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 William Lee Irwin III
14
#include <linux/ring_buffer.h>
15
#include <generated/utsrelease.h>
16
#include <linux/stacktrace.h>
17
#include <linux/writeback.h>
18
#include <linux/kallsyms.h>
19
#include <linux/seq_file.h>
20
#include <linux/notifier.h>
21
#include <linux/irqflags.h>
22
#include <linux/debugfs.h>
23
#include <linux/pagemap.h>
24
#include <linux/hardirq.h>
25
#include <linux/linkage.h>
26
#include <linux/uaccess.h>
27
#include <linux/kprobes.h>
28
#include <linux/ftrace.h>
29
#include <linux/module.h>
30
#include <linux/percpu.h>
31
#include <linux/splice.h>
32
#include <linux/kdebug.h>
33
#include <linux/string.h>
34
#include <linux/rwsem.h>
35
#include <linux/slab.h>
36
#include <linux/ctype.h>
37
#include <linux/init.h>
38
#include <linux/poll.h>
42
#include "trace_output.h"
45
* On boot up, the ring buffer is set to the minimum size, so that
46
* we do not waste memory on systems that are not using tracing.
48
int ring_buffer_expanded;
51
* We need to change this state when a selftest is running.
52
* A selftest will lurk into the ring-buffer to count the
53
* entries inserted during the selftest although some concurrent
54
* insertions into the ring-buffer such as trace_printk could occurred
55
* at the same time, giving false positive or negative results.
57
static bool __read_mostly tracing_selftest_running;
60
* If a tracer is running, we do not want to run SELFTEST.
62
bool __read_mostly tracing_selftest_disabled;
64
/* For tracers that don't implement custom flags */
65
static struct tracer_opt dummy_tracer_opt[] = {
69
static struct tracer_flags dummy_tracer_flags = {
71
.opts = dummy_tracer_opt
74
static int dummy_set_flag(u32 old_flags, u32 bit, int set)
80
* Kill all tracing for good (never come back).
81
* It is initialized to 1 but will turn to zero if the initialization
82
* of the tracer is successful. But that is the only place that sets
85
static int tracing_disabled = 1;
87
DEFINE_PER_CPU(int, ftrace_cpu_disabled);
89
static inline void ftrace_disable_cpu(void)
92
__this_cpu_inc(ftrace_cpu_disabled);
95
static inline void ftrace_enable_cpu(void)
97
__this_cpu_dec(ftrace_cpu_disabled);
101
cpumask_var_t __read_mostly tracing_buffer_mask;
104
* ftrace_dump_on_oops - variable to dump ftrace buffer on oops
106
* If there is an oops (or kernel panic) and the ftrace_dump_on_oops
107
* is set, then ftrace_dump is called. This will output the contents
108
* of the ftrace buffers to the console. This is very useful for
109
* capturing traces that lead to crashes and outputing it to a
112
* It is default off, but you can enable it with either specifying
113
* "ftrace_dump_on_oops" in the kernel command line, or setting
114
* /proc/sys/kernel/ftrace_dump_on_oops
115
* Set 1 if you want to dump buffers of all CPUs
116
* Set 2 if you want to dump the buffer of the CPU that triggered oops
119
enum ftrace_dump_mode ftrace_dump_on_oops;
121
static int tracing_set_tracer(const char *buf);
123
#define MAX_TRACER_SIZE 100
124
static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
125
static char *default_bootup_tracer;
127
static int __init set_cmdline_ftrace(char *str)
129
strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
130
default_bootup_tracer = bootup_tracer_buf;
131
/* We are using ftrace early, expand it */
132
ring_buffer_expanded = 1;
135
__setup("ftrace=", set_cmdline_ftrace);
137
static int __init set_ftrace_dump_on_oops(char *str)
139
if (*str++ != '=' || !*str) {
140
ftrace_dump_on_oops = DUMP_ALL;
144
if (!strcmp("orig_cpu", str)) {
145
ftrace_dump_on_oops = DUMP_ORIG;
151
__setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
153
unsigned long long ns2usecs(cycle_t nsec)
161
* The global_trace is the descriptor that holds the tracing
162
* buffers for the live tracing. For each CPU, it contains
163
* a link list of pages that will store trace entries. The
164
* page descriptor of the pages in the memory is used to hold
165
* the link list by linking the lru item in the page descriptor
166
* to each of the pages in the buffer per CPU.
168
* For each active CPU there is a data field that holds the
169
* pages for the buffer for that CPU. Each CPU has the same number
170
* of pages allocated for its buffer.
172
static struct trace_array global_trace;
174
static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
176
int filter_current_check_discard(struct ring_buffer *buffer,
177
struct ftrace_event_call *call, void *rec,
178
struct ring_buffer_event *event)
180
return filter_check_discard(call, rec, buffer, event);
182
EXPORT_SYMBOL_GPL(filter_current_check_discard);
184
cycle_t ftrace_now(int cpu)
188
/* Early boot up does not have a buffer yet */
189
if (!global_trace.buffer)
190
return trace_clock_local();
192
ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
193
ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
199
* The max_tr is used to snapshot the global_trace when a maximum
200
* latency is reached. Some tracers will use this to store a maximum
201
* trace while it continues examining live traces.
203
* The buffers for the max_tr are set up the same as the global_trace.
204
* When a snapshot is taken, the link list of the max_tr is swapped
205
* with the link list of the global_trace and the buffers are reset for
206
* the global_trace so the tracing can continue.
208
static struct trace_array max_tr;
210
static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
212
/* tracer_enabled is used to toggle activation of a tracer */
213
static int tracer_enabled = 1;
216
* tracing_is_enabled - return tracer_enabled status
218
* This function is used by other tracers to know the status
219
* of the tracer_enabled flag. Tracers may use this function
220
* to know if it should enable their features when starting
221
* up. See irqsoff tracer for an example (start_irqsoff_tracer).
223
int tracing_is_enabled(void)
225
return tracer_enabled;
229
* trace_buf_size is the size in bytes that is allocated
230
* for a buffer. Note, the number of bytes is always rounded
233
* This number is purposely set to a low number of 16384.
234
* If the dump on oops happens, it will be much appreciated
235
* to not have to wait for all that output. Anyway this can be
236
* boot time and run time configurable.
238
#define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
240
static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
242
/* trace_types holds a link list of available tracers. */
243
static struct tracer *trace_types __read_mostly;
245
/* current_trace points to the tracer that is currently active */
246
static struct tracer *current_trace __read_mostly;
249
* trace_types_lock is used to protect the trace_types list.
251
static DEFINE_MUTEX(trace_types_lock);
254
* serialize the access of the ring buffer
256
* ring buffer serializes readers, but it is low level protection.
257
* The validity of the events (which returns by ring_buffer_peek() ..etc)
258
* are not protected by ring buffer.
260
* The content of events may become garbage if we allow other process consumes
261
* these events concurrently:
262
* A) the page of the consumed events may become a normal page
263
* (not reader page) in ring buffer, and this page will be rewrited
264
* by events producer.
265
* B) The page of the consumed events may become a page for splice_read,
266
* and this page will be returned to system.
268
* These primitives allow multi process access to different cpu ring buffer
271
* These primitives don't distinguish read-only and read-consume access.
272
* Multi read-only access are also serialized.
276
static DECLARE_RWSEM(all_cpu_access_lock);
277
static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
279
static inline void trace_access_lock(int cpu)
281
if (cpu == TRACE_PIPE_ALL_CPU) {
282
/* gain it for accessing the whole ring buffer. */
283
down_write(&all_cpu_access_lock);
285
/* gain it for accessing a cpu ring buffer. */
287
/* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
288
down_read(&all_cpu_access_lock);
290
/* Secondly block other access to this @cpu ring buffer. */
291
mutex_lock(&per_cpu(cpu_access_lock, cpu));
295
static inline void trace_access_unlock(int cpu)
297
if (cpu == TRACE_PIPE_ALL_CPU) {
298
up_write(&all_cpu_access_lock);
300
mutex_unlock(&per_cpu(cpu_access_lock, cpu));
301
up_read(&all_cpu_access_lock);
305
static inline void trace_access_lock_init(void)
309
for_each_possible_cpu(cpu)
310
mutex_init(&per_cpu(cpu_access_lock, cpu));
315
static DEFINE_MUTEX(access_lock);
317
static inline void trace_access_lock(int cpu)
320
mutex_lock(&access_lock);
323
static inline void trace_access_unlock(int cpu)
326
mutex_unlock(&access_lock);
329
static inline void trace_access_lock_init(void)
335
/* trace_wait is a waitqueue for tasks blocked on trace_poll */
336
static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
338
/* trace_flags holds trace_options default values */
339
unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
340
TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
341
TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE;
343
static int trace_stop_count;
344
static DEFINE_RAW_SPINLOCK(tracing_start_lock);
346
static void wakeup_work_handler(struct work_struct *work)
348
wake_up(&trace_wait);
351
static DECLARE_DELAYED_WORK(wakeup_work, wakeup_work_handler);
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
void trace_wake_up(void)
362
const unsigned long delay = msecs_to_jiffies(2);
364
if (trace_flags & TRACE_ITER_BLOCK)
366
schedule_delayed_work(&wakeup_work, delay);
369
static int __init set_buf_size(char *str)
371
unsigned long buf_size;
375
buf_size = memparse(str, &str);
376
/* nr_entries can not be zero */
379
trace_buf_size = buf_size;
382
__setup("trace_buf_size=", set_buf_size);
384
static int __init set_tracing_thresh(char *str)
386
unsigned long threshhold;
391
ret = strict_strtoul(str, 0, &threshhold);
394
tracing_thresh = threshhold * 1000;
397
__setup("tracing_thresh=", set_tracing_thresh);
399
unsigned long nsecs_to_usecs(unsigned long nsecs)
404
/* These must match the bit postions in trace_iterator_flags */
405
static const char *trace_options[] = {
436
{ trace_clock_local, "local" },
437
{ trace_clock_global, "global" },
438
{ trace_clock_counter, "counter" },
444
* trace_parser_get_init - gets the buffer for trace parser
446
int trace_parser_get_init(struct trace_parser *parser, int size)
448
memset(parser, 0, sizeof(*parser));
450
parser->buffer = kmalloc(size, GFP_KERNEL);
459
* trace_parser_put - frees the buffer for trace parser
461
void trace_parser_put(struct trace_parser *parser)
463
kfree(parser->buffer);
467
* trace_get_user - reads the user input string separated by space
468
* (matched by isspace(ch))
470
* For each string found the 'struct trace_parser' is updated,
471
* and the function returns.
473
* Returns number of bytes read.
475
* See kernel/trace/trace.h for 'struct trace_parser' details.
477
int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
478
size_t cnt, loff_t *ppos)
485
trace_parser_clear(parser);
487
ret = get_user(ch, ubuf++);
495
* The parser is not finished with the last write,
496
* continue reading the user input without skipping spaces.
499
/* skip white space */
500
while (cnt && isspace(ch)) {
501
ret = get_user(ch, ubuf++);
508
/* only spaces were written */
518
/* read the non-space input */
519
while (cnt && !isspace(ch)) {
520
if (parser->idx < parser->size - 1)
521
parser->buffer[parser->idx++] = ch;
526
ret = get_user(ch, ubuf++);
533
/* We either got finished input or we have to wait for another call. */
535
parser->buffer[parser->idx] = 0;
536
parser->cont = false;
539
parser->buffer[parser->idx++] = ch;
549
ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
557
if (s->len <= s->readpos)
560
len = s->len - s->readpos;
563
ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
573
static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
578
if (s->len <= s->readpos)
581
len = s->len - s->readpos;
584
ret = memcpy(buf, s->buffer + s->readpos, cnt);
593
* ftrace_max_lock is used to protect the swapping of buffers
594
* when taking a max snapshot. The buffers themselves are
595
* protected by per_cpu spinlocks. But the action of the swap
596
* needs its own lock.
598
* This is defined as a arch_spinlock_t in order to help
599
* with performance when lockdep debugging is enabled.
601
* It is also used in other places outside the update_max_tr
602
* so it needs to be defined outside of the
603
* CONFIG_TRACER_MAX_TRACE.
605
static arch_spinlock_t ftrace_max_lock =
606
(arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
608
unsigned long __read_mostly tracing_thresh;
610
#ifdef CONFIG_TRACER_MAX_TRACE
611
unsigned long __read_mostly tracing_max_latency;
614
* Copy the new maximum trace into the separate maximum-trace
615
* structure. (this way the maximum trace is permanently saved,
616
* for later retrieval via /sys/kernel/debug/tracing/latency_trace)
619
__update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
621
struct trace_array_cpu *data = tr->data[cpu];
622
struct trace_array_cpu *max_data;
625
max_tr.time_start = data->preempt_timestamp;
627
max_data = max_tr.data[cpu];
628
max_data->saved_latency = tracing_max_latency;
629
max_data->critical_start = data->critical_start;
630
max_data->critical_end = data->critical_end;
632
memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
633
max_data->pid = tsk->pid;
634
max_data->uid = task_uid(tsk);
635
max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
636
max_data->policy = tsk->policy;
637
max_data->rt_priority = tsk->rt_priority;
639
/* record this tasks comm */
640
tracing_record_cmdline(tsk);
644
* update_max_tr - snapshot all trace buffers from global_trace to max_tr
646
* @tsk: the task with the latency
647
* @cpu: The cpu that initiated the trace.
649
* Flip the buffers between the @tr and the max_tr and record information
650
* about which task was the cause of this latency.
653
update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
655
struct ring_buffer *buf = tr->buffer;
657
if (trace_stop_count)
660
WARN_ON_ONCE(!irqs_disabled());
661
if (!current_trace->use_max_tr) {
665
arch_spin_lock(&ftrace_max_lock);
667
tr->buffer = max_tr.buffer;
670
__update_max_tr(tr, tsk, cpu);
671
arch_spin_unlock(&ftrace_max_lock);
675
* update_max_tr_single - only copy one trace over, and reset the rest
677
* @tsk - task with the latency
678
* @cpu - the cpu of the buffer to copy.
680
* Flip the trace of a single CPU buffer between the @tr and the max_tr.
683
update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
687
if (trace_stop_count)
690
WARN_ON_ONCE(!irqs_disabled());
691
if (!current_trace->use_max_tr) {
696
arch_spin_lock(&ftrace_max_lock);
698
ftrace_disable_cpu();
700
ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
704
* We failed to swap the buffer due to a commit taking
705
* place on this CPU. We fail to record, but we reset
706
* the max trace buffer (no one writes directly to it)
707
* and flag that it failed.
709
trace_array_printk(&max_tr, _THIS_IP_,
710
"Failed to swap buffers due to commit in progress\n");
715
WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
717
__update_max_tr(tr, tsk, cpu);
718
arch_spin_unlock(&ftrace_max_lock);
720
#endif /* CONFIG_TRACER_MAX_TRACE */
723
* register_tracer - register a tracer with the ftrace system.
724
* @type - the plugin for the tracer
726
* Register a new plugin tracer.
728
int register_tracer(struct tracer *type)
729
__releases(kernel_lock)
730
__acquires(kernel_lock)
736
pr_info("Tracer must have a name\n");
740
if (strlen(type->name) >= MAX_TRACER_SIZE) {
741
pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
745
mutex_lock(&trace_types_lock);
747
tracing_selftest_running = true;
749
for (t = trace_types; t; t = t->next) {
750
if (strcmp(type->name, t->name) == 0) {
752
pr_info("Tracer %s already registered\n",
760
type->set_flag = &dummy_set_flag;
762
type->flags = &dummy_tracer_flags;
764
if (!type->flags->opts)
765
type->flags->opts = dummy_tracer_opt;
766
if (!type->wait_pipe)
767
type->wait_pipe = default_wait_pipe;
770
#ifdef CONFIG_FTRACE_STARTUP_TEST
771
if (type->selftest && !tracing_selftest_disabled) {
772
struct tracer *saved_tracer = current_trace;
773
struct trace_array *tr = &global_trace;
776
* Run a selftest on this tracer.
777
* Here we reset the trace buffer, and set the current
778
* tracer to be this tracer. The tracer can then run some
779
* internal tracing to verify that everything is in order.
780
* If we fail, we do not register this tracer.
782
tracing_reset_online_cpus(tr);
784
current_trace = type;
786
/* If we expanded the buffers, make sure the max is expanded too */
787
if (ring_buffer_expanded && type->use_max_tr)
788
ring_buffer_resize(max_tr.buffer, trace_buf_size);
790
/* the test is responsible for initializing and enabling */
791
pr_info("Testing tracer %s: ", type->name);
792
ret = type->selftest(type, tr);
793
/* the test is responsible for resetting too */
794
current_trace = saved_tracer;
796
printk(KERN_CONT "FAILED!\n");
799
/* Only reset on passing, to avoid touching corrupted buffers */
800
tracing_reset_online_cpus(tr);
802
/* Shrink the max buffer again */
803
if (ring_buffer_expanded && type->use_max_tr)
804
ring_buffer_resize(max_tr.buffer, 1);
806
printk(KERN_CONT "PASSED\n");
810
type->next = trace_types;
814
tracing_selftest_running = false;
815
mutex_unlock(&trace_types_lock);
817
if (ret || !default_bootup_tracer)
820
if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
823
printk(KERN_INFO "Starting tracer '%s'\n", type->name);
824
/* Do we want this tracer to start on bootup? */
825
tracing_set_tracer(type->name);
826
default_bootup_tracer = NULL;
827
/* disable other selftests, since this will break it. */
828
tracing_selftest_disabled = 1;
829
#ifdef CONFIG_FTRACE_STARTUP_TEST
830
printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
838
void unregister_tracer(struct tracer *type)
842
mutex_lock(&trace_types_lock);
843
for (t = &trace_types; *t; t = &(*t)->next) {
847
pr_info("Tracer %s not registered\n", type->name);
853
if (type == current_trace && tracer_enabled) {
856
if (current_trace->stop)
857
current_trace->stop(&global_trace);
858
current_trace = &nop_trace;
861
mutex_unlock(&trace_types_lock);
864
static void __tracing_reset(struct ring_buffer *buffer, int cpu)
866
ftrace_disable_cpu();
867
ring_buffer_reset_cpu(buffer, cpu);
871
void tracing_reset(struct trace_array *tr, int cpu)
873
struct ring_buffer *buffer = tr->buffer;
875
ring_buffer_record_disable(buffer);
877
/* Make sure all commits have finished */
879
__tracing_reset(buffer, cpu);
881
ring_buffer_record_enable(buffer);
884
void tracing_reset_online_cpus(struct trace_array *tr)
886
struct ring_buffer *buffer = tr->buffer;
889
ring_buffer_record_disable(buffer);
891
/* Make sure all commits have finished */
894
tr->time_start = ftrace_now(tr->cpu);
896
for_each_online_cpu(cpu)
897
__tracing_reset(buffer, cpu);
899
ring_buffer_record_enable(buffer);
902
void tracing_reset_current(int cpu)
904
tracing_reset(&global_trace, cpu);
907
void tracing_reset_current_online_cpus(void)
909
tracing_reset_online_cpus(&global_trace);
912
#define SAVED_CMDLINES 128
913
#define NO_CMDLINE_MAP UINT_MAX
914
static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
915
static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
916
static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
917
static int cmdline_idx;
918
static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
920
/* temporary disable recording */
921
static atomic_t trace_record_cmdline_disabled __read_mostly;
923
static void trace_init_cmdlines(void)
925
memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
926
memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
930
int is_tracing_stopped(void)
932
return trace_stop_count;
936
* ftrace_off_permanent - disable all ftrace code permanently
938
* This should only be called when a serious anomally has
939
* been detected. This will turn off the function tracing,
940
* ring buffers, and other tracing utilites. It takes no
941
* locks and can be called from any context.
943
void ftrace_off_permanent(void)
945
tracing_disabled = 1;
947
tracing_off_permanent();
951
* tracing_start - quick start of the tracer
953
* If tracing is enabled but was stopped by tracing_stop,
954
* this will start the tracer back up.
956
void tracing_start(void)
958
struct ring_buffer *buffer;
961
if (tracing_disabled)
964
raw_spin_lock_irqsave(&tracing_start_lock, flags);
965
if (--trace_stop_count) {
966
if (trace_stop_count < 0) {
967
/* Someone screwed up their debugging */
969
trace_stop_count = 0;
974
/* Prevent the buffers from switching */
975
arch_spin_lock(&ftrace_max_lock);
977
buffer = global_trace.buffer;
979
ring_buffer_record_enable(buffer);
981
buffer = max_tr.buffer;
983
ring_buffer_record_enable(buffer);
985
arch_spin_unlock(&ftrace_max_lock);
989
raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
993
* tracing_stop - quick stop of the tracer
995
* Light weight way to stop tracing. Use in conjunction with
998
void tracing_stop(void)
1000
struct ring_buffer *buffer;
1001
unsigned long flags;
1004
raw_spin_lock_irqsave(&tracing_start_lock, flags);
1005
if (trace_stop_count++)
1008
/* Prevent the buffers from switching */
1009
arch_spin_lock(&ftrace_max_lock);
1011
buffer = global_trace.buffer;
1013
ring_buffer_record_disable(buffer);
1015
buffer = max_tr.buffer;
1017
ring_buffer_record_disable(buffer);
1019
arch_spin_unlock(&ftrace_max_lock);
1022
raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1025
void trace_stop_cmdline_recording(void);
1027
static void trace_save_cmdline(struct task_struct *tsk)
1031
if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1035
* It's not the end of the world if we don't get
1036
* the lock, but we also don't want to spin
1037
* nor do we want to disable interrupts,
1038
* so if we miss here, then better luck next time.
1040
if (!arch_spin_trylock(&trace_cmdline_lock))
1043
idx = map_pid_to_cmdline[tsk->pid];
1044
if (idx == NO_CMDLINE_MAP) {
1045
idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1048
* Check whether the cmdline buffer at idx has a pid
1049
* mapped. We are going to overwrite that entry so we
1050
* need to clear the map_pid_to_cmdline. Otherwise we
1051
* would read the new comm for the old pid.
1053
pid = map_cmdline_to_pid[idx];
1054
if (pid != NO_CMDLINE_MAP)
1055
map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1057
map_cmdline_to_pid[idx] = tsk->pid;
1058
map_pid_to_cmdline[tsk->pid] = idx;
1063
memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1065
arch_spin_unlock(&trace_cmdline_lock);
1068
void trace_find_cmdline(int pid, char comm[])
1073
strcpy(comm, "<idle>");
1077
if (WARN_ON_ONCE(pid < 0)) {
1078
strcpy(comm, "<XXX>");
1082
if (pid > PID_MAX_DEFAULT) {
1083
strcpy(comm, "<...>");
1088
arch_spin_lock(&trace_cmdline_lock);
1089
map = map_pid_to_cmdline[pid];
1090
if (map != NO_CMDLINE_MAP)
1091
strcpy(comm, saved_cmdlines[map]);
1093
strcpy(comm, "<...>");
1095
arch_spin_unlock(&trace_cmdline_lock);
1099
void tracing_record_cmdline(struct task_struct *tsk)
1101
if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1105
trace_save_cmdline(tsk);
1109
tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1112
struct task_struct *tsk = current;
1114
entry->preempt_count = pc & 0xff;
1115
entry->pid = (tsk) ? tsk->pid : 0;
1118
#ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1119
(irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1121
TRACE_FLAG_IRQS_NOSUPPORT |
1123
((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1124
((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1125
(need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1127
EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1129
struct ring_buffer_event *
1130
trace_buffer_lock_reserve(struct ring_buffer *buffer,
1133
unsigned long flags, int pc)
1135
struct ring_buffer_event *event;
1137
event = ring_buffer_lock_reserve(buffer, len);
1138
if (event != NULL) {
1139
struct trace_entry *ent = ring_buffer_event_data(event);
1141
tracing_generic_entry_update(ent, flags, pc);
1149
__trace_buffer_unlock_commit(struct ring_buffer *buffer,
1150
struct ring_buffer_event *event,
1151
unsigned long flags, int pc,
1154
ring_buffer_unlock_commit(buffer, event);
1156
ftrace_trace_stack(buffer, flags, 6, pc);
1157
ftrace_trace_userstack(buffer, flags, pc);
1163
void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1164
struct ring_buffer_event *event,
1165
unsigned long flags, int pc)
1167
__trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1170
struct ring_buffer_event *
1171
trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1172
int type, unsigned long len,
1173
unsigned long flags, int pc)
1175
*current_rb = global_trace.buffer;
1176
return trace_buffer_lock_reserve(*current_rb,
1177
type, len, flags, pc);
1179
EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1181
void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1182
struct ring_buffer_event *event,
1183
unsigned long flags, int pc)
1185
__trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1187
EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1189
void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1190
struct ring_buffer_event *event,
1191
unsigned long flags, int pc)
1193
__trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1195
EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1197
void trace_nowake_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1198
struct ring_buffer_event *event,
1199
unsigned long flags, int pc,
1200
struct pt_regs *regs)
1202
ring_buffer_unlock_commit(buffer, event);
1204
ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1205
ftrace_trace_userstack(buffer, flags, pc);
1207
EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit_regs);
1209
void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1210
struct ring_buffer_event *event)
1212
ring_buffer_discard_commit(buffer, event);
1214
EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1217
trace_function(struct trace_array *tr,
1218
unsigned long ip, unsigned long parent_ip, unsigned long flags,
1221
struct ftrace_event_call *call = &event_function;
1222
struct ring_buffer *buffer = tr->buffer;
1223
struct ring_buffer_event *event;
1224
struct ftrace_entry *entry;
1226
/* If we are reading the ring buffer, don't trace */
1227
if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1230
event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1234
entry = ring_buffer_event_data(event);
1236
entry->parent_ip = parent_ip;
1238
if (!filter_check_discard(call, entry, buffer, event))
1239
ring_buffer_unlock_commit(buffer, event);
1243
ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1244
unsigned long ip, unsigned long parent_ip, unsigned long flags,
1247
if (likely(!atomic_read(&data->disabled)))
1248
trace_function(tr, ip, parent_ip, flags, pc);
1251
#ifdef CONFIG_STACKTRACE
1253
#define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1254
struct ftrace_stack {
1255
unsigned long calls[FTRACE_STACK_MAX_ENTRIES];
1258
static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1259
static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1261
static void __ftrace_trace_stack(struct ring_buffer *buffer,
1262
unsigned long flags,
1263
int skip, int pc, struct pt_regs *regs)
1265
struct ftrace_event_call *call = &event_kernel_stack;
1266
struct ring_buffer_event *event;
1267
struct stack_entry *entry;
1268
struct stack_trace trace;
1270
int size = FTRACE_STACK_ENTRIES;
1272
trace.nr_entries = 0;
1276
* Since events can happen in NMIs there's no safe way to
1277
* use the per cpu ftrace_stacks. We reserve it and if an interrupt
1278
* or NMI comes in, it will just have to use the default
1279
* FTRACE_STACK_SIZE.
1281
preempt_disable_notrace();
1283
use_stack = ++__get_cpu_var(ftrace_stack_reserve);
1285
* We don't need any atomic variables, just a barrier.
1286
* If an interrupt comes in, we don't care, because it would
1287
* have exited and put the counter back to what we want.
1288
* We just need a barrier to keep gcc from moving things
1292
if (use_stack == 1) {
1293
trace.entries = &__get_cpu_var(ftrace_stack).calls[0];
1294
trace.max_entries = FTRACE_STACK_MAX_ENTRIES;
1297
save_stack_trace_regs(regs, &trace);
1299
save_stack_trace(&trace);
1301
if (trace.nr_entries > size)
1302
size = trace.nr_entries;
1304
/* From now on, use_stack is a boolean */
1307
size *= sizeof(unsigned long);
1309
event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1310
sizeof(*entry) + size, flags, pc);
1313
entry = ring_buffer_event_data(event);
1315
memset(&entry->caller, 0, size);
1318
memcpy(&entry->caller, trace.entries,
1319
trace.nr_entries * sizeof(unsigned long));
1321
trace.max_entries = FTRACE_STACK_ENTRIES;
1322
trace.entries = entry->caller;
1324
save_stack_trace_regs(regs, &trace);
1326
save_stack_trace(&trace);
1329
entry->size = trace.nr_entries;
1331
if (!filter_check_discard(call, entry, buffer, event))
1332
ring_buffer_unlock_commit(buffer, event);
1335
/* Again, don't let gcc optimize things here */
1337
__get_cpu_var(ftrace_stack_reserve)--;
1338
preempt_enable_notrace();
1342
void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1343
int skip, int pc, struct pt_regs *regs)
1345
if (!(trace_flags & TRACE_ITER_STACKTRACE))
1348
__ftrace_trace_stack(buffer, flags, skip, pc, regs);
1351
void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1354
if (!(trace_flags & TRACE_ITER_STACKTRACE))
1357
__ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1360
void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1363
__ftrace_trace_stack(tr->buffer, flags, skip, pc, NULL);
1367
* trace_dump_stack - record a stack back trace in the trace buffer
1369
void trace_dump_stack(void)
1371
unsigned long flags;
1373
if (tracing_disabled || tracing_selftest_running)
1376
local_save_flags(flags);
1378
/* skipping 3 traces, seems to get us at the caller of this function */
1379
__ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count(), NULL);
1382
static DEFINE_PER_CPU(int, user_stack_count);
1385
ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1387
struct ftrace_event_call *call = &event_user_stack;
1388
struct ring_buffer_event *event;
1389
struct userstack_entry *entry;
1390
struct stack_trace trace;
1392
if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1396
* NMIs can not handle page faults, even with fix ups.
1397
* The save user stack can (and often does) fault.
1399
if (unlikely(in_nmi()))
1403
* prevent recursion, since the user stack tracing may
1404
* trigger other kernel events.
1407
if (__this_cpu_read(user_stack_count))
1410
__this_cpu_inc(user_stack_count);
1412
event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1413
sizeof(*entry), flags, pc);
1415
goto out_drop_count;
1416
entry = ring_buffer_event_data(event);
1418
entry->tgid = current->tgid;
1419
memset(&entry->caller, 0, sizeof(entry->caller));
1421
trace.nr_entries = 0;
1422
trace.max_entries = FTRACE_STACK_ENTRIES;
1424
trace.entries = entry->caller;
1426
save_stack_trace_user(&trace);
1427
if (!filter_check_discard(call, entry, buffer, event))
1428
ring_buffer_unlock_commit(buffer, event);
1431
__this_cpu_dec(user_stack_count);
1437
static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1439
ftrace_trace_userstack(tr, flags, preempt_count());
1443
#endif /* CONFIG_STACKTRACE */
1446
* trace_vbprintk - write binary msg to tracing buffer
1449
int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1451
static arch_spinlock_t trace_buf_lock =
1452
(arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1453
static u32 trace_buf[TRACE_BUF_SIZE];
1455
struct ftrace_event_call *call = &event_bprint;
1456
struct ring_buffer_event *event;
1457
struct ring_buffer *buffer;
1458
struct trace_array *tr = &global_trace;
1459
struct trace_array_cpu *data;
1460
struct bprint_entry *entry;
1461
unsigned long flags;
1463
int cpu, len = 0, size, pc;
1465
if (unlikely(tracing_selftest_running || tracing_disabled))
1468
/* Don't pollute graph traces with trace_vprintk internals */
1469
pause_graph_tracing();
1471
pc = preempt_count();
1472
preempt_disable_notrace();
1473
cpu = raw_smp_processor_id();
1474
data = tr->data[cpu];
1476
disable = atomic_inc_return(&data->disabled);
1477
if (unlikely(disable != 1))
1480
/* Lockdep uses trace_printk for lock tracing */
1481
local_irq_save(flags);
1482
arch_spin_lock(&trace_buf_lock);
1483
len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1485
if (len > TRACE_BUF_SIZE || len < 0)
1488
size = sizeof(*entry) + sizeof(u32) * len;
1489
buffer = tr->buffer;
1490
event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1494
entry = ring_buffer_event_data(event);
1498
memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1499
if (!filter_check_discard(call, entry, buffer, event)) {
1500
ring_buffer_unlock_commit(buffer, event);
1501
ftrace_trace_stack(buffer, flags, 6, pc);
1505
arch_spin_unlock(&trace_buf_lock);
1506
local_irq_restore(flags);
1509
atomic_dec_return(&data->disabled);
1510
preempt_enable_notrace();
1511
unpause_graph_tracing();
1515
EXPORT_SYMBOL_GPL(trace_vbprintk);
1517
int trace_array_printk(struct trace_array *tr,
1518
unsigned long ip, const char *fmt, ...)
1523
if (!(trace_flags & TRACE_ITER_PRINTK))
1527
ret = trace_array_vprintk(tr, ip, fmt, ap);
1532
int trace_array_vprintk(struct trace_array *tr,
1533
unsigned long ip, const char *fmt, va_list args)
1535
static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1536
static char trace_buf[TRACE_BUF_SIZE];
1538
struct ftrace_event_call *call = &event_print;
1539
struct ring_buffer_event *event;
1540
struct ring_buffer *buffer;
1541
struct trace_array_cpu *data;
1542
int cpu, len = 0, size, pc;
1543
struct print_entry *entry;
1544
unsigned long irq_flags;
1547
if (tracing_disabled || tracing_selftest_running)
1550
pc = preempt_count();
1551
preempt_disable_notrace();
1552
cpu = raw_smp_processor_id();
1553
data = tr->data[cpu];
1555
disable = atomic_inc_return(&data->disabled);
1556
if (unlikely(disable != 1))
1559
pause_graph_tracing();
1560
raw_local_irq_save(irq_flags);
1561
arch_spin_lock(&trace_buf_lock);
1562
len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1564
size = sizeof(*entry) + len + 1;
1565
buffer = tr->buffer;
1566
event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1570
entry = ring_buffer_event_data(event);
1573
memcpy(&entry->buf, trace_buf, len);
1574
entry->buf[len] = '\0';
1575
if (!filter_check_discard(call, entry, buffer, event)) {
1576
ring_buffer_unlock_commit(buffer, event);
1577
ftrace_trace_stack(buffer, irq_flags, 6, pc);
1581
arch_spin_unlock(&trace_buf_lock);
1582
raw_local_irq_restore(irq_flags);
1583
unpause_graph_tracing();
1585
atomic_dec_return(&data->disabled);
1586
preempt_enable_notrace();
1591
int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1593
return trace_array_vprintk(&global_trace, ip, fmt, args);
1595
EXPORT_SYMBOL_GPL(trace_vprintk);
1597
static void trace_iterator_increment(struct trace_iterator *iter)
1599
/* Don't allow ftrace to trace into the ring buffers */
1600
ftrace_disable_cpu();
1603
if (iter->buffer_iter[iter->cpu])
1604
ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1606
ftrace_enable_cpu();
1609
static struct trace_entry *
1610
peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1611
unsigned long *lost_events)
1613
struct ring_buffer_event *event;
1614
struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1616
/* Don't allow ftrace to trace into the ring buffers */
1617
ftrace_disable_cpu();
1620
event = ring_buffer_iter_peek(buf_iter, ts);
1622
event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1625
ftrace_enable_cpu();
1628
iter->ent_size = ring_buffer_event_length(event);
1629
return ring_buffer_event_data(event);
1635
static struct trace_entry *
1636
__find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1637
unsigned long *missing_events, u64 *ent_ts)
1639
struct ring_buffer *buffer = iter->tr->buffer;
1640
struct trace_entry *ent, *next = NULL;
1641
unsigned long lost_events = 0, next_lost = 0;
1642
int cpu_file = iter->cpu_file;
1643
u64 next_ts = 0, ts;
1648
* If we are in a per_cpu trace file, don't bother by iterating over
1649
* all cpu and peek directly.
1651
if (cpu_file > TRACE_PIPE_ALL_CPU) {
1652
if (ring_buffer_empty_cpu(buffer, cpu_file))
1654
ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1656
*ent_cpu = cpu_file;
1661
for_each_tracing_cpu(cpu) {
1663
if (ring_buffer_empty_cpu(buffer, cpu))
1666
ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1669
* Pick the entry with the smallest timestamp:
1671
if (ent && (!next || ts < next_ts)) {
1675
next_lost = lost_events;
1680
*ent_cpu = next_cpu;
1686
*missing_events = next_lost;
1691
/* Find the next real entry, without updating the iterator itself */
1692
struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1693
int *ent_cpu, u64 *ent_ts)
1695
return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1698
/* Find the next real entry, and increment the iterator to the next entry */
1699
void *trace_find_next_entry_inc(struct trace_iterator *iter)
1701
iter->ent = __find_next_entry(iter, &iter->cpu,
1702
&iter->lost_events, &iter->ts);
1705
trace_iterator_increment(iter);
1707
return iter->ent ? iter : NULL;
1710
static void trace_consume(struct trace_iterator *iter)
1712
/* Don't allow ftrace to trace into the ring buffers */
1713
ftrace_disable_cpu();
1714
ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1715
&iter->lost_events);
1716
ftrace_enable_cpu();
1719
static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1721
struct trace_iterator *iter = m->private;
1725
WARN_ON_ONCE(iter->leftover);
1729
/* can't go backwards */
1734
ent = trace_find_next_entry_inc(iter);
1738
while (ent && iter->idx < i)
1739
ent = trace_find_next_entry_inc(iter);
1746
void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1748
struct trace_array *tr = iter->tr;
1749
struct ring_buffer_event *event;
1750
struct ring_buffer_iter *buf_iter;
1751
unsigned long entries = 0;
1754
tr->data[cpu]->skipped_entries = 0;
1756
if (!iter->buffer_iter[cpu])
1759
buf_iter = iter->buffer_iter[cpu];
1760
ring_buffer_iter_reset(buf_iter);
1763
* We could have the case with the max latency tracers
1764
* that a reset never took place on a cpu. This is evident
1765
* by the timestamp being before the start of the buffer.
1767
while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1768
if (ts >= iter->tr->time_start)
1771
ring_buffer_read(buf_iter, NULL);
1774
tr->data[cpu]->skipped_entries = entries;
1778
* The current tracer is copied to avoid a global locking
1781
static void *s_start(struct seq_file *m, loff_t *pos)
1783
struct trace_iterator *iter = m->private;
1784
static struct tracer *old_tracer;
1785
int cpu_file = iter->cpu_file;
1790
/* copy the tracer to avoid using a global lock all around */
1791
mutex_lock(&trace_types_lock);
1792
if (unlikely(old_tracer != current_trace && current_trace)) {
1793
old_tracer = current_trace;
1794
*iter->trace = *current_trace;
1796
mutex_unlock(&trace_types_lock);
1798
atomic_inc(&trace_record_cmdline_disabled);
1800
if (*pos != iter->pos) {
1805
ftrace_disable_cpu();
1807
if (cpu_file == TRACE_PIPE_ALL_CPU) {
1808
for_each_tracing_cpu(cpu)
1809
tracing_iter_reset(iter, cpu);
1811
tracing_iter_reset(iter, cpu_file);
1813
ftrace_enable_cpu();
1816
for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1821
* If we overflowed the seq_file before, then we want
1822
* to just reuse the trace_seq buffer again.
1828
p = s_next(m, p, &l);
1832
trace_event_read_lock();
1833
trace_access_lock(cpu_file);
1837
static void s_stop(struct seq_file *m, void *p)
1839
struct trace_iterator *iter = m->private;
1841
atomic_dec(&trace_record_cmdline_disabled);
1842
trace_access_unlock(iter->cpu_file);
1843
trace_event_read_unlock();
1846
static void print_lat_help_header(struct seq_file *m)
1848
seq_puts(m, "# _------=> CPU# \n");
1849
seq_puts(m, "# / _-----=> irqs-off \n");
1850
seq_puts(m, "# | / _----=> need-resched \n");
1851
seq_puts(m, "# || / _---=> hardirq/softirq \n");
1852
seq_puts(m, "# ||| / _--=> preempt-depth \n");
1853
seq_puts(m, "# |||| / delay \n");
1854
seq_puts(m, "# cmd pid ||||| time | caller \n");
1855
seq_puts(m, "# \\ / ||||| \\ | / \n");
1858
static void print_func_help_header(struct seq_file *m)
1860
seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1861
seq_puts(m, "# | | | | |\n");
1866
print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1868
unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1869
struct trace_array *tr = iter->tr;
1870
struct trace_array_cpu *data = tr->data[tr->cpu];
1871
struct tracer *type = current_trace;
1872
unsigned long entries = 0;
1873
unsigned long total = 0;
1874
unsigned long count;
1875
const char *name = "preemption";
1882
for_each_tracing_cpu(cpu) {
1883
count = ring_buffer_entries_cpu(tr->buffer, cpu);
1885
* If this buffer has skipped entries, then we hold all
1886
* entries for the trace and we need to ignore the
1887
* ones before the time stamp.
1889
if (tr->data[cpu]->skipped_entries) {
1890
count -= tr->data[cpu]->skipped_entries;
1891
/* total is the same as the entries */
1895
ring_buffer_overrun_cpu(tr->buffer, cpu);
1899
seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1901
seq_puts(m, "# -----------------------------------"
1902
"---------------------------------\n");
1903
seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1904
" (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1905
nsecs_to_usecs(data->saved_latency),
1909
#if defined(CONFIG_PREEMPT_NONE)
1911
#elif defined(CONFIG_PREEMPT_VOLUNTARY)
1913
#elif defined(CONFIG_PREEMPT)
1918
/* These are reserved for later use */
1921
seq_printf(m, " #P:%d)\n", num_online_cpus());
1925
seq_puts(m, "# -----------------\n");
1926
seq_printf(m, "# | task: %.16s-%d "
1927
"(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1928
data->comm, data->pid, data->uid, data->nice,
1929
data->policy, data->rt_priority);
1930
seq_puts(m, "# -----------------\n");
1932
if (data->critical_start) {
1933
seq_puts(m, "# => started at: ");
1934
seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1935
trace_print_seq(m, &iter->seq);
1936
seq_puts(m, "\n# => ended at: ");
1937
seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1938
trace_print_seq(m, &iter->seq);
1939
seq_puts(m, "\n#\n");
1945
static void test_cpu_buff_start(struct trace_iterator *iter)
1947
struct trace_seq *s = &iter->seq;
1949
if (!(trace_flags & TRACE_ITER_ANNOTATE))
1952
if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1955
if (cpumask_test_cpu(iter->cpu, iter->started))
1958
if (iter->tr->data[iter->cpu]->skipped_entries)
1961
cpumask_set_cpu(iter->cpu, iter->started);
1963
/* Don't print started cpu buffer for the first entry of the trace */
1965
trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1969
static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1971
struct trace_seq *s = &iter->seq;
1972
unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1973
struct trace_entry *entry;
1974
struct trace_event *event;
1978
test_cpu_buff_start(iter);
1980
event = ftrace_find_event(entry->type);
1982
if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1983
if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1984
if (!trace_print_lat_context(iter))
1987
if (!trace_print_context(iter))
1993
return event->funcs->trace(iter, sym_flags, event);
1995
if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1998
return TRACE_TYPE_HANDLED;
2000
return TRACE_TYPE_PARTIAL_LINE;
2003
static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2005
struct trace_seq *s = &iter->seq;
2006
struct trace_entry *entry;
2007
struct trace_event *event;
2011
if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2012
if (!trace_seq_printf(s, "%d %d %llu ",
2013
entry->pid, iter->cpu, iter->ts))
2017
event = ftrace_find_event(entry->type);
2019
return event->funcs->raw(iter, 0, event);
2021
if (!trace_seq_printf(s, "%d ?\n", entry->type))
2024
return TRACE_TYPE_HANDLED;
2026
return TRACE_TYPE_PARTIAL_LINE;
2029
static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2031
struct trace_seq *s = &iter->seq;
2032
unsigned char newline = '\n';
2033
struct trace_entry *entry;
2034
struct trace_event *event;
2038
if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2039
SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2040
SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2041
SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2044
event = ftrace_find_event(entry->type);
2046
enum print_line_t ret = event->funcs->hex(iter, 0, event);
2047
if (ret != TRACE_TYPE_HANDLED)
2051
SEQ_PUT_FIELD_RET(s, newline);
2053
return TRACE_TYPE_HANDLED;
2056
static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2058
struct trace_seq *s = &iter->seq;
2059
struct trace_entry *entry;
2060
struct trace_event *event;
2064
if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2065
SEQ_PUT_FIELD_RET(s, entry->pid);
2066
SEQ_PUT_FIELD_RET(s, iter->cpu);
2067
SEQ_PUT_FIELD_RET(s, iter->ts);
2070
event = ftrace_find_event(entry->type);
2071
return event ? event->funcs->binary(iter, 0, event) :
2075
int trace_empty(struct trace_iterator *iter)
2079
/* If we are looking at one CPU buffer, only check that one */
2080
if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2081
cpu = iter->cpu_file;
2082
if (iter->buffer_iter[cpu]) {
2083
if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2086
if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2092
for_each_tracing_cpu(cpu) {
2093
if (iter->buffer_iter[cpu]) {
2094
if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2097
if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2105
/* Called with trace_event_read_lock() held. */
2106
enum print_line_t print_trace_line(struct trace_iterator *iter)
2108
enum print_line_t ret;
2110
if (iter->lost_events &&
2111
!trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2112
iter->cpu, iter->lost_events))
2113
return TRACE_TYPE_PARTIAL_LINE;
2115
if (iter->trace && iter->trace->print_line) {
2116
ret = iter->trace->print_line(iter);
2117
if (ret != TRACE_TYPE_UNHANDLED)
2121
if (iter->ent->type == TRACE_BPRINT &&
2122
trace_flags & TRACE_ITER_PRINTK &&
2123
trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2124
return trace_print_bprintk_msg_only(iter);
2126
if (iter->ent->type == TRACE_PRINT &&
2127
trace_flags & TRACE_ITER_PRINTK &&
2128
trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2129
return trace_print_printk_msg_only(iter);
2131
if (trace_flags & TRACE_ITER_BIN)
2132
return print_bin_fmt(iter);
2134
if (trace_flags & TRACE_ITER_HEX)
2135
return print_hex_fmt(iter);
2137
if (trace_flags & TRACE_ITER_RAW)
2138
return print_raw_fmt(iter);
2140
return print_trace_fmt(iter);
2143
void trace_default_header(struct seq_file *m)
2145
struct trace_iterator *iter = m->private;
2147
if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2150
if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2151
/* print nothing if the buffers are empty */
2152
if (trace_empty(iter))
2154
print_trace_header(m, iter);
2155
if (!(trace_flags & TRACE_ITER_VERBOSE))
2156
print_lat_help_header(m);
2158
if (!(trace_flags & TRACE_ITER_VERBOSE))
2159
print_func_help_header(m);
2163
static void test_ftrace_alive(struct seq_file *m)
2165
if (!ftrace_is_dead())
2167
seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2168
seq_printf(m, "# MAY BE MISSING FUNCTION EVENTS\n");
2171
static int s_show(struct seq_file *m, void *v)
2173
struct trace_iterator *iter = v;
2176
if (iter->ent == NULL) {
2178
seq_printf(m, "# tracer: %s\n", iter->trace->name);
2180
test_ftrace_alive(m);
2182
if (iter->trace && iter->trace->print_header)
2183
iter->trace->print_header(m);
2185
trace_default_header(m);
2187
} else if (iter->leftover) {
2189
* If we filled the seq_file buffer earlier, we
2190
* want to just show it now.
2192
ret = trace_print_seq(m, &iter->seq);
2194
/* ret should this time be zero, but you never know */
2195
iter->leftover = ret;
2198
print_trace_line(iter);
2199
ret = trace_print_seq(m, &iter->seq);
2201
* If we overflow the seq_file buffer, then it will
2202
* ask us for this data again at start up.
2204
* ret is 0 if seq_file write succeeded.
2207
iter->leftover = ret;
2213
static const struct seq_operations tracer_seq_ops = {
2220
static struct trace_iterator *
2221
__tracing_open(struct inode *inode, struct file *file)
2223
long cpu_file = (long) inode->i_private;
2224
void *fail_ret = ERR_PTR(-ENOMEM);
2225
struct trace_iterator *iter;
2229
if (tracing_disabled)
2230
return ERR_PTR(-ENODEV);
2232
iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2234
return ERR_PTR(-ENOMEM);
2237
* We make a copy of the current tracer to avoid concurrent
2238
* changes on it while we are reading.
2240
mutex_lock(&trace_types_lock);
2241
iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2246
*iter->trace = *current_trace;
2248
if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2251
if (current_trace && current_trace->print_max)
2254
iter->tr = &global_trace;
2256
mutex_init(&iter->mutex);
2257
iter->cpu_file = cpu_file;
2259
/* Notify the tracer early; before we stop tracing. */
2260
if (iter->trace && iter->trace->open)
2261
iter->trace->open(iter);
2263
/* Annotate start of buffers if we had overruns */
2264
if (ring_buffer_overruns(iter->tr->buffer))
2265
iter->iter_flags |= TRACE_FILE_ANNOTATE;
2267
/* stop the trace while dumping */
2270
if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2271
for_each_tracing_cpu(cpu) {
2272
iter->buffer_iter[cpu] =
2273
ring_buffer_read_prepare(iter->tr->buffer, cpu);
2275
ring_buffer_read_prepare_sync();
2276
for_each_tracing_cpu(cpu) {
2277
ring_buffer_read_start(iter->buffer_iter[cpu]);
2278
tracing_iter_reset(iter, cpu);
2281
cpu = iter->cpu_file;
2282
iter->buffer_iter[cpu] =
2283
ring_buffer_read_prepare(iter->tr->buffer, cpu);
2284
ring_buffer_read_prepare_sync();
2285
ring_buffer_read_start(iter->buffer_iter[cpu]);
2286
tracing_iter_reset(iter, cpu);
2289
ret = seq_open(file, &tracer_seq_ops);
2291
fail_ret = ERR_PTR(ret);
2295
m = file->private_data;
2298
mutex_unlock(&trace_types_lock);
2303
for_each_tracing_cpu(cpu) {
2304
if (iter->buffer_iter[cpu])
2305
ring_buffer_read_finish(iter->buffer_iter[cpu]);
2307
free_cpumask_var(iter->started);
2310
mutex_unlock(&trace_types_lock);
2317
int tracing_open_generic(struct inode *inode, struct file *filp)
2319
if (tracing_disabled)
2322
filp->private_data = inode->i_private;
2326
static int tracing_release(struct inode *inode, struct file *file)
2328
struct seq_file *m = file->private_data;
2329
struct trace_iterator *iter;
2332
if (!(file->f_mode & FMODE_READ))
2337
mutex_lock(&trace_types_lock);
2338
for_each_tracing_cpu(cpu) {
2339
if (iter->buffer_iter[cpu])
2340
ring_buffer_read_finish(iter->buffer_iter[cpu]);
2343
if (iter->trace && iter->trace->close)
2344
iter->trace->close(iter);
2346
/* reenable tracing if it was previously enabled */
2348
mutex_unlock(&trace_types_lock);
2350
seq_release(inode, file);
2351
mutex_destroy(&iter->mutex);
2352
free_cpumask_var(iter->started);
2358
static int tracing_open(struct inode *inode, struct file *file)
2360
struct trace_iterator *iter;
2363
/* If this file was open for write, then erase contents */
2364
if ((file->f_mode & FMODE_WRITE) &&
2365
(file->f_flags & O_TRUNC)) {
2366
long cpu = (long) inode->i_private;
2368
if (cpu == TRACE_PIPE_ALL_CPU)
2369
tracing_reset_online_cpus(&global_trace);
2371
tracing_reset(&global_trace, cpu);
2374
if (file->f_mode & FMODE_READ) {
2375
iter = __tracing_open(inode, file);
2377
ret = PTR_ERR(iter);
2378
else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2379
iter->iter_flags |= TRACE_FILE_LAT_FMT;
2385
t_next(struct seq_file *m, void *v, loff_t *pos)
2387
struct tracer *t = v;
2397
static void *t_start(struct seq_file *m, loff_t *pos)
2402
mutex_lock(&trace_types_lock);
2403
for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2409
static void t_stop(struct seq_file *m, void *p)
2411
mutex_unlock(&trace_types_lock);
2414
static int t_show(struct seq_file *m, void *v)
2416
struct tracer *t = v;
2421
seq_printf(m, "%s", t->name);
2430
static const struct seq_operations show_traces_seq_ops = {
2437
static int show_traces_open(struct inode *inode, struct file *file)
2439
if (tracing_disabled)
2442
return seq_open(file, &show_traces_seq_ops);
2446
tracing_write_stub(struct file *filp, const char __user *ubuf,
2447
size_t count, loff_t *ppos)
2452
static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2454
if (file->f_mode & FMODE_READ)
2455
return seq_lseek(file, offset, origin);
2460
static const struct file_operations tracing_fops = {
2461
.open = tracing_open,
2463
.write = tracing_write_stub,
2464
.llseek = tracing_seek,
2465
.release = tracing_release,
2468
static const struct file_operations show_traces_fops = {
2469
.open = show_traces_open,
2471
.release = seq_release,
2472
.llseek = seq_lseek,
2476
* Only trace on a CPU if the bitmask is set:
2478
static cpumask_var_t tracing_cpumask;
2481
* The tracer itself will not take this lock, but still we want
2482
* to provide a consistent cpumask to user-space:
2484
static DEFINE_MUTEX(tracing_cpumask_update_lock);
2487
* Temporary storage for the character representation of the
2488
* CPU bitmask (and one more byte for the newline):
2490
static char mask_str[NR_CPUS + 1];
2493
tracing_cpumask_read(struct file *filp, char __user *ubuf,
2494
size_t count, loff_t *ppos)
2498
mutex_lock(&tracing_cpumask_update_lock);
2500
len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2501
if (count - len < 2) {
2505
len += sprintf(mask_str + len, "\n");
2506
count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2509
mutex_unlock(&tracing_cpumask_update_lock);
2515
tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2516
size_t count, loff_t *ppos)
2519
cpumask_var_t tracing_cpumask_new;
2521
if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2524
err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2528
mutex_lock(&tracing_cpumask_update_lock);
2530
local_irq_disable();
2531
arch_spin_lock(&ftrace_max_lock);
2532
for_each_tracing_cpu(cpu) {
2534
* Increase/decrease the disabled counter if we are
2535
* about to flip a bit in the cpumask:
2537
if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2538
!cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2539
atomic_inc(&global_trace.data[cpu]->disabled);
2541
if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2542
cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2543
atomic_dec(&global_trace.data[cpu]->disabled);
2546
arch_spin_unlock(&ftrace_max_lock);
2549
cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2551
mutex_unlock(&tracing_cpumask_update_lock);
2552
free_cpumask_var(tracing_cpumask_new);
2557
free_cpumask_var(tracing_cpumask_new);
2562
static const struct file_operations tracing_cpumask_fops = {
2563
.open = tracing_open_generic,
2564
.read = tracing_cpumask_read,
2565
.write = tracing_cpumask_write,
2566
.llseek = generic_file_llseek,
2569
static int tracing_trace_options_show(struct seq_file *m, void *v)
2571
struct tracer_opt *trace_opts;
2575
mutex_lock(&trace_types_lock);
2576
tracer_flags = current_trace->flags->val;
2577
trace_opts = current_trace->flags->opts;
2579
for (i = 0; trace_options[i]; i++) {
2580
if (trace_flags & (1 << i))
2581
seq_printf(m, "%s\n", trace_options[i]);
2583
seq_printf(m, "no%s\n", trace_options[i]);
2586
for (i = 0; trace_opts[i].name; i++) {
2587
if (tracer_flags & trace_opts[i].bit)
2588
seq_printf(m, "%s\n", trace_opts[i].name);
2590
seq_printf(m, "no%s\n", trace_opts[i].name);
2592
mutex_unlock(&trace_types_lock);
2597
static int __set_tracer_option(struct tracer *trace,
2598
struct tracer_flags *tracer_flags,
2599
struct tracer_opt *opts, int neg)
2603
ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2608
tracer_flags->val &= ~opts->bit;
2610
tracer_flags->val |= opts->bit;
2614
/* Try to assign a tracer specific option */
2615
static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2617
struct tracer_flags *tracer_flags = trace->flags;
2618
struct tracer_opt *opts = NULL;
2621
for (i = 0; tracer_flags->opts[i].name; i++) {
2622
opts = &tracer_flags->opts[i];
2624
if (strcmp(cmp, opts->name) == 0)
2625
return __set_tracer_option(trace, trace->flags,
2632
static void set_tracer_flags(unsigned int mask, int enabled)
2634
/* do nothing if flag is already set */
2635
if (!!(trace_flags & mask) == !!enabled)
2639
trace_flags |= mask;
2641
trace_flags &= ~mask;
2643
if (mask == TRACE_ITER_RECORD_CMD)
2644
trace_event_enable_cmd_record(enabled);
2646
if (mask == TRACE_ITER_OVERWRITE)
2647
ring_buffer_change_overwrite(global_trace.buffer, enabled);
2651
tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2652
size_t cnt, loff_t *ppos)
2660
if (cnt >= sizeof(buf))
2663
if (copy_from_user(&buf, ubuf, cnt))
2667
cmp = strstrip(buf);
2669
if (strncmp(cmp, "no", 2) == 0) {
2674
for (i = 0; trace_options[i]; i++) {
2675
if (strcmp(cmp, trace_options[i]) == 0) {
2676
set_tracer_flags(1 << i, !neg);
2681
/* If no option could be set, test the specific tracer options */
2682
if (!trace_options[i]) {
2683
mutex_lock(&trace_types_lock);
2684
ret = set_tracer_option(current_trace, cmp, neg);
2685
mutex_unlock(&trace_types_lock);
2695
static int tracing_trace_options_open(struct inode *inode, struct file *file)
2697
if (tracing_disabled)
2699
return single_open(file, tracing_trace_options_show, NULL);
2702
static const struct file_operations tracing_iter_fops = {
2703
.open = tracing_trace_options_open,
2705
.llseek = seq_lseek,
2706
.release = single_release,
2707
.write = tracing_trace_options_write,
2710
static const char readme_msg[] =
2711
"tracing mini-HOWTO:\n\n"
2712
"# mount -t debugfs nodev /sys/kernel/debug\n\n"
2713
"# cat /sys/kernel/debug/tracing/available_tracers\n"
2714
"wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2715
"# cat /sys/kernel/debug/tracing/current_tracer\n"
2717
"# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2718
"# cat /sys/kernel/debug/tracing/current_tracer\n"
2720
"# cat /sys/kernel/debug/tracing/trace_options\n"
2721
"noprint-parent nosym-offset nosym-addr noverbose\n"
2722
"# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2723
"# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2724
"# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2725
"# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2729
tracing_readme_read(struct file *filp, char __user *ubuf,
2730
size_t cnt, loff_t *ppos)
2732
return simple_read_from_buffer(ubuf, cnt, ppos,
2733
readme_msg, strlen(readme_msg));
2736
static const struct file_operations tracing_readme_fops = {
2737
.open = tracing_open_generic,
2738
.read = tracing_readme_read,
2739
.llseek = generic_file_llseek,
2743
tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2744
size_t cnt, loff_t *ppos)
2753
file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2757
buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2765
for (i = 0; i < SAVED_CMDLINES; i++) {
2768
pid = map_cmdline_to_pid[i];
2769
if (pid == -1 || pid == NO_CMDLINE_MAP)
2772
trace_find_cmdline(pid, buf_comm);
2773
r = sprintf(buf, "%d %s\n", pid, buf_comm);
2778
len = simple_read_from_buffer(ubuf, cnt, ppos,
2787
static const struct file_operations tracing_saved_cmdlines_fops = {
2788
.open = tracing_open_generic,
2789
.read = tracing_saved_cmdlines_read,
2790
.llseek = generic_file_llseek,
2794
tracing_ctrl_read(struct file *filp, char __user *ubuf,
2795
size_t cnt, loff_t *ppos)
2800
r = sprintf(buf, "%u\n", tracer_enabled);
2801
return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2805
tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2806
size_t cnt, loff_t *ppos)
2808
struct trace_array *tr = filp->private_data;
2812
ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
2818
mutex_lock(&trace_types_lock);
2819
if (tracer_enabled ^ val) {
2821
/* Only need to warn if this is used to change the state */
2822
WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on");
2826
if (current_trace->start)
2827
current_trace->start(tr);
2832
if (current_trace->stop)
2833
current_trace->stop(tr);
2836
mutex_unlock(&trace_types_lock);
2844
tracing_set_trace_read(struct file *filp, char __user *ubuf,
2845
size_t cnt, loff_t *ppos)
2847
char buf[MAX_TRACER_SIZE+2];
2850
mutex_lock(&trace_types_lock);
2852
r = sprintf(buf, "%s\n", current_trace->name);
2854
r = sprintf(buf, "\n");
2855
mutex_unlock(&trace_types_lock);
2857
return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2860
int tracer_init(struct tracer *t, struct trace_array *tr)
2862
tracing_reset_online_cpus(tr);
2866
static int __tracing_resize_ring_buffer(unsigned long size)
2871
* If kernel or user changes the size of the ring buffer
2872
* we use the size that was given, and we can forget about
2873
* expanding it later.
2875
ring_buffer_expanded = 1;
2877
ret = ring_buffer_resize(global_trace.buffer, size);
2881
if (!current_trace->use_max_tr)
2884
ret = ring_buffer_resize(max_tr.buffer, size);
2888
r = ring_buffer_resize(global_trace.buffer,
2889
global_trace.entries);
2892
* AARGH! We are left with different
2893
* size max buffer!!!!
2894
* The max buffer is our "snapshot" buffer.
2895
* When a tracer needs a snapshot (one of the
2896
* latency tracers), it swaps the max buffer
2897
* with the saved snap shot. We succeeded to
2898
* update the size of the main buffer, but failed to
2899
* update the size of the max buffer. But when we tried
2900
* to reset the main buffer to the original size, we
2901
* failed there too. This is very unlikely to
2902
* happen, but if it does, warn and kill all
2906
tracing_disabled = 1;
2911
max_tr.entries = size;
2913
global_trace.entries = size;
2918
static ssize_t tracing_resize_ring_buffer(unsigned long size)
2920
int cpu, ret = size;
2922
mutex_lock(&trace_types_lock);
2926
/* disable all cpu buffers */
2927
for_each_tracing_cpu(cpu) {
2928
if (global_trace.data[cpu])
2929
atomic_inc(&global_trace.data[cpu]->disabled);
2930
if (max_tr.data[cpu])
2931
atomic_inc(&max_tr.data[cpu]->disabled);
2934
if (size != global_trace.entries)
2935
ret = __tracing_resize_ring_buffer(size);
2940
for_each_tracing_cpu(cpu) {
2941
if (global_trace.data[cpu])
2942
atomic_dec(&global_trace.data[cpu]->disabled);
2943
if (max_tr.data[cpu])
2944
atomic_dec(&max_tr.data[cpu]->disabled);
2948
mutex_unlock(&trace_types_lock);
2955
* tracing_update_buffers - used by tracing facility to expand ring buffers
2957
* To save on memory when the tracing is never used on a system with it
2958
* configured in. The ring buffers are set to a minimum size. But once
2959
* a user starts to use the tracing facility, then they need to grow
2960
* to their default size.
2962
* This function is to be called when a tracer is about to be used.
2964
int tracing_update_buffers(void)
2968
mutex_lock(&trace_types_lock);
2969
if (!ring_buffer_expanded)
2970
ret = __tracing_resize_ring_buffer(trace_buf_size);
2971
mutex_unlock(&trace_types_lock);
2976
struct trace_option_dentry;
2978
static struct trace_option_dentry *
2979
create_trace_option_files(struct tracer *tracer);
2982
destroy_trace_option_files(struct trace_option_dentry *topts);
2984
static int tracing_set_tracer(const char *buf)
2986
static struct trace_option_dentry *topts;
2987
struct trace_array *tr = &global_trace;
2991
mutex_lock(&trace_types_lock);
2993
if (!ring_buffer_expanded) {
2994
ret = __tracing_resize_ring_buffer(trace_buf_size);
3000
for (t = trace_types; t; t = t->next) {
3001
if (strcmp(t->name, buf) == 0)
3008
if (t == current_trace)
3011
trace_branch_disable();
3012
if (current_trace && current_trace->reset)
3013
current_trace->reset(tr);
3014
if (current_trace && current_trace->use_max_tr) {
3016
* We don't free the ring buffer. instead, resize it because
3017
* The max_tr ring buffer has some state (e.g. ring->clock) and
3018
* we want preserve it.
3020
ring_buffer_resize(max_tr.buffer, 1);
3023
destroy_trace_option_files(topts);
3027
topts = create_trace_option_files(current_trace);
3028
if (current_trace->use_max_tr) {
3029
ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
3032
max_tr.entries = global_trace.entries;
3036
ret = tracer_init(t, tr);
3041
trace_branch_enable(tr);
3043
mutex_unlock(&trace_types_lock);
3049
tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3050
size_t cnt, loff_t *ppos)
3052
char buf[MAX_TRACER_SIZE+1];
3059
if (cnt > MAX_TRACER_SIZE)
3060
cnt = MAX_TRACER_SIZE;
3062
if (copy_from_user(&buf, ubuf, cnt))
3067
/* strip ending whitespace. */
3068
for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3071
err = tracing_set_tracer(buf);
3081
tracing_max_lat_read(struct file *filp, char __user *ubuf,
3082
size_t cnt, loff_t *ppos)
3084
unsigned long *ptr = filp->private_data;
3088
r = snprintf(buf, sizeof(buf), "%ld\n",
3089
*ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3090
if (r > sizeof(buf))
3092
return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3096
tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3097
size_t cnt, loff_t *ppos)
3099
unsigned long *ptr = filp->private_data;
3103
ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3112
static int tracing_open_pipe(struct inode *inode, struct file *filp)
3114
long cpu_file = (long) inode->i_private;
3115
struct trace_iterator *iter;
3118
if (tracing_disabled)
3121
mutex_lock(&trace_types_lock);
3123
/* create a buffer to store the information to pass to userspace */
3124
iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3131
* We make a copy of the current tracer to avoid concurrent
3132
* changes on it while we are reading.
3134
iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3140
*iter->trace = *current_trace;
3142
if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3147
/* trace pipe does not show start of buffer */
3148
cpumask_setall(iter->started);
3150
if (trace_flags & TRACE_ITER_LATENCY_FMT)
3151
iter->iter_flags |= TRACE_FILE_LAT_FMT;
3153
iter->cpu_file = cpu_file;
3154
iter->tr = &global_trace;
3155
mutex_init(&iter->mutex);
3156
filp->private_data = iter;
3158
if (iter->trace->pipe_open)
3159
iter->trace->pipe_open(iter);
3161
nonseekable_open(inode, filp);
3163
mutex_unlock(&trace_types_lock);
3169
mutex_unlock(&trace_types_lock);
3173
static int tracing_release_pipe(struct inode *inode, struct file *file)
3175
struct trace_iterator *iter = file->private_data;
3177
mutex_lock(&trace_types_lock);
3179
if (iter->trace->pipe_close)
3180
iter->trace->pipe_close(iter);
3182
mutex_unlock(&trace_types_lock);
3184
free_cpumask_var(iter->started);
3185
mutex_destroy(&iter->mutex);
3193
tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3195
struct trace_iterator *iter = filp->private_data;
3197
if (trace_flags & TRACE_ITER_BLOCK) {
3199
* Always select as readable when in blocking mode
3201
return POLLIN | POLLRDNORM;
3203
if (!trace_empty(iter))
3204
return POLLIN | POLLRDNORM;
3205
poll_wait(filp, &trace_wait, poll_table);
3206
if (!trace_empty(iter))
3207
return POLLIN | POLLRDNORM;
3214
void default_wait_pipe(struct trace_iterator *iter)
3218
prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3220
if (trace_empty(iter))
3223
finish_wait(&trace_wait, &wait);
3227
* This is a make-shift waitqueue.
3228
* A tracer might use this callback on some rare cases:
3230
* 1) the current tracer might hold the runqueue lock when it wakes up
3231
* a reader, hence a deadlock (sched, function, and function graph tracers)
3232
* 2) the function tracers, trace all functions, we don't want
3233
* the overhead of calling wake_up and friends
3234
* (and tracing them too)
3236
* Anyway, this is really very primitive wakeup.
3238
void poll_wait_pipe(struct trace_iterator *iter)
3240
set_current_state(TASK_INTERRUPTIBLE);
3241
/* sleep for 100 msecs, and try again. */
3242
schedule_timeout(HZ / 10);
3245
/* Must be called with trace_types_lock mutex held. */
3246
static int tracing_wait_pipe(struct file *filp)
3248
struct trace_iterator *iter = filp->private_data;
3250
while (trace_empty(iter)) {
3252
if ((filp->f_flags & O_NONBLOCK)) {
3256
mutex_unlock(&iter->mutex);
3258
iter->trace->wait_pipe(iter);
3260
mutex_lock(&iter->mutex);
3262
if (signal_pending(current))
3266
* We block until we read something and tracing is disabled.
3267
* We still block if tracing is disabled, but we have never
3268
* read anything. This allows a user to cat this file, and
3269
* then enable tracing. But after we have read something,
3270
* we give an EOF when tracing is again disabled.
3272
* iter->pos will be 0 if we haven't read anything.
3274
if (!tracer_enabled && iter->pos)
3285
tracing_read_pipe(struct file *filp, char __user *ubuf,
3286
size_t cnt, loff_t *ppos)
3288
struct trace_iterator *iter = filp->private_data;
3289
static struct tracer *old_tracer;
3292
/* return any leftover data */
3293
sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3297
trace_seq_init(&iter->seq);
3299
/* copy the tracer to avoid using a global lock all around */
3300
mutex_lock(&trace_types_lock);
3301
if (unlikely(old_tracer != current_trace && current_trace)) {
3302
old_tracer = current_trace;
3303
*iter->trace = *current_trace;
3305
mutex_unlock(&trace_types_lock);
3308
* Avoid more than one consumer on a single file descriptor
3309
* This is just a matter of traces coherency, the ring buffer itself
3312
mutex_lock(&iter->mutex);
3313
if (iter->trace->read) {
3314
sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3320
sret = tracing_wait_pipe(filp);
3324
/* stop when tracing is finished */
3325
if (trace_empty(iter)) {
3330
if (cnt >= PAGE_SIZE)
3331
cnt = PAGE_SIZE - 1;
3333
/* reset all but tr, trace, and overruns */
3334
memset(&iter->seq, 0,
3335
sizeof(struct trace_iterator) -
3336
offsetof(struct trace_iterator, seq));
3339
trace_event_read_lock();
3340
trace_access_lock(iter->cpu_file);
3341
while (trace_find_next_entry_inc(iter) != NULL) {
3342
enum print_line_t ret;
3343
int len = iter->seq.len;
3345
ret = print_trace_line(iter);
3346
if (ret == TRACE_TYPE_PARTIAL_LINE) {
3347
/* don't print partial lines */
3348
iter->seq.len = len;
3351
if (ret != TRACE_TYPE_NO_CONSUME)
3352
trace_consume(iter);
3354
if (iter->seq.len >= cnt)
3358
* Setting the full flag means we reached the trace_seq buffer
3359
* size and we should leave by partial output condition above.
3360
* One of the trace_seq_* functions is not used properly.
3362
WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3365
trace_access_unlock(iter->cpu_file);
3366
trace_event_read_unlock();
3368
/* Now copy what we have to the user */
3369
sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3370
if (iter->seq.readpos >= iter->seq.len)
3371
trace_seq_init(&iter->seq);
3374
* If there was nothing to send to user, in spite of consuming trace
3375
* entries, go back to wait for more entries.
3381
mutex_unlock(&iter->mutex);
3386
static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3387
struct pipe_buffer *buf)
3389
__free_page(buf->page);
3392
static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3395
__free_page(spd->pages[idx]);
3398
static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3400
.map = generic_pipe_buf_map,
3401
.unmap = generic_pipe_buf_unmap,
3402
.confirm = generic_pipe_buf_confirm,
3403
.release = tracing_pipe_buf_release,
3404
.steal = generic_pipe_buf_steal,
3405
.get = generic_pipe_buf_get,
3409
tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3414
/* Seq buffer is page-sized, exactly what we need. */
3416
count = iter->seq.len;
3417
ret = print_trace_line(iter);
3418
count = iter->seq.len - count;
3421
iter->seq.len -= count;
3424
if (ret == TRACE_TYPE_PARTIAL_LINE) {
3425
iter->seq.len -= count;
3429
if (ret != TRACE_TYPE_NO_CONSUME)
3430
trace_consume(iter);
3432
if (!trace_find_next_entry_inc(iter)) {
3442
static ssize_t tracing_splice_read_pipe(struct file *filp,
3444
struct pipe_inode_info *pipe,
3448
struct page *pages_def[PIPE_DEF_BUFFERS];
3449
struct partial_page partial_def[PIPE_DEF_BUFFERS];
3450
struct trace_iterator *iter = filp->private_data;
3451
struct splice_pipe_desc spd = {
3453
.partial = partial_def,
3454
.nr_pages = 0, /* This gets updated below. */
3456
.ops = &tracing_pipe_buf_ops,
3457
.spd_release = tracing_spd_release_pipe,
3459
static struct tracer *old_tracer;
3464
if (splice_grow_spd(pipe, &spd))
3467
/* copy the tracer to avoid using a global lock all around */
3468
mutex_lock(&trace_types_lock);
3469
if (unlikely(old_tracer != current_trace && current_trace)) {
3470
old_tracer = current_trace;
3471
*iter->trace = *current_trace;
3473
mutex_unlock(&trace_types_lock);
3475
mutex_lock(&iter->mutex);
3477
if (iter->trace->splice_read) {
3478
ret = iter->trace->splice_read(iter, filp,
3479
ppos, pipe, len, flags);
3484
ret = tracing_wait_pipe(filp);
3488
if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3493
trace_event_read_lock();
3494
trace_access_lock(iter->cpu_file);
3496
/* Fill as many pages as possible. */
3497
for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3498
spd.pages[i] = alloc_page(GFP_KERNEL);
3502
rem = tracing_fill_pipe_page(rem, iter);
3504
/* Copy the data into the page, so we can start over. */
3505
ret = trace_seq_to_buffer(&iter->seq,
3506
page_address(spd.pages[i]),
3509
__free_page(spd.pages[i]);
3512
spd.partial[i].offset = 0;
3513
spd.partial[i].len = iter->seq.len;
3515
trace_seq_init(&iter->seq);
3518
trace_access_unlock(iter->cpu_file);
3519
trace_event_read_unlock();
3520
mutex_unlock(&iter->mutex);
3524
ret = splice_to_pipe(pipe, &spd);
3526
splice_shrink_spd(pipe, &spd);
3530
mutex_unlock(&iter->mutex);
3535
tracing_entries_read(struct file *filp, char __user *ubuf,
3536
size_t cnt, loff_t *ppos)
3538
struct trace_array *tr = filp->private_data;
3542
mutex_lock(&trace_types_lock);
3543
if (!ring_buffer_expanded)
3544
r = sprintf(buf, "%lu (expanded: %lu)\n",
3546
trace_buf_size >> 10);
3548
r = sprintf(buf, "%lu\n", tr->entries >> 10);
3549
mutex_unlock(&trace_types_lock);
3551
return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3555
tracing_entries_write(struct file *filp, const char __user *ubuf,
3556
size_t cnt, loff_t *ppos)
3561
ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3565
/* must have at least 1 entry */
3569
/* value is in KB */
3572
ret = tracing_resize_ring_buffer(val);
3582
tracing_total_entries_read(struct file *filp, char __user *ubuf,
3583
size_t cnt, loff_t *ppos)
3585
struct trace_array *tr = filp->private_data;
3588
unsigned long size = 0, expanded_size = 0;
3590
mutex_lock(&trace_types_lock);
3591
for_each_tracing_cpu(cpu) {
3592
size += tr->entries >> 10;
3593
if (!ring_buffer_expanded)
3594
expanded_size += trace_buf_size >> 10;
3596
if (ring_buffer_expanded)
3597
r = sprintf(buf, "%lu\n", size);
3599
r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
3600
mutex_unlock(&trace_types_lock);
3602
return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3606
tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
3607
size_t cnt, loff_t *ppos)
3610
* There is no need to read what the user has written, this function
3611
* is just to make sure that there is no error when "echo" is used
3620
tracing_free_buffer_release(struct inode *inode, struct file *filp)
3622
/* disable tracing ? */
3623
if (trace_flags & TRACE_ITER_STOP_ON_FREE)
3625
/* resize the ring buffer to 0 */
3626
tracing_resize_ring_buffer(0);
3632
tracing_mark_write(struct file *filp, const char __user *ubuf,
3633
size_t cnt, loff_t *fpos)
3635
unsigned long addr = (unsigned long)ubuf;
3636
struct ring_buffer_event *event;
3637
struct ring_buffer *buffer;
3638
struct print_entry *entry;
3639
unsigned long irq_flags;
3640
struct page *pages[2];
3650
if (tracing_disabled)
3653
if (cnt > TRACE_BUF_SIZE)
3654
cnt = TRACE_BUF_SIZE;
3657
* Userspace is injecting traces into the kernel trace buffer.
3658
* We want to be as non intrusive as possible.
3659
* To do so, we do not want to allocate any special buffers
3660
* or take any locks, but instead write the userspace data
3661
* straight into the ring buffer.
3663
* First we need to pin the userspace buffer into memory,
3664
* which, most likely it is, because it just referenced it.
3665
* But there's no guarantee that it is. By using get_user_pages_fast()
3666
* and kmap_atomic/kunmap_atomic() we can get access to the
3667
* pages directly. We then write the data directly into the
3670
BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
3672
/* check if we cross pages */
3673
if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
3676
offset = addr & (PAGE_SIZE - 1);
3679
ret = get_user_pages_fast(addr, nr_pages, 0, pages);
3680
if (ret < nr_pages) {
3682
put_page(pages[ret]);
3687
page1 = kmap_atomic(pages[0]);
3689
page2 = kmap_atomic(pages[1]);
3691
local_save_flags(irq_flags);
3692
size = sizeof(*entry) + cnt + 2; /* possible \n added */
3693
buffer = global_trace.buffer;
3694
event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3695
irq_flags, preempt_count());
3697
/* Ring buffer disabled, return as if not open for write */
3702
entry = ring_buffer_event_data(event);
3703
entry->ip = _THIS_IP_;
3705
if (nr_pages == 2) {
3706
len = PAGE_SIZE - offset;
3707
memcpy(&entry->buf, page1 + offset, len);
3708
memcpy(&entry->buf[len], page2, cnt - len);
3710
memcpy(&entry->buf, page1 + offset, cnt);
3712
if (entry->buf[cnt - 1] != '\n') {
3713
entry->buf[cnt] = '\n';
3714
entry->buf[cnt + 1] = '\0';
3716
entry->buf[cnt] = '\0';
3718
ring_buffer_unlock_commit(buffer, event);
3726
kunmap_atomic(page2);
3727
kunmap_atomic(page1);
3728
while (nr_pages > 0)
3729
put_page(pages[--nr_pages]);
3734
static int tracing_clock_show(struct seq_file *m, void *v)
3738
for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3740
"%s%s%s%s", i ? " " : "",
3741
i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3742
i == trace_clock_id ? "]" : "");
3748
static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3749
size_t cnt, loff_t *fpos)
3752
const char *clockstr;
3755
if (cnt >= sizeof(buf))
3758
if (copy_from_user(&buf, ubuf, cnt))
3763
clockstr = strstrip(buf);
3765
for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3766
if (strcmp(trace_clocks[i].name, clockstr) == 0)
3769
if (i == ARRAY_SIZE(trace_clocks))
3774
mutex_lock(&trace_types_lock);
3776
ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3778
ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3780
mutex_unlock(&trace_types_lock);
3787
static int tracing_clock_open(struct inode *inode, struct file *file)
3789
if (tracing_disabled)
3791
return single_open(file, tracing_clock_show, NULL);
3794
static const struct file_operations tracing_max_lat_fops = {
3795
.open = tracing_open_generic,
3796
.read = tracing_max_lat_read,
3797
.write = tracing_max_lat_write,
3798
.llseek = generic_file_llseek,
3801
static const struct file_operations tracing_ctrl_fops = {
3802
.open = tracing_open_generic,
3803
.read = tracing_ctrl_read,
3804
.write = tracing_ctrl_write,
3805
.llseek = generic_file_llseek,
3808
static const struct file_operations set_tracer_fops = {
3809
.open = tracing_open_generic,
3810
.read = tracing_set_trace_read,
3811
.write = tracing_set_trace_write,
3812
.llseek = generic_file_llseek,
3815
static const struct file_operations tracing_pipe_fops = {
3816
.open = tracing_open_pipe,
3817
.poll = tracing_poll_pipe,
3818
.read = tracing_read_pipe,
3819
.splice_read = tracing_splice_read_pipe,
3820
.release = tracing_release_pipe,
3821
.llseek = no_llseek,
3824
static const struct file_operations tracing_entries_fops = {
3825
.open = tracing_open_generic,
3826
.read = tracing_entries_read,
3827
.write = tracing_entries_write,
3828
.llseek = generic_file_llseek,
3831
static const struct file_operations tracing_total_entries_fops = {
3832
.open = tracing_open_generic,
3833
.read = tracing_total_entries_read,
3834
.llseek = generic_file_llseek,
3837
static const struct file_operations tracing_free_buffer_fops = {
3838
.write = tracing_free_buffer_write,
3839
.release = tracing_free_buffer_release,
3842
static const struct file_operations tracing_mark_fops = {
3843
.open = tracing_open_generic,
3844
.write = tracing_mark_write,
3845
.llseek = generic_file_llseek,
3848
static const struct file_operations trace_clock_fops = {
3849
.open = tracing_clock_open,
3851
.llseek = seq_lseek,
3852
.release = single_release,
3853
.write = tracing_clock_write,
3856
struct ftrace_buffer_info {
3857
struct trace_array *tr;
3863
static int tracing_buffers_open(struct inode *inode, struct file *filp)
3865
int cpu = (int)(long)inode->i_private;
3866
struct ftrace_buffer_info *info;
3868
if (tracing_disabled)
3871
info = kzalloc(sizeof(*info), GFP_KERNEL);
3875
info->tr = &global_trace;
3878
/* Force reading ring buffer for first read */
3879
info->read = (unsigned int)-1;
3881
filp->private_data = info;
3883
return nonseekable_open(inode, filp);
3887
tracing_buffers_read(struct file *filp, char __user *ubuf,
3888
size_t count, loff_t *ppos)
3890
struct ftrace_buffer_info *info = filp->private_data;
3898
info->spare = ring_buffer_alloc_read_page(info->tr->buffer, info->cpu);
3902
/* Do we have previous read data to read? */
3903
if (info->read < PAGE_SIZE)
3906
trace_access_lock(info->cpu);
3907
ret = ring_buffer_read_page(info->tr->buffer,
3911
trace_access_unlock(info->cpu);
3918
size = PAGE_SIZE - info->read;
3922
ret = copy_to_user(ubuf, info->spare + info->read, size);
3933
static int tracing_buffers_release(struct inode *inode, struct file *file)
3935
struct ftrace_buffer_info *info = file->private_data;
3938
ring_buffer_free_read_page(info->tr->buffer, info->spare);
3945
struct ring_buffer *buffer;
3950
static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3951
struct pipe_buffer *buf)
3953
struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3958
ring_buffer_free_read_page(ref->buffer, ref->page);
3963
static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3964
struct pipe_buffer *buf)
3969
static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3970
struct pipe_buffer *buf)
3972
struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3977
/* Pipe buffer operations for a buffer. */
3978
static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3980
.map = generic_pipe_buf_map,
3981
.unmap = generic_pipe_buf_unmap,
3982
.confirm = generic_pipe_buf_confirm,
3983
.release = buffer_pipe_buf_release,
3984
.steal = buffer_pipe_buf_steal,
3985
.get = buffer_pipe_buf_get,
3989
* Callback from splice_to_pipe(), if we need to release some pages
3990
* at the end of the spd in case we error'ed out in filling the pipe.
3992
static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3994
struct buffer_ref *ref =
3995
(struct buffer_ref *)spd->partial[i].private;
4000
ring_buffer_free_read_page(ref->buffer, ref->page);
4002
spd->partial[i].private = 0;
4006
tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4007
struct pipe_inode_info *pipe, size_t len,
4010
struct ftrace_buffer_info *info = file->private_data;
4011
struct partial_page partial_def[PIPE_DEF_BUFFERS];
4012
struct page *pages_def[PIPE_DEF_BUFFERS];
4013
struct splice_pipe_desc spd = {
4015
.partial = partial_def,
4017
.ops = &buffer_pipe_buf_ops,
4018
.spd_release = buffer_spd_release,
4020
struct buffer_ref *ref;
4021
int entries, size, i;
4024
if (splice_grow_spd(pipe, &spd))
4027
if (*ppos & (PAGE_SIZE - 1)) {
4028
WARN_ONCE(1, "Ftrace: previous read must page-align\n");
4033
if (len & (PAGE_SIZE - 1)) {
4034
WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
4035
if (len < PAGE_SIZE) {
4042
trace_access_lock(info->cpu);
4043
entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4045
for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4049
ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4054
ref->buffer = info->tr->buffer;
4055
ref->page = ring_buffer_alloc_read_page(ref->buffer, info->cpu);
4061
r = ring_buffer_read_page(ref->buffer, &ref->page,
4064
ring_buffer_free_read_page(ref->buffer, ref->page);
4070
* zero out any left over data, this is going to
4073
size = ring_buffer_page_len(ref->page);
4074
if (size < PAGE_SIZE)
4075
memset(ref->page + size, 0, PAGE_SIZE - size);
4077
page = virt_to_page(ref->page);
4079
spd.pages[i] = page;
4080
spd.partial[i].len = PAGE_SIZE;
4081
spd.partial[i].offset = 0;
4082
spd.partial[i].private = (unsigned long)ref;
4086
entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4089
trace_access_unlock(info->cpu);
4092
/* did we read anything? */
4093
if (!spd.nr_pages) {
4094
if (flags & SPLICE_F_NONBLOCK)
4102
ret = splice_to_pipe(pipe, &spd);
4103
splice_shrink_spd(pipe, &spd);
4108
static const struct file_operations tracing_buffers_fops = {
4109
.open = tracing_buffers_open,
4110
.read = tracing_buffers_read,
4111
.release = tracing_buffers_release,
4112
.splice_read = tracing_buffers_splice_read,
4113
.llseek = no_llseek,
4117
tracing_stats_read(struct file *filp, char __user *ubuf,
4118
size_t count, loff_t *ppos)
4120
unsigned long cpu = (unsigned long)filp->private_data;
4121
struct trace_array *tr = &global_trace;
4122
struct trace_seq *s;
4124
unsigned long long t;
4125
unsigned long usec_rem;
4127
s = kmalloc(sizeof(*s), GFP_KERNEL);
4133
cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
4134
trace_seq_printf(s, "entries: %ld\n", cnt);
4136
cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
4137
trace_seq_printf(s, "overrun: %ld\n", cnt);
4139
cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
4140
trace_seq_printf(s, "commit overrun: %ld\n", cnt);
4142
cnt = ring_buffer_bytes_cpu(tr->buffer, cpu);
4143
trace_seq_printf(s, "bytes: %ld\n", cnt);
4145
t = ns2usecs(ring_buffer_oldest_event_ts(tr->buffer, cpu));
4146
usec_rem = do_div(t, USEC_PER_SEC);
4147
trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n", t, usec_rem);
4149
t = ns2usecs(ring_buffer_time_stamp(tr->buffer, cpu));
4150
usec_rem = do_div(t, USEC_PER_SEC);
4151
trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
4153
count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
4160
static const struct file_operations tracing_stats_fops = {
4161
.open = tracing_open_generic,
4162
.read = tracing_stats_read,
4163
.llseek = generic_file_llseek,
4166
#ifdef CONFIG_DYNAMIC_FTRACE
4168
int __weak ftrace_arch_read_dyn_info(char *buf, int size)
4174
tracing_read_dyn_info(struct file *filp, char __user *ubuf,
4175
size_t cnt, loff_t *ppos)
4177
static char ftrace_dyn_info_buffer[1024];
4178
static DEFINE_MUTEX(dyn_info_mutex);
4179
unsigned long *p = filp->private_data;
4180
char *buf = ftrace_dyn_info_buffer;
4181
int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
4184
mutex_lock(&dyn_info_mutex);
4185
r = sprintf(buf, "%ld ", *p);
4187
r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
4190
r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4192
mutex_unlock(&dyn_info_mutex);
4197
static const struct file_operations tracing_dyn_info_fops = {
4198
.open = tracing_open_generic,
4199
.read = tracing_read_dyn_info,
4200
.llseek = generic_file_llseek,
4204
static struct dentry *d_tracer;
4206
struct dentry *tracing_init_dentry(void)
4213
if (!debugfs_initialized())
4216
d_tracer = debugfs_create_dir("tracing", NULL);
4218
if (!d_tracer && !once) {
4220
pr_warning("Could not create debugfs directory 'tracing'\n");
4227
static struct dentry *d_percpu;
4229
struct dentry *tracing_dentry_percpu(void)
4232
struct dentry *d_tracer;
4237
d_tracer = tracing_init_dentry();
4242
d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4244
if (!d_percpu && !once) {
4246
pr_warning("Could not create debugfs directory 'per_cpu'\n");
4253
static void tracing_init_debugfs_percpu(long cpu)
4255
struct dentry *d_percpu = tracing_dentry_percpu();
4256
struct dentry *d_cpu;
4257
char cpu_dir[30]; /* 30 characters should be more than enough */
4259
snprintf(cpu_dir, 30, "cpu%ld", cpu);
4260
d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4262
pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4266
/* per cpu trace_pipe */
4267
trace_create_file("trace_pipe", 0444, d_cpu,
4268
(void *) cpu, &tracing_pipe_fops);
4271
trace_create_file("trace", 0644, d_cpu,
4272
(void *) cpu, &tracing_fops);
4274
trace_create_file("trace_pipe_raw", 0444, d_cpu,
4275
(void *) cpu, &tracing_buffers_fops);
4277
trace_create_file("stats", 0444, d_cpu,
4278
(void *) cpu, &tracing_stats_fops);
4281
#ifdef CONFIG_FTRACE_SELFTEST
4282
/* Let selftest have access to static functions in this file */
4283
#include "trace_selftest.c"
4286
struct trace_option_dentry {
4287
struct tracer_opt *opt;
4288
struct tracer_flags *flags;
4289
struct dentry *entry;
4293
trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4296
struct trace_option_dentry *topt = filp->private_data;
4299
if (topt->flags->val & topt->opt->bit)
4304
return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4308
trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4311
struct trace_option_dentry *topt = filp->private_data;
4315
ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4319
if (val != 0 && val != 1)
4322
if (!!(topt->flags->val & topt->opt->bit) != val) {
4323
mutex_lock(&trace_types_lock);
4324
ret = __set_tracer_option(current_trace, topt->flags,
4326
mutex_unlock(&trace_types_lock);
4337
static const struct file_operations trace_options_fops = {
4338
.open = tracing_open_generic,
4339
.read = trace_options_read,
4340
.write = trace_options_write,
4341
.llseek = generic_file_llseek,
4345
trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4348
long index = (long)filp->private_data;
4351
if (trace_flags & (1 << index))
4356
return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4360
trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4363
long index = (long)filp->private_data;
4367
ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4371
if (val != 0 && val != 1)
4373
set_tracer_flags(1 << index, val);
4380
static const struct file_operations trace_options_core_fops = {
4381
.open = tracing_open_generic,
4382
.read = trace_options_core_read,
4383
.write = trace_options_core_write,
4384
.llseek = generic_file_llseek,
4387
struct dentry *trace_create_file(const char *name,
4389
struct dentry *parent,
4391
const struct file_operations *fops)
4395
ret = debugfs_create_file(name, mode, parent, data, fops);
4397
pr_warning("Could not create debugfs '%s' entry\n", name);
4403
static struct dentry *trace_options_init_dentry(void)
4405
struct dentry *d_tracer;
4406
static struct dentry *t_options;
4411
d_tracer = tracing_init_dentry();
4415
t_options = debugfs_create_dir("options", d_tracer);
4417
pr_warning("Could not create debugfs directory 'options'\n");
4425
create_trace_option_file(struct trace_option_dentry *topt,
4426
struct tracer_flags *flags,
4427
struct tracer_opt *opt)
4429
struct dentry *t_options;
4431
t_options = trace_options_init_dentry();
4435
topt->flags = flags;
4438
topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4439
&trace_options_fops);
4443
static struct trace_option_dentry *
4444
create_trace_option_files(struct tracer *tracer)
4446
struct trace_option_dentry *topts;
4447
struct tracer_flags *flags;
4448
struct tracer_opt *opts;
4454
flags = tracer->flags;
4456
if (!flags || !flags->opts)
4461
for (cnt = 0; opts[cnt].name; cnt++)
4464
topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4468
for (cnt = 0; opts[cnt].name; cnt++)
4469
create_trace_option_file(&topts[cnt], flags,
4476
destroy_trace_option_files(struct trace_option_dentry *topts)
4483
for (cnt = 0; topts[cnt].opt; cnt++) {
4484
if (topts[cnt].entry)
4485
debugfs_remove(topts[cnt].entry);
4491
static struct dentry *
4492
create_trace_option_core_file(const char *option, long index)
4494
struct dentry *t_options;
4496
t_options = trace_options_init_dentry();
4500
return trace_create_file(option, 0644, t_options, (void *)index,
4501
&trace_options_core_fops);
4504
static __init void create_trace_options_dir(void)
4506
struct dentry *t_options;
4509
t_options = trace_options_init_dentry();
4513
for (i = 0; trace_options[i]; i++)
4514
create_trace_option_core_file(trace_options[i], i);
4517
static __init int tracer_init_debugfs(void)
4519
struct dentry *d_tracer;
4522
trace_access_lock_init();
4524
d_tracer = tracing_init_dentry();
4526
trace_create_file("tracing_enabled", 0644, d_tracer,
4527
&global_trace, &tracing_ctrl_fops);
4529
trace_create_file("trace_options", 0644, d_tracer,
4530
NULL, &tracing_iter_fops);
4532
trace_create_file("tracing_cpumask", 0644, d_tracer,
4533
NULL, &tracing_cpumask_fops);
4535
trace_create_file("trace", 0644, d_tracer,
4536
(void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4538
trace_create_file("available_tracers", 0444, d_tracer,
4539
&global_trace, &show_traces_fops);
4541
trace_create_file("current_tracer", 0644, d_tracer,
4542
&global_trace, &set_tracer_fops);
4544
#ifdef CONFIG_TRACER_MAX_TRACE
4545
trace_create_file("tracing_max_latency", 0644, d_tracer,
4546
&tracing_max_latency, &tracing_max_lat_fops);
4549
trace_create_file("tracing_thresh", 0644, d_tracer,
4550
&tracing_thresh, &tracing_max_lat_fops);
4552
trace_create_file("README", 0444, d_tracer,
4553
NULL, &tracing_readme_fops);
4555
trace_create_file("trace_pipe", 0444, d_tracer,
4556
(void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4558
trace_create_file("buffer_size_kb", 0644, d_tracer,
4559
&global_trace, &tracing_entries_fops);
4561
trace_create_file("buffer_total_size_kb", 0444, d_tracer,
4562
&global_trace, &tracing_total_entries_fops);
4564
trace_create_file("free_buffer", 0644, d_tracer,
4565
&global_trace, &tracing_free_buffer_fops);
4567
trace_create_file("trace_marker", 0220, d_tracer,
4568
NULL, &tracing_mark_fops);
4570
trace_create_file("saved_cmdlines", 0444, d_tracer,
4571
NULL, &tracing_saved_cmdlines_fops);
4573
trace_create_file("trace_clock", 0644, d_tracer, NULL,
4576
#ifdef CONFIG_DYNAMIC_FTRACE
4577
trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4578
&ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4581
create_trace_options_dir();
4583
for_each_tracing_cpu(cpu)
4584
tracing_init_debugfs_percpu(cpu);
4589
static int trace_panic_handler(struct notifier_block *this,
4590
unsigned long event, void *unused)
4592
if (ftrace_dump_on_oops)
4593
ftrace_dump(ftrace_dump_on_oops);
4597
static struct notifier_block trace_panic_notifier = {
4598
.notifier_call = trace_panic_handler,
4600
.priority = 150 /* priority: INT_MAX >= x >= 0 */
4603
static int trace_die_handler(struct notifier_block *self,
4609
if (ftrace_dump_on_oops)
4610
ftrace_dump(ftrace_dump_on_oops);
4618
static struct notifier_block trace_die_notifier = {
4619
.notifier_call = trace_die_handler,
4624
* printk is set to max of 1024, we really don't need it that big.
4625
* Nothing should be printing 1000 characters anyway.
4627
#define TRACE_MAX_PRINT 1000
4630
* Define here KERN_TRACE so that we have one place to modify
4631
* it if we decide to change what log level the ftrace dump
4634
#define KERN_TRACE KERN_EMERG
4637
trace_printk_seq(struct trace_seq *s)
4639
/* Probably should print a warning here. */
4643
/* should be zero ended, but we are paranoid. */
4644
s->buffer[s->len] = 0;
4646
printk(KERN_TRACE "%s", s->buffer);
4651
void trace_init_global_iter(struct trace_iterator *iter)
4653
iter->tr = &global_trace;
4654
iter->trace = current_trace;
4655
iter->cpu_file = TRACE_PIPE_ALL_CPU;
4659
__ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4661
static arch_spinlock_t ftrace_dump_lock =
4662
(arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4663
/* use static because iter can be a bit big for the stack */
4664
static struct trace_iterator iter;
4665
unsigned int old_userobj;
4666
static int dump_ran;
4667
unsigned long flags;
4671
local_irq_save(flags);
4672
arch_spin_lock(&ftrace_dump_lock);
4680
/* Did function tracer already get disabled? */
4681
if (ftrace_is_dead()) {
4682
printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
4683
printk("# MAY BE MISSING FUNCTION EVENTS\n");
4686
if (disable_tracing)
4689
trace_init_global_iter(&iter);
4691
for_each_tracing_cpu(cpu) {
4692
atomic_inc(&iter.tr->data[cpu]->disabled);
4695
old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4697
/* don't look at user memory in panic mode */
4698
trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4700
/* Simulate the iterator */
4701
iter.tr = &global_trace;
4702
iter.trace = current_trace;
4704
switch (oops_dump_mode) {
4706
iter.cpu_file = TRACE_PIPE_ALL_CPU;
4709
iter.cpu_file = raw_smp_processor_id();
4714
printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4715
iter.cpu_file = TRACE_PIPE_ALL_CPU;
4718
printk(KERN_TRACE "Dumping ftrace buffer:\n");
4721
* We need to stop all tracing on all CPUS to read the
4722
* the next buffer. This is a bit expensive, but is
4723
* not done often. We fill all what we can read,
4724
* and then release the locks again.
4727
while (!trace_empty(&iter)) {
4730
printk(KERN_TRACE "---------------------------------\n");
4734
/* reset all but tr, trace, and overruns */
4735
memset(&iter.seq, 0,
4736
sizeof(struct trace_iterator) -
4737
offsetof(struct trace_iterator, seq));
4738
iter.iter_flags |= TRACE_FILE_LAT_FMT;
4741
if (trace_find_next_entry_inc(&iter) != NULL) {
4744
ret = print_trace_line(&iter);
4745
if (ret != TRACE_TYPE_NO_CONSUME)
4746
trace_consume(&iter);
4749
trace_printk_seq(&iter.seq);
4753
printk(KERN_TRACE " (ftrace buffer empty)\n");
4755
printk(KERN_TRACE "---------------------------------\n");
4758
/* Re-enable tracing if requested */
4759
if (!disable_tracing) {
4760
trace_flags |= old_userobj;
4762
for_each_tracing_cpu(cpu) {
4763
atomic_dec(&iter.tr->data[cpu]->disabled);
4769
arch_spin_unlock(&ftrace_dump_lock);
4770
local_irq_restore(flags);
4773
/* By default: disable tracing after the dump */
4774
void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4776
__ftrace_dump(true, oops_dump_mode);
4779
__init static int tracer_alloc_buffers(void)
4782
enum ring_buffer_flags rb_flags;
4787
if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4790
if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4791
goto out_free_buffer_mask;
4793
/* To save memory, keep the ring buffer size to its minimum */
4794
if (ring_buffer_expanded)
4795
ring_buf_size = trace_buf_size;
4799
rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
4801
cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4802
cpumask_copy(tracing_cpumask, cpu_all_mask);
4804
/* TODO: make the number of buffers hot pluggable with CPUS */
4805
global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
4806
if (!global_trace.buffer) {
4807
printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4809
goto out_free_cpumask;
4811
global_trace.entries = ring_buffer_size(global_trace.buffer);
4814
#ifdef CONFIG_TRACER_MAX_TRACE
4815
max_tr.buffer = ring_buffer_alloc(1, rb_flags);
4816
if (!max_tr.buffer) {
4817
printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4819
ring_buffer_free(global_trace.buffer);
4820
goto out_free_cpumask;
4825
/* Allocate the first page for all buffers */
4826
for_each_tracing_cpu(i) {
4827
global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4828
max_tr.data[i] = &per_cpu(max_tr_data, i);
4831
trace_init_cmdlines();
4833
register_tracer(&nop_trace);
4834
current_trace = &nop_trace;
4835
/* All seems OK, enable tracing */
4836
tracing_disabled = 0;
4838
atomic_notifier_chain_register(&panic_notifier_list,
4839
&trace_panic_notifier);
4841
register_die_notifier(&trace_die_notifier);
4846
free_cpumask_var(tracing_cpumask);
4847
out_free_buffer_mask:
4848
free_cpumask_var(tracing_buffer_mask);
4853
__init static int clear_boot_tracer(void)
4856
* The default tracer at boot buffer is an init section.
4857
* This function is called in lateinit. If we did not
4858
* find the boot tracer, then clear it out, to prevent
4859
* later registration from accessing the buffer that is
4860
* about to be freed.
4862
if (!default_bootup_tracer)
4865
printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4866
default_bootup_tracer);
4867
default_bootup_tracer = NULL;
4872
early_initcall(tracer_alloc_buffers);
4873
fs_initcall(tracer_init_debugfs);
4874
late_initcall(clear_boot_tracer);