
&& !had_max_tr
) {
3244 /* we need to make per cpu buffer sizes equivalent */
3245 ret
= resize_buffer_duplicate_size(&max_tr
, &global_trace
,
3246 RING_BUFFER_ALL_CPUS
);
3252 ret
= tracer_init(t
, tr
);
3258 trace_branch_enable(tr
);
3260 mutex_unlock(&trace_types_lock
);
3266 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
3267 size_t cnt
, loff_t
*ppos
)
3269 char buf
[MAX_TRACER_SIZE
+1];
3276 if (cnt
> MAX_TRACER_SIZE
)
3277 cnt
= MAX_TRACER_SIZE
;
3279 if (copy_from_user(&buf
, ubuf
, cnt
))
3284 /* strip ending whitespace. */
3285 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
3288 err
= tracing_set_tracer(buf
);
3298 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
3299 size_t cnt
, loff_t
*ppos
)
3301 unsigned long *ptr
= filp
->private_data
;
3305 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
3306 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
3307 if (r
> sizeof(buf
))
3309 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3313 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
3314 size_t cnt
, loff_t
*ppos
)
3316 unsigned long *ptr
= filp
->private_data
;
3320 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3329 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
3331 long cpu_file
= (long) inode
->i_private
;
3332 struct trace_iterator
*iter
;
3335 if (tracing_disabled
)
3338 mutex_lock(&trace_types_lock
);
3340 /* create a buffer to store the information to pass to userspace */
3341 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
3348 * We make a copy of the current tracer to avoid concurrent
3349 * changes on it while we are reading.
3351 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
3357 *iter
->trace
= *current_trace
;
3359 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
3364 /* trace pipe does not show start of buffer */
3365 cpumask_setall(iter
->started
);
3367 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
3368 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
3370 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3371 if (trace_clocks
[trace_clock_id
].in_ns
)
3372 iter
->iter_flags
|= TRACE_FILE_TIME_IN_NS
;
3374 iter
->cpu_file
= cpu_file
;
3375 iter
->tr
= &global_trace
;
3376 mutex_init(&iter
->mutex
);
3377 filp
->private_data
= iter
;
3379 if (iter
->trace
->pipe_open
)
3380 iter
->trace
->pipe_open(iter
);
3382 nonseekable_open(inode
, filp
);
3384 mutex_unlock(&trace_types_lock
);
3390 mutex_unlock(&trace_types_lock
);
3394 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
3396 struct trace_iterator
*iter
= file
->private_data
;
3398 mutex_lock(&trace_types_lock
);
3400 if (iter
->trace
->pipe_close
)
3401 iter
->trace
->pipe_close(iter
);
3403 mutex_unlock(&trace_types_lock
);
3405 free_cpumask_var(iter
->started
);
3406 mutex_destroy(&iter
->mutex
);
3414 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
3416 struct trace_iterator
*iter
= filp
->private_data
;
3418 if (trace_flags
& TRACE_ITER_BLOCK
) {
3420 * Always select as readable when in blocking mode
3422 return POLLIN
| POLLRDNORM
;
3424 if (!trace_empty(iter
))
3425 return POLLIN
| POLLRDNORM
;
3426 poll_wait(filp
, &trace_wait
, poll_table
);
3427 if (!trace_empty(iter
))
3428 return POLLIN
| POLLRDNORM
;
3435 * This is a make-shift waitqueue.
3436 * A tracer might use this callback on some rare cases:
3438 * 1) the current tracer might hold the runqueue lock when it wakes up
3439 * a reader, hence a deadlock (sched, function, and function graph tracers)
3440 * 2) the function tracers, trace all functions, we don't want
3441 * the overhead of calling wake_up and friends
3442 * (and tracing them too)
3444 * Anyway, this is really very primitive wakeup.
3446 void poll_wait_pipe(struct trace_iterator
*iter
)
3448 set_current_state(TASK_INTERRUPTIBLE
);
3449 /* sleep for 100 msecs, and try again. */
3450 schedule_timeout(HZ
/ 10);
3453 /* Must be called with trace_types_lock mutex held. */
3454 static int tracing_wait_pipe(struct file
*filp
)
3456 struct trace_iterator
*iter
= filp
->private_data
;
3458 while (trace_empty(iter
)) {
3460 if ((filp
->f_flags
& O_NONBLOCK
)) {
3464 mutex_unlock(&iter
->mutex
);
3466 iter
->trace
->wait_pipe(iter
);
3468 mutex_lock(&iter
->mutex
);
3470 if (signal_pending(current
))
3474 * We block until we read something and tracing is disabled.
3475 * We still block if tracing is disabled, but we have never
3476 * read anything. This allows a user to cat this file, and
3477 * then enable tracing. But after we have read something,
3478 * we give an EOF when tracing is again disabled.
3480 * iter->pos will be 0 if we haven't read anything.
3482 if (!tracing_is_enabled() && iter
->pos
)
3493 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
3494 size_t cnt
, loff_t
*ppos
)
3496 struct trace_iterator
*iter
= filp
->private_data
;
3497 static struct tracer
*old_tracer
;
3500 /* return any leftover data */
3501 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3505 trace_seq_init(&iter
->seq
);
3507 /* copy the tracer to avoid using a global lock all around */
3508 mutex_lock(&trace_types_lock
);
3509 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3510 old_tracer
= current_trace
;
3511 *iter
->trace
= *current_trace
;
3513 mutex_unlock(&trace_types_lock
);
3516 * Avoid more than one consumer on a single file descriptor
3517 * This is just a matter of traces coherency, the ring buffer itself
3520 mutex_lock(&iter
->mutex
);
3521 if (iter
->trace
->read
) {
3522 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
3528 sret
= tracing_wait_pipe(filp
);
3532 /* stop when tracing is finished */
3533 if (trace_empty(iter
)) {
3538 if (cnt
>= PAGE_SIZE
)
3539 cnt
= PAGE_SIZE
- 1;
3541 /* reset all but tr, trace, and overruns */
3542 memset(&iter
->seq
, 0,
3543 sizeof(struct trace_iterator
) -
3544 offsetof(struct trace_iterator
, seq
));
3547 trace_event_read_lock();
3548 trace_access_lock(iter
->cpu_file
);
3549 while (trace_find_next_entry_inc(iter
) != NULL
) {
3550 enum print_line_t ret
;
3551 int len
= iter
->seq
.len
;
3553 ret
= print_trace_line(iter
);
3554 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3555 /* don't print partial lines */
3556 iter
->seq
.len
= len
;
3559 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3560 trace_consume(iter
);
3562 if (iter
->seq
.len
>= cnt
)
3566 * Setting the full flag means we reached the trace_seq buffer
3567 * size and we should leave by partial output condition above.
3568 * One of the trace_seq_* functions is not used properly.
3570 WARN_ONCE(iter
->seq
.full
, "full flag set for trace type %d",
3573 trace_access_unlock(iter
->cpu_file
);
3574 trace_event_read_unlock();
3576 /* Now copy what we have to the user */
3577 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3578 if (iter
->seq
.readpos
>= iter
->seq
.len
)
3579 trace_seq_init(&iter
->seq
);
3582 * If there was nothing to send to user, in spite of consuming trace
3583 * entries, go back to wait for more entries.
3589 mutex_unlock(&iter
->mutex
);
3594 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
3595 struct pipe_buffer
*buf
)
3597 __free_page(buf
->page
);
3600 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
3603 __free_page(spd
->pages
[idx
]);
3606 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
3608 .map
= generic_pipe_buf_map
,
3609 .unmap
= generic_pipe_buf_unmap
,
3610 .confirm
= generic_pipe_buf_confirm
,
3611 .release
= tracing_pipe_buf_release
,
3612 .steal
= generic_pipe_buf_steal
,
3613 .get
= generic_pipe_buf_get
,
3617 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
3622 /* Seq buffer is page-sized, exactly what we need. */
3624 count
= iter
->seq
.len
;
3625 ret
= print_trace_line(iter
);
3626 count
= iter
->seq
.len
- count
;
3629 iter
->seq
.len
-= count
;
3632 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3633 iter
->seq
.len
-= count
;
3637 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3638 trace_consume(iter
);
3640 if (!trace_find_next_entry_inc(iter
)) {
3650 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3652 struct pipe_inode_info
*pipe
,
3656 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3657 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3658 struct trace_iterator
*iter
= filp
->private_data
;
3659 struct splice_pipe_desc spd
= {
3661 .partial
= partial_def
,
3662 .nr_pages
= 0, /* This gets updated below. */
3663 .nr_pages_max
= PIPE_DEF_BUFFERS
,
3665 .ops
= &tracing_pipe_buf_ops
,
3666 .spd_release
= tracing_spd_release_pipe
,
3668 static struct tracer
*old_tracer
;
3673 if (splice_grow_spd(pipe
, &spd
))
3676 /* copy the tracer to avoid using a global lock all around */
3677 mutex_lock(&trace_types_lock
);
3678 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3679 old_tracer
= current_trace
;
3680 *iter
->trace
= *current_trace
;
3682 mutex_unlock(&trace_types_lock
);
3684 mutex_lock(&iter
->mutex
);
3686 if (iter
->trace
->splice_read
) {
3687 ret
= iter
->trace
->splice_read(iter
, filp
,
3688 ppos
, pipe
, len
, flags
);
3693 ret
= tracing_wait_pipe(filp
);
3697 if (!iter
->ent
&& !trace_find_next_entry_inc(iter
)) {
3702 trace_event_read_lock();
3703 trace_access_lock(iter
->cpu_file
);
3705 /* Fill as many pages as possible. */
3706 for (i
= 0, rem
= len
; i
< pipe
->buffers
&& rem
; i
++) {
3707 spd
.pages
[i
] = alloc_page(GFP_KERNEL
);
3711 rem
= tracing_fill_pipe_page(rem
, iter
);
3713 /* Copy the data into the page, so we can start over. */
3714 ret
= trace_seq_to_buffer(&iter
->seq
,
3715 page_address(spd
.pages
[i
]),
3718 __free_page(spd
.pages
[i
]);
3721 spd
.partial
[i
].offset
= 0;
3722 spd
.partial
[i
].len
= iter
->seq
.len
;
3724 trace_seq_init(&iter
->seq
);
3727 trace_access_unlock(iter
->cpu_file
);
3728 trace_event_read_unlock();
3729 mutex_unlock(&iter
->mutex
);
3733 ret
= splice_to_pipe(pipe
, &spd
);
3735 splice_shrink_spd(&spd
);
3739 mutex_unlock(&iter
->mutex
);
3743 struct ftrace_entries_info
{
3744 struct trace_array
*tr
;
3748 static int tracing_entries_open(struct inode
*inode
, struct file
*filp
)
3750 struct ftrace_entries_info
*info
;
3752 if (tracing_disabled
)
3755 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3759 info
->tr
= &global_trace
;
3760 info
->cpu
= (unsigned long)inode
->i_private
;
3762 filp
->private_data
= info
;
3768 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3769 size_t cnt
, loff_t
*ppos
)
3771 struct ftrace_entries_info
*info
= filp
->private_data
;
3772 struct trace_array
*tr
= info
->tr
;
3777 mutex_lock(&trace_types_lock
);
3779 if (info
->cpu
== RING_BUFFER_ALL_CPUS
) {
3780 int cpu
, buf_size_same
;
3785 /* check if all cpu sizes are same */
3786 for_each_tracing_cpu(cpu
) {
3787 /* fill in the size from first enabled cpu */
3789 size
= tr
->data
[cpu
]->entries
;
3790 if (size
!= tr
->data
[cpu
]->entries
) {
3796 if (buf_size_same
) {
3797 if (!ring_buffer_expanded
)
3798 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3800 trace_buf_size
>> 10);
3802 r
= sprintf(buf
, "%lu\n", size
>> 10);
3804 r
= sprintf(buf
, "X\n");
3806 r
= sprintf(buf
, "%lu\n", tr
->data
[info
->cpu
]->entries
>> 10);
3808 mutex_unlock(&trace_types_lock
);
3810 ret
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3815 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3816 size_t cnt
, loff_t
*ppos
)
3818 struct ftrace_entries_info
*info
= filp
->private_data
;
3822 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3826 /* must have at least 1 entry */
3830 /* value is in KB */
3833 ret
= tracing_resize_ring_buffer(val
, info
->cpu
);
3843 tracing_entries_release(struct inode
*inode
, struct file
*filp
)
3845 struct ftrace_entries_info
*info
= filp
->private_data
;
3853 tracing_total_entries_read(struct file
*filp
, char __user
*ubuf
,
3854 size_t cnt
, loff_t
*ppos
)
3856 struct trace_array
*tr
= filp
->private_data
;
3859 unsigned long size
= 0, expanded_size
= 0;
3861 mutex_lock(&trace_types_lock
);
3862 for_each_tracing_cpu(cpu
) {
3863 size
+= tr
->data
[cpu
]->entries
>> 10;
3864 if (!ring_buffer_expanded
)
3865 expanded_size
+= trace_buf_size
>> 10;
3867 if (ring_buffer_expanded
)
3868 r
= sprintf(buf
, "%lu\n", size
);
3870 r
= sprintf(buf
, "%lu (expanded: %lu)\n", size
, expanded_size
);
3871 mutex_unlock(&trace_types_lock
);
3873 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3877 tracing_free_buffer_write(struct file
*filp
, const char __user
*ubuf
,
3878 size_t cnt
, loff_t
*ppos
)
3881 * There is no need to read what the user has written, this function
3882 * is just to make sure that there is no error when "echo" is used
3891 tracing_free_buffer_release(struct inode
*inode
, struct file
*filp
)
3893 /* disable tracing ? */
3894 if (trace_flags
& TRACE_ITER_STOP_ON_FREE
)
3896 /* resize the ring buffer to 0 */
3897 tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS
);
3903 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3904 size_t cnt
, loff_t
*fpos
)
3906 unsigned long addr
= (unsigned long)ubuf
;
3907 struct ring_buffer_event
*event
;
3908 struct ring_buffer
*buffer
;
3909 struct print_entry
*entry
;
3910 unsigned long irq_flags
;
3911 struct page
*pages
[2];
3921 if (tracing_disabled
)
3924 if (!(trace_flags
& TRACE_ITER_MARKERS
))
3927 if (cnt
> TRACE_BUF_SIZE
)
3928 cnt
= TRACE_BUF_SIZE
;
3931 * Userspace is injecting traces into the kernel trace buffer.
3932 * We want to be as non intrusive as possible.
3933 * To do so, we do not want to allocate any special buffers
3934 * or take any locks, but instead write the userspace data
3935 * straight into the ring buffer.
3937 * First we need to pin the userspace buffer into memory,
3938 * which, most likely it is, because it just referenced it.
3939 * But there's no guarantee that it is. By using get_user_pages_fast()
3940 * and kmap_atomic/kunmap_atomic() we can get access to the
3941 * pages directly. We then write the data directly into the
3944 BUILD_BUG_ON(TRACE_BUF_SIZE
>= PAGE_SIZE
);
3946 /* check if we cross pages */
3947 if ((addr
& PAGE_MASK
) != ((addr
+ cnt
) & PAGE_MASK
))
3950 offset
= addr
& (PAGE_SIZE
- 1);
3953 ret
= get_user_pages_fast(addr
, nr_pages
, 0, pages
);
3954 if (ret
< nr_pages
) {
3956 put_page(pages
[ret
]);
3961 for (i
= 0; i
< nr_pages
; i
++)
3962 map_page
[i
] = kmap_atomic(pages
[i
]);
3964 local_save_flags(irq_flags
);
3965 size
= sizeof(*entry
) + cnt
+ 2; /* possible \n added */
3966 buffer
= global_trace
.buffer
;
3967 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
3968 irq_flags
, preempt_count());
3970 /* Ring buffer disabled, return as if not open for write */
3975 entry
= ring_buffer_event_data(event
);
3976 entry
->ip
= _THIS_IP_
;
3978 if (nr_pages
== 2) {
3979 len
= PAGE_SIZE
- offset
;
3980 memcpy(&entry
->buf
, map_page
[0] + offset
, len
);
3981 memcpy(&entry
->buf
[len
], map_page
[1], cnt
- len
);
3983 memcpy(&entry
->buf
, map_page
[0] + offset
, cnt
);
3985 if (entry
->buf
[cnt
- 1] != '\n') {
3986 entry
->buf
[cnt
] = '\n';
3987 entry
->buf
[cnt
+ 1] = '\0';
3989 entry
->buf
[cnt
] = '\0';
3991 __buffer_unlock_commit(buffer
, event
);
3998 for (i
= 0; i
< nr_pages
; i
++){
3999 kunmap_atomic(map_page
[i
]);
4006 static int tracing_clock_show(struct seq_file
*m
, void *v
)
4010 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
4012 "%s%s%s%s", i
? " " : "",
4013 i
== trace_clock_id
? "[" : "", trace_clocks
[i
].name
,
4014 i
== trace_clock_id
? "]" : "");
4020 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
4021 size_t cnt
, loff_t
*fpos
)
4024 const char *clockstr
;
4027 if (cnt
>= sizeof(buf
))
4030 if (copy_from_user(&buf
, ubuf
, cnt
))
4035 clockstr
= strstrip(buf
);
4037 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
4038 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
4041 if (i
== ARRAY_SIZE(trace_clocks
))
4046 mutex_lock(&trace_types_lock
);
4048 ring_buffer_set_clock(global_trace
.buffer
, trace_clocks
[i
].func
);
4050 ring_buffer_set_clock(max_tr
.buffer
, trace_clocks
[i
].func
);
4053 * New clock may not be consistent with the previous clock.
4054 * Reset the buffer so that it doesn't have incomparable timestamps.
4056 tracing_reset_online_cpus(&global_trace
);
4057 tracing_reset_online_cpus(&max_tr
);
4059 mutex_unlock(&trace_types_lock
);
4066 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
4068 if (tracing_disabled
)
4070 return single_open(file
, tracing_clock_show
, NULL
);
4073 static const struct file_operations tracing_max_lat_fops
= {
4074 .open
= tracing_open_generic
,
4075 .read
= tracing_max_lat_read
,
4076 .write
= tracing_max_lat_write
,
4077 .llseek
= generic_file_llseek
,
4080 static const struct file_operations set_tracer_fops
= {
4081 .open
= tracing_open_generic
,
4082 .read
= tracing_set_trace_read
,
4083 .write
= tracing_set_trace_write
,
4084 .llseek
= generic_file_llseek
,
4087 static const struct file_operations tracing_pipe_fops
= {
4088 .open
= tracing_open_pipe
,
4089 .poll
= tracing_poll_pipe
,
4090 .read
= tracing_read_pipe
,
4091 .splice_read
= tracing_splice_read_pipe
,
4092 .release
= tracing_release_pipe
,
4093 .llseek
= no_llseek
,
4096 static const struct file_operations tracing_entries_fops
= {
4097 .open
= tracing_entries_open
,
4098 .read
= tracing_entries_read
,
4099 .write
= tracing_entries_write
,
4100 .release
= tracing_entries_release
,
4101 .llseek
= generic_file_llseek
,
4104 static const struct file_operations tracing_total_entries_fops
= {
4105 .open
= tracing_open_generic
,
4106 .read
= tracing_total_entries_read
,
4107 .llseek
= generic_file_llseek
,
4110 static const struct file_operations tracing_free_buffer_fops
= {
4111 .write
= tracing_free_buffer_write
,
4112 .release
= tracing_free_buffer_release
,
4115 static const struct file_operations tracing_mark_fops
= {
4116 .open
= tracing_open_generic
,
4117 .write
= tracing_mark_write
,
4118 .llseek
= generic_file_llseek
,
4121 static const struct file_operations trace_clock_fops
= {
4122 .open
= tracing_clock_open
,
4124 .llseek
= seq_lseek
,
4125 .release
= single_release
,
4126 .write
= tracing_clock_write
,
4129 struct ftrace_buffer_info
{
4130 struct trace_array
*tr
;
4136 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
4138 int cpu
= (int)(long)inode
->i_private
;
4139 struct ftrace_buffer_info
*info
;
4141 if (tracing_disabled
)
4144 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
4148 info
->tr
= &global_trace
;
4151 /* Force reading ring buffer for first read */
4152 info
->read
= (unsigned int)-1;
4154 filp
->private_data
= info
;
4156 return nonseekable_open(inode
, filp
);
4160 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
4161 size_t count
, loff_t
*ppos
)
4163 struct ftrace_buffer_info
*info
= filp
->private_data
;
4171 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
, info
->cpu
);
4175 /* Do we have previous read data to read? */
4176 if (info
->read
< PAGE_SIZE
)
4179 trace_access_lock(info
->cpu
);
4180 ret
= ring_buffer_read_page(info
->tr
->buffer
,
4184 trace_access_unlock(info
->cpu
);
4191 size
= PAGE_SIZE
- info
->read
;
4195 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
4206 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
4208 struct ftrace_buffer_info
*info
= file
->private_data
;
4211 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
4218 struct ring_buffer
*buffer
;
4223 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
4224 struct pipe_buffer
*buf
)
4226 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
4231 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4236 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
4237 struct pipe_buffer
*buf
)
4239 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
4244 /* Pipe buffer operations for a buffer. */
4245 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
4247 .map
= generic_pipe_buf_map
,
4248 .unmap
= generic_pipe_buf_unmap
,
4249 .confirm
= generic_pipe_buf_confirm
,
4250 .release
= buffer_pipe_buf_release
,
4251 .steal
= generic_pipe_buf_steal
,
4252 .get
= buffer_pipe_buf_get
,
4256 * Callback from splice_to_pipe(), if we need to release some pages
4257 * at the end of the spd in case we error'ed out in filling the pipe.
4259 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
4261 struct buffer_ref
*ref
=
4262 (struct buffer_ref
*)spd
->partial
[i
].private;
4267 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4269 spd
->partial
[i
].private = 0;
4273 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
4274 struct pipe_inode_info
*pipe
, size_t len
,
4277 struct ftrace_buffer_info
*info
= file
->private_data
;
4278 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
4279 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
4280 struct splice_pipe_desc spd
= {
4282 .partial
= partial_def
,
4283 .nr_pages_max
= PIPE_DEF_BUFFERS
,
4285 .ops
= &buffer_pipe_buf_ops
,
4286 .spd_release
= buffer_spd_release
,
4288 struct buffer_ref
*ref
;
4289 int entries
, size
, i
;
4292 if (splice_grow_spd(pipe
, &spd
))
4295 if (*ppos
& (PAGE_SIZE
- 1)) {
4300 if (len
& (PAGE_SIZE
- 1)) {
4301 if (len
< PAGE_SIZE
) {
4308 trace_access_lock(info
->cpu
);
4309 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4311 for (i
= 0; i
< pipe
->buffers
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
4315 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
4320 ref
->buffer
= info
->tr
->buffer
;
4321 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
, info
->cpu
);
4327 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
4330 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4336 * zero out any left over data, this is going to
4339 size
= ring_buffer_page_len(ref
->page
);
4340 if (size
< PAGE_SIZE
)
4341 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
4343 page
= virt_to_page(ref
->page
);
4345 spd
.pages
[i
] = page
;
4346 spd
.partial
[i
].len
= PAGE_SIZE
;
4347 spd
.partial
[i
].offset
= 0;
4348 spd
.partial
[i
].private = (unsigned long)ref
;
4352 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4355 trace_access_unlock(info
->cpu
);
4358 /* did we read anything? */
4359 if (!spd
.nr_pages
) {
4360 if (flags
& SPLICE_F_NONBLOCK
)
4368 ret
= splice_to_pipe(pipe
, &spd
);
4369 splice_shrink_spd(&spd
);
4374 static const struct file_operations tracing_buffers_fops
= {
4375 .open
= tracing_buffers_open
,
4376 .read
= tracing_buffers_read
,
4377 .release
= tracing_buffers_release
,
4378 .splice_read
= tracing_buffers_splice_read
,
4379 .llseek
= no_llseek
,
4383 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
4384 size_t count
, loff_t
*ppos
)
4386 unsigned long cpu
= (unsigned long)filp
->private_data
;
4387 struct trace_array
*tr
= &global_trace
;
4388 struct trace_seq
*s
;
4390 unsigned long long t
;
4391 unsigned long usec_rem
;
4393 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
4399 cnt
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
4400 trace_seq_printf(s
, "entries: %ld\n", cnt
);
4402 cnt
= ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
4403 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
4405 cnt
= ring_buffer_commit_overrun_cpu(tr
->buffer
, cpu
);
4406 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
4408 cnt
= ring_buffer_bytes_cpu(tr
->buffer
, cpu
);
4409 trace_seq_printf(s
, "bytes: %ld\n", cnt
);
4411 if (trace_clocks
[trace_clock_id
].in_ns
) {
4412 /* local or global for trace_clock */
4413 t
= ns2usecs(ring_buffer_oldest_event_ts(tr
->buffer
, cpu
));
4414 usec_rem
= do_div(t
, USEC_PER_SEC
);
4415 trace_seq_printf(s
, "oldest event ts: %5llu.%06lu\n",
4418 t
= ns2usecs(ring_buffer_time_stamp(tr
->buffer
, cpu
));
4419 usec_rem
= do_div(t
, USEC_PER_SEC
);
4420 trace_seq_printf(s
, "now ts: %5llu.%06lu\n", t
, usec_rem
);
4422 /* counter or tsc mode for trace_clock */
4423 trace_seq_printf(s
, "oldest event ts: %llu\n",
4424 ring_buffer_oldest_event_ts(tr
->buffer
, cpu
));
4426 trace_seq_printf(s
, "now ts: %llu\n",
4427 ring_buffer_time_stamp(tr
->buffer
, cpu
));
4430 cnt
= ring_buffer_dropped_events_cpu(tr
->buffer
, cpu
);
4431 trace_seq_printf(s
, "dropped events: %ld\n", cnt
);
4433 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
4440 static const struct file_operations tracing_stats_fops
= {
4441 .open
= tracing_open_generic
,
4442 .read
= tracing_stats_read
,
4443 .llseek
= generic_file_llseek
,
4446 #ifdef CONFIG_DYNAMIC_FTRACE
4448 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
4454 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
4455 size_t cnt
, loff_t
*ppos
)
4457 static char ftrace_dyn_info_buffer
[1024];
4458 static DEFINE_MUTEX(dyn_info_mutex
);
4459 unsigned long *p
= filp
->private_data
;
4460 char *buf
= ftrace_dyn_info_buffer
;
4461 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
4464 mutex_lock(&dyn_info_mutex
);
4465 r
= sprintf(buf
, "%ld ", *p
);
4467 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
4470 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4472 mutex_unlock(&dyn_info_mutex
);
4477 static const struct file_operations tracing_dyn_info_fops
= {
4478 .open
= tracing_open_generic
,
4479 .read
= tracing_read_dyn_info
,
4480 .llseek
= generic_file_llseek
,
4484 static struct dentry
*d_tracer
;
4486 struct dentry
*tracing_init_dentry(void)
4493 if (!debugfs_initialized())
4496 d_tracer
= debugfs_create_dir("tracing", NULL
);
4498 if (!d_tracer
&& !once
) {
4500 pr_warning("Could not create debugfs directory 'tracing'\n");
4507 static struct dentry
*d_percpu
;
4509 struct dentry
*tracing_dentry_percpu(void)
4512 struct dentry
*d_tracer
;
4517 d_tracer
= tracing_init_dentry();
4522 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
4524 if (!d_percpu
&& !once
) {
4526 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4533 static void tracing_init_debugfs_percpu(long cpu
)
4535 struct dentry
*d_percpu
= tracing_dentry_percpu();
4536 struct dentry
*d_cpu
;
4537 char cpu_dir
[30]; /* 30 characters should be more than enough */
4542 snprintf(cpu_dir
, 30, "cpu%ld", cpu
);
4543 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
4545 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
4549 /* per cpu trace_pipe */
4550 trace_create_file("trace_pipe", 0444, d_cpu
,
4551 (void *) cpu
, &tracing_pipe_fops
);
4554 trace_create_file("trace", 0644, d_cpu
,
4555 (void *) cpu
, &tracing_fops
);
4557 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
4558 (void *) cpu
, &tracing_buffers_fops
);
4560 trace_create_file("stats", 0444, d_cpu
,
4561 (void *) cpu
, &tracing_stats_fops
);
4563 trace_create_file("buffer_size_kb", 0444, d_cpu
,
4564 (void *) cpu
, &tracing_entries_fops
);
4567 #ifdef CONFIG_FTRACE_SELFTEST
4568 /* Let selftest have access to static functions in this file */
4569 #include "trace_selftest.c"
4572 struct trace_option_dentry
{
4573 struct tracer_opt
*opt
;
4574 struct tracer_flags
*flags
;
4575 struct dentry
*entry
;
4579 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4582 struct trace_option_dentry
*topt
= filp
->private_data
;
4585 if (topt
->flags
->val
& topt
->opt
->bit
)
4590 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4594 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4597 struct trace_option_dentry
*topt
= filp
->private_data
;
4601 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4605 if (val
!= 0 && val
!= 1)
4608 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
4609 mutex_lock(&trace_types_lock
);
4610 ret
= __set_tracer_option(current_trace
, topt
->flags
,
4612 mutex_unlock(&trace_types_lock
);
4623 static const struct file_operations trace_options_fops
= {
4624 .open
= tracing_open_generic
,
4625 .read
= trace_options_read
,
4626 .write
= trace_options_write
,
4627 .llseek
= generic_file_llseek
,
4631 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4634 long index
= (long)filp
->private_data
;
4637 if (trace_flags
& (1 << index
))
4642 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4646 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4649 long index
= (long)filp
->private_data
;
4653 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4657 if (val
!= 0 && val
!= 1)
4659 set_tracer_flags(1 << index
, val
);
4666 static const struct file_operations trace_options_core_fops
= {
4667 .open
= tracing_open_generic
,
4668 .read
= trace_options_core_read
,
4669 .write
= trace_options_core_write
,
4670 .llseek
= generic_file_llseek
,
4673 struct dentry
*trace_create_file(const char *name
,
4675 struct dentry
*parent
,
4677 const struct file_operations
*fops
)
4681 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
4683 pr_warning("Could not create debugfs '%s' entry\n", name
);
4689 static struct dentry
*trace_options_init_dentry(void)
4691 struct dentry
*d_tracer
;
4692 static struct dentry
*t_options
;
4697 d_tracer
= tracing_init_dentry();
4701 t_options
= debugfs_create_dir("options", d_tracer
);
4703 pr_warning("Could not create debugfs directory 'options'\n");
4711 create_trace_option_file(struct trace_option_dentry
*topt
,
4712 struct tracer_flags
*flags
,
4713 struct tracer_opt
*opt
)
4715 struct dentry
*t_options
;
4717 t_options
= trace_options_init_dentry();
4721 topt
->flags
= flags
;
4724 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
4725 &trace_options_fops
);
4729 static struct trace_option_dentry
*
4730 create_trace_option_files(struct tracer
*tracer
)
4732 struct trace_option_dentry
*topts
;
4733 struct tracer_flags
*flags
;
4734 struct tracer_opt
*opts
;
4740 flags
= tracer
->flags
;
4742 if (!flags
|| !flags
->opts
)
4747 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4750 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
4754 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4755 create_trace_option_file(&topts
[cnt
], flags
,
4762 destroy_trace_option_files(struct trace_option_dentry
*topts
)
4769 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
4770 if (topts
[cnt
].entry
)
4771 debugfs_remove(topts
[cnt
].entry
);
4777 static struct dentry
*
4778 create_trace_option_core_file(const char *option
, long index
)
4780 struct dentry
*t_options
;
4782 t_options
= trace_options_init_dentry();
4786 return trace_create_file(option
, 0644, t_options
, (void *)index
,
4787 &trace_options_core_fops
);
4790 static __init
void create_trace_options_dir(void)
4792 struct dentry
*t_options
;
4795 t_options
= trace_options_init_dentry();
4799 for (i
= 0; trace_options
[i
]; i
++)
4800 create_trace_option_core_file(trace_options
[i
], i
);
4804 rb_simple_read(struct file
*filp
, char __user
*ubuf
,
4805 size_t cnt
, loff_t
*ppos
)
4807 struct trace_array
*tr
= filp
->private_data
;
4808 struct ring_buffer
*buffer
= tr
->buffer
;
4813 r
= ring_buffer_record_is_on(buffer
);
4817 r
= sprintf(buf
, "%d\n", r
);
4819 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4823 rb_simple_write(struct file
*filp
, const char __user
*ubuf
,
4824 size_t cnt
, loff_t
*ppos
)
4826 struct trace_array
*tr
= filp
->private_data
;
4827 struct ring_buffer
*buffer
= tr
->buffer
;
4831 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4836 mutex_lock(&trace_types_lock
);
4838 ring_buffer_record_on(buffer
);
4839 if (current_trace
->start
)
4840 current_trace
->start(tr
);
4842 ring_buffer_record_off(buffer
);
4843 if (current_trace
->stop
)
4844 current_trace
->stop(tr
);
4846 mutex_unlock(&trace_types_lock
);
4854 static const struct file_operations rb_simple_fops
= {
4855 .open
= tracing_open_generic
,
4856 .read
= rb_simple_read
,
4857 .write
= rb_simple_write
,
4858 .llseek
= default_llseek
,
4861 static __init
int tracer_init_debugfs(void)
4863 struct dentry
*d_tracer
;
4866 trace_access_lock_init();
4868 d_tracer
= tracing_init_dentry();
4870 trace_create_file("trace_options", 0644, d_tracer
,
4871 NULL
, &tracing_iter_fops
);
4873 trace_create_file("tracing_cpumask", 0644, d_tracer
,
4874 NULL
, &tracing_cpumask_fops
);
4876 trace_create_file("trace", 0644, d_tracer
,
4877 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
4879 trace_create_file("available_tracers", 0444, d_tracer
,
4880 &global_trace
, &show_traces_fops
);
4882 trace_create_file("current_tracer", 0644, d_tracer
,
4883 &global_trace
, &set_tracer_fops
);
4885 #ifdef CONFIG_TRACER_MAX_TRACE
4886 trace_create_file("tracing_max_latency", 0644, d_tracer
,
4887 &tracing_max_latency
, &tracing_max_lat_fops
);
4890 trace_create_file("tracing_thresh", 0644, d_tracer
,
4891 &tracing_thresh
, &tracing_max_lat_fops
);
4893 trace_create_file("README", 0444, d_tracer
,
4894 NULL
, &tracing_readme_fops
);
4896 trace_create_file("trace_pipe", 0444, d_tracer
,
4897 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
4899 trace_create_file("buffer_size_kb", 0644, d_tracer
,
4900 (void *) RING_BUFFER_ALL_CPUS
, &tracing_entries_fops
);
4902 trace_create_file("buffer_total_size_kb", 0444, d_tracer
,
4903 &global_trace
, &tracing_total_entries_fops
);
4905 trace_create_file("free_buffer", 0644, d_tracer
,
4906 &global_trace
, &tracing_free_buffer_fops
);
4908 trace_create_file("trace_marker", 0220, d_tracer
,
4909 NULL
, &tracing_mark_fops
);
4911 trace_create_file("saved_cmdlines", 0444, d_tracer
,
4912 NULL
, &tracing_saved_cmdlines_fops
);
4914 trace_create_file("trace_clock", 0644, d_tracer
, NULL
,
4917 trace_create_file("tracing_on", 0644, d_tracer
,
4918 &global_trace
, &rb_simple_fops
);
4920 #ifdef CONFIG_DYNAMIC_FTRACE
4921 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
4922 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
4925 create_trace_options_dir();
4927 for_each_tracing_cpu(cpu
)
4928 tracing_init_debugfs_percpu(cpu
);
4933 static int trace_panic_handler(struct notifier_block
*this,
4934 unsigned long event
, void *unused
)
4936 if (ftrace_dump_on_oops
)
4937 ftrace_dump(ftrace_dump_on_oops
);
4941 static struct notifier_block trace_panic_notifier
= {
4942 .notifier_call
= trace_panic_handler
,
4944 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
4947 static int trace_die_handler(struct notifier_block
*self
,
4953 if (ftrace_dump_on_oops
)
4954 ftrace_dump(ftrace_dump_on_oops
);
4962 static struct notifier_block trace_die_notifier
= {
4963 .notifier_call
= trace_die_handler
,
4968 * printk is set to max of 1024, we really don't need it that big.
4969 * Nothing should be printing 1000 characters anyway.
4971 #define TRACE_MAX_PRINT 1000
4974 * Define here KERN_TRACE so that we have one place to modify
4975 * it if we decide to change what log level the ftrace dump
4978 #define KERN_TRACE KERN_EMERG
4981 trace_printk_seq(struct trace_seq
*s
)
4983 /* Probably should print a warning here. */
4987 /* should be zero ended, but we are paranoid. */
4988 s
->buffer
[s
->len
] = 0;
4990 printk(KERN_TRACE
"%s", s
->buffer
);
4995 void trace_init_global_iter(struct trace_iterator
*iter
)
4997 iter
->tr
= &global_trace
;
4998 iter
->trace
= current_trace
;
4999 iter
->cpu_file
= TRACE_PIPE_ALL_CPU
;
5003 __ftrace_dump(bool disable_tracing
, enum ftrace_dump_mode oops_dump_mode
)
5005 static arch_spinlock_t ftrace_dump_lock
=
5006 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
5007 /* use static because iter can be a bit big for the stack */
5008 static struct trace_iterator iter
;
5009 unsigned int old_userobj
;
5010 static int dump_ran
;
5011 unsigned long flags
;
5015 local_irq_save(flags
);
5016 arch_spin_lock(&ftrace_dump_lock
);
5024 /* Did function tracer already get disabled? */
5025 if (ftrace_is_dead()) {
5026 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
5027 printk("# MAY BE MISSING FUNCTION EVENTS\n");
5030 if (disable_tracing
)
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 /* Simulate the iterator */
5045 iter
.tr
= &global_trace
;
5046 iter
.trace
= current_trace
;
5048 switch (oops_dump_mode
) {
5050 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
5053 iter
.cpu_file
= raw_smp_processor_id();
5058 printk(KERN_TRACE
"Bad dumping mode, switching to all CPUs dump\n");
5059 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
5062 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
5065 * We need to stop all tracing on all CPUS to read the
5066 * the next buffer. This is a bit expensive, but is
5067 * not done often. We fill all what we can read,
5068 * and then release the locks again.
5071 while (!trace_empty(&iter
)) {
5074 printk(KERN_TRACE
"---------------------------------\n");
5078 /* reset all but tr, trace, and overruns */
5079 memset(&iter
.seq
, 0,
5080 sizeof(struct trace_iterator
) -
5081 offsetof(struct trace_iterator
, seq
));
5082 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
5085 if (trace_find_next_entry_inc(&iter
) != NULL
) {
5088 ret
= print_trace_line(&iter
);
5089 if (ret
!= TRACE_TYPE_NO_CONSUME
)
5090 trace_consume(&iter
);
5092 touch_nmi_watchdog();
5094 trace_printk_seq(&iter
.seq
);
5098 printk(KERN_TRACE
" (ftrace buffer empty)\n");
5100 printk(KERN_TRACE
"---------------------------------\n");
5103 /* Re-enable tracing if requested */
5104 if (!disable_tracing
) {
5105 trace_flags
|= old_userobj
;
5107 for_each_tracing_cpu(cpu
) {
5108 atomic_dec(&iter
.tr
->data
[cpu
]->disabled
);
5114 arch_spin_unlock(&ftrace_dump_lock
);
5115 local_irq_restore(flags
);
5118 /* By default: disable tracing after the dump */
5119 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode
)
5121 __ftrace_dump(true, oops_dump_mode
);
5123 EXPORT_SYMBOL_GPL(ftrace_dump
);
5125 __init
static int tracer_alloc_buffers(void)
5128 enum ring_buffer_flags rb_flags
;
5133 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
5136 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
5137 goto out_free_buffer_mask
;
5139 /* Only allocate trace_printk buffers if a trace_printk exists */
5140 if (__stop___trace_bprintk_fmt
!= __start___trace_bprintk_fmt
)
5141 /* Must be called before global_trace.buffer is allocated */
5142 trace_printk_init_buffers();
5144 /* To save memory, keep the ring buffer size to its minimum */
5145 if (ring_buffer_expanded
)
5146 ring_buf_size
= trace_buf_size
;
5150 rb_flags
= trace_flags
& TRACE_ITER_OVERWRITE
? RB_FL_OVERWRITE
: 0;
5152 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
5153 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
5155 /* TODO: make the number of buffers hot pluggable with CPUS */
5156 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
, rb_flags
);
5157 if (!global_trace
.buffer
) {
5158 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
5160 goto out_free_cpumask
;
5162 if (global_trace
.buffer_disabled
)
5166 #ifdef CONFIG_TRACER_MAX_TRACE
5167 max_tr
.buffer
= ring_buffer_alloc(1, rb_flags
);
5168 if (!max_tr
.buffer
) {
5169 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
5171 ring_buffer_free(global_trace
.buffer
);
5172 goto out_free_cpumask
;
5176 /* Allocate the first page for all buffers */
5177 for_each_tracing_cpu(i
) {
5178 global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
5179 max_tr
.data
[i
] = &per_cpu(max_tr_data
, i
);
5182 set_buffer_entries(&global_trace
,
5183 ring_buffer_size(global_trace
.buffer
, 0));
5184 #ifdef CONFIG_TRACER_MAX_TRACE
5185 set_buffer_entries(&max_tr
, 1);
5188 trace_init_cmdlines();
5189 init_irq_work(&trace_work_wakeup
, trace_wake_up
);
5191 register_tracer(&nop_trace
);
5192 current_trace
= &nop_trace
;
5193 /* All seems OK, enable tracing */
5194 tracing_disabled
= 0;
5196 atomic_notifier_chain_register(&panic_notifier_list
,
5197 &trace_panic_notifier
);
5199 register_die_notifier(&trace_die_notifier
);
5201 while (trace_boot_options
) {
5204 option
= strsep(&trace_boot_options
, ",");
5205 trace_set_options(option
);
5211 free_cpumask_var(tracing_cpumask
);
5212 out_free_buffer_mask
:
5213 free_cpumask_var(tracing_buffer_mask
);
5218 __init
static int clear_boot_tracer(void)
5221 * The default tracer at boot buffer is an init section.
5222 * This function is called in lateinit. If we did not
5223 * find the boot tracer, then clear it out, to prevent
5224 * later registration from accessing the buffer that is
5225 * about to be freed.
5227 if (!default_bootup_tracer
)
5230 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
5231 default_bootup_tracer
);
5232 default_bootup_tracer
= NULL
;
5237 early_initcall(tracer_alloc_buffers
);
5238 fs_initcall(tracer_init_debugfs
);
5239 late_initcall(clear_boot_tracer
);