2 * ring buffer based function tracer
4 * Copyright (C) 2007-2012 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/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/nmi.h>
41 #include <linux/sched/rt.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 bool 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 * Kill all tracing for good (never come back).
90 * It is initialized to 1 but will turn to zero if the initialization
91 * of the tracer is successful. But that is the only place that sets
94 static int tracing_disabled
= 1;
96 DEFINE_PER_CPU(int, ftrace_cpu_disabled
);
98 cpumask_var_t __read_mostly tracing_buffer_mask
;
101 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
103 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
104 * is set, then ftrace_dump is called. This will output the contents
105 * of the ftrace buffers to the console. This is very useful for
106 * capturing traces that lead to crashes and outputing it to a
109 * It is default off, but you can enable it with either specifying
110 * "ftrace_dump_on_oops" in the kernel command line, or setting
111 * /proc/sys/kernel/ftrace_dump_on_oops
112 * Set 1 if you want to dump buffers of all CPUs
113 * Set 2 if you want to dump the buffer of the CPU that triggered oops
116 enum ftrace_dump_mode ftrace_dump_on_oops
;
118 static int tracing_set_tracer(const char *buf
);
120 #define MAX_TRACER_SIZE 100
121 static char bootup_tracer_buf
[MAX_TRACER_SIZE
] __initdata
;
122 static char *default_bootup_tracer
;
124 static bool allocate_snapshot
;
126 static int __init
set_cmdline_ftrace(char *str
)
128 strlcpy(bootup_tracer_buf
, str
, MAX_TRACER_SIZE
);
129 default_bootup_tracer
= bootup_tracer_buf
;
130 /* We are using ftrace early, expand it */
131 ring_buffer_expanded
= true;
134 __setup("ftrace=", set_cmdline_ftrace
);
136 static int __init
set_ftrace_dump_on_oops(char *str
)
138 if (*str
++ != '=' || !*str
) {
139 ftrace_dump_on_oops
= DUMP_ALL
;
143 if (!strcmp("orig_cpu", str
)) {
144 ftrace_dump_on_oops
= DUMP_ORIG
;
150 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops
);
152 static int __init
boot_alloc_snapshot(char *str
)
154 allocate_snapshot
= true;
155 /* We also need the main ring buffer expanded */
156 ring_buffer_expanded
= true;
159 __setup("alloc_snapshot", boot_alloc_snapshot
);
162 static char trace_boot_options_buf
[MAX_TRACER_SIZE
] __initdata
;
163 static char *trace_boot_options __initdata
;
165 static int __init
set_trace_boot_options(char *str
)
167 strlcpy(trace_boot_options_buf
, str
, MAX_TRACER_SIZE
);
168 trace_boot_options
= trace_boot_options_buf
;
171 __setup("trace_options=", set_trace_boot_options
);
173 unsigned long long ns2usecs(cycle_t nsec
)
181 * The global_trace is the descriptor that holds the tracing
182 * buffers for the live tracing. For each CPU, it contains
183 * a link list of pages that will store trace entries. The
184 * page descriptor of the pages in the memory is used to hold
185 * the link list by linking the lru item in the page descriptor
186 * to each of the pages in the buffer per CPU.
188 * For each active CPU there is a data field that holds the
189 * pages for the buffer for that CPU. Each CPU has the same number
190 * of pages allocated for its buffer.
192 static struct trace_array global_trace
;
194 LIST_HEAD(ftrace_trace_arrays
);
196 int trace_array_get(struct trace_array
*this_tr
)
198 struct trace_array
*tr
;
201 mutex_lock(&trace_types_lock
);
202 list_for_each_entry(tr
, &ftrace_trace_arrays
, list
) {
209 mutex_unlock(&trace_types_lock
);
214 static void __trace_array_put(struct trace_array
*this_tr
)
216 WARN_ON(!this_tr
->ref
);
220 void trace_array_put(struct trace_array
*this_tr
)
222 mutex_lock(&trace_types_lock
);
223 __trace_array_put(this_tr
);
224 mutex_unlock(&trace_types_lock
);
227 int filter_current_check_discard(struct ring_buffer
*buffer
,
228 struct ftrace_event_call
*call
, void *rec
,
229 struct ring_buffer_event
*event
)
231 return filter_check_discard(call
, rec
, buffer
, event
);
233 EXPORT_SYMBOL_GPL(filter_current_check_discard
);
235 cycle_t
buffer_ftrace_now(struct trace_buffer
*buf
, int cpu
)
239 /* Early boot up does not have a buffer yet */
241 return trace_clock_local();
243 ts
= ring_buffer_time_stamp(buf
->buffer
, cpu
);
244 ring_buffer_normalize_time_stamp(buf
->buffer
, cpu
, &ts
);
249 cycle_t
ftrace_now(int cpu
)
251 return buffer_ftrace_now(&global_trace
.trace_buffer
, cpu
);
255 * tracing_is_enabled - Show if global_trace has been disabled
257 * Shows if the global trace has been enabled or not. It uses the
258 * mirror flag "buffer_disabled" to be used in fast paths such as for
259 * the irqsoff tracer. But it may be inaccurate due to races. If you
260 * need to know the accurate state, use tracing_is_on() which is a little
261 * slower, but accurate.
263 int tracing_is_enabled(void)
266 * For quick access (irqsoff uses this in fast path), just
267 * return the mirror variable of the state of the ring buffer.
268 * It's a little racy, but we don't really care.
271 return !global_trace
.buffer_disabled
;
275 * trace_buf_size is the size in bytes that is allocated
276 * for a buffer. Note, the number of bytes is always rounded
279 * This number is purposely set to a low number of 16384.
280 * If the dump on oops happens, it will be much appreciated
281 * to not have to wait for all that output. Anyway this can be
282 * boot time and run time configurable.
284 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
286 static unsigned long trace_buf_size
= TRACE_BUF_SIZE_DEFAULT
;
288 /* trace_types holds a link list of available tracers. */
289 static struct tracer
*trace_types __read_mostly
;
292 * trace_types_lock is used to protect the trace_types list.
294 DEFINE_MUTEX(trace_types_lock
);
297 * serialize the access of the ring buffer
299 * ring buffer serializes readers, but it is low level protection.
300 * The validity of the events (which returns by ring_buffer_peek() ..etc)
301 * are not protected by ring buffer.
303 * The content of events may become garbage if we allow other process consumes
304 * these events concurrently:
305 * A) the page of the consumed events may become a normal page
306 * (not reader page) in ring buffer, and this page will be rewrited
307 * by events producer.
308 * B) The page of the consumed events may become a page for splice_read,
309 * and this page will be returned to system.
311 * These primitives allow multi process access to different cpu ring buffer
314 * These primitives don't distinguish read-only and read-consume access.
315 * Multi read-only access are also serialized.
319 static DECLARE_RWSEM(all_cpu_access_lock
);
320 static DEFINE_PER_CPU(struct mutex
, cpu_access_lock
);
322 static inline void trace_access_lock(int cpu
)
324 if (cpu
== RING_BUFFER_ALL_CPUS
) {
325 /* gain it for accessing the whole ring buffer. */
326 down_write(&all_cpu_access_lock
);
328 /* gain it for accessing a cpu ring buffer. */
330 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
331 down_read(&all_cpu_access_lock
);
333 /* Secondly block other access to this @cpu ring buffer. */
334 mutex_lock(&per_cpu(cpu_access_lock
, cpu
));
338 static inline void trace_access_unlock(int cpu
)
340 if (cpu
== RING_BUFFER_ALL_CPUS
) {
341 up_write(&all_cpu_access_lock
);
343 mutex_unlock(&per_cpu(cpu_access_lock
, cpu
));
344 up_read(&all_cpu_access_lock
);
348 static inline void trace_access_lock_init(void)
352 for_each_possible_cpu(cpu
)
353 mutex_init(&per_cpu(cpu_access_lock
, cpu
));
358 static DEFINE_MUTEX(access_lock
);
360 static inline void trace_access_lock(int cpu
)
363 mutex_lock(&access_lock
);
366 static inline void trace_access_unlock(int cpu
)
369 mutex_unlock(&access_lock
);
372 static inline void trace_access_lock_init(void)
378 /* trace_flags holds trace_options default values */
379 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
| TRACE_ITER_PRINTK
|
380 TRACE_ITER_ANNOTATE
| TRACE_ITER_CONTEXT_INFO
| TRACE_ITER_SLEEP_TIME
|
381 TRACE_ITER_GRAPH_TIME
| TRACE_ITER_RECORD_CMD
| TRACE_ITER_OVERWRITE
|
382 TRACE_ITER_IRQ_INFO
| TRACE_ITER_MARKERS
| TRACE_ITER_FUNCTION
;
384 void tracer_tracing_on(struct trace_array
*tr
)
386 if (tr
->trace_buffer
.buffer
)
387 ring_buffer_record_on(tr
->trace_buffer
.buffer
);
389 * This flag is looked at when buffers haven't been allocated
390 * yet, or by some tracers (like irqsoff), that just want to
391 * know if the ring buffer has been disabled, but it can handle
392 * races of where it gets disabled but we still do a record.
393 * As the check is in the fast path of the tracers, it is more
394 * important to be fast than accurate.
396 tr
->buffer_disabled
= 0;
397 /* Make the flag seen by readers */
402 * tracing_on - enable tracing buffers
404 * This function enables tracing buffers that may have been
405 * disabled with tracing_off.
407 void tracing_on(void)
409 tracer_tracing_on(&global_trace
);
411 EXPORT_SYMBOL_GPL(tracing_on
);
414 * __trace_puts - write a constant string into the trace buffer.
415 * @ip: The address of the caller
416 * @str: The constant string to write
417 * @size: The size of the string.
419 int __trace_puts(unsigned long ip
, const char *str
, int size
)
421 struct ring_buffer_event
*event
;
422 struct ring_buffer
*buffer
;
423 struct print_entry
*entry
;
424 unsigned long irq_flags
;
428 pc
= preempt_count();
430 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
433 alloc
= sizeof(*entry
) + size
+ 2; /* possible \n added */
435 local_save_flags(irq_flags
);
436 buffer
= global_trace
.trace_buffer
.buffer
;
437 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, alloc
,
442 entry
= ring_buffer_event_data(event
);
445 memcpy(&entry
->buf
, str
, size
);
447 /* Add a newline if necessary */
448 if (entry
->buf
[size
- 1] != '\n') {
449 entry
->buf
[size
] = '\n';
450 entry
->buf
[size
+ 1] = '\0';
452 entry
->buf
[size
] = '\0';
454 __buffer_unlock_commit(buffer
, event
);
455 ftrace_trace_stack(buffer
, irq_flags
, 4, pc
);
459 EXPORT_SYMBOL_GPL(__trace_puts
);
462 * __trace_bputs - write the pointer to a constant string into trace buffer
463 * @ip: The address of the caller
464 * @str: The constant string to write to the buffer to
466 int __trace_bputs(unsigned long ip
, const char *str
)
468 struct ring_buffer_event
*event
;
469 struct ring_buffer
*buffer
;
470 struct bputs_entry
*entry
;
471 unsigned long irq_flags
;
472 int size
= sizeof(struct bputs_entry
);
475 pc
= preempt_count();
477 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
480 local_save_flags(irq_flags
);
481 buffer
= global_trace
.trace_buffer
.buffer
;
482 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPUTS
, size
,
487 entry
= ring_buffer_event_data(event
);
491 __buffer_unlock_commit(buffer
, event
);
492 ftrace_trace_stack(buffer
, irq_flags
, 4, pc
);
496 EXPORT_SYMBOL_GPL(__trace_bputs
);
498 #ifdef CONFIG_TRACER_SNAPSHOT
500 * trace_snapshot - take a snapshot of the current buffer.
502 * This causes a swap between the snapshot buffer and the current live
503 * tracing buffer. You can use this to take snapshots of the live
504 * trace when some condition is triggered, but continue to trace.
506 * Note, make sure to allocate the snapshot with either
507 * a tracing_snapshot_alloc(), or by doing it manually
508 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
510 * If the snapshot buffer is not allocated, it will stop tracing.
511 * Basically making a permanent snapshot.
513 void tracing_snapshot(void)
515 struct trace_array
*tr
= &global_trace
;
516 struct tracer
*tracer
= tr
->current_trace
;
520 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
521 internal_trace_puts("*** snapshot is being ignored ***\n");
525 if (!tr
->allocated_snapshot
) {
526 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
527 internal_trace_puts("*** stopping trace here! ***\n");
532 /* Note, snapshot can not be used when the tracer uses it */
533 if (tracer
->use_max_tr
) {
534 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
535 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
539 local_irq_save(flags
);
540 update_max_tr(tr
, current
, smp_processor_id());
541 local_irq_restore(flags
);
543 EXPORT_SYMBOL_GPL(tracing_snapshot
);
545 static int resize_buffer_duplicate_size(struct trace_buffer
*trace_buf
,
546 struct trace_buffer
*size_buf
, int cpu_id
);
547 static void set_buffer_entries(struct trace_buffer
*buf
, unsigned long val
);
549 static int alloc_snapshot(struct trace_array
*tr
)
553 if (!tr
->allocated_snapshot
) {
555 /* allocate spare buffer */
556 ret
= resize_buffer_duplicate_size(&tr
->max_buffer
,
557 &tr
->trace_buffer
, RING_BUFFER_ALL_CPUS
);
561 tr
->allocated_snapshot
= true;
567 void free_snapshot(struct trace_array
*tr
)
570 * We don't free the ring buffer. instead, resize it because
571 * The max_tr ring buffer has some state (e.g. ring->clock) and
572 * we want preserve it.
574 ring_buffer_resize(tr
->max_buffer
.buffer
, 1, RING_BUFFER_ALL_CPUS
);
575 set_buffer_entries(&tr
->max_buffer
, 1);
576 tracing_reset_online_cpus(&tr
->max_buffer
);
577 tr
->allocated_snapshot
= false;
581 * trace_snapshot_alloc - allocate and take a snapshot of the current buffer.
583 * This is similar to trace_snapshot(), but it will allocate the
584 * snapshot buffer if it isn't already allocated. Use this only
585 * where it is safe to sleep, as the allocation may sleep.
587 * This causes a swap between the snapshot buffer and the current live
588 * tracing buffer. You can use this to take snapshots of the live
589 * trace when some condition is triggered, but continue to trace.
591 void tracing_snapshot_alloc(void)
593 struct trace_array
*tr
= &global_trace
;
596 ret
= alloc_snapshot(tr
);
597 if (WARN_ON(ret
< 0))
602 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc
);
604 void tracing_snapshot(void)
606 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
608 EXPORT_SYMBOL_GPL(tracing_snapshot
);
609 void tracing_snapshot_alloc(void)
614 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc
);
615 #endif /* CONFIG_TRACER_SNAPSHOT */
617 void tracer_tracing_off(struct trace_array
*tr
)
619 if (tr
->trace_buffer
.buffer
)
620 ring_buffer_record_off(tr
->trace_buffer
.buffer
);
622 * This flag is looked at when buffers haven't been allocated
623 * yet, or by some tracers (like irqsoff), that just want to
624 * know if the ring buffer has been disabled, but it can handle
625 * races of where it gets disabled but we still do a record.
626 * As the check is in the fast path of the tracers, it is more
627 * important to be fast than accurate.
629 tr
->buffer_disabled
= 1;
630 /* Make the flag seen by readers */
635 * tracing_off - turn off tracing buffers
637 * This function stops the tracing buffers from recording data.
638 * It does not disable any overhead the tracers themselves may
639 * be causing. This function simply causes all recording to
640 * the ring buffers to fail.
642 void tracing_off(void)
644 tracer_tracing_off(&global_trace
);
646 EXPORT_SYMBOL_GPL(tracing_off
);
649 * tracer_tracing_is_on - show real state of ring buffer enabled
650 * @tr : the trace array to know if ring buffer is enabled
652 * Shows real state of the ring buffer if it is enabled or not.
654 int tracer_tracing_is_on(struct trace_array
*tr
)
656 if (tr
->trace_buffer
.buffer
)
657 return ring_buffer_record_is_on(tr
->trace_buffer
.buffer
);
658 return !tr
->buffer_disabled
;
662 * tracing_is_on - show state of ring buffers enabled
664 int tracing_is_on(void)
666 return tracer_tracing_is_on(&global_trace
);
668 EXPORT_SYMBOL_GPL(tracing_is_on
);
670 static int __init
set_buf_size(char *str
)
672 unsigned long buf_size
;
676 buf_size
= memparse(str
, &str
);
677 /* nr_entries can not be zero */
680 trace_buf_size
= buf_size
;
683 __setup("trace_buf_size=", set_buf_size
);
685 static int __init
set_tracing_thresh(char *str
)
687 unsigned long threshold
;
692 ret
= kstrtoul(str
, 0, &threshold
);
695 tracing_thresh
= threshold
* 1000;
698 __setup("tracing_thresh=", set_tracing_thresh
);
700 unsigned long nsecs_to_usecs(unsigned long nsecs
)
705 /* These must match the bit postions in trace_iterator_flags */
706 static const char *trace_options
[] = {
740 int in_ns
; /* is this clock in nanoseconds? */
742 { trace_clock_local
, "local", 1 },
743 { trace_clock_global
, "global", 1 },
744 { trace_clock_counter
, "counter", 0 },
745 { trace_clock_jiffies
, "uptime", 0 },
746 { trace_clock
, "perf", 1 },
751 * trace_parser_get_init - gets the buffer for trace parser
753 int trace_parser_get_init(struct trace_parser
*parser
, int size
)
755 memset(parser
, 0, sizeof(*parser
));
757 parser
->buffer
= kmalloc(size
, GFP_KERNEL
);
766 * trace_parser_put - frees the buffer for trace parser
768 void trace_parser_put(struct trace_parser
*parser
)
770 kfree(parser
->buffer
);
774 * trace_get_user - reads the user input string separated by space
775 * (matched by isspace(ch))
777 * For each string found the 'struct trace_parser' is updated,
778 * and the function returns.
780 * Returns number of bytes read.
782 * See kernel/trace/trace.h for 'struct trace_parser' details.
784 int trace_get_user(struct trace_parser
*parser
, const char __user
*ubuf
,
785 size_t cnt
, loff_t
*ppos
)
792 trace_parser_clear(parser
);
794 ret
= get_user(ch
, ubuf
++);
802 * The parser is not finished with the last write,
803 * continue reading the user input without skipping spaces.
806 /* skip white space */
807 while (cnt
&& isspace(ch
)) {
808 ret
= get_user(ch
, ubuf
++);
815 /* only spaces were written */
825 /* read the non-space input */
826 while (cnt
&& !isspace(ch
)) {
827 if (parser
->idx
< parser
->size
- 1)
828 parser
->buffer
[parser
->idx
++] = ch
;
833 ret
= get_user(ch
, ubuf
++);
840 /* We either got finished input or we have to wait for another call. */
842 parser
->buffer
[parser
->idx
] = 0;
843 parser
->cont
= false;
844 } else if (parser
->idx
< parser
->size
- 1) {
846 parser
->buffer
[parser
->idx
++] = ch
;
859 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
867 if (s
->len
<= s
->readpos
)
870 len
= s
->len
- s
->readpos
;
873 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
883 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
887 if (s
->len
<= s
->readpos
)
890 len
= s
->len
- s
->readpos
;
893 memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
900 * ftrace_max_lock is used to protect the swapping of buffers
901 * when taking a max snapshot. The buffers themselves are
902 * protected by per_cpu spinlocks. But the action of the swap
903 * needs its own lock.
905 * This is defined as a arch_spinlock_t in order to help
906 * with performance when lockdep debugging is enabled.
908 * It is also used in other places outside the update_max_tr
909 * so it needs to be defined outside of the
910 * CONFIG_TRACER_MAX_TRACE.
912 static arch_spinlock_t ftrace_max_lock
=
913 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
915 unsigned long __read_mostly tracing_thresh
;
917 #ifdef CONFIG_TRACER_MAX_TRACE
918 unsigned long __read_mostly tracing_max_latency
;
921 * Copy the new maximum trace into the separate maximum-trace
922 * structure. (this way the maximum trace is permanently saved,
923 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
926 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
928 struct trace_buffer
*trace_buf
= &tr
->trace_buffer
;
929 struct trace_buffer
*max_buf
= &tr
->max_buffer
;
930 struct trace_array_cpu
*data
= per_cpu_ptr(trace_buf
->data
, cpu
);
931 struct trace_array_cpu
*max_data
= per_cpu_ptr(max_buf
->data
, cpu
);
934 max_buf
->time_start
= data
->preempt_timestamp
;
936 max_data
->saved_latency
= tracing_max_latency
;
937 max_data
->critical_start
= data
->critical_start
;
938 max_data
->critical_end
= data
->critical_end
;
940 memcpy(max_data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
941 max_data
->pid
= tsk
->pid
;
943 * If tsk == current, then use current_uid(), as that does not use
944 * RCU. The irq tracer can be called out of RCU scope.
947 max_data
->uid
= current_uid();
949 max_data
->uid
= task_uid(tsk
);
951 max_data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
952 max_data
->policy
= tsk
->policy
;
953 max_data
->rt_priority
= tsk
->rt_priority
;
955 /* record this tasks comm */
956 tracing_record_cmdline(tsk
);
960 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
962 * @tsk: the task with the latency
963 * @cpu: The cpu that initiated the trace.
965 * Flip the buffers between the @tr and the max_tr and record information
966 * about which task was the cause of this latency.
969 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
971 struct ring_buffer
*buf
;
976 WARN_ON_ONCE(!irqs_disabled());
978 if (!tr
->allocated_snapshot
) {
979 /* Only the nop tracer should hit this when disabling */
980 WARN_ON_ONCE(tr
->current_trace
!= &nop_trace
);
984 arch_spin_lock(&ftrace_max_lock
);
986 buf
= tr
->trace_buffer
.buffer
;
987 tr
->trace_buffer
.buffer
= tr
->max_buffer
.buffer
;
988 tr
->max_buffer
.buffer
= buf
;
990 __update_max_tr(tr
, tsk
, cpu
);
991 arch_spin_unlock(&ftrace_max_lock
);
995 * update_max_tr_single - only copy one trace over, and reset the rest
997 * @tsk - task with the latency
998 * @cpu - the cpu of the buffer to copy.
1000 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1003 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
1010 WARN_ON_ONCE(!irqs_disabled());
1011 if (!tr
->allocated_snapshot
) {
1012 /* Only the nop tracer should hit this when disabling */
1013 WARN_ON_ONCE(tr
->current_trace
!= &nop_trace
);
1017 arch_spin_lock(&ftrace_max_lock
);
1019 ret
= ring_buffer_swap_cpu(tr
->max_buffer
.buffer
, tr
->trace_buffer
.buffer
, cpu
);
1021 if (ret
== -EBUSY
) {
1023 * We failed to swap the buffer due to a commit taking
1024 * place on this CPU. We fail to record, but we reset
1025 * the max trace buffer (no one writes directly to it)
1026 * and flag that it failed.
1028 trace_array_printk_buf(tr
->max_buffer
.buffer
, _THIS_IP_
,
1029 "Failed to swap buffers due to commit in progress\n");
1032 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
&& ret
!= -EBUSY
);
1034 __update_max_tr(tr
, tsk
, cpu
);
1035 arch_spin_unlock(&ftrace_max_lock
);
1037 #endif /* CONFIG_TRACER_MAX_TRACE */
1039 static int default_wait_pipe(struct trace_iterator
*iter
)
1041 /* Iterators are static, they should be filled or empty */
1042 if (trace_buffer_iter(iter
, iter
->cpu_file
))
1045 return ring_buffer_wait(iter
->trace_buffer
->buffer
, iter
->cpu_file
);
1048 #ifdef CONFIG_FTRACE_STARTUP_TEST
1049 static int run_tracer_selftest(struct tracer
*type
)
1051 struct trace_array
*tr
= &global_trace
;
1052 struct tracer
*saved_tracer
= tr
->current_trace
;
1055 if (!type
->selftest
|| tracing_selftest_disabled
)
1059 * Run a selftest on this tracer.
1060 * Here we reset the trace buffer, and set the current
1061 * tracer to be this tracer. The tracer can then run some
1062 * internal tracing to verify that everything is in order.
1063 * If we fail, we do not register this tracer.
1065 tracing_reset_online_cpus(&tr
->trace_buffer
);
1067 tr
->current_trace
= type
;
1069 #ifdef CONFIG_TRACER_MAX_TRACE
1070 if (type
->use_max_tr
) {
1071 /* If we expanded the buffers, make sure the max is expanded too */
1072 if (ring_buffer_expanded
)
1073 ring_buffer_resize(tr
->max_buffer
.buffer
, trace_buf_size
,
1074 RING_BUFFER_ALL_CPUS
);
1075 tr
->allocated_snapshot
= true;
1079 /* the test is responsible for initializing and enabling */
1080 pr_info("Testing tracer %s: ", type
->name
);
1081 ret
= type
->selftest(type
, tr
);
1082 /* the test is responsible for resetting too */
1083 tr
->current_trace
= saved_tracer
;
1085 printk(KERN_CONT
"FAILED!\n");
1086 /* Add the warning after printing 'FAILED' */
1090 /* Only reset on passing, to avoid touching corrupted buffers */
1091 tracing_reset_online_cpus(&tr
->trace_buffer
);
1093 #ifdef CONFIG_TRACER_MAX_TRACE
1094 if (type
->use_max_tr
) {
1095 tr
->allocated_snapshot
= false;
1097 /* Shrink the max buffer again */
1098 if (ring_buffer_expanded
)
1099 ring_buffer_resize(tr
->max_buffer
.buffer
, 1,
1100 RING_BUFFER_ALL_CPUS
);
1104 printk(KERN_CONT
"PASSED\n");
1108 static inline int run_tracer_selftest(struct tracer
*type
)
1112 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1115 * register_tracer - register a tracer with the ftrace system.
1116 * @type - the plugin for the tracer
1118 * Register a new plugin tracer.
1120 int register_tracer(struct tracer
*type
)
1126 pr_info("Tracer must have a name\n");
1130 if (strlen(type
->name
) >= MAX_TRACER_SIZE
) {
1131 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE
);
1135 mutex_lock(&trace_types_lock
);
1137 tracing_selftest_running
= true;
1139 for (t
= trace_types
; t
; t
= t
->next
) {
1140 if (strcmp(type
->name
, t
->name
) == 0) {
1142 pr_info("Tracer %s already registered\n",
1149 if (!type
->set_flag
)
1150 type
->set_flag
= &dummy_set_flag
;
1152 type
->flags
= &dummy_tracer_flags
;
1154 if (!type
->flags
->opts
)
1155 type
->flags
->opts
= dummy_tracer_opt
;
1156 if (!type
->wait_pipe
)
1157 type
->wait_pipe
= default_wait_pipe
;
1159 ret
= run_tracer_selftest(type
);
1163 type
->next
= trace_types
;
1167 tracing_selftest_running
= false;
1168 mutex_unlock(&trace_types_lock
);
1170 if (ret
|| !default_bootup_tracer
)
1173 if (strncmp(default_bootup_tracer
, type
->name
, MAX_TRACER_SIZE
))
1176 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
1177 /* Do we want this tracer to start on bootup? */
1178 tracing_set_tracer(type
->name
);
1179 default_bootup_tracer
= NULL
;
1180 /* disable other selftests, since this will break it. */
1181 tracing_selftest_disabled
= true;
1182 #ifdef CONFIG_FTRACE_STARTUP_TEST
1183 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
1191 void tracing_reset(struct trace_buffer
*buf
, int cpu
)
1193 struct ring_buffer
*buffer
= buf
->buffer
;
1198 ring_buffer_record_disable(buffer
);
1200 /* Make sure all commits have finished */
1201 synchronize_sched();
1202 ring_buffer_reset_cpu(buffer
, cpu
);
1204 ring_buffer_record_enable(buffer
);
1207 void tracing_reset_online_cpus(struct trace_buffer
*buf
)
1209 struct ring_buffer
*buffer
= buf
->buffer
;
1215 ring_buffer_record_disable(buffer
);
1217 /* Make sure all commits have finished */
1218 synchronize_sched();
1220 buf
->time_start
= buffer_ftrace_now(buf
, buf
->cpu
);
1222 for_each_online_cpu(cpu
)
1223 ring_buffer_reset_cpu(buffer
, cpu
);
1225 ring_buffer_record_enable(buffer
);
1228 /* Must have trace_types_lock held */
1229 void tracing_reset_all_online_cpus(void)
1231 struct trace_array
*tr
;
1233 list_for_each_entry(tr
, &ftrace_trace_arrays
, list
) {
1234 tracing_reset_online_cpus(&tr
->trace_buffer
);
1235 #ifdef CONFIG_TRACER_MAX_TRACE
1236 tracing_reset_online_cpus(&tr
->max_buffer
);
1241 #define SAVED_CMDLINES 128
1242 #define NO_CMDLINE_MAP UINT_MAX
1243 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
1244 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
1245 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
1246 static unsigned saved_tgids
[SAVED_CMDLINES
];
1247 static int cmdline_idx
;
1248 static arch_spinlock_t trace_cmdline_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
1250 /* temporary disable recording */
1251 static atomic_t trace_record_cmdline_disabled __read_mostly
;
1253 static void trace_init_cmdlines(void)
1255 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
1256 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
1260 int is_tracing_stopped(void)
1262 return global_trace
.stop_count
;
1266 * ftrace_off_permanent - disable all ftrace code permanently
1268 * This should only be called when a serious anomally has
1269 * been detected. This will turn off the function tracing,
1270 * ring buffers, and other tracing utilites. It takes no
1271 * locks and can be called from any context.
1273 void ftrace_off_permanent(void)
1275 tracing_disabled
= 1;
1277 tracing_off_permanent();
1281 * tracing_start - quick start of the tracer
1283 * If tracing is enabled but was stopped by tracing_stop,
1284 * this will start the tracer back up.
1286 void tracing_start(void)
1288 struct ring_buffer
*buffer
;
1289 unsigned long flags
;
1291 if (tracing_disabled
)
1294 raw_spin_lock_irqsave(&global_trace
.start_lock
, flags
);
1295 if (--global_trace
.stop_count
) {
1296 if (global_trace
.stop_count
< 0) {
1297 /* Someone screwed up their debugging */
1299 global_trace
.stop_count
= 0;
1304 /* Prevent the buffers from switching */
1305 arch_spin_lock(&ftrace_max_lock
);
1307 buffer
= global_trace
.trace_buffer
.buffer
;
1309 ring_buffer_record_enable(buffer
);
1311 #ifdef CONFIG_TRACER_MAX_TRACE
1312 buffer
= global_trace
.max_buffer
.buffer
;
1314 ring_buffer_record_enable(buffer
);
1317 arch_spin_unlock(&ftrace_max_lock
);
1320 raw_spin_unlock_irqrestore(&global_trace
.start_lock
, flags
);
1323 static void tracing_start_tr(struct trace_array
*tr
)
1325 struct ring_buffer
*buffer
;
1326 unsigned long flags
;
1328 if (tracing_disabled
)
1331 /* If global, we need to also start the max tracer */
1332 if (tr
->flags
& TRACE_ARRAY_FL_GLOBAL
)
1333 return tracing_start();
1335 raw_spin_lock_irqsave(&tr
->start_lock
, flags
);
1337 if (--tr
->stop_count
) {
1338 if (tr
->stop_count
< 0) {
1339 /* Someone screwed up their debugging */
1346 buffer
= tr
->trace_buffer
.buffer
;
1348 ring_buffer_record_enable(buffer
);
1351 raw_spin_unlock_irqrestore(&tr
->start_lock
, flags
);
1355 * tracing_stop - quick stop of the tracer
1357 * Light weight way to stop tracing. Use in conjunction with
1360 void tracing_stop(void)
1362 struct ring_buffer
*buffer
;
1363 unsigned long flags
;
1365 raw_spin_lock_irqsave(&global_trace
.start_lock
, flags
);
1366 if (global_trace
.stop_count
++)
1369 /* Prevent the buffers from switching */
1370 arch_spin_lock(&ftrace_max_lock
);
1372 buffer
= global_trace
.trace_buffer
.buffer
;
1374 ring_buffer_record_disable(buffer
);
1376 #ifdef CONFIG_TRACER_MAX_TRACE
1377 buffer
= global_trace
.max_buffer
.buffer
;
1379 ring_buffer_record_disable(buffer
);
1382 arch_spin_unlock(&ftrace_max_lock
);
1385 raw_spin_unlock_irqrestore(&global_trace
.start_lock
, flags
);
1388 static void tracing_stop_tr(struct trace_array
*tr
)
1390 struct ring_buffer
*buffer
;
1391 unsigned long flags
;
1393 /* If global, we need to also stop the max tracer */
1394 if (tr
->flags
& TRACE_ARRAY_FL_GLOBAL
)
1395 return tracing_stop();
1397 raw_spin_lock_irqsave(&tr
->start_lock
, flags
);
1398 if (tr
->stop_count
++)
1401 buffer
= tr
->trace_buffer
.buffer
;
1403 ring_buffer_record_disable(buffer
);
1406 raw_spin_unlock_irqrestore(&tr
->start_lock
, flags
);
1409 void trace_stop_cmdline_recording(void);
1411 static int trace_save_cmdline(struct task_struct
*tsk
)
1415 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
1419 * It's not the end of the world if we don't get
1420 * the lock, but we also don't want to spin
1421 * nor do we want to disable interrupts,
1422 * so if we miss here, then better luck next time.
1424 if (!arch_spin_trylock(&trace_cmdline_lock
))
1427 idx
= map_pid_to_cmdline
[tsk
->pid
];
1428 if (idx
== NO_CMDLINE_MAP
) {
1429 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
1432 * Check whether the cmdline buffer at idx has a pid
1433 * mapped. We are going to overwrite that entry so we
1434 * need to clear the map_pid_to_cmdline. Otherwise we
1435 * would read the new comm for the old pid.
1437 pid
= map_cmdline_to_pid
[idx
];
1438 if (pid
!= NO_CMDLINE_MAP
)
1439 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
1441 map_cmdline_to_pid
[idx
] = tsk
->pid
;
1442 map_pid_to_cmdline
[tsk
->pid
] = idx
;
1447 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
1448 saved_tgids
[idx
] = tsk
->tgid
;
1450 arch_spin_unlock(&trace_cmdline_lock
);
1455 void trace_find_cmdline(int pid
, char comm
[])
1460 strcpy(comm
, "<idle>");
1464 if (WARN_ON_ONCE(pid
< 0)) {
1465 strcpy(comm
, "<XXX>");
1469 if (pid
> PID_MAX_DEFAULT
) {
1470 strcpy(comm
, "<...>");
1475 arch_spin_lock(&trace_cmdline_lock
);
1476 map
= map_pid_to_cmdline
[pid
];
1477 if (map
!= NO_CMDLINE_MAP
)
1478 strcpy(comm
, saved_cmdlines
[map
]);
1480 strcpy(comm
, "<...>");
1482 arch_spin_unlock(&trace_cmdline_lock
);
1486 int trace_find_tgid(int pid
)
1492 arch_spin_lock(&trace_cmdline_lock
);
1493 map
= map_pid_to_cmdline
[pid
];
1494 if (map
!= NO_CMDLINE_MAP
)
1495 tgid
= saved_tgids
[map
];
1499 arch_spin_unlock(&trace_cmdline_lock
);
1505 void tracing_record_cmdline(struct task_struct
*tsk
)
1507 if (atomic_read(&trace_record_cmdline_disabled
) || !tracing_is_on())
1510 if (!__this_cpu_read(trace_cmdline_save
))
1513 if (trace_save_cmdline(tsk
))
1514 __this_cpu_write(trace_cmdline_save
, false);
1518 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
1521 struct task_struct
*tsk
= current
;
1523 entry
->preempt_count
= pc
& 0xff;
1524 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
1526 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1527 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
1529 TRACE_FLAG_IRQS_NOSUPPORT
|
1531 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
1532 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
1533 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
1535 EXPORT_SYMBOL_GPL(tracing_generic_entry_update
);
1537 struct ring_buffer_event
*
1538 trace_buffer_lock_reserve(struct ring_buffer
*buffer
,
1541 unsigned long flags
, int pc
)
1543 struct ring_buffer_event
*event
;
1545 event
= ring_buffer_lock_reserve(buffer
, len
);
1546 if (event
!= NULL
) {
1547 struct trace_entry
*ent
= ring_buffer_event_data(event
);
1549 tracing_generic_entry_update(ent
, flags
, pc
);
1557 __buffer_unlock_commit(struct ring_buffer
*buffer
, struct ring_buffer_event
*event
)
1559 __this_cpu_write(trace_cmdline_save
, true);
1560 ring_buffer_unlock_commit(buffer
, event
);
1564 __trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1565 struct ring_buffer_event
*event
,
1566 unsigned long flags
, int pc
)
1568 __buffer_unlock_commit(buffer
, event
);
1570 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1571 ftrace_trace_userstack(buffer
, flags
, pc
);
1574 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1575 struct ring_buffer_event
*event
,
1576 unsigned long flags
, int pc
)
1578 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
);
1580 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit
);
1582 struct ring_buffer_event
*
1583 trace_event_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1584 struct ftrace_event_file
*ftrace_file
,
1585 int type
, unsigned long len
,
1586 unsigned long flags
, int pc
)
1588 *current_rb
= ftrace_file
->tr
->trace_buffer
.buffer
;
1589 return trace_buffer_lock_reserve(*current_rb
,
1590 type
, len
, flags
, pc
);
1592 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve
);
1594 struct ring_buffer_event
*
1595 trace_current_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1596 int type
, unsigned long len
,
1597 unsigned long flags
, int pc
)
1599 *current_rb
= global_trace
.trace_buffer
.buffer
;
1600 return trace_buffer_lock_reserve(*current_rb
,
1601 type
, len
, flags
, pc
);
1603 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve
);
1605 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
1606 struct ring_buffer_event
*event
,
1607 unsigned long flags
, int pc
)
1609 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
);
1611 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit
);
1613 void trace_buffer_unlock_commit_regs(struct ring_buffer
*buffer
,
1614 struct ring_buffer_event
*event
,
1615 unsigned long flags
, int pc
,
1616 struct pt_regs
*regs
)
1618 __buffer_unlock_commit(buffer
, event
);
1620 ftrace_trace_stack_regs(buffer
, flags
, 0, pc
, regs
);
1621 ftrace_trace_userstack(buffer
, flags
, pc
);
1623 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs
);
1625 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
1626 struct ring_buffer_event
*event
)
1628 ring_buffer_discard_commit(buffer
, event
);
1630 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit
);
1633 trace_function(struct trace_array
*tr
,
1634 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1637 struct ftrace_event_call
*call
= &event_function
;
1638 struct ring_buffer
*buffer
= tr
->trace_buffer
.buffer
;
1639 struct ring_buffer_event
*event
;
1640 struct ftrace_entry
*entry
;
1642 /* If we are reading the ring buffer, don't trace */
1643 if (unlikely(__this_cpu_read(ftrace_cpu_disabled
)))
1646 event
= trace_buffer_lock_reserve(buffer
, TRACE_FN
, sizeof(*entry
),
1650 entry
= ring_buffer_event_data(event
);
1652 entry
->parent_ip
= parent_ip
;
1654 if (!filter_check_discard(call
, entry
, buffer
, event
))
1655 __buffer_unlock_commit(buffer
, event
);
1659 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1660 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1663 if (likely(!atomic_read(&data
->disabled
)))
1664 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
1667 #ifdef CONFIG_STACKTRACE
1669 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1670 struct ftrace_stack
{
1671 unsigned long calls
[FTRACE_STACK_MAX_ENTRIES
];
1674 static DEFINE_PER_CPU(struct ftrace_stack
, ftrace_stack
);
1675 static DEFINE_PER_CPU(int, ftrace_stack_reserve
);
1677 static void __ftrace_trace_stack(struct ring_buffer
*buffer
,
1678 unsigned long flags
,
1679 int skip
, int pc
, struct pt_regs
*regs
)
1681 struct ftrace_event_call
*call
= &event_kernel_stack
;
1682 struct ring_buffer_event
*event
;
1683 struct stack_entry
*entry
;
1684 struct stack_trace trace
;
1686 int size
= FTRACE_STACK_ENTRIES
;
1688 trace
.nr_entries
= 0;
1692 * Since events can happen in NMIs there's no safe way to
1693 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1694 * or NMI comes in, it will just have to use the default
1695 * FTRACE_STACK_SIZE.
1697 preempt_disable_notrace();
1699 use_stack
= __this_cpu_inc_return(ftrace_stack_reserve
);
1701 * We don't need any atomic variables, just a barrier.
1702 * If an interrupt comes in, we don't care, because it would
1703 * have exited and put the counter back to what we want.
1704 * We just need a barrier to keep gcc from moving things
1708 if (use_stack
== 1) {
1709 trace
.entries
= &__get_cpu_var(ftrace_stack
).calls
[0];
1710 trace
.max_entries
= FTRACE_STACK_MAX_ENTRIES
;
1713 save_stack_trace_regs(regs
, &trace
);
1715 save_stack_trace(&trace
);
1717 if (trace
.nr_entries
> size
)
1718 size
= trace
.nr_entries
;
1720 /* From now on, use_stack is a boolean */
1723 size
*= sizeof(unsigned long);
1725 event
= trace_buffer_lock_reserve(buffer
, TRACE_STACK
,
1726 sizeof(*entry
) + size
, flags
, pc
);
1729 entry
= ring_buffer_event_data(event
);
1731 memset(&entry
->caller
, 0, size
);
1734 memcpy(&entry
->caller
, trace
.entries
,
1735 trace
.nr_entries
* sizeof(unsigned long));
1737 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1738 trace
.entries
= entry
->caller
;
1740 save_stack_trace_regs(regs
, &trace
);
1742 save_stack_trace(&trace
);
1745 entry
->size
= trace
.nr_entries
;
1747 if (!filter_check_discard(call
, entry
, buffer
, event
))
1748 __buffer_unlock_commit(buffer
, event
);
1751 /* Again, don't let gcc optimize things here */
1753 __this_cpu_dec(ftrace_stack_reserve
);
1754 preempt_enable_notrace();
1758 void ftrace_trace_stack_regs(struct ring_buffer
*buffer
, unsigned long flags
,
1759 int skip
, int pc
, struct pt_regs
*regs
)
1761 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1764 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, regs
);
1767 void ftrace_trace_stack(struct ring_buffer
*buffer
, unsigned long flags
,
1770 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1773 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, NULL
);
1776 void __trace_stack(struct trace_array
*tr
, unsigned long flags
, int skip
,
1779 __ftrace_trace_stack(tr
->trace_buffer
.buffer
, flags
, skip
, pc
, NULL
);
1783 * trace_dump_stack - record a stack back trace in the trace buffer
1784 * @skip: Number of functions to skip (helper handlers)
1786 void trace_dump_stack(int skip
)
1788 unsigned long flags
;
1790 if (tracing_disabled
|| tracing_selftest_running
)
1793 local_save_flags(flags
);
1796 * Skip 3 more, seems to get us at the caller of
1800 __ftrace_trace_stack(global_trace
.trace_buffer
.buffer
,
1801 flags
, skip
, preempt_count(), NULL
);
1804 static DEFINE_PER_CPU(int, user_stack_count
);
1807 ftrace_trace_userstack(struct ring_buffer
*buffer
, unsigned long flags
, int pc
)
1809 struct ftrace_event_call
*call
= &event_user_stack
;
1810 struct ring_buffer_event
*event
;
1811 struct userstack_entry
*entry
;
1812 struct stack_trace trace
;
1814 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1818 * NMIs can not handle page faults, even with fix ups.
1819 * The save user stack can (and often does) fault.
1821 if (unlikely(in_nmi()))
1825 * prevent recursion, since the user stack tracing may
1826 * trigger other kernel events.
1829 if (__this_cpu_read(user_stack_count
))
1832 __this_cpu_inc(user_stack_count
);
1834 event
= trace_buffer_lock_reserve(buffer
, TRACE_USER_STACK
,
1835 sizeof(*entry
), flags
, pc
);
1837 goto out_drop_count
;
1838 entry
= ring_buffer_event_data(event
);
1840 entry
->tgid
= current
->tgid
;
1841 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1843 trace
.nr_entries
= 0;
1844 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1846 trace
.entries
= entry
->caller
;
1848 save_stack_trace_user(&trace
);
1849 if (!filter_check_discard(call
, entry
, buffer
, event
))
1850 __buffer_unlock_commit(buffer
, event
);
1853 __this_cpu_dec(user_stack_count
);
1859 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1861 ftrace_trace_userstack(tr
, flags
, preempt_count());
1865 #endif /* CONFIG_STACKTRACE */
1867 /* created for use with alloc_percpu */
1868 struct trace_buffer_struct
{
1869 char buffer
[TRACE_BUF_SIZE
];
1872 static struct trace_buffer_struct
*trace_percpu_buffer
;
1873 static struct trace_buffer_struct
*trace_percpu_sirq_buffer
;
1874 static struct trace_buffer_struct
*trace_percpu_irq_buffer
;
1875 static struct trace_buffer_struct
*trace_percpu_nmi_buffer
;
1878 * The buffer used is dependent on the context. There is a per cpu
1879 * buffer for normal context, softirq contex, hard irq context and
1880 * for NMI context. Thise allows for lockless recording.
1882 * Note, if the buffers failed to be allocated, then this returns NULL
1884 static char *get_trace_buf(void)
1886 struct trace_buffer_struct
*percpu_buffer
;
1889 * If we have allocated per cpu buffers, then we do not
1890 * need to do any locking.
1893 percpu_buffer
= trace_percpu_nmi_buffer
;
1895 percpu_buffer
= trace_percpu_irq_buffer
;
1896 else if (in_softirq())
1897 percpu_buffer
= trace_percpu_sirq_buffer
;
1899 percpu_buffer
= trace_percpu_buffer
;
1904 return this_cpu_ptr(&percpu_buffer
->buffer
[0]);
1907 static int alloc_percpu_trace_buffer(void)
1909 struct trace_buffer_struct
*buffers
;
1910 struct trace_buffer_struct
*sirq_buffers
;
1911 struct trace_buffer_struct
*irq_buffers
;
1912 struct trace_buffer_struct
*nmi_buffers
;
1914 buffers
= alloc_percpu(struct trace_buffer_struct
);
1918 sirq_buffers
= alloc_percpu(struct trace_buffer_struct
);
1922 irq_buffers
= alloc_percpu(struct trace_buffer_struct
);
1926 nmi_buffers
= alloc_percpu(struct trace_buffer_struct
);
1930 trace_percpu_buffer
= buffers
;
1931 trace_percpu_sirq_buffer
= sirq_buffers
;
1932 trace_percpu_irq_buffer
= irq_buffers
;
1933 trace_percpu_nmi_buffer
= nmi_buffers
;
1938 free_percpu(irq_buffers
);
1940 free_percpu(sirq_buffers
);
1942 free_percpu(buffers
);
1944 WARN(1, "Could not allocate percpu trace_printk buffer");
1948 static int buffers_allocated
;
1950 void trace_printk_init_buffers(void)
1952 if (buffers_allocated
)
1955 if (alloc_percpu_trace_buffer())
1958 pr_info("ftrace: Allocated trace_printk buffers\n");
1960 /* Expand the buffers to set size */
1961 tracing_update_buffers();
1963 buffers_allocated
= 1;
1966 * trace_printk_init_buffers() can be called by modules.
1967 * If that happens, then we need to start cmdline recording
1968 * directly here. If the global_trace.buffer is already
1969 * allocated here, then this was called by module code.
1971 if (global_trace
.trace_buffer
.buffer
)
1972 tracing_start_cmdline_record();
1975 void trace_printk_start_comm(void)
1977 /* Start tracing comms if trace printk is set */
1978 if (!buffers_allocated
)
1980 tracing_start_cmdline_record();
1983 static void trace_printk_start_stop_comm(int enabled
)
1985 if (!buffers_allocated
)
1989 tracing_start_cmdline_record();
1991 tracing_stop_cmdline_record();
1995 * trace_vbprintk - write binary msg to tracing buffer
1998 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
2000 struct ftrace_event_call
*call
= &event_bprint
;
2001 struct ring_buffer_event
*event
;
2002 struct ring_buffer
*buffer
;
2003 struct trace_array
*tr
= &global_trace
;
2004 struct bprint_entry
*entry
;
2005 unsigned long flags
;
2007 int len
= 0, size
, pc
;
2009 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
2012 /* Don't pollute graph traces with trace_vprintk internals */
2013 pause_graph_tracing();
2015 pc
= preempt_count();
2016 preempt_disable_notrace();
2018 tbuffer
= get_trace_buf();
2024 len
= vbin_printf((u32
*)tbuffer
, TRACE_BUF_SIZE
/sizeof(int), fmt
, args
);
2026 if (len
> TRACE_BUF_SIZE
/sizeof(int) || len
< 0)
2029 local_save_flags(flags
);
2030 size
= sizeof(*entry
) + sizeof(u32
) * len
;
2031 buffer
= tr
->trace_buffer
.buffer
;
2032 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
2036 entry
= ring_buffer_event_data(event
);
2040 memcpy(entry
->buf
, tbuffer
, sizeof(u32
) * len
);
2041 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
2042 __buffer_unlock_commit(buffer
, event
);
2043 ftrace_trace_stack(buffer
, flags
, 6, pc
);
2047 preempt_enable_notrace();
2048 unpause_graph_tracing();
2052 EXPORT_SYMBOL_GPL(trace_vbprintk
);
2055 __trace_array_vprintk(struct ring_buffer
*buffer
,
2056 unsigned long ip
, const char *fmt
, va_list args
)
2058 struct ftrace_event_call
*call
= &event_print
;
2059 struct ring_buffer_event
*event
;
2060 int len
= 0, size
, pc
;
2061 struct print_entry
*entry
;
2062 unsigned long flags
;
2065 if (tracing_disabled
|| tracing_selftest_running
)
2068 /* Don't pollute graph traces with trace_vprintk internals */
2069 pause_graph_tracing();
2071 pc
= preempt_count();
2072 preempt_disable_notrace();
2075 tbuffer
= get_trace_buf();
2081 len
= vsnprintf(tbuffer
, TRACE_BUF_SIZE
, fmt
, args
);
2082 if (len
> TRACE_BUF_SIZE
)
2085 local_save_flags(flags
);
2086 size
= sizeof(*entry
) + len
+ 1;
2087 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
2091 entry
= ring_buffer_event_data(event
);
2094 memcpy(&entry
->buf
, tbuffer
, len
);
2095 entry
->buf
[len
] = '\0';
2096 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
2097 __buffer_unlock_commit(buffer
, event
);
2098 ftrace_trace_stack(buffer
, flags
, 6, pc
);
2101 preempt_enable_notrace();
2102 unpause_graph_tracing();
2107 int trace_array_vprintk(struct trace_array
*tr
,
2108 unsigned long ip
, const char *fmt
, va_list args
)
2110 return __trace_array_vprintk(tr
->trace_buffer
.buffer
, ip
, fmt
, args
);
2113 int trace_array_printk(struct trace_array
*tr
,
2114 unsigned long ip
, const char *fmt
, ...)
2119 if (!(trace_flags
& TRACE_ITER_PRINTK
))
2123 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
2128 int trace_array_printk_buf(struct ring_buffer
*buffer
,
2129 unsigned long ip
, const char *fmt
, ...)
2134 if (!(trace_flags
& TRACE_ITER_PRINTK
))
2138 ret
= __trace_array_vprintk(buffer
, ip
, fmt
, ap
);
2143 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
2145 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
2147 EXPORT_SYMBOL_GPL(trace_vprintk
);
2149 static void trace_iterator_increment(struct trace_iterator
*iter
)
2151 struct ring_buffer_iter
*buf_iter
= trace_buffer_iter(iter
, iter
->cpu
);
2155 ring_buffer_read(buf_iter
, NULL
);
2158 static struct trace_entry
*
2159 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
,
2160 unsigned long *lost_events
)
2162 struct ring_buffer_event
*event
;
2163 struct ring_buffer_iter
*buf_iter
= trace_buffer_iter(iter
, cpu
);
2166 event
= ring_buffer_iter_peek(buf_iter
, ts
);
2168 event
= ring_buffer_peek(iter
->trace_buffer
->buffer
, cpu
, ts
,
2172 iter
->ent_size
= ring_buffer_event_length(event
);
2173 return ring_buffer_event_data(event
);
2179 static struct trace_entry
*
2180 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
,
2181 unsigned long *missing_events
, u64
*ent_ts
)
2183 struct ring_buffer
*buffer
= iter
->trace_buffer
->buffer
;
2184 struct trace_entry
*ent
, *next
= NULL
;
2185 unsigned long lost_events
= 0, next_lost
= 0;
2186 int cpu_file
= iter
->cpu_file
;
2187 u64 next_ts
= 0, ts
;
2193 * If we are in a per_cpu trace file, don't bother by iterating over
2194 * all cpu and peek directly.
2196 if (cpu_file
> RING_BUFFER_ALL_CPUS
) {
2197 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
2199 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
, missing_events
);
2201 *ent_cpu
= cpu_file
;
2206 for_each_tracing_cpu(cpu
) {
2208 if (ring_buffer_empty_cpu(buffer
, cpu
))
2211 ent
= peek_next_entry(iter
, cpu
, &ts
, &lost_events
);
2214 * Pick the entry with the smallest timestamp:
2216 if (ent
&& (!next
|| ts
< next_ts
)) {
2220 next_lost
= lost_events
;
2221 next_size
= iter
->ent_size
;
2225 iter
->ent_size
= next_size
;
2228 *ent_cpu
= next_cpu
;
2234 *missing_events
= next_lost
;
2239 /* Find the next real entry, without updating the iterator itself */
2240 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
2241 int *ent_cpu
, u64
*ent_ts
)
2243 return __find_next_entry(iter
, ent_cpu
, NULL
, ent_ts
);
2246 /* Find the next real entry, and increment the iterator to the next entry */
2247 void *trace_find_next_entry_inc(struct trace_iterator
*iter
)
2249 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
,
2250 &iter
->lost_events
, &iter
->ts
);
2253 trace_iterator_increment(iter
);
2255 return iter
->ent
? iter
: NULL
;
2258 static void trace_consume(struct trace_iterator
*iter
)
2260 ring_buffer_consume(iter
->trace_buffer
->buffer
, iter
->cpu
, &iter
->ts
,
2261 &iter
->lost_events
);
2264 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2266 struct trace_iterator
*iter
= m
->private;
2270 WARN_ON_ONCE(iter
->leftover
);
2274 /* can't go backwards */
2279 ent
= trace_find_next_entry_inc(iter
);
2283 while (ent
&& iter
->idx
< i
)
2284 ent
= trace_find_next_entry_inc(iter
);
2291 void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
2293 struct ring_buffer_event
*event
;
2294 struct ring_buffer_iter
*buf_iter
;
2295 unsigned long entries
= 0;
2298 per_cpu_ptr(iter
->trace_buffer
->data
, cpu
)->skipped_entries
= 0;
2300 buf_iter
= trace_buffer_iter(iter
, cpu
);
2304 ring_buffer_iter_reset(buf_iter
);
2307 * We could have the case with the max latency tracers
2308 * that a reset never took place on a cpu. This is evident
2309 * by the timestamp being before the start of the buffer.
2311 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
2312 if (ts
>= iter
->trace_buffer
->time_start
)
2315 ring_buffer_read(buf_iter
, NULL
);
2318 per_cpu_ptr(iter
->trace_buffer
->data
, cpu
)->skipped_entries
= entries
;
2322 * The current tracer is copied to avoid a global locking
2325 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
2327 struct trace_iterator
*iter
= m
->private;
2328 struct trace_array
*tr
= iter
->tr
;
2329 int cpu_file
= iter
->cpu_file
;
2335 * copy the tracer to avoid using a global lock all around.
2336 * iter->trace is a copy of current_trace, the pointer to the
2337 * name may be used instead of a strcmp(), as iter->trace->name
2338 * will point to the same string as current_trace->name.
2340 mutex_lock(&trace_types_lock
);
2341 if (unlikely(tr
->current_trace
&& iter
->trace
->name
!= tr
->current_trace
->name
))
2342 *iter
->trace
= *tr
->current_trace
;
2343 mutex_unlock(&trace_types_lock
);
2345 #ifdef CONFIG_TRACER_MAX_TRACE
2346 if (iter
->snapshot
&& iter
->trace
->use_max_tr
)
2347 return ERR_PTR(-EBUSY
);
2350 if (!iter
->snapshot
)
2351 atomic_inc(&trace_record_cmdline_disabled
);
2353 if (*pos
!= iter
->pos
) {
2358 if (cpu_file
== RING_BUFFER_ALL_CPUS
) {
2359 for_each_tracing_cpu(cpu
)
2360 tracing_iter_reset(iter
, cpu
);
2362 tracing_iter_reset(iter
, cpu_file
);
2365 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
2370 * If we overflowed the seq_file before, then we want
2371 * to just reuse the trace_seq buffer again.
2377 p
= s_next(m
, p
, &l
);
2381 trace_event_read_lock();
2382 trace_access_lock(cpu_file
);
2386 static void s_stop(struct seq_file
*m
, void *p
)
2388 struct trace_iterator
*iter
= m
->private;
2390 #ifdef CONFIG_TRACER_MAX_TRACE
2391 if (iter
->snapshot
&& iter
->trace
->use_max_tr
)
2395 if (!iter
->snapshot
)
2396 atomic_dec(&trace_record_cmdline_disabled
);
2398 trace_access_unlock(iter
->cpu_file
);
2399 trace_event_read_unlock();
2403 get_total_entries(struct trace_buffer
*buf
,
2404 unsigned long *total
, unsigned long *entries
)
2406 unsigned long count
;
2412 for_each_tracing_cpu(cpu
) {
2413 count
= ring_buffer_entries_cpu(buf
->buffer
, cpu
);
2415 * If this buffer has skipped entries, then we hold all
2416 * entries for the trace and we need to ignore the
2417 * ones before the time stamp.
2419 if (per_cpu_ptr(buf
->data
, cpu
)->skipped_entries
) {
2420 count
-= per_cpu_ptr(buf
->data
, cpu
)->skipped_entries
;
2421 /* total is the same as the entries */
2425 ring_buffer_overrun_cpu(buf
->buffer
, cpu
);
2430 static void print_lat_help_header(struct seq_file
*m
)
2432 seq_puts(m
, "# _------=> CPU# \n");
2433 seq_puts(m
, "# / _-----=> irqs-off \n");
2434 seq_puts(m
, "# | / _----=> need-resched \n");
2435 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
2436 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
2437 seq_puts(m
, "# |||| / delay \n");
2438 seq_puts(m
, "# cmd pid ||||| time | caller \n");
2439 seq_puts(m
, "# \\ / ||||| \\ | / \n");
2442 static void print_event_info(struct trace_buffer
*buf
, struct seq_file
*m
)
2444 unsigned long total
;
2445 unsigned long entries
;
2447 get_total_entries(buf
, &total
, &entries
);
2448 seq_printf(m
, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2449 entries
, total
, num_online_cpus());
2453 static void print_func_help_header(struct trace_buffer
*buf
, struct seq_file
*m
)
2455 print_event_info(buf
, m
);
2456 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2457 seq_puts(m
, "# | | | | |\n");
2460 static void print_func_help_header_tgid(struct trace_buffer
*buf
, struct seq_file
*m
)
2462 print_event_info(buf
, m
);
2463 seq_puts(m
, "# TASK-PID TGID CPU# TIMESTAMP FUNCTION\n");
2464 seq_puts(m
, "# | | | | | |\n");
2467 static void print_func_help_header_irq(struct trace_buffer
*buf
, struct seq_file
*m
)
2469 print_event_info(buf
, m
);
2470 seq_puts(m
, "# _-----=> irqs-off\n");
2471 seq_puts(m
, "# / _----=> need-resched\n");
2472 seq_puts(m
, "# | / _---=> hardirq/softirq\n");
2473 seq_puts(m
, "# || / _--=> preempt-depth\n");
2474 seq_puts(m
, "# ||| / delay\n");
2475 seq_puts(m
, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2476 seq_puts(m
, "# | | | |||| | |\n");
2479 static void print_func_help_header_irq_tgid(struct trace_buffer
*buf
, struct seq_file
*m
)
2481 print_event_info(buf
, m
);
2482 seq_puts(m
, "# _-----=> irqs-off\n");
2483 seq_puts(m
, "# / _----=> need-resched\n");
2484 seq_puts(m
, "# | / _---=> hardirq/softirq\n");
2485 seq_puts(m
, "# || / _--=> preempt-depth\n");
2486 seq_puts(m
, "# ||| / delay\n");
2487 seq_puts(m
, "# TASK-PID TGID CPU# |||| TIMESTAMP FUNCTION\n");
2488 seq_puts(m
, "# | | | | |||| | |\n");
2492 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
2494 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2495 struct trace_buffer
*buf
= iter
->trace_buffer
;
2496 struct trace_array_cpu
*data
= per_cpu_ptr(buf
->data
, buf
->cpu
);
2497 struct tracer
*type
= iter
->trace
;
2498 unsigned long entries
;
2499 unsigned long total
;
2500 const char *name
= "preemption";
2504 get_total_entries(buf
, &total
, &entries
);
2506 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
2508 seq_puts(m
, "# -----------------------------------"
2509 "---------------------------------\n");
2510 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2511 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2512 nsecs_to_usecs(data
->saved_latency
),
2516 #if defined(CONFIG_PREEMPT_NONE)
2518 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2520 #elif defined(CONFIG_PREEMPT)
2525 /* These are reserved for later use */
2528 seq_printf(m
, " #P:%d)\n", num_online_cpus());
2532 seq_puts(m
, "# -----------------\n");
2533 seq_printf(m
, "# | task: %.16s-%d "
2534 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2535 data
->comm
, data
->pid
,
2536 from_kuid_munged(seq_user_ns(m
), data
->uid
), data
->nice
,
2537 data
->policy
, data
->rt_priority
);
2538 seq_puts(m
, "# -----------------\n");
2540 if (data
->critical_start
) {
2541 seq_puts(m
, "# => started at: ");
2542 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
2543 trace_print_seq(m
, &iter
->seq
);
2544 seq_puts(m
, "\n# => ended at: ");
2545 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
2546 trace_print_seq(m
, &iter
->seq
);
2547 seq_puts(m
, "\n#\n");
2553 static void test_cpu_buff_start(struct trace_iterator
*iter
)
2555 struct trace_seq
*s
= &iter
->seq
;
2557 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
2560 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
2563 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
2566 if (per_cpu_ptr(iter
->trace_buffer
->data
, iter
->cpu
)->skipped_entries
)
2569 cpumask_set_cpu(iter
->cpu
, iter
->started
);
2571 /* Don't print started cpu buffer for the first entry of the trace */
2573 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
2577 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
2579 struct trace_seq
*s
= &iter
->seq
;
2580 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2581 struct trace_entry
*entry
;
2582 struct trace_event
*event
;
2586 test_cpu_buff_start(iter
);
2588 event
= ftrace_find_event(entry
->type
);
2590 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2591 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2592 if (!trace_print_lat_context(iter
))
2595 if (!trace_print_context(iter
))
2601 return event
->funcs
->trace(iter
, sym_flags
, event
);
2603 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
2606 return TRACE_TYPE_HANDLED
;
2608 return TRACE_TYPE_PARTIAL_LINE
;
2611 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
2613 struct trace_seq
*s
= &iter
->seq
;
2614 struct trace_entry
*entry
;
2615 struct trace_event
*event
;
2619 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2620 if (!trace_seq_printf(s
, "%d %d %llu ",
2621 entry
->pid
, iter
->cpu
, iter
->ts
))
2625 event
= ftrace_find_event(entry
->type
);
2627 return event
->funcs
->raw(iter
, 0, event
);
2629 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
2632 return TRACE_TYPE_HANDLED
;
2634 return TRACE_TYPE_PARTIAL_LINE
;
2637 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
2639 struct trace_seq
*s
= &iter
->seq
;
2640 unsigned char newline
= '\n';
2641 struct trace_entry
*entry
;
2642 struct trace_event
*event
;
2646 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2647 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
2648 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
2649 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
2652 event
= ftrace_find_event(entry
->type
);
2654 enum print_line_t ret
= event
->funcs
->hex(iter
, 0, event
);
2655 if (ret
!= TRACE_TYPE_HANDLED
)
2659 SEQ_PUT_FIELD_RET(s
, newline
);
2661 return TRACE_TYPE_HANDLED
;
2664 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
2666 struct trace_seq
*s
= &iter
->seq
;
2667 struct trace_entry
*entry
;
2668 struct trace_event
*event
;
2672 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2673 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
2674 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
2675 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
2678 event
= ftrace_find_event(entry
->type
);
2679 return event
? event
->funcs
->binary(iter
, 0, event
) :
2683 int trace_empty(struct trace_iterator
*iter
)
2685 struct ring_buffer_iter
*buf_iter
;
2688 /* If we are looking at one CPU buffer, only check that one */
2689 if (iter
->cpu_file
!= RING_BUFFER_ALL_CPUS
) {
2690 cpu
= iter
->cpu_file
;
2691 buf_iter
= trace_buffer_iter(iter
, cpu
);
2693 if (!ring_buffer_iter_empty(buf_iter
))
2696 if (!ring_buffer_empty_cpu(iter
->trace_buffer
->buffer
, cpu
))
2702 for_each_tracing_cpu(cpu
) {
2703 buf_iter
= trace_buffer_iter(iter
, cpu
);
2705 if (!ring_buffer_iter_empty(buf_iter
))
2708 if (!ring_buffer_empty_cpu(iter
->trace_buffer
->buffer
, cpu
))
2716 /* Called with trace_event_read_lock() held. */
2717 enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
2719 enum print_line_t ret
;
2721 if (iter
->lost_events
&&
2722 !trace_seq_printf(&iter
->seq
, "CPU:%d [LOST %lu EVENTS]\n",
2723 iter
->cpu
, iter
->lost_events
))
2724 return TRACE_TYPE_PARTIAL_LINE
;
2726 if (iter
->trace
&& iter
->trace
->print_line
) {
2727 ret
= iter
->trace
->print_line(iter
);
2728 if (ret
!= TRACE_TYPE_UNHANDLED
)
2732 if (iter
->ent
->type
== TRACE_BPUTS
&&
2733 trace_flags
& TRACE_ITER_PRINTK
&&
2734 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2735 return trace_print_bputs_msg_only(iter
);
2737 if (iter
->ent
->type
== TRACE_BPRINT
&&
2738 trace_flags
& TRACE_ITER_PRINTK
&&
2739 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2740 return trace_print_bprintk_msg_only(iter
);
2742 if (iter
->ent
->type
== TRACE_PRINT
&&
2743 trace_flags
& TRACE_ITER_PRINTK
&&
2744 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2745 return trace_print_printk_msg_only(iter
);
2747 if (trace_flags
& TRACE_ITER_BIN
)
2748 return print_bin_fmt(iter
);
2750 if (trace_flags
& TRACE_ITER_HEX
)
2751 return print_hex_fmt(iter
);
2753 if (trace_flags
& TRACE_ITER_RAW
)
2754 return print_raw_fmt(iter
);
2756 return print_trace_fmt(iter
);
2759 void trace_latency_header(struct seq_file
*m
)
2761 struct trace_iterator
*iter
= m
->private;
2763 /* print nothing if the buffers are empty */
2764 if (trace_empty(iter
))
2767 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
2768 print_trace_header(m
, iter
);
2770 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2771 print_lat_help_header(m
);
2774 void trace_default_header(struct seq_file
*m
)
2776 struct trace_iterator
*iter
= m
->private;
2778 if (!(trace_flags
& TRACE_ITER_CONTEXT_INFO
))
2781 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2782 /* print nothing if the buffers are empty */
2783 if (trace_empty(iter
))
2785 print_trace_header(m
, iter
);
2786 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2787 print_lat_help_header(m
);
2789 if (!(trace_flags
& TRACE_ITER_VERBOSE
)) {
2790 if (trace_flags
& TRACE_ITER_IRQ_INFO
)
2791 if (trace_flags
& TRACE_ITER_TGID
)
2792 print_func_help_header_irq_tgid(iter
->trace_buffer
, m
);
2794 print_func_help_header_irq(iter
->trace_buffer
, m
);
2796 if (trace_flags
& TRACE_ITER_TGID
)
2797 print_func_help_header_tgid(iter
->trace_buffer
, m
);
2799 print_func_help_header(iter
->trace_buffer
, m
);
2804 static void test_ftrace_alive(struct seq_file
*m
)
2806 if (!ftrace_is_dead())
2808 seq_printf(m
, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2809 seq_printf(m
, "# MAY BE MISSING FUNCTION EVENTS\n");
2812 #ifdef CONFIG_TRACER_MAX_TRACE
2813 static void show_snapshot_main_help(struct seq_file
*m
)
2815 seq_printf(m
, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2816 seq_printf(m
, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2817 seq_printf(m
, "# Takes a snapshot of the main buffer.\n");
2818 seq_printf(m
, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate)\n");
2819 seq_printf(m
, "# (Doesn't have to be '2' works with any number that\n");
2820 seq_printf(m
, "# is not a '0' or '1')\n");
2823 static void show_snapshot_percpu_help(struct seq_file
*m
)
2825 seq_printf(m
, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2826 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2827 seq_printf(m
, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2828 seq_printf(m
, "# Takes a snapshot of the main buffer for this cpu.\n");
2830 seq_printf(m
, "# echo 1 > snapshot : Not supported with this kernel.\n");
2831 seq_printf(m
, "# Must use main snapshot file to allocate.\n");
2833 seq_printf(m
, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2834 seq_printf(m
, "# (Doesn't have to be '2' works with any number that\n");
2835 seq_printf(m
, "# is not a '0' or '1')\n");
2838 static void print_snapshot_help(struct seq_file
*m
, struct trace_iterator
*iter
)
2840 if (iter
->tr
->allocated_snapshot
)
2841 seq_printf(m
, "#\n# * Snapshot is allocated *\n#\n");
2843 seq_printf(m
, "#\n# * Snapshot is freed *\n#\n");
2845 seq_printf(m
, "# Snapshot commands:\n");
2846 if (iter
->cpu_file
== RING_BUFFER_ALL_CPUS
)
2847 show_snapshot_main_help(m
);
2849 show_snapshot_percpu_help(m
);
2852 /* Should never be called */
2853 static inline void print_snapshot_help(struct seq_file
*m
, struct trace_iterator
*iter
) { }
2856 static int s_show(struct seq_file
*m
, void *v
)
2858 struct trace_iterator
*iter
= v
;
2861 if (iter
->ent
== NULL
) {
2863 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
2865 test_ftrace_alive(m
);
2867 if (iter
->snapshot
&& trace_empty(iter
))
2868 print_snapshot_help(m
, iter
);
2869 else if (iter
->trace
&& iter
->trace
->print_header
)
2870 iter
->trace
->print_header(m
);
2872 trace_default_header(m
);
2874 } else if (iter
->leftover
) {
2876 * If we filled the seq_file buffer earlier, we
2877 * want to just show it now.
2879 ret
= trace_print_seq(m
, &iter
->seq
);
2881 /* ret should this time be zero, but you never know */
2882 iter
->leftover
= ret
;
2885 print_trace_line(iter
);
2886 ret
= trace_print_seq(m
, &iter
->seq
);
2888 * If we overflow the seq_file buffer, then it will
2889 * ask us for this data again at start up.
2891 * ret is 0 if seq_file write succeeded.
2894 iter
->leftover
= ret
;
2901 * Should be used after trace_array_get(), trace_types_lock
2902 * ensures that i_cdev was already initialized.
2904 static inline int tracing_get_cpu(struct inode
*inode
)
2906 if (inode
->i_cdev
) /* See trace_create_cpu_file() */
2907 return (long)inode
->i_cdev
- 1;
2908 return RING_BUFFER_ALL_CPUS
;
2911 static const struct seq_operations tracer_seq_ops
= {
2918 static struct trace_iterator
*
2919 __tracing_open(struct inode
*inode
, struct file
*file
, bool snapshot
)
2921 struct trace_array
*tr
= inode
->i_private
;
2922 struct trace_iterator
*iter
;
2925 if (tracing_disabled
)
2926 return ERR_PTR(-ENODEV
);
2928 iter
= __seq_open_private(file
, &tracer_seq_ops
, sizeof(*iter
));
2930 return ERR_PTR(-ENOMEM
);
2932 iter
->buffer_iter
= kzalloc(sizeof(*iter
->buffer_iter
) * num_possible_cpus(),
2934 if (!iter
->buffer_iter
)
2938 * We make a copy of the current tracer to avoid concurrent
2939 * changes on it while we are reading.
2941 mutex_lock(&trace_types_lock
);
2942 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2946 *iter
->trace
= *tr
->current_trace
;
2948 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2953 #ifdef CONFIG_TRACER_MAX_TRACE
2954 /* Currently only the top directory has a snapshot */
2955 if (tr
->current_trace
->print_max
|| snapshot
)
2956 iter
->trace_buffer
= &tr
->max_buffer
;
2959 iter
->trace_buffer
= &tr
->trace_buffer
;
2960 iter
->snapshot
= snapshot
;
2962 iter
->cpu_file
= tracing_get_cpu(inode
);
2963 mutex_init(&iter
->mutex
);
2965 /* Notify the tracer early; before we stop tracing. */
2966 if (iter
->trace
&& iter
->trace
->open
)
2967 iter
->trace
->open(iter
);
2969 /* Annotate start of buffers if we had overruns */
2970 if (ring_buffer_overruns(iter
->trace_buffer
->buffer
))
2971 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2973 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2974 if (trace_clocks
[tr
->clock_id
].in_ns
)
2975 iter
->iter_flags
|= TRACE_FILE_TIME_IN_NS
;
2977 /* stop the trace while dumping if we are not opening "snapshot" */
2978 if (!iter
->snapshot
)
2979 tracing_stop_tr(tr
);
2981 if (iter
->cpu_file
== RING_BUFFER_ALL_CPUS
) {
2982 for_each_tracing_cpu(cpu
) {
2983 iter
->buffer_iter
[cpu
] =
2984 ring_buffer_read_prepare(iter
->trace_buffer
->buffer
, cpu
);
2986 ring_buffer_read_prepare_sync();
2987 for_each_tracing_cpu(cpu
) {
2988 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2989 tracing_iter_reset(iter
, cpu
);
2992 cpu
= iter
->cpu_file
;
2993 iter
->buffer_iter
[cpu
] =
2994 ring_buffer_read_prepare(iter
->trace_buffer
->buffer
, cpu
);
2995 ring_buffer_read_prepare_sync();
2996 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2997 tracing_iter_reset(iter
, cpu
);
3000 mutex_unlock(&trace_types_lock
);
3005 mutex_unlock(&trace_types_lock
);
3007 kfree(iter
->buffer_iter
);
3009 seq_release_private(inode
, file
);
3010 return ERR_PTR(-ENOMEM
);
3013 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
3015 if (tracing_disabled
)
3018 filp
->private_data
= inode
->i_private
;
3023 * Open and update trace_array ref count.
3024 * Must have the current trace_array passed to it.
3026 int tracing_open_generic_tr(struct inode
*inode
, struct file
*filp
)
3028 struct trace_array
*tr
= inode
->i_private
;
3030 if (tracing_disabled
)
3033 if (trace_array_get(tr
) < 0)
3036 filp
->private_data
= inode
->i_private
;
3041 static int tracing_release(struct inode
*inode
, struct file
*file
)
3043 struct trace_array
*tr
= inode
->i_private
;
3044 struct seq_file
*m
= file
->private_data
;
3045 struct trace_iterator
*iter
;
3048 if (!(file
->f_mode
& FMODE_READ
)) {
3049 trace_array_put(tr
);
3053 /* Writes do not use seq_file */
3055 mutex_lock(&trace_types_lock
);
3057 for_each_tracing_cpu(cpu
) {
3058 if (iter
->buffer_iter
[cpu
])
3059 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
3062 if (iter
->trace
&& iter
->trace
->close
)
3063 iter
->trace
->close(iter
);
3065 if (!iter
->snapshot
)
3066 /* reenable tracing if it was previously enabled */
3067 tracing_start_tr(tr
);
3069 __trace_array_put(tr
);
3071 mutex_unlock(&trace_types_lock
);
3073 mutex_destroy(&iter
->mutex
);
3074 free_cpumask_var(iter
->started
);
3076 kfree(iter
->buffer_iter
);
3077 seq_release_private(inode
, file
);
3082 static int tracing_release_generic_tr(struct inode
*inode
, struct file
*file
)
3084 struct trace_array
*tr
= inode
->i_private
;
3086 trace_array_put(tr
);
3090 static int tracing_single_release_tr(struct inode
*inode
, struct file
*file
)
3092 struct trace_array
*tr
= inode
->i_private
;
3094 trace_array_put(tr
);
3096 return single_release(inode
, file
);
3099 static int tracing_open(struct inode
*inode
, struct file
*file
)
3101 struct trace_array
*tr
= inode
->i_private
;
3102 struct trace_iterator
*iter
;
3105 if (trace_array_get(tr
) < 0)
3108 /* If this file was open for write, then erase contents */
3109 if ((file
->f_mode
& FMODE_WRITE
) && (file
->f_flags
& O_TRUNC
)) {
3110 int cpu
= tracing_get_cpu(inode
);
3112 if (cpu
== RING_BUFFER_ALL_CPUS
)
3113 tracing_reset_online_cpus(&tr
->trace_buffer
);
3115 tracing_reset(&tr
->trace_buffer
, cpu
);
3118 if (file
->f_mode
& FMODE_READ
) {
3119 iter
= __tracing_open(inode
, file
, false);
3121 ret
= PTR_ERR(iter
);
3122 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
3123 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
3127 trace_array_put(tr
);
3133 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
3135 struct tracer
*t
= v
;
3145 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
3150 mutex_lock(&trace_types_lock
);
3151 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
3157 static void t_stop(struct seq_file
*m
, void *p
)
3159 mutex_unlock(&trace_types_lock
);
3162 static int t_show(struct seq_file
*m
, void *v
)
3164 struct tracer
*t
= v
;
3169 seq_printf(m
, "%s", t
->name
);
3178 static const struct seq_operations show_traces_seq_ops
= {
3185 static int show_traces_open(struct inode
*inode
, struct file
*file
)
3187 if (tracing_disabled
)
3190 return seq_open(file
, &show_traces_seq_ops
);
3194 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
3195 size_t count
, loff_t
*ppos
)
3200 static loff_t
tracing_seek(struct file
*file
, loff_t offset
, int origin
)
3202 if (file
->f_mode
& FMODE_READ
)
3203 return seq_lseek(file
, offset
, origin
);
3208 static const struct file_operations tracing_fops
= {
3209 .open
= tracing_open
,
3211 .write
= tracing_write_stub
,
3212 .llseek
= tracing_seek
,
3213 .release
= tracing_release
,
3216 static const struct file_operations show_traces_fops
= {
3217 .open
= show_traces_open
,
3219 .release
= seq_release
,
3220 .llseek
= seq_lseek
,
3224 * Only trace on a CPU if the bitmask is set:
3226 static cpumask_var_t tracing_cpumask
;
3229 * The tracer itself will not take this lock, but still we want
3230 * to provide a consistent cpumask to user-space:
3232 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
3235 * Temporary storage for the character representation of the
3236 * CPU bitmask (and one more byte for the newline):
3238 static char mask_str
[NR_CPUS
+ 1];
3241 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
3242 size_t count
, loff_t
*ppos
)
3246 mutex_lock(&tracing_cpumask_update_lock
);
3248 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
3249 if (count
- len
< 2) {
3253 len
+= sprintf(mask_str
+ len
, "\n");
3254 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
3257 mutex_unlock(&tracing_cpumask_update_lock
);
3263 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
3264 size_t count
, loff_t
*ppos
)
3266 struct trace_array
*tr
= filp
->private_data
;
3267 cpumask_var_t tracing_cpumask_new
;
3270 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
3273 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
3277 mutex_lock(&tracing_cpumask_update_lock
);
3279 local_irq_disable();
3280 arch_spin_lock(&ftrace_max_lock
);
3281 for_each_tracing_cpu(cpu
) {
3283 * Increase/decrease the disabled counter if we are
3284 * about to flip a bit in the cpumask:
3286 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
3287 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
3288 atomic_inc(&per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->disabled
);
3289 ring_buffer_record_disable_cpu(tr
->trace_buffer
.buffer
, cpu
);
3291 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
3292 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
3293 atomic_dec(&per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->disabled
);
3294 ring_buffer_record_enable_cpu(tr
->trace_buffer
.buffer
, cpu
);
3297 arch_spin_unlock(&ftrace_max_lock
);
3300 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
3302 mutex_unlock(&tracing_cpumask_update_lock
);
3303 free_cpumask_var(tracing_cpumask_new
);
3308 free_cpumask_var(tracing_cpumask_new
);
3313 static const struct file_operations tracing_cpumask_fops
= {
3314 .open
= tracing_open_generic
,
3315 .read
= tracing_cpumask_read
,
3316 .write
= tracing_cpumask_write
,
3317 .llseek
= generic_file_llseek
,
3320 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
3322 struct tracer_opt
*trace_opts
;
3323 struct trace_array
*tr
= m
->private;
3327 mutex_lock(&trace_types_lock
);
3328 tracer_flags
= tr
->current_trace
->flags
->val
;
3329 trace_opts
= tr
->current_trace
->flags
->opts
;
3331 for (i
= 0; trace_options
[i
]; i
++) {
3332 if (trace_flags
& (1 << i
))
3333 seq_printf(m
, "%s\n", trace_options
[i
]);
3335 seq_printf(m
, "no%s\n", trace_options
[i
]);
3338 for (i
= 0; trace_opts
[i
].name
; i
++) {
3339 if (tracer_flags
& trace_opts
[i
].bit
)
3340 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
3342 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
3344 mutex_unlock(&trace_types_lock
);
3349 static int __set_tracer_option(struct tracer
*trace
,
3350 struct tracer_flags
*tracer_flags
,
3351 struct tracer_opt
*opts
, int neg
)
3355 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
3360 tracer_flags
->val
&= ~opts
->bit
;
3362 tracer_flags
->val
|= opts
->bit
;
3366 /* Try to assign a tracer specific option */
3367 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
3369 struct tracer_flags
*tracer_flags
= trace
->flags
;
3370 struct tracer_opt
*opts
= NULL
;
3373 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
3374 opts
= &tracer_flags
->opts
[i
];
3376 if (strcmp(cmp
, opts
->name
) == 0)
3377 return __set_tracer_option(trace
, trace
->flags
,
3384 /* Some tracers require overwrite to stay enabled */
3385 int trace_keep_overwrite(struct tracer
*tracer
, u32 mask
, int set
)
3387 if (tracer
->enabled
&& (mask
& TRACE_ITER_OVERWRITE
) && !set
)
3393 int set_tracer_flag(struct trace_array
*tr
, unsigned int mask
, int enabled
)
3395 /* do nothing if flag is already set */
3396 if (!!(trace_flags
& mask
) == !!enabled
)
3399 /* Give the tracer a chance to approve the change */
3400 if (tr
->current_trace
->flag_changed
)
3401 if (tr
->current_trace
->flag_changed(tr
->current_trace
, mask
, !!enabled
))
3405 trace_flags
|= mask
;
3407 trace_flags
&= ~mask
;
3409 if (mask
== TRACE_ITER_RECORD_CMD
)
3410 trace_event_enable_cmd_record(enabled
);
3412 if (mask
== TRACE_ITER_OVERWRITE
) {
3413 ring_buffer_change_overwrite(tr
->trace_buffer
.buffer
, enabled
);
3414 #ifdef CONFIG_TRACER_MAX_TRACE
3415 ring_buffer_change_overwrite(tr
->max_buffer
.buffer
, enabled
);
3419 if (mask
== TRACE_ITER_PRINTK
)
3420 trace_printk_start_stop_comm(enabled
);
3425 static int trace_set_options(struct trace_array
*tr
, char *option
)
3432 cmp
= strstrip(option
);
3434 if (strncmp(cmp
, "no", 2) == 0) {
3439 mutex_lock(&trace_types_lock
);
3441 for (i
= 0; trace_options
[i
]; i
++) {
3442 if (strcmp(cmp
, trace_options
[i
]) == 0) {
3443 ret
= set_tracer_flag(tr
, 1 << i
, !neg
);
3448 /* If no option could be set, test the specific tracer options */
3449 if (!trace_options
[i
])
3450 ret
= set_tracer_option(tr
->current_trace
, cmp
, neg
);
3452 mutex_unlock(&trace_types_lock
);
3458 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
3459 size_t cnt
, loff_t
*ppos
)
3461 struct seq_file
*m
= filp
->private_data
;
3462 struct trace_array
*tr
= m
->private;
3466 if (cnt
>= sizeof(buf
))
3469 if (copy_from_user(&buf
, ubuf
, cnt
))
3474 ret
= trace_set_options(tr
, buf
);
3483 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
3485 struct trace_array
*tr
= inode
->i_private
;
3488 if (tracing_disabled
)
3491 if (trace_array_get(tr
) < 0)
3494 ret
= single_open(file
, tracing_trace_options_show
, inode
->i_private
);
3496 trace_array_put(tr
);
3501 static const struct file_operations tracing_iter_fops
= {
3502 .open
= tracing_trace_options_open
,
3504 .llseek
= seq_lseek
,
3505 .release
= tracing_single_release_tr
,
3506 .write
= tracing_trace_options_write
,
3509 static const char readme_msg
[] =
3510 "tracing mini-HOWTO:\n\n"
3511 "# echo 0 > tracing_on : quick way to disable tracing\n"
3512 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3513 " Important files:\n"
3514 " trace\t\t\t- The static contents of the buffer\n"
3515 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3516 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3517 " current_tracer\t- function and latency tracers\n"
3518 " available_tracers\t- list of configured tracers for current_tracer\n"
3519 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3520 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3521 " trace_clock\t\t-change the clock used to order events\n"
3522 " local: Per cpu clock but may not be synced across CPUs\n"
3523 " global: Synced across CPUs but slows tracing down.\n"
3524 " counter: Not a clock, but just an increment\n"
3525 " uptime: Jiffy counter from time of boot\n"
3526 " perf: Same clock that perf events use\n"
3527 #ifdef CONFIG_X86_64
3528 " x86-tsc: TSC cycle counter\n"
3530 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3531 " tracing_cpumask\t- Limit which CPUs to trace\n"
3532 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3533 "\t\t\t Remove sub-buffer with rmdir\n"
3534 " trace_options\t\t- Set format or modify how tracing happens\n"
3535 "\t\t\t Disable an option by adding a suffix 'no' to the option name\n"
3536 #ifdef CONFIG_DYNAMIC_FTRACE
3537 "\n available_filter_functions - list of functions that can be filtered on\n"
3538 " set_ftrace_filter\t- echo function name in here to only trace these functions\n"
3539 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3540 " modules: Can select a group via module\n"
3541 " Format: :mod:<module-name>\n"
3542 " example: echo :mod:ext3 > set_ftrace_filter\n"
3543 " triggers: a command to perform when function is hit\n"
3544 " Format: <function>:<trigger>[:count]\n"
3545 " trigger: traceon, traceoff\n"
3546 " enable_event:<system>:<event>\n"
3547 " disable_event:<system>:<event>\n"
3548 #ifdef CONFIG_STACKTRACE
3551 #ifdef CONFIG_TRACER_SNAPSHOT
3554 " example: echo do_fault:traceoff > set_ftrace_filter\n"
3555 " echo do_trap:traceoff:3 > set_ftrace_filter\n"
3556 " The first one will disable tracing every time do_fault is hit\n"
3557 " The second will disable tracing at most 3 times when do_trap is hit\n"
3558 " The first time do trap is hit and it disables tracing, the counter\n"
3559 " will decrement to 2. If tracing is already disabled, the counter\n"
3560 " will not decrement. It only decrements when the trigger did work\n"
3561 " To remove trigger without count:\n"
3562 " echo '!<function>:<trigger> > set_ftrace_filter\n"
3563 " To remove trigger with a count:\n"
3564 " echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3565 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3566 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3567 " modules: Can select a group via module command :mod:\n"
3568 " Does not accept triggers\n"
3569 #endif /* CONFIG_DYNAMIC_FTRACE */
3570 #ifdef CONFIG_FUNCTION_TRACER
3571 " set_ftrace_pid\t- Write pid(s) to only function trace those pids (function)\n"
3573 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3574 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3575 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3577 #ifdef CONFIG_TRACER_SNAPSHOT
3578 "\n snapshot\t\t- Like 'trace' but shows the content of the static snapshot buffer\n"
3579 "\t\t\t Read the contents for more information\n"
3581 #ifdef CONFIG_STACKTRACE
3582 " stack_trace\t\t- Shows the max stack trace when active\n"
3583 " stack_max_size\t- Shows current max stack size that was traced\n"
3584 "\t\t\t Write into this file to reset the max size (trigger a new trace)\n"
3585 #ifdef CONFIG_DYNAMIC_FTRACE
3586 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace traces\n"
3588 #endif /* CONFIG_STACKTRACE */
3592 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
3593 size_t cnt
, loff_t
*ppos
)
3595 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
3596 readme_msg
, strlen(readme_msg
));
3599 static const struct file_operations tracing_readme_fops
= {
3600 .open
= tracing_open_generic
,
3601 .read
= tracing_readme_read
,
3602 .llseek
= generic_file_llseek
,
3606 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
3607 size_t cnt
, loff_t
*ppos
)
3616 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
3620 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
3628 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
3631 pid
= map_cmdline_to_pid
[i
];
3632 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
3635 trace_find_cmdline(pid
, buf_comm
);
3636 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
3641 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
3650 static const struct file_operations tracing_saved_cmdlines_fops
= {
3651 .open
= tracing_open_generic
,
3652 .read
= tracing_saved_cmdlines_read
,
3653 .llseek
= generic_file_llseek
,
3657 tracing_saved_tgids_read(struct file
*file
, char __user
*ubuf
,
3658 size_t cnt
, loff_t
*ppos
)
3666 file_buf
= kmalloc(SAVED_CMDLINES
*(16+1+16), GFP_KERNEL
);
3672 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
3676 pid
= map_cmdline_to_pid
[i
];
3677 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
3680 tgid
= trace_find_tgid(pid
);
3681 r
= sprintf(buf
, "%d %d\n", pid
, tgid
);
3686 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
3694 static const struct file_operations tracing_saved_tgids_fops
= {
3695 .open
= tracing_open_generic
,
3696 .read
= tracing_saved_tgids_read
,
3697 .llseek
= generic_file_llseek
,
3701 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
3702 size_t cnt
, loff_t
*ppos
)
3704 struct trace_array
*tr
= filp
->private_data
;
3705 char buf
[MAX_TRACER_SIZE
+2];
3708 mutex_lock(&trace_types_lock
);
3709 r
= sprintf(buf
, "%s\n", tr
->current_trace
->name
);
3710 mutex_unlock(&trace_types_lock
);
3712 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3715 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
3717 tracing_reset_online_cpus(&tr
->trace_buffer
);
3721 static void set_buffer_entries(struct trace_buffer
*buf
, unsigned long val
)
3725 for_each_tracing_cpu(cpu
)
3726 per_cpu_ptr(buf
->data
, cpu
)->entries
= val
;
3729 #ifdef CONFIG_TRACER_MAX_TRACE
3730 /* resize @tr's buffer to the size of @size_tr's entries */
3731 static int resize_buffer_duplicate_size(struct trace_buffer
*trace_buf
,
3732 struct trace_buffer
*size_buf
, int cpu_id
)
3736 if (cpu_id
== RING_BUFFER_ALL_CPUS
) {
3737 for_each_tracing_cpu(cpu
) {
3738 ret
= ring_buffer_resize(trace_buf
->buffer
,
3739 per_cpu_ptr(size_buf
->data
, cpu
)->entries
, cpu
);
3742 per_cpu_ptr(trace_buf
->data
, cpu
)->entries
=
3743 per_cpu_ptr(size_buf
->data
, cpu
)->entries
;
3746 ret
= ring_buffer_resize(trace_buf
->buffer
,
3747 per_cpu_ptr(size_buf
->data
, cpu_id
)->entries
, cpu_id
);
3749 per_cpu_ptr(trace_buf
->data
, cpu_id
)->entries
=
3750 per_cpu_ptr(size_buf
->data
, cpu_id
)->entries
;
3755 #endif /* CONFIG_TRACER_MAX_TRACE */
3757 static int __tracing_resize_ring_buffer(struct trace_array
*tr
,
3758 unsigned long size
, int cpu
)
3763 * If kernel or user changes the size of the ring buffer
3764 * we use the size that was given, and we can forget about
3765 * expanding it later.
3767 ring_buffer_expanded
= true;
3769 /* May be called before buffers are initialized */
3770 if (!tr
->trace_buffer
.buffer
)
3773 ret
= ring_buffer_resize(tr
->trace_buffer
.buffer
, size
, cpu
);
3777 #ifdef CONFIG_TRACER_MAX_TRACE
3778 if (!(tr
->flags
& TRACE_ARRAY_FL_GLOBAL
) ||
3779 !tr
->current_trace
->use_max_tr
)
3782 ret
= ring_buffer_resize(tr
->max_buffer
.buffer
, size
, cpu
);
3784 int r
= resize_buffer_duplicate_size(&tr
->trace_buffer
,
3785 &tr
->trace_buffer
, cpu
);
3788 * AARGH! We are left with different
3789 * size max buffer!!!!
3790 * The max buffer is our "snapshot" buffer.
3791 * When a tracer needs a snapshot (one of the
3792 * latency tracers), it swaps the max buffer
3793 * with the saved snap shot. We succeeded to
3794 * update the size of the main buffer, but failed to
3795 * update the size of the max buffer. But when we tried
3796 * to reset the main buffer to the original size, we
3797 * failed there too. This is very unlikely to
3798 * happen, but if it does, warn and kill all
3802 tracing_disabled
= 1;
3807 if (cpu
== RING_BUFFER_ALL_CPUS
)
3808 set_buffer_entries(&tr
->max_buffer
, size
);
3810 per_cpu_ptr(tr
->max_buffer
.data
, cpu
)->entries
= size
;
3813 #endif /* CONFIG_TRACER_MAX_TRACE */
3815 if (cpu
== RING_BUFFER_ALL_CPUS
)
3816 set_buffer_entries(&tr
->trace_buffer
, size
);
3818 per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->entries
= size
;
3823 static ssize_t
tracing_resize_ring_buffer(struct trace_array
*tr
,
3824 unsigned long size
, int cpu_id
)
3828 mutex_lock(&trace_types_lock
);
3830 if (cpu_id
!= RING_BUFFER_ALL_CPUS
) {
3831 /* make sure, this cpu is enabled in the mask */
3832 if (!cpumask_test_cpu(cpu_id
, tracing_buffer_mask
)) {
3838 ret
= __tracing_resize_ring_buffer(tr
, size
, cpu_id
);
3843 mutex_unlock(&trace_types_lock
);
3850 * tracing_update_buffers - used by tracing facility to expand ring buffers
3852 * To save on memory when the tracing is never used on a system with it
3853 * configured in. The ring buffers are set to a minimum size. But once
3854 * a user starts to use the tracing facility, then they need to grow
3855 * to their default size.
3857 * This function is to be called when a tracer is about to be used.
3859 int tracing_update_buffers(void)
3863 mutex_lock(&trace_types_lock
);
3864 if (!ring_buffer_expanded
)
3865 ret
= __tracing_resize_ring_buffer(&global_trace
, trace_buf_size
,
3866 RING_BUFFER_ALL_CPUS
);
3867 mutex_unlock(&trace_types_lock
);
3872 struct trace_option_dentry
;
3874 static struct trace_option_dentry
*
3875 create_trace_option_files(struct trace_array
*tr
, struct tracer
*tracer
);
3878 destroy_trace_option_files(struct trace_option_dentry
*topts
);
3880 static int tracing_set_tracer(const char *buf
)
3882 static struct trace_option_dentry
*topts
;
3883 struct trace_array
*tr
= &global_trace
;
3885 #ifdef CONFIG_TRACER_MAX_TRACE
3890 mutex_lock(&trace_types_lock
);
3892 if (!ring_buffer_expanded
) {
3893 ret
= __tracing_resize_ring_buffer(tr
, trace_buf_size
,
3894 RING_BUFFER_ALL_CPUS
);
3900 for (t
= trace_types
; t
; t
= t
->next
) {
3901 if (strcmp(t
->name
, buf
) == 0)
3908 if (t
== tr
->current_trace
)
3911 trace_branch_disable();
3913 tr
->current_trace
->enabled
= false;
3915 if (tr
->current_trace
->reset
)
3916 tr
->current_trace
->reset(tr
);
3918 /* Current trace needs to be nop_trace before synchronize_sched */
3919 tr
->current_trace
= &nop_trace
;
3921 #ifdef CONFIG_TRACER_MAX_TRACE
3922 had_max_tr
= tr
->allocated_snapshot
;
3924 if (had_max_tr
&& !t
->use_max_tr
) {
3926 * We need to make sure that the update_max_tr sees that
3927 * current_trace changed to nop_trace to keep it from
3928 * swapping the buffers after we resize it.
3929 * The update_max_tr is called from interrupts disabled
3930 * so a synchronized_sched() is sufficient.
3932 synchronize_sched();
3936 destroy_trace_option_files(topts
);
3938 topts
= create_trace_option_files(tr
, t
);
3940 #ifdef CONFIG_TRACER_MAX_TRACE
3941 if (t
->use_max_tr
&& !had_max_tr
) {
3942 ret
= alloc_snapshot(tr
);
3949 ret
= tracer_init(t
, tr
);
3954 tr
->current_trace
= t
;
3955 tr
->current_trace
->enabled
= true;
3956 trace_branch_enable(tr
);
3958 mutex_unlock(&trace_types_lock
);
3964 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
3965 size_t cnt
, loff_t
*ppos
)
3967 char buf
[MAX_TRACER_SIZE
+1];
3974 if (cnt
> MAX_TRACER_SIZE
)
3975 cnt
= MAX_TRACER_SIZE
;
3977 if (copy_from_user(&buf
, ubuf
, cnt
))
3982 /* strip ending whitespace. */
3983 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
3986 err
= tracing_set_tracer(buf
);
3996 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
3997 size_t cnt
, loff_t
*ppos
)
3999 unsigned long *ptr
= filp
->private_data
;
4003 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
4004 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
4005 if (r
> sizeof(buf
))
4007 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4011 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
4012 size_t cnt
, loff_t
*ppos
)
4014 unsigned long *ptr
= filp
->private_data
;
4018 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4027 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
4029 struct trace_array
*tr
= inode
->i_private
;
4030 struct trace_iterator
*iter
;
4033 if (tracing_disabled
)
4036 if (trace_array_get(tr
) < 0)
4039 mutex_lock(&trace_types_lock
);
4041 /* create a buffer to store the information to pass to userspace */
4042 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
4045 __trace_array_put(tr
);
4050 * We make a copy of the current tracer to avoid concurrent
4051 * changes on it while we are reading.
4053 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
4058 *iter
->trace
= *tr
->current_trace
;
4060 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
4065 /* trace pipe does not show start of buffer */
4066 cpumask_setall(iter
->started
);
4068 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
4069 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
4071 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4072 if (trace_clocks
[tr
->clock_id
].in_ns
)
4073 iter
->iter_flags
|= TRACE_FILE_TIME_IN_NS
;
4076 iter
->trace_buffer
= &tr
->trace_buffer
;
4077 iter
->cpu_file
= tracing_get_cpu(inode
);
4078 mutex_init(&iter
->mutex
);
4079 filp
->private_data
= iter
;
4081 if (iter
->trace
->pipe_open
)
4082 iter
->trace
->pipe_open(iter
);
4084 nonseekable_open(inode
, filp
);
4086 mutex_unlock(&trace_types_lock
);
4092 __trace_array_put(tr
);
4093 mutex_unlock(&trace_types_lock
);
4097 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
4099 struct trace_iterator
*iter
= file
->private_data
;
4100 struct trace_array
*tr
= inode
->i_private
;
4102 mutex_lock(&trace_types_lock
);
4104 if (iter
->trace
->pipe_close
)
4105 iter
->trace
->pipe_close(iter
);
4107 mutex_unlock(&trace_types_lock
);
4109 free_cpumask_var(iter
->started
);
4110 mutex_destroy(&iter
->mutex
);
4114 trace_array_put(tr
);
4120 trace_poll(struct trace_iterator
*iter
, struct file
*filp
, poll_table
*poll_table
)
4122 /* Iterators are static, they should be filled or empty */
4123 if (trace_buffer_iter(iter
, iter
->cpu_file
))
4124 return POLLIN
| POLLRDNORM
;
4126 if (trace_flags
& TRACE_ITER_BLOCK
)
4128 * Always select as readable when in blocking mode
4130 return POLLIN
| POLLRDNORM
;
4132 return ring_buffer_poll_wait(iter
->trace_buffer
->buffer
, iter
->cpu_file
,
4137 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
4139 struct trace_iterator
*iter
= filp
->private_data
;
4141 return trace_poll(iter
, filp
, poll_table
);
4145 * This is a make-shift waitqueue.
4146 * A tracer might use this callback on some rare cases:
4148 * 1) the current tracer might hold the runqueue lock when it wakes up
4149 * a reader, hence a deadlock (sched, function, and function graph tracers)
4150 * 2) the function tracers, trace all functions, we don't want
4151 * the overhead of calling wake_up and friends
4152 * (and tracing them too)
4154 * Anyway, this is really very primitive wakeup.
4156 int poll_wait_pipe(struct trace_iterator
*iter
)
4158 set_current_state(TASK_INTERRUPTIBLE
);
4159 /* sleep for 100 msecs, and try again. */
4160 schedule_timeout(HZ
/ 10);
4164 /* Must be called with trace_types_lock mutex held. */
4165 static int tracing_wait_pipe(struct file
*filp
)
4167 struct trace_iterator
*iter
= filp
->private_data
;
4170 while (trace_empty(iter
)) {
4172 if ((filp
->f_flags
& O_NONBLOCK
)) {
4176 mutex_unlock(&iter
->mutex
);
4178 ret
= iter
->trace
->wait_pipe(iter
);
4180 mutex_lock(&iter
->mutex
);
4185 if (signal_pending(current
))
4189 * We block until we read something and tracing is disabled.
4190 * We still block if tracing is disabled, but we have never
4191 * read anything. This allows a user to cat this file, and
4192 * then enable tracing. But after we have read something,
4193 * we give an EOF when tracing is again disabled.
4195 * iter->pos will be 0 if we haven't read anything.
4197 if (!tracing_is_on() && iter
->pos
)
4208 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
4209 size_t cnt
, loff_t
*ppos
)
4211 struct trace_iterator
*iter
= filp
->private_data
;
4212 struct trace_array
*tr
= iter
->tr
;
4215 /* copy the tracer to avoid using a global lock all around */
4216 mutex_lock(&trace_types_lock
);
4217 if (unlikely(iter
->trace
->name
!= tr
->current_trace
->name
))
4218 *iter
->trace
= *tr
->current_trace
;
4219 mutex_unlock(&trace_types_lock
);
4222 * Avoid more than one consumer on a single file descriptor
4223 * This is just a matter of traces coherency, the ring buffer itself
4226 mutex_lock(&iter
->mutex
);
4228 /* return any leftover data */
4229 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
4233 trace_seq_init(&iter
->seq
);
4235 if (iter
->trace
->read
) {
4236 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
4242 sret
= tracing_wait_pipe(filp
);
4246 /* stop when tracing is finished */
4247 if (trace_empty(iter
)) {
4252 if (cnt
>= PAGE_SIZE
)
4253 cnt
= PAGE_SIZE
- 1;
4255 /* reset all but tr, trace, and overruns */
4256 memset(&iter
->seq
, 0,
4257 sizeof(struct trace_iterator
) -
4258 offsetof(struct trace_iterator
, seq
));
4259 cpumask_clear(iter
->started
);
4262 trace_event_read_lock();
4263 trace_access_lock(iter
->cpu_file
);
4264 while (trace_find_next_entry_inc(iter
) != NULL
) {
4265 enum print_line_t ret
;
4266 int len
= iter
->seq
.len
;
4268 ret
= print_trace_line(iter
);
4269 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
4270 /* don't print partial lines */
4271 iter
->seq
.len
= len
;
4274 if (ret
!= TRACE_TYPE_NO_CONSUME
)
4275 trace_consume(iter
);
4277 if (iter
->seq
.len
>= cnt
)
4281 * Setting the full flag means we reached the trace_seq buffer
4282 * size and we should leave by partial output condition above.
4283 * One of the trace_seq_* functions is not used properly.
4285 WARN_ONCE(iter
->seq
.full
, "full flag set for trace type %d",
4288 trace_access_unlock(iter
->cpu_file
);
4289 trace_event_read_unlock();
4291 /* Now copy what we have to the user */
4292 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
4293 if (iter
->seq
.readpos
>= iter
->seq
.len
)
4294 trace_seq_init(&iter
->seq
);
4297 * If there was nothing to send to user, in spite of consuming trace
4298 * entries, go back to wait for more entries.
4304 mutex_unlock(&iter
->mutex
);
4309 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
4310 struct pipe_buffer
*buf
)
4312 __free_page(buf
->page
);
4315 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
4318 __free_page(spd
->pages
[idx
]);
4321 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
4323 .map
= generic_pipe_buf_map
,
4324 .unmap
= generic_pipe_buf_unmap
,
4325 .confirm
= generic_pipe_buf_confirm
,
4326 .release
= tracing_pipe_buf_release
,
4327 .steal
= generic_pipe_buf_steal
,
4328 .get
= generic_pipe_buf_get
,
4332 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
4337 /* Seq buffer is page-sized, exactly what we need. */
4339 count
= iter
->seq
.len
;
4340 ret
= print_trace_line(iter
);
4341 count
= iter
->seq
.len
- count
;
4344 iter
->seq
.len
-= count
;
4347 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
4348 iter
->seq
.len
-= count
;
4352 if (ret
!= TRACE_TYPE_NO_CONSUME
)
4353 trace_consume(iter
);
4355 if (!trace_find_next_entry_inc(iter
)) {
4365 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
4367 struct pipe_inode_info
*pipe
,
4371 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
4372 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
4373 struct trace_iterator
*iter
= filp
->private_data
;
4374 struct splice_pipe_desc spd
= {
4376 .partial
= partial_def
,
4377 .nr_pages
= 0, /* This gets updated below. */
4378 .nr_pages_max
= PIPE_DEF_BUFFERS
,
4380 .ops
= &tracing_pipe_buf_ops
,
4381 .spd_release
= tracing_spd_release_pipe
,
4383 struct trace_array
*tr
= iter
->tr
;
4388 if (splice_grow_spd(pipe
, &spd
))
4391 /* copy the tracer to avoid using a global lock all around */
4392 mutex_lock(&trace_types_lock
);
4393 if (unlikely(iter
->trace
->name
!= tr
->current_trace
->name
))
4394 *iter
->trace
= *tr
->current_trace
;
4395 mutex_unlock(&trace_types_lock
);
4397 mutex_lock(&iter
->mutex
);
4399 if (iter
->trace
->splice_read
) {
4400 ret
= iter
->trace
->splice_read(iter
, filp
,
4401 ppos
, pipe
, len
, flags
);
4406 ret
= tracing_wait_pipe(filp
);
4410 if (!iter
->ent
&& !trace_find_next_entry_inc(iter
)) {
4415 trace_event_read_lock();
4416 trace_access_lock(iter
->cpu_file
);
4418 /* Fill as many pages as possible. */
4419 for (i
= 0, rem
= len
; i
< pipe
->buffers
&& rem
; i
++) {
4420 spd
.pages
[i
] = alloc_page(GFP_KERNEL
);
4424 rem
= tracing_fill_pipe_page(rem
, iter
);
4426 /* Copy the data into the page, so we can start over. */
4427 ret
= trace_seq_to_buffer(&iter
->seq
,
4428 page_address(spd
.pages
[i
]),
4431 __free_page(spd
.pages
[i
]);
4434 spd
.partial
[i
].offset
= 0;
4435 spd
.partial
[i
].len
= iter
->seq
.len
;
4437 trace_seq_init(&iter
->seq
);
4440 trace_access_unlock(iter
->cpu_file
);
4441 trace_event_read_unlock();
4442 mutex_unlock(&iter
->mutex
);
4447 ret
= splice_to_pipe(pipe
, &spd
);
4451 splice_shrink_spd(&spd
);
4455 mutex_unlock(&iter
->mutex
);
4460 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
4461 size_t cnt
, loff_t
*ppos
)
4463 struct inode
*inode
= file_inode(filp
);
4464 struct trace_array
*tr
= inode
->i_private
;
4465 int cpu
= tracing_get_cpu(inode
);
4470 mutex_lock(&trace_types_lock
);
4472 if (cpu
== RING_BUFFER_ALL_CPUS
) {
4473 int cpu
, buf_size_same
;
4478 /* check if all cpu sizes are same */
4479 for_each_tracing_cpu(cpu
) {
4480 /* fill in the size from first enabled cpu */
4482 size
= per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->entries
;
4483 if (size
!= per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->entries
) {
4489 if (buf_size_same
) {
4490 if (!ring_buffer_expanded
)
4491 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
4493 trace_buf_size
>> 10);
4495 r
= sprintf(buf
, "%lu\n", size
>> 10);
4497 r
= sprintf(buf
, "X\n");
4499 r
= sprintf(buf
, "%lu\n", per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->entries
>> 10);
4501 mutex_unlock(&trace_types_lock
);
4503 ret
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4508 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
4509 size_t cnt
, loff_t
*ppos
)
4511 struct inode
*inode
= file_inode(filp
);
4512 struct trace_array
*tr
= inode
->i_private
;
4516 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4520 /* must have at least 1 entry */
4524 /* value is in KB */
4526 ret
= tracing_resize_ring_buffer(tr
, val
, tracing_get_cpu(inode
));
4536 tracing_total_entries_read(struct file
*filp
, char __user
*ubuf
,
4537 size_t cnt
, loff_t
*ppos
)
4539 struct trace_array
*tr
= filp
->private_data
;
4542 unsigned long size
= 0, expanded_size
= 0;
4544 mutex_lock(&trace_types_lock
);
4545 for_each_tracing_cpu(cpu
) {
4546 size
+= per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->entries
>> 10;
4547 if (!ring_buffer_expanded
)
4548 expanded_size
+= trace_buf_size
>> 10;
4550 if (ring_buffer_expanded
)
4551 r
= sprintf(buf
, "%lu\n", size
);
4553 r
= sprintf(buf
, "%lu (expanded: %lu)\n", size
, expanded_size
);
4554 mutex_unlock(&trace_types_lock
);
4556 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4560 tracing_free_buffer_write(struct file
*filp
, const char __user
*ubuf
,
4561 size_t cnt
, loff_t
*ppos
)
4564 * There is no need to read what the user has written, this function
4565 * is just to make sure that there is no error when "echo" is used
4574 tracing_free_buffer_release(struct inode
*inode
, struct file
*filp
)
4576 struct trace_array
*tr
= inode
->i_private
;
4578 /* disable tracing ? */
4579 if (trace_flags
& TRACE_ITER_STOP_ON_FREE
)
4580 tracer_tracing_off(tr
);
4581 /* resize the ring buffer to 0 */
4582 tracing_resize_ring_buffer(tr
, 0, RING_BUFFER_ALL_CPUS
);
4584 trace_array_put(tr
);
4590 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
4591 size_t cnt
, loff_t
*fpos
)
4593 unsigned long addr
= (unsigned long)ubuf
;
4594 struct trace_array
*tr
= filp
->private_data
;
4595 struct ring_buffer_event
*event
;
4596 struct ring_buffer
*buffer
;
4597 struct print_entry
*entry
;
4598 unsigned long irq_flags
;
4599 struct page
*pages
[2];
4609 if (tracing_disabled
)
4612 if (!(trace_flags
& TRACE_ITER_MARKERS
))
4615 if (cnt
> TRACE_BUF_SIZE
)
4616 cnt
= TRACE_BUF_SIZE
;
4619 * Userspace is injecting traces into the kernel trace buffer.
4620 * We want to be as non intrusive as possible.
4621 * To do so, we do not want to allocate any special buffers
4622 * or take any locks, but instead write the userspace data
4623 * straight into the ring buffer.
4625 * First we need to pin the userspace buffer into memory,
4626 * which, most likely it is, because it just referenced it.
4627 * But there's no guarantee that it is. By using get_user_pages_fast()
4628 * and kmap_atomic/kunmap_atomic() we can get access to the
4629 * pages directly. We then write the data directly into the
4632 BUILD_BUG_ON(TRACE_BUF_SIZE
>= PAGE_SIZE
);
4634 /* check if we cross pages */
4635 if ((addr
& PAGE_MASK
) != ((addr
+ cnt
) & PAGE_MASK
))
4638 offset
= addr
& (PAGE_SIZE
- 1);
4641 ret
= get_user_pages_fast(addr
, nr_pages
, 0, pages
);
4642 if (ret
< nr_pages
) {
4644 put_page(pages
[ret
]);
4649 for (i
= 0; i
< nr_pages
; i
++)
4650 map_page
[i
] = kmap_atomic(pages
[i
]);
4652 local_save_flags(irq_flags
);
4653 size
= sizeof(*entry
) + cnt
+ 2; /* possible \n added */
4654 buffer
= tr
->trace_buffer
.buffer
;
4655 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
4656 irq_flags
, preempt_count());
4658 /* Ring buffer disabled, return as if not open for write */
4663 entry
= ring_buffer_event_data(event
);
4664 entry
->ip
= _THIS_IP_
;
4666 if (nr_pages
== 2) {
4667 len
= PAGE_SIZE
- offset
;
4668 memcpy(&entry
->buf
, map_page
[0] + offset
, len
);
4669 memcpy(&entry
->buf
[len
], map_page
[1], cnt
- len
);
4671 memcpy(&entry
->buf
, map_page
[0] + offset
, cnt
);
4673 if (entry
->buf
[cnt
- 1] != '\n') {
4674 entry
->buf
[cnt
] = '\n';
4675 entry
->buf
[cnt
+ 1] = '\0';
4677 entry
->buf
[cnt
] = '\0';
4679 __buffer_unlock_commit(buffer
, event
);
4686 for (i
= nr_pages
- 1; i
>= 0; i
--) {
4687 kunmap_atomic(map_page
[i
]);
4694 static int tracing_clock_show(struct seq_file
*m
, void *v
)
4696 struct trace_array
*tr
= m
->private;
4699 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
4701 "%s%s%s%s", i
? " " : "",
4702 i
== tr
->clock_id
? "[" : "", trace_clocks
[i
].name
,
4703 i
== tr
->clock_id
? "]" : "");
4709 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
4710 size_t cnt
, loff_t
*fpos
)
4712 struct seq_file
*m
= filp
->private_data
;
4713 struct trace_array
*tr
= m
->private;
4715 const char *clockstr
;
4718 if (cnt
>= sizeof(buf
))
4721 if (copy_from_user(&buf
, ubuf
, cnt
))
4726 clockstr
= strstrip(buf
);
4728 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
4729 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
4732 if (i
== ARRAY_SIZE(trace_clocks
))
4735 mutex_lock(&trace_types_lock
);
4739 ring_buffer_set_clock(tr
->trace_buffer
.buffer
, trace_clocks
[i
].func
);
4742 * New clock may not be consistent with the previous clock.
4743 * Reset the buffer so that it doesn't have incomparable timestamps.
4745 tracing_reset_online_cpus(&tr
->trace_buffer
);
4747 #ifdef CONFIG_TRACER_MAX_TRACE
4748 if (tr
->flags
& TRACE_ARRAY_FL_GLOBAL
&& tr
->max_buffer
.buffer
)
4749 ring_buffer_set_clock(tr
->max_buffer
.buffer
, trace_clocks
[i
].func
);
4750 tracing_reset_online_cpus(&tr
->max_buffer
);
4753 mutex_unlock(&trace_types_lock
);
4760 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
4762 struct trace_array
*tr
= inode
->i_private
;
4765 if (tracing_disabled
)
4768 if (trace_array_get(tr
))
4771 ret
= single_open(file
, tracing_clock_show
, inode
->i_private
);
4773 trace_array_put(tr
);
4778 struct ftrace_buffer_info
{
4779 struct trace_iterator iter
;
4784 #ifdef CONFIG_TRACER_SNAPSHOT
4785 static int tracing_snapshot_open(struct inode
*inode
, struct file
*file
)
4787 struct trace_array
*tr
= inode
->i_private
;
4788 struct trace_iterator
*iter
;
4792 if (trace_array_get(tr
) < 0)
4795 if (file
->f_mode
& FMODE_READ
) {
4796 iter
= __tracing_open(inode
, file
, true);
4798 ret
= PTR_ERR(iter
);
4800 /* Writes still need the seq_file to hold the private data */
4802 m
= kzalloc(sizeof(*m
), GFP_KERNEL
);
4805 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
4813 iter
->trace_buffer
= &tr
->max_buffer
;
4814 iter
->cpu_file
= tracing_get_cpu(inode
);
4816 file
->private_data
= m
;
4820 trace_array_put(tr
);
4826 tracing_snapshot_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4829 struct seq_file
*m
= filp
->private_data
;
4830 struct trace_iterator
*iter
= m
->private;
4831 struct trace_array
*tr
= iter
->tr
;
4835 ret
= tracing_update_buffers();
4839 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4843 mutex_lock(&trace_types_lock
);
4845 if (tr
->current_trace
->use_max_tr
) {
4852 if (iter
->cpu_file
!= RING_BUFFER_ALL_CPUS
) {
4856 if (tr
->allocated_snapshot
)
4860 /* Only allow per-cpu swap if the ring buffer supports it */
4861 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4862 if (iter
->cpu_file
!= RING_BUFFER_ALL_CPUS
) {
4867 if (!tr
->allocated_snapshot
) {
4868 ret
= alloc_snapshot(tr
);
4872 local_irq_disable();
4873 /* Now, we're going to swap */
4874 if (iter
->cpu_file
== RING_BUFFER_ALL_CPUS
)
4875 update_max_tr(tr
, current
, smp_processor_id());
4877 update_max_tr_single(tr
, current
, iter
->cpu_file
);
4881 if (tr
->allocated_snapshot
) {
4882 if (iter
->cpu_file
== RING_BUFFER_ALL_CPUS
)
4883 tracing_reset_online_cpus(&tr
->max_buffer
);
4885 tracing_reset(&tr
->max_buffer
, iter
->cpu_file
);
4895 mutex_unlock(&trace_types_lock
);
4899 static int tracing_snapshot_release(struct inode
*inode
, struct file
*file
)
4901 struct seq_file
*m
= file
->private_data
;
4904 ret
= tracing_release(inode
, file
);
4906 if (file
->f_mode
& FMODE_READ
)
4909 /* If write only, the seq_file is just a stub */
4917 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
);
4918 static ssize_t
tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
4919 size_t count
, loff_t
*ppos
);
4920 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
);
4921 static ssize_t
tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
4922 struct pipe_inode_info
*pipe
, size_t len
, unsigned int flags
);
4924 static int snapshot_raw_open(struct inode
*inode
, struct file
*filp
)
4926 struct ftrace_buffer_info
*info
;
4929 ret
= tracing_buffers_open(inode
, filp
);
4933 info
= filp
->private_data
;
4935 if (info
->iter
.trace
->use_max_tr
) {
4936 tracing_buffers_release(inode
, filp
);
4940 info
->iter
.snapshot
= true;
4941 info
->iter
.trace_buffer
= &info
->iter
.tr
->max_buffer
;
4946 #endif /* CONFIG_TRACER_SNAPSHOT */
4949 static const struct file_operations tracing_max_lat_fops
= {
4950 .open
= tracing_open_generic
,
4951 .read
= tracing_max_lat_read
,
4952 .write
= tracing_max_lat_write
,
4953 .llseek
= generic_file_llseek
,
4956 static const struct file_operations set_tracer_fops
= {
4957 .open
= tracing_open_generic
,
4958 .read
= tracing_set_trace_read
,
4959 .write
= tracing_set_trace_write
,
4960 .llseek
= generic_file_llseek
,
4963 static const struct file_operations tracing_pipe_fops
= {
4964 .open
= tracing_open_pipe
,
4965 .poll
= tracing_poll_pipe
,
4966 .read
= tracing_read_pipe
,
4967 .splice_read
= tracing_splice_read_pipe
,
4968 .release
= tracing_release_pipe
,
4969 .llseek
= no_llseek
,
4972 static const struct file_operations tracing_entries_fops
= {
4973 .open
= tracing_open_generic_tr
,
4974 .read
= tracing_entries_read
,
4975 .write
= tracing_entries_write
,
4976 .llseek
= generic_file_llseek
,
4977 .release
= tracing_release_generic_tr
,
4980 static const struct file_operations tracing_total_entries_fops
= {
4981 .open
= tracing_open_generic_tr
,
4982 .read
= tracing_total_entries_read
,
4983 .llseek
= generic_file_llseek
,
4984 .release
= tracing_release_generic_tr
,
4987 static const struct file_operations tracing_free_buffer_fops
= {
4988 .open
= tracing_open_generic_tr
,
4989 .write
= tracing_free_buffer_write
,
4990 .release
= tracing_free_buffer_release
,
4993 static const struct file_operations tracing_mark_fops
= {
4994 .open
= tracing_open_generic_tr
,
4995 .write
= tracing_mark_write
,
4996 .llseek
= generic_file_llseek
,
4997 .release
= tracing_release_generic_tr
,
5000 static const struct file_operations trace_clock_fops
= {
5001 .open
= tracing_clock_open
,
5003 .llseek
= seq_lseek
,
5004 .release
= tracing_single_release_tr
,
5005 .write
= tracing_clock_write
,
5008 #ifdef CONFIG_TRACER_SNAPSHOT
5009 static const struct file_operations snapshot_fops
= {
5010 .open
= tracing_snapshot_open
,
5012 .write
= tracing_snapshot_write
,
5013 .llseek
= tracing_seek
,
5014 .release
= tracing_snapshot_release
,
5017 static const struct file_operations snapshot_raw_fops
= {
5018 .open
= snapshot_raw_open
,
5019 .read
= tracing_buffers_read
,
5020 .release
= tracing_buffers_release
,
5021 .splice_read
= tracing_buffers_splice_read
,
5022 .llseek
= no_llseek
,
5025 #endif /* CONFIG_TRACER_SNAPSHOT */
5027 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
5029 struct trace_array
*tr
= inode
->i_private
;
5030 struct ftrace_buffer_info
*info
;
5033 if (tracing_disabled
)
5036 if (trace_array_get(tr
) < 0)
5039 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
5041 trace_array_put(tr
);
5045 mutex_lock(&trace_types_lock
);
5048 info
->iter
.cpu_file
= tracing_get_cpu(inode
);
5049 info
->iter
.trace
= tr
->current_trace
;
5050 info
->iter
.trace_buffer
= &tr
->trace_buffer
;
5052 /* Force reading ring buffer for first read */
5053 info
->read
= (unsigned int)-1;
5055 filp
->private_data
= info
;
5057 mutex_unlock(&trace_types_lock
);
5059 ret
= nonseekable_open(inode
, filp
);
5061 trace_array_put(tr
);
5067 tracing_buffers_poll(struct file
*filp
, poll_table
*poll_table
)
5069 struct ftrace_buffer_info
*info
= filp
->private_data
;
5070 struct trace_iterator
*iter
= &info
->iter
;
5072 return trace_poll(iter
, filp
, poll_table
);
5076 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
5077 size_t count
, loff_t
*ppos
)
5079 struct ftrace_buffer_info
*info
= filp
->private_data
;
5080 struct trace_iterator
*iter
= &info
->iter
;
5087 mutex_lock(&trace_types_lock
);
5089 #ifdef CONFIG_TRACER_MAX_TRACE
5090 if (iter
->snapshot
&& iter
->tr
->current_trace
->use_max_tr
) {
5097 info
->spare
= ring_buffer_alloc_read_page(iter
->trace_buffer
->buffer
,
5103 /* Do we have previous read data to read? */
5104 if (info
->read
< PAGE_SIZE
)
5108 trace_access_lock(iter
->cpu_file
);
5109 ret
= ring_buffer_read_page(iter
->trace_buffer
->buffer
,
5113 trace_access_unlock(iter
->cpu_file
);
5116 if (trace_empty(iter
)) {
5117 if ((filp
->f_flags
& O_NONBLOCK
)) {
5121 mutex_unlock(&trace_types_lock
);
5122 ret
= iter
->trace
->wait_pipe(iter
);
5123 mutex_lock(&trace_types_lock
);
5128 if (signal_pending(current
)) {
5140 size
= PAGE_SIZE
- info
->read
;
5144 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
5155 mutex_unlock(&trace_types_lock
);
5160 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
5162 struct ftrace_buffer_info
*info
= file
->private_data
;
5163 struct trace_iterator
*iter
= &info
->iter
;
5165 mutex_lock(&trace_types_lock
);
5167 __trace_array_put(iter
->tr
);
5170 ring_buffer_free_read_page(iter
->trace_buffer
->buffer
, info
->spare
);
5173 mutex_unlock(&trace_types_lock
);
5179 struct ring_buffer
*buffer
;
5184 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
5185 struct pipe_buffer
*buf
)
5187 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
5192 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
5197 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
5198 struct pipe_buffer
*buf
)
5200 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
5205 /* Pipe buffer operations for a buffer. */
5206 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
5208 .map
= generic_pipe_buf_map
,
5209 .unmap
= generic_pipe_buf_unmap
,
5210 .confirm
= generic_pipe_buf_confirm
,
5211 .release
= buffer_pipe_buf_release
,
5212 .steal
= generic_pipe_buf_steal
,
5213 .get
= buffer_pipe_buf_get
,
5217 * Callback from splice_to_pipe(), if we need to release some pages
5218 * at the end of the spd in case we error'ed out in filling the pipe.
5220 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
5222 struct buffer_ref
*ref
=
5223 (struct buffer_ref
*)spd
->partial
[i
].private;
5228 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
5230 spd
->partial
[i
].private = 0;
5234 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
5235 struct pipe_inode_info
*pipe
, size_t len
,
5238 struct ftrace_buffer_info
*info
= file
->private_data
;
5239 struct trace_iterator
*iter
= &info
->iter
;
5240 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
5241 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
5242 struct splice_pipe_desc spd
= {
5244 .partial
= partial_def
,
5245 .nr_pages_max
= PIPE_DEF_BUFFERS
,
5247 .ops
= &buffer_pipe_buf_ops
,
5248 .spd_release
= buffer_spd_release
,
5250 struct buffer_ref
*ref
;
5251 int entries
, size
, i
;
5254 mutex_lock(&trace_types_lock
);
5256 #ifdef CONFIG_TRACER_MAX_TRACE
5257 if (iter
->snapshot
&& iter
->tr
->current_trace
->use_max_tr
) {
5263 if (*ppos
& (PAGE_SIZE
- 1)) {
5268 if (len
& (PAGE_SIZE
- 1)) {
5269 if (len
< PAGE_SIZE
) {
5276 if (splice_grow_spd(pipe
, &spd
)) {
5282 trace_access_lock(iter
->cpu_file
);
5283 entries
= ring_buffer_entries_cpu(iter
->trace_buffer
->buffer
, iter
->cpu_file
);
5285 for (i
= 0; i
< pipe
->buffers
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
5289 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
5294 ref
->buffer
= iter
->trace_buffer
->buffer
;
5295 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
, iter
->cpu_file
);
5301 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
5302 len
, iter
->cpu_file
, 1);
5304 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
5310 * zero out any left over data, this is going to
5313 size
= ring_buffer_page_len(ref
->page
);
5314 if (size
< PAGE_SIZE
)
5315 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
5317 page
= virt_to_page(ref
->page
);
5319 spd
.pages
[i
] = page
;
5320 spd
.partial
[i
].len
= PAGE_SIZE
;
5321 spd
.partial
[i
].offset
= 0;
5322 spd
.partial
[i
].private = (unsigned long)ref
;
5326 entries
= ring_buffer_entries_cpu(iter
->trace_buffer
->buffer
, iter
->cpu_file
);
5329 trace_access_unlock(iter
->cpu_file
);
5332 /* did we read anything? */
5333 if (!spd
.nr_pages
) {
5334 if ((file
->f_flags
& O_NONBLOCK
) || (flags
& SPLICE_F_NONBLOCK
)) {
5338 mutex_unlock(&trace_types_lock
);
5339 ret
= iter
->trace
->wait_pipe(iter
);
5340 mutex_lock(&trace_types_lock
);
5343 if (signal_pending(current
)) {
5350 ret
= splice_to_pipe(pipe
, &spd
);
5352 splice_shrink_spd(&spd
);
5354 mutex_unlock(&trace_types_lock
);
5359 static const struct file_operations tracing_buffers_fops
= {
5360 .open
= tracing_buffers_open
,
5361 .read
= tracing_buffers_read
,
5362 .poll
= tracing_buffers_poll
,
5363 .release
= tracing_buffers_release
,
5364 .splice_read
= tracing_buffers_splice_read
,
5365 .llseek
= no_llseek
,
5369 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
5370 size_t count
, loff_t
*ppos
)
5372 struct inode
*inode
= file_inode(filp
);
5373 struct trace_array
*tr
= inode
->i_private
;
5374 struct trace_buffer
*trace_buf
= &tr
->trace_buffer
;
5375 int cpu
= tracing_get_cpu(inode
);
5376 struct trace_seq
*s
;
5378 unsigned long long t
;
5379 unsigned long usec_rem
;
5381 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
5387 cnt
= ring_buffer_entries_cpu(trace_buf
->buffer
, cpu
);
5388 trace_seq_printf(s
, "entries: %ld\n", cnt
);
5390 cnt
= ring_buffer_overrun_cpu(trace_buf
->buffer
, cpu
);
5391 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
5393 cnt
= ring_buffer_commit_overrun_cpu(trace_buf
->buffer
, cpu
);
5394 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
5396 cnt
= ring_buffer_bytes_cpu(trace_buf
->buffer
, cpu
);
5397 trace_seq_printf(s
, "bytes: %ld\n", cnt
);
5399 if (trace_clocks
[tr
->clock_id
].in_ns
) {
5400 /* local or global for trace_clock */
5401 t
= ns2usecs(ring_buffer_oldest_event_ts(trace_buf
->buffer
, cpu
));
5402 usec_rem
= do_div(t
, USEC_PER_SEC
);
5403 trace_seq_printf(s
, "oldest event ts: %5llu.%06lu\n",
5406 t
= ns2usecs(ring_buffer_time_stamp(trace_buf
->buffer
, cpu
));
5407 usec_rem
= do_div(t
, USEC_PER_SEC
);
5408 trace_seq_printf(s
, "now ts: %5llu.%06lu\n", t
, usec_rem
);
5410 /* counter or tsc mode for trace_clock */
5411 trace_seq_printf(s
, "oldest event ts: %llu\n",
5412 ring_buffer_oldest_event_ts(trace_buf
->buffer
, cpu
));
5414 trace_seq_printf(s
, "now ts: %llu\n",
5415 ring_buffer_time_stamp(trace_buf
->buffer
, cpu
));
5418 cnt
= ring_buffer_dropped_events_cpu(trace_buf
->buffer
, cpu
);
5419 trace_seq_printf(s
, "dropped events: %ld\n", cnt
);
5421 cnt
= ring_buffer_read_events_cpu(trace_buf
->buffer
, cpu
);
5422 trace_seq_printf(s
, "read events: %ld\n", cnt
);
5424 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
5431 static const struct file_operations tracing_stats_fops
= {
5432 .open
= tracing_open_generic_tr
,
5433 .read
= tracing_stats_read
,
5434 .llseek
= generic_file_llseek
,
5435 .release
= tracing_release_generic_tr
,
5438 #ifdef CONFIG_DYNAMIC_FTRACE
5440 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
5446 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
5447 size_t cnt
, loff_t
*ppos
)
5449 static char ftrace_dyn_info_buffer
[1024];
5450 static DEFINE_MUTEX(dyn_info_mutex
);
5451 unsigned long *p
= filp
->private_data
;
5452 char *buf
= ftrace_dyn_info_buffer
;
5453 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
5456 mutex_lock(&dyn_info_mutex
);
5457 r
= sprintf(buf
, "%ld ", *p
);
5459 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
5462 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
5464 mutex_unlock(&dyn_info_mutex
);
5469 static const struct file_operations tracing_dyn_info_fops
= {
5470 .open
= tracing_open_generic
,
5471 .read
= tracing_read_dyn_info
,
5472 .llseek
= generic_file_llseek
,
5474 #endif /* CONFIG_DYNAMIC_FTRACE */
5476 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5478 ftrace_snapshot(unsigned long ip
, unsigned long parent_ip
, void **data
)
5484 ftrace_count_snapshot(unsigned long ip
, unsigned long parent_ip
, void **data
)
5486 unsigned long *count
= (long *)data
;
5498 ftrace_snapshot_print(struct seq_file
*m
, unsigned long ip
,
5499 struct ftrace_probe_ops
*ops
, void *data
)
5501 long count
= (long)data
;
5503 seq_printf(m
, "%ps:", (void *)ip
);
5505 seq_printf(m
, "snapshot");
5508 seq_printf(m
, ":unlimited\n");
5510 seq_printf(m
, ":count=%ld\n", count
);
5515 static struct ftrace_probe_ops snapshot_probe_ops
= {
5516 .func
= ftrace_snapshot
,
5517 .print
= ftrace_snapshot_print
,
5520 static struct ftrace_probe_ops snapshot_count_probe_ops
= {
5521 .func
= ftrace_count_snapshot
,
5522 .print
= ftrace_snapshot_print
,
5526 ftrace_trace_snapshot_callback(struct ftrace_hash
*hash
,
5527 char *glob
, char *cmd
, char *param
, int enable
)
5529 struct ftrace_probe_ops
*ops
;
5530 void *count
= (void *)-1;
5534 /* hash funcs only work with set_ftrace_filter */
5538 ops
= param
? &snapshot_count_probe_ops
: &snapshot_probe_ops
;
5540 if (glob
[0] == '!') {
5541 unregister_ftrace_function_probe_func(glob
+1, ops
);
5548 number
= strsep(¶m
, ":");
5550 if (!strlen(number
))
5554 * We use the callback data field (which is a pointer)
5557 ret
= kstrtoul(number
, 0, (unsigned long *)&count
);
5562 ret
= register_ftrace_function_probe(glob
, ops
, count
);
5565 alloc_snapshot(&global_trace
);
5567 return ret
< 0 ? ret
: 0;
5570 static struct ftrace_func_command ftrace_snapshot_cmd
= {
5572 .func
= ftrace_trace_snapshot_callback
,
5575 static int register_snapshot_cmd(void)
5577 return register_ftrace_command(&ftrace_snapshot_cmd
);
5580 static inline int register_snapshot_cmd(void) { return 0; }
5581 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5583 struct dentry
*tracing_init_dentry_tr(struct trace_array
*tr
)
5588 if (!debugfs_initialized())
5591 if (tr
->flags
& TRACE_ARRAY_FL_GLOBAL
)
5592 tr
->dir
= debugfs_create_dir("tracing", NULL
);
5595 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5600 struct dentry
*tracing_init_dentry(void)
5602 return tracing_init_dentry_tr(&global_trace
);
5605 static struct dentry
*tracing_dentry_percpu(struct trace_array
*tr
, int cpu
)
5607 struct dentry
*d_tracer
;
5610 return tr
->percpu_dir
;
5612 d_tracer
= tracing_init_dentry_tr(tr
);
5616 tr
->percpu_dir
= debugfs_create_dir("per_cpu", d_tracer
);
5618 WARN_ONCE(!tr
->percpu_dir
,
5619 "Could not create debugfs directory 'per_cpu/%d'\n", cpu
);
5621 return tr
->percpu_dir
;
5624 static struct dentry
*
5625 trace_create_cpu_file(const char *name
, umode_t mode
, struct dentry
*parent
,
5626 void *data
, long cpu
, const struct file_operations
*fops
)
5628 struct dentry
*ret
= trace_create_file(name
, mode
, parent
, data
, fops
);
5630 if (ret
) /* See tracing_get_cpu() */
5631 ret
->d_inode
->i_cdev
= (void *)(cpu
+ 1);
5636 tracing_init_debugfs_percpu(struct trace_array
*tr
, long cpu
)
5638 struct dentry
*d_percpu
= tracing_dentry_percpu(tr
, cpu
);
5639 struct dentry
*d_cpu
;
5640 char cpu_dir
[30]; /* 30 characters should be more than enough */
5645 snprintf(cpu_dir
, 30, "cpu%ld", cpu
);
5646 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
5648 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
5652 /* per cpu trace_pipe */
5653 trace_create_cpu_file("trace_pipe", 0444, d_cpu
,
5654 tr
, cpu
, &tracing_pipe_fops
);
5657 trace_create_cpu_file("trace", 0644, d_cpu
,
5658 tr
, cpu
, &tracing_fops
);
5660 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu
,
5661 tr
, cpu
, &tracing_buffers_fops
);
5663 trace_create_cpu_file("stats", 0444, d_cpu
,
5664 tr
, cpu
, &tracing_stats_fops
);
5666 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu
,
5667 tr
, cpu
, &tracing_entries_fops
);
5669 #ifdef CONFIG_TRACER_SNAPSHOT
5670 trace_create_cpu_file("snapshot", 0644, d_cpu
,
5671 tr
, cpu
, &snapshot_fops
);
5673 trace_create_cpu_file("snapshot_raw", 0444, d_cpu
,
5674 tr
, cpu
, &snapshot_raw_fops
);
5678 #ifdef CONFIG_FTRACE_SELFTEST
5679 /* Let selftest have access to static functions in this file */
5680 #include "trace_selftest.c"
5683 struct trace_option_dentry
{
5684 struct tracer_opt
*opt
;
5685 struct tracer_flags
*flags
;
5686 struct trace_array
*tr
;
5687 struct dentry
*entry
;
5691 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
5694 struct trace_option_dentry
*topt
= filp
->private_data
;
5697 if (topt
->flags
->val
& topt
->opt
->bit
)
5702 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
5706 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
5709 struct trace_option_dentry
*topt
= filp
->private_data
;
5713 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
5717 if (val
!= 0 && val
!= 1)
5720 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
5721 mutex_lock(&trace_types_lock
);
5722 ret
= __set_tracer_option(topt
->tr
->current_trace
, topt
->flags
,
5724 mutex_unlock(&trace_types_lock
);
5735 static const struct file_operations trace_options_fops
= {
5736 .open
= tracing_open_generic
,
5737 .read
= trace_options_read
,
5738 .write
= trace_options_write
,
5739 .llseek
= generic_file_llseek
,
5743 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
5746 long index
= (long)filp
->private_data
;
5749 if (trace_flags
& (1 << index
))
5754 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
5758 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
5761 struct trace_array
*tr
= &global_trace
;
5762 long index
= (long)filp
->private_data
;
5766 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
5770 if (val
!= 0 && val
!= 1)
5773 mutex_lock(&trace_types_lock
);
5774 ret
= set_tracer_flag(tr
, 1 << index
, val
);
5775 mutex_unlock(&trace_types_lock
);
5785 static const struct file_operations trace_options_core_fops
= {
5786 .open
= tracing_open_generic
,
5787 .read
= trace_options_core_read
,
5788 .write
= trace_options_core_write
,
5789 .llseek
= generic_file_llseek
,
5792 struct dentry
*trace_create_file(const char *name
,
5794 struct dentry
*parent
,
5796 const struct file_operations
*fops
)
5800 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
5802 pr_warning("Could not create debugfs '%s' entry\n", name
);
5808 static struct dentry
*trace_options_init_dentry(struct trace_array
*tr
)
5810 struct dentry
*d_tracer
;
5815 d_tracer
= tracing_init_dentry_tr(tr
);
5819 tr
->options
= debugfs_create_dir("options", d_tracer
);
5821 pr_warning("Could not create debugfs directory 'options'\n");
5829 create_trace_option_file(struct trace_array
*tr
,
5830 struct trace_option_dentry
*topt
,
5831 struct tracer_flags
*flags
,
5832 struct tracer_opt
*opt
)
5834 struct dentry
*t_options
;
5836 t_options
= trace_options_init_dentry(tr
);
5840 topt
->flags
= flags
;
5844 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
5845 &trace_options_fops
);
5849 static struct trace_option_dentry
*
5850 create_trace_option_files(struct trace_array
*tr
, struct tracer
*tracer
)
5852 struct trace_option_dentry
*topts
;
5853 struct tracer_flags
*flags
;
5854 struct tracer_opt
*opts
;
5860 flags
= tracer
->flags
;
5862 if (!flags
|| !flags
->opts
)
5867 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
5870 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
5874 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
5875 create_trace_option_file(tr
, &topts
[cnt
], flags
,
5882 destroy_trace_option_files(struct trace_option_dentry
*topts
)
5889 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
5890 if (topts
[cnt
].entry
)
5891 debugfs_remove(topts
[cnt
].entry
);
5897 static struct dentry
*
5898 create_trace_option_core_file(struct trace_array
*tr
,
5899 const char *option
, long index
)
5901 struct dentry
*t_options
;
5903 t_options
= trace_options_init_dentry(tr
);
5907 return trace_create_file(option
, 0644, t_options
, (void *)index
,
5908 &trace_options_core_fops
);
5911 static __init
void create_trace_options_dir(struct trace_array
*tr
)
5913 struct dentry
*t_options
;
5916 t_options
= trace_options_init_dentry(tr
);
5920 for (i
= 0; trace_options
[i
]; i
++)
5921 create_trace_option_core_file(tr
, trace_options
[i
], i
);
5925 rb_simple_read(struct file
*filp
, char __user
*ubuf
,
5926 size_t cnt
, loff_t
*ppos
)
5928 struct trace_array
*tr
= filp
->private_data
;
5932 r
= tracer_tracing_is_on(tr
);
5933 r
= sprintf(buf
, "%d\n", r
);
5935 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
5939 rb_simple_write(struct file
*filp
, const char __user
*ubuf
,
5940 size_t cnt
, loff_t
*ppos
)
5942 struct trace_array
*tr
= filp
->private_data
;
5943 struct ring_buffer
*buffer
= tr
->trace_buffer
.buffer
;
5947 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
5952 mutex_lock(&trace_types_lock
);
5954 tracer_tracing_on(tr
);
5955 if (tr
->current_trace
->start
)
5956 tr
->current_trace
->start(tr
);
5958 tracer_tracing_off(tr
);
5959 if (tr
->current_trace
->stop
)
5960 tr
->current_trace
->stop(tr
);
5962 mutex_unlock(&trace_types_lock
);
5970 static const struct file_operations rb_simple_fops
= {
5971 .open
= tracing_open_generic_tr
,
5972 .read
= rb_simple_read
,
5973 .write
= rb_simple_write
,
5974 .release
= tracing_release_generic_tr
,
5975 .llseek
= default_llseek
,
5978 struct dentry
*trace_instance_dir
;
5981 init_tracer_debugfs(struct trace_array
*tr
, struct dentry
*d_tracer
);
5983 static void init_trace_buffers(struct trace_array
*tr
, struct trace_buffer
*buf
)
5987 for_each_tracing_cpu(cpu
) {
5988 memset(per_cpu_ptr(buf
->data
, cpu
), 0, sizeof(struct trace_array_cpu
));
5989 per_cpu_ptr(buf
->data
, cpu
)->trace_cpu
.cpu
= cpu
;
5990 per_cpu_ptr(buf
->data
, cpu
)->trace_cpu
.tr
= tr
;
5995 allocate_trace_buffer(struct trace_array
*tr
, struct trace_buffer
*buf
, int size
)
5997 enum ring_buffer_flags rb_flags
;
5999 rb_flags
= trace_flags
& TRACE_ITER_OVERWRITE
? RB_FL_OVERWRITE
: 0;
6003 buf
->buffer
= ring_buffer_alloc(size
, rb_flags
);
6007 buf
->data
= alloc_percpu(struct trace_array_cpu
);
6009 ring_buffer_free(buf
->buffer
);
6013 init_trace_buffers(tr
, buf
);
6015 /* Allocate the first page for all buffers */
6016 set_buffer_entries(&tr
->trace_buffer
,
6017 ring_buffer_size(tr
->trace_buffer
.buffer
, 0));
6022 static int allocate_trace_buffers(struct trace_array
*tr
, int size
)
6026 ret
= allocate_trace_buffer(tr
, &tr
->trace_buffer
, size
);
6030 #ifdef CONFIG_TRACER_MAX_TRACE
6031 ret
= allocate_trace_buffer(tr
, &tr
->max_buffer
,
6032 allocate_snapshot
? size
: 1);
6034 ring_buffer_free(tr
->trace_buffer
.buffer
);
6035 free_percpu(tr
->trace_buffer
.data
);
6038 tr
->allocated_snapshot
= allocate_snapshot
;
6041 * Only the top level trace array gets its snapshot allocated
6042 * from the kernel command line.
6044 allocate_snapshot
= false;
6049 static int new_instance_create(const char *name
)
6051 struct trace_array
*tr
;
6054 mutex_lock(&trace_types_lock
);
6057 list_for_each_entry(tr
, &ftrace_trace_arrays
, list
) {
6058 if (tr
->name
&& strcmp(tr
->name
, name
) == 0)
6063 tr
= kzalloc(sizeof(*tr
), GFP_KERNEL
);
6067 tr
->name
= kstrdup(name
, GFP_KERNEL
);
6071 raw_spin_lock_init(&tr
->start_lock
);
6073 tr
->current_trace
= &nop_trace
;
6075 INIT_LIST_HEAD(&tr
->systems
);
6076 INIT_LIST_HEAD(&tr
->events
);
6078 if (allocate_trace_buffers(tr
, trace_buf_size
) < 0)
6081 /* Holder for file callbacks */
6082 tr
->trace_cpu
.cpu
= RING_BUFFER_ALL_CPUS
;
6083 tr
->trace_cpu
.tr
= tr
;
6085 tr
->dir
= debugfs_create_dir(name
, trace_instance_dir
);
6089 ret
= event_trace_add_tracer(tr
->dir
, tr
);
6091 debugfs_remove_recursive(tr
->dir
);
6095 init_tracer_debugfs(tr
, tr
->dir
);
6097 list_add(&tr
->list
, &ftrace_trace_arrays
);
6099 mutex_unlock(&trace_types_lock
);
6104 if (tr
->trace_buffer
.buffer
)
6105 ring_buffer_free(tr
->trace_buffer
.buffer
);
6110 mutex_unlock(&trace_types_lock
);
6116 static int instance_delete(const char *name
)
6118 struct trace_array
*tr
;
6122 mutex_lock(&trace_types_lock
);
6125 list_for_each_entry(tr
, &ftrace_trace_arrays
, list
) {
6126 if (tr
->name
&& strcmp(tr
->name
, name
) == 0) {
6138 list_del(&tr
->list
);
6140 event_trace_del_tracer(tr
);
6141 debugfs_remove_recursive(tr
->dir
);
6142 free_percpu(tr
->trace_buffer
.data
);
6143 ring_buffer_free(tr
->trace_buffer
.buffer
);
6151 mutex_unlock(&trace_types_lock
);
6156 static int instance_mkdir (struct inode
*inode
, struct dentry
*dentry
, umode_t mode
)
6158 struct dentry
*parent
;
6161 /* Paranoid: Make sure the parent is the "instances" directory */
6162 parent
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
6163 if (WARN_ON_ONCE(parent
!= trace_instance_dir
))
6167 * The inode mutex is locked, but debugfs_create_dir() will also
6168 * take the mutex. As the instances directory can not be destroyed
6169 * or changed in any other way, it is safe to unlock it, and
6170 * let the dentry try. If two users try to make the same dir at
6171 * the same time, then the new_instance_create() will determine the
6174 mutex_unlock(&inode
->i_mutex
);
6176 ret
= new_instance_create(dentry
->d_iname
);
6178 mutex_lock(&inode
->i_mutex
);
6183 static int instance_rmdir(struct inode
*inode
, struct dentry
*dentry
)
6185 struct dentry
*parent
;
6188 /* Paranoid: Make sure the parent is the "instances" directory */
6189 parent
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
6190 if (WARN_ON_ONCE(parent
!= trace_instance_dir
))
6193 /* The caller did a dget() on dentry */
6194 mutex_unlock(&dentry
->d_inode
->i_mutex
);
6197 * The inode mutex is locked, but debugfs_create_dir() will also
6198 * take the mutex. As the instances directory can not be destroyed
6199 * or changed in any other way, it is safe to unlock it, and
6200 * let the dentry try. If two users try to make the same dir at
6201 * the same time, then the instance_delete() will determine the
6204 mutex_unlock(&inode
->i_mutex
);
6206 ret
= instance_delete(dentry
->d_iname
);
6208 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_PARENT
);
6209 mutex_lock(&dentry
->d_inode
->i_mutex
);
6214 static const struct inode_operations instance_dir_inode_operations
= {
6215 .lookup
= simple_lookup
,
6216 .mkdir
= instance_mkdir
,
6217 .rmdir
= instance_rmdir
,
6220 static __init
void create_trace_instances(struct dentry
*d_tracer
)
6222 trace_instance_dir
= debugfs_create_dir("instances", d_tracer
);
6223 if (WARN_ON(!trace_instance_dir
))
6226 /* Hijack the dir inode operations, to allow mkdir */
6227 trace_instance_dir
->d_inode
->i_op
= &instance_dir_inode_operations
;
6231 init_tracer_debugfs(struct trace_array
*tr
, struct dentry
*d_tracer
)
6235 trace_create_file("trace_options", 0644, d_tracer
,
6236 tr
, &tracing_iter_fops
);
6238 trace_create_file("trace", 0644, d_tracer
,
6241 trace_create_file("trace_pipe", 0444, d_tracer
,
6242 tr
, &tracing_pipe_fops
);
6244 trace_create_file("buffer_size_kb", 0644, d_tracer
,
6245 tr
, &tracing_entries_fops
);
6247 trace_create_file("buffer_total_size_kb", 0444, d_tracer
,
6248 tr
, &tracing_total_entries_fops
);
6250 trace_create_file("free_buffer", 0644, d_tracer
,
6251 tr
, &tracing_free_buffer_fops
);
6253 trace_create_file("trace_marker", 0220, d_tracer
,
6254 tr
, &tracing_mark_fops
);
6256 trace_create_file("saved_tgids", 0444, d_tracer
,
6257 tr
, &tracing_saved_tgids_fops
);
6259 trace_create_file("trace_clock", 0644, d_tracer
, tr
,
6262 trace_create_file("tracing_on", 0644, d_tracer
,
6263 tr
, &rb_simple_fops
);
6265 #ifdef CONFIG_TRACER_SNAPSHOT
6266 trace_create_file("snapshot", 0644, d_tracer
,
6267 tr
, &snapshot_fops
);
6270 for_each_tracing_cpu(cpu
)
6271 tracing_init_debugfs_percpu(tr
, cpu
);
6275 static __init
int tracer_init_debugfs(void)
6277 struct dentry
*d_tracer
;
6279 trace_access_lock_init();
6281 d_tracer
= tracing_init_dentry();
6285 init_tracer_debugfs(&global_trace
, d_tracer
);
6287 trace_create_file("tracing_cpumask", 0644, d_tracer
,
6288 &global_trace
, &tracing_cpumask_fops
);
6290 trace_create_file("available_tracers", 0444, d_tracer
,
6291 &global_trace
, &show_traces_fops
);
6293 trace_create_file("current_tracer", 0644, d_tracer
,
6294 &global_trace
, &set_tracer_fops
);
6296 #ifdef CONFIG_TRACER_MAX_TRACE
6297 trace_create_file("tracing_max_latency", 0644, d_tracer
,
6298 &tracing_max_latency
, &tracing_max_lat_fops
);
6301 trace_create_file("tracing_thresh", 0644, d_tracer
,
6302 &tracing_thresh
, &tracing_max_lat_fops
);
6304 trace_create_file("README", 0444, d_tracer
,
6305 NULL
, &tracing_readme_fops
);
6307 trace_create_file("saved_cmdlines", 0444, d_tracer
,
6308 NULL
, &tracing_saved_cmdlines_fops
);
6310 #ifdef CONFIG_DYNAMIC_FTRACE
6311 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
6312 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
6315 create_trace_instances(d_tracer
);
6317 create_trace_options_dir(&global_trace
);
6322 static int trace_panic_handler(struct notifier_block
*this,
6323 unsigned long event
, void *unused
)
6325 if (ftrace_dump_on_oops
)
6326 ftrace_dump(ftrace_dump_on_oops
);
6330 static struct notifier_block trace_panic_notifier
= {
6331 .notifier_call
= trace_panic_handler
,
6333 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
6336 static int trace_die_handler(struct notifier_block
*self
,
6342 if (ftrace_dump_on_oops
)
6343 ftrace_dump(ftrace_dump_on_oops
);
6351 static struct notifier_block trace_die_notifier
= {
6352 .notifier_call
= trace_die_handler
,
6357 * printk is set to max of 1024, we really don't need it that big.
6358 * Nothing should be printing 1000 characters anyway.
6360 #define TRACE_MAX_PRINT 1000
6363 * Define here KERN_TRACE so that we have one place to modify
6364 * it if we decide to change what log level the ftrace dump
6367 #define KERN_TRACE KERN_EMERG
6370 trace_printk_seq(struct trace_seq
*s
)
6372 /* Probably should print a warning here. */
6373 if (s
->len
>= TRACE_MAX_PRINT
)
6374 s
->len
= TRACE_MAX_PRINT
;
6376 /* should be zero ended, but we are paranoid. */
6377 s
->buffer
[s
->len
] = 0;
6379 printk(KERN_TRACE
"%s", s
->buffer
);
6384 void trace_init_global_iter(struct trace_iterator
*iter
)
6386 iter
->tr
= &global_trace
;
6387 iter
->trace
= iter
->tr
->current_trace
;
6388 iter
->cpu_file
= RING_BUFFER_ALL_CPUS
;
6389 iter
->trace_buffer
= &global_trace
.trace_buffer
;
6392 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode
)
6394 /* use static because iter can be a bit big for the stack */
6395 static struct trace_iterator iter
;
6396 static atomic_t dump_running
;
6397 unsigned int old_userobj
;
6398 unsigned long flags
;
6401 /* Only allow one dump user at a time. */
6402 if (atomic_inc_return(&dump_running
) != 1) {
6403 atomic_dec(&dump_running
);
6408 * Always turn off tracing when we dump.
6409 * We don't need to show trace output of what happens
6410 * between multiple crashes.
6412 * If the user does a sysrq-z, then they can re-enable
6413 * tracing with echo 1 > tracing_on.
6417 local_irq_save(flags
);
6419 /* Simulate the iterator */
6420 trace_init_global_iter(&iter
);
6422 for_each_tracing_cpu(cpu
) {
6423 atomic_inc(&per_cpu_ptr(iter
.tr
->trace_buffer
.data
, cpu
)->disabled
);
6426 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
6428 /* don't look at user memory in panic mode */
6429 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
6431 switch (oops_dump_mode
) {
6433 iter
.cpu_file
= RING_BUFFER_ALL_CPUS
;
6436 iter
.cpu_file
= raw_smp_processor_id();
6441 printk(KERN_TRACE
"Bad dumping mode, switching to all CPUs dump\n");
6442 iter
.cpu_file
= RING_BUFFER_ALL_CPUS
;
6445 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
6447 /* Did function tracer already get disabled? */
6448 if (ftrace_is_dead()) {
6449 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6450 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6454 * We need to stop all tracing on all CPUS to read the
6455 * the next buffer. This is a bit expensive, but is
6456 * not done often. We fill all what we can read,
6457 * and then release the locks again.
6460 while (!trace_empty(&iter
)) {
6463 printk(KERN_TRACE
"---------------------------------\n");
6467 /* reset all but tr, trace, and overruns */
6468 memset(&iter
.seq
, 0,
6469 sizeof(struct trace_iterator
) -
6470 offsetof(struct trace_iterator
, seq
));
6471 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
6474 if (trace_find_next_entry_inc(&iter
) != NULL
) {
6477 ret
= print_trace_line(&iter
);
6478 if (ret
!= TRACE_TYPE_NO_CONSUME
)
6479 trace_consume(&iter
);
6481 touch_nmi_watchdog();
6483 trace_printk_seq(&iter
.seq
);
6487 printk(KERN_TRACE
" (ftrace buffer empty)\n");
6489 printk(KERN_TRACE
"---------------------------------\n");
6492 trace_flags
|= old_userobj
;
6494 for_each_tracing_cpu(cpu
) {
6495 atomic_dec(&per_cpu_ptr(iter
.trace_buffer
->data
, cpu
)->disabled
);
6497 atomic_dec(&dump_running
);
6498 local_irq_restore(flags
);
6500 EXPORT_SYMBOL_GPL(ftrace_dump
);
6502 __init
static int tracer_alloc_buffers(void)
6508 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
6511 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
6512 goto out_free_buffer_mask
;
6514 /* Only allocate trace_printk buffers if a trace_printk exists */
6515 if (__stop___trace_bprintk_fmt
!= __start___trace_bprintk_fmt
)
6516 /* Must be called before global_trace.buffer is allocated */
6517 trace_printk_init_buffers();
6519 /* To save memory, keep the ring buffer size to its minimum */
6520 if (ring_buffer_expanded
)
6521 ring_buf_size
= trace_buf_size
;
6525 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
6526 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
6528 raw_spin_lock_init(&global_trace
.start_lock
);
6530 /* TODO: make the number of buffers hot pluggable with CPUS */
6531 if (allocate_trace_buffers(&global_trace
, ring_buf_size
) < 0) {
6532 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
6534 goto out_free_cpumask
;
6537 if (global_trace
.buffer_disabled
)
6540 trace_init_cmdlines();
6543 * register_tracer() might reference current_trace, so it
6544 * needs to be set before we register anything. This is
6545 * just a bootstrap of current_trace anyway.
6547 global_trace
.current_trace
= &nop_trace
;
6549 register_tracer(&nop_trace
);
6551 /* All seems OK, enable tracing */
6552 tracing_disabled
= 0;
6554 atomic_notifier_chain_register(&panic_notifier_list
,
6555 &trace_panic_notifier
);
6557 register_die_notifier(&trace_die_notifier
);
6559 global_trace
.flags
= TRACE_ARRAY_FL_GLOBAL
;
6561 /* Holder for file callbacks */
6562 global_trace
.trace_cpu
.cpu
= RING_BUFFER_ALL_CPUS
;
6563 global_trace
.trace_cpu
.tr
= &global_trace
;
6565 INIT_LIST_HEAD(&global_trace
.systems
);
6566 INIT_LIST_HEAD(&global_trace
.events
);
6567 list_add(&global_trace
.list
, &ftrace_trace_arrays
);
6569 while (trace_boot_options
) {
6572 option
= strsep(&trace_boot_options
, ",");
6573 trace_set_options(&global_trace
, option
);
6576 register_snapshot_cmd();
6581 free_percpu(global_trace
.trace_buffer
.data
);
6582 #ifdef CONFIG_TRACER_MAX_TRACE
6583 free_percpu(global_trace
.max_buffer
.data
);
6585 free_cpumask_var(tracing_cpumask
);
6586 out_free_buffer_mask
:
6587 free_cpumask_var(tracing_buffer_mask
);
6592 __init
static int clear_boot_tracer(void)
6595 * The default tracer at boot buffer is an init section.
6596 * This function is called in lateinit. If we did not
6597 * find the boot tracer, then clear it out, to prevent
6598 * later registration from accessing the buffer that is
6599 * about to be freed.
6601 if (!default_bootup_tracer
)
6604 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
6605 default_bootup_tracer
);
6606 default_bootup_tracer
= NULL
;
6611 early_initcall(tracer_alloc_buffers
);
6612 fs_initcall(tracer_init_debugfs
);
6613 late_initcall(clear_boot_tracer
);