2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Based on code from the latency_tracer, that is:
11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 Nadia Yvette Chambers
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/irq_work.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/rwsem.h>
36 #include <linux/slab.h>
37 #include <linux/ctype.h>
38 #include <linux/init.h>
39 #include <linux/poll.h>
40 #include <linux/nmi.h>
44 #include "trace_output.h"
47 * On boot up, the ring buffer is set to the minimum size, so that
48 * we do not waste memory on systems that are not using tracing.
50 int ring_buffer_expanded
;
53 * We need to change this state when a selftest is running.
54 * A selftest will lurk into the ring-buffer to count the
55 * entries inserted during the selftest although some concurrent
56 * insertions into the ring-buffer such as trace_printk could occurred
57 * at the same time, giving false positive or negative results.
59 static bool __read_mostly tracing_selftest_running
;
62 * If a tracer is running, we do not want to run SELFTEST.
64 bool __read_mostly tracing_selftest_disabled
;
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt
[] = {
71 static struct tracer_flags dummy_tracer_flags
= {
73 .opts
= dummy_tracer_opt
76 static int dummy_set_flag(u32 old_flags
, u32 bit
, int set
)
82 * To prevent the comm cache from being overwritten when no
83 * tracing is active, only save the comm when a trace event
86 static DEFINE_PER_CPU(bool, trace_cmdline_save
);
89 * When a reader is waiting for data, then this variable is
92 static bool trace_wakeup_needed
;
94 static struct irq_work trace_work_wakeup
;
97 * Kill all tracing for good (never come back).
98 * It is initialized to 1 but will turn to zero if the initialization
99 * of the tracer is successful. But that is the only place that sets
102 static int tracing_disabled
= 1;
104 DEFINE_PER_CPU(int, ftrace_cpu_disabled
);
106 cpumask_var_t __read_mostly tracing_buffer_mask
;
109 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
111 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
112 * is set, then ftrace_dump is called. This will output the contents
113 * of the ftrace buffers to the console. This is very useful for
114 * capturing traces that lead to crashes and outputing it to a
117 * It is default off, but you can enable it with either specifying
118 * "ftrace_dump_on_oops" in the kernel command line, or setting
119 * /proc/sys/kernel/ftrace_dump_on_oops
120 * Set 1 if you want to dump buffers of all CPUs
121 * Set 2 if you want to dump the buffer of the CPU that triggered oops
124 enum ftrace_dump_mode ftrace_dump_on_oops
;
126 static int tracing_set_tracer(const char *buf
);
128 #define MAX_TRACER_SIZE 100
129 static char bootup_tracer_buf
[MAX_TRACER_SIZE
] __initdata
;
130 static char *default_bootup_tracer
;
132 static int __init
set_cmdline_ftrace(char *str
)
134 strncpy(bootup_tracer_buf
, str
, MAX_TRACER_SIZE
);
135 default_bootup_tracer
= bootup_tracer_buf
;
136 /* We are using ftrace early, expand it */
137 ring_buffer_expanded
= 1;
140 __setup("ftrace=", set_cmdline_ftrace
);
142 static int __init
set_ftrace_dump_on_oops(char *str
)
144 if (*str
++ != '=' || !*str
) {
145 ftrace_dump_on_oops
= DUMP_ALL
;
149 if (!strcmp("orig_cpu", str
)) {
150 ftrace_dump_on_oops
= DUMP_ORIG
;
156 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops
);
159 static char trace_boot_options_buf
[MAX_TRACER_SIZE
] __initdata
;
160 static char *trace_boot_options __initdata
;
162 static int __init
set_trace_boot_options(char *str
)
164 strncpy(trace_boot_options_buf
, str
, MAX_TRACER_SIZE
);
165 trace_boot_options
= trace_boot_options_buf
;
168 __setup("trace_options=", set_trace_boot_options
);
170 unsigned long long ns2usecs(cycle_t nsec
)
178 * The global_trace is the descriptor that holds the tracing
179 * buffers for the live tracing. For each CPU, it contains
180 * a link list of pages that will store trace entries. The
181 * page descriptor of the pages in the memory is used to hold
182 * the link list by linking the lru item in the page descriptor
183 * to each of the pages in the buffer per CPU.
185 * For each active CPU there is a data field that holds the
186 * pages for the buffer for that CPU. Each CPU has the same number
187 * of pages allocated for its buffer.
189 static struct trace_array global_trace
;
191 static DEFINE_PER_CPU(struct trace_array_cpu
, global_trace_cpu
);
193 int filter_current_check_discard(struct ring_buffer
*buffer
,
194 struct ftrace_event_call
*call
, void *rec
,
195 struct ring_buffer_event
*event
)
197 return filter_check_discard(call
, rec
, buffer
, event
);
199 EXPORT_SYMBOL_GPL(filter_current_check_discard
);
201 cycle_t
ftrace_now(int cpu
)
205 /* Early boot up does not have a buffer yet */
206 if (!global_trace
.buffer
)
207 return trace_clock_local();
209 ts
= ring_buffer_time_stamp(global_trace
.buffer
, cpu
);
210 ring_buffer_normalize_time_stamp(global_trace
.buffer
, cpu
, &ts
);
216 * The max_tr is used to snapshot the global_trace when a maximum
217 * latency is reached. Some tracers will use this to store a maximum
218 * trace while it continues examining live traces.
220 * The buffers for the max_tr are set up the same as the global_trace.
221 * When a snapshot is taken, the link list of the max_tr is swapped
222 * with the link list of the global_trace and the buffers are reset for
223 * the global_trace so the tracing can continue.
225 static struct trace_array max_tr
;
227 static DEFINE_PER_CPU(struct trace_array_cpu
, max_tr_data
);
229 int tracing_is_enabled(void)
231 return tracing_is_on();
235 * trace_buf_size is the size in bytes that is allocated
236 * for a buffer. Note, the number of bytes is always rounded
239 * This number is purposely set to a low number of 16384.
240 * If the dump on oops happens, it will be much appreciated
241 * to not have to wait for all that output. Anyway this can be
242 * boot time and run time configurable.
244 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
246 static unsigned long trace_buf_size
= TRACE_BUF_SIZE_DEFAULT
;
248 /* trace_types holds a link list of available tracers. */
249 static struct tracer
*trace_types __read_mostly
;
251 /* current_trace points to the tracer that is currently active */
252 static struct tracer
*current_trace __read_mostly
;
255 * trace_types_lock is used to protect the trace_types list.
257 static DEFINE_MUTEX(trace_types_lock
);
260 * serialize the access of the ring buffer
262 * ring buffer serializes readers, but it is low level protection.
263 * The validity of the events (which returns by ring_buffer_peek() ..etc)
264 * are not protected by ring buffer.
266 * The content of events may become garbage if we allow other process consumes
267 * these events concurrently:
268 * A) the page of the consumed events may become a normal page
269 * (not reader page) in ring buffer, and this page will be rewrited
270 * by events producer.
271 * B) The page of the consumed events may become a page for splice_read,
272 * and this page will be returned to system.
274 * These primitives allow multi process access to different cpu ring buffer
277 * These primitives don't distinguish read-only and read-consume access.
278 * Multi read-only access are also serialized.
282 static DECLARE_RWSEM(all_cpu_access_lock
);
283 static DEFINE_PER_CPU(struct mutex
, cpu_access_lock
);
285 static inline void trace_access_lock(int cpu
)
287 if (cpu
== TRACE_PIPE_ALL_CPU
) {
288 /* gain it for accessing the whole ring buffer. */
289 down_write(&all_cpu_access_lock
);
291 /* gain it for accessing a cpu ring buffer. */
293 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
294 down_read(&all_cpu_access_lock
);
296 /* Secondly block other access to this @cpu ring buffer. */
297 mutex_lock(&per_cpu(cpu_access_lock
, cpu
));
301 static inline void trace_access_unlock(int cpu
)
303 if (cpu
== TRACE_PIPE_ALL_CPU
) {
304 up_write(&all_cpu_access_lock
);
306 mutex_unlock(&per_cpu(cpu_access_lock
, cpu
));
307 up_read(&all_cpu_access_lock
);
311 static inline void trace_access_lock_init(void)
315 for_each_possible_cpu(cpu
)
316 mutex_init(&per_cpu(cpu_access_lock
, cpu
));
321 static DEFINE_MUTEX(access_lock
);
323 static inline void trace_access_lock(int cpu
)
326 mutex_lock(&access_lock
);
329 static inline void trace_access_unlock(int cpu
)
332 mutex_unlock(&access_lock
);
335 static inline void trace_access_lock_init(void)
341 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
342 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
344 /* trace_flags holds trace_options default values */
345 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
| TRACE_ITER_PRINTK
|
346 TRACE_ITER_ANNOTATE
| TRACE_ITER_CONTEXT_INFO
| TRACE_ITER_SLEEP_TIME
|
347 TRACE_ITER_GRAPH_TIME
| TRACE_ITER_RECORD_CMD
| TRACE_ITER_OVERWRITE
|
348 TRACE_ITER_IRQ_INFO
| TRACE_ITER_MARKERS
;
350 static int trace_stop_count
;
351 static DEFINE_RAW_SPINLOCK(tracing_start_lock
);
354 * trace_wake_up - wake up tasks waiting for trace input
356 * Schedules a delayed work to wake up any task that is blocked on the
357 * trace_wait queue. These is used with trace_poll for tasks polling the
360 static void trace_wake_up(struct irq_work
*work
)
362 wake_up_all(&trace_wait
);
367 * tracing_on - enable tracing buffers
369 * This function enables tracing buffers that may have been
370 * disabled with tracing_off.
372 void tracing_on(void)
374 if (global_trace
.buffer
)
375 ring_buffer_record_on(global_trace
.buffer
);
377 * This flag is only looked at when buffers haven't been
378 * allocated yet. We don't really care about the race
379 * between setting this flag and actually turning
382 global_trace
.buffer_disabled
= 0;
384 EXPORT_SYMBOL_GPL(tracing_on
);
387 * tracing_off - turn off tracing buffers
389 * This function stops the tracing buffers from recording data.
390 * It does not disable any overhead the tracers themselves may
391 * be causing. This function simply causes all recording to
392 * the ring buffers to fail.
394 void tracing_off(void)
396 if (global_trace
.buffer
)
397 ring_buffer_record_off(global_trace
.buffer
);
399 * This flag is only looked at when buffers haven't been
400 * allocated yet. We don't really care about the race
401 * between setting this flag and actually turning
404 global_trace
.buffer_disabled
= 1;
406 EXPORT_SYMBOL_GPL(tracing_off
);
409 * tracing_is_on - show state of ring buffers enabled
411 int tracing_is_on(void)
413 if (global_trace
.buffer
)
414 return ring_buffer_record_is_on(global_trace
.buffer
);
415 return !global_trace
.buffer_disabled
;
417 EXPORT_SYMBOL_GPL(tracing_is_on
);
419 static int __init
set_buf_size(char *str
)
421 unsigned long buf_size
;
425 buf_size
= memparse(str
, &str
);
426 /* nr_entries can not be zero */
429 trace_buf_size
= buf_size
;
432 __setup("trace_buf_size=", set_buf_size
);
434 static int __init
set_tracing_thresh(char *str
)
436 unsigned long threshold
;
441 ret
= kstrtoul(str
, 0, &threshold
);
444 tracing_thresh
= threshold
* 1000;
447 __setup("tracing_thresh=", set_tracing_thresh
);
449 unsigned long nsecs_to_usecs(unsigned long nsecs
)
454 /* These must match the bit postions in trace_iterator_flags */
455 static const char *trace_options
[] = {
487 int in_ns
; /* is this clock in nanoseconds? */
489 { trace_clock_local
, "local", 1 },
490 { trace_clock_global
, "global", 1 },
491 { trace_clock_counter
, "counter", 0 },
498 * trace_parser_get_init - gets the buffer for trace parser
500 int trace_parser_get_init(struct trace_parser
*parser
, int size
)
502 memset(parser
, 0, sizeof(*parser
));
504 parser
->buffer
= kmalloc(size
, GFP_KERNEL
);
513 * trace_parser_put - frees the buffer for trace parser
515 void trace_parser_put(struct trace_parser
*parser
)
517 kfree(parser
->buffer
);
521 * trace_get_user - reads the user input string separated by space
522 * (matched by isspace(ch))
524 * For each string found the 'struct trace_parser' is updated,
525 * and the function returns.
527 * Returns number of bytes read.
529 * See kernel/trace/trace.h for 'struct trace_parser' details.
531 int trace_get_user(struct trace_parser
*parser
, const char __user
*ubuf
,
532 size_t cnt
, loff_t
*ppos
)
539 trace_parser_clear(parser
);
541 ret
= get_user(ch
, ubuf
++);
549 * The parser is not finished with the last write,
550 * continue reading the user input without skipping spaces.
553 /* skip white space */
554 while (cnt
&& isspace(ch
)) {
555 ret
= get_user(ch
, ubuf
++);
562 /* only spaces were written */
572 /* read the non-space input */
573 while (cnt
&& !isspace(ch
)) {
574 if (parser
->idx
< parser
->size
- 1)
575 parser
->buffer
[parser
->idx
++] = ch
;
580 ret
= get_user(ch
, ubuf
++);
587 /* We either got finished input or we have to wait for another call. */
589 parser
->buffer
[parser
->idx
] = 0;
590 parser
->cont
= false;
593 parser
->buffer
[parser
->idx
++] = ch
;
603 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
611 if (s
->len
<= s
->readpos
)
614 len
= s
->len
- s
->readpos
;
617 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
627 static ssize_t
trace_seq_to_buffer(struct trace_seq
*s
, void *buf
, size_t cnt
)
631 if (s
->len
<= s
->readpos
)
634 len
= s
->len
- s
->readpos
;
637 memcpy(buf
, s
->buffer
+ s
->readpos
, cnt
);
644 * ftrace_max_lock is used to protect the swapping of buffers
645 * when taking a max snapshot. The buffers themselves are
646 * protected by per_cpu spinlocks. But the action of the swap
647 * needs its own lock.
649 * This is defined as a arch_spinlock_t in order to help
650 * with performance when lockdep debugging is enabled.
652 * It is also used in other places outside the update_max_tr
653 * so it needs to be defined outside of the
654 * CONFIG_TRACER_MAX_TRACE.
656 static arch_spinlock_t ftrace_max_lock
=
657 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
659 unsigned long __read_mostly tracing_thresh
;
661 #ifdef CONFIG_TRACER_MAX_TRACE
662 unsigned long __read_mostly tracing_max_latency
;
665 * Copy the new maximum trace into the separate maximum-trace
666 * structure. (this way the maximum trace is permanently saved,
667 * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
670 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
672 struct trace_array_cpu
*data
= tr
->data
[cpu
];
673 struct trace_array_cpu
*max_data
;
676 max_tr
.time_start
= data
->preempt_timestamp
;
678 max_data
= max_tr
.data
[cpu
];
679 max_data
->saved_latency
= tracing_max_latency
;
680 max_data
->critical_start
= data
->critical_start
;
681 max_data
->critical_end
= data
->critical_end
;
683 memcpy(max_data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
684 max_data
->pid
= tsk
->pid
;
685 max_data
->uid
= task_uid(tsk
);
686 max_data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
687 max_data
->policy
= tsk
->policy
;
688 max_data
->rt_priority
= tsk
->rt_priority
;
690 /* record this tasks comm */
691 tracing_record_cmdline(tsk
);
695 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
697 * @tsk: the task with the latency
698 * @cpu: The cpu that initiated the trace.
700 * Flip the buffers between the @tr and the max_tr and record information
701 * about which task was the cause of this latency.
704 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
706 struct ring_buffer
*buf
= tr
->buffer
;
708 if (trace_stop_count
)
711 WARN_ON_ONCE(!irqs_disabled());
712 if (!current_trace
->use_max_tr
) {
716 arch_spin_lock(&ftrace_max_lock
);
718 tr
->buffer
= max_tr
.buffer
;
721 __update_max_tr(tr
, tsk
, cpu
);
722 arch_spin_unlock(&ftrace_max_lock
);
726 * update_max_tr_single - only copy one trace over, and reset the rest
728 * @tsk - task with the latency
729 * @cpu - the cpu of the buffer to copy.
731 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
734 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
738 if (trace_stop_count
)
741 WARN_ON_ONCE(!irqs_disabled());
742 if (!current_trace
->use_max_tr
) {
747 arch_spin_lock(&ftrace_max_lock
);
749 ret
= ring_buffer_swap_cpu(max_tr
.buffer
, tr
->buffer
, cpu
);
753 * We failed to swap the buffer due to a commit taking
754 * place on this CPU. We fail to record, but we reset
755 * the max trace buffer (no one writes directly to it)
756 * and flag that it failed.
758 trace_array_printk(&max_tr
, _THIS_IP_
,
759 "Failed to swap buffers due to commit in progress\n");
762 WARN_ON_ONCE(ret
&& ret
!= -EAGAIN
&& ret
!= -EBUSY
);
764 __update_max_tr(tr
, tsk
, cpu
);
765 arch_spin_unlock(&ftrace_max_lock
);
767 #endif /* CONFIG_TRACER_MAX_TRACE */
769 static void default_wait_pipe(struct trace_iterator
*iter
)
773 prepare_to_wait(&trace_wait
, &wait
, TASK_INTERRUPTIBLE
);
776 * The events can happen in critical sections where
777 * checking a work queue can cause deadlocks.
778 * After adding a task to the queue, this flag is set
779 * only to notify events to try to wake up the queue
782 * We don't clear it even if the buffer is no longer
783 * empty. The flag only causes the next event to run
784 * irq_work to do the work queue wake up. The worse
785 * that can happen if we race with !trace_empty() is that
786 * an event will cause an irq_work to try to wake up
789 * There's no reason to protect this flag either, as
790 * the work queue and irq_work logic will do the necessary
791 * synchronization for the wake ups. The only thing
792 * that is necessary is that the wake up happens after
793 * a task has been queued. It's OK for spurious wake ups.
795 trace_wakeup_needed
= true;
797 if (trace_empty(iter
))
800 finish_wait(&trace_wait
, &wait
);
804 * register_tracer - register a tracer with the ftrace system.
805 * @type - the plugin for the tracer
807 * Register a new plugin tracer.
809 int register_tracer(struct tracer
*type
)
815 pr_info("Tracer must have a name\n");
819 if (strlen(type
->name
) >= MAX_TRACER_SIZE
) {
820 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE
);
824 mutex_lock(&trace_types_lock
);
826 tracing_selftest_running
= true;
828 for (t
= trace_types
; t
; t
= t
->next
) {
829 if (strcmp(type
->name
, t
->name
) == 0) {
831 pr_info("Tracer %s already registered\n",
839 type
->set_flag
= &dummy_set_flag
;
841 type
->flags
= &dummy_tracer_flags
;
843 if (!type
->flags
->opts
)
844 type
->flags
->opts
= dummy_tracer_opt
;
845 if (!type
->wait_pipe
)
846 type
->wait_pipe
= default_wait_pipe
;
849 #ifdef CONFIG_FTRACE_STARTUP_TEST
850 if (type
->selftest
&& !tracing_selftest_disabled
) {
851 struct tracer
*saved_tracer
= current_trace
;
852 struct trace_array
*tr
= &global_trace
;
855 * Run a selftest on this tracer.
856 * Here we reset the trace buffer, and set the current
857 * tracer to be this tracer. The tracer can then run some
858 * internal tracing to verify that everything is in order.
859 * If we fail, we do not register this tracer.
861 tracing_reset_online_cpus(tr
);
863 current_trace
= type
;
865 /* If we expanded the buffers, make sure the max is expanded too */
866 if (ring_buffer_expanded
&& type
->use_max_tr
)
867 ring_buffer_resize(max_tr
.buffer
, trace_buf_size
,
868 RING_BUFFER_ALL_CPUS
);
870 /* the test is responsible for initializing and enabling */
871 pr_info("Testing tracer %s: ", type
->name
);
872 ret
= type
->selftest(type
, tr
);
873 /* the test is responsible for resetting too */
874 current_trace
= saved_tracer
;
876 printk(KERN_CONT
"FAILED!\n");
877 /* Add the warning after printing 'FAILED' */
881 /* Only reset on passing, to avoid touching corrupted buffers */
882 tracing_reset_online_cpus(tr
);
884 /* Shrink the max buffer again */
885 if (ring_buffer_expanded
&& type
->use_max_tr
)
886 ring_buffer_resize(max_tr
.buffer
, 1,
887 RING_BUFFER_ALL_CPUS
);
889 printk(KERN_CONT
"PASSED\n");
893 type
->next
= trace_types
;
897 tracing_selftest_running
= false;
898 mutex_unlock(&trace_types_lock
);
900 if (ret
|| !default_bootup_tracer
)
903 if (strncmp(default_bootup_tracer
, type
->name
, MAX_TRACER_SIZE
))
906 printk(KERN_INFO
"Starting tracer '%s'\n", type
->name
);
907 /* Do we want this tracer to start on bootup? */
908 tracing_set_tracer(type
->name
);
909 default_bootup_tracer
= NULL
;
910 /* disable other selftests, since this will break it. */
911 tracing_selftest_disabled
= 1;
912 #ifdef CONFIG_FTRACE_STARTUP_TEST
913 printk(KERN_INFO
"Disabling FTRACE selftests due to running tracer '%s'\n",
921 void tracing_reset(struct trace_array
*tr
, int cpu
)
923 struct ring_buffer
*buffer
= tr
->buffer
;
925 ring_buffer_record_disable(buffer
);
927 /* Make sure all commits have finished */
929 ring_buffer_reset_cpu(buffer
, cpu
);
931 ring_buffer_record_enable(buffer
);
934 void tracing_reset_online_cpus(struct trace_array
*tr
)
936 struct ring_buffer
*buffer
= tr
->buffer
;
939 ring_buffer_record_disable(buffer
);
941 /* Make sure all commits have finished */
944 tr
->time_start
= ftrace_now(tr
->cpu
);
946 for_each_online_cpu(cpu
)
947 ring_buffer_reset_cpu(buffer
, cpu
);
949 ring_buffer_record_enable(buffer
);
952 void tracing_reset_current(int cpu
)
954 tracing_reset(&global_trace
, cpu
);
957 void tracing_reset_current_online_cpus(void)
959 tracing_reset_online_cpus(&global_trace
);
962 #define SAVED_CMDLINES 128
963 #define NO_CMDLINE_MAP UINT_MAX
964 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
965 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
966 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
967 static int cmdline_idx
;
968 static arch_spinlock_t trace_cmdline_lock
= __ARCH_SPIN_LOCK_UNLOCKED
;
970 /* temporary disable recording */
971 static atomic_t trace_record_cmdline_disabled __read_mostly
;
973 static void trace_init_cmdlines(void)
975 memset(&map_pid_to_cmdline
, NO_CMDLINE_MAP
, sizeof(map_pid_to_cmdline
));
976 memset(&map_cmdline_to_pid
, NO_CMDLINE_MAP
, sizeof(map_cmdline_to_pid
));
980 int is_tracing_stopped(void)
982 return trace_stop_count
;
986 * ftrace_off_permanent - disable all ftrace code permanently
988 * This should only be called when a serious anomally has
989 * been detected. This will turn off the function tracing,
990 * ring buffers, and other tracing utilites. It takes no
991 * locks and can be called from any context.
993 void ftrace_off_permanent(void)
995 tracing_disabled
= 1;
997 tracing_off_permanent();
1001 * tracing_start - quick start of the tracer
1003 * If tracing is enabled but was stopped by tracing_stop,
1004 * this will start the tracer back up.
1006 void tracing_start(void)
1008 struct ring_buffer
*buffer
;
1009 unsigned long flags
;
1011 if (tracing_disabled
)
1014 raw_spin_lock_irqsave(&tracing_start_lock
, flags
);
1015 if (--trace_stop_count
) {
1016 if (trace_stop_count
< 0) {
1017 /* Someone screwed up their debugging */
1019 trace_stop_count
= 0;
1024 /* Prevent the buffers from switching */
1025 arch_spin_lock(&ftrace_max_lock
);
1027 buffer
= global_trace
.buffer
;
1029 ring_buffer_record_enable(buffer
);
1031 buffer
= max_tr
.buffer
;
1033 ring_buffer_record_enable(buffer
);
1035 arch_spin_unlock(&ftrace_max_lock
);
1039 raw_spin_unlock_irqrestore(&tracing_start_lock
, flags
);
1043 * tracing_stop - quick stop of the tracer
1045 * Light weight way to stop tracing. Use in conjunction with
1048 void tracing_stop(void)
1050 struct ring_buffer
*buffer
;
1051 unsigned long flags
;
1054 raw_spin_lock_irqsave(&tracing_start_lock
, flags
);
1055 if (trace_stop_count
++)
1058 /* Prevent the buffers from switching */
1059 arch_spin_lock(&ftrace_max_lock
);
1061 buffer
= global_trace
.buffer
;
1063 ring_buffer_record_disable(buffer
);
1065 buffer
= max_tr
.buffer
;
1067 ring_buffer_record_disable(buffer
);
1069 arch_spin_unlock(&ftrace_max_lock
);
1072 raw_spin_unlock_irqrestore(&tracing_start_lock
, flags
);
1075 void trace_stop_cmdline_recording(void);
1077 static void trace_save_cmdline(struct task_struct
*tsk
)
1081 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
1085 * It's not the end of the world if we don't get
1086 * the lock, but we also don't want to spin
1087 * nor do we want to disable interrupts,
1088 * so if we miss here, then better luck next time.
1090 if (!arch_spin_trylock(&trace_cmdline_lock
))
1093 idx
= map_pid_to_cmdline
[tsk
->pid
];
1094 if (idx
== NO_CMDLINE_MAP
) {
1095 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
1098 * Check whether the cmdline buffer at idx has a pid
1099 * mapped. We are going to overwrite that entry so we
1100 * need to clear the map_pid_to_cmdline. Otherwise we
1101 * would read the new comm for the old pid.
1103 pid
= map_cmdline_to_pid
[idx
];
1104 if (pid
!= NO_CMDLINE_MAP
)
1105 map_pid_to_cmdline
[pid
] = NO_CMDLINE_MAP
;
1107 map_cmdline_to_pid
[idx
] = tsk
->pid
;
1108 map_pid_to_cmdline
[tsk
->pid
] = idx
;
1113 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
1115 arch_spin_unlock(&trace_cmdline_lock
);
1118 void trace_find_cmdline(int pid
, char comm
[])
1123 strcpy(comm
, "<idle>");
1127 if (WARN_ON_ONCE(pid
< 0)) {
1128 strcpy(comm
, "<XXX>");
1132 if (pid
> PID_MAX_DEFAULT
) {
1133 strcpy(comm
, "<...>");
1138 arch_spin_lock(&trace_cmdline_lock
);
1139 map
= map_pid_to_cmdline
[pid
];
1140 if (map
!= NO_CMDLINE_MAP
)
1141 strcpy(comm
, saved_cmdlines
[map
]);
1143 strcpy(comm
, "<...>");
1145 arch_spin_unlock(&trace_cmdline_lock
);
1149 void tracing_record_cmdline(struct task_struct
*tsk
)
1151 if (atomic_read(&trace_record_cmdline_disabled
) || !tracing_is_on())
1154 if (!__this_cpu_read(trace_cmdline_save
))
1157 __this_cpu_write(trace_cmdline_save
, false);
1159 trace_save_cmdline(tsk
);
1163 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
,
1166 struct task_struct
*tsk
= current
;
1168 entry
->preempt_count
= pc
& 0xff;
1169 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
1172 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1173 (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
1175 TRACE_FLAG_IRQS_NOSUPPORT
|
1177 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
1178 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
1179 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
1181 EXPORT_SYMBOL_GPL(tracing_generic_entry_update
);
1183 struct ring_buffer_event
*
1184 trace_buffer_lock_reserve(struct ring_buffer
*buffer
,
1187 unsigned long flags
, int pc
)
1189 struct ring_buffer_event
*event
;
1191 event
= ring_buffer_lock_reserve(buffer
, len
);
1192 if (event
!= NULL
) {
1193 struct trace_entry
*ent
= ring_buffer_event_data(event
);
1195 tracing_generic_entry_update(ent
, flags
, pc
);
1203 __buffer_unlock_commit(struct ring_buffer
*buffer
, struct ring_buffer_event
*event
)
1205 __this_cpu_write(trace_cmdline_save
, true);
1206 if (trace_wakeup_needed
) {
1207 trace_wakeup_needed
= false;
1208 /* irq_work_queue() supplies it's own memory barriers */
1209 irq_work_queue(&trace_work_wakeup
);
1211 ring_buffer_unlock_commit(buffer
, event
);
1215 __trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1216 struct ring_buffer_event
*event
,
1217 unsigned long flags
, int pc
)
1219 __buffer_unlock_commit(buffer
, event
);
1221 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1222 ftrace_trace_userstack(buffer
, flags
, pc
);
1225 void trace_buffer_unlock_commit(struct ring_buffer
*buffer
,
1226 struct ring_buffer_event
*event
,
1227 unsigned long flags
, int pc
)
1229 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
);
1231 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit
);
1233 struct ring_buffer_event
*
1234 trace_current_buffer_lock_reserve(struct ring_buffer
**current_rb
,
1235 int type
, unsigned long len
,
1236 unsigned long flags
, int pc
)
1238 *current_rb
= global_trace
.buffer
;
1239 return trace_buffer_lock_reserve(*current_rb
,
1240 type
, len
, flags
, pc
);
1242 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve
);
1244 void trace_current_buffer_unlock_commit(struct ring_buffer
*buffer
,
1245 struct ring_buffer_event
*event
,
1246 unsigned long flags
, int pc
)
1248 __trace_buffer_unlock_commit(buffer
, event
, flags
, pc
);
1250 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit
);
1252 void trace_buffer_unlock_commit_regs(struct ring_buffer
*buffer
,
1253 struct ring_buffer_event
*event
,
1254 unsigned long flags
, int pc
,
1255 struct pt_regs
*regs
)
1257 __buffer_unlock_commit(buffer
, event
);
1259 ftrace_trace_stack_regs(buffer
, flags
, 0, pc
, regs
);
1260 ftrace_trace_userstack(buffer
, flags
, pc
);
1262 EXPORT_SYMBOL_GPL(trace_buffer_unlock_commit_regs
);
1264 void trace_current_buffer_discard_commit(struct ring_buffer
*buffer
,
1265 struct ring_buffer_event
*event
)
1267 ring_buffer_discard_commit(buffer
, event
);
1269 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit
);
1272 trace_function(struct trace_array
*tr
,
1273 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1276 struct ftrace_event_call
*call
= &event_function
;
1277 struct ring_buffer
*buffer
= tr
->buffer
;
1278 struct ring_buffer_event
*event
;
1279 struct ftrace_entry
*entry
;
1281 /* If we are reading the ring buffer, don't trace */
1282 if (unlikely(__this_cpu_read(ftrace_cpu_disabled
)))
1285 event
= trace_buffer_lock_reserve(buffer
, TRACE_FN
, sizeof(*entry
),
1289 entry
= ring_buffer_event_data(event
);
1291 entry
->parent_ip
= parent_ip
;
1293 if (!filter_check_discard(call
, entry
, buffer
, event
))
1294 __buffer_unlock_commit(buffer
, event
);
1298 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1299 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
,
1302 if (likely(!atomic_read(&data
->disabled
)))
1303 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
1306 #ifdef CONFIG_STACKTRACE
1308 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1309 struct ftrace_stack
{
1310 unsigned long calls
[FTRACE_STACK_MAX_ENTRIES
];
1313 static DEFINE_PER_CPU(struct ftrace_stack
, ftrace_stack
);
1314 static DEFINE_PER_CPU(int, ftrace_stack_reserve
);
1316 static void __ftrace_trace_stack(struct ring_buffer
*buffer
,
1317 unsigned long flags
,
1318 int skip
, int pc
, struct pt_regs
*regs
)
1320 struct ftrace_event_call
*call
= &event_kernel_stack
;
1321 struct ring_buffer_event
*event
;
1322 struct stack_entry
*entry
;
1323 struct stack_trace trace
;
1325 int size
= FTRACE_STACK_ENTRIES
;
1327 trace
.nr_entries
= 0;
1331 * Since events can happen in NMIs there's no safe way to
1332 * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1333 * or NMI comes in, it will just have to use the default
1334 * FTRACE_STACK_SIZE.
1336 preempt_disable_notrace();
1338 use_stack
= ++__get_cpu_var(ftrace_stack_reserve
);
1340 * We don't need any atomic variables, just a barrier.
1341 * If an interrupt comes in, we don't care, because it would
1342 * have exited and put the counter back to what we want.
1343 * We just need a barrier to keep gcc from moving things
1347 if (use_stack
== 1) {
1348 trace
.entries
= &__get_cpu_var(ftrace_stack
).calls
[0];
1349 trace
.max_entries
= FTRACE_STACK_MAX_ENTRIES
;
1352 save_stack_trace_regs(regs
, &trace
);
1354 save_stack_trace(&trace
);
1356 if (trace
.nr_entries
> size
)
1357 size
= trace
.nr_entries
;
1359 /* From now on, use_stack is a boolean */
1362 size
*= sizeof(unsigned long);
1364 event
= trace_buffer_lock_reserve(buffer
, TRACE_STACK
,
1365 sizeof(*entry
) + size
, flags
, pc
);
1368 entry
= ring_buffer_event_data(event
);
1370 memset(&entry
->caller
, 0, size
);
1373 memcpy(&entry
->caller
, trace
.entries
,
1374 trace
.nr_entries
* sizeof(unsigned long));
1376 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1377 trace
.entries
= entry
->caller
;
1379 save_stack_trace_regs(regs
, &trace
);
1381 save_stack_trace(&trace
);
1384 entry
->size
= trace
.nr_entries
;
1386 if (!filter_check_discard(call
, entry
, buffer
, event
))
1387 __buffer_unlock_commit(buffer
, event
);
1390 /* Again, don't let gcc optimize things here */
1392 __get_cpu_var(ftrace_stack_reserve
)--;
1393 preempt_enable_notrace();
1397 void ftrace_trace_stack_regs(struct ring_buffer
*buffer
, unsigned long flags
,
1398 int skip
, int pc
, struct pt_regs
*regs
)
1400 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1403 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, regs
);
1406 void ftrace_trace_stack(struct ring_buffer
*buffer
, unsigned long flags
,
1409 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
1412 __ftrace_trace_stack(buffer
, flags
, skip
, pc
, NULL
);
1415 void __trace_stack(struct trace_array
*tr
, unsigned long flags
, int skip
,
1418 __ftrace_trace_stack(tr
->buffer
, flags
, skip
, pc
, NULL
);
1422 * trace_dump_stack - record a stack back trace in the trace buffer
1424 void trace_dump_stack(void)
1426 unsigned long flags
;
1428 if (tracing_disabled
|| tracing_selftest_running
)
1431 local_save_flags(flags
);
1433 /* skipping 3 traces, seems to get us at the caller of this function */
1434 __ftrace_trace_stack(global_trace
.buffer
, flags
, 3, preempt_count(), NULL
);
1437 static DEFINE_PER_CPU(int, user_stack_count
);
1440 ftrace_trace_userstack(struct ring_buffer
*buffer
, unsigned long flags
, int pc
)
1442 struct ftrace_event_call
*call
= &event_user_stack
;
1443 struct ring_buffer_event
*event
;
1444 struct userstack_entry
*entry
;
1445 struct stack_trace trace
;
1447 if (!(trace_flags
& TRACE_ITER_USERSTACKTRACE
))
1451 * NMIs can not handle page faults, even with fix ups.
1452 * The save user stack can (and often does) fault.
1454 if (unlikely(in_nmi()))
1458 * prevent recursion, since the user stack tracing may
1459 * trigger other kernel events.
1462 if (__this_cpu_read(user_stack_count
))
1465 __this_cpu_inc(user_stack_count
);
1467 event
= trace_buffer_lock_reserve(buffer
, TRACE_USER_STACK
,
1468 sizeof(*entry
), flags
, pc
);
1470 goto out_drop_count
;
1471 entry
= ring_buffer_event_data(event
);
1473 entry
->tgid
= current
->tgid
;
1474 memset(&entry
->caller
, 0, sizeof(entry
->caller
));
1476 trace
.nr_entries
= 0;
1477 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
1479 trace
.entries
= entry
->caller
;
1481 save_stack_trace_user(&trace
);
1482 if (!filter_check_discard(call
, entry
, buffer
, event
))
1483 __buffer_unlock_commit(buffer
, event
);
1486 __this_cpu_dec(user_stack_count
);
1492 static void __trace_userstack(struct trace_array
*tr
, unsigned long flags
)
1494 ftrace_trace_userstack(tr
, flags
, preempt_count());
1498 #endif /* CONFIG_STACKTRACE */
1500 /* created for use with alloc_percpu */
1501 struct trace_buffer_struct
{
1502 char buffer
[TRACE_BUF_SIZE
];
1505 static struct trace_buffer_struct
*trace_percpu_buffer
;
1506 static struct trace_buffer_struct
*trace_percpu_sirq_buffer
;
1507 static struct trace_buffer_struct
*trace_percpu_irq_buffer
;
1508 static struct trace_buffer_struct
*trace_percpu_nmi_buffer
;
1511 * The buffer used is dependent on the context. There is a per cpu
1512 * buffer for normal context, softirq contex, hard irq context and
1513 * for NMI context. Thise allows for lockless recording.
1515 * Note, if the buffers failed to be allocated, then this returns NULL
1517 static char *get_trace_buf(void)
1519 struct trace_buffer_struct
*percpu_buffer
;
1520 struct trace_buffer_struct
*buffer
;
1523 * If we have allocated per cpu buffers, then we do not
1524 * need to do any locking.
1527 percpu_buffer
= trace_percpu_nmi_buffer
;
1529 percpu_buffer
= trace_percpu_irq_buffer
;
1530 else if (in_softirq())
1531 percpu_buffer
= trace_percpu_sirq_buffer
;
1533 percpu_buffer
= trace_percpu_buffer
;
1538 buffer
= per_cpu_ptr(percpu_buffer
, smp_processor_id());
1540 return buffer
->buffer
;
1543 static int alloc_percpu_trace_buffer(void)
1545 struct trace_buffer_struct
*buffers
;
1546 struct trace_buffer_struct
*sirq_buffers
;
1547 struct trace_buffer_struct
*irq_buffers
;
1548 struct trace_buffer_struct
*nmi_buffers
;
1550 buffers
= alloc_percpu(struct trace_buffer_struct
);
1554 sirq_buffers
= alloc_percpu(struct trace_buffer_struct
);
1558 irq_buffers
= alloc_percpu(struct trace_buffer_struct
);
1562 nmi_buffers
= alloc_percpu(struct trace_buffer_struct
);
1566 trace_percpu_buffer
= buffers
;
1567 trace_percpu_sirq_buffer
= sirq_buffers
;
1568 trace_percpu_irq_buffer
= irq_buffers
;
1569 trace_percpu_nmi_buffer
= nmi_buffers
;
1574 free_percpu(irq_buffers
);
1576 free_percpu(sirq_buffers
);
1578 free_percpu(buffers
);
1580 WARN(1, "Could not allocate percpu trace_printk buffer");
1584 static int buffers_allocated
;
1586 void trace_printk_init_buffers(void)
1588 if (buffers_allocated
)
1591 if (alloc_percpu_trace_buffer())
1594 pr_info("ftrace: Allocated trace_printk buffers\n");
1596 /* Expand the buffers to set size */
1597 tracing_update_buffers();
1599 buffers_allocated
= 1;
1602 * trace_printk_init_buffers() can be called by modules.
1603 * If that happens, then we need to start cmdline recording
1604 * directly here. If the global_trace.buffer is already
1605 * allocated here, then this was called by module code.
1607 if (global_trace
.buffer
)
1608 tracing_start_cmdline_record();
1611 void trace_printk_start_comm(void)
1613 /* Start tracing comms if trace printk is set */
1614 if (!buffers_allocated
)
1616 tracing_start_cmdline_record();
1619 static void trace_printk_start_stop_comm(int enabled
)
1621 if (!buffers_allocated
)
1625 tracing_start_cmdline_record();
1627 tracing_stop_cmdline_record();
1631 * trace_vbprintk - write binary msg to tracing buffer
1634 int trace_vbprintk(unsigned long ip
, const char *fmt
, va_list args
)
1636 struct ftrace_event_call
*call
= &event_bprint
;
1637 struct ring_buffer_event
*event
;
1638 struct ring_buffer
*buffer
;
1639 struct trace_array
*tr
= &global_trace
;
1640 struct bprint_entry
*entry
;
1641 unsigned long flags
;
1643 int len
= 0, size
, pc
;
1645 if (unlikely(tracing_selftest_running
|| tracing_disabled
))
1648 /* Don't pollute graph traces with trace_vprintk internals */
1649 pause_graph_tracing();
1651 pc
= preempt_count();
1652 preempt_disable_notrace();
1654 tbuffer
= get_trace_buf();
1660 len
= vbin_printf((u32
*)tbuffer
, TRACE_BUF_SIZE
/sizeof(int), fmt
, args
);
1662 if (len
> TRACE_BUF_SIZE
/sizeof(int) || len
< 0)
1665 local_save_flags(flags
);
1666 size
= sizeof(*entry
) + sizeof(u32
) * len
;
1667 buffer
= tr
->buffer
;
1668 event
= trace_buffer_lock_reserve(buffer
, TRACE_BPRINT
, size
,
1672 entry
= ring_buffer_event_data(event
);
1676 memcpy(entry
->buf
, tbuffer
, sizeof(u32
) * len
);
1677 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1678 __buffer_unlock_commit(buffer
, event
);
1679 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1683 preempt_enable_notrace();
1684 unpause_graph_tracing();
1688 EXPORT_SYMBOL_GPL(trace_vbprintk
);
1690 int trace_array_printk(struct trace_array
*tr
,
1691 unsigned long ip
, const char *fmt
, ...)
1696 if (!(trace_flags
& TRACE_ITER_PRINTK
))
1700 ret
= trace_array_vprintk(tr
, ip
, fmt
, ap
);
1705 int trace_array_vprintk(struct trace_array
*tr
,
1706 unsigned long ip
, const char *fmt
, va_list args
)
1708 struct ftrace_event_call
*call
= &event_print
;
1709 struct ring_buffer_event
*event
;
1710 struct ring_buffer
*buffer
;
1711 int len
= 0, size
, pc
;
1712 struct print_entry
*entry
;
1713 unsigned long flags
;
1716 if (tracing_disabled
|| tracing_selftest_running
)
1719 /* Don't pollute graph traces with trace_vprintk internals */
1720 pause_graph_tracing();
1722 pc
= preempt_count();
1723 preempt_disable_notrace();
1726 tbuffer
= get_trace_buf();
1732 len
= vsnprintf(tbuffer
, TRACE_BUF_SIZE
, fmt
, args
);
1733 if (len
> TRACE_BUF_SIZE
)
1736 local_save_flags(flags
);
1737 size
= sizeof(*entry
) + len
+ 1;
1738 buffer
= tr
->buffer
;
1739 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
1743 entry
= ring_buffer_event_data(event
);
1746 memcpy(&entry
->buf
, tbuffer
, len
);
1747 entry
->buf
[len
] = '\0';
1748 if (!filter_check_discard(call
, entry
, buffer
, event
)) {
1749 __buffer_unlock_commit(buffer
, event
);
1750 ftrace_trace_stack(buffer
, flags
, 6, pc
);
1753 preempt_enable_notrace();
1754 unpause_graph_tracing();
1759 int trace_vprintk(unsigned long ip
, const char *fmt
, va_list args
)
1761 return trace_array_vprintk(&global_trace
, ip
, fmt
, args
);
1763 EXPORT_SYMBOL_GPL(trace_vprintk
);
1765 static void trace_iterator_increment(struct trace_iterator
*iter
)
1767 struct ring_buffer_iter
*buf_iter
= trace_buffer_iter(iter
, iter
->cpu
);
1771 ring_buffer_read(buf_iter
, NULL
);
1774 static struct trace_entry
*
1775 peek_next_entry(struct trace_iterator
*iter
, int cpu
, u64
*ts
,
1776 unsigned long *lost_events
)
1778 struct ring_buffer_event
*event
;
1779 struct ring_buffer_iter
*buf_iter
= trace_buffer_iter(iter
, cpu
);
1782 event
= ring_buffer_iter_peek(buf_iter
, ts
);
1784 event
= ring_buffer_peek(iter
->tr
->buffer
, cpu
, ts
,
1788 iter
->ent_size
= ring_buffer_event_length(event
);
1789 return ring_buffer_event_data(event
);
1795 static struct trace_entry
*
1796 __find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
,
1797 unsigned long *missing_events
, u64
*ent_ts
)
1799 struct ring_buffer
*buffer
= iter
->tr
->buffer
;
1800 struct trace_entry
*ent
, *next
= NULL
;
1801 unsigned long lost_events
= 0, next_lost
= 0;
1802 int cpu_file
= iter
->cpu_file
;
1803 u64 next_ts
= 0, ts
;
1809 * If we are in a per_cpu trace file, don't bother by iterating over
1810 * all cpu and peek directly.
1812 if (cpu_file
> TRACE_PIPE_ALL_CPU
) {
1813 if (ring_buffer_empty_cpu(buffer
, cpu_file
))
1815 ent
= peek_next_entry(iter
, cpu_file
, ent_ts
, missing_events
);
1817 *ent_cpu
= cpu_file
;
1822 for_each_tracing_cpu(cpu
) {
1824 if (ring_buffer_empty_cpu(buffer
, cpu
))
1827 ent
= peek_next_entry(iter
, cpu
, &ts
, &lost_events
);
1830 * Pick the entry with the smallest timestamp:
1832 if (ent
&& (!next
|| ts
< next_ts
)) {
1836 next_lost
= lost_events
;
1837 next_size
= iter
->ent_size
;
1841 iter
->ent_size
= next_size
;
1844 *ent_cpu
= next_cpu
;
1850 *missing_events
= next_lost
;
1855 /* Find the next real entry, without updating the iterator itself */
1856 struct trace_entry
*trace_find_next_entry(struct trace_iterator
*iter
,
1857 int *ent_cpu
, u64
*ent_ts
)
1859 return __find_next_entry(iter
, ent_cpu
, NULL
, ent_ts
);
1862 /* Find the next real entry, and increment the iterator to the next entry */
1863 void *trace_find_next_entry_inc(struct trace_iterator
*iter
)
1865 iter
->ent
= __find_next_entry(iter
, &iter
->cpu
,
1866 &iter
->lost_events
, &iter
->ts
);
1869 trace_iterator_increment(iter
);
1871 return iter
->ent
? iter
: NULL
;
1874 static void trace_consume(struct trace_iterator
*iter
)
1876 ring_buffer_consume(iter
->tr
->buffer
, iter
->cpu
, &iter
->ts
,
1877 &iter
->lost_events
);
1880 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1882 struct trace_iterator
*iter
= m
->private;
1886 WARN_ON_ONCE(iter
->leftover
);
1890 /* can't go backwards */
1895 ent
= trace_find_next_entry_inc(iter
);
1899 while (ent
&& iter
->idx
< i
)
1900 ent
= trace_find_next_entry_inc(iter
);
1907 void tracing_iter_reset(struct trace_iterator
*iter
, int cpu
)
1909 struct trace_array
*tr
= iter
->tr
;
1910 struct ring_buffer_event
*event
;
1911 struct ring_buffer_iter
*buf_iter
;
1912 unsigned long entries
= 0;
1915 tr
->data
[cpu
]->skipped_entries
= 0;
1917 buf_iter
= trace_buffer_iter(iter
, cpu
);
1921 ring_buffer_iter_reset(buf_iter
);
1924 * We could have the case with the max latency tracers
1925 * that a reset never took place on a cpu. This is evident
1926 * by the timestamp being before the start of the buffer.
1928 while ((event
= ring_buffer_iter_peek(buf_iter
, &ts
))) {
1929 if (ts
>= iter
->tr
->time_start
)
1932 ring_buffer_read(buf_iter
, NULL
);
1935 tr
->data
[cpu
]->skipped_entries
= entries
;
1939 * The current tracer is copied to avoid a global locking
1942 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1944 struct trace_iterator
*iter
= m
->private;
1945 static struct tracer
*old_tracer
;
1946 int cpu_file
= iter
->cpu_file
;
1951 /* copy the tracer to avoid using a global lock all around */
1952 mutex_lock(&trace_types_lock
);
1953 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
1954 old_tracer
= current_trace
;
1955 *iter
->trace
= *current_trace
;
1957 mutex_unlock(&trace_types_lock
);
1959 atomic_inc(&trace_record_cmdline_disabled
);
1961 if (*pos
!= iter
->pos
) {
1966 if (cpu_file
== TRACE_PIPE_ALL_CPU
) {
1967 for_each_tracing_cpu(cpu
)
1968 tracing_iter_reset(iter
, cpu
);
1970 tracing_iter_reset(iter
, cpu_file
);
1973 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1978 * If we overflowed the seq_file before, then we want
1979 * to just reuse the trace_seq buffer again.
1985 p
= s_next(m
, p
, &l
);
1989 trace_event_read_lock();
1990 trace_access_lock(cpu_file
);
1994 static void s_stop(struct seq_file
*m
, void *p
)
1996 struct trace_iterator
*iter
= m
->private;
1998 atomic_dec(&trace_record_cmdline_disabled
);
1999 trace_access_unlock(iter
->cpu_file
);
2000 trace_event_read_unlock();
2004 get_total_entries(struct trace_array
*tr
, unsigned long *total
, unsigned long *entries
)
2006 unsigned long count
;
2012 for_each_tracing_cpu(cpu
) {
2013 count
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
2015 * If this buffer has skipped entries, then we hold all
2016 * entries for the trace and we need to ignore the
2017 * ones before the time stamp.
2019 if (tr
->data
[cpu
]->skipped_entries
) {
2020 count
-= tr
->data
[cpu
]->skipped_entries
;
2021 /* total is the same as the entries */
2025 ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
2030 static void print_lat_help_header(struct seq_file
*m
)
2032 seq_puts(m
, "# _------=> CPU# \n");
2033 seq_puts(m
, "# / _-----=> irqs-off \n");
2034 seq_puts(m
, "# | / _----=> need-resched \n");
2035 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
2036 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
2037 seq_puts(m
, "# |||| / delay \n");
2038 seq_puts(m
, "# cmd pid ||||| time | caller \n");
2039 seq_puts(m
, "# \\ / ||||| \\ | / \n");
2042 static void print_event_info(struct trace_array
*tr
, struct seq_file
*m
)
2044 unsigned long total
;
2045 unsigned long entries
;
2047 get_total_entries(tr
, &total
, &entries
);
2048 seq_printf(m
, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
2049 entries
, total
, num_online_cpus());
2053 static void print_func_help_header(struct trace_array
*tr
, struct seq_file
*m
)
2055 print_event_info(tr
, m
);
2056 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
2057 seq_puts(m
, "# | | | | |\n");
2060 static void print_func_help_header_irq(struct trace_array
*tr
, struct seq_file
*m
)
2062 print_event_info(tr
, m
);
2063 seq_puts(m
, "# _-----=> irqs-off\n");
2064 seq_puts(m
, "# / _----=> need-resched\n");
2065 seq_puts(m
, "# | / _---=> hardirq/softirq\n");
2066 seq_puts(m
, "# || / _--=> preempt-depth\n");
2067 seq_puts(m
, "# ||| / delay\n");
2068 seq_puts(m
, "# TASK-PID CPU# |||| TIMESTAMP FUNCTION\n");
2069 seq_puts(m
, "# | | | |||| | |\n");
2073 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
2075 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2076 struct trace_array
*tr
= iter
->tr
;
2077 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
2078 struct tracer
*type
= current_trace
;
2079 unsigned long entries
;
2080 unsigned long total
;
2081 const char *name
= "preemption";
2086 get_total_entries(tr
, &total
, &entries
);
2088 seq_printf(m
, "# %s latency trace v1.1.5 on %s\n",
2090 seq_puts(m
, "# -----------------------------------"
2091 "---------------------------------\n");
2092 seq_printf(m
, "# latency: %lu us, #%lu/%lu, CPU#%d |"
2093 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
2094 nsecs_to_usecs(data
->saved_latency
),
2098 #if defined(CONFIG_PREEMPT_NONE)
2100 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2102 #elif defined(CONFIG_PREEMPT)
2107 /* These are reserved for later use */
2110 seq_printf(m
, " #P:%d)\n", num_online_cpus());
2114 seq_puts(m
, "# -----------------\n");
2115 seq_printf(m
, "# | task: %.16s-%d "
2116 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2117 data
->comm
, data
->pid
,
2118 from_kuid_munged(seq_user_ns(m
), data
->uid
), data
->nice
,
2119 data
->policy
, data
->rt_priority
);
2120 seq_puts(m
, "# -----------------\n");
2122 if (data
->critical_start
) {
2123 seq_puts(m
, "# => started at: ");
2124 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
2125 trace_print_seq(m
, &iter
->seq
);
2126 seq_puts(m
, "\n# => ended at: ");
2127 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
2128 trace_print_seq(m
, &iter
->seq
);
2129 seq_puts(m
, "\n#\n");
2135 static void test_cpu_buff_start(struct trace_iterator
*iter
)
2137 struct trace_seq
*s
= &iter
->seq
;
2139 if (!(trace_flags
& TRACE_ITER_ANNOTATE
))
2142 if (!(iter
->iter_flags
& TRACE_FILE_ANNOTATE
))
2145 if (cpumask_test_cpu(iter
->cpu
, iter
->started
))
2148 if (iter
->tr
->data
[iter
->cpu
]->skipped_entries
)
2151 cpumask_set_cpu(iter
->cpu
, iter
->started
);
2153 /* Don't print started cpu buffer for the first entry of the trace */
2155 trace_seq_printf(s
, "##### CPU %u buffer started ####\n",
2159 static enum print_line_t
print_trace_fmt(struct trace_iterator
*iter
)
2161 struct trace_seq
*s
= &iter
->seq
;
2162 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
2163 struct trace_entry
*entry
;
2164 struct trace_event
*event
;
2168 test_cpu_buff_start(iter
);
2170 event
= ftrace_find_event(entry
->type
);
2172 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2173 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2174 if (!trace_print_lat_context(iter
))
2177 if (!trace_print_context(iter
))
2183 return event
->funcs
->trace(iter
, sym_flags
, event
);
2185 if (!trace_seq_printf(s
, "Unknown type %d\n", entry
->type
))
2188 return TRACE_TYPE_HANDLED
;
2190 return TRACE_TYPE_PARTIAL_LINE
;
2193 static enum print_line_t
print_raw_fmt(struct trace_iterator
*iter
)
2195 struct trace_seq
*s
= &iter
->seq
;
2196 struct trace_entry
*entry
;
2197 struct trace_event
*event
;
2201 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2202 if (!trace_seq_printf(s
, "%d %d %llu ",
2203 entry
->pid
, iter
->cpu
, iter
->ts
))
2207 event
= ftrace_find_event(entry
->type
);
2209 return event
->funcs
->raw(iter
, 0, event
);
2211 if (!trace_seq_printf(s
, "%d ?\n", entry
->type
))
2214 return TRACE_TYPE_HANDLED
;
2216 return TRACE_TYPE_PARTIAL_LINE
;
2219 static enum print_line_t
print_hex_fmt(struct trace_iterator
*iter
)
2221 struct trace_seq
*s
= &iter
->seq
;
2222 unsigned char newline
= '\n';
2223 struct trace_entry
*entry
;
2224 struct trace_event
*event
;
2228 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2229 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
2230 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
2231 SEQ_PUT_HEX_FIELD_RET(s
, iter
->ts
);
2234 event
= ftrace_find_event(entry
->type
);
2236 enum print_line_t ret
= event
->funcs
->hex(iter
, 0, event
);
2237 if (ret
!= TRACE_TYPE_HANDLED
)
2241 SEQ_PUT_FIELD_RET(s
, newline
);
2243 return TRACE_TYPE_HANDLED
;
2246 static enum print_line_t
print_bin_fmt(struct trace_iterator
*iter
)
2248 struct trace_seq
*s
= &iter
->seq
;
2249 struct trace_entry
*entry
;
2250 struct trace_event
*event
;
2254 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
2255 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
2256 SEQ_PUT_FIELD_RET(s
, iter
->cpu
);
2257 SEQ_PUT_FIELD_RET(s
, iter
->ts
);
2260 event
= ftrace_find_event(entry
->type
);
2261 return event
? event
->funcs
->binary(iter
, 0, event
) :
2265 int trace_empty(struct trace_iterator
*iter
)
2267 struct ring_buffer_iter
*buf_iter
;
2270 /* If we are looking at one CPU buffer, only check that one */
2271 if (iter
->cpu_file
!= TRACE_PIPE_ALL_CPU
) {
2272 cpu
= iter
->cpu_file
;
2273 buf_iter
= trace_buffer_iter(iter
, cpu
);
2275 if (!ring_buffer_iter_empty(buf_iter
))
2278 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2284 for_each_tracing_cpu(cpu
) {
2285 buf_iter
= trace_buffer_iter(iter
, cpu
);
2287 if (!ring_buffer_iter_empty(buf_iter
))
2290 if (!ring_buffer_empty_cpu(iter
->tr
->buffer
, cpu
))
2298 /* Called with trace_event_read_lock() held. */
2299 enum print_line_t
print_trace_line(struct trace_iterator
*iter
)
2301 enum print_line_t ret
;
2303 if (iter
->lost_events
&&
2304 !trace_seq_printf(&iter
->seq
, "CPU:%d [LOST %lu EVENTS]\n",
2305 iter
->cpu
, iter
->lost_events
))
2306 return TRACE_TYPE_PARTIAL_LINE
;
2308 if (iter
->trace
&& iter
->trace
->print_line
) {
2309 ret
= iter
->trace
->print_line(iter
);
2310 if (ret
!= TRACE_TYPE_UNHANDLED
)
2314 if (iter
->ent
->type
== TRACE_BPRINT
&&
2315 trace_flags
& TRACE_ITER_PRINTK
&&
2316 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2317 return trace_print_bprintk_msg_only(iter
);
2319 if (iter
->ent
->type
== TRACE_PRINT
&&
2320 trace_flags
& TRACE_ITER_PRINTK
&&
2321 trace_flags
& TRACE_ITER_PRINTK_MSGONLY
)
2322 return trace_print_printk_msg_only(iter
);
2324 if (trace_flags
& TRACE_ITER_BIN
)
2325 return print_bin_fmt(iter
);
2327 if (trace_flags
& TRACE_ITER_HEX
)
2328 return print_hex_fmt(iter
);
2330 if (trace_flags
& TRACE_ITER_RAW
)
2331 return print_raw_fmt(iter
);
2333 return print_trace_fmt(iter
);
2336 void trace_latency_header(struct seq_file
*m
)
2338 struct trace_iterator
*iter
= m
->private;
2340 /* print nothing if the buffers are empty */
2341 if (trace_empty(iter
))
2344 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
2345 print_trace_header(m
, iter
);
2347 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2348 print_lat_help_header(m
);
2351 void trace_default_header(struct seq_file
*m
)
2353 struct trace_iterator
*iter
= m
->private;
2355 if (!(trace_flags
& TRACE_ITER_CONTEXT_INFO
))
2358 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
2359 /* print nothing if the buffers are empty */
2360 if (trace_empty(iter
))
2362 print_trace_header(m
, iter
);
2363 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
2364 print_lat_help_header(m
);
2366 if (!(trace_flags
& TRACE_ITER_VERBOSE
)) {
2367 if (trace_flags
& TRACE_ITER_IRQ_INFO
)
2368 print_func_help_header_irq(iter
->tr
, m
);
2370 print_func_help_header(iter
->tr
, m
);
2375 static void test_ftrace_alive(struct seq_file
*m
)
2377 if (!ftrace_is_dead())
2379 seq_printf(m
, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2380 seq_printf(m
, "# MAY BE MISSING FUNCTION EVENTS\n");
2383 static int s_show(struct seq_file
*m
, void *v
)
2385 struct trace_iterator
*iter
= v
;
2388 if (iter
->ent
== NULL
) {
2390 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
2392 test_ftrace_alive(m
);
2394 if (iter
->trace
&& iter
->trace
->print_header
)
2395 iter
->trace
->print_header(m
);
2397 trace_default_header(m
);
2399 } else if (iter
->leftover
) {
2401 * If we filled the seq_file buffer earlier, we
2402 * want to just show it now.
2404 ret
= trace_print_seq(m
, &iter
->seq
);
2406 /* ret should this time be zero, but you never know */
2407 iter
->leftover
= ret
;
2410 print_trace_line(iter
);
2411 ret
= trace_print_seq(m
, &iter
->seq
);
2413 * If we overflow the seq_file buffer, then it will
2414 * ask us for this data again at start up.
2416 * ret is 0 if seq_file write succeeded.
2419 iter
->leftover
= ret
;
2425 static const struct seq_operations tracer_seq_ops
= {
2432 static struct trace_iterator
*
2433 __tracing_open(struct inode
*inode
, struct file
*file
)
2435 long cpu_file
= (long) inode
->i_private
;
2436 struct trace_iterator
*iter
;
2439 if (tracing_disabled
)
2440 return ERR_PTR(-ENODEV
);
2442 iter
= __seq_open_private(file
, &tracer_seq_ops
, sizeof(*iter
));
2444 return ERR_PTR(-ENOMEM
);
2446 iter
->buffer_iter
= kzalloc(sizeof(*iter
->buffer_iter
) * num_possible_cpus(),
2448 if (!iter
->buffer_iter
)
2452 * We make a copy of the current tracer to avoid concurrent
2453 * changes on it while we are reading.
2455 mutex_lock(&trace_types_lock
);
2456 iter
->trace
= kzalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
2461 *iter
->trace
= *current_trace
;
2463 if (!zalloc_cpumask_var(&iter
->started
, GFP_KERNEL
))
2466 if (current_trace
&& current_trace
->print_max
)
2469 iter
->tr
= &global_trace
;
2471 mutex_init(&iter
->mutex
);
2472 iter
->cpu_file
= cpu_file
;
2474 /* Notify the tracer early; before we stop tracing. */
2475 if (iter
->trace
&& iter
->trace
->open
)
2476 iter
->trace
->open(iter
);
2478 /* Annotate start of buffers if we had overruns */
2479 if (ring_buffer_overruns(iter
->tr
->buffer
))
2480 iter
->iter_flags
|= TRACE_FILE_ANNOTATE
;
2482 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
2483 if (trace_clocks
[trace_clock_id
].in_ns
)
2484 iter
->iter_flags
|= TRACE_FILE_TIME_IN_NS
;
2486 /* stop the trace while dumping */
2489 if (iter
->cpu_file
== TRACE_PIPE_ALL_CPU
) {
2490 for_each_tracing_cpu(cpu
) {
2491 iter
->buffer_iter
[cpu
] =
2492 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2494 ring_buffer_read_prepare_sync();
2495 for_each_tracing_cpu(cpu
) {
2496 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2497 tracing_iter_reset(iter
, cpu
);
2500 cpu
= iter
->cpu_file
;
2501 iter
->buffer_iter
[cpu
] =
2502 ring_buffer_read_prepare(iter
->tr
->buffer
, cpu
);
2503 ring_buffer_read_prepare_sync();
2504 ring_buffer_read_start(iter
->buffer_iter
[cpu
]);
2505 tracing_iter_reset(iter
, cpu
);
2508 mutex_unlock(&trace_types_lock
);
2513 mutex_unlock(&trace_types_lock
);
2515 kfree(iter
->buffer_iter
);
2517 seq_release_private(inode
, file
);
2518 return ERR_PTR(-ENOMEM
);
2521 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
2523 if (tracing_disabled
)
2526 filp
->private_data
= inode
->i_private
;
2530 static int tracing_release(struct inode
*inode
, struct file
*file
)
2532 struct seq_file
*m
= file
->private_data
;
2533 struct trace_iterator
*iter
;
2536 if (!(file
->f_mode
& FMODE_READ
))
2541 mutex_lock(&trace_types_lock
);
2542 for_each_tracing_cpu(cpu
) {
2543 if (iter
->buffer_iter
[cpu
])
2544 ring_buffer_read_finish(iter
->buffer_iter
[cpu
]);
2547 if (iter
->trace
&& iter
->trace
->close
)
2548 iter
->trace
->close(iter
);
2550 /* reenable tracing if it was previously enabled */
2552 mutex_unlock(&trace_types_lock
);
2554 mutex_destroy(&iter
->mutex
);
2555 free_cpumask_var(iter
->started
);
2557 kfree(iter
->buffer_iter
);
2558 seq_release_private(inode
, file
);
2562 static int tracing_open(struct inode
*inode
, struct file
*file
)
2564 struct trace_iterator
*iter
;
2567 /* If this file was open for write, then erase contents */
2568 if ((file
->f_mode
& FMODE_WRITE
) &&
2569 (file
->f_flags
& O_TRUNC
)) {
2570 long cpu
= (long) inode
->i_private
;
2572 if (cpu
== TRACE_PIPE_ALL_CPU
)
2573 tracing_reset_online_cpus(&global_trace
);
2575 tracing_reset(&global_trace
, cpu
);
2578 if (file
->f_mode
& FMODE_READ
) {
2579 iter
= __tracing_open(inode
, file
);
2581 ret
= PTR_ERR(iter
);
2582 else if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
2583 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
2589 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
2591 struct tracer
*t
= v
;
2601 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
2606 mutex_lock(&trace_types_lock
);
2607 for (t
= trace_types
; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2613 static void t_stop(struct seq_file
*m
, void *p
)
2615 mutex_unlock(&trace_types_lock
);
2618 static int t_show(struct seq_file
*m
, void *v
)
2620 struct tracer
*t
= v
;
2625 seq_printf(m
, "%s", t
->name
);
2634 static const struct seq_operations show_traces_seq_ops
= {
2641 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2643 if (tracing_disabled
)
2646 return seq_open(file
, &show_traces_seq_ops
);
2650 tracing_write_stub(struct file
*filp
, const char __user
*ubuf
,
2651 size_t count
, loff_t
*ppos
)
2656 static loff_t
tracing_seek(struct file
*file
, loff_t offset
, int origin
)
2658 if (file
->f_mode
& FMODE_READ
)
2659 return seq_lseek(file
, offset
, origin
);
2664 static const struct file_operations tracing_fops
= {
2665 .open
= tracing_open
,
2667 .write
= tracing_write_stub
,
2668 .llseek
= tracing_seek
,
2669 .release
= tracing_release
,
2672 static const struct file_operations show_traces_fops
= {
2673 .open
= show_traces_open
,
2675 .release
= seq_release
,
2676 .llseek
= seq_lseek
,
2680 * Only trace on a CPU if the bitmask is set:
2682 static cpumask_var_t tracing_cpumask
;
2685 * The tracer itself will not take this lock, but still we want
2686 * to provide a consistent cpumask to user-space:
2688 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2691 * Temporary storage for the character representation of the
2692 * CPU bitmask (and one more byte for the newline):
2694 static char mask_str
[NR_CPUS
+ 1];
2697 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2698 size_t count
, loff_t
*ppos
)
2702 mutex_lock(&tracing_cpumask_update_lock
);
2704 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2705 if (count
- len
< 2) {
2709 len
+= sprintf(mask_str
+ len
, "\n");
2710 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2713 mutex_unlock(&tracing_cpumask_update_lock
);
2719 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2720 size_t count
, loff_t
*ppos
)
2723 cpumask_var_t tracing_cpumask_new
;
2725 if (!alloc_cpumask_var(&tracing_cpumask_new
, GFP_KERNEL
))
2728 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2732 mutex_lock(&tracing_cpumask_update_lock
);
2734 local_irq_disable();
2735 arch_spin_lock(&ftrace_max_lock
);
2736 for_each_tracing_cpu(cpu
) {
2738 * Increase/decrease the disabled counter if we are
2739 * about to flip a bit in the cpumask:
2741 if (cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2742 !cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2743 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2744 ring_buffer_record_disable_cpu(global_trace
.buffer
, cpu
);
2746 if (!cpumask_test_cpu(cpu
, tracing_cpumask
) &&
2747 cpumask_test_cpu(cpu
, tracing_cpumask_new
)) {
2748 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2749 ring_buffer_record_enable_cpu(global_trace
.buffer
, cpu
);
2752 arch_spin_unlock(&ftrace_max_lock
);
2755 cpumask_copy(tracing_cpumask
, tracing_cpumask_new
);
2757 mutex_unlock(&tracing_cpumask_update_lock
);
2758 free_cpumask_var(tracing_cpumask_new
);
2763 free_cpumask_var(tracing_cpumask_new
);
2768 static const struct file_operations tracing_cpumask_fops
= {
2769 .open
= tracing_open_generic
,
2770 .read
= tracing_cpumask_read
,
2771 .write
= tracing_cpumask_write
,
2772 .llseek
= generic_file_llseek
,
2775 static int tracing_trace_options_show(struct seq_file
*m
, void *v
)
2777 struct tracer_opt
*trace_opts
;
2781 mutex_lock(&trace_types_lock
);
2782 tracer_flags
= current_trace
->flags
->val
;
2783 trace_opts
= current_trace
->flags
->opts
;
2785 for (i
= 0; trace_options
[i
]; i
++) {
2786 if (trace_flags
& (1 << i
))
2787 seq_printf(m
, "%s\n", trace_options
[i
]);
2789 seq_printf(m
, "no%s\n", trace_options
[i
]);
2792 for (i
= 0; trace_opts
[i
].name
; i
++) {
2793 if (tracer_flags
& trace_opts
[i
].bit
)
2794 seq_printf(m
, "%s\n", trace_opts
[i
].name
);
2796 seq_printf(m
, "no%s\n", trace_opts
[i
].name
);
2798 mutex_unlock(&trace_types_lock
);
2803 static int __set_tracer_option(struct tracer
*trace
,
2804 struct tracer_flags
*tracer_flags
,
2805 struct tracer_opt
*opts
, int neg
)
2809 ret
= trace
->set_flag(tracer_flags
->val
, opts
->bit
, !neg
);
2814 tracer_flags
->val
&= ~opts
->bit
;
2816 tracer_flags
->val
|= opts
->bit
;
2820 /* Try to assign a tracer specific option */
2821 static int set_tracer_option(struct tracer
*trace
, char *cmp
, int neg
)
2823 struct tracer_flags
*tracer_flags
= trace
->flags
;
2824 struct tracer_opt
*opts
= NULL
;
2827 for (i
= 0; tracer_flags
->opts
[i
].name
; i
++) {
2828 opts
= &tracer_flags
->opts
[i
];
2830 if (strcmp(cmp
, opts
->name
) == 0)
2831 return __set_tracer_option(trace
, trace
->flags
,
2838 static void set_tracer_flags(unsigned int mask
, int enabled
)
2840 /* do nothing if flag is already set */
2841 if (!!(trace_flags
& mask
) == !!enabled
)
2845 trace_flags
|= mask
;
2847 trace_flags
&= ~mask
;
2849 if (mask
== TRACE_ITER_RECORD_CMD
)
2850 trace_event_enable_cmd_record(enabled
);
2852 if (mask
== TRACE_ITER_OVERWRITE
)
2853 ring_buffer_change_overwrite(global_trace
.buffer
, enabled
);
2855 if (mask
== TRACE_ITER_PRINTK
)
2856 trace_printk_start_stop_comm(enabled
);
2859 static int trace_set_options(char *option
)
2866 cmp
= strstrip(option
);
2868 if (strncmp(cmp
, "no", 2) == 0) {
2873 for (i
= 0; trace_options
[i
]; i
++) {
2874 if (strcmp(cmp
, trace_options
[i
]) == 0) {
2875 set_tracer_flags(1 << i
, !neg
);
2880 /* If no option could be set, test the specific tracer options */
2881 if (!trace_options
[i
]) {
2882 mutex_lock(&trace_types_lock
);
2883 ret
= set_tracer_option(current_trace
, cmp
, neg
);
2884 mutex_unlock(&trace_types_lock
);
2891 tracing_trace_options_write(struct file
*filp
, const char __user
*ubuf
,
2892 size_t cnt
, loff_t
*ppos
)
2896 if (cnt
>= sizeof(buf
))
2899 if (copy_from_user(&buf
, ubuf
, cnt
))
2902 trace_set_options(buf
);
2909 static int tracing_trace_options_open(struct inode
*inode
, struct file
*file
)
2911 if (tracing_disabled
)
2913 return single_open(file
, tracing_trace_options_show
, NULL
);
2916 static const struct file_operations tracing_iter_fops
= {
2917 .open
= tracing_trace_options_open
,
2919 .llseek
= seq_lseek
,
2920 .release
= single_release
,
2921 .write
= tracing_trace_options_write
,
2924 static const char readme_msg
[] =
2925 "tracing mini-HOWTO:\n\n"
2926 "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2927 "# cat /sys/kernel/debug/tracing/available_tracers\n"
2928 "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2929 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2931 "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2932 "# cat /sys/kernel/debug/tracing/current_tracer\n"
2934 "# cat /sys/kernel/debug/tracing/trace_options\n"
2935 "noprint-parent nosym-offset nosym-addr noverbose\n"
2936 "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2937 "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2938 "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2939 "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2943 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2944 size_t cnt
, loff_t
*ppos
)
2946 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2947 readme_msg
, strlen(readme_msg
));
2950 static const struct file_operations tracing_readme_fops
= {
2951 .open
= tracing_open_generic
,
2952 .read
= tracing_readme_read
,
2953 .llseek
= generic_file_llseek
,
2957 tracing_saved_cmdlines_read(struct file
*file
, char __user
*ubuf
,
2958 size_t cnt
, loff_t
*ppos
)
2967 file_buf
= kmalloc(SAVED_CMDLINES
*(16+TASK_COMM_LEN
), GFP_KERNEL
);
2971 buf_comm
= kmalloc(TASK_COMM_LEN
, GFP_KERNEL
);
2979 for (i
= 0; i
< SAVED_CMDLINES
; i
++) {
2982 pid
= map_cmdline_to_pid
[i
];
2983 if (pid
== -1 || pid
== NO_CMDLINE_MAP
)
2986 trace_find_cmdline(pid
, buf_comm
);
2987 r
= sprintf(buf
, "%d %s\n", pid
, buf_comm
);
2992 len
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
3001 static const struct file_operations tracing_saved_cmdlines_fops
= {
3002 .open
= tracing_open_generic
,
3003 .read
= tracing_saved_cmdlines_read
,
3004 .llseek
= generic_file_llseek
,
3008 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
3009 size_t cnt
, loff_t
*ppos
)
3011 char buf
[MAX_TRACER_SIZE
+2];
3014 mutex_lock(&trace_types_lock
);
3016 r
= sprintf(buf
, "%s\n", current_trace
->name
);
3018 r
= sprintf(buf
, "\n");
3019 mutex_unlock(&trace_types_lock
);
3021 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3024 int tracer_init(struct tracer
*t
, struct trace_array
*tr
)
3026 tracing_reset_online_cpus(tr
);
3030 static void set_buffer_entries(struct trace_array
*tr
, unsigned long val
)
3033 for_each_tracing_cpu(cpu
)
3034 tr
->data
[cpu
]->entries
= val
;
3037 /* resize @tr's buffer to the size of @size_tr's entries */
3038 static int resize_buffer_duplicate_size(struct trace_array
*tr
,
3039 struct trace_array
*size_tr
, int cpu_id
)
3043 if (cpu_id
== RING_BUFFER_ALL_CPUS
) {
3044 for_each_tracing_cpu(cpu
) {
3045 ret
= ring_buffer_resize(tr
->buffer
,
3046 size_tr
->data
[cpu
]->entries
, cpu
);
3049 tr
->data
[cpu
]->entries
= size_tr
->data
[cpu
]->entries
;
3052 ret
= ring_buffer_resize(tr
->buffer
,
3053 size_tr
->data
[cpu_id
]->entries
, cpu_id
);
3055 tr
->data
[cpu_id
]->entries
=
3056 size_tr
->data
[cpu_id
]->entries
;
3062 static int __tracing_resize_ring_buffer(unsigned long size
, int cpu
)
3067 * If kernel or user changes the size of the ring buffer
3068 * we use the size that was given, and we can forget about
3069 * expanding it later.
3071 ring_buffer_expanded
= 1;
3073 /* May be called before buffers are initialized */
3074 if (!global_trace
.buffer
)
3077 ret
= ring_buffer_resize(global_trace
.buffer
, size
, cpu
);
3081 if (!current_trace
->use_max_tr
)
3084 ret
= ring_buffer_resize(max_tr
.buffer
, size
, cpu
);
3086 int r
= resize_buffer_duplicate_size(&global_trace
,
3087 &global_trace
, cpu
);
3090 * AARGH! We are left with different
3091 * size max buffer!!!!
3092 * The max buffer is our "snapshot" buffer.
3093 * When a tracer needs a snapshot (one of the
3094 * latency tracers), it swaps the max buffer
3095 * with the saved snap shot. We succeeded to
3096 * update the size of the main buffer, but failed to
3097 * update the size of the max buffer. But when we tried
3098 * to reset the main buffer to the original size, we
3099 * failed there too. This is very unlikely to
3100 * happen, but if it does, warn and kill all
3104 tracing_disabled
= 1;
3109 if (cpu
== RING_BUFFER_ALL_CPUS
)
3110 set_buffer_entries(&max_tr
, size
);
3112 max_tr
.data
[cpu
]->entries
= size
;
3115 if (cpu
== RING_BUFFER_ALL_CPUS
)
3116 set_buffer_entries(&global_trace
, size
);
3118 global_trace
.data
[cpu
]->entries
= size
;
3123 static ssize_t
tracing_resize_ring_buffer(unsigned long size
, int cpu_id
)
3127 mutex_lock(&trace_types_lock
);
3129 if (cpu_id
!= RING_BUFFER_ALL_CPUS
) {
3130 /* make sure, this cpu is enabled in the mask */
3131 if (!cpumask_test_cpu(cpu_id
, tracing_buffer_mask
)) {
3137 ret
= __tracing_resize_ring_buffer(size
, cpu_id
);
3142 mutex_unlock(&trace_types_lock
);
3149 * tracing_update_buffers - used by tracing facility to expand ring buffers
3151 * To save on memory when the tracing is never used on a system with it
3152 * configured in. The ring buffers are set to a minimum size. But once
3153 * a user starts to use the tracing facility, then they need to grow
3154 * to their default size.
3156 * This function is to be called when a tracer is about to be used.
3158 int tracing_update_buffers(void)
3162 mutex_lock(&trace_types_lock
);
3163 if (!ring_buffer_expanded
)
3164 ret
= __tracing_resize_ring_buffer(trace_buf_size
,
3165 RING_BUFFER_ALL_CPUS
);
3166 mutex_unlock(&trace_types_lock
);
3171 struct trace_option_dentry
;
3173 static struct trace_option_dentry
*
3174 create_trace_option_files(struct tracer
*tracer
);
3177 destroy_trace_option_files(struct trace_option_dentry
*topts
);
3179 static int tracing_set_tracer(const char *buf
)
3181 static struct trace_option_dentry
*topts
;
3182 struct trace_array
*tr
= &global_trace
;
3186 mutex_lock(&trace_types_lock
);
3188 if (!ring_buffer_expanded
) {
3189 ret
= __tracing_resize_ring_buffer(trace_buf_size
,
3190 RING_BUFFER_ALL_CPUS
);
3196 for (t
= trace_types
; t
; t
= t
->next
) {
3197 if (strcmp(t
->name
, buf
) == 0)
3204 if (t
== current_trace
)
3207 trace_branch_disable();
3208 if (current_trace
&& current_trace
->reset
)
3209 current_trace
->reset(tr
);
3210 if (current_trace
&& current_trace
->use_max_tr
) {
3212 * We don't free the ring buffer. instead, resize it because
3213 * The max_tr ring buffer has some state (e.g. ring->clock) and
3214 * we want preserve it.
3216 ring_buffer_resize(max_tr
.buffer
, 1, RING_BUFFER_ALL_CPUS
);
3217 set_buffer_entries(&max_tr
, 1);
3219 destroy_trace_option_files(topts
);
3221 current_trace
= &nop_trace
;
3223 topts
= create_trace_option_files(t
);
3224 if (t
->use_max_tr
) {
3225 /* we need to make per cpu buffer sizes equivalent */
3226 ret
= resize_buffer_duplicate_size(&max_tr
, &global_trace
,
3227 RING_BUFFER_ALL_CPUS
);
3233 ret
= tracer_init(t
, tr
);
3239 trace_branch_enable(tr
);
3241 mutex_unlock(&trace_types_lock
);
3247 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
3248 size_t cnt
, loff_t
*ppos
)
3250 char buf
[MAX_TRACER_SIZE
+1];
3257 if (cnt
> MAX_TRACER_SIZE
)
3258 cnt
= MAX_TRACER_SIZE
;
3260 if (copy_from_user(&buf
, ubuf
, cnt
))
3265 /* strip ending whitespace. */
3266 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
3269 err
= tracing_set_tracer(buf
);
3279 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
3280 size_t cnt
, loff_t
*ppos
)
3282 unsigned long *ptr
= filp
->private_data
;
3286 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
3287 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
3288 if (r
> sizeof(buf
))
3290 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3294 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
3295 size_t cnt
, loff_t
*ppos
)
3297 unsigned long *ptr
= filp
->private_data
;
3301 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3310 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
3312 long cpu_file
= (long) inode
->i_private
;
3313 struct trace_iterator
*iter
;
3316 if (tracing_disabled
)
3319 mutex_lock(&trace_types_lock
);
3321 /* create a buffer to store the information to pass to userspace */
3322 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
3329 * We make a copy of the current tracer to avoid concurrent
3330 * changes on it while we are reading.
3332 iter
->trace
= kmalloc(sizeof(*iter
->trace
), GFP_KERNEL
);
3338 *iter
->trace
= *current_trace
;
3340 if (!alloc_cpumask_var(&iter
->started
, GFP_KERNEL
)) {
3345 /* trace pipe does not show start of buffer */
3346 cpumask_setall(iter
->started
);
3348 if (trace_flags
& TRACE_ITER_LATENCY_FMT
)
3349 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
3351 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
3352 if (trace_clocks
[trace_clock_id
].in_ns
)
3353 iter
->iter_flags
|= TRACE_FILE_TIME_IN_NS
;
3355 iter
->cpu_file
= cpu_file
;
3356 iter
->tr
= &global_trace
;
3357 mutex_init(&iter
->mutex
);
3358 filp
->private_data
= iter
;
3360 if (iter
->trace
->pipe_open
)
3361 iter
->trace
->pipe_open(iter
);
3363 nonseekable_open(inode
, filp
);
3365 mutex_unlock(&trace_types_lock
);
3371 mutex_unlock(&trace_types_lock
);
3375 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
3377 struct trace_iterator
*iter
= file
->private_data
;
3379 mutex_lock(&trace_types_lock
);
3381 if (iter
->trace
->pipe_close
)
3382 iter
->trace
->pipe_close(iter
);
3384 mutex_unlock(&trace_types_lock
);
3386 free_cpumask_var(iter
->started
);
3387 mutex_destroy(&iter
->mutex
);
3395 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
3397 struct trace_iterator
*iter
= filp
->private_data
;
3399 if (trace_flags
& TRACE_ITER_BLOCK
) {
3401 * Always select as readable when in blocking mode
3403 return POLLIN
| POLLRDNORM
;
3405 if (!trace_empty(iter
))
3406 return POLLIN
| POLLRDNORM
;
3407 poll_wait(filp
, &trace_wait
, poll_table
);
3408 if (!trace_empty(iter
))
3409 return POLLIN
| POLLRDNORM
;
3416 * This is a make-shift waitqueue.
3417 * A tracer might use this callback on some rare cases:
3419 * 1) the current tracer might hold the runqueue lock when it wakes up
3420 * a reader, hence a deadlock (sched, function, and function graph tracers)
3421 * 2) the function tracers, trace all functions, we don't want
3422 * the overhead of calling wake_up and friends
3423 * (and tracing them too)
3425 * Anyway, this is really very primitive wakeup.
3427 void poll_wait_pipe(struct trace_iterator
*iter
)
3429 set_current_state(TASK_INTERRUPTIBLE
);
3430 /* sleep for 100 msecs, and try again. */
3431 schedule_timeout(HZ
/ 10);
3434 /* Must be called with trace_types_lock mutex held. */
3435 static int tracing_wait_pipe(struct file
*filp
)
3437 struct trace_iterator
*iter
= filp
->private_data
;
3439 while (trace_empty(iter
)) {
3441 if ((filp
->f_flags
& O_NONBLOCK
)) {
3445 mutex_unlock(&iter
->mutex
);
3447 iter
->trace
->wait_pipe(iter
);
3449 mutex_lock(&iter
->mutex
);
3451 if (signal_pending(current
))
3455 * We block until we read something and tracing is enabled.
3456 * We still block if tracing is disabled, but we have never
3457 * read anything. This allows a user to cat this file, and
3458 * then enable tracing. But after we have read something,
3459 * we give an EOF when tracing is again disabled.
3461 * iter->pos will be 0 if we haven't read anything.
3463 if (tracing_is_enabled() && iter
->pos
)
3474 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
3475 size_t cnt
, loff_t
*ppos
)
3477 struct trace_iterator
*iter
= filp
->private_data
;
3478 static struct tracer
*old_tracer
;
3481 /* return any leftover data */
3482 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3486 trace_seq_init(&iter
->seq
);
3488 /* copy the tracer to avoid using a global lock all around */
3489 mutex_lock(&trace_types_lock
);
3490 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3491 old_tracer
= current_trace
;
3492 *iter
->trace
= *current_trace
;
3494 mutex_unlock(&trace_types_lock
);
3497 * Avoid more than one consumer on a single file descriptor
3498 * This is just a matter of traces coherency, the ring buffer itself
3501 mutex_lock(&iter
->mutex
);
3502 if (iter
->trace
->read
) {
3503 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
3509 sret
= tracing_wait_pipe(filp
);
3513 /* stop when tracing is finished */
3514 if (trace_empty(iter
)) {
3519 if (cnt
>= PAGE_SIZE
)
3520 cnt
= PAGE_SIZE
- 1;
3522 /* reset all but tr, trace, and overruns */
3523 memset(&iter
->seq
, 0,
3524 sizeof(struct trace_iterator
) -
3525 offsetof(struct trace_iterator
, seq
));
3528 trace_event_read_lock();
3529 trace_access_lock(iter
->cpu_file
);
3530 while (trace_find_next_entry_inc(iter
) != NULL
) {
3531 enum print_line_t ret
;
3532 int len
= iter
->seq
.len
;
3534 ret
= print_trace_line(iter
);
3535 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3536 /* don't print partial lines */
3537 iter
->seq
.len
= len
;
3540 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3541 trace_consume(iter
);
3543 if (iter
->seq
.len
>= cnt
)
3547 * Setting the full flag means we reached the trace_seq buffer
3548 * size and we should leave by partial output condition above.
3549 * One of the trace_seq_* functions is not used properly.
3551 WARN_ONCE(iter
->seq
.full
, "full flag set for trace type %d",
3554 trace_access_unlock(iter
->cpu_file
);
3555 trace_event_read_unlock();
3557 /* Now copy what we have to the user */
3558 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
3559 if (iter
->seq
.readpos
>= iter
->seq
.len
)
3560 trace_seq_init(&iter
->seq
);
3563 * If there was nothing to send to user, in spite of consuming trace
3564 * entries, go back to wait for more entries.
3570 mutex_unlock(&iter
->mutex
);
3575 static void tracing_pipe_buf_release(struct pipe_inode_info
*pipe
,
3576 struct pipe_buffer
*buf
)
3578 __free_page(buf
->page
);
3581 static void tracing_spd_release_pipe(struct splice_pipe_desc
*spd
,
3584 __free_page(spd
->pages
[idx
]);
3587 static const struct pipe_buf_operations tracing_pipe_buf_ops
= {
3589 .map
= generic_pipe_buf_map
,
3590 .unmap
= generic_pipe_buf_unmap
,
3591 .confirm
= generic_pipe_buf_confirm
,
3592 .release
= tracing_pipe_buf_release
,
3593 .steal
= generic_pipe_buf_steal
,
3594 .get
= generic_pipe_buf_get
,
3598 tracing_fill_pipe_page(size_t rem
, struct trace_iterator
*iter
)
3603 /* Seq buffer is page-sized, exactly what we need. */
3605 count
= iter
->seq
.len
;
3606 ret
= print_trace_line(iter
);
3607 count
= iter
->seq
.len
- count
;
3610 iter
->seq
.len
-= count
;
3613 if (ret
== TRACE_TYPE_PARTIAL_LINE
) {
3614 iter
->seq
.len
-= count
;
3618 if (ret
!= TRACE_TYPE_NO_CONSUME
)
3619 trace_consume(iter
);
3621 if (!trace_find_next_entry_inc(iter
)) {
3631 static ssize_t
tracing_splice_read_pipe(struct file
*filp
,
3633 struct pipe_inode_info
*pipe
,
3637 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
3638 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
3639 struct trace_iterator
*iter
= filp
->private_data
;
3640 struct splice_pipe_desc spd
= {
3642 .partial
= partial_def
,
3643 .nr_pages
= 0, /* This gets updated below. */
3644 .nr_pages_max
= PIPE_DEF_BUFFERS
,
3646 .ops
= &tracing_pipe_buf_ops
,
3647 .spd_release
= tracing_spd_release_pipe
,
3649 static struct tracer
*old_tracer
;
3654 if (splice_grow_spd(pipe
, &spd
))
3657 /* copy the tracer to avoid using a global lock all around */
3658 mutex_lock(&trace_types_lock
);
3659 if (unlikely(old_tracer
!= current_trace
&& current_trace
)) {
3660 old_tracer
= current_trace
;
3661 *iter
->trace
= *current_trace
;
3663 mutex_unlock(&trace_types_lock
);
3665 mutex_lock(&iter
->mutex
);
3667 if (iter
->trace
->splice_read
) {
3668 ret
= iter
->trace
->splice_read(iter
, filp
,
3669 ppos
, pipe
, len
, flags
);
3674 ret
= tracing_wait_pipe(filp
);
3678 if (!iter
->ent
&& !trace_find_next_entry_inc(iter
)) {
3683 trace_event_read_lock();
3684 trace_access_lock(iter
->cpu_file
);
3686 /* Fill as many pages as possible. */
3687 for (i
= 0, rem
= len
; i
< pipe
->buffers
&& rem
; i
++) {
3688 spd
.pages
[i
] = alloc_page(GFP_KERNEL
);
3692 rem
= tracing_fill_pipe_page(rem
, iter
);
3694 /* Copy the data into the page, so we can start over. */
3695 ret
= trace_seq_to_buffer(&iter
->seq
,
3696 page_address(spd
.pages
[i
]),
3699 __free_page(spd
.pages
[i
]);
3702 spd
.partial
[i
].offset
= 0;
3703 spd
.partial
[i
].len
= iter
->seq
.len
;
3705 trace_seq_init(&iter
->seq
);
3708 trace_access_unlock(iter
->cpu_file
);
3709 trace_event_read_unlock();
3710 mutex_unlock(&iter
->mutex
);
3714 ret
= splice_to_pipe(pipe
, &spd
);
3716 splice_shrink_spd(&spd
);
3720 mutex_unlock(&iter
->mutex
);
3724 struct ftrace_entries_info
{
3725 struct trace_array
*tr
;
3729 static int tracing_entries_open(struct inode
*inode
, struct file
*filp
)
3731 struct ftrace_entries_info
*info
;
3733 if (tracing_disabled
)
3736 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
3740 info
->tr
= &global_trace
;
3741 info
->cpu
= (unsigned long)inode
->i_private
;
3743 filp
->private_data
= info
;
3749 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
3750 size_t cnt
, loff_t
*ppos
)
3752 struct ftrace_entries_info
*info
= filp
->private_data
;
3753 struct trace_array
*tr
= info
->tr
;
3758 mutex_lock(&trace_types_lock
);
3760 if (info
->cpu
== RING_BUFFER_ALL_CPUS
) {
3761 int cpu
, buf_size_same
;
3766 /* check if all cpu sizes are same */
3767 for_each_tracing_cpu(cpu
) {
3768 /* fill in the size from first enabled cpu */
3770 size
= tr
->data
[cpu
]->entries
;
3771 if (size
!= tr
->data
[cpu
]->entries
) {
3777 if (buf_size_same
) {
3778 if (!ring_buffer_expanded
)
3779 r
= sprintf(buf
, "%lu (expanded: %lu)\n",
3781 trace_buf_size
>> 10);
3783 r
= sprintf(buf
, "%lu\n", size
>> 10);
3785 r
= sprintf(buf
, "X\n");
3787 r
= sprintf(buf
, "%lu\n", tr
->data
[info
->cpu
]->entries
>> 10);
3789 mutex_unlock(&trace_types_lock
);
3791 ret
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3796 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
3797 size_t cnt
, loff_t
*ppos
)
3799 struct ftrace_entries_info
*info
= filp
->private_data
;
3803 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
3807 /* must have at least 1 entry */
3811 /* value is in KB */
3814 ret
= tracing_resize_ring_buffer(val
, info
->cpu
);
3824 tracing_entries_release(struct inode
*inode
, struct file
*filp
)
3826 struct ftrace_entries_info
*info
= filp
->private_data
;
3834 tracing_total_entries_read(struct file
*filp
, char __user
*ubuf
,
3835 size_t cnt
, loff_t
*ppos
)
3837 struct trace_array
*tr
= filp
->private_data
;
3840 unsigned long size
= 0, expanded_size
= 0;
3842 mutex_lock(&trace_types_lock
);
3843 for_each_tracing_cpu(cpu
) {
3844 size
+= tr
->data
[cpu
]->entries
>> 10;
3845 if (!ring_buffer_expanded
)
3846 expanded_size
+= trace_buf_size
>> 10;
3848 if (ring_buffer_expanded
)
3849 r
= sprintf(buf
, "%lu\n", size
);
3851 r
= sprintf(buf
, "%lu (expanded: %lu)\n", size
, expanded_size
);
3852 mutex_unlock(&trace_types_lock
);
3854 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3858 tracing_free_buffer_write(struct file
*filp
, const char __user
*ubuf
,
3859 size_t cnt
, loff_t
*ppos
)
3862 * There is no need to read what the user has written, this function
3863 * is just to make sure that there is no error when "echo" is used
3872 tracing_free_buffer_release(struct inode
*inode
, struct file
*filp
)
3874 /* disable tracing ? */
3875 if (trace_flags
& TRACE_ITER_STOP_ON_FREE
)
3877 /* resize the ring buffer to 0 */
3878 tracing_resize_ring_buffer(0, RING_BUFFER_ALL_CPUS
);
3884 tracing_mark_write(struct file
*filp
, const char __user
*ubuf
,
3885 size_t cnt
, loff_t
*fpos
)
3887 unsigned long addr
= (unsigned long)ubuf
;
3888 struct ring_buffer_event
*event
;
3889 struct ring_buffer
*buffer
;
3890 struct print_entry
*entry
;
3891 unsigned long irq_flags
;
3892 struct page
*pages
[2];
3902 if (tracing_disabled
)
3905 if (!(trace_flags
& TRACE_ITER_MARKERS
))
3908 if (cnt
> TRACE_BUF_SIZE
)
3909 cnt
= TRACE_BUF_SIZE
;
3912 * Userspace is injecting traces into the kernel trace buffer.
3913 * We want to be as non intrusive as possible.
3914 * To do so, we do not want to allocate any special buffers
3915 * or take any locks, but instead write the userspace data
3916 * straight into the ring buffer.
3918 * First we need to pin the userspace buffer into memory,
3919 * which, most likely it is, because it just referenced it.
3920 * But there's no guarantee that it is. By using get_user_pages_fast()
3921 * and kmap_atomic/kunmap_atomic() we can get access to the
3922 * pages directly. We then write the data directly into the
3925 BUILD_BUG_ON(TRACE_BUF_SIZE
>= PAGE_SIZE
);
3927 /* check if we cross pages */
3928 if ((addr
& PAGE_MASK
) != ((addr
+ cnt
) & PAGE_MASK
))
3931 offset
= addr
& (PAGE_SIZE
- 1);
3934 ret
= get_user_pages_fast(addr
, nr_pages
, 0, pages
);
3935 if (ret
< nr_pages
) {
3937 put_page(pages
[ret
]);
3942 for (i
= 0; i
< nr_pages
; i
++)
3943 map_page
[i
] = kmap_atomic(pages
[i
]);
3945 local_save_flags(irq_flags
);
3946 size
= sizeof(*entry
) + cnt
+ 2; /* possible \n added */
3947 buffer
= global_trace
.buffer
;
3948 event
= trace_buffer_lock_reserve(buffer
, TRACE_PRINT
, size
,
3949 irq_flags
, preempt_count());
3951 /* Ring buffer disabled, return as if not open for write */
3956 entry
= ring_buffer_event_data(event
);
3957 entry
->ip
= _THIS_IP_
;
3959 if (nr_pages
== 2) {
3960 len
= PAGE_SIZE
- offset
;
3961 memcpy(&entry
->buf
, map_page
[0] + offset
, len
);
3962 memcpy(&entry
->buf
[len
], map_page
[1], cnt
- len
);
3964 memcpy(&entry
->buf
, map_page
[0] + offset
, cnt
);
3966 if (entry
->buf
[cnt
- 1] != '\n') {
3967 entry
->buf
[cnt
] = '\n';
3968 entry
->buf
[cnt
+ 1] = '\0';
3970 entry
->buf
[cnt
] = '\0';
3972 __buffer_unlock_commit(buffer
, event
);
3979 for (i
= 0; i
< nr_pages
; i
++){
3980 kunmap_atomic(map_page
[i
]);
3987 static int tracing_clock_show(struct seq_file
*m
, void *v
)
3991 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++)
3993 "%s%s%s%s", i
? " " : "",
3994 i
== trace_clock_id
? "[" : "", trace_clocks
[i
].name
,
3995 i
== trace_clock_id
? "]" : "");
4001 static ssize_t
tracing_clock_write(struct file
*filp
, const char __user
*ubuf
,
4002 size_t cnt
, loff_t
*fpos
)
4005 const char *clockstr
;
4008 if (cnt
>= sizeof(buf
))
4011 if (copy_from_user(&buf
, ubuf
, cnt
))
4016 clockstr
= strstrip(buf
);
4018 for (i
= 0; i
< ARRAY_SIZE(trace_clocks
); i
++) {
4019 if (strcmp(trace_clocks
[i
].name
, clockstr
) == 0)
4022 if (i
== ARRAY_SIZE(trace_clocks
))
4027 mutex_lock(&trace_types_lock
);
4029 ring_buffer_set_clock(global_trace
.buffer
, trace_clocks
[i
].func
);
4031 ring_buffer_set_clock(max_tr
.buffer
, trace_clocks
[i
].func
);
4034 * New clock may not be consistent with the previous clock.
4035 * Reset the buffer so that it doesn't have incomparable timestamps.
4037 tracing_reset_online_cpus(&global_trace
);
4039 tracing_reset_online_cpus(&max_tr
);
4041 mutex_unlock(&trace_types_lock
);
4048 static int tracing_clock_open(struct inode
*inode
, struct file
*file
)
4050 if (tracing_disabled
)
4052 return single_open(file
, tracing_clock_show
, NULL
);
4055 static const struct file_operations tracing_max_lat_fops
= {
4056 .open
= tracing_open_generic
,
4057 .read
= tracing_max_lat_read
,
4058 .write
= tracing_max_lat_write
,
4059 .llseek
= generic_file_llseek
,
4062 static const struct file_operations set_tracer_fops
= {
4063 .open
= tracing_open_generic
,
4064 .read
= tracing_set_trace_read
,
4065 .write
= tracing_set_trace_write
,
4066 .llseek
= generic_file_llseek
,
4069 static const struct file_operations tracing_pipe_fops
= {
4070 .open
= tracing_open_pipe
,
4071 .poll
= tracing_poll_pipe
,
4072 .read
= tracing_read_pipe
,
4073 .splice_read
= tracing_splice_read_pipe
,
4074 .release
= tracing_release_pipe
,
4075 .llseek
= no_llseek
,
4078 static const struct file_operations tracing_entries_fops
= {
4079 .open
= tracing_entries_open
,
4080 .read
= tracing_entries_read
,
4081 .write
= tracing_entries_write
,
4082 .release
= tracing_entries_release
,
4083 .llseek
= generic_file_llseek
,
4086 static const struct file_operations tracing_total_entries_fops
= {
4087 .open
= tracing_open_generic
,
4088 .read
= tracing_total_entries_read
,
4089 .llseek
= generic_file_llseek
,
4092 static const struct file_operations tracing_free_buffer_fops
= {
4093 .write
= tracing_free_buffer_write
,
4094 .release
= tracing_free_buffer_release
,
4097 static const struct file_operations tracing_mark_fops
= {
4098 .open
= tracing_open_generic
,
4099 .write
= tracing_mark_write
,
4100 .llseek
= generic_file_llseek
,
4103 static const struct file_operations trace_clock_fops
= {
4104 .open
= tracing_clock_open
,
4106 .llseek
= seq_lseek
,
4107 .release
= single_release
,
4108 .write
= tracing_clock_write
,
4111 struct ftrace_buffer_info
{
4112 struct trace_array
*tr
;
4118 static int tracing_buffers_open(struct inode
*inode
, struct file
*filp
)
4120 int cpu
= (int)(long)inode
->i_private
;
4121 struct ftrace_buffer_info
*info
;
4123 if (tracing_disabled
)
4126 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
4130 info
->tr
= &global_trace
;
4133 /* Force reading ring buffer for first read */
4134 info
->read
= (unsigned int)-1;
4136 filp
->private_data
= info
;
4138 return nonseekable_open(inode
, filp
);
4142 tracing_buffers_read(struct file
*filp
, char __user
*ubuf
,
4143 size_t count
, loff_t
*ppos
)
4145 struct ftrace_buffer_info
*info
= filp
->private_data
;
4153 info
->spare
= ring_buffer_alloc_read_page(info
->tr
->buffer
, info
->cpu
);
4157 /* Do we have previous read data to read? */
4158 if (info
->read
< PAGE_SIZE
)
4161 trace_access_lock(info
->cpu
);
4162 ret
= ring_buffer_read_page(info
->tr
->buffer
,
4166 trace_access_unlock(info
->cpu
);
4173 size
= PAGE_SIZE
- info
->read
;
4177 ret
= copy_to_user(ubuf
, info
->spare
+ info
->read
, size
);
4188 static int tracing_buffers_release(struct inode
*inode
, struct file
*file
)
4190 struct ftrace_buffer_info
*info
= file
->private_data
;
4193 ring_buffer_free_read_page(info
->tr
->buffer
, info
->spare
);
4200 struct ring_buffer
*buffer
;
4205 static void buffer_pipe_buf_release(struct pipe_inode_info
*pipe
,
4206 struct pipe_buffer
*buf
)
4208 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
4213 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4218 static void buffer_pipe_buf_get(struct pipe_inode_info
*pipe
,
4219 struct pipe_buffer
*buf
)
4221 struct buffer_ref
*ref
= (struct buffer_ref
*)buf
->private;
4226 /* Pipe buffer operations for a buffer. */
4227 static const struct pipe_buf_operations buffer_pipe_buf_ops
= {
4229 .map
= generic_pipe_buf_map
,
4230 .unmap
= generic_pipe_buf_unmap
,
4231 .confirm
= generic_pipe_buf_confirm
,
4232 .release
= buffer_pipe_buf_release
,
4233 .steal
= generic_pipe_buf_steal
,
4234 .get
= buffer_pipe_buf_get
,
4238 * Callback from splice_to_pipe(), if we need to release some pages
4239 * at the end of the spd in case we error'ed out in filling the pipe.
4241 static void buffer_spd_release(struct splice_pipe_desc
*spd
, unsigned int i
)
4243 struct buffer_ref
*ref
=
4244 (struct buffer_ref
*)spd
->partial
[i
].private;
4249 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4251 spd
->partial
[i
].private = 0;
4255 tracing_buffers_splice_read(struct file
*file
, loff_t
*ppos
,
4256 struct pipe_inode_info
*pipe
, size_t len
,
4259 struct ftrace_buffer_info
*info
= file
->private_data
;
4260 struct partial_page partial_def
[PIPE_DEF_BUFFERS
];
4261 struct page
*pages_def
[PIPE_DEF_BUFFERS
];
4262 struct splice_pipe_desc spd
= {
4264 .partial
= partial_def
,
4265 .nr_pages_max
= PIPE_DEF_BUFFERS
,
4267 .ops
= &buffer_pipe_buf_ops
,
4268 .spd_release
= buffer_spd_release
,
4270 struct buffer_ref
*ref
;
4271 int entries
, size
, i
;
4274 if (splice_grow_spd(pipe
, &spd
))
4277 if (*ppos
& (PAGE_SIZE
- 1)) {
4282 if (len
& (PAGE_SIZE
- 1)) {
4283 if (len
< PAGE_SIZE
) {
4290 trace_access_lock(info
->cpu
);
4291 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4293 for (i
= 0; i
< pipe
->buffers
&& len
&& entries
; i
++, len
-= PAGE_SIZE
) {
4297 ref
= kzalloc(sizeof(*ref
), GFP_KERNEL
);
4302 ref
->buffer
= info
->tr
->buffer
;
4303 ref
->page
= ring_buffer_alloc_read_page(ref
->buffer
, info
->cpu
);
4309 r
= ring_buffer_read_page(ref
->buffer
, &ref
->page
,
4312 ring_buffer_free_read_page(ref
->buffer
, ref
->page
);
4318 * zero out any left over data, this is going to
4321 size
= ring_buffer_page_len(ref
->page
);
4322 if (size
< PAGE_SIZE
)
4323 memset(ref
->page
+ size
, 0, PAGE_SIZE
- size
);
4325 page
= virt_to_page(ref
->page
);
4327 spd
.pages
[i
] = page
;
4328 spd
.partial
[i
].len
= PAGE_SIZE
;
4329 spd
.partial
[i
].offset
= 0;
4330 spd
.partial
[i
].private = (unsigned long)ref
;
4334 entries
= ring_buffer_entries_cpu(info
->tr
->buffer
, info
->cpu
);
4337 trace_access_unlock(info
->cpu
);
4340 /* did we read anything? */
4341 if (!spd
.nr_pages
) {
4342 if (flags
& SPLICE_F_NONBLOCK
)
4350 ret
= splice_to_pipe(pipe
, &spd
);
4351 splice_shrink_spd(&spd
);
4356 static const struct file_operations tracing_buffers_fops
= {
4357 .open
= tracing_buffers_open
,
4358 .read
= tracing_buffers_read
,
4359 .release
= tracing_buffers_release
,
4360 .splice_read
= tracing_buffers_splice_read
,
4361 .llseek
= no_llseek
,
4365 tracing_stats_read(struct file
*filp
, char __user
*ubuf
,
4366 size_t count
, loff_t
*ppos
)
4368 unsigned long cpu
= (unsigned long)filp
->private_data
;
4369 struct trace_array
*tr
= &global_trace
;
4370 struct trace_seq
*s
;
4372 unsigned long long t
;
4373 unsigned long usec_rem
;
4375 s
= kmalloc(sizeof(*s
), GFP_KERNEL
);
4381 cnt
= ring_buffer_entries_cpu(tr
->buffer
, cpu
);
4382 trace_seq_printf(s
, "entries: %ld\n", cnt
);
4384 cnt
= ring_buffer_overrun_cpu(tr
->buffer
, cpu
);
4385 trace_seq_printf(s
, "overrun: %ld\n", cnt
);
4387 cnt
= ring_buffer_commit_overrun_cpu(tr
->buffer
, cpu
);
4388 trace_seq_printf(s
, "commit overrun: %ld\n", cnt
);
4390 cnt
= ring_buffer_bytes_cpu(tr
->buffer
, cpu
);
4391 trace_seq_printf(s
, "bytes: %ld\n", cnt
);
4393 if (trace_clocks
[trace_clock_id
].in_ns
) {
4394 /* local or global for trace_clock */
4395 t
= ns2usecs(ring_buffer_oldest_event_ts(tr
->buffer
, cpu
));
4396 usec_rem
= do_div(t
, USEC_PER_SEC
);
4397 trace_seq_printf(s
, "oldest event ts: %5llu.%06lu\n",
4400 t
= ns2usecs(ring_buffer_time_stamp(tr
->buffer
, cpu
));
4401 usec_rem
= do_div(t
, USEC_PER_SEC
);
4402 trace_seq_printf(s
, "now ts: %5llu.%06lu\n", t
, usec_rem
);
4404 /* counter or tsc mode for trace_clock */
4405 trace_seq_printf(s
, "oldest event ts: %llu\n",
4406 ring_buffer_oldest_event_ts(tr
->buffer
, cpu
));
4408 trace_seq_printf(s
, "now ts: %llu\n",
4409 ring_buffer_time_stamp(tr
->buffer
, cpu
));
4412 cnt
= ring_buffer_dropped_events_cpu(tr
->buffer
, cpu
);
4413 trace_seq_printf(s
, "dropped events: %ld\n", cnt
);
4415 count
= simple_read_from_buffer(ubuf
, count
, ppos
, s
->buffer
, s
->len
);
4422 static const struct file_operations tracing_stats_fops
= {
4423 .open
= tracing_open_generic
,
4424 .read
= tracing_stats_read
,
4425 .llseek
= generic_file_llseek
,
4428 #ifdef CONFIG_DYNAMIC_FTRACE
4430 int __weak
ftrace_arch_read_dyn_info(char *buf
, int size
)
4436 tracing_read_dyn_info(struct file
*filp
, char __user
*ubuf
,
4437 size_t cnt
, loff_t
*ppos
)
4439 static char ftrace_dyn_info_buffer
[1024];
4440 static DEFINE_MUTEX(dyn_info_mutex
);
4441 unsigned long *p
= filp
->private_data
;
4442 char *buf
= ftrace_dyn_info_buffer
;
4443 int size
= ARRAY_SIZE(ftrace_dyn_info_buffer
);
4446 mutex_lock(&dyn_info_mutex
);
4447 r
= sprintf(buf
, "%ld ", *p
);
4449 r
+= ftrace_arch_read_dyn_info(buf
+r
, (size
-1)-r
);
4452 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4454 mutex_unlock(&dyn_info_mutex
);
4459 static const struct file_operations tracing_dyn_info_fops
= {
4460 .open
= tracing_open_generic
,
4461 .read
= tracing_read_dyn_info
,
4462 .llseek
= generic_file_llseek
,
4466 static struct dentry
*d_tracer
;
4468 struct dentry
*tracing_init_dentry(void)
4475 if (!debugfs_initialized())
4478 d_tracer
= debugfs_create_dir("tracing", NULL
);
4480 if (!d_tracer
&& !once
) {
4482 pr_warning("Could not create debugfs directory 'tracing'\n");
4489 static struct dentry
*d_percpu
;
4491 struct dentry
*tracing_dentry_percpu(void)
4494 struct dentry
*d_tracer
;
4499 d_tracer
= tracing_init_dentry();
4504 d_percpu
= debugfs_create_dir("per_cpu", d_tracer
);
4506 if (!d_percpu
&& !once
) {
4508 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4515 static void tracing_init_debugfs_percpu(long cpu
)
4517 struct dentry
*d_percpu
= tracing_dentry_percpu();
4518 struct dentry
*d_cpu
;
4519 char cpu_dir
[30]; /* 30 characters should be more than enough */
4524 snprintf(cpu_dir
, 30, "cpu%ld", cpu
);
4525 d_cpu
= debugfs_create_dir(cpu_dir
, d_percpu
);
4527 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir
);
4531 /* per cpu trace_pipe */
4532 trace_create_file("trace_pipe", 0444, d_cpu
,
4533 (void *) cpu
, &tracing_pipe_fops
);
4536 trace_create_file("trace", 0644, d_cpu
,
4537 (void *) cpu
, &tracing_fops
);
4539 trace_create_file("trace_pipe_raw", 0444, d_cpu
,
4540 (void *) cpu
, &tracing_buffers_fops
);
4542 trace_create_file("stats", 0444, d_cpu
,
4543 (void *) cpu
, &tracing_stats_fops
);
4545 trace_create_file("buffer_size_kb", 0444, d_cpu
,
4546 (void *) cpu
, &tracing_entries_fops
);
4549 #ifdef CONFIG_FTRACE_SELFTEST
4550 /* Let selftest have access to static functions in this file */
4551 #include "trace_selftest.c"
4554 struct trace_option_dentry
{
4555 struct tracer_opt
*opt
;
4556 struct tracer_flags
*flags
;
4557 struct dentry
*entry
;
4561 trace_options_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4564 struct trace_option_dentry
*topt
= filp
->private_data
;
4567 if (topt
->flags
->val
& topt
->opt
->bit
)
4572 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4576 trace_options_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4579 struct trace_option_dentry
*topt
= filp
->private_data
;
4583 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4587 if (val
!= 0 && val
!= 1)
4590 if (!!(topt
->flags
->val
& topt
->opt
->bit
) != val
) {
4591 mutex_lock(&trace_types_lock
);
4592 ret
= __set_tracer_option(current_trace
, topt
->flags
,
4594 mutex_unlock(&trace_types_lock
);
4605 static const struct file_operations trace_options_fops
= {
4606 .open
= tracing_open_generic
,
4607 .read
= trace_options_read
,
4608 .write
= trace_options_write
,
4609 .llseek
= generic_file_llseek
,
4613 trace_options_core_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
4616 long index
= (long)filp
->private_data
;
4619 if (trace_flags
& (1 << index
))
4624 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, 2);
4628 trace_options_core_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
4631 long index
= (long)filp
->private_data
;
4635 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4639 if (val
!= 0 && val
!= 1)
4641 set_tracer_flags(1 << index
, val
);
4648 static const struct file_operations trace_options_core_fops
= {
4649 .open
= tracing_open_generic
,
4650 .read
= trace_options_core_read
,
4651 .write
= trace_options_core_write
,
4652 .llseek
= generic_file_llseek
,
4655 struct dentry
*trace_create_file(const char *name
,
4657 struct dentry
*parent
,
4659 const struct file_operations
*fops
)
4663 ret
= debugfs_create_file(name
, mode
, parent
, data
, fops
);
4665 pr_warning("Could not create debugfs '%s' entry\n", name
);
4671 static struct dentry
*trace_options_init_dentry(void)
4673 struct dentry
*d_tracer
;
4674 static struct dentry
*t_options
;
4679 d_tracer
= tracing_init_dentry();
4683 t_options
= debugfs_create_dir("options", d_tracer
);
4685 pr_warning("Could not create debugfs directory 'options'\n");
4693 create_trace_option_file(struct trace_option_dentry
*topt
,
4694 struct tracer_flags
*flags
,
4695 struct tracer_opt
*opt
)
4697 struct dentry
*t_options
;
4699 t_options
= trace_options_init_dentry();
4703 topt
->flags
= flags
;
4706 topt
->entry
= trace_create_file(opt
->name
, 0644, t_options
, topt
,
4707 &trace_options_fops
);
4711 static struct trace_option_dentry
*
4712 create_trace_option_files(struct tracer
*tracer
)
4714 struct trace_option_dentry
*topts
;
4715 struct tracer_flags
*flags
;
4716 struct tracer_opt
*opts
;
4722 flags
= tracer
->flags
;
4724 if (!flags
|| !flags
->opts
)
4729 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4732 topts
= kcalloc(cnt
+ 1, sizeof(*topts
), GFP_KERNEL
);
4736 for (cnt
= 0; opts
[cnt
].name
; cnt
++)
4737 create_trace_option_file(&topts
[cnt
], flags
,
4744 destroy_trace_option_files(struct trace_option_dentry
*topts
)
4751 for (cnt
= 0; topts
[cnt
].opt
; cnt
++) {
4752 if (topts
[cnt
].entry
)
4753 debugfs_remove(topts
[cnt
].entry
);
4759 static struct dentry
*
4760 create_trace_option_core_file(const char *option
, long index
)
4762 struct dentry
*t_options
;
4764 t_options
= trace_options_init_dentry();
4768 return trace_create_file(option
, 0644, t_options
, (void *)index
,
4769 &trace_options_core_fops
);
4772 static __init
void create_trace_options_dir(void)
4774 struct dentry
*t_options
;
4777 t_options
= trace_options_init_dentry();
4781 for (i
= 0; trace_options
[i
]; i
++)
4782 create_trace_option_core_file(trace_options
[i
], i
);
4786 rb_simple_read(struct file
*filp
, char __user
*ubuf
,
4787 size_t cnt
, loff_t
*ppos
)
4789 struct trace_array
*tr
= filp
->private_data
;
4790 struct ring_buffer
*buffer
= tr
->buffer
;
4795 r
= ring_buffer_record_is_on(buffer
);
4799 r
= sprintf(buf
, "%d\n", r
);
4801 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
4805 rb_simple_write(struct file
*filp
, const char __user
*ubuf
,
4806 size_t cnt
, loff_t
*ppos
)
4808 struct trace_array
*tr
= filp
->private_data
;
4809 struct ring_buffer
*buffer
= tr
->buffer
;
4813 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
4819 ring_buffer_record_on(buffer
);
4821 ring_buffer_record_off(buffer
);
4829 static const struct file_operations rb_simple_fops
= {
4830 .open
= tracing_open_generic
,
4831 .read
= rb_simple_read
,
4832 .write
= rb_simple_write
,
4833 .llseek
= default_llseek
,
4836 static __init
int tracer_init_debugfs(void)
4838 struct dentry
*d_tracer
;
4841 trace_access_lock_init();
4843 d_tracer
= tracing_init_dentry();
4845 trace_create_file("trace_options", 0644, d_tracer
,
4846 NULL
, &tracing_iter_fops
);
4848 trace_create_file("tracing_cpumask", 0644, d_tracer
,
4849 NULL
, &tracing_cpumask_fops
);
4851 trace_create_file("trace", 0644, d_tracer
,
4852 (void *) TRACE_PIPE_ALL_CPU
, &tracing_fops
);
4854 trace_create_file("available_tracers", 0444, d_tracer
,
4855 &global_trace
, &show_traces_fops
);
4857 trace_create_file("current_tracer", 0644, d_tracer
,
4858 &global_trace
, &set_tracer_fops
);
4860 #ifdef CONFIG_TRACER_MAX_TRACE
4861 trace_create_file("tracing_max_latency", 0644, d_tracer
,
4862 &tracing_max_latency
, &tracing_max_lat_fops
);
4865 trace_create_file("tracing_thresh", 0644, d_tracer
,
4866 &tracing_thresh
, &tracing_max_lat_fops
);
4868 trace_create_file("README", 0444, d_tracer
,
4869 NULL
, &tracing_readme_fops
);
4871 trace_create_file("trace_pipe", 0444, d_tracer
,
4872 (void *) TRACE_PIPE_ALL_CPU
, &tracing_pipe_fops
);
4874 trace_create_file("buffer_size_kb", 0644, d_tracer
,
4875 (void *) RING_BUFFER_ALL_CPUS
, &tracing_entries_fops
);
4877 trace_create_file("buffer_total_size_kb", 0444, d_tracer
,
4878 &global_trace
, &tracing_total_entries_fops
);
4880 trace_create_file("free_buffer", 0644, d_tracer
,
4881 &global_trace
, &tracing_free_buffer_fops
);
4883 trace_create_file("trace_marker", 0220, d_tracer
,
4884 NULL
, &tracing_mark_fops
);
4886 trace_create_file("saved_cmdlines", 0444, d_tracer
,
4887 NULL
, &tracing_saved_cmdlines_fops
);
4889 trace_create_file("trace_clock", 0644, d_tracer
, NULL
,
4892 trace_create_file("tracing_on", 0644, d_tracer
,
4893 &global_trace
, &rb_simple_fops
);
4895 #ifdef CONFIG_DYNAMIC_FTRACE
4896 trace_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
4897 &ftrace_update_tot_cnt
, &tracing_dyn_info_fops
);
4900 create_trace_options_dir();
4902 for_each_tracing_cpu(cpu
)
4903 tracing_init_debugfs_percpu(cpu
);
4908 static int trace_panic_handler(struct notifier_block
*this,
4909 unsigned long event
, void *unused
)
4911 if (ftrace_dump_on_oops
)
4912 ftrace_dump(ftrace_dump_on_oops
);
4916 static struct notifier_block trace_panic_notifier
= {
4917 .notifier_call
= trace_panic_handler
,
4919 .priority
= 150 /* priority: INT_MAX >= x >= 0 */
4922 static int trace_die_handler(struct notifier_block
*self
,
4928 if (ftrace_dump_on_oops
)
4929 ftrace_dump(ftrace_dump_on_oops
);
4937 static struct notifier_block trace_die_notifier
= {
4938 .notifier_call
= trace_die_handler
,
4943 * printk is set to max of 1024, we really don't need it that big.
4944 * Nothing should be printing 1000 characters anyway.
4946 #define TRACE_MAX_PRINT 1000
4949 * Define here KERN_TRACE so that we have one place to modify
4950 * it if we decide to change what log level the ftrace dump
4953 #define KERN_TRACE KERN_EMERG
4956 trace_printk_seq(struct trace_seq
*s
)
4958 /* Probably should print a warning here. */
4962 /* should be zero ended, but we are paranoid. */
4963 s
->buffer
[s
->len
] = 0;
4965 printk(KERN_TRACE
"%s", s
->buffer
);
4970 void trace_init_global_iter(struct trace_iterator
*iter
)
4972 iter
->tr
= &global_trace
;
4973 iter
->trace
= current_trace
;
4974 iter
->cpu_file
= TRACE_PIPE_ALL_CPU
;
4978 __ftrace_dump(bool disable_tracing
, enum ftrace_dump_mode oops_dump_mode
)
4980 static arch_spinlock_t ftrace_dump_lock
=
4981 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
4982 /* use static because iter can be a bit big for the stack */
4983 static struct trace_iterator iter
;
4984 unsigned int old_userobj
;
4985 static int dump_ran
;
4986 unsigned long flags
;
4990 local_irq_save(flags
);
4991 arch_spin_lock(&ftrace_dump_lock
);
4999 /* Did function tracer already get disabled? */
5000 if (ftrace_is_dead()) {
5001 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
5002 printk("# MAY BE MISSING FUNCTION EVENTS\n");
5005 if (disable_tracing
)
5008 trace_init_global_iter(&iter
);
5010 for_each_tracing_cpu(cpu
) {
5011 atomic_inc(&iter
.tr
->data
[cpu
]->disabled
);
5014 old_userobj
= trace_flags
& TRACE_ITER_SYM_USEROBJ
;
5016 /* don't look at user memory in panic mode */
5017 trace_flags
&= ~TRACE_ITER_SYM_USEROBJ
;
5019 /* Simulate the iterator */
5020 iter
.tr
= &global_trace
;
5021 iter
.trace
= current_trace
;
5023 switch (oops_dump_mode
) {
5025 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
5028 iter
.cpu_file
= raw_smp_processor_id();
5033 printk(KERN_TRACE
"Bad dumping mode, switching to all CPUs dump\n");
5034 iter
.cpu_file
= TRACE_PIPE_ALL_CPU
;
5037 printk(KERN_TRACE
"Dumping ftrace buffer:\n");
5040 * We need to stop all tracing on all CPUS to read the
5041 * the next buffer. This is a bit expensive, but is
5042 * not done often. We fill all what we can read,
5043 * and then release the locks again.
5046 while (!trace_empty(&iter
)) {
5049 printk(KERN_TRACE
"---------------------------------\n");
5053 /* reset all but tr, trace, and overruns */
5054 memset(&iter
.seq
, 0,
5055 sizeof(struct trace_iterator
) -
5056 offsetof(struct trace_iterator
, seq
));
5057 iter
.iter_flags
|= TRACE_FILE_LAT_FMT
;
5060 if (trace_find_next_entry_inc(&iter
) != NULL
) {
5063 ret
= print_trace_line(&iter
);
5064 if (ret
!= TRACE_TYPE_NO_CONSUME
)
5065 trace_consume(&iter
);
5067 touch_nmi_watchdog();
5069 trace_printk_seq(&iter
.seq
);
5073 printk(KERN_TRACE
" (ftrace buffer empty)\n");
5075 printk(KERN_TRACE
"---------------------------------\n");
5078 /* Re-enable tracing if requested */
5079 if (!disable_tracing
) {
5080 trace_flags
|= old_userobj
;
5082 for_each_tracing_cpu(cpu
) {
5083 atomic_dec(&iter
.tr
->data
[cpu
]->disabled
);
5089 arch_spin_unlock(&ftrace_dump_lock
);
5090 local_irq_restore(flags
);
5093 /* By default: disable tracing after the dump */
5094 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode
)
5096 __ftrace_dump(true, oops_dump_mode
);
5098 EXPORT_SYMBOL_GPL(ftrace_dump
);
5100 __init
static int tracer_alloc_buffers(void)
5103 enum ring_buffer_flags rb_flags
;
5108 if (!alloc_cpumask_var(&tracing_buffer_mask
, GFP_KERNEL
))
5111 if (!alloc_cpumask_var(&tracing_cpumask
, GFP_KERNEL
))
5112 goto out_free_buffer_mask
;
5114 /* Only allocate trace_printk buffers if a trace_printk exists */
5115 if (__stop___trace_bprintk_fmt
!= __start___trace_bprintk_fmt
)
5116 /* Must be called before global_trace.buffer is allocated */
5117 trace_printk_init_buffers();
5119 /* To save memory, keep the ring buffer size to its minimum */
5120 if (ring_buffer_expanded
)
5121 ring_buf_size
= trace_buf_size
;
5125 rb_flags
= trace_flags
& TRACE_ITER_OVERWRITE
? RB_FL_OVERWRITE
: 0;
5127 cpumask_copy(tracing_buffer_mask
, cpu_possible_mask
);
5128 cpumask_copy(tracing_cpumask
, cpu_all_mask
);
5130 /* TODO: make the number of buffers hot pluggable with CPUS */
5131 global_trace
.buffer
= ring_buffer_alloc(ring_buf_size
, rb_flags
);
5132 if (!global_trace
.buffer
) {
5133 printk(KERN_ERR
"tracer: failed to allocate ring buffer!\n");
5135 goto out_free_cpumask
;
5137 if (global_trace
.buffer_disabled
)
5141 #ifdef CONFIG_TRACER_MAX_TRACE
5142 max_tr
.buffer
= ring_buffer_alloc(1, rb_flags
);
5143 if (!max_tr
.buffer
) {
5144 printk(KERN_ERR
"tracer: failed to allocate max ring buffer!\n");
5146 ring_buffer_free(global_trace
.buffer
);
5147 goto out_free_cpumask
;
5151 /* Allocate the first page for all buffers */
5152 for_each_tracing_cpu(i
) {
5153 global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
5154 max_tr
.data
[i
] = &per_cpu(max_tr_data
, i
);
5157 set_buffer_entries(&global_trace
,
5158 ring_buffer_size(global_trace
.buffer
, 0));
5159 #ifdef CONFIG_TRACER_MAX_TRACE
5160 set_buffer_entries(&max_tr
, 1);
5163 trace_init_cmdlines();
5164 init_irq_work(&trace_work_wakeup
, trace_wake_up
);
5166 register_tracer(&nop_trace
);
5167 current_trace
= &nop_trace
;
5168 /* All seems OK, enable tracing */
5169 tracing_disabled
= 0;
5171 atomic_notifier_chain_register(&panic_notifier_list
,
5172 &trace_panic_notifier
);
5174 register_die_notifier(&trace_die_notifier
);
5176 while (trace_boot_options
) {
5179 option
= strsep(&trace_boot_options
, ",");
5180 trace_set_options(option
);
5186 free_cpumask_var(tracing_cpumask
);
5187 out_free_buffer_mask
:
5188 free_cpumask_var(tracing_buffer_mask
);
5193 __init
static int clear_boot_tracer(void)
5196 * The default tracer at boot buffer is an init section.
5197 * This function is called in lateinit. If we did not
5198 * find the boot tracer, then clear it out, to prevent
5199 * later registration from accessing the buffer that is
5200 * about to be freed.
5202 if (!default_bootup_tracer
)
5205 printk(KERN_INFO
"ftrace bootup tracer '%s' not registered.\n",
5206 default_bootup_tracer
);
5207 default_bootup_tracer
= NULL
;
5212 early_initcall(tracer_alloc_buffers
);
5213 fs_initcall(tracer_init_debugfs
);
5214 late_initcall(clear_boot_tracer
);