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
;
427 alloc
= sizeof(*entry
) + size
+ 2; /* possible \n added */
429 local_save_flags(irq_flags
);
430 buffer
= global_trace
.trace_buffer
.buffer
;
431 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, alloc
,
432 irq_flags
, preempt_count());
436 entry
= ring_buffer_event_data(event
);
439 memcpy(&entry
->buf
, str
, size
);
441 /* Add a newline if necessary */
442 if (entry
->buf
[size
- 1] != '\n') {
443 entry
->buf
[size
] = '\n';
444 entry
->buf
[size
+ 1] = '\0';
446 entry
->buf
[size
] = '\0';
448 __buffer_unlock_commit(buffer
, event
);
452 EXPORT_SYMBOL_GPL(__trace_puts
);
455 * __trace_bputs - write the pointer to a constant string into trace buffer
456 * @ip: The address of the caller
457 * @str: The constant string to write to the buffer to
459 int __trace_bputs(unsigned long ip
, const char *str
)
461 struct ring_buffer_event
*event
;
462 struct ring_buffer
*buffer
;
463 struct bputs_entry
*entry
;
464 unsigned long irq_flags
;
465 int size
= sizeof(struct bputs_entry
);
467 local_save_flags(irq_flags
);
468 buffer
= global_trace
.trace_buffer
.buffer
;
469 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPUTS
, size
,
470 irq_flags
, preempt_count());
474 entry
= ring_buffer_event_data(event
);
478 __buffer_unlock_commit(buffer
, event
);
482 EXPORT_SYMBOL_GPL(__trace_bputs
);
484 #ifdef CONFIG_TRACER_SNAPSHOT
486 * trace_snapshot - take a snapshot of the current buffer.
488 * This causes a swap between the snapshot buffer and the current live
489 * tracing buffer. You can use this to take snapshots of the live
490 * trace when some condition is triggered, but continue to trace.
492 * Note, make sure to allocate the snapshot with either
493 * a tracing_snapshot_alloc(), or by doing it manually
494 * with: echo 1 > /sys/kernel/debug/tracing/snapshot
496 * If the snapshot buffer is not allocated, it will stop tracing.
497 * Basically making a permanent snapshot.
499 void tracing_snapshot(void)
501 struct trace_array
*tr
= &global_trace
;
502 struct tracer
*tracer
= tr
->current_trace
;
506 internal_trace_puts("*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
507 internal_trace_puts("*** snapshot is being ignored ***\n");
511 if (!tr
->allocated_snapshot
) {
512 internal_trace_puts("*** SNAPSHOT NOT ALLOCATED ***\n");
513 internal_trace_puts("*** stopping trace here! ***\n");
518 /* Note, snapshot can not be used when the tracer uses it */
519 if (tracer
->use_max_tr
) {
520 internal_trace_puts("*** LATENCY TRACER ACTIVE ***\n");
521 internal_trace_puts("*** Can not use snapshot (sorry) ***\n");
525 local_irq_save(flags
);
526 update_max_tr(tr
, current
, smp_processor_id());
527 local_irq_restore(flags
);
529 EXPORT_SYMBOL_GPL(tracing_snapshot
);
531 static int resize_buffer_duplicate_size(struct trace_buffer
*trace_buf
,
532 struct trace_buffer
*size_buf
, int cpu_id
);
533 static void set_buffer_entries(struct trace_buffer
*buf
, unsigned long val
);
535 static int alloc_snapshot(struct trace_array
*tr
)
539 if (!tr
->allocated_snapshot
) {
541 /* allocate spare buffer */
542 ret
= resize_buffer_duplicate_size(&tr
->max_buffer
,
543 &tr
->trace_buffer
, RING_BUFFER_ALL_CPUS
);
547 tr
->allocated_snapshot
= true;
553 void free_snapshot(struct trace_array
*tr
)
556 * We don't free the ring buffer. instead, resize it because
557 * The max_tr ring buffer has some state (e.g. ring->clock) and
558 * we want preserve it.
560 ring_buffer_resize(tr
->max_buffer
.buffer
, 1, RING_BUFFER_ALL_CPUS
);
561 set_buffer_entries(&tr
->max_buffer
, 1);
562 tracing_reset_online_cpus(&tr
->max_buffer
);
563 tr
->allocated_snapshot
= false;
567 * trace_snapshot_alloc - allocate and take a snapshot of the current buffer.
569 * This is similar to trace_snapshot(), but it will allocate the
570 * snapshot buffer if it isn't already allocated. Use this only
571 * where it is safe to sleep, as the allocation may sleep.
573 * This causes a swap between the snapshot buffer and the current live
574 * tracing buffer. You can use this to take snapshots of the live
575 * trace when some condition is triggered, but continue to trace.
577 void tracing_snapshot_alloc(void)
579 struct trace_array
*tr
= &global_trace
;
582 ret
= alloc_snapshot(tr
);
583 if (WARN_ON(ret
< 0))
588 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc
);
590 void tracing_snapshot(void)
592 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
594 EXPORT_SYMBOL_GPL(tracing_snapshot
);
595 void tracing_snapshot_alloc(void)
600 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc
);
601 #endif /* CONFIG_TRACER_SNAPSHOT */
603 void tracer_tracing_off(struct trace_array
*tr
)
605 if (tr
->trace_buffer
.buffer
)
606 ring_buffer_record_off(tr
->trace_buffer
.buffer
);
608 * This flag is looked at when buffers haven't been allocated
609 * yet, or by some tracers (like irqsoff), that just want to
610 * know if the ring buffer has been disabled, but it can handle
611 * races of where it gets disabled but we still do a record.
612 * As the check is in the fast path of the tracers, it is more
613 * important to be fast than accurate.
615 tr
->buffer_disabled
= 1;
616 /* Make the flag seen by readers */
621 * tracing_off - turn off tracing buffers
623 * This function stops the tracing buffers from recording data.
624 * It does not disable any overhead the tracers themselves may
625 * be causing. This function simply causes all recording to
626 * the ring buffers to fail.
628 void tracing_off(void)
630 tracer_tracing_off(&global_trace
);
632 EXPORT_SYMBOL_GPL(tracing_off
);
635 * tracer_tracing_is_on - show real state of ring buffer enabled
636 * @tr : the trace array to know if ring buffer is enabled
638 * Shows real state of the ring buffer if it is enabled or not.
640 int tracer_tracing_is_on(struct trace_array
*tr
)
642 if (tr
->trace_buffer
.buffer
)
643 return ring_buffer_record_is_on(tr
->trace_buffer
.buffer
);
644 return !tr
->buffer_disabled
;
648 * tracing_is_on - show state of ring buffers enabled
650 int tracing_is_on(void)
652 return tracer_tracing_is_on(&global_trace
);
654 EXPORT_SYMBOL_GPL(tracing_is_on
);
656 static int __init
set_buf_size(char *str
)
658 unsigned long buf_size
;
662 buf_size
= memparse(str
, &str
);
663 /* nr_entries can not be zero */
666 trace_buf_size
= buf_size
;
669 __setup("trace_buf_size=", set_buf_size
);
671 static int __init
set_tracing_thresh(char *str
)
673 unsigned long threshold
;
678 ret
= kstrtoul(str
, 0, &threshold
);
681 tracing_thresh
= threshold
* 1000;
684 __setup("tracing_thresh=", set_tracing_thresh
);
686 unsigned long nsecs_to_usecs(unsigned long nsecs
)
691 /* These must match the bit postions in trace_iterator_flags */
692 static const char *trace_options
[] = {
725 int in_ns
; /* is this clock in nanoseconds? */
727 { trace_clock_local
, "local", 1 },
728 { trace_clock_global
, "global", 1 },
729 { trace_clock_counter
, "counter", 0 },
730 { trace_clock_jiffies
, "uptime", 1 },
731 { trace_clock
, "perf", 1 },
736 * trace_parser_get_init - gets the buffer for trace parser
738 int trace_parser_get_init(struct trace_parser
*parser
, int size
)
740 memset(parser
, 0, sizeof(*parser
));
742 parser
->buffer
= kmalloc(size
, GFP_KERNEL
);
751 * trace_parser_put - frees the buffer for trace parser
753 void trace_parser_put(struct trace_parser
*parser
)
755 kfree(parser
->buffer
);
759 * trace_get_user - reads the user input string separated by space
760 * (matched by isspace(ch))
762 * For each string found the 'struct trace_parser' is updated,
763 * and the function returns.
765 * Returns number of bytes read.
767 * See kernel/trace/trace.h for 'struct trace_parser' details.
769 int trace_get_user(struct trace_parser
*parser
, const char __user
*ubuf
,
770 size_t cnt
, loff_t
*ppos
)
777 trace_parser_clear(parser
);
779 ret
= get_user(ch
, ubuf
++);
787 * The parser is not finished with the last write,
788 * continue reading the user input without skipping spaces.
791 /* skip white space */
792 while (cnt
&& isspace(ch
)) {
793 ret
= get_user(ch
, ubuf
++);
800 /* only spaces were written */
810 /* read the non-space input */
811 while (cnt
&& !isspace(ch
)) {
812 if (parser
->idx
< parser
->size
- 1)
813 parser
->buffer
[parser
->idx
++] = ch
;
818 ret
= get_user(ch
, ubuf
++);
825 /* We either got finished input or we have to wait for another call. */
827 parser
->buffer
[parser
->idx
] = 0;
828 parser
->cont
= false;
831 parser
->buffer
[parser
->idx
++] = ch
;
841 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
849 if (s
->len
<= s
->readpos
)
852 len
= s
->len
- s
->readpos
;
855 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
865 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
869 if (s
->len
<= s
->readpos
)
872 len
= s
->len
- s
->readpos
;
875 memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
882 * ftrace_max_lock is used to protect the swapping of buffers
883 * when taking a max snapshot. The buffers themselves are
884 * protected by per_cpu spinlocks. But the action of the swap
885 * needs its own lock.
887 * This is defined as a arch_spinlock_t in order to help
888 * with performance when lockdep debugging is enabled.
890 * It is also used in other places outside the update_max_tr
891 * so it needs to be defined outside of the
892 * CONFIG_TRACER_MAX_TRACE.
894 static arch_spinlock_t ftrace_max_lock
=
895 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
897 unsigned long __read_mostly tracing_thresh
;
899 #ifdef CONFIG_TRACER_MAX_TRACE
900 unsigned long __read_mostly tracing_max_latency
;
903 * Copy the new maximum trace into the separate maximum-trace
904 * structure. (this way the maximum trace is permanently saved,
905 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
908 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
910 struct trace_buffer
*trace_buf
= &tr
->trace_buffer
;
911 struct trace_buffer
*max_buf
= &tr
->max_buffer
;
912 struct trace_array_cpu
*data
= per_cpu_ptr(trace_buf
->data
, cpu
);
913 struct trace_array_cpu
*max_data
= per_cpu_ptr(max_buf
->data
, cpu
);
916 max_buf
->time_start
= data
->preempt_timestamp
;
918 max_data
->saved_latency
= tracing_max_latency
;
919 max_data
->critical_start
= data
->critical_start
;
920 max_data
->critical_end
= data
->critical_end
;
922 memcpy(max_data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
923 max_data
->pid
= tsk
->pid
;
925 * If tsk == current, then use current_uid(), as that does not use
926 * RCU. The irq tracer can be called out of RCU scope.
929 max_data
->uid
= current_uid();
931 max_data
->uid
= task_uid(tsk
);
933 max_data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
934 max_data
->policy
= tsk
->policy
;
935 max_data
->rt_priority
= tsk
->rt_priority
;
937 /* record this tasks comm */
938 tracing_record_cmdline(tsk
);
942 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
944 * @tsk: the task with the latency
945 * @cpu: The cpu that initiated the trace.
947 * Flip the buffers between the @tr and the max_tr and record information
948 * about which task was the cause of this latency.
951 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
953 struct ring_buffer
*buf
;
958 WARN_ON_ONCE(!irqs_disabled());
960 if (!tr
->allocated_snapshot
) {
961 /* Only the nop tracer should hit this when disabling */
962 WARN_ON_ONCE(tr
->current_trace
!= &nop_trace
);
966 arch_spin_lock(&ftrace_max_lock
);
968 buf
= tr
->trace_buffer
.buffer
;
969 tr
->trace_buffer
.buffer
= tr
->max_buffer
.buffer
;
970 tr
->max_buffer
.buffer
= buf
;
972 __update_max_tr(tr
, tsk
, cpu
);
973 arch_spin_unlock(&ftrace_max_lock
);
977 * update_max_tr_single - only copy one trace over, and reset the rest
979 * @tsk - task with the latency
980 * @cpu - the cpu of the buffer to copy.
982 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
985 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
992 WARN_ON_ONCE(!irqs_disabled());
993 if (!tr
->allocated_snapshot
) {
994 /* Only the nop tracer should hit this when disabling */
995 WARN_ON_ONCE(tr
->current_trace
!= &nop_trace
);
999 arch_spin_lock(&ftrace_max_lock
);
1001 ret
= ring_buffer_swap_cpu(tr
->max_buffer
.buffer
, tr
->trace_buffer
.buffer
, cpu
);
1003 if (ret
== -EBUSY
) {
1005 * We failed to swap the buffer due to a commit taking
1006 * place on this CPU. We fail to record, but we reset
1007 * the max trace buffer (no one writes directly to it)
1008 * and flag that it failed.
1010 trace_array_printk_buf(tr
->max_buffer
.buffer
, _THIS_IP_
,
1011 "Failed to swap buffers due to commit in progress\n");
1014 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
&& ret
!= -EBUSY
);
1016 __update_max_tr(tr
, tsk
, cpu
);
1017 arch_spin_unlock(&ftrace_max_lock
);
1019 #endif /* CONFIG_TRACER_MAX_TRACE */
1021 static void default_wait_pipe(struct trace_iterator
*iter
)
1023 /* Iterators are static, they should be filled or empty */
1024 if (trace_buffer_iter(iter
, iter
->cpu_file
))
1027 ring_buffer_wait(iter
->trace_buffer
->buffer
, iter
->cpu_file
);
1030 #ifdef CONFIG_FTRACE_STARTUP_TEST
1031 static int run_tracer_selftest(struct tracer
*type
)
1033 struct trace_array
*tr
= &global_trace
;
1034 struct tracer
*saved_tracer
= tr
->current_trace
;
1037 if (!type
->selftest
|| tracing_selftest_disabled
)
1041 * Run a selftest on this tracer.
1042 * Here we reset the trace buffer, and set the current
1043 * tracer to be this tracer. The tracer can then run some
1044 * internal tracing to verify that everything is in order.
1045 * If we fail, we do not register this tracer.
1047 tracing_reset_online_cpus(&tr
->trace_buffer
);
1049 tr
->current_trace
= type
;
1051 #ifdef CONFIG_TRACER_MAX_TRACE
1052 if (type
->use_max_tr
) {
1053 /* If we expanded the buffers, make sure the max is expanded too */
1054 if (ring_buffer_expanded
)
1055 ring_buffer_resize(tr
->max_buffer
.buffer
, trace_buf_size
,
1056 RING_BUFFER_ALL_CPUS
);
1057 tr
->allocated_snapshot
= true;
1061 /* the test is responsible for initializing and enabling */
1062 pr_info("Testing tracer %s: ", type
->name
);
1063 ret
= type
->selftest(type
, tr
);
1064 /* the test is responsible for resetting too */
1065 tr
->current_trace
= saved_tracer
;
1067 printk(KERN_CONT
"FAILED!\n");
1068 /* Add the warning after printing 'FAILED' */
1072 /* Only reset on passing, to avoid touching corrupted buffers */
1073 tracing_reset_online_cpus(&tr
->trace_buffer
);
1075 #ifdef CONFIG_TRACER_MAX_TRACE
1076 if (type
->use_max_tr
) {
1077 tr
->allocated_snapshot
= false;
1079 /* Shrink the max buffer again */
1080 if (ring_buffer_expanded
)
1081 ring_buffer_resize(tr
->max_buffer
.buffer
, 1,
1082 RING_BUFFER_ALL_CPUS
);
1086 printk(KERN_CONT
"PASSED\n");
1090 static inline int run_tracer_selftest(struct tracer
*type
)
1094 #endif /* CONFIG_FTRACE_STARTUP_TEST */
1097 * register_tracer - register a tracer with the ftrace system.
1098 * @type - the plugin for the tracer
1100 * Register a new plugin tracer.
1102 int register_tracer(struct tracer
*type
)
1108 pr_info("Tracer must have a name\n");
1112 if (strlen(type
->name
) >= MAX_TRACER_SIZE
) {
1113 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE
);
1117 mutex_lock(&trace_types_lock
);
1119 tracing_selftest_running
= true;
1121 for (t
= trace_types
; t
; t
= t
->next
) {
1122 if (strcmp(type
->name
, t
->name
) == 0) {
1124 pr_info("Tracer %s already registered\n",
1131 if (!type
->set_flag
)
1132 type
->set_flag
= &dummy_set_flag
;
1134 type
->flags
= &dummy_tracer_flags
;
1136 if (!type
->flags
->opts
)
1137 type
->flags
->opts
= dummy_tracer_opt
;
1138 if (!type
->wait_pipe
)
1139 type
->wait_pipe
= default_wait_pipe
;
1141 ret
= run_tracer_selftest(type
);
1145 type
->next
= trace_types
;
1149 tracing_selftest_running
= false;
1150 mutex_unlock(&trace_types_lock
);
1152 if (ret
|| !default_bootup_tracer
)
1155 if (strncmp(default_bootup_tracer
, type
->name
, MAX_TRACER_SIZE
))
1158 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
1159 /* Do we want this tracer to start on bootup? */
1160 tracing_set_tracer(type
->name
);
1161 default_bootup_tracer
= NULL
;
1162 /* disable other selftests, since this will break it. */
1163 tracing_selftest_disabled
= true;
1164 #ifdef CONFIG_FTRACE_STARTUP_TEST
1165 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
1173 void tracing_reset(struct trace_buffer
*buf
, int cpu
)
1175 struct ring_buffer
*buffer
= buf
->buffer
;
1180 ring_buffer_record_disable(buffer
);
1182 /* Make sure all commits have finished */
1183 synchronize_sched();
1184 ring_buffer_reset_cpu(buffer
, cpu
);
1186 ring_buffer_record_enable(buffer
);
1189 void tracing_reset_online_cpus(struct trace_buffer
*buf
)
1191 struct ring_buffer
*buffer
= buf
->buffer
;
1197 ring_buffer_record_disable(buffer
);
1199 /* Make sure all commits have finished */
1200 synchronize_sched();
1202 buf
->time_start
= buffer_ftrace_now(buf
, buf
->cpu
);
1204 for_each_online_cpu(cpu
)
1205 ring_buffer_reset_cpu(buffer
, cpu
);
1207 ring_buffer_record_enable(buffer
);
1210 /* Must have trace_types_lock held */
1211 void tracing_reset_all_online_cpus(void)
1213 struct trace_array
*tr
;
1215 list_for_each_entry(tr
, &ftrace_trace_arrays
, list
) {
1216 tracing_reset_online_cpus(&tr
->trace_buffer
);
1217 #ifdef CONFIG_TRACER_MAX_TRACE
1218 tracing_reset_online_cpus(&tr
->max_buffer
);
1223 #define SAVED_CMDLINES 128
1224 #define NO_CMDLINE_MAP UINT_MAX
1225 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
1226 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
1227 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
1228 static int cmdline_idx
;
1229 static arch_spinlock_t trace_cmdline_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
1231 /* temporary disable recording */
1232 static atomic_t trace_record_cmdline_disabled __read_mostly
;
1234 static void trace_init_cmdlines(void)
1236 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
1237 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
1241 int is_tracing_stopped(void)
1243 return global_trace
.stop_count
;
1247 * ftrace_off_permanent - disable all ftrace code permanently
1249 * This should only be called when a serious anomally has
1250 * been detected. This will turn off the function tracing,
1251 * ring buffers, and other tracing utilites. It takes no
1252 * locks and can be called from any context.
1254 void ftrace_off_permanent(void)
1256 tracing_disabled
= 1;
1258 tracing_off_permanent();
1262 * tracing_start - quick start of the tracer
1264 * If tracing is enabled but was stopped by tracing_stop,
1265 * this will start the tracer back up.
1267 void tracing_start(void)
1269 struct ring_buffer
*buffer
;
1270 unsigned long flags
;
1272 if (tracing_disabled
)
1275 raw_spin_lock_irqsave(&global_trace
.start_lock
, flags
);
1276 if (--global_trace
.stop_count
) {
1277 if (global_trace
.stop_count
< 0) {
1278 /* Someone screwed up their debugging */
1280 global_trace
.stop_count
= 0;
1285 /* Prevent the buffers from switching */
1286 arch_spin_lock(&ftrace_max_lock
);
1288 buffer
= global_trace
.trace_buffer
.buffer
;
1290 ring_buffer_record_enable(buffer
);
1292 #ifdef CONFIG_TRACER_MAX_TRACE
1293 buffer
= global_trace
.max_buffer
.buffer
;
1295 ring_buffer_record_enable(buffer
);
1298 arch_spin_unlock(&ftrace_max_lock
);
1302 raw_spin_unlock_irqrestore(&global_trace
.start_lock
, flags
);
1305 static void tracing_start_tr(struct trace_array
*tr
)
1307 struct ring_buffer
*buffer
;
1308 unsigned long flags
;
1310 if (tracing_disabled
)
1313 /* If global, we need to also start the max tracer */
1314 if (tr
->flags
& TRACE_ARRAY_FL_GLOBAL
)
1315 return tracing_start();
1317 raw_spin_lock_irqsave(&tr
->start_lock
, flags
);
1319 if (--tr
->stop_count
) {
1320 if (tr
->stop_count
< 0) {
1321 /* Someone screwed up their debugging */
1328 buffer
= tr
->trace_buffer
.buffer
;
1330 ring_buffer_record_enable(buffer
);
1333 raw_spin_unlock_irqrestore(&tr
->start_lock
, flags
);
1337 * tracing_stop - quick stop of the tracer
1339 * Light weight way to stop tracing. Use in conjunction with
1342 void tracing_stop(void)
1344 struct ring_buffer
*buffer
;
1345 unsigned long flags
;
1348 raw_spin_lock_irqsave(&global_trace
.start_lock
, flags
);
1349 if (global_trace
.stop_count
++)
1352 /* Prevent the buffers from switching */
1353 arch_spin_lock(&ftrace_max_lock
);
1355 buffer
= global_trace
.trace_buffer
.buffer
;
1357 ring_buffer_record_disable(buffer
);
1359 #ifdef CONFIG_TRACER_MAX_TRACE
1360 buffer
= global_trace
.max_buffer
.buffer
;
1362 ring_buffer_record_disable(buffer
);
1365 arch_spin_unlock(&ftrace_max_lock
);
1368 raw_spin_unlock_irqrestore(&global_trace
.start_lock
, flags
);
1371 static void tracing_stop_tr(struct trace_array
*tr
)
1373 struct ring_buffer
*buffer
;
1374 unsigned long flags
;
1376 /* If global, we need to also stop the max tracer */
1377 if (tr
->flags
& TRACE_ARRAY_FL_GLOBAL
)
1378 return tracing_stop();
1380 raw_spin_lock_irqsave(&tr
->start_lock
, flags
);
1381 if (tr
->stop_count
++)
1384 buffer
= tr
->trace_buffer
.buffer
;
1386 ring_buffer_record_disable(buffer
);
1389 raw_spin_unlock_irqrestore(&tr
->start_lock
, flags
);
1392 void trace_stop_cmdline_recording(void);
1394 static void trace_save_cmdline(struct task_struct
*tsk
)
1398 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
1402 * It's not the end of the world if we don't get
1403 * the lock, but we also don't want to spin
1404 * nor do we want to disable interrupts,
1405 * so if we miss here, then better luck next time.
1407 if (!arch_spin_trylock(&trace_cmdline_lock
))
1410 idx
= map_pid_to_cmdline
[tsk
->pid
];
1411 if (idx
== NO_CMDLINE_MAP
) {
1412 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
1415 * Check whether the cmdline buffer at idx has a pid
1416 * mapped. We are going to overwrite that entry so we
1417 * need to clear the map_pid_to_cmdline. Otherwise we
1418 * would read the new comm for the old pid.
1420 pid
= map_cmdline_to_pid
[idx
];
1421 if (pid
!= NO_CMDLINE_MAP
)
1422 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
1424 map_cmdline_to_pid
[idx
] = tsk
->pid
;
1425 map_pid_to_cmdline
[tsk
->pid
] = idx
;
1430 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
1432 arch_spin_unlock(&trace_cmdline_lock
);
1435 void trace_find_cmdline(int pid
, char comm
[])
1440 strcpy(comm
, "<idle>");
1444 if (WARN_ON_ONCE(pid
< 0)) {
1445 strcpy(comm
, "<XXX>");
1449 if (pid
> PID_MAX_DEFAULT
) {
1450 strcpy(comm
, "<...>");
1455 arch_spin_lock(&trace_cmdline_lock
);
1456 map
= map_pid_to_cmdline
[pid
];
1457 if (map
!= NO_CMDLINE_MAP
)
1458 strcpy(comm
, saved_cmdlines
[map
]);
1460 strcpy(comm
, "<...>");
1462 arch_spin_unlock(&trace_cmdline_lock
);
1466 void tracing_record_cmdline(struct task_struct
*tsk
)
1468 if (atomic_read(&trace_record_cmdline_disabled
) || !tracing_is_on())
1471 if (!__this_cpu_read(trace_cmdline_save
))
1474 __this_cpu_write(trace_cmdline_save
, false);
1476 trace_save_cmdline(tsk
);
1480 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
1483 struct task_struct
*tsk
= current
;
1485 entry
->preempt_count
= pc
& 0xff;
1486 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
1488 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1489 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
1491 TRACE_FLAG_IRQS_NOSUPPORT
|
1493 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
1494 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
1495 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
1497 EXPORT_SYMBOL_GPL(tracing_generic_entry_update
);
1499 struct ring_buffer_event
*
1500 trace_buffer_lock_reserve(struct ring_buffer
*buffer
,
1503 unsigned long flags
, int pc
)
1505 struct ring_buffer_event
*event
;
1507 event
= ring_buffer_lock_reserve(buffer
, len
);
1508 if (event
!= NULL
) {
1509 struct trace_entry
*ent
= ring_buffer_event_data(event
);
1511 tracing_generic_entry_update(ent
, flags
, pc
);
1519 __buffer_unlock_commit(struct ring_buffer
*buffer
, struct ring_buffer_event
*event
)
1521 __this_cpu_write(trace_cmdline_save
, true);
1522 ring_buffer_unlock_commit(buffer
, event
);
1526 __trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1527 struct ring_buffer_event
*event
,
1528 unsigned long flags
, int pc
)
1530 __buffer_unlock_commit(buffer
, event
);
1532 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1533 ftrace_trace_userstack(buffer
, flags
, pc
);
1536 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1537 struct ring_buffer_event
*event
,
1538 unsigned long flags
, int pc
)
1540 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
);
1542 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit
);
1544 struct ring_buffer_event
*
1545 trace_event_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1546 struct ftrace_event_file
*ftrace_file
,
1547 int type
, unsigned long len
,
1548 unsigned long flags
, int pc
)
1550 *current_rb
= ftrace_file
->tr
->trace_buffer
.buffer
;
1551 return trace_buffer_lock_reserve(*current_rb
,
1552 type
, len
, flags
, pc
);
1554 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve
);
1556 struct ring_buffer_event
*
1557 trace_current_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1558 int type
, unsigned long len
,
1559 unsigned long flags
, int pc
)
1561 *current_rb
= global_trace
.trace_buffer
.buffer
;
1562 return trace_buffer_lock_reserve(*current_rb
,
1563 type
, len
, flags
, pc
);
1565 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve
);
1567 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
1568 struct ring_buffer_event
*event
,
1569 unsigned long flags
, int pc
)
1571 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
);
1573 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit
);
1575 void trace_buffer_unlock_commit_regs(struct ring_buffer
*buffer
,
1576 struct ring_buffer_event
*event
,
1577 unsigned long flags
, int pc
,
1578 struct pt_regs
*regs
)
1580 __buffer_unlock_commit(buffer
, event
);
1582 ftrace_trace_stack_regs(buffer
, flags
, 0, pc
, regs
);
1583 ftrace_trace_userstack(buffer
, flags
, pc
);
1585 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs
);
1587 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
1588 struct ring_buffer_event
*event
)
1590 ring_buffer_discard_commit(buffer
, event
);
1592 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit
);
1595 trace_function(struct trace_array
*tr
,
1596 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1599 struct ftrace_event_call
*call
= &event_function
;
1600 struct ring_buffer
*buffer
= tr
->trace_buffer
.buffer
;
1601 struct ring_buffer_event
*event
;
1602 struct ftrace_entry
*entry
;
1604 /* If we are reading the ring buffer, don't trace */
1605 if (unlikely(__this_cpu_read(ftrace_cpu_disabled
)))
1608 event
= trace_buffer_lock_reserve(buffer
, TRACE_FN
, sizeof(*entry
),
1612 entry
= ring_buffer_event_data(event
);
1614 entry
->parent_ip
= parent_ip
;
1616 if (!filter_check_discard(call
, entry
, buffer
, event
))
1617 __buffer_unlock_commit(buffer
, event
);
1621 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1622 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1625 if (likely(!atomic_read(&data
->disabled
)))
1626 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
1629 #ifdef CONFIG_STACKTRACE
1631 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1632 struct ftrace_stack
{
1633 unsigned long calls
[FTRACE_STACK_MAX_ENTRIES
];
1636 static DEFINE_PER_CPU(struct ftrace_stack
, ftrace_stack
);
1637 static DEFINE_PER_CPU(int, ftrace_stack_reserve
);
1639 static void __ftrace_trace_stack(struct ring_buffer
*buffer
,
1640 unsigned long flags
,
1641 int skip
, int pc
, struct pt_regs
*regs
)
1643 struct ftrace_event_call
*call
= &event_kernel_stack
;
1644 struct ring_buffer_event
*event
;
1645 struct stack_entry
*entry
;
1646 struct stack_trace trace
;
1648 int size
= FTRACE_STACK_ENTRIES
;
1650 trace
.nr_entries
= 0;
1654 * Since events can happen in NMIs there's no safe way to
1655 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1656 * or NMI comes in, it will just have to use the default
1657 * FTRACE_STACK_SIZE.
1659 preempt_disable_notrace();
1661 use_stack
= __this_cpu_inc_return(ftrace_stack_reserve
);
1663 * We don't need any atomic variables, just a barrier.
1664 * If an interrupt comes in, we don't care, because it would
1665 * have exited and put the counter back to what we want.
1666 * We just need a barrier to keep gcc from moving things
1670 if (use_stack
== 1) {
1671 trace
.entries
= &__get_cpu_var(ftrace_stack
).calls
[0];
1672 trace
.max_entries
= FTRACE_STACK_MAX_ENTRIES
;
1675 save_stack_trace_regs(regs
, &trace
);
1677 save_stack_trace(&trace
);
1679 if (trace
.nr_entries
> size
)
1680 size
= trace
.nr_entries
;
1682 /* From now on, use_stack is a boolean */
1685 size
*= sizeof(unsigned long);
1687 event
= trace_buffer_lock_reserve(buffer
, TRACE_STACK
,
1688 sizeof(*entry
) + size
, flags
, pc
);
1691 entry
= ring_buffer_event_data(event
);
1693 memset(&entry
->caller
, 0, size
);
1696 memcpy(&entry
->caller
, trace
.entries
,
1697 trace
.nr_entries
* sizeof(unsigned long));
1699 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1700 trace
.entries
= entry
->caller
;
1702 save_stack_trace_regs(regs
, &trace
);
1704 save_stack_trace(&trace
);
1707 entry
->size
= trace
.nr_entries
;
1709 if (!filter_check_discard(call
, entry
, buffer
, event
))
1710 __buffer_unlock_commit(buffer
, event
);
1713 /* Again, don't let gcc optimize things here */
1715 __this_cpu_dec(ftrace_stack_reserve
);
1716 preempt_enable_notrace();
1720 void ftrace_trace_stack_regs(struct ring_buffer
*buffer
, unsigned long flags
,
1721 int skip
, int pc
, struct pt_regs
*regs
)
1723 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1726 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, regs
);
1729 void ftrace_trace_stack(struct ring_buffer
*buffer
, unsigned long flags
,
1732 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1735 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, NULL
);
1738 void __trace_stack(struct trace_array
*tr
, unsigned long flags
, int skip
,
1741 __ftrace_trace_stack(tr
->trace_buffer
.buffer
, flags
, skip
, pc
, NULL
);
1745 * trace_dump_stack - record a stack back trace in the trace buffer
1746 * @skip: Number of functions to skip (helper handlers)
1748 void trace_dump_stack(int skip
)
1750 unsigned long flags
;
1752 if (tracing_disabled
|| tracing_selftest_running
)
1755 local_save_flags(flags
);
1758 * Skip 3 more, seems to get us at the caller of
1762 __ftrace_trace_stack(global_trace
.trace_buffer
.buffer
,
1763 flags
, skip
, preempt_count(), NULL
);
1766 static DEFINE_PER_CPU(int, user_stack_count
);
1769 ftrace_trace_userstack(struct ring_buffer
*buffer
, unsigned long flags
, int pc
)
1771 struct ftrace_event_call
*call
= &event_user_stack
;
1772 struct ring_buffer_event
*event
;
1773 struct userstack_entry
*entry
;
1774 struct stack_trace trace
;
1776 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1780 * NMIs can not handle page faults, even with fix ups.
1781 * The save user stack can (and often does) fault.
1783 if (unlikely(in_nmi()))
1787 * prevent recursion, since the user stack tracing may
1788 * trigger other kernel events.
1791 if (__this_cpu_read(user_stack_count
))
1794 __this_cpu_inc(user_stack_count
);
1796 event
= trace_buffer_lock_reserve(buffer
, TRACE_USER_STACK
,
1797 sizeof(*entry
), flags
, pc
);
1799 goto out_drop_count
;
1800 entry
= ring_buffer_event_data(event
);
1802 entry
->tgid
= current
->tgid
;
1803 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1805 trace
.nr_entries
= 0;
1806 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1808 trace
.entries
= entry
->caller
;
1810 save_stack_trace_user(&trace
);
1811 if (!filter_check_discard(call
, entry
, buffer
, event
))
1812 __buffer_unlock_commit(buffer
, event
);
1815 __this_cpu_dec(user_stack_count
);
1821 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1823 ftrace_trace_userstack(tr
, flags
, preempt_count());
1827 #endif /* CONFIG_STACKTRACE */
1829 /* created for use with alloc_percpu */
1830 struct trace_buffer_struct
{
1831 char buffer
[TRACE_BUF_SIZE
];
1834 static struct trace_buffer_struct
*trace_percpu_buffer
;
1835 static struct trace_buffer_struct
*trace_percpu_sirq_buffer
;
1836 static struct trace_buffer_struct
*trace_percpu_irq_buffer
;
1837 static struct trace_buffer_struct
*trace_percpu_nmi_buffer
;
1840 * The buffer used is dependent on the context. There is a per cpu
1841 * buffer for normal context, softirq contex, hard irq context and
1842 * for NMI context. Thise allows for lockless recording.
1844 * Note, if the buffers failed to be allocated, then this returns NULL
1846 static char *get_trace_buf(void)
1848 struct trace_buffer_struct
*percpu_buffer
;
1851 * If we have allocated per cpu buffers, then we do not
1852 * need to do any locking.
1855 percpu_buffer
= trace_percpu_nmi_buffer
;
1857 percpu_buffer
= trace_percpu_irq_buffer
;
1858 else if (in_softirq())
1859 percpu_buffer
= trace_percpu_sirq_buffer
;
1861 percpu_buffer
= trace_percpu_buffer
;
1866 return this_cpu_ptr(&percpu_buffer
->buffer
[0]);
1869 static int alloc_percpu_trace_buffer(void)
1871 struct trace_buffer_struct
*buffers
;
1872 struct trace_buffer_struct
*sirq_buffers
;
1873 struct trace_buffer_struct
*irq_buffers
;
1874 struct trace_buffer_struct
*nmi_buffers
;
1876 buffers
= alloc_percpu(struct trace_buffer_struct
);
1880 sirq_buffers
= alloc_percpu(struct trace_buffer_struct
);
1884 irq_buffers
= alloc_percpu(struct trace_buffer_struct
);
1888 nmi_buffers
= alloc_percpu(struct trace_buffer_struct
);
1892 trace_percpu_buffer
= buffers
;
1893 trace_percpu_sirq_buffer
= sirq_buffers
;
1894 trace_percpu_irq_buffer
= irq_buffers
;
1895 trace_percpu_nmi_buffer
= nmi_buffers
;
1900 free_percpu(irq_buffers
);
1902 free_percpu(sirq_buffers
);
1904 free_percpu(buffers
);
1906 WARN(1, "Could not allocate percpu trace_printk buffer");
1910 static int buffers_allocated
;
1912 void trace_printk_init_buffers(void)
1914 if (buffers_allocated
)
1917 if (alloc_percpu_trace_buffer())
1920 pr_info("ftrace: Allocated trace_printk buffers\n");
1922 /* Expand the buffers to set size */
1923 tracing_update_buffers();
1925 buffers_allocated
= 1;
1928 * trace_printk_init_buffers() can be called by modules.
1929 * If that happens, then we need to start cmdline recording
1930 * directly here. If the global_trace.buffer is already
1931 * allocated here, then this was called by module code.
1933 if (global_trace
.trace_buffer
.buffer
)
1934 tracing_start_cmdline_record();
1937 void trace_printk_start_comm(void)
1939 /* Start tracing comms if trace printk is set */
1940 if (!buffers_allocated
)
1942 tracing_start_cmdline_record();
1945 static void trace_printk_start_stop_comm(int enabled
)
1947 if (!buffers_allocated
)
1951 tracing_start_cmdline_record();
1953 tracing_stop_cmdline_record();
1957 * trace_vbprintk - write binary msg to tracing buffer
1960 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1962 struct ftrace_event_call
*call
= &event_bprint
;
1963 struct ring_buffer_event
*event
;
1964 struct ring_buffer
*buffer
;
1965 struct trace_array
*tr
= &global_trace
;
1966 struct bprint_entry
*entry
;
1967 unsigned long flags
;
1969 int len
= 0, size
, pc
;
1971 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1974 /* Don't pollute graph traces with trace_vprintk internals */
1975 pause_graph_tracing();
1977 pc
= preempt_count();
1978 preempt_disable_notrace();
1980 tbuffer
= get_trace_buf();
1986 len
= vbin_printf((u32
*)tbuffer
, TRACE_BUF_SIZE
/sizeof(int), fmt
, args
);
1988 if (len
> TRACE_BUF_SIZE
/sizeof(int) || len
< 0)
1991 local_save_flags(flags
);
1992 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1993 buffer
= tr
->trace_buffer
.buffer
;
1994 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
1998 entry
= ring_buffer_event_data(event
);
2002 memcpy(entry
->buf
, tbuffer
, sizeof(u32
) * len
);
2003 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
2004 __buffer_unlock_commit(buffer
, event
);
2005 ftrace_trace_stack(buffer
, flags
, 6, pc
);
2009 preempt_enable_notrace();
2010 unpause_graph_tracing();
2014 EXPORT_SYMBOL_GPL(trace_vbprintk
);
2017 __trace_array_vprintk(struct ring_buffer
*buffer
,
2018 unsigned long ip
, const char *fmt
, va_list args
)
2020 struct ftrace_event_call
*call
= &event_print
;
2021 struct ring_buffer_event
*event
;
2022 int len
= 0, size
, pc
;
2023 struct print_entry
*entry
;
2024 unsigned long flags
;
2027 if (tracing_disabled
|| tracing_selftest_running
)
2030 /* Don't pollute graph traces with trace_vprintk internals */
2031 pause_graph_tracing();
2033 pc
= preempt_count();
2034 preempt_disable_notrace();
2037 tbuffer
= get_trace_buf();
2043 len
= vsnprintf(tbuffer
, TRACE_BUF_SIZE
, fmt
, args
);
2044 if (len
> TRACE_BUF_SIZE
)
2047 local_save_flags(flags
);
2048 size
= sizeof(*entry
) + len
+ 1;
2049 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
2053 entry
= ring_buffer_event_data(event
);
2056 memcpy(&entry
->buf
, tbuffer
, len
);
2057 entry
->buf
[len
] = '\0';
2058 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
2059 __buffer_unlock_commit(buffer
, event
);
2060 ftrace_trace_stack(buffer
, flags
, 6, pc
);
2063 preempt_enable_notrace();
2064 unpause_graph_tracing();
2069 int trace_array_vprintk(struct trace_array
*tr
,
2070 unsigned long ip
, const char *fmt
, va_list args
)
2072 return __trace_array_vprintk(tr
->trace_buffer
.buffer
, ip
, fmt
, args
);
2075 int trace_array_printk(struct trace_array
*tr
,
2076 unsigned long ip
, const char *fmt
, ...)
2081 if (!(trace_flags
& TRACE_ITER_PRINTK
))
2085 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
2090 int trace_array_printk_buf(struct ring_buffer
*buffer
,
2091 unsigned long ip
, const char *fmt
, ...)
2096 if (!(trace_flags
& TRACE_ITER_PRINTK
))
2100 ret
= __trace_array_vprintk(buffer
, ip
, fmt
, ap
);
2105 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
2107 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
2109 EXPORT_SYMBOL_GPL(trace_vprintk
);
2111 static void trace_iterator_increment(struct trace_iterator
*iter
)
2113 struct ring_buffer_iter
*buf_iter
= trace_buffer_iter(iter
, iter
->cpu
);
2117 ring_buffer_read(buf_iter
, NULL
);
2120 static struct trace_entry
*
2121 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
,
2122 unsigned long *lost_events
)
2124 struct ring_buffer_event
*event
;
2125 struct ring_buffer_iter
*buf_iter
= trace_buffer_iter(iter
, cpu
);
2128 event
= ring_buffer_iter_peek(buf_iter
, ts
);
2130 event
= ring_buffer_peek(iter
->trace_buffer
->buffer
, cpu
, ts
,
2134 iter
->ent_size
= ring_buffer_event_length(event
);
2135 return ring_buffer_event_data(event
);
2141 static struct trace_entry
*
2142 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
,
2143 unsigned long *missing_events
, u64
*ent_ts
)
2145 struct ring_buffer
*buffer
= iter
->trace_buffer
->buffer
;
2146 struct trace_entry
*ent
, *next
= NULL
;
2147 unsigned long lost_events
= 0, next_lost
= 0;
2148 int cpu_file
= iter
->cpu_file
;
2149 u64 next_ts
= 0, ts
;
2155 * If we are in a per_cpu trace file, don't bother by iterating over
2156 * all cpu and peek directly.
2158 if (cpu_file
> RING_BUFFER_ALL_CPUS
) {
2159 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
2161 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
, missing_events
);
2163 *ent_cpu
= cpu_file
;
2168 for_each_tracing_cpu(cpu
) {
2170 if (ring_buffer_empty_cpu(buffer
, cpu
))
2173 ent
= peek_next_entry(iter
, cpu
, &ts
, &lost_events
);
2176 * Pick the entry with the smallest timestamp:
2178 if (ent
&& (!next
|| ts
< next_ts
)) {
2182 next_lost
= lost_events
;
2183 next_size
= iter
->ent_size
;
2187 iter
->ent_size
= next_size
;
2190 *ent_cpu
= next_cpu
;
2196 *missing_events
= next_lost
;
2201 /* Find the next real entry, without updating the iterator itself */
2202 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
2203 int *ent_cpu
, u64
*ent_ts
)
2205 return __find_next_entry(iter
, ent_cpu
, NULL
, ent_ts
);
2208 /* Find the next real entry, and increment the iterator to the next entry */
2209 void *trace_find_next_entry_inc(struct trace_iterator
*iter
)
2211 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
,
2212 &iter
->lost_events
, &iter
->ts
);
2215 trace_iterator_increment(iter
);
2217 return iter
->ent
? iter
: NULL
;
2220 static void trace_consume(struct trace_iterator
*iter
)
2222 ring_buffer_consume(iter
->trace_buffer
->buffer
, iter
->cpu
, &iter
->ts
,
2223 &iter
->lost_events
);
2226 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2228 struct trace_iterator
*iter
= m
->private;
2232 WARN_ON_ONCE(iter
->leftover
);
2236 /* can't go backwards */
2241 ent
= trace_find_next_entry_inc(iter
);
2245 while (ent
&& iter
->idx
< i
)
2246 ent
= trace_find_next_entry_inc(iter
);
2253 void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
2255 struct ring_buffer_event
*event
;
2256 struct ring_buffer_iter
*buf_iter
;
2257 unsigned long entries
= 0;
2260 per_cpu_ptr(iter
->trace_buffer
->data
, cpu
)->skipped_entries
= 0;
2262 buf_iter
= trace_buffer_iter(iter
, cpu
);
2266 ring_buffer_iter_reset(buf_iter
);
2269 * We could have the case with the max latency tracers
2270 * that a reset never took place on a cpu. This is evident
2271 * by the timestamp being before the start of the buffer.
2273 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
2274 if (ts
>= iter
->trace_buffer
->time_start
)
2277 ring_buffer_read(buf_iter
, NULL
);
2280 per_cpu_ptr(iter
->trace_buffer
->data
, cpu
)->skipped_entries
= entries
;
2284 * The current tracer is copied to avoid a global locking
2287 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
2289 struct trace_iterator
*iter
= m
->private;
2290 struct trace_array
*tr
= iter
->tr
;
2291 int cpu_file
= iter
->cpu_file
;
2297 * copy the tracer to avoid using a global lock all around.
2298 * iter->trace is a copy of current_trace, the pointer to the
2299 * name may be used instead of a strcmp(), as iter->trace->name
2300 * will point to the same string as current_trace->name.
2302 mutex_lock(&trace_types_lock
);
2303 if (unlikely(tr
->current_trace
&& iter
->trace
->name
!= tr
->current_trace
->name
))
2304 *iter
->trace
= *tr
->current_trace
;
2305 mutex_unlock(&trace_types_lock
);
2307 #ifdef CONFIG_TRACER_MAX_TRACE
2308 if (iter
->snapshot
&& iter
->trace
->use_max_tr
)
2309 return ERR_PTR(-EBUSY
);
2312 if (!iter
->snapshot
)
2313 atomic_inc(&trace_record_cmdline_disabled
);
2315 if (*pos
!= iter
->pos
) {
2320 if (cpu_file
== RING_BUFFER_ALL_CPUS
) {
2321 for_each_tracing_cpu(cpu
)
2322 tracing_iter_reset(iter
, cpu
);
2324 tracing_iter_reset(iter
, cpu_file
);
2327 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
2332 * If we overflowed the seq_file before, then we want
2333 * to just reuse the trace_seq buffer again.
2339 p
= s_next(m
, p
, &l
);
2343 trace_event_read_lock();
2344 trace_access_lock(cpu_file
);
2348 static void s_stop(struct seq_file
*m
, void *p
)
2350 struct trace_iterator
*iter
= m
->private;
2352 #ifdef CONFIG_TRACER_MAX_TRACE
2353 if (iter
->snapshot
&& iter
->trace
->use_max_tr
)
2357 if (!iter
->snapshot
)
2358 atomic_dec(&trace_record_cmdline_disabled
);
2360 trace_access_unlock(iter
->cpu_file
);
2361 trace_event_read_unlock();
2365 get_total_entries(struct trace_buffer
*buf
,
2366 unsigned long *total
, unsigned long *entries
)
2368 unsigned long count
;
2374 for_each_tracing_cpu(cpu
) {
2375 count
= ring_buffer_entries_cpu(buf
->buffer
, cpu
);
2377 * If this buffer has skipped entries, then we hold all
2378 * entries for the trace and we need to ignore the
2379 * ones before the time stamp.
2381 if (per_cpu_ptr(buf
->data
, cpu
)->skipped_entries
) {
2382 count
-= per_cpu_ptr(buf
->data
, cpu
)->skipped_entries
;
2383 /* total is the same as the entries */
2387 ring_buffer_overrun_cpu(buf
->buffer
, cpu
);
2392 static void print_lat_help_header(struct seq_file
*m
)
2394 seq_puts(m
, "# _------=> CPU# \n");
2395 seq_puts(m
, "# / _-----=> irqs-off \n");
2396 seq_puts(m
, "# | / _----=> need-resched \n");
2397 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
2398 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
2399 seq_puts(m
, "# |||| / delay \n");
2400 seq_puts(m
, "# cmd pid ||||| time | caller \n");
2401 seq_puts(m
, "# \\ / ||||| \\ | / \n");
2404 static void print_event_info(struct trace_buffer
*buf
, struct seq_file
*m
)
2406 unsigned long total
;
2407 unsigned long entries
;
2409 get_total_entries(buf
, &total
, &entries
);
2410 seq_printf(m
, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2411 entries
, total
, num_online_cpus());
2415 static void print_func_help_header(struct trace_buffer
*buf
, struct seq_file
*m
)
2417 print_event_info(buf
, m
);
2418 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2419 seq_puts(m
, "# | | | | |\n");
2422 static void print_func_help_header_irq(struct trace_buffer
*buf
, struct seq_file
*m
)
2424 print_event_info(buf
, m
);
2425 seq_puts(m
, "# _-----=> irqs-off\n");
2426 seq_puts(m
, "# / _----=> need-resched\n");
2427 seq_puts(m
, "# | / _---=> hardirq/softirq\n");
2428 seq_puts(m
, "# || / _--=> preempt-depth\n");
2429 seq_puts(m
, "# ||| / delay\n");
2430 seq_puts(m
, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2431 seq_puts(m
, "# | | | |||| | |\n");
2435 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
2437 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2438 struct trace_buffer
*buf
= iter
->trace_buffer
;
2439 struct trace_array_cpu
*data
= per_cpu_ptr(buf
->data
, buf
->cpu
);
2440 struct tracer
*type
= iter
->trace
;
2441 unsigned long entries
;
2442 unsigned long total
;
2443 const char *name
= "preemption";
2447 get_total_entries(buf
, &total
, &entries
);
2449 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
2451 seq_puts(m
, "# -----------------------------------"
2452 "---------------------------------\n");
2453 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2454 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2455 nsecs_to_usecs(data
->saved_latency
),
2459 #if defined(CONFIG_PREEMPT_NONE)
2461 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2463 #elif defined(CONFIG_PREEMPT)
2468 /* These are reserved for later use */
2471 seq_printf(m
, " #P:%d)\n", num_online_cpus());
2475 seq_puts(m
, "# -----------------\n");
2476 seq_printf(m
, "# | task: %.16s-%d "
2477 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2478 data
->comm
, data
->pid
,
2479 from_kuid_munged(seq_user_ns(m
), data
->uid
), data
->nice
,
2480 data
->policy
, data
->rt_priority
);
2481 seq_puts(m
, "# -----------------\n");
2483 if (data
->critical_start
) {
2484 seq_puts(m
, "# => started at: ");
2485 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
2486 trace_print_seq(m
, &iter
->seq
);
2487 seq_puts(m
, "\n# => ended at: ");
2488 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
2489 trace_print_seq(m
, &iter
->seq
);
2490 seq_puts(m
, "\n#\n");
2496 static void test_cpu_buff_start(struct trace_iterator
*iter
)
2498 struct trace_seq
*s
= &iter
->seq
;
2500 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
2503 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
2506 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
2509 if (per_cpu_ptr(iter
->trace_buffer
->data
, iter
->cpu
)->skipped_entries
)
2512 cpumask_set_cpu(iter
->cpu
, iter
->started
);
2514 /* Don't print started cpu buffer for the first entry of the trace */
2516 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
2520 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
2522 struct trace_seq
*s
= &iter
->seq
;
2523 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2524 struct trace_entry
*entry
;
2525 struct trace_event
*event
;
2529 test_cpu_buff_start(iter
);
2531 event
= ftrace_find_event(entry
->type
);
2533 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2534 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2535 if (!trace_print_lat_context(iter
))
2538 if (!trace_print_context(iter
))
2544 return event
->funcs
->trace(iter
, sym_flags
, event
);
2546 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
2549 return TRACE_TYPE_HANDLED
;
2551 return TRACE_TYPE_PARTIAL_LINE
;
2554 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
2556 struct trace_seq
*s
= &iter
->seq
;
2557 struct trace_entry
*entry
;
2558 struct trace_event
*event
;
2562 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2563 if (!trace_seq_printf(s
, "%d %d %llu ",
2564 entry
->pid
, iter
->cpu
, iter
->ts
))
2568 event
= ftrace_find_event(entry
->type
);
2570 return event
->funcs
->raw(iter
, 0, event
);
2572 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
2575 return TRACE_TYPE_HANDLED
;
2577 return TRACE_TYPE_PARTIAL_LINE
;
2580 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
2582 struct trace_seq
*s
= &iter
->seq
;
2583 unsigned char newline
= '\n';
2584 struct trace_entry
*entry
;
2585 struct trace_event
*event
;
2589 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2590 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
2591 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
2592 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
2595 event
= ftrace_find_event(entry
->type
);
2597 enum print_line_t ret
= event
->funcs
->hex(iter
, 0, event
);
2598 if (ret
!= TRACE_TYPE_HANDLED
)
2602 SEQ_PUT_FIELD_RET(s
, newline
);
2604 return TRACE_TYPE_HANDLED
;
2607 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
2609 struct trace_seq
*s
= &iter
->seq
;
2610 struct trace_entry
*entry
;
2611 struct trace_event
*event
;
2615 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2616 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
2617 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
2618 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
2621 event
= ftrace_find_event(entry
->type
);
2622 return event
? event
->funcs
->binary(iter
, 0, event
) :
2626 int trace_empty(struct trace_iterator
*iter
)
2628 struct ring_buffer_iter
*buf_iter
;
2631 /* If we are looking at one CPU buffer, only check that one */
2632 if (iter
->cpu_file
!= RING_BUFFER_ALL_CPUS
) {
2633 cpu
= iter
->cpu_file
;
2634 buf_iter
= trace_buffer_iter(iter
, cpu
);
2636 if (!ring_buffer_iter_empty(buf_iter
))
2639 if (!ring_buffer_empty_cpu(iter
->trace_buffer
->buffer
, cpu
))
2645 for_each_tracing_cpu(cpu
) {
2646 buf_iter
= trace_buffer_iter(iter
, cpu
);
2648 if (!ring_buffer_iter_empty(buf_iter
))
2651 if (!ring_buffer_empty_cpu(iter
->trace_buffer
->buffer
, cpu
))
2659 /* Called with trace_event_read_lock() held. */
2660 enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
2662 enum print_line_t ret
;
2664 if (iter
->lost_events
&&
2665 !trace_seq_printf(&iter
->seq
, "CPU:%d [LOST %lu EVENTS]\n",
2666 iter
->cpu
, iter
->lost_events
))
2667 return TRACE_TYPE_PARTIAL_LINE
;
2669 if (iter
->trace
&& iter
->trace
->print_line
) {
2670 ret
= iter
->trace
->print_line(iter
);
2671 if (ret
!= TRACE_TYPE_UNHANDLED
)
2675 if (iter
->ent
->type
== TRACE_BPUTS
&&
2676 trace_flags
& TRACE_ITER_PRINTK
&&
2677 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2678 return trace_print_bputs_msg_only(iter
);
2680 if (iter
->ent
->type
== TRACE_BPRINT
&&
2681 trace_flags
& TRACE_ITER_PRINTK
&&
2682 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2683 return trace_print_bprintk_msg_only(iter
);
2685 if (iter
->ent
->type
== TRACE_PRINT
&&
2686 trace_flags
& TRACE_ITER_PRINTK
&&
2687 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2688 return trace_print_printk_msg_only(iter
);
2690 if (trace_flags
& TRACE_ITER_BIN
)
2691 return print_bin_fmt(iter
);
2693 if (trace_flags
& TRACE_ITER_HEX
)
2694 return print_hex_fmt(iter
);
2696 if (trace_flags
& TRACE_ITER_RAW
)
2697 return print_raw_fmt(iter
);
2699 return print_trace_fmt(iter
);
2702 void trace_latency_header(struct seq_file
*m
)
2704 struct trace_iterator
*iter
= m
->private;
2706 /* print nothing if the buffers are empty */
2707 if (trace_empty(iter
))
2710 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
2711 print_trace_header(m
, iter
);
2713 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2714 print_lat_help_header(m
);
2717 void trace_default_header(struct seq_file
*m
)
2719 struct trace_iterator
*iter
= m
->private;
2721 if (!(trace_flags
& TRACE_ITER_CONTEXT_INFO
))
2724 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2725 /* print nothing if the buffers are empty */
2726 if (trace_empty(iter
))
2728 print_trace_header(m
, iter
);
2729 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2730 print_lat_help_header(m
);
2732 if (!(trace_flags
& TRACE_ITER_VERBOSE
)) {
2733 if (trace_flags
& TRACE_ITER_IRQ_INFO
)
2734 print_func_help_header_irq(iter
->trace_buffer
, m
);
2736 print_func_help_header(iter
->trace_buffer
, m
);
2741 static void test_ftrace_alive(struct seq_file
*m
)
2743 if (!ftrace_is_dead())
2745 seq_printf(m
, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2746 seq_printf(m
, "# MAY BE MISSING FUNCTION EVENTS\n");
2749 #ifdef CONFIG_TRACER_MAX_TRACE
2750 static void show_snapshot_main_help(struct seq_file
*m
)
2752 seq_printf(m
, "# echo 0 > snapshot : Clears and frees snapshot buffer\n");
2753 seq_printf(m
, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2754 seq_printf(m
, "# Takes a snapshot of the main buffer.\n");
2755 seq_printf(m
, "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate)\n");
2756 seq_printf(m
, "# (Doesn't have to be '2' works with any number that\n");
2757 seq_printf(m
, "# is not a '0' or '1')\n");
2760 static void show_snapshot_percpu_help(struct seq_file
*m
)
2762 seq_printf(m
, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
2763 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2764 seq_printf(m
, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n");
2765 seq_printf(m
, "# Takes a snapshot of the main buffer for this cpu.\n");
2767 seq_printf(m
, "# echo 1 > snapshot : Not supported with this kernel.\n");
2768 seq_printf(m
, "# Must use main snapshot file to allocate.\n");
2770 seq_printf(m
, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n");
2771 seq_printf(m
, "# (Doesn't have to be '2' works with any number that\n");
2772 seq_printf(m
, "# is not a '0' or '1')\n");
2775 static void print_snapshot_help(struct seq_file
*m
, struct trace_iterator
*iter
)
2777 if (iter
->tr
->allocated_snapshot
)
2778 seq_printf(m
, "#\n# * Snapshot is allocated *\n#\n");
2780 seq_printf(m
, "#\n# * Snapshot is freed *\n#\n");
2782 seq_printf(m
, "# Snapshot commands:\n");
2783 if (iter
->cpu_file
== RING_BUFFER_ALL_CPUS
)
2784 show_snapshot_main_help(m
);
2786 show_snapshot_percpu_help(m
);
2789 /* Should never be called */
2790 static inline void print_snapshot_help(struct seq_file
*m
, struct trace_iterator
*iter
) { }
2793 static int s_show(struct seq_file
*m
, void *v
)
2795 struct trace_iterator
*iter
= v
;
2798 if (iter
->ent
== NULL
) {
2800 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
2802 test_ftrace_alive(m
);
2804 if (iter
->snapshot
&& trace_empty(iter
))
2805 print_snapshot_help(m
, iter
);
2806 else if (iter
->trace
&& iter
->trace
->print_header
)
2807 iter
->trace
->print_header(m
);
2809 trace_default_header(m
);
2811 } else if (iter
->leftover
) {
2813 * If we filled the seq_file buffer earlier, we
2814 * want to just show it now.
2816 ret
= trace_print_seq(m
, &iter
->seq
);
2818 /* ret should this time be zero, but you never know */
2819 iter
->leftover
= ret
;
2822 print_trace_line(iter
);
2823 ret
= trace_print_seq(m
, &iter
->seq
);
2825 * If we overflow the seq_file buffer, then it will
2826 * ask us for this data again at start up.
2828 * ret is 0 if seq_file write succeeded.
2831 iter
->leftover
= ret
;
2838 * Should be used after trace_array_get(), trace_types_lock
2839 * ensures that i_cdev was already initialized.
2841 static inline int tracing_get_cpu(struct inode
*inode
)
2843 if (inode
->i_cdev
) /* See trace_create_cpu_file() */
2844 return (long)inode
->i_cdev
- 1;
2845 return RING_BUFFER_ALL_CPUS
;
2848 static const struct seq_operations tracer_seq_ops
= {
2855 static struct trace_iterator
*
2856 __tracing_open(struct trace_array
*tr
, struct trace_cpu
*tc
,
2857 struct inode
*inode
, struct file
*file
, bool snapshot
)
2859 struct trace_iterator
*iter
;
2862 if (tracing_disabled
)
2863 return ERR_PTR(-ENODEV
);
2865 iter
= __seq_open_private(file
, &tracer_seq_ops
, sizeof(*iter
));
2867 return ERR_PTR(-ENOMEM
);
2869 iter
->buffer_iter
= kzalloc(sizeof(*iter
->buffer_iter
) * num_possible_cpus(),
2871 if (!iter
->buffer_iter
)
2875 * We make a copy of the current tracer to avoid concurrent
2876 * changes on it while we are reading.
2878 mutex_lock(&trace_types_lock
);
2879 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2883 *iter
->trace
= *tr
->current_trace
;
2885 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2890 #ifdef CONFIG_TRACER_MAX_TRACE
2891 /* Currently only the top directory has a snapshot */
2892 if (tr
->current_trace
->print_max
|| snapshot
)
2893 iter
->trace_buffer
= &tr
->max_buffer
;
2896 iter
->trace_buffer
= &tr
->trace_buffer
;
2897 iter
->snapshot
= snapshot
;
2899 mutex_init(&iter
->mutex
);
2900 iter
->cpu_file
= tc
->cpu
;
2902 /* Notify the tracer early; before we stop tracing. */
2903 if (iter
->trace
&& iter
->trace
->open
)
2904 iter
->trace
->open(iter
);
2906 /* Annotate start of buffers if we had overruns */
2907 if (ring_buffer_overruns(iter
->trace_buffer
->buffer
))
2908 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2910 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2911 if (trace_clocks
[tr
->clock_id
].in_ns
)
2912 iter
->iter_flags
|= TRACE_FILE_TIME_IN_NS
;
2914 /* stop the trace while dumping if we are not opening "snapshot" */
2915 if (!iter
->snapshot
)
2916 tracing_stop_tr(tr
);
2918 if (iter
->cpu_file
== RING_BUFFER_ALL_CPUS
) {
2919 for_each_tracing_cpu(cpu
) {
2920 iter
->buffer_iter
[cpu
] =
2921 ring_buffer_read_prepare(iter
->trace_buffer
->buffer
, cpu
);
2923 ring_buffer_read_prepare_sync();
2924 for_each_tracing_cpu(cpu
) {
2925 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2926 tracing_iter_reset(iter
, cpu
);
2929 cpu
= iter
->cpu_file
;
2930 iter
->buffer_iter
[cpu
] =
2931 ring_buffer_read_prepare(iter
->trace_buffer
->buffer
, cpu
);
2932 ring_buffer_read_prepare_sync();
2933 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2934 tracing_iter_reset(iter
, cpu
);
2937 mutex_unlock(&trace_types_lock
);
2942 mutex_unlock(&trace_types_lock
);
2944 kfree(iter
->buffer_iter
);
2946 seq_release_private(inode
, file
);
2947 return ERR_PTR(-ENOMEM
);
2950 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2952 if (tracing_disabled
)
2955 filp
->private_data
= inode
->i_private
;
2960 * Open and update trace_array ref count.
2961 * Must have the current trace_array passed to it.
2963 int tracing_open_generic_tr(struct inode
*inode
, struct file
*filp
)
2965 struct trace_array
*tr
= inode
->i_private
;
2967 if (tracing_disabled
)
2970 if (trace_array_get(tr
) < 0)
2973 filp
->private_data
= inode
->i_private
;
2979 int tracing_open_generic_tc(struct inode
*inode
, struct file
*filp
)
2981 struct trace_cpu
*tc
= inode
->i_private
;
2982 struct trace_array
*tr
= tc
->tr
;
2984 if (tracing_disabled
)
2987 if (trace_array_get(tr
) < 0)
2990 filp
->private_data
= inode
->i_private
;
2996 static int tracing_release(struct inode
*inode
, struct file
*file
)
2998 struct seq_file
*m
= file
->private_data
;
2999 struct trace_iterator
*iter
;
3000 struct trace_array
*tr
;
3003 /* Writes do not use seq_file, need to grab tr from inode */
3004 if (!(file
->f_mode
& FMODE_READ
)) {
3005 struct trace_cpu
*tc
= inode
->i_private
;
3007 trace_array_put(tc
->tr
);
3014 mutex_lock(&trace_types_lock
);
3016 for_each_tracing_cpu(cpu
) {
3017 if (iter
->buffer_iter
[cpu
])
3018 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
3021 if (iter
->trace
&& iter
->trace
->close
)
3022 iter
->trace
->close(iter
);
3024 if (!iter
->snapshot
)
3025 /* reenable tracing if it was previously enabled */
3026 tracing_start_tr(tr
);
3028 __trace_array_put(tr
);
3030 mutex_unlock(&trace_types_lock
);
3032 mutex_destroy(&iter
->mutex
);
3033 free_cpumask_var(iter
->started
);
3035 kfree(iter
->buffer_iter
);
3036 seq_release_private(inode
, file
);
3041 static int tracing_release_generic_tr(struct inode
*inode
, struct file
*file
)
3043 struct trace_array
*tr
= inode
->i_private
;
3045 trace_array_put(tr
);
3049 static int tracing_release_generic_tc(struct inode
*inode
, struct file
*file
)
3051 struct trace_cpu
*tc
= inode
->i_private
;
3052 struct trace_array
*tr
= tc
->tr
;
3054 trace_array_put(tr
);
3058 static int tracing_single_release_tr(struct inode
*inode
, struct file
*file
)
3060 struct trace_array
*tr
= inode
->i_private
;
3062 trace_array_put(tr
);
3064 return single_release(inode
, file
);
3067 static int tracing_open(struct inode
*inode
, struct file
*file
)
3069 struct trace_cpu
*tc
= inode
->i_private
;
3070 struct trace_array
*tr
= tc
->tr
;
3071 struct trace_iterator
*iter
;
3074 if (trace_array_get(tr
) < 0)
3077 /* If this file was open for write, then erase contents */
3078 if ((file
->f_mode
& FMODE_WRITE
) &&
3079 (file
->f_flags
& O_TRUNC
)) {
3080 if (tc
->cpu
== RING_BUFFER_ALL_CPUS
)
3081 tracing_reset_online_cpus(&tr
->trace_buffer
);
3083 tracing_reset(&tr
->trace_buffer
, tc
->cpu
);
3086 if (file
->f_mode
& FMODE_READ
) {
3087 iter
= __tracing_open(tr
, tc
, inode
, file
, false);
3089 ret
= PTR_ERR(iter
);
3090 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
3091 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
3095 trace_array_put(tr
);
3101 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
3103 struct tracer
*t
= v
;
3113 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
3118 mutex_lock(&trace_types_lock
);
3119 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
3125 static void t_stop(struct seq_file
*m
, void *p
)
3127 mutex_unlock(&trace_types_lock
);
3130 static int t_show(struct seq_file
*m
, void *v
)
3132 struct tracer
*t
= v
;
3137 seq_printf(m
, "%s", t
->name
);
3146 static const struct seq_operations show_traces_seq_ops
= {
3153 static int show_traces_open(struct inode
*inode
, struct file
*file
)
3155 if (tracing_disabled
)
3158 return seq_open(file
, &show_traces_seq_ops
);
3162 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
3163 size_t count
, loff_t
*ppos
)
3168 static loff_t
tracing_seek(struct file
*file
, loff_t offset
, int origin
)
3170 if (file
->f_mode
& FMODE_READ
)
3171 return seq_lseek(file
, offset
, origin
);
3176 static const struct file_operations tracing_fops
= {
3177 .open
= tracing_open
,
3179 .write
= tracing_write_stub
,
3180 .llseek
= tracing_seek
,
3181 .release
= tracing_release
,
3184 static const struct file_operations show_traces_fops
= {
3185 .open
= show_traces_open
,
3187 .release
= seq_release
,
3188 .llseek
= seq_lseek
,
3192 * Only trace on a CPU if the bitmask is set:
3194 static cpumask_var_t tracing_cpumask
;
3197 * The tracer itself will not take this lock, but still we want
3198 * to provide a consistent cpumask to user-space:
3200 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
3203 * Temporary storage for the character representation of the
3204 * CPU bitmask (and one more byte for the newline):
3206 static char mask_str
[NR_CPUS
+ 1];
3209 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
3210 size_t count
, loff_t
*ppos
)
3214 mutex_lock(&tracing_cpumask_update_lock
);
3216 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
3217 if (count
- len
< 2) {
3221 len
+= sprintf(mask_str
+ len
, "\n");
3222 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
3225 mutex_unlock(&tracing_cpumask_update_lock
);
3231 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
3232 size_t count
, loff_t
*ppos
)
3234 struct trace_array
*tr
= filp
->private_data
;
3235 cpumask_var_t tracing_cpumask_new
;
3238 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
3241 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
3245 mutex_lock(&tracing_cpumask_update_lock
);
3247 local_irq_disable();
3248 arch_spin_lock(&ftrace_max_lock
);
3249 for_each_tracing_cpu(cpu
) {
3251 * Increase/decrease the disabled counter if we are
3252 * about to flip a bit in the cpumask:
3254 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
3255 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
3256 atomic_inc(&per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->disabled
);
3257 ring_buffer_record_disable_cpu(tr
->trace_buffer
.buffer
, cpu
);
3259 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
3260 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
3261 atomic_dec(&per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->disabled
);
3262 ring_buffer_record_enable_cpu(tr
->trace_buffer
.buffer
, cpu
);
3265 arch_spin_unlock(&ftrace_max_lock
);
3268 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
3270 mutex_unlock(&tracing_cpumask_update_lock
);
3271 free_cpumask_var(tracing_cpumask_new
);
3276 free_cpumask_var(tracing_cpumask_new
);
3281 static const struct file_operations tracing_cpumask_fops
= {
3282 .open
= tracing_open_generic
,
3283 .read
= tracing_cpumask_read
,
3284 .write
= tracing_cpumask_write
,
3285 .llseek
= generic_file_llseek
,
3288 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
3290 struct tracer_opt
*trace_opts
;
3291 struct trace_array
*tr
= m
->private;
3295 mutex_lock(&trace_types_lock
);
3296 tracer_flags
= tr
->current_trace
->flags
->val
;
3297 trace_opts
= tr
->current_trace
->flags
->opts
;
3299 for (i
= 0; trace_options
[i
]; i
++) {
3300 if (trace_flags
& (1 << i
))
3301 seq_printf(m
, "%s\n", trace_options
[i
]);
3303 seq_printf(m
, "no%s\n", trace_options
[i
]);
3306 for (i
= 0; trace_opts
[i
].name
; i
++) {
3307 if (tracer_flags
& trace_opts
[i
].bit
)
3308 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
3310 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
3312 mutex_unlock(&trace_types_lock
);
3317 static int __set_tracer_option(struct tracer
*trace
,
3318 struct tracer_flags
*tracer_flags
,
3319 struct tracer_opt
*opts
, int neg
)
3323 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
3328 tracer_flags
->val
&= ~opts
->bit
;
3330 tracer_flags
->val
|= opts
->bit
;
3334 /* Try to assign a tracer specific option */
3335 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
3337 struct tracer_flags
*tracer_flags
= trace
->flags
;
3338 struct tracer_opt
*opts
= NULL
;
3341 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
3342 opts
= &tracer_flags
->opts
[i
];
3344 if (strcmp(cmp
, opts
->name
) == 0)
3345 return __set_tracer_option(trace
, trace
->flags
,
3352 /* Some tracers require overwrite to stay enabled */
3353 int trace_keep_overwrite(struct tracer
*tracer
, u32 mask
, int set
)
3355 if (tracer
->enabled
&& (mask
& TRACE_ITER_OVERWRITE
) && !set
)
3361 int set_tracer_flag(struct trace_array
*tr
, unsigned int mask
, int enabled
)
3363 /* do nothing if flag is already set */
3364 if (!!(trace_flags
& mask
) == !!enabled
)
3367 /* Give the tracer a chance to approve the change */
3368 if (tr
->current_trace
->flag_changed
)
3369 if (tr
->current_trace
->flag_changed(tr
->current_trace
, mask
, !!enabled
))
3373 trace_flags
|= mask
;
3375 trace_flags
&= ~mask
;
3377 if (mask
== TRACE_ITER_RECORD_CMD
)
3378 trace_event_enable_cmd_record(enabled
);
3380 if (mask
== TRACE_ITER_OVERWRITE
) {
3381 ring_buffer_change_overwrite(tr
->trace_buffer
.buffer
, enabled
);
3382 #ifdef CONFIG_TRACER_MAX_TRACE
3383 ring_buffer_change_overwrite(tr
->max_buffer
.buffer
, enabled
);
3387 if (mask
== TRACE_ITER_PRINTK
)
3388 trace_printk_start_stop_comm(enabled
);
3393 static int trace_set_options(struct trace_array
*tr
, char *option
)
3400 cmp
= strstrip(option
);
3402 if (strncmp(cmp
, "no", 2) == 0) {
3407 mutex_lock(&trace_types_lock
);
3409 for (i
= 0; trace_options
[i
]; i
++) {
3410 if (strcmp(cmp
, trace_options
[i
]) == 0) {
3411 ret
= set_tracer_flag(tr
, 1 << i
, !neg
);
3416 /* If no option could be set, test the specific tracer options */
3417 if (!trace_options
[i
])
3418 ret
= set_tracer_option(tr
->current_trace
, cmp
, neg
);
3420 mutex_unlock(&trace_types_lock
);
3426 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
3427 size_t cnt
, loff_t
*ppos
)
3429 struct seq_file
*m
= filp
->private_data
;
3430 struct trace_array
*tr
= m
->private;
3434 if (cnt
>= sizeof(buf
))
3437 if (copy_from_user(&buf
, ubuf
, cnt
))
3442 ret
= trace_set_options(tr
, buf
);
3451 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
3453 struct trace_array
*tr
= inode
->i_private
;
3456 if (tracing_disabled
)
3459 if (trace_array_get(tr
) < 0)
3462 ret
= single_open(file
, tracing_trace_options_show
, inode
->i_private
);
3464 trace_array_put(tr
);
3469 static const struct file_operations tracing_iter_fops
= {
3470 .open
= tracing_trace_options_open
,
3472 .llseek
= seq_lseek
,
3473 .release
= tracing_single_release_tr
,
3474 .write
= tracing_trace_options_write
,
3477 static const char readme_msg
[] =
3478 "tracing mini-HOWTO:\n\n"
3479 "# echo 0 > tracing_on : quick way to disable tracing\n"
3480 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
3481 " Important files:\n"
3482 " trace\t\t\t- The static contents of the buffer\n"
3483 "\t\t\t To clear the buffer write into this file: echo > trace\n"
3484 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
3485 " current_tracer\t- function and latency tracers\n"
3486 " available_tracers\t- list of configured tracers for current_tracer\n"
3487 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
3488 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
3489 " trace_clock\t\t-change the clock used to order events\n"
3490 " local: Per cpu clock but may not be synced across CPUs\n"
3491 " global: Synced across CPUs but slows tracing down.\n"
3492 " counter: Not a clock, but just an increment\n"
3493 " uptime: Jiffy counter from time of boot\n"
3494 " perf: Same clock that perf events use\n"
3495 #ifdef CONFIG_X86_64
3496 " x86-tsc: TSC cycle counter\n"
3498 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
3499 " tracing_cpumask\t- Limit which CPUs to trace\n"
3500 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
3501 "\t\t\t Remove sub-buffer with rmdir\n"
3502 " trace_options\t\t- Set format or modify how tracing happens\n"
3503 "\t\t\t Disable an option by adding a suffix 'no' to the option name\n"
3504 #ifdef CONFIG_DYNAMIC_FTRACE
3505 "\n available_filter_functions - list of functions that can be filtered on\n"
3506 " set_ftrace_filter\t- echo function name in here to only trace these functions\n"
3507 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3508 " modules: Can select a group via module\n"
3509 " Format: :mod:<module-name>\n"
3510 " example: echo :mod:ext3 > set_ftrace_filter\n"
3511 " triggers: a command to perform when function is hit\n"
3512 " Format: <function>:<trigger>[:count]\n"
3513 " trigger: traceon, traceoff\n"
3514 " enable_event:<system>:<event>\n"
3515 " disable_event:<system>:<event>\n"
3516 #ifdef CONFIG_STACKTRACE
3519 #ifdef CONFIG_TRACER_SNAPSHOT
3522 " example: echo do_fault:traceoff > set_ftrace_filter\n"
3523 " echo do_trap:traceoff:3 > set_ftrace_filter\n"
3524 " The first one will disable tracing every time do_fault is hit\n"
3525 " The second will disable tracing at most 3 times when do_trap is hit\n"
3526 " The first time do trap is hit and it disables tracing, the counter\n"
3527 " will decrement to 2. If tracing is already disabled, the counter\n"
3528 " will not decrement. It only decrements when the trigger did work\n"
3529 " To remove trigger without count:\n"
3530 " echo '!<function>:<trigger> > set_ftrace_filter\n"
3531 " To remove trigger with a count:\n"
3532 " echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
3533 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
3534 " accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
3535 " modules: Can select a group via module command :mod:\n"
3536 " Does not accept triggers\n"
3537 #endif /* CONFIG_DYNAMIC_FTRACE */
3538 #ifdef CONFIG_FUNCTION_TRACER
3539 " set_ftrace_pid\t- Write pid(s) to only function trace those pids (function)\n"
3541 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3542 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
3543 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
3545 #ifdef CONFIG_TRACER_SNAPSHOT
3546 "\n snapshot\t\t- Like 'trace' but shows the content of the static snapshot buffer\n"
3547 "\t\t\t Read the contents for more information\n"
3549 #ifdef CONFIG_STACKTRACE
3550 " stack_trace\t\t- Shows the max stack trace when active\n"
3551 " stack_max_size\t- Shows current max stack size that was traced\n"
3552 "\t\t\t Write into this file to reset the max size (trigger a new trace)\n"
3553 #ifdef CONFIG_DYNAMIC_FTRACE
3554 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace traces\n"
3556 #endif /* CONFIG_STACKTRACE */
3560 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
3561 size_t cnt
, loff_t
*ppos
)
3563 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
3564 readme_msg
, strlen(readme_msg
));
3567 static const struct file_operations tracing_readme_fops
= {
3568 .open
= tracing_open_generic
,
3569 .read
= tracing_readme_read
,
3570 .llseek
= generic_file_llseek
,
3574 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
3575 size_t cnt
, loff_t
*ppos
)
3584 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
3588 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
3596 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
3599 pid
= map_cmdline_to_pid
[i
];
3600 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
3603 trace_find_cmdline(pid
, buf_comm
);
3604 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
3609 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
3618 static const struct file_operations tracing_saved_cmdlines_fops
= {
3619 .open
= tracing_open_generic
,
3620 .read
= tracing_saved_cmdlines_read
,
3621 .llseek
= generic_file_llseek
,
3625 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
3626 size_t cnt
, loff_t
*ppos
)
3628 struct trace_array
*tr
= filp
->private_data
;
3629 char buf
[MAX_TRACER_SIZE
+2];
3632 mutex_lock(&trace_types_lock
);
3633 r
= sprintf(buf
, "%s\n", tr
->current_trace
->name
);
3634 mutex_unlock(&trace_types_lock
);
3636 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3639 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
3641 tracing_reset_online_cpus(&tr
->trace_buffer
);
3645 static void set_buffer_entries(struct trace_buffer
*buf
, unsigned long val
)
3649 for_each_tracing_cpu(cpu
)
3650 per_cpu_ptr(buf
->data
, cpu
)->entries
= val
;
3653 #ifdef CONFIG_TRACER_MAX_TRACE
3654 /* resize @tr's buffer to the size of @size_tr's entries */
3655 static int resize_buffer_duplicate_size(struct trace_buffer
*trace_buf
,
3656 struct trace_buffer
*size_buf
, int cpu_id
)
3660 if (cpu_id
== RING_BUFFER_ALL_CPUS
) {
3661 for_each_tracing_cpu(cpu
) {
3662 ret
= ring_buffer_resize(trace_buf
->buffer
,
3663 per_cpu_ptr(size_buf
->data
, cpu
)->entries
, cpu
);
3666 per_cpu_ptr(trace_buf
->data
, cpu
)->entries
=
3667 per_cpu_ptr(size_buf
->data
, cpu
)->entries
;
3670 ret
= ring_buffer_resize(trace_buf
->buffer
,
3671 per_cpu_ptr(size_buf
->data
, cpu_id
)->entries
, cpu_id
);
3673 per_cpu_ptr(trace_buf
->data
, cpu_id
)->entries
=
3674 per_cpu_ptr(size_buf
->data
, cpu_id
)->entries
;
3679 #endif /* CONFIG_TRACER_MAX_TRACE */
3681 static int __tracing_resize_ring_buffer(struct trace_array
*tr
,
3682 unsigned long size
, int cpu
)
3687 * If kernel or user changes the size of the ring buffer
3688 * we use the size that was given, and we can forget about
3689 * expanding it later.
3691 ring_buffer_expanded
= true;
3693 /* May be called before buffers are initialized */
3694 if (!tr
->trace_buffer
.buffer
)
3697 ret
= ring_buffer_resize(tr
->trace_buffer
.buffer
, size
, cpu
);
3701 #ifdef CONFIG_TRACER_MAX_TRACE
3702 if (!(tr
->flags
& TRACE_ARRAY_FL_GLOBAL
) ||
3703 !tr
->current_trace
->use_max_tr
)
3706 ret
= ring_buffer_resize(tr
->max_buffer
.buffer
, size
, cpu
);
3708 int r
= resize_buffer_duplicate_size(&tr
->trace_buffer
,
3709 &tr
->trace_buffer
, cpu
);
3712 * AARGH! We are left with different
3713 * size max buffer!!!!
3714 * The max buffer is our "snapshot" buffer.
3715 * When a tracer needs a snapshot (one of the
3716 * latency tracers), it swaps the max buffer
3717 * with the saved snap shot. We succeeded to
3718 * update the size of the main buffer, but failed to
3719 * update the size of the max buffer. But when we tried
3720 * to reset the main buffer to the original size, we
3721 * failed there too. This is very unlikely to
3722 * happen, but if it does, warn and kill all
3726 tracing_disabled
= 1;
3731 if (cpu
== RING_BUFFER_ALL_CPUS
)
3732 set_buffer_entries(&tr
->max_buffer
, size
);
3734 per_cpu_ptr(tr
->max_buffer
.data
, cpu
)->entries
= size
;
3737 #endif /* CONFIG_TRACER_MAX_TRACE */
3739 if (cpu
== RING_BUFFER_ALL_CPUS
)
3740 set_buffer_entries(&tr
->trace_buffer
, size
);
3742 per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->entries
= size
;
3747 static ssize_t
tracing_resize_ring_buffer(struct trace_array
*tr
,
3748 unsigned long size
, int cpu_id
)
3752 mutex_lock(&trace_types_lock
);
3754 if (cpu_id
!= RING_BUFFER_ALL_CPUS
) {
3755 /* make sure, this cpu is enabled in the mask */
3756 if (!cpumask_test_cpu(cpu_id
, tracing_buffer_mask
)) {
3762 ret
= __tracing_resize_ring_buffer(tr
, size
, cpu_id
);
3767 mutex_unlock(&trace_types_lock
);
3774 * tracing_update_buffers - used by tracing facility to expand ring buffers
3776 * To save on memory when the tracing is never used on a system with it
3777 * configured in. The ring buffers are set to a minimum size. But once
3778 * a user starts to use the tracing facility, then they need to grow
3779 * to their default size.
3781 * This function is to be called when a tracer is about to be used.
3783 int tracing_update_buffers(void)
3787 mutex_lock(&trace_types_lock
);
3788 if (!ring_buffer_expanded
)
3789 ret
= __tracing_resize_ring_buffer(&global_trace
, trace_buf_size
,
3790 RING_BUFFER_ALL_CPUS
);
3791 mutex_unlock(&trace_types_lock
);
3796 struct trace_option_dentry
;
3798 static struct trace_option_dentry
*
3799 create_trace_option_files(struct trace_array
*tr
, struct tracer
*tracer
);
3802 destroy_trace_option_files(struct trace_option_dentry
*topts
);
3804 static int tracing_set_tracer(const char *buf
)
3806 static struct trace_option_dentry
*topts
;
3807 struct trace_array
*tr
= &global_trace
;
3809 #ifdef CONFIG_TRACER_MAX_TRACE
3814 mutex_lock(&trace_types_lock
);
3816 if (!ring_buffer_expanded
) {
3817 ret
= __tracing_resize_ring_buffer(tr
, trace_buf_size
,
3818 RING_BUFFER_ALL_CPUS
);
3824 for (t
= trace_types
; t
; t
= t
->next
) {
3825 if (strcmp(t
->name
, buf
) == 0)
3832 if (t
== tr
->current_trace
)
3835 trace_branch_disable();
3837 tr
->current_trace
->enabled
= false;
3839 if (tr
->current_trace
->reset
)
3840 tr
->current_trace
->reset(tr
);
3842 /* Current trace needs to be nop_trace before synchronize_sched */
3843 tr
->current_trace
= &nop_trace
;
3845 #ifdef CONFIG_TRACER_MAX_TRACE
3846 had_max_tr
= tr
->allocated_snapshot
;
3848 if (had_max_tr
&& !t
->use_max_tr
) {
3850 * We need to make sure that the update_max_tr sees that
3851 * current_trace changed to nop_trace to keep it from
3852 * swapping the buffers after we resize it.
3853 * The update_max_tr is called from interrupts disabled
3854 * so a synchronized_sched() is sufficient.
3856 synchronize_sched();
3860 destroy_trace_option_files(topts
);
3862 topts
= create_trace_option_files(tr
, t
);
3864 #ifdef CONFIG_TRACER_MAX_TRACE
3865 if (t
->use_max_tr
&& !had_max_tr
) {
3866 ret
= alloc_snapshot(tr
);
3873 ret
= tracer_init(t
, tr
);
3878 tr
->current_trace
= t
;
3879 tr
->current_trace
->enabled
= true;
3880 trace_branch_enable(tr
);
3882 mutex_unlock(&trace_types_lock
);
3888 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
3889 size_t cnt
, loff_t
*ppos
)
3891 char buf
[MAX_TRACER_SIZE
+1];
3898 if (cnt
> MAX_TRACER_SIZE
)
3899 cnt
= MAX_TRACER_SIZE
;
3901 if (copy_from_user(&buf
, ubuf
, cnt
))
3906 /* strip ending whitespace. */
3907 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
3910 err
= tracing_set_tracer(buf
);
3920 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
3921 size_t cnt
, loff_t
*ppos
)
3923 unsigned long *ptr
= filp
->private_data
;
3927 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
3928 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
3929 if (r
> sizeof(buf
))
3931 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3935 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
3936 size_t cnt
, loff_t
*ppos
)
3938 unsigned long *ptr
= filp
->private_data
;
3942 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3951 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
3953 struct trace_array
*tr
= inode
->i_private
;
3954 struct trace_iterator
*iter
;
3957 if (tracing_disabled
)
3960 if (trace_array_get(tr
) < 0)
3963 mutex_lock(&trace_types_lock
);
3965 /* create a buffer to store the information to pass to userspace */
3966 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
3969 __trace_array_put(tr
);
3974 * We make a copy of the current tracer to avoid concurrent
3975 * changes on it while we are reading.
3977 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
3982 *iter
->trace
= *tr
->current_trace
;
3984 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
3989 /* trace pipe does not show start of buffer */
3990 cpumask_setall(iter
->started
);
3992 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
3993 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
3995 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3996 if (trace_clocks
[tr
->clock_id
].in_ns
)
3997 iter
->iter_flags
|= TRACE_FILE_TIME_IN_NS
;
4000 iter
->trace_buffer
= &tr
->trace_buffer
;
4001 iter
->cpu_file
= tracing_get_cpu(inode
);
4002 mutex_init(&iter
->mutex
);
4003 filp
->private_data
= iter
;
4005 if (iter
->trace
->pipe_open
)
4006 iter
->trace
->pipe_open(iter
);
4008 nonseekable_open(inode
, filp
);
4010 mutex_unlock(&trace_types_lock
);
4016 __trace_array_put(tr
);
4017 mutex_unlock(&trace_types_lock
);
4021 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
4023 struct trace_iterator
*iter
= file
->private_data
;
4024 struct trace_array
*tr
= inode
->i_private
;
4026 mutex_lock(&trace_types_lock
);
4028 if (iter
->trace
->pipe_close
)
4029 iter
->trace
->pipe_close(iter
);
4031 mutex_unlock(&trace_types_lock
);
4033 free_cpumask_var(iter
->started
);
4034 mutex_destroy(&iter
->mutex
);
4038 trace_array_put(tr
);
4044 trace_poll(struct trace_iterator
*iter
, struct file
*filp
, poll_table
*poll_table
)
4046 /* Iterators are static, they should be filled or empty */
4047 if (trace_buffer_iter(iter
, iter
->cpu_file
))
4048 return POLLIN
| POLLRDNORM
;
4050 if (trace_flags
& TRACE_ITER_BLOCK
)
4052 * Always select as readable when in blocking mode
4054 return POLLIN
| POLLRDNORM
;
4056 return ring_buffer_poll_wait(iter
->trace_buffer
->buffer
, iter
->cpu_file
,
4061 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
4063 struct trace_iterator
*iter
= filp
->private_data
;
4065 return trace_poll(iter
, filp
, poll_table
);
4069 * This is a make-shift waitqueue.
4070 * A tracer might use this callback on some rare cases:
4072 * 1) the current tracer might hold the runqueue lock when it wakes up
4073 * a reader, hence a deadlock (sched, function, and function graph tracers)
4074 * 2) the function tracers, trace all functions, we don't want
4075 * the overhead of calling wake_up and friends
4076 * (and tracing them too)
4078 * Anyway, this is really very primitive wakeup.
4080 void poll_wait_pipe(struct trace_iterator
*iter
)
4082 set_current_state(TASK_INTERRUPTIBLE
);
4083 /* sleep for 100 msecs, and try again. */
4084 schedule_timeout(HZ
/ 10);
4087 /* Must be called with trace_types_lock mutex held. */
4088 static int tracing_wait_pipe(struct file
*filp
)
4090 struct trace_iterator
*iter
= filp
->private_data
;
4092 while (trace_empty(iter
)) {
4094 if ((filp
->f_flags
& O_NONBLOCK
)) {
4098 mutex_unlock(&iter
->mutex
);
4100 iter
->trace
->wait_pipe(iter
);
4102 mutex_lock(&iter
->mutex
);
4104 if (signal_pending(current
))
4108 * We block until we read something and tracing is disabled.
4109 * We still block if tracing is disabled, but we have never
4110 * read anything. This allows a user to cat this file, and
4111 * then enable tracing. But after we have read something,
4112 * we give an EOF when tracing is again disabled.
4114 * iter->pos will be 0 if we haven't read anything.
4116 if (!tracing_is_on() && iter
->pos
)
4127 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
4128 size_t cnt
, loff_t
*ppos
)
4130 struct trace_iterator
*iter
= filp
->private_data
;
4131 struct trace_array
*tr
= iter
->tr
;
4134 /* return any leftover data */
4135 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
4139 trace_seq_init(&iter
->seq
);
4141 /* copy the tracer to avoid using a global lock all around */
4142 mutex_lock(&trace_types_lock
);
4143 if (unlikely(iter
->trace
->name
!= tr
->current_trace
->name
))
4144 *iter
->trace
= *tr
->current_trace
;
4145 mutex_unlock(&trace_types_lock
);
4148 * Avoid more than one consumer on a single file descriptor
4149 * This is just a matter of traces coherency, the ring buffer itself
4152 mutex_lock(&iter
->mutex
);
4153 if (iter
->trace
->read
) {
4154 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
4160 sret
= tracing_wait_pipe(filp
);
4164 /* stop when tracing is finished */
4165 if (trace_empty(iter
)) {
4170 if (cnt
>= PAGE_SIZE
)
4171 cnt
= PAGE_SIZE
- 1;
4173 /* reset all but tr, trace, and overruns */
4174 memset(&iter
->seq
, 0,
4175 sizeof(struct trace_iterator
) -
4176 offsetof(struct trace_iterator
, seq
));
4177 cpumask_clear(iter
->started
);
4180 trace_event_read_lock();
4181 trace_access_lock(iter
->cpu_file
);
4182 while (trace_find_next_entry_inc(iter
) != NULL
) {
4183 enum print_line_t ret
;
4184 int len
= iter
->seq
.len
;
4186 ret
= print_trace_line(iter
);
4187 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
4188 /* don't print partial lines */
4189 iter
->seq
.len
= len
;
4192 if (ret
!= TRACE_TYPE_NO_CONSUME
)
4193 trace_consume(iter
);
4195 if (iter
->seq
.len
>= cnt
)
4199 * Setting the full flag means we reached the trace_seq buffer
4200 * size and we should leave by partial output condition above.
4201 * One of the trace_seq_* functions is not used properly.
4203 WARN_ONCE(iter
->seq
.full
, "full flag set for trace type %d",
4206 trace_access_unlock(iter
->cpu_file
);
4207 trace_event_read_unlock();
4209 /* Now copy what we have to the user */
4210 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
4211 if (iter
->seq
.readpos
>= iter
->seq
.len
)
4212 trace_seq_init(&iter
->seq
);
4215 * If there was nothing to send to user, in spite of consuming trace
4216 * entries, go back to wait for more entries.
4222 mutex_unlock(&iter
->mutex
);
4227 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
4228 struct pipe_buffer
*buf
)
4230 __free_page(buf
->page
);
4233 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
4236 __free_page(spd
->pages
[idx
]);
4239 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
4241 .map
= generic_pipe_buf_map
,
4242 .unmap
= generic_pipe_buf_unmap
,
4243 .confirm
= generic_pipe_buf_confirm
,
4244 .release
= tracing_pipe_buf_release
,
4245 .steal
= generic_pipe_buf_steal
,
4246 .get
= generic_pipe_buf_get
,
4250 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
4255 /* Seq buffer is page-sized, exactly what we need. */
4257 count
= iter
->seq
.len
;
4258 ret
= print_trace_line(iter
);
4259 count
= iter
->seq
.len
- count
;
4262 iter
->seq
.len
-= count
;
4265 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
4266 iter
->seq
.len
-= count
;
4270 if (ret
!= TRACE_TYPE_NO_CONSUME
)
4271 trace_consume(iter
);
4273 if (!trace_find_next_entry_inc(iter
)) {
4283 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
4285 struct pipe_inode_info
*pipe
,
4289 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
4290 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
4291 struct trace_iterator
*iter
= filp
->private_data
;
4292 struct splice_pipe_desc spd
= {
4294 .partial
= partial_def
,
4295 .nr_pages
= 0, /* This gets updated below. */
4296 .nr_pages_max
= PIPE_DEF_BUFFERS
,
4298 .ops
= &tracing_pipe_buf_ops
,
4299 .spd_release
= tracing_spd_release_pipe
,
4301 struct trace_array
*tr
= iter
->tr
;
4306 if (splice_grow_spd(pipe
, &spd
))
4309 /* copy the tracer to avoid using a global lock all around */
4310 mutex_lock(&trace_types_lock
);
4311 if (unlikely(iter
->trace
->name
!= tr
->current_trace
->name
))
4312 *iter
->trace
= *tr
->current_trace
;
4313 mutex_unlock(&trace_types_lock
);
4315 mutex_lock(&iter
->mutex
);
4317 if (iter
->trace
->splice_read
) {
4318 ret
= iter
->trace
->splice_read(iter
, filp
,
4319 ppos
, pipe
, len
, flags
);
4324 ret
= tracing_wait_pipe(filp
);
4328 if (!iter
->ent
&& !trace_find_next_entry_inc(iter
)) {
4333 trace_event_read_lock();
4334 trace_access_lock(iter
->cpu_file
);
4336 /* Fill as many pages as possible. */
4337 for (i
= 0, rem
= len
; i
< pipe
->buffers
&& rem
; i
++) {
4338 spd
.pages
[i
] = alloc_page(GFP_KERNEL
);
4342 rem
= tracing_fill_pipe_page(rem
, iter
);
4344 /* Copy the data into the page, so we can start over. */
4345 ret
= trace_seq_to_buffer(&iter
->seq
,
4346 page_address(spd
.pages
[i
]),
4349 __free_page(spd
.pages
[i
]);
4352 spd
.partial
[i
].offset
= 0;
4353 spd
.partial
[i
].len
= iter
->seq
.len
;
4355 trace_seq_init(&iter
->seq
);
4358 trace_access_unlock(iter
->cpu_file
);
4359 trace_event_read_unlock();
4360 mutex_unlock(&iter
->mutex
);
4364 ret
= splice_to_pipe(pipe
, &spd
);
4366 splice_shrink_spd(&spd
);
4370 mutex_unlock(&iter
->mutex
);
4375 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
4376 size_t cnt
, loff_t
*ppos
)
4378 struct trace_cpu
*tc
= filp
->private_data
;
4379 struct trace_array
*tr
= tc
->tr
;
4384 mutex_lock(&trace_types_lock
);
4386 if (tc
->cpu
== RING_BUFFER_ALL_CPUS
) {
4387 int cpu
, buf_size_same
;
4392 /* check if all cpu sizes are same */
4393 for_each_tracing_cpu(cpu
) {
4394 /* fill in the size from first enabled cpu */
4396 size
= per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->entries
;
4397 if (size
!= per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->entries
) {
4403 if (buf_size_same
) {
4404 if (!ring_buffer_expanded
)
4405 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
4407 trace_buf_size
>> 10);
4409 r
= sprintf(buf
, "%lu\n", size
>> 10);
4411 r
= sprintf(buf
, "X\n");
4413 r
= sprintf(buf
, "%lu\n", per_cpu_ptr(tr
->trace_buffer
.data
, tc
->cpu
)->entries
>> 10);
4415 mutex_unlock(&trace_types_lock
);
4417 ret
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4422 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
4423 size_t cnt
, loff_t
*ppos
)
4425 struct trace_cpu
*tc
= filp
->private_data
;
4429 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4433 /* must have at least 1 entry */
4437 /* value is in KB */
4440 ret
= tracing_resize_ring_buffer(tc
->tr
, val
, tc
->cpu
);
4450 tracing_total_entries_read(struct file
*filp
, char __user
*ubuf
,
4451 size_t cnt
, loff_t
*ppos
)
4453 struct trace_array
*tr
= filp
->private_data
;
4456 unsigned long size
= 0, expanded_size
= 0;
4458 mutex_lock(&trace_types_lock
);
4459 for_each_tracing_cpu(cpu
) {
4460 size
+= per_cpu_ptr(tr
->trace_buffer
.data
, cpu
)->entries
>> 10;
4461 if (!ring_buffer_expanded
)
4462 expanded_size
+= trace_buf_size
>> 10;
4464 if (ring_buffer_expanded
)
4465 r
= sprintf(buf
, "%lu\n", size
);
4467 r
= sprintf(buf
, "%lu (expanded: %lu)\n", size
, expanded_size
);
4468 mutex_unlock(&trace_types_lock
);
4470 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4474 tracing_free_buffer_write(struct file
*filp
, const char __user
*ubuf
,
4475 size_t cnt
, loff_t
*ppos
)
4478 * There is no need to read what the user has written, this function
4479 * is just to make sure that there is no error when "echo" is used
4488 tracing_free_buffer_release(struct inode
*inode
, struct file
*filp
)
4490 struct trace_array
*tr
= inode
->i_private
;
4492 /* disable tracing ? */
4493 if (trace_flags
& TRACE_ITER_STOP_ON_FREE
)
4494 tracer_tracing_off(tr
);
4495 /* resize the ring buffer to 0 */
4496 tracing_resize_ring_buffer(tr
, 0, RING_BUFFER_ALL_CPUS
);
4498 trace_array_put(tr
);
4504 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
4505 size_t cnt
, loff_t
*fpos
)
4507 unsigned long addr
= (unsigned long)ubuf
;
4508 struct trace_array
*tr
= filp
->private_data
;
4509 struct ring_buffer_event
*event
;
4510 struct ring_buffer
*buffer
;
4511 struct print_entry
*entry
;
4512 unsigned long irq_flags
;
4513 struct page
*pages
[2];
4523 if (tracing_disabled
)
4526 if (!(trace_flags
& TRACE_ITER_MARKERS
))
4529 if (cnt
> TRACE_BUF_SIZE
)
4530 cnt
= TRACE_BUF_SIZE
;
4533 * Userspace is injecting traces into the kernel trace buffer.
4534 * We want to be as non intrusive as possible.
4535 * To do so, we do not want to allocate any special buffers
4536 * or take any locks, but instead write the userspace data
4537 * straight into the ring buffer.
4539 * First we need to pin the userspace buffer into memory,
4540 * which, most likely it is, because it just referenced it.
4541 * But there's no guarantee that it is. By using get_user_pages_fast()
4542 * and kmap_atomic/kunmap_atomic() we can get access to the
4543 * pages directly. We then write the data directly into the
4546 BUILD_BUG_ON(TRACE_BUF_SIZE
>= PAGE_SIZE
);
4548 /* check if we cross pages */
4549 if ((addr
& PAGE_MASK
) != ((addr
+ cnt
) & PAGE_MASK
))
4552 offset
= addr
& (PAGE_SIZE
- 1);
4555 ret
= get_user_pages_fast(addr
, nr_pages
, 0, pages
);
4556 if (ret
< nr_pages
) {
4558 put_page(pages
[ret
]);
4563 for (i
= 0; i
< nr_pages
; i
++)
4564 map_page
[i
] = kmap_atomic(pages
[i
]);
4566 local_save_flags(irq_flags
);
4567 size
= sizeof(*entry
) + cnt
+ 2; /* possible \n added */
4568 buffer
= tr
->trace_buffer
.buffer
;
4569 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
4570 irq_flags
, preempt_count());
4572 /* Ring buffer disabled, return as if not open for write */
4577 entry
= ring_buffer_event_data(event
);
4578 entry
->ip
= _THIS_IP_
;
4580 if (nr_pages
== 2) {
4581 len
= PAGE_SIZE
- offset
;
4582 memcpy(&entry
->buf
, map_page
[0] + offset
, len
);
4583 memcpy(&entry
->buf
[len
], map_page
[1], cnt
- len
);
4585 memcpy(&entry
->buf
, map_page
[0] + offset
, cnt
);
4587 if (entry
->buf
[cnt
- 1] != '\n') {
4588 entry
->buf
[cnt
] = '\n';
4589 entry
->buf
[cnt
+ 1] = '\0';
4591 entry
->buf
[cnt
] = '\0';
4593 __buffer_unlock_commit(buffer
, event
);
4600 for (i
= 0; i
< nr_pages
; i
++){
4601 kunmap_atomic(map_page
[i
]);
4608 static int tracing_clock_show(struct seq_file
*m
, void *v
)
4610 struct trace_array
*tr
= m
->private;
4613 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
4615 "%s%s%s%s", i
? " " : "",
4616 i
== tr
->clock_id
? "[" : "", trace_clocks
[i
].name
,
4617 i
== tr
->clock_id
? "]" : "");
4623 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
4624 size_t cnt
, loff_t
*fpos
)
4626 struct seq_file
*m
= filp
->private_data
;
4627 struct trace_array
*tr
= m
->private;
4629 const char *clockstr
;
4632 if (cnt
>= sizeof(buf
))
4635 if (copy_from_user(&buf
, ubuf
, cnt
))
4640 clockstr
= strstrip(buf
);
4642 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
4643 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
4646 if (i
== ARRAY_SIZE(trace_clocks
))
4649 mutex_lock(&trace_types_lock
);
4653 ring_buffer_set_clock(tr
->trace_buffer
.buffer
, trace_clocks
[i
].func
);
4656 * New clock may not be consistent with the previous clock.
4657 * Reset the buffer so that it doesn't have incomparable timestamps.
4659 tracing_reset_online_cpus(&tr
->trace_buffer
);
4661 #ifdef CONFIG_TRACER_MAX_TRACE
4662 if (tr
->flags
& TRACE_ARRAY_FL_GLOBAL
&& tr
->max_buffer
.buffer
)
4663 ring_buffer_set_clock(tr
->max_buffer
.buffer
, trace_clocks
[i
].func
);
4664 tracing_reset_online_cpus(&tr
->max_buffer
);
4667 mutex_unlock(&trace_types_lock
);
4674 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
4676 struct trace_array
*tr
= inode
->i_private
;
4679 if (tracing_disabled
)
4682 if (trace_array_get(tr
))
4685 ret
= single_open(file
, tracing_clock_show
, inode
->i_private
);
4687 trace_array_put(tr
);
4692 struct ftrace_buffer_info
{
4693 struct trace_iterator iter
;
4698 #ifdef CONFIG_TRACER_SNAPSHOT
4699 static int tracing_snapshot_open(struct inode
*inode
, struct file
*file
)
4701 struct trace_cpu
*tc
= inode
->i_private
;
4702 struct trace_array
*tr
= tc
->tr
;
4703 struct trace_iterator
*iter
;
4707 if (trace_array_get(tr
) < 0)
4710 if (file
->f_mode
& FMODE_READ
) {
4711 iter
= __tracing_open(tr
, tc
, inode
, file
, true);
4713 ret
= PTR_ERR(iter
);
4715 /* Writes still need the seq_file to hold the private data */
4717 m
= kzalloc(sizeof(*m
), GFP_KERNEL
);
4720 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
4728 iter
->trace_buffer
= &tc
->tr
->max_buffer
;
4729 iter
->cpu_file
= tc
->cpu
;
4731 file
->private_data
= m
;
4735 trace_array_put(tr
);
4741 tracing_snapshot_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4744 struct seq_file
*m
= filp
->private_data
;
4745 struct trace_iterator
*iter
= m
->private;
4746 struct trace_array
*tr
= iter
->tr
;
4750 ret
= tracing_update_buffers();
4754 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4758 mutex_lock(&trace_types_lock
);
4760 if (tr
->current_trace
->use_max_tr
) {
4767 if (iter
->cpu_file
!= RING_BUFFER_ALL_CPUS
) {
4771 if (tr
->allocated_snapshot
)
4775 /* Only allow per-cpu swap if the ring buffer supports it */
4776 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
4777 if (iter
->cpu_file
!= RING_BUFFER_ALL_CPUS
) {
4782 if (!tr
->allocated_snapshot
) {
4783 ret
= alloc_snapshot(tr
);
4787 local_irq_disable();
4788 /* Now, we're going to swap */
4789 if (iter
->cpu_file
== RING_BUFFER_ALL_CPUS
)
4790 update_max_tr(tr
, current
, smp_processor_id());
4792 update_max_tr_single(tr
, current
, iter
->cpu_file
);
4796 if (tr
->allocated_snapshot
) {
4797 if (iter
->cpu_file
== RING_BUFFER_ALL_CPUS
)
4798 tracing_reset_online_cpus(&tr
->max_buffer
);
4800 tracing_reset(&tr
->max_buffer
, iter
->cpu_file
);
4810 mutex_unlock(&trace_types_lock
);
4814 static int tracing_snapshot_release(struct inode
*inode
, struct file
*file
)
4816 struct seq_file
*m
= file
->private_data
;
4819 ret
= tracing_release(inode
, file
);
4821 if (file
->f_mode
& FMODE_READ
)
4824 /* If write only, the seq_file is just a stub */
4832 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
);
4833 static ssize_t
tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
4834 size_t count
, loff_t
*ppos
);
4835 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
);
4836 static ssize_t
tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
4837 struct pipe_inode_info
*pipe
, size_t len
, unsigned int flags
);
4839 static int snapshot_raw_open(struct inode
*inode
, struct file
*filp
)
4841 struct ftrace_buffer_info
*info
;
4844 ret
= tracing_buffers_open(inode
, filp
);
4848 info
= filp
->private_data
;
4850 if (info
->iter
.trace
->use_max_tr
) {
4851 tracing_buffers_release(inode
, filp
);
4855 info
->iter
.snapshot
= true;
4856 info
->iter
.trace_buffer
= &info
->iter
.tr
->max_buffer
;
4861 #endif /* CONFIG_TRACER_SNAPSHOT */
4864 static const struct file_operations tracing_max_lat_fops
= {
4865 .open
= tracing_open_generic
,
4866 .read
= tracing_max_lat_read
,
4867 .write
= tracing_max_lat_write
,
4868 .llseek
= generic_file_llseek
,
4871 static const struct file_operations set_tracer_fops
= {
4872 .open
= tracing_open_generic
,
4873 .read
= tracing_set_trace_read
,
4874 .write
= tracing_set_trace_write
,
4875 .llseek
= generic_file_llseek
,
4878 static const struct file_operations tracing_pipe_fops
= {
4879 .open
= tracing_open_pipe
,
4880 .poll
= tracing_poll_pipe
,
4881 .read
= tracing_read_pipe
,
4882 .splice_read
= tracing_splice_read_pipe
,
4883 .release
= tracing_release_pipe
,
4884 .llseek
= no_llseek
,
4887 static const struct file_operations tracing_entries_fops
= {
4888 .open
= tracing_open_generic_tc
,
4889 .read
= tracing_entries_read
,
4890 .write
= tracing_entries_write
,
4891 .llseek
= generic_file_llseek
,
4892 .release
= tracing_release_generic_tc
,
4895 static const struct file_operations tracing_total_entries_fops
= {
4896 .open
= tracing_open_generic_tr
,
4897 .read
= tracing_total_entries_read
,
4898 .llseek
= generic_file_llseek
,
4899 .release
= tracing_release_generic_tr
,
4902 static const struct file_operations tracing_free_buffer_fops
= {
4903 .open
= tracing_open_generic_tr
,
4904 .write
= tracing_free_buffer_write
,
4905 .release
= tracing_free_buffer_release
,
4908 static const struct file_operations tracing_mark_fops
= {
4909 .open
= tracing_open_generic_tr
,
4910 .write
= tracing_mark_write
,
4911 .llseek
= generic_file_llseek
,
4912 .release
= tracing_release_generic_tr
,
4915 static const struct file_operations trace_clock_fops
= {
4916 .open
= tracing_clock_open
,
4918 .llseek
= seq_lseek
,
4919 .release
= tracing_single_release_tr
,
4920 .write
= tracing_clock_write
,
4923 #ifdef CONFIG_TRACER_SNAPSHOT
4924 static const struct file_operations snapshot_fops
= {
4925 .open
= tracing_snapshot_open
,
4927 .write
= tracing_snapshot_write
,
4928 .llseek
= tracing_seek
,
4929 .release
= tracing_snapshot_release
,
4932 static const struct file_operations snapshot_raw_fops
= {
4933 .open
= snapshot_raw_open
,
4934 .read
= tracing_buffers_read
,
4935 .release
= tracing_buffers_release
,
4936 .splice_read
= tracing_buffers_splice_read
,
4937 .llseek
= no_llseek
,
4940 #endif /* CONFIG_TRACER_SNAPSHOT */
4942 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
4944 struct trace_cpu
*tc
= inode
->i_private
;
4945 struct trace_array
*tr
= tc
->tr
;
4946 struct ftrace_buffer_info
*info
;
4949 if (tracing_disabled
)
4952 if (trace_array_get(tr
) < 0)
4955 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
4957 trace_array_put(tr
);
4961 mutex_lock(&trace_types_lock
);
4964 info
->iter
.cpu_file
= tc
->cpu
;
4965 info
->iter
.trace
= tr
->current_trace
;
4966 info
->iter
.trace_buffer
= &tr
->trace_buffer
;
4968 /* Force reading ring buffer for first read */
4969 info
->read
= (unsigned int)-1;
4971 filp
->private_data
= info
;
4973 mutex_unlock(&trace_types_lock
);
4975 ret
= nonseekable_open(inode
, filp
);
4977 trace_array_put(tr
);
4983 tracing_buffers_poll(struct file
*filp
, poll_table
*poll_table
)
4985 struct ftrace_buffer_info
*info
= filp
->private_data
;
4986 struct trace_iterator
*iter
= &info
->iter
;
4988 return trace_poll(iter
, filp
, poll_table
);
4992 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
4993 size_t count
, loff_t
*ppos
)
4995 struct ftrace_buffer_info
*info
= filp
->private_data
;
4996 struct trace_iterator
*iter
= &info
->iter
;
5003 mutex_lock(&trace_types_lock
);
5005 #ifdef CONFIG_TRACER_MAX_TRACE
5006 if (iter
->snapshot
&& iter
->tr
->current_trace
->use_max_tr
) {
5013 info
->spare
= ring_buffer_alloc_read_page(iter
->trace_buffer
->buffer
,
5019 /* Do we have previous read data to read? */
5020 if (info
->read
< PAGE_SIZE
)
5024 trace_access_lock(iter
->cpu_file
);
5025 ret
= ring_buffer_read_page(iter
->trace_buffer
->buffer
,
5029 trace_access_unlock(iter
->cpu_file
);
5032 if (trace_empty(iter
)) {
5033 if ((filp
->f_flags
& O_NONBLOCK
)) {
5037 mutex_unlock(&trace_types_lock
);
5038 iter
->trace
->wait_pipe(iter
);
5039 mutex_lock(&trace_types_lock
);
5040 if (signal_pending(current
)) {
5052 size
= PAGE_SIZE
- info
->read
;
5056 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
5067 mutex_unlock(&trace_types_lock
);
5072 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
5074 struct ftrace_buffer_info
*info
= file
->private_data
;
5075 struct trace_iterator
*iter
= &info
->iter
;
5077 mutex_lock(&trace_types_lock
);
5079 __trace_array_put(iter
->tr
);
5082 ring_buffer_free_read_page(iter
->trace_buffer
->buffer
, info
->spare
);
5085 mutex_unlock(&trace_types_lock
);
5091 struct ring_buffer
*buffer
;
5096 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
5097 struct pipe_buffer
*buf
)
5099 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
5104 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
5109 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
5110 struct pipe_buffer
*buf
)
5112 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
5117 /* Pipe buffer operations for a buffer. */
5118 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
5120 .map
= generic_pipe_buf_map
,
5121 .unmap
= generic_pipe_buf_unmap
,
5122 .confirm
= generic_pipe_buf_confirm
,
5123 .release
= buffer_pipe_buf_release
,
5124 .steal
= generic_pipe_buf_steal
,
5125 .get
= buffer_pipe_buf_get
,
5129 * Callback from splice_to_pipe(), if we need to release some pages
5130 * at the end of the spd in case we error'ed out in filling the pipe.
5132 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
5134 struct buffer_ref
*ref
=
5135 (struct buffer_ref
*)spd
->partial
[i
].private;
5140 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
5142 spd
->partial
[i
].private = 0;
5146 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
5147 struct pipe_inode_info
*pipe
, size_t len
,
5150 struct ftrace_buffer_info
*info
= file
->private_data
;
5151 struct trace_iterator
*iter
= &info
->iter
;
5152 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
5153 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
5154 struct splice_pipe_desc spd
= {
5156 .partial
= partial_def
,
5157 .nr_pages_max
= PIPE_DEF_BUFFERS
,
5159 .ops
= &buffer_pipe_buf_ops
,
5160 .spd_release
= buffer_spd_release
,
5162 struct buffer_ref
*ref
;
5163 int entries
, size
, i
;
5166 mutex_lock(&trace_types_lock
);
5168 #ifdef CONFIG_TRACER_MAX_TRACE
5169 if (iter
->snapshot
&& iter
->tr
->current_trace
->use_max_tr
) {
5175 if (splice_grow_spd(pipe
, &spd
)) {
5180 if (*ppos
& (PAGE_SIZE
- 1)) {
5185 if (len
& (PAGE_SIZE
- 1)) {
5186 if (len
< PAGE_SIZE
) {
5194 trace_access_lock(iter
->cpu_file
);
5195 entries
= ring_buffer_entries_cpu(iter
->trace_buffer
->buffer
, iter
->cpu_file
);
5197 for (i
= 0; i
< pipe
->buffers
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
5201 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
5206 ref
->buffer
= iter
->trace_buffer
->buffer
;
5207 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
, iter
->cpu_file
);
5213 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
5214 len
, iter
->cpu_file
, 1);
5216 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
5222 * zero out any left over data, this is going to
5225 size
= ring_buffer_page_len(ref
->page
);
5226 if (size
< PAGE_SIZE
)
5227 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
5229 page
= virt_to_page(ref
->page
);
5231 spd
.pages
[i
] = page
;
5232 spd
.partial
[i
].len
= PAGE_SIZE
;
5233 spd
.partial
[i
].offset
= 0;
5234 spd
.partial
[i
].private = (unsigned long)ref
;
5238 entries
= ring_buffer_entries_cpu(iter
->trace_buffer
->buffer
, iter
->cpu_file
);
5241 trace_access_unlock(iter
->cpu_file
);
5244 /* did we read anything? */
5245 if (!spd
.nr_pages
) {
5246 if ((file
->f_flags
& O_NONBLOCK
) || (flags
& SPLICE_F_NONBLOCK
)) {
5250 mutex_unlock(&trace_types_lock
);
5251 iter
->trace
->wait_pipe(iter
);
5252 mutex_lock(&trace_types_lock
);
5253 if (signal_pending(current
)) {
5260 ret
= splice_to_pipe(pipe
, &spd
);
5261 splice_shrink_spd(&spd
);
5263 mutex_unlock(&trace_types_lock
);
5268 static const struct file_operations tracing_buffers_fops
= {
5269 .open
= tracing_buffers_open
,
5270 .read
= tracing_buffers_read
,
5271 .poll
= tracing_buffers_poll
,
5272 .release
= tracing_buffers_release
,
5273 .splice_read
= tracing_buffers_splice_read
,
5274 .llseek
= no_llseek
,
5278 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
5279 size_t count
, loff_t
*ppos
)
5281 struct trace_cpu
*tc
= filp
->private_data
;
5282 struct trace_array
*tr
= tc
->tr
;
5283 struct trace_buffer
*trace_buf
= &tr
->trace_buffer
;
5284 struct trace_seq
*s
;
5286 unsigned long long t
;
5287 unsigned long usec_rem
;
5290 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
5296 cnt
= ring_buffer_entries_cpu(trace_buf
->buffer
, cpu
);
5297 trace_seq_printf(s
, "entries: %ld\n", cnt
);
5299 cnt
= ring_buffer_overrun_cpu(trace_buf
->buffer
, cpu
);
5300 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
5302 cnt
= ring_buffer_commit_overrun_cpu(trace_buf
->buffer
, cpu
);
5303 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
5305 cnt
= ring_buffer_bytes_cpu(trace_buf
->buffer
, cpu
);
5306 trace_seq_printf(s
, "bytes: %ld\n", cnt
);
5308 if (trace_clocks
[tr
->clock_id
].in_ns
) {
5309 /* local or global for trace_clock */
5310 t
= ns2usecs(ring_buffer_oldest_event_ts(trace_buf
->buffer
, cpu
));
5311 usec_rem
= do_div(t
, USEC_PER_SEC
);
5312 trace_seq_printf(s
, "oldest event ts: %5llu.%06lu\n",
5315 t
= ns2usecs(ring_buffer_time_stamp(trace_buf
->buffer
, cpu
));
5316 usec_rem
= do_div(t
, USEC_PER_SEC
);
5317 trace_seq_printf(s
, "now ts: %5llu.%06lu\n", t
, usec_rem
);
5319 /* counter or tsc mode for trace_clock */
5320 trace_seq_printf(s
, "oldest event ts: %llu\n",
5321 ring_buffer_oldest_event_ts(trace_buf
->buffer
, cpu
));
5323 trace_seq_printf(s
, "now ts: %llu\n",
5324 ring_buffer_time_stamp(trace_buf
->buffer
, cpu
));
5327 cnt
= ring_buffer_dropped_events_cpu(trace_buf
->buffer
, cpu
);
5328 trace_seq_printf(s
, "dropped events: %ld\n", cnt
);
5330 cnt
= ring_buffer_read_events_cpu(trace_buf
->buffer
, cpu
);
5331 trace_seq_printf(s
, "read events: %ld\n", cnt
);
5333 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
5340 static const struct file_operations tracing_stats_fops
= {
5341 .open
= tracing_open_generic_tc
,
5342 .read
= tracing_stats_read
,
5343 .llseek
= generic_file_llseek
,
5344 .release
= tracing_release_generic_tc
,
5347 #ifdef CONFIG_DYNAMIC_FTRACE
5349 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
5355 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
5356 size_t cnt
, loff_t
*ppos
)
5358 static char ftrace_dyn_info_buffer
[1024];
5359 static DEFINE_MUTEX(dyn_info_mutex
);
5360 unsigned long *p
= filp
->private_data
;
5361 char *buf
= ftrace_dyn_info_buffer
;
5362 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
5365 mutex_lock(&dyn_info_mutex
);
5366 r
= sprintf(buf
, "%ld ", *p
);
5368 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
5371 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
5373 mutex_unlock(&dyn_info_mutex
);
5378 static const struct file_operations tracing_dyn_info_fops
= {
5379 .open
= tracing_open_generic
,
5380 .read
= tracing_read_dyn_info
,
5381 .llseek
= generic_file_llseek
,
5383 #endif /* CONFIG_DYNAMIC_FTRACE */
5385 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
5387 ftrace_snapshot(unsigned long ip
, unsigned long parent_ip
, void **data
)
5393 ftrace_count_snapshot(unsigned long ip
, unsigned long parent_ip
, void **data
)
5395 unsigned long *count
= (long *)data
;
5407 ftrace_snapshot_print(struct seq_file
*m
, unsigned long ip
,
5408 struct ftrace_probe_ops
*ops
, void *data
)
5410 long count
= (long)data
;
5412 seq_printf(m
, "%ps:", (void *)ip
);
5414 seq_printf(m
, "snapshot");
5417 seq_printf(m
, ":unlimited\n");
5419 seq_printf(m
, ":count=%ld\n", count
);
5424 static struct ftrace_probe_ops snapshot_probe_ops
= {
5425 .func
= ftrace_snapshot
,
5426 .print
= ftrace_snapshot_print
,
5429 static struct ftrace_probe_ops snapshot_count_probe_ops
= {
5430 .func
= ftrace_count_snapshot
,
5431 .print
= ftrace_snapshot_print
,
5435 ftrace_trace_snapshot_callback(struct ftrace_hash
*hash
,
5436 char *glob
, char *cmd
, char *param
, int enable
)
5438 struct ftrace_probe_ops
*ops
;
5439 void *count
= (void *)-1;
5443 /* hash funcs only work with set_ftrace_filter */
5447 ops
= param
? &snapshot_count_probe_ops
: &snapshot_probe_ops
;
5449 if (glob
[0] == '!') {
5450 unregister_ftrace_function_probe_func(glob
+1, ops
);
5457 number
= strsep(¶m
, ":");
5459 if (!strlen(number
))
5463 * We use the callback data field (which is a pointer)
5466 ret
= kstrtoul(number
, 0, (unsigned long *)&count
);
5471 ret
= register_ftrace_function_probe(glob
, ops
, count
);
5474 alloc_snapshot(&global_trace
);
5476 return ret
< 0 ? ret
: 0;
5479 static struct ftrace_func_command ftrace_snapshot_cmd
= {
5481 .func
= ftrace_trace_snapshot_callback
,
5484 static int register_snapshot_cmd(void)
5486 return register_ftrace_command(&ftrace_snapshot_cmd
);
5489 static inline int register_snapshot_cmd(void) { return 0; }
5490 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
5492 struct dentry
*tracing_init_dentry_tr(struct trace_array
*tr
)
5497 if (!debugfs_initialized())
5500 if (tr
->flags
& TRACE_ARRAY_FL_GLOBAL
)
5501 tr
->dir
= debugfs_create_dir("tracing", NULL
);
5504 pr_warn_once("Could not create debugfs directory 'tracing'\n");
5509 struct dentry
*tracing_init_dentry(void)
5511 return tracing_init_dentry_tr(&global_trace
);
5514 static struct dentry
*tracing_dentry_percpu(struct trace_array
*tr
, int cpu
)
5516 struct dentry
*d_tracer
;
5519 return tr
->percpu_dir
;
5521 d_tracer
= tracing_init_dentry_tr(tr
);
5525 tr
->percpu_dir
= debugfs_create_dir("per_cpu", d_tracer
);
5527 WARN_ONCE(!tr
->percpu_dir
,
5528 "Could not create debugfs directory 'per_cpu/%d'\n", cpu
);
5530 return tr
->percpu_dir
;
5533 static struct dentry
*
5534 trace_create_cpu_file(const char *name
, umode_t mode
, struct dentry
*parent
,
5535 void *data
, long cpu
, const struct file_operations
*fops
)
5537 struct dentry
*ret
= trace_create_file(name
, mode
, parent
, data
, fops
);
5539 if (ret
) /* See tracing_get_cpu() */
5540 ret
->d_inode
->i_cdev
= (void *)(cpu
+ 1);
5545 tracing_init_debugfs_percpu(struct trace_array
*tr
, long cpu
)
5547 struct trace_array_cpu
*data
= per_cpu_ptr(tr
->trace_buffer
.data
, cpu
);
5548 struct dentry
*d_percpu
= tracing_dentry_percpu(tr
, cpu
);
5549 struct dentry
*d_cpu
;
5550 char cpu_dir
[30]; /* 30 characters should be more than enough */
5555 snprintf(cpu_dir
, 30, "cpu%ld", cpu
);
5556 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
5558 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
5562 /* per cpu trace_pipe */
5563 trace_create_cpu_file("trace_pipe", 0444, d_cpu
,
5564 tr
, cpu
, &tracing_pipe_fops
);
5567 trace_create_cpu_file("trace", 0644, d_cpu
,
5568 &data
->trace_cpu
, cpu
, &tracing_fops
);
5570 trace_create_cpu_file("trace_pipe_raw", 0444, d_cpu
,
5571 &data
->trace_cpu
, cpu
, &tracing_buffers_fops
);
5573 trace_create_cpu_file("stats", 0444, d_cpu
,
5574 &data
->trace_cpu
, cpu
, &tracing_stats_fops
);
5576 trace_create_cpu_file("buffer_size_kb", 0444, d_cpu
,
5577 &data
->trace_cpu
, cpu
, &tracing_entries_fops
);
5579 #ifdef CONFIG_TRACER_SNAPSHOT
5580 trace_create_cpu_file("snapshot", 0644, d_cpu
,
5581 &data
->trace_cpu
, cpu
, &snapshot_fops
);
5583 trace_create_cpu_file("snapshot_raw", 0444, d_cpu
,
5584 &data
->trace_cpu
, cpu
, &snapshot_raw_fops
);
5588 #ifdef CONFIG_FTRACE_SELFTEST
5589 /* Let selftest have access to static functions in this file */
5590 #include "trace_selftest.c"
5593 struct trace_option_dentry
{
5594 struct tracer_opt
*opt
;
5595 struct tracer_flags
*flags
;
5596 struct trace_array
*tr
;
5597 struct dentry
*entry
;
5601 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
5604 struct trace_option_dentry
*topt
= filp
->private_data
;
5607 if (topt
->flags
->val
& topt
->opt
->bit
)
5612 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
5616 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
5619 struct trace_option_dentry
*topt
= filp
->private_data
;
5623 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
5627 if (val
!= 0 && val
!= 1)
5630 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
5631 mutex_lock(&trace_types_lock
);
5632 ret
= __set_tracer_option(topt
->tr
->current_trace
, topt
->flags
,
5634 mutex_unlock(&trace_types_lock
);
5645 static const struct file_operations trace_options_fops
= {
5646 .open
= tracing_open_generic
,
5647 .read
= trace_options_read
,
5648 .write
= trace_options_write
,
5649 .llseek
= generic_file_llseek
,
5653 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
5656 long index
= (long)filp
->private_data
;
5659 if (trace_flags
& (1 << index
))
5664 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
5668 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
5671 struct trace_array
*tr
= &global_trace
;
5672 long index
= (long)filp
->private_data
;
5676 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
5680 if (val
!= 0 && val
!= 1)
5683 mutex_lock(&trace_types_lock
);
5684 ret
= set_tracer_flag(tr
, 1 << index
, val
);
5685 mutex_unlock(&trace_types_lock
);
5695 static const struct file_operations trace_options_core_fops
= {
5696 .open
= tracing_open_generic
,
5697 .read
= trace_options_core_read
,
5698 .write
= trace_options_core_write
,
5699 .llseek
= generic_file_llseek
,
5702 struct dentry
*trace_create_file(const char *name
,
5704 struct dentry
*parent
,
5706 const struct file_operations
*fops
)
5710 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
5712 pr_warning("Could not create debugfs '%s' entry\n", name
);
5718 static struct dentry
*trace_options_init_dentry(struct trace_array
*tr
)
5720 struct dentry
*d_tracer
;
5725 d_tracer
= tracing_init_dentry_tr(tr
);
5729 tr
->options
= debugfs_create_dir("options", d_tracer
);
5731 pr_warning("Could not create debugfs directory 'options'\n");
5739 create_trace_option_file(struct trace_array
*tr
,
5740 struct trace_option_dentry
*topt
,
5741 struct tracer_flags
*flags
,
5742 struct tracer_opt
*opt
)
5744 struct dentry
*t_options
;
5746 t_options
= trace_options_init_dentry(tr
);
5750 topt
->flags
= flags
;
5754 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
5755 &trace_options_fops
);
5759 static struct trace_option_dentry
*
5760 create_trace_option_files(struct trace_array
*tr
, struct tracer
*tracer
)
5762 struct trace_option_dentry
*topts
;
5763 struct tracer_flags
*flags
;
5764 struct tracer_opt
*opts
;
5770 flags
= tracer
->flags
;
5772 if (!flags
|| !flags
->opts
)
5777 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
5780 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
5784 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
5785 create_trace_option_file(tr
, &topts
[cnt
], flags
,
5792 destroy_trace_option_files(struct trace_option_dentry
*topts
)
5799 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
5800 if (topts
[cnt
].entry
)
5801 debugfs_remove(topts
[cnt
].entry
);
5807 static struct dentry
*
5808 create_trace_option_core_file(struct trace_array
*tr
,
5809 const char *option
, long index
)
5811 struct dentry
*t_options
;
5813 t_options
= trace_options_init_dentry(tr
);
5817 return trace_create_file(option
, 0644, t_options
, (void *)index
,
5818 &trace_options_core_fops
);
5821 static __init
void create_trace_options_dir(struct trace_array
*tr
)
5823 struct dentry
*t_options
;
5826 t_options
= trace_options_init_dentry(tr
);
5830 for (i
= 0; trace_options
[i
]; i
++)
5831 create_trace_option_core_file(tr
, trace_options
[i
], i
);
5835 rb_simple_read(struct file
*filp
, char __user
*ubuf
,
5836 size_t cnt
, loff_t
*ppos
)
5838 struct trace_array
*tr
= filp
->private_data
;
5842 r
= tracer_tracing_is_on(tr
);
5843 r
= sprintf(buf
, "%d\n", r
);
5845 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
5849 rb_simple_write(struct file
*filp
, const char __user
*ubuf
,
5850 size_t cnt
, loff_t
*ppos
)
5852 struct trace_array
*tr
= filp
->private_data
;
5853 struct ring_buffer
*buffer
= tr
->trace_buffer
.buffer
;
5857 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
5862 mutex_lock(&trace_types_lock
);
5864 tracer_tracing_on(tr
);
5865 if (tr
->current_trace
->start
)
5866 tr
->current_trace
->start(tr
);
5868 tracer_tracing_off(tr
);
5869 if (tr
->current_trace
->stop
)
5870 tr
->current_trace
->stop(tr
);
5872 mutex_unlock(&trace_types_lock
);
5880 static const struct file_operations rb_simple_fops
= {
5881 .open
= tracing_open_generic_tr
,
5882 .read
= rb_simple_read
,
5883 .write
= rb_simple_write
,
5884 .release
= tracing_release_generic_tr
,
5885 .llseek
= default_llseek
,
5888 struct dentry
*trace_instance_dir
;
5891 init_tracer_debugfs(struct trace_array
*tr
, struct dentry
*d_tracer
);
5893 static void init_trace_buffers(struct trace_array
*tr
, struct trace_buffer
*buf
)
5897 for_each_tracing_cpu(cpu
) {
5898 memset(per_cpu_ptr(buf
->data
, cpu
), 0, sizeof(struct trace_array_cpu
));
5899 per_cpu_ptr(buf
->data
, cpu
)->trace_cpu
.cpu
= cpu
;
5900 per_cpu_ptr(buf
->data
, cpu
)->trace_cpu
.tr
= tr
;
5905 allocate_trace_buffer(struct trace_array
*tr
, struct trace_buffer
*buf
, int size
)
5907 enum ring_buffer_flags rb_flags
;
5909 rb_flags
= trace_flags
& TRACE_ITER_OVERWRITE
? RB_FL_OVERWRITE
: 0;
5911 buf
->buffer
= ring_buffer_alloc(size
, rb_flags
);
5915 buf
->data
= alloc_percpu(struct trace_array_cpu
);
5917 ring_buffer_free(buf
->buffer
);
5921 init_trace_buffers(tr
, buf
);
5923 /* Allocate the first page for all buffers */
5924 set_buffer_entries(&tr
->trace_buffer
,
5925 ring_buffer_size(tr
->trace_buffer
.buffer
, 0));
5930 static int allocate_trace_buffers(struct trace_array
*tr
, int size
)
5934 ret
= allocate_trace_buffer(tr
, &tr
->trace_buffer
, size
);
5938 #ifdef CONFIG_TRACER_MAX_TRACE
5939 ret
= allocate_trace_buffer(tr
, &tr
->max_buffer
,
5940 allocate_snapshot
? size
: 1);
5942 ring_buffer_free(tr
->trace_buffer
.buffer
);
5943 free_percpu(tr
->trace_buffer
.data
);
5946 tr
->allocated_snapshot
= allocate_snapshot
;
5949 * Only the top level trace array gets its snapshot allocated
5950 * from the kernel command line.
5952 allocate_snapshot
= false;
5957 static int new_instance_create(const char *name
)
5959 struct trace_array
*tr
;
5962 mutex_lock(&trace_types_lock
);
5965 list_for_each_entry(tr
, &ftrace_trace_arrays
, list
) {
5966 if (tr
->name
&& strcmp(tr
->name
, name
) == 0)
5971 tr
= kzalloc(sizeof(*tr
), GFP_KERNEL
);
5975 tr
->name
= kstrdup(name
, GFP_KERNEL
);
5979 raw_spin_lock_init(&tr
->start_lock
);
5981 tr
->current_trace
= &nop_trace
;
5983 INIT_LIST_HEAD(&tr
->systems
);
5984 INIT_LIST_HEAD(&tr
->events
);
5986 if (allocate_trace_buffers(tr
, trace_buf_size
) < 0)
5989 /* Holder for file callbacks */
5990 tr
->trace_cpu
.cpu
= RING_BUFFER_ALL_CPUS
;
5991 tr
->trace_cpu
.tr
= tr
;
5993 tr
->dir
= debugfs_create_dir(name
, trace_instance_dir
);
5997 ret
= event_trace_add_tracer(tr
->dir
, tr
);
5999 debugfs_remove_recursive(tr
->dir
);
6003 init_tracer_debugfs(tr
, tr
->dir
);
6005 list_add(&tr
->list
, &ftrace_trace_arrays
);
6007 mutex_unlock(&trace_types_lock
);
6012 if (tr
->trace_buffer
.buffer
)
6013 ring_buffer_free(tr
->trace_buffer
.buffer
);
6018 mutex_unlock(&trace_types_lock
);
6024 static int instance_delete(const char *name
)
6026 struct trace_array
*tr
;
6030 mutex_lock(&trace_types_lock
);
6033 list_for_each_entry(tr
, &ftrace_trace_arrays
, list
) {
6034 if (tr
->name
&& strcmp(tr
->name
, name
) == 0) {
6046 list_del(&tr
->list
);
6048 event_trace_del_tracer(tr
);
6049 debugfs_remove_recursive(tr
->dir
);
6050 free_percpu(tr
->trace_buffer
.data
);
6051 ring_buffer_free(tr
->trace_buffer
.buffer
);
6059 mutex_unlock(&trace_types_lock
);
6064 static int instance_mkdir (struct inode
*inode
, struct dentry
*dentry
, umode_t mode
)
6066 struct dentry
*parent
;
6069 /* Paranoid: Make sure the parent is the "instances" directory */
6070 parent
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_alias
);
6071 if (WARN_ON_ONCE(parent
!= trace_instance_dir
))
6075 * The inode mutex is locked, but debugfs_create_dir() will also
6076 * take the mutex. As the instances directory can not be destroyed
6077 * or changed in any other way, it is safe to unlock it, and
6078 * let the dentry try. If two users try to make the same dir at
6079 * the same time, then the new_instance_create() will determine the
6082 mutex_unlock(&inode
->i_mutex
);
6084 ret
= new_instance_create(dentry
->d_iname
);
6086 mutex_lock(&inode
->i_mutex
);
6091 static int instance_rmdir(struct inode
*inode
, struct dentry
*dentry
)
6093 struct dentry
*parent
;
6096 /* Paranoid: Make sure the parent is the "instances" directory */
6097 parent
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_alias
);
6098 if (WARN_ON_ONCE(parent
!= trace_instance_dir
))
6101 /* The caller did a dget() on dentry */
6102 mutex_unlock(&dentry
->d_inode
->i_mutex
);
6105 * The inode mutex is locked, but debugfs_create_dir() will also
6106 * take the mutex. As the instances directory can not be destroyed
6107 * or changed in any other way, it is safe to unlock it, and
6108 * let the dentry try. If two users try to make the same dir at
6109 * the same time, then the instance_delete() will determine the
6112 mutex_unlock(&inode
->i_mutex
);
6114 ret
= instance_delete(dentry
->d_iname
);
6116 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_PARENT
);
6117 mutex_lock(&dentry
->d_inode
->i_mutex
);
6122 static const struct inode_operations instance_dir_inode_operations
= {
6123 .lookup
= simple_lookup
,
6124 .mkdir
= instance_mkdir
,
6125 .rmdir
= instance_rmdir
,
6128 static __init
void create_trace_instances(struct dentry
*d_tracer
)
6130 trace_instance_dir
= debugfs_create_dir("instances", d_tracer
);
6131 if (WARN_ON(!trace_instance_dir
))
6134 /* Hijack the dir inode operations, to allow mkdir */
6135 trace_instance_dir
->d_inode
->i_op
= &instance_dir_inode_operations
;
6139 init_tracer_debugfs(struct trace_array
*tr
, struct dentry
*d_tracer
)
6143 trace_create_file("trace_options", 0644, d_tracer
,
6144 tr
, &tracing_iter_fops
);
6146 trace_create_file("trace", 0644, d_tracer
,
6147 (void *)&tr
->trace_cpu
, &tracing_fops
);
6149 trace_create_file("trace_pipe", 0444, d_tracer
,
6150 tr
, &tracing_pipe_fops
);
6152 trace_create_file("buffer_size_kb", 0644, d_tracer
,
6153 (void *)&tr
->trace_cpu
, &tracing_entries_fops
);
6155 trace_create_file("buffer_total_size_kb", 0444, d_tracer
,
6156 tr
, &tracing_total_entries_fops
);
6158 trace_create_file("free_buffer", 0644, d_tracer
,
6159 tr
, &tracing_free_buffer_fops
);
6161 trace_create_file("trace_marker", 0220, d_tracer
,
6162 tr
, &tracing_mark_fops
);
6164 trace_create_file("trace_clock", 0644, d_tracer
, tr
,
6167 trace_create_file("tracing_on", 0644, d_tracer
,
6168 tr
, &rb_simple_fops
);
6170 #ifdef CONFIG_TRACER_SNAPSHOT
6171 trace_create_file("snapshot", 0644, d_tracer
,
6172 (void *)&tr
->trace_cpu
, &snapshot_fops
);
6175 for_each_tracing_cpu(cpu
)
6176 tracing_init_debugfs_percpu(tr
, cpu
);
6180 static __init
int tracer_init_debugfs(void)
6182 struct dentry
*d_tracer
;
6184 trace_access_lock_init();
6186 d_tracer
= tracing_init_dentry();
6190 init_tracer_debugfs(&global_trace
, d_tracer
);
6192 trace_create_file("tracing_cpumask", 0644, d_tracer
,
6193 &global_trace
, &tracing_cpumask_fops
);
6195 trace_create_file("available_tracers", 0444, d_tracer
,
6196 &global_trace
, &show_traces_fops
);
6198 trace_create_file("current_tracer", 0644, d_tracer
,
6199 &global_trace
, &set_tracer_fops
);
6201 #ifdef CONFIG_TRACER_MAX_TRACE
6202 trace_create_file("tracing_max_latency", 0644, d_tracer
,
6203 &tracing_max_latency
, &tracing_max_lat_fops
);
6206 trace_create_file("tracing_thresh", 0644, d_tracer
,
6207 &tracing_thresh
, &tracing_max_lat_fops
);
6209 trace_create_file("README", 0444, d_tracer
,
6210 NULL
, &tracing_readme_fops
);
6212 trace_create_file("saved_cmdlines", 0444, d_tracer
,
6213 NULL
, &tracing_saved_cmdlines_fops
);
6215 #ifdef CONFIG_DYNAMIC_FTRACE
6216 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
6217 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
6220 create_trace_instances(d_tracer
);
6222 create_trace_options_dir(&global_trace
);
6227 static int trace_panic_handler(struct notifier_block
*this,
6228 unsigned long event
, void *unused
)
6230 if (ftrace_dump_on_oops
)
6231 ftrace_dump(ftrace_dump_on_oops
);
6235 static struct notifier_block trace_panic_notifier
= {
6236 .notifier_call
= trace_panic_handler
,
6238 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
6241 static int trace_die_handler(struct notifier_block
*self
,
6247 if (ftrace_dump_on_oops
)
6248 ftrace_dump(ftrace_dump_on_oops
);
6256 static struct notifier_block trace_die_notifier
= {
6257 .notifier_call
= trace_die_handler
,
6262 * printk is set to max of 1024, we really don't need it that big.
6263 * Nothing should be printing 1000 characters anyway.
6265 #define TRACE_MAX_PRINT 1000
6268 * Define here KERN_TRACE so that we have one place to modify
6269 * it if we decide to change what log level the ftrace dump
6272 #define KERN_TRACE KERN_EMERG
6275 trace_printk_seq(struct trace_seq
*s
)
6277 /* Probably should print a warning here. */
6278 if (s
->len
>= TRACE_MAX_PRINT
)
6279 s
->len
= TRACE_MAX_PRINT
;
6281 /* should be zero ended, but we are paranoid. */
6282 s
->buffer
[s
->len
] = 0;
6284 printk(KERN_TRACE
"%s", s
->buffer
);
6289 void trace_init_global_iter(struct trace_iterator
*iter
)
6291 iter
->tr
= &global_trace
;
6292 iter
->trace
= iter
->tr
->current_trace
;
6293 iter
->cpu_file
= RING_BUFFER_ALL_CPUS
;
6294 iter
->trace_buffer
= &global_trace
.trace_buffer
;
6297 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode
)
6299 /* use static because iter can be a bit big for the stack */
6300 static struct trace_iterator iter
;
6301 static atomic_t dump_running
;
6302 unsigned int old_userobj
;
6303 unsigned long flags
;
6306 /* Only allow one dump user at a time. */
6307 if (atomic_inc_return(&dump_running
) != 1) {
6308 atomic_dec(&dump_running
);
6313 * Always turn off tracing when we dump.
6314 * We don't need to show trace output of what happens
6315 * between multiple crashes.
6317 * If the user does a sysrq-z, then they can re-enable
6318 * tracing with echo 1 > tracing_on.
6322 local_irq_save(flags
);
6324 /* Simulate the iterator */
6325 trace_init_global_iter(&iter
);
6327 for_each_tracing_cpu(cpu
) {
6328 atomic_inc(&per_cpu_ptr(iter
.tr
->trace_buffer
.data
, cpu
)->disabled
);
6331 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
6333 /* don't look at user memory in panic mode */
6334 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
6336 switch (oops_dump_mode
) {
6338 iter
.cpu_file
= RING_BUFFER_ALL_CPUS
;
6341 iter
.cpu_file
= raw_smp_processor_id();
6346 printk(KERN_TRACE
"Bad dumping mode, switching to all CPUs dump\n");
6347 iter
.cpu_file
= RING_BUFFER_ALL_CPUS
;
6350 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
6352 /* Did function tracer already get disabled? */
6353 if (ftrace_is_dead()) {
6354 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
6355 printk("# MAY BE MISSING FUNCTION EVENTS\n");
6359 * We need to stop all tracing on all CPUS to read the
6360 * the next buffer. This is a bit expensive, but is
6361 * not done often. We fill all what we can read,
6362 * and then release the locks again.
6365 while (!trace_empty(&iter
)) {
6368 printk(KERN_TRACE
"---------------------------------\n");
6372 /* reset all but tr, trace, and overruns */
6373 memset(&iter
.seq
, 0,
6374 sizeof(struct trace_iterator
) -
6375 offsetof(struct trace_iterator
, seq
));
6376 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
6379 if (trace_find_next_entry_inc(&iter
) != NULL
) {
6382 ret
= print_trace_line(&iter
);
6383 if (ret
!= TRACE_TYPE_NO_CONSUME
)
6384 trace_consume(&iter
);
6386 touch_nmi_watchdog();
6388 trace_printk_seq(&iter
.seq
);
6392 printk(KERN_TRACE
" (ftrace buffer empty)\n");
6394 printk(KERN_TRACE
"---------------------------------\n");
6397 trace_flags
|= old_userobj
;
6399 for_each_tracing_cpu(cpu
) {
6400 atomic_dec(&per_cpu_ptr(iter
.trace_buffer
->data
, cpu
)->disabled
);
6402 atomic_dec(&dump_running
);
6403 local_irq_restore(flags
);
6405 EXPORT_SYMBOL_GPL(ftrace_dump
);
6407 __init
static int tracer_alloc_buffers(void)
6413 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
6416 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
6417 goto out_free_buffer_mask
;
6419 /* Only allocate trace_printk buffers if a trace_printk exists */
6420 if (__stop___trace_bprintk_fmt
!= __start___trace_bprintk_fmt
)
6421 /* Must be called before global_trace.buffer is allocated */
6422 trace_printk_init_buffers();
6424 /* To save memory, keep the ring buffer size to its minimum */
6425 if (ring_buffer_expanded
)
6426 ring_buf_size
= trace_buf_size
;
6430 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
6431 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
6433 raw_spin_lock_init(&global_trace
.start_lock
);
6435 /* TODO: make the number of buffers hot pluggable with CPUS */
6436 if (allocate_trace_buffers(&global_trace
, ring_buf_size
) < 0) {
6437 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
6439 goto out_free_cpumask
;
6442 if (global_trace
.buffer_disabled
)
6445 trace_init_cmdlines();
6448 * register_tracer() might reference current_trace, so it
6449 * needs to be set before we register anything. This is
6450 * just a bootstrap of current_trace anyway.
6452 global_trace
.current_trace
= &nop_trace
;
6454 register_tracer(&nop_trace
);
6456 /* All seems OK, enable tracing */
6457 tracing_disabled
= 0;
6459 atomic_notifier_chain_register(&panic_notifier_list
,
6460 &trace_panic_notifier
);
6462 register_die_notifier(&trace_die_notifier
);
6464 global_trace
.flags
= TRACE_ARRAY_FL_GLOBAL
;
6466 /* Holder for file callbacks */
6467 global_trace
.trace_cpu
.cpu
= RING_BUFFER_ALL_CPUS
;
6468 global_trace
.trace_cpu
.tr
= &global_trace
;
6470 INIT_LIST_HEAD(&global_trace
.systems
);
6471 INIT_LIST_HEAD(&global_trace
.events
);
6472 list_add(&global_trace
.list
, &ftrace_trace_arrays
);
6474 while (trace_boot_options
) {
6477 option
= strsep(&trace_boot_options
, ",");
6478 trace_set_options(&global_trace
, option
);
6481 register_snapshot_cmd();
6486 free_percpu(global_trace
.trace_buffer
.data
);
6487 #ifdef CONFIG_TRACER_MAX_TRACE
6488 free_percpu(global_trace
.max_buffer
.data
);
6490 free_cpumask_var(tracing_cpumask
);
6491 out_free_buffer_mask
:
6492 free_cpumask_var(tracing_buffer_mask
);
6497 __init
static int clear_boot_tracer(void)
6500 * The default tracer at boot buffer is an init section.
6501 * This function is called in lateinit. If we did not
6502 * find the boot tracer, then clear it out, to prevent
6503 * later registration from accessing the buffer that is
6504 * about to be freed.
6506 if (!default_bootup_tracer
)
6509 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
6510 default_bootup_tracer
);
6511 default_bootup_tracer
= NULL
;
6516 early_initcall(tracer_alloc_buffers
);
6517 fs_initcall(tracer_init_debugfs
);
6518 late_initcall(clear_boot_tracer
);