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/utsrelease.h>
15 #include <linux/kallsyms.h>
16 #include <linux/seq_file.h>
17 #include <linux/debugfs.h>
18 #include <linux/pagemap.h>
19 #include <linux/hardirq.h>
20 #include <linux/linkage.h>
21 #include <linux/uaccess.h>
22 #include <linux/ftrace.h>
23 #include <linux/module.h>
24 #include <linux/percpu.h>
25 #include <linux/ctype.h>
26 #include <linux/init.h>
27 #include <linux/gfp.h>
32 unsigned long __read_mostly tracing_max_latency
= (cycle_t
)ULONG_MAX
;
33 unsigned long __read_mostly tracing_thresh
;
35 static int tracing_disabled
= 1;
38 ns2usecs(cycle_t nsec
)
45 static const int time_sync_freq_max
= 128;
46 static const cycle_t time_sync_thresh
= 100000;
48 static DEFINE_PER_CPU(cycle_t
, time_offset
);
49 static DEFINE_PER_CPU(cycle_t
, prev_cpu_time
);
50 static DEFINE_PER_CPU(int, time_sync_count
);
51 static DEFINE_PER_CPU(int, time_sync_freq
);
54 * Global lock which we take every now and then to synchronize
55 * the CPUs time. This method is not warp-safe, but it's good
56 * enough to synchronize slowly diverging time sources and thus
57 * it's good enough for tracing:
59 static DEFINE_SPINLOCK(time_sync_lock
);
60 static cycle_t prev_global_time
;
62 static notrace cycle_t
__ftrace_now_sync(cycles_t time
, int cpu
)
66 spin_lock_irqsave(&time_sync_lock
, flags
);
69 * Update the synchronization frequency:
71 if (per_cpu(time_sync_freq
, cpu
) < time_sync_freq_max
)
72 per_cpu(time_sync_freq
, cpu
) *= 2;
73 per_cpu(time_sync_count
, cpu
) = per_cpu(time_sync_freq
, cpu
);
75 if (time
< prev_global_time
) {
76 per_cpu(time_offset
, cpu
) += prev_global_time
- time
;
77 time
= prev_global_time
;
79 prev_global_time
= time
;
82 spin_unlock_irqrestore(&time_sync_lock
, flags
);
87 notrace cycle_t
ftrace_now(int cpu
)
89 cycle_t prev_cpu_time
, time
, delta_time
;
91 prev_cpu_time
= per_cpu(prev_cpu_time
, cpu
);
92 time
= sched_clock() + per_cpu(time_offset
, cpu
);
93 delta_time
= time
-prev_cpu_time
;
95 if (unlikely(delta_time
> time_sync_thresh
||
96 --per_cpu(time_sync_count
, cpu
) <= 0))
97 time
= __ftrace_now_sync(time
, cpu
);
102 static struct trace_array global_trace
;
104 static DEFINE_PER_CPU(struct trace_array_cpu
, global_trace_cpu
);
106 static struct trace_array max_tr
;
108 static DEFINE_PER_CPU(struct trace_array_cpu
, max_data
);
110 static int tracer_enabled
= 1;
111 static unsigned long trace_nr_entries
= 16384UL;
113 static struct tracer
*trace_types __read_mostly
;
114 static struct tracer
*current_trace __read_mostly
;
115 static int max_tracer_type_len
;
117 static DEFINE_MUTEX(trace_types_lock
);
119 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
121 static int __init
set_nr_entries(char *str
)
125 trace_nr_entries
= simple_strtoul(str
, &str
, 0);
128 __setup("trace_entries=", set_nr_entries
);
130 unsigned long nsecs_to_usecs(unsigned long nsecs
)
136 __TRACE_FIRST_TYPE
= 0,
145 enum trace_flag_type
{
146 TRACE_FLAG_IRQS_OFF
= 0x01,
147 TRACE_FLAG_NEED_RESCHED
= 0x02,
148 TRACE_FLAG_HARDIRQ
= 0x04,
149 TRACE_FLAG_SOFTIRQ
= 0x08,
152 enum trace_iterator_flags
{
153 TRACE_ITER_PRINT_PARENT
= 0x01,
154 TRACE_ITER_SYM_OFFSET
= 0x02,
155 TRACE_ITER_SYM_ADDR
= 0x04,
156 TRACE_ITER_VERBOSE
= 0x08,
157 TRACE_ITER_RAW
= 0x10,
158 TRACE_ITER_BIN
= 0x20,
161 #define TRACE_ITER_SYM_MASK \
162 (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
164 /* These must match the bit postions above */
165 static const char *trace_options
[] = {
175 static unsigned trace_flags
;
177 static DEFINE_SPINLOCK(ftrace_max_lock
);
180 * Copy the new maximum trace into the separate maximum-trace
181 * structure. (this way the maximum trace is permanently saved,
182 * for later retrieval via /debugfs/tracing/latency_trace)
185 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
187 struct trace_array_cpu
*data
= tr
->data
[cpu
];
190 max_tr
.time_start
= data
->preempt_timestamp
;
192 data
= max_tr
.data
[cpu
];
193 data
->saved_latency
= tracing_max_latency
;
195 memcpy(data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
196 data
->pid
= tsk
->pid
;
197 data
->uid
= tsk
->uid
;
198 data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
199 data
->policy
= tsk
->policy
;
200 data
->rt_priority
= tsk
->rt_priority
;
202 /* record this tasks comm */
203 tracing_record_cmdline(current
);
206 void check_pages(struct trace_array_cpu
*data
)
208 struct page
*page
, *tmp
;
210 BUG_ON(data
->trace_pages
.next
->prev
!= &data
->trace_pages
);
211 BUG_ON(data
->trace_pages
.prev
->next
!= &data
->trace_pages
);
213 list_for_each_entry_safe(page
, tmp
, &data
->trace_pages
, lru
) {
214 BUG_ON(page
->lru
.next
->prev
!= &page
->lru
);
215 BUG_ON(page
->lru
.prev
->next
!= &page
->lru
);
219 void *head_page(struct trace_array_cpu
*data
)
224 if (list_empty(&data
->trace_pages
))
227 page
= list_entry(data
->trace_pages
.next
, struct page
, lru
);
228 BUG_ON(&page
->lru
== &data
->trace_pages
);
230 return page_address(page
);
234 trace_seq_printf(struct trace_seq
*s
, const char *fmt
, ...)
236 int len
= (PAGE_SIZE
- 1) - s
->len
;
244 ret
= vsnprintf(s
->buffer
+ s
->len
, len
, fmt
, ap
);
247 /* If we can't write it all, don't bother writing anything */
257 trace_seq_puts(struct trace_seq
*s
, const char *str
)
259 int len
= strlen(str
);
261 if (len
> ((PAGE_SIZE
- 1) - s
->len
))
264 memcpy(s
->buffer
+ s
->len
, str
, len
);
271 trace_seq_putc(struct trace_seq
*s
, unsigned char c
)
273 if (s
->len
>= (PAGE_SIZE
- 1))
276 s
->buffer
[s
->len
++] = c
;
282 trace_seq_putmem(struct trace_seq
*s
, void *mem
, size_t len
)
284 if (len
> ((PAGE_SIZE
- 1) - s
->len
))
287 memcpy(s
->buffer
+ s
->len
, mem
, len
);
294 trace_seq_reset(struct trace_seq
*s
)
300 trace_print_seq(struct seq_file
*m
, struct trace_seq
*s
)
302 int len
= s
->len
>= PAGE_SIZE
? PAGE_SIZE
- 1 : s
->len
;
305 seq_puts(m
, s
->buffer
);
311 flip_trace(struct trace_array_cpu
*tr1
, struct trace_array_cpu
*tr2
)
313 struct list_head flip_pages
;
315 INIT_LIST_HEAD(&flip_pages
);
317 memcpy(&tr1
->trace_head_idx
, &tr2
->trace_head_idx
,
318 sizeof(struct trace_array_cpu
) -
319 offsetof(struct trace_array_cpu
, trace_head_idx
));
323 list_splice_init(&tr1
->trace_pages
, &flip_pages
);
324 list_splice_init(&tr2
->trace_pages
, &tr1
->trace_pages
);
325 list_splice_init(&flip_pages
, &tr2
->trace_pages
);
326 BUG_ON(!list_empty(&flip_pages
));
332 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
334 struct trace_array_cpu
*data
;
337 WARN_ON_ONCE(!irqs_disabled());
338 spin_lock(&ftrace_max_lock
);
339 /* clear out all the previous traces */
340 for_each_possible_cpu(i
) {
342 flip_trace(max_tr
.data
[i
], data
);
346 __update_max_tr(tr
, tsk
, cpu
);
347 spin_unlock(&ftrace_max_lock
);
351 * update_max_tr_single - only copy one trace over, and reset the rest
353 * @tsk - task with the latency
354 * @cpu - the cpu of the buffer to copy.
357 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
359 struct trace_array_cpu
*data
= tr
->data
[cpu
];
362 WARN_ON_ONCE(!irqs_disabled());
363 spin_lock(&ftrace_max_lock
);
364 for_each_possible_cpu(i
)
365 tracing_reset(max_tr
.data
[i
]);
367 flip_trace(max_tr
.data
[cpu
], data
);
370 __update_max_tr(tr
, tsk
, cpu
);
371 spin_unlock(&ftrace_max_lock
);
374 int register_tracer(struct tracer
*type
)
381 pr_info("Tracer must have a name\n");
385 mutex_lock(&trace_types_lock
);
386 for (t
= trace_types
; t
; t
= t
->next
) {
387 if (strcmp(type
->name
, t
->name
) == 0) {
389 pr_info("Trace %s already registered\n",
396 #ifdef CONFIG_FTRACE_STARTUP_TEST
397 if (type
->selftest
) {
398 struct tracer
*saved_tracer
= current_trace
;
399 struct trace_array_cpu
*data
;
400 struct trace_array
*tr
= &global_trace
;
401 int saved_ctrl
= tr
->ctrl
;
404 * Run a selftest on this tracer.
405 * Here we reset the trace buffer, and set the current
406 * tracer to be this tracer. The tracer can then run some
407 * internal tracing to verify that everything is in order.
408 * If we fail, we do not register this tracer.
410 for_each_possible_cpu(i
) {
412 if (!head_page(data
))
416 current_trace
= type
;
418 /* the test is responsible for initializing and enabling */
419 pr_info("Testing tracer %s: ", type
->name
);
420 ret
= type
->selftest(type
, tr
);
421 /* the test is responsible for resetting too */
422 current_trace
= saved_tracer
;
423 tr
->ctrl
= saved_ctrl
;
425 printk(KERN_CONT
"FAILED!\n");
428 /* Only reset on passing, to avoid touching corrupted buffers */
429 for_each_possible_cpu(i
) {
431 if (!head_page(data
))
435 printk(KERN_CONT
"PASSED\n");
439 type
->next
= trace_types
;
441 len
= strlen(type
->name
);
442 if (len
> max_tracer_type_len
)
443 max_tracer_type_len
= len
;
446 mutex_unlock(&trace_types_lock
);
451 void unregister_tracer(struct tracer
*type
)
456 mutex_lock(&trace_types_lock
);
457 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
461 pr_info("Trace %s not registered\n", type
->name
);
466 if (strlen(type
->name
) != max_tracer_type_len
)
469 max_tracer_type_len
= 0;
470 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
471 len
= strlen((*t
)->name
);
472 if (len
> max_tracer_type_len
)
473 max_tracer_type_len
= len
;
476 mutex_unlock(&trace_types_lock
);
479 notrace
void tracing_reset(struct trace_array_cpu
*data
)
482 data
->trace_head
= data
->trace_tail
= head_page(data
);
483 data
->trace_head_idx
= 0;
484 data
->trace_tail_idx
= 0;
489 function_trace_call(unsigned long ip
, unsigned long parent_ip
)
491 struct trace_array
*tr
= &global_trace
;
492 struct trace_array_cpu
*data
;
497 if (unlikely(!tracer_enabled
))
500 local_irq_save(flags
);
501 cpu
= raw_smp_processor_id();
502 data
= tr
->data
[cpu
];
503 disabled
= atomic_inc_return(&data
->disabled
);
505 if (likely(disabled
== 1))
506 ftrace(tr
, data
, ip
, parent_ip
, flags
);
508 atomic_dec(&data
->disabled
);
509 local_irq_restore(flags
);
512 static struct ftrace_ops trace_ops __read_mostly
=
514 .func
= function_trace_call
,
518 notrace
void tracing_start_function_trace(void)
520 register_ftrace_function(&trace_ops
);
523 notrace
void tracing_stop_function_trace(void)
525 unregister_ftrace_function(&trace_ops
);
528 #define SAVED_CMDLINES 128
529 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
530 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
531 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
532 static int cmdline_idx
;
533 static DEFINE_SPINLOCK(trace_cmdline_lock
);
534 atomic_t trace_record_cmdline_disabled
;
536 static void trace_init_cmdlines(void)
538 memset(&map_pid_to_cmdline
, -1, sizeof(map_pid_to_cmdline
));
539 memset(&map_cmdline_to_pid
, -1, sizeof(map_cmdline_to_pid
));
543 notrace
void trace_stop_cmdline_recording(void);
545 static notrace
void trace_save_cmdline(struct task_struct
*tsk
)
550 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
554 * It's not the end of the world if we don't get
555 * the lock, but we also don't want to spin
556 * nor do we want to disable interrupts,
557 * so if we miss here, then better luck next time.
559 if (!spin_trylock(&trace_cmdline_lock
))
562 idx
= map_pid_to_cmdline
[tsk
->pid
];
563 if (idx
>= SAVED_CMDLINES
) {
564 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
566 map
= map_cmdline_to_pid
[idx
];
567 if (map
<= PID_MAX_DEFAULT
)
568 map_pid_to_cmdline
[map
] = (unsigned)-1;
570 map_pid_to_cmdline
[tsk
->pid
] = idx
;
575 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
577 spin_unlock(&trace_cmdline_lock
);
580 static notrace
char *trace_find_cmdline(int pid
)
582 char *cmdline
= "<...>";
588 if (pid
> PID_MAX_DEFAULT
)
591 map
= map_pid_to_cmdline
[pid
];
592 if (map
>= SAVED_CMDLINES
)
595 cmdline
= saved_cmdlines
[map
];
601 notrace
void tracing_record_cmdline(struct task_struct
*tsk
)
603 if (atomic_read(&trace_record_cmdline_disabled
))
606 trace_save_cmdline(tsk
);
609 static inline notrace
struct list_head
*
610 trace_next_list(struct trace_array_cpu
*data
, struct list_head
*next
)
613 * Roundrobin - but skip the head (which is not a real page):
616 if (unlikely(next
== &data
->trace_pages
))
618 BUG_ON(next
== &data
->trace_pages
);
623 static inline notrace
void *
624 trace_next_page(struct trace_array_cpu
*data
, void *addr
)
626 struct list_head
*next
;
629 page
= virt_to_page(addr
);
631 next
= trace_next_list(data
, &page
->lru
);
632 page
= list_entry(next
, struct page
, lru
);
634 return page_address(page
);
637 static inline notrace
struct trace_entry
*
638 tracing_get_trace_entry(struct trace_array
*tr
, struct trace_array_cpu
*data
)
640 unsigned long idx
, idx_next
;
641 struct trace_entry
*entry
;
644 idx
= data
->trace_head_idx
;
647 BUG_ON(idx
* TRACE_ENTRY_SIZE
>= PAGE_SIZE
);
649 entry
= data
->trace_head
+ idx
* TRACE_ENTRY_SIZE
;
651 if (unlikely(idx_next
>= ENTRIES_PER_PAGE
)) {
652 data
->trace_head
= trace_next_page(data
, data
->trace_head
);
656 if (data
->trace_head
== data
->trace_tail
&&
657 idx_next
== data
->trace_tail_idx
) {
659 data
->trace_tail_idx
++;
660 if (data
->trace_tail_idx
>= ENTRIES_PER_PAGE
) {
662 trace_next_page(data
, data
->trace_tail
);
663 data
->trace_tail_idx
= 0;
667 data
->trace_head_idx
= idx_next
;
672 static inline notrace
void
673 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
)
675 struct task_struct
*tsk
= current
;
678 pc
= preempt_count();
680 entry
->preempt_count
= pc
& 0xff;
681 entry
->pid
= tsk
->pid
;
682 entry
->t
= ftrace_now(raw_smp_processor_id());
683 entry
->flags
= (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
684 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
685 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
686 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
690 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
691 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
)
693 struct trace_entry
*entry
;
694 unsigned long irq_flags
;
696 spin_lock_irqsave(&data
->lock
, irq_flags
);
697 entry
= tracing_get_trace_entry(tr
, data
);
698 tracing_generic_entry_update(entry
, flags
);
699 entry
->type
= TRACE_FN
;
701 entry
->fn
.parent_ip
= parent_ip
;
702 spin_unlock_irqrestore(&data
->lock
, irq_flags
);
706 trace_special(struct trace_array
*tr
, struct trace_array_cpu
*data
,
707 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
709 struct trace_entry
*entry
;
710 unsigned long irq_flags
;
712 spin_lock_irqsave(&data
->lock
, irq_flags
);
713 entry
= tracing_get_trace_entry(tr
, data
);
714 tracing_generic_entry_update(entry
, 0);
715 entry
->type
= TRACE_SPECIAL
;
716 entry
->special
.arg1
= arg1
;
717 entry
->special
.arg2
= arg2
;
718 entry
->special
.arg3
= arg3
;
719 spin_unlock_irqrestore(&data
->lock
, irq_flags
);
723 tracing_sched_switch_trace(struct trace_array
*tr
,
724 struct trace_array_cpu
*data
,
725 struct task_struct
*prev
, struct task_struct
*next
,
728 struct trace_entry
*entry
;
729 unsigned long irq_flags
;
731 spin_lock_irqsave(&data
->lock
, irq_flags
);
732 entry
= tracing_get_trace_entry(tr
, data
);
733 tracing_generic_entry_update(entry
, flags
);
734 entry
->type
= TRACE_CTX
;
735 entry
->ctx
.prev_pid
= prev
->pid
;
736 entry
->ctx
.prev_prio
= prev
->prio
;
737 entry
->ctx
.prev_state
= prev
->state
;
738 entry
->ctx
.next_pid
= next
->pid
;
739 entry
->ctx
.next_prio
= next
->prio
;
740 spin_unlock_irqrestore(&data
->lock
, irq_flags
);
743 enum trace_file_type
{
744 TRACE_FILE_LAT_FMT
= 1,
747 static struct trace_entry
*
748 trace_entry_idx(struct trace_array
*tr
, struct trace_array_cpu
*data
,
749 struct trace_iterator
*iter
, int cpu
)
752 struct trace_entry
*array
;
754 if (iter
->next_idx
[cpu
] >= tr
->entries
||
755 iter
->next_idx
[cpu
] >= data
->trace_idx
||
756 (data
->trace_head
== data
->trace_tail
&&
757 data
->trace_head_idx
== data
->trace_tail_idx
))
760 if (!iter
->next_page
[cpu
]) {
761 /* Initialize the iterator for this cpu trace buffer */
762 WARN_ON(!data
->trace_tail
);
763 page
= virt_to_page(data
->trace_tail
);
764 iter
->next_page
[cpu
] = &page
->lru
;
765 iter
->next_page_idx
[cpu
] = data
->trace_tail_idx
;
768 page
= list_entry(iter
->next_page
[cpu
], struct page
, lru
);
769 BUG_ON(&data
->trace_pages
== &page
->lru
);
771 array
= page_address(page
);
773 WARN_ON(iter
->next_page_idx
[cpu
] >= ENTRIES_PER_PAGE
);
774 return &array
[iter
->next_page_idx
[cpu
]];
777 static struct notrace trace_entry
*
778 find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
)
780 struct trace_array
*tr
= iter
->tr
;
781 struct trace_entry
*ent
, *next
= NULL
;
785 for_each_possible_cpu(cpu
) {
786 if (!head_page(tr
->data
[cpu
]))
788 ent
= trace_entry_idx(tr
, tr
->data
[cpu
], iter
, cpu
);
790 * Pick the entry with the smallest timestamp:
792 if (ent
&& (!next
|| ent
->t
< next
->t
)) {
804 static notrace
void trace_iterator_increment(struct trace_iterator
*iter
)
807 iter
->next_idx
[iter
->cpu
]++;
808 iter
->next_page_idx
[iter
->cpu
]++;
810 if (iter
->next_page_idx
[iter
->cpu
] >= ENTRIES_PER_PAGE
) {
811 struct trace_array_cpu
*data
= iter
->tr
->data
[iter
->cpu
];
813 iter
->next_page_idx
[iter
->cpu
] = 0;
814 iter
->next_page
[iter
->cpu
] =
815 trace_next_list(data
, iter
->next_page
[iter
->cpu
]);
819 static notrace
void trace_consume(struct trace_iterator
*iter
)
821 struct trace_array_cpu
*data
= iter
->tr
->data
[iter
->cpu
];
823 data
->trace_tail_idx
++;
824 if (data
->trace_tail_idx
>= ENTRIES_PER_PAGE
) {
825 data
->trace_tail
= trace_next_page(data
, data
->trace_tail
);
826 data
->trace_tail_idx
= 0;
829 /* Check if we empty it, then reset the index */
830 if (data
->trace_head
== data
->trace_tail
&&
831 data
->trace_head_idx
== data
->trace_tail_idx
)
835 static notrace
void *find_next_entry_inc(struct trace_iterator
*iter
)
837 struct trace_entry
*next
;
840 next
= find_next_entry(iter
, &next_cpu
);
842 iter
->prev_ent
= iter
->ent
;
843 iter
->prev_cpu
= iter
->cpu
;
846 iter
->cpu
= next_cpu
;
849 trace_iterator_increment(iter
);
851 return next
? iter
: NULL
;
854 static notrace
void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
856 struct trace_iterator
*iter
= m
->private;
857 void *last_ent
= iter
->ent
;
863 /* can't go backwards */
868 ent
= find_next_entry_inc(iter
);
872 while (ent
&& iter
->idx
< i
)
873 ent
= find_next_entry_inc(iter
);
877 if (last_ent
&& !ent
)
878 seq_puts(m
, "\n\nvim:ft=help\n");
883 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
885 struct trace_iterator
*iter
= m
->private;
890 mutex_lock(&trace_types_lock
);
892 if (!current_trace
|| current_trace
!= iter
->trace
)
895 atomic_inc(&trace_record_cmdline_disabled
);
897 /* let the tracer grab locks here if needed */
898 if (current_trace
->start
)
899 current_trace
->start(iter
);
901 if (*pos
!= iter
->pos
) {
905 iter
->prev_ent
= NULL
;
908 for_each_possible_cpu(i
) {
909 iter
->next_idx
[i
] = 0;
910 iter
->next_page
[i
] = NULL
;
913 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
918 p
= s_next(m
, p
, &l
);
924 static void s_stop(struct seq_file
*m
, void *p
)
926 struct trace_iterator
*iter
= m
->private;
928 atomic_dec(&trace_record_cmdline_disabled
);
930 /* let the tracer release locks here if needed */
931 if (current_trace
&& current_trace
== iter
->trace
&& iter
->trace
->stop
)
932 iter
->trace
->stop(iter
);
934 mutex_unlock(&trace_types_lock
);
938 seq_print_sym_short(struct trace_seq
*s
, const char *fmt
, unsigned long address
)
940 #ifdef CONFIG_KALLSYMS
941 char str
[KSYM_SYMBOL_LEN
];
943 kallsyms_lookup(address
, NULL
, NULL
, NULL
, str
);
945 return trace_seq_printf(s
, fmt
, str
);
951 seq_print_sym_offset(struct trace_seq
*s
, const char *fmt
,
952 unsigned long address
)
954 #ifdef CONFIG_KALLSYMS
955 char str
[KSYM_SYMBOL_LEN
];
957 sprint_symbol(str
, address
);
958 return trace_seq_printf(s
, fmt
, str
);
964 # define IP_FMT "%08lx"
966 # define IP_FMT "%016lx"
970 seq_print_ip_sym(struct trace_seq
*s
, unsigned long ip
, unsigned long sym_flags
)
975 return trace_seq_printf(s
, "0");
977 if (sym_flags
& TRACE_ITER_SYM_OFFSET
)
978 ret
= seq_print_sym_offset(s
, "%s", ip
);
980 ret
= seq_print_sym_short(s
, "%s", ip
);
985 if (sym_flags
& TRACE_ITER_SYM_ADDR
)
986 ret
= trace_seq_printf(s
, " <" IP_FMT
">", ip
);
990 static notrace
void print_lat_help_header(struct seq_file
*m
)
992 seq_puts(m
, "# _------=> CPU# \n");
993 seq_puts(m
, "# / _-----=> irqs-off \n");
994 seq_puts(m
, "# | / _----=> need-resched \n");
995 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
996 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
997 seq_puts(m
, "# |||| / \n");
998 seq_puts(m
, "# ||||| delay \n");
999 seq_puts(m
, "# cmd pid ||||| time | caller \n");
1000 seq_puts(m
, "# \\ / ||||| \\ | / \n");
1003 static notrace
void print_func_help_header(struct seq_file
*m
)
1005 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1006 seq_puts(m
, "# | | | | |\n");
1011 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1013 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1014 struct trace_array
*tr
= iter
->tr
;
1015 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1016 struct tracer
*type
= current_trace
;
1017 unsigned long total
= 0;
1018 unsigned long entries
= 0;
1020 const char *name
= "preemption";
1025 for_each_possible_cpu(cpu
) {
1026 if (head_page(tr
->data
[cpu
])) {
1027 total
+= tr
->data
[cpu
]->trace_idx
;
1028 if (tr
->data
[cpu
]->trace_idx
> tr
->entries
)
1029 entries
+= tr
->entries
;
1031 entries
+= tr
->data
[cpu
]->trace_idx
;
1035 seq_printf(m
, "%s latency trace v1.1.5 on %s\n",
1037 seq_puts(m
, "-----------------------------------"
1038 "---------------------------------\n");
1039 seq_printf(m
, " latency: %lu us, #%lu/%lu, CPU#%d |"
1040 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1041 nsecs_to_usecs(data
->saved_latency
),
1045 #if defined(CONFIG_PREEMPT_NONE)
1047 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1049 #elif defined(CONFIG_PREEMPT_DESKTOP)
1054 /* These are reserved for later use */
1057 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1061 seq_puts(m
, " -----------------\n");
1062 seq_printf(m
, " | task: %.16s-%d "
1063 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1064 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1065 data
->policy
, data
->rt_priority
);
1066 seq_puts(m
, " -----------------\n");
1068 if (data
->critical_start
) {
1069 seq_puts(m
, " => started at: ");
1070 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1071 trace_print_seq(m
, &iter
->seq
);
1072 seq_puts(m
, "\n => ended at: ");
1073 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1074 trace_print_seq(m
, &iter
->seq
);
1082 lat_print_generic(struct trace_seq
*s
, struct trace_entry
*entry
, int cpu
)
1084 int hardirq
, softirq
;
1087 comm
= trace_find_cmdline(entry
->pid
);
1089 trace_seq_printf(s
, "%8.8s-%-5d ", comm
, entry
->pid
);
1090 trace_seq_printf(s
, "%d", cpu
);
1091 trace_seq_printf(s
, "%c%c",
1092 (entry
->flags
& TRACE_FLAG_IRQS_OFF
) ? 'd' : '.',
1093 ((entry
->flags
& TRACE_FLAG_NEED_RESCHED
) ? 'N' : '.'));
1095 hardirq
= entry
->flags
& TRACE_FLAG_HARDIRQ
;
1096 softirq
= entry
->flags
& TRACE_FLAG_SOFTIRQ
;
1097 if (hardirq
&& softirq
)
1098 trace_seq_putc(s
, 'H');
1101 trace_seq_putc(s
, 'h');
1104 trace_seq_putc(s
, 's');
1106 trace_seq_putc(s
, '.');
1110 if (entry
->preempt_count
)
1111 trace_seq_printf(s
, "%x", entry
->preempt_count
);
1113 trace_seq_puts(s
, ".");
1116 unsigned long preempt_mark_thresh
= 100;
1119 lat_print_timestamp(struct trace_seq
*s
, unsigned long long abs_usecs
,
1120 unsigned long rel_usecs
)
1122 trace_seq_printf(s
, " %4lldus", abs_usecs
);
1123 if (rel_usecs
> preempt_mark_thresh
)
1124 trace_seq_puts(s
, "!: ");
1125 else if (rel_usecs
> 1)
1126 trace_seq_puts(s
, "+: ");
1128 trace_seq_puts(s
, " : ");
1131 static const char state_to_char
[] = TASK_STATE_TO_CHAR_STR
;
1134 print_lat_fmt(struct trace_iterator
*iter
, unsigned int trace_idx
, int cpu
)
1136 struct trace_seq
*s
= &iter
->seq
;
1137 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1138 struct trace_entry
*next_entry
= find_next_entry(iter
, NULL
);
1139 unsigned long verbose
= (trace_flags
& TRACE_ITER_VERBOSE
);
1140 struct trace_entry
*entry
= iter
->ent
;
1141 unsigned long abs_usecs
;
1142 unsigned long rel_usecs
;
1148 rel_usecs
= ns2usecs(next_entry
->t
- entry
->t
);
1149 abs_usecs
= ns2usecs(entry
->t
- iter
->tr
->time_start
);
1152 comm
= trace_find_cmdline(entry
->pid
);
1153 trace_seq_printf(s
, "%16s %5d %d %d %08x %08x [%08lx]"
1154 " %ld.%03ldms (+%ld.%03ldms): ",
1156 entry
->pid
, cpu
, entry
->flags
,
1157 entry
->preempt_count
, trace_idx
,
1160 abs_usecs
% 1000, rel_usecs
/1000,
1163 lat_print_generic(s
, entry
, cpu
);
1164 lat_print_timestamp(s
, abs_usecs
, rel_usecs
);
1166 switch (entry
->type
) {
1168 seq_print_ip_sym(s
, entry
->fn
.ip
, sym_flags
);
1169 trace_seq_puts(s
, " (");
1170 seq_print_ip_sym(s
, entry
->fn
.parent_ip
, sym_flags
);
1171 trace_seq_puts(s
, ")\n");
1174 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1175 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1176 comm
= trace_find_cmdline(entry
->ctx
.next_pid
);
1177 trace_seq_printf(s
, " %d:%d:%c --> %d:%d %s\n",
1178 entry
->ctx
.prev_pid
,
1179 entry
->ctx
.prev_prio
,
1181 entry
->ctx
.next_pid
,
1182 entry
->ctx
.next_prio
,
1186 trace_seq_printf(s
, " %lx %lx %lx\n",
1187 entry
->special
.arg1
,
1188 entry
->special
.arg2
,
1189 entry
->special
.arg3
);
1192 trace_seq_printf(s
, "Unknown type %d\n", entry
->type
);
1197 static notrace
int print_trace_fmt(struct trace_iterator
*iter
)
1199 struct trace_seq
*s
= &iter
->seq
;
1200 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1201 struct trace_entry
*entry
;
1202 unsigned long usec_rem
;
1203 unsigned long long t
;
1211 comm
= trace_find_cmdline(iter
->ent
->pid
);
1213 t
= ns2usecs(entry
->t
);
1214 usec_rem
= do_div(t
, 1000000ULL);
1215 secs
= (unsigned long)t
;
1217 ret
= trace_seq_printf(s
, "%16s-%-5d ", comm
, entry
->pid
);
1220 ret
= trace_seq_printf(s
, "[%02d] ", iter
->cpu
);
1223 ret
= trace_seq_printf(s
, "%5lu.%06lu: ", secs
, usec_rem
);
1227 switch (entry
->type
) {
1229 ret
= seq_print_ip_sym(s
, entry
->fn
.ip
, sym_flags
);
1232 if ((sym_flags
& TRACE_ITER_PRINT_PARENT
) &&
1233 entry
->fn
.parent_ip
) {
1234 ret
= trace_seq_printf(s
, " <-");
1237 ret
= seq_print_ip_sym(s
, entry
->fn
.parent_ip
,
1242 ret
= trace_seq_printf(s
, "\n");
1247 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1248 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1249 ret
= trace_seq_printf(s
, " %d:%d:%c ==> %d:%d\n",
1250 entry
->ctx
.prev_pid
,
1251 entry
->ctx
.prev_prio
,
1253 entry
->ctx
.next_pid
,
1254 entry
->ctx
.next_prio
);
1259 ret
= trace_seq_printf(s
, " %lx %lx %lx\n",
1260 entry
->special
.arg1
,
1261 entry
->special
.arg2
,
1262 entry
->special
.arg3
);
1270 static notrace
int print_raw_fmt(struct trace_iterator
*iter
)
1272 struct trace_seq
*s
= &iter
->seq
;
1273 struct trace_entry
*entry
;
1279 ret
= trace_seq_printf(s
, "%d %d %llu ",
1280 entry
->pid
, iter
->cpu
, entry
->t
);
1284 switch (entry
->type
) {
1286 ret
= trace_seq_printf(s
, "%x %x\n",
1287 entry
->fn
.ip
, entry
->fn
.parent_ip
);
1292 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1293 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1294 ret
= trace_seq_printf(s
, "%d %d %c %d %d\n",
1295 entry
->ctx
.prev_pid
,
1296 entry
->ctx
.prev_prio
,
1298 entry
->ctx
.next_pid
,
1299 entry
->ctx
.next_prio
);
1304 ret
= trace_seq_printf(s
, " %lx %lx %lx\n",
1305 entry
->special
.arg1
,
1306 entry
->special
.arg2
,
1307 entry
->special
.arg3
);
1315 #define SEQ_PUT_FIELD_RET(s, x) \
1317 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1321 static notrace
int print_bin_fmt(struct trace_iterator
*iter
)
1323 struct trace_seq
*s
= &iter
->seq
;
1324 struct trace_entry
*entry
;
1328 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1329 SEQ_PUT_FIELD_RET(s
, entry
->cpu
);
1330 SEQ_PUT_FIELD_RET(s
, entry
->t
);
1332 switch (entry
->type
) {
1334 SEQ_PUT_FIELD_RET(s
, entry
->fn
.ip
);
1335 SEQ_PUT_FIELD_RET(s
, entry
->fn
.parent_ip
);
1338 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_pid
);
1339 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_prio
);
1340 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_state
);
1341 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_pid
);
1342 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_prio
);
1345 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg1
);
1346 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg2
);
1347 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg3
);
1353 static int trace_empty(struct trace_iterator
*iter
)
1355 struct trace_array_cpu
*data
;
1358 for_each_possible_cpu(cpu
) {
1359 data
= iter
->tr
->data
[cpu
];
1361 if (head_page(data
) && data
->trace_idx
&&
1362 (data
->trace_tail
!= data
->trace_head
||
1363 data
->trace_tail_idx
!= data
->trace_head_idx
))
1369 static int print_trace_line(struct trace_iterator
*iter
)
1371 if (trace_flags
& TRACE_ITER_BIN
)
1372 return print_bin_fmt(iter
);
1374 if (trace_flags
& TRACE_ITER_RAW
)
1375 return print_raw_fmt(iter
);
1377 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
1378 return print_lat_fmt(iter
, iter
->idx
, iter
->cpu
);
1380 return print_trace_fmt(iter
);
1383 static int s_show(struct seq_file
*m
, void *v
)
1385 struct trace_iterator
*iter
= v
;
1387 if (iter
->ent
== NULL
) {
1389 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
1392 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1393 /* print nothing if the buffers are empty */
1394 if (trace_empty(iter
))
1396 print_trace_header(m
, iter
);
1397 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1398 print_lat_help_header(m
);
1400 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1401 print_func_help_header(m
);
1404 print_trace_line(iter
);
1405 trace_print_seq(m
, &iter
->seq
);
1411 static struct seq_operations tracer_seq_ops
= {
1418 static struct trace_iterator notrace
*
1419 __tracing_open(struct inode
*inode
, struct file
*file
, int *ret
)
1421 struct trace_iterator
*iter
;
1423 if (tracing_disabled
) {
1428 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
1434 mutex_lock(&trace_types_lock
);
1435 if (current_trace
&& current_trace
->print_max
)
1438 iter
->tr
= inode
->i_private
;
1439 iter
->trace
= current_trace
;
1442 /* TODO stop tracer */
1443 *ret
= seq_open(file
, &tracer_seq_ops
);
1445 struct seq_file
*m
= file
->private_data
;
1448 /* stop the trace while dumping */
1452 if (iter
->trace
&& iter
->trace
->open
)
1453 iter
->trace
->open(iter
);
1458 mutex_unlock(&trace_types_lock
);
1464 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
1466 if (tracing_disabled
)
1469 filp
->private_data
= inode
->i_private
;
1473 int tracing_release(struct inode
*inode
, struct file
*file
)
1475 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
1476 struct trace_iterator
*iter
= m
->private;
1478 mutex_lock(&trace_types_lock
);
1479 if (iter
->trace
&& iter
->trace
->close
)
1480 iter
->trace
->close(iter
);
1482 /* reenable tracing if it was previously enabled */
1485 mutex_unlock(&trace_types_lock
);
1487 seq_release(inode
, file
);
1492 static int tracing_open(struct inode
*inode
, struct file
*file
)
1496 __tracing_open(inode
, file
, &ret
);
1501 static int tracing_lt_open(struct inode
*inode
, struct file
*file
)
1503 struct trace_iterator
*iter
;
1506 iter
= __tracing_open(inode
, file
, &ret
);
1509 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
1515 static notrace
void *
1516 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1518 struct tracer
*t
= m
->private;
1530 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
1532 struct tracer
*t
= m
->private;
1535 mutex_lock(&trace_types_lock
);
1536 for (; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
1542 static void t_stop(struct seq_file
*m
, void *p
)
1544 mutex_unlock(&trace_types_lock
);
1547 static int t_show(struct seq_file
*m
, void *v
)
1549 struct tracer
*t
= v
;
1554 seq_printf(m
, "%s", t
->name
);
1563 static struct seq_operations show_traces_seq_ops
= {
1570 static int show_traces_open(struct inode
*inode
, struct file
*file
)
1574 if (tracing_disabled
)
1577 ret
= seq_open(file
, &show_traces_seq_ops
);
1579 struct seq_file
*m
= file
->private_data
;
1580 m
->private = trace_types
;
1586 static struct file_operations tracing_fops
= {
1587 .open
= tracing_open
,
1589 .llseek
= seq_lseek
,
1590 .release
= tracing_release
,
1593 static struct file_operations tracing_lt_fops
= {
1594 .open
= tracing_lt_open
,
1596 .llseek
= seq_lseek
,
1597 .release
= tracing_release
,
1600 static struct file_operations show_traces_fops
= {
1601 .open
= show_traces_open
,
1603 .release
= seq_release
,
1607 tracing_iter_ctrl_read(struct file
*filp
, char __user
*ubuf
,
1608 size_t cnt
, loff_t
*ppos
)
1615 /* calulate max size */
1616 for (i
= 0; trace_options
[i
]; i
++) {
1617 len
+= strlen(trace_options
[i
]);
1618 len
+= 3; /* "no" and space */
1621 /* +2 for \n and \0 */
1622 buf
= kmalloc(len
+ 2, GFP_KERNEL
);
1626 for (i
= 0; trace_options
[i
]; i
++) {
1627 if (trace_flags
& (1 << i
))
1628 r
+= sprintf(buf
+ r
, "%s ", trace_options
[i
]);
1630 r
+= sprintf(buf
+ r
, "no%s ", trace_options
[i
]);
1633 r
+= sprintf(buf
+ r
, "\n");
1634 WARN_ON(r
>= len
+ 2);
1636 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
,
1645 tracing_iter_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
1646 size_t cnt
, loff_t
*ppos
)
1656 if (copy_from_user(&buf
, ubuf
, cnt
))
1661 if (strncmp(buf
, "no", 2) == 0) {
1666 for (i
= 0; trace_options
[i
]; i
++) {
1667 int len
= strlen(trace_options
[i
]);
1669 if (strncmp(cmp
, trace_options
[i
], len
) == 0) {
1671 trace_flags
&= ~(1 << i
);
1673 trace_flags
|= (1 << i
);
1683 static struct file_operations tracing_iter_fops
= {
1684 .open
= tracing_open_generic
,
1685 .read
= tracing_iter_ctrl_read
,
1686 .write
= tracing_iter_ctrl_write
,
1689 static const char readme_msg
[] =
1690 "tracing mini-HOWTO:\n\n"
1692 "# mount -t debugfs nodev /debug\n\n"
1693 "# cat /debug/tracing/available_tracers\n"
1694 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1695 "# cat /debug/tracing/current_tracer\n"
1697 "# echo sched_switch > /debug/tracing/current_tracer\n"
1698 "# cat /debug/tracing/current_tracer\n"
1700 "# cat /debug/tracing/iter_ctrl\n"
1701 "noprint-parent nosym-offset nosym-addr noverbose\n"
1702 "# echo print-parent > /debug/tracing/iter_ctrl\n"
1703 "# echo 1 > /debug/tracing/tracing_enabled\n"
1704 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1705 "echo 0 > /debug/tracing/tracing_enabled\n"
1709 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
1710 size_t cnt
, loff_t
*ppos
)
1712 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
1713 readme_msg
, strlen(readme_msg
));
1716 static struct file_operations tracing_readme_fops
= {
1717 .open
= tracing_open_generic
,
1718 .read
= tracing_readme_read
,
1723 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
1724 size_t cnt
, loff_t
*ppos
)
1726 struct trace_array
*tr
= filp
->private_data
;
1730 r
= sprintf(buf
, "%ld\n", tr
->ctrl
);
1731 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
1735 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
1736 size_t cnt
, loff_t
*ppos
)
1738 struct trace_array
*tr
= filp
->private_data
;
1745 if (copy_from_user(&buf
, ubuf
, cnt
))
1750 val
= simple_strtoul(buf
, NULL
, 10);
1754 mutex_lock(&trace_types_lock
);
1755 if (tr
->ctrl
^ val
) {
1763 if (current_trace
&& current_trace
->ctrl_update
)
1764 current_trace
->ctrl_update(tr
);
1766 mutex_unlock(&trace_types_lock
);
1774 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
1775 size_t cnt
, loff_t
*ppos
)
1777 char buf
[max_tracer_type_len
+2];
1780 mutex_lock(&trace_types_lock
);
1782 r
= sprintf(buf
, "%s\n", current_trace
->name
);
1784 r
= sprintf(buf
, "\n");
1785 mutex_unlock(&trace_types_lock
);
1787 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
1791 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
1792 size_t cnt
, loff_t
*ppos
)
1794 struct trace_array
*tr
= &global_trace
;
1796 char buf
[max_tracer_type_len
+1];
1799 if (cnt
> max_tracer_type_len
)
1800 cnt
= max_tracer_type_len
;
1802 if (copy_from_user(&buf
, ubuf
, cnt
))
1807 /* strip ending whitespace. */
1808 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
1811 mutex_lock(&trace_types_lock
);
1812 for (t
= trace_types
; t
; t
= t
->next
) {
1813 if (strcmp(t
->name
, buf
) == 0)
1816 if (!t
|| t
== current_trace
)
1819 if (current_trace
&& current_trace
->reset
)
1820 current_trace
->reset(tr
);
1827 mutex_unlock(&trace_types_lock
);
1835 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
1836 size_t cnt
, loff_t
*ppos
)
1838 unsigned long *ptr
= filp
->private_data
;
1842 r
= snprintf(buf
, 64, "%ld\n",
1843 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
1846 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
1850 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
1851 size_t cnt
, loff_t
*ppos
)
1853 long *ptr
= filp
->private_data
;
1860 if (copy_from_user(&buf
, ubuf
, cnt
))
1865 val
= simple_strtoul(buf
, NULL
, 10);
1872 static atomic_t tracing_reader
;
1874 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
1876 struct trace_iterator
*iter
;
1878 if (tracing_disabled
)
1881 /* We only allow for reader of the pipe */
1882 if (atomic_inc_return(&tracing_reader
) != 1) {
1883 atomic_dec(&tracing_reader
);
1887 /* create a buffer to store the information to pass to userspace */
1888 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
1892 iter
->tr
= &global_trace
;
1894 filp
->private_data
= iter
;
1899 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
1901 struct trace_iterator
*iter
= file
->private_data
;
1904 atomic_dec(&tracing_reader
);
1913 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
1914 size_t cnt
, loff_t
*ppos
)
1916 struct trace_iterator
*iter
= filp
->private_data
;
1917 struct trace_array_cpu
*data
;
1918 static cpumask_t mask
;
1920 unsigned long flags
;
1926 /* return any leftover data */
1927 if (iter
->seq
.len
> start
) {
1928 len
= iter
->seq
.len
- start
;
1931 ret
= copy_to_user(ubuf
, iter
->seq
.buffer
+ start
, cnt
);
1940 trace_seq_reset(&iter
->seq
);
1943 while (trace_empty(iter
)) {
1945 * This is a make-shift waitqueue. The reason we don't use
1946 * an actual wait queue is because:
1947 * 1) we only ever have one waiter
1948 * 2) the tracing, traces all functions, we don't want
1949 * the overhead of calling wake_up and friends
1950 * (and tracing them too)
1951 * Anyway, this is really very primitive wakeup.
1953 set_current_state(TASK_INTERRUPTIBLE
);
1954 iter
->tr
->waiter
= current
;
1956 /* sleep for one second, and try again. */
1957 schedule_timeout(HZ
);
1959 iter
->tr
->waiter
= NULL
;
1961 if (signal_pending(current
))
1965 * We block until we read something and tracing is disabled.
1966 * We still block if tracing is disabled, but we have never
1967 * read anything. This allows a user to cat this file, and
1968 * then enable tracing. But after we have read something,
1969 * we give an EOF when tracing is again disabled.
1971 * iter->pos will be 0 if we haven't read anything.
1973 if (!tracer_enabled
&& iter
->pos
)
1979 /* stop when tracing is finished */
1980 if (trace_empty(iter
))
1983 if (cnt
>= PAGE_SIZE
)
1984 cnt
= PAGE_SIZE
- 1;
1986 memset(iter
, 0, sizeof(*iter
));
1987 iter
->tr
= &global_trace
;
1991 * We need to stop all tracing on all CPUS to read the
1992 * the next buffer. This is a bit expensive, but is
1993 * not done often. We fill all what we can read,
1994 * and then release the locks again.
1998 local_irq_save(flags
);
1999 for_each_possible_cpu(cpu
) {
2000 data
= iter
->tr
->data
[cpu
];
2002 if (!head_page(data
) || !data
->trace_idx
)
2005 atomic_inc(&data
->disabled
);
2006 spin_lock(&data
->lock
);
2010 while (find_next_entry_inc(iter
) != NULL
) {
2011 int len
= iter
->seq
.len
;
2013 ret
= print_trace_line(iter
);
2015 /* don't print partial lines */
2016 iter
->seq
.len
= len
;
2020 trace_consume(iter
);
2022 if (iter
->seq
.len
>= cnt
)
2026 for_each_cpu_mask(cpu
, mask
) {
2027 data
= iter
->tr
->data
[cpu
];
2028 spin_unlock(&data
->lock
);
2029 atomic_dec(&data
->disabled
);
2031 local_irq_restore(flags
);
2033 /* Now copy what we have to the user */
2034 read
= iter
->seq
.len
;
2038 ret
= copy_to_user(ubuf
, iter
->seq
.buffer
, read
);
2040 if (read
< iter
->seq
.len
)
2043 trace_seq_reset(&iter
->seq
);
2051 static struct file_operations tracing_max_lat_fops
= {
2052 .open
= tracing_open_generic
,
2053 .read
= tracing_max_lat_read
,
2054 .write
= tracing_max_lat_write
,
2057 static struct file_operations tracing_ctrl_fops
= {
2058 .open
= tracing_open_generic
,
2059 .read
= tracing_ctrl_read
,
2060 .write
= tracing_ctrl_write
,
2063 static struct file_operations set_tracer_fops
= {
2064 .open
= tracing_open_generic
,
2065 .read
= tracing_set_trace_read
,
2066 .write
= tracing_set_trace_write
,
2069 static struct file_operations tracing_pipe_fops
= {
2070 .open
= tracing_open_pipe
,
2071 .read
= tracing_read_pipe
,
2072 .release
= tracing_release_pipe
,
2075 #ifdef CONFIG_DYNAMIC_FTRACE
2078 tracing_read_long(struct file
*filp
, char __user
*ubuf
,
2079 size_t cnt
, loff_t
*ppos
)
2081 unsigned long *p
= filp
->private_data
;
2085 r
= sprintf(buf
, "%ld\n", *p
);
2087 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2090 static struct file_operations tracing_read_long_fops
= {
2091 .open
= tracing_open_generic
,
2092 .read
= tracing_read_long
,
2096 static struct dentry
*d_tracer
;
2098 struct dentry
*tracing_init_dentry(void)
2105 d_tracer
= debugfs_create_dir("tracing", NULL
);
2107 if (!d_tracer
&& !once
) {
2109 pr_warning("Could not create debugfs directory 'tracing'\n");
2116 #ifdef CONFIG_FTRACE_SELFTEST
2117 /* Let selftest have access to static functions in this file */
2118 #include "trace_selftest.c"
2121 static __init
void tracer_init_debugfs(void)
2123 struct dentry
*d_tracer
;
2124 struct dentry
*entry
;
2126 d_tracer
= tracing_init_dentry();
2128 entry
= debugfs_create_file("tracing_enabled", 0644, d_tracer
,
2129 &global_trace
, &tracing_ctrl_fops
);
2131 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2133 entry
= debugfs_create_file("iter_ctrl", 0644, d_tracer
,
2134 NULL
, &tracing_iter_fops
);
2136 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2138 entry
= debugfs_create_file("latency_trace", 0444, d_tracer
,
2139 &global_trace
, &tracing_lt_fops
);
2141 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2143 entry
= debugfs_create_file("trace", 0444, d_tracer
,
2144 &global_trace
, &tracing_fops
);
2146 pr_warning("Could not create debugfs 'trace' entry\n");
2148 entry
= debugfs_create_file("available_tracers", 0444, d_tracer
,
2149 &global_trace
, &show_traces_fops
);
2151 pr_warning("Could not create debugfs 'trace' entry\n");
2153 entry
= debugfs_create_file("current_tracer", 0444, d_tracer
,
2154 &global_trace
, &set_tracer_fops
);
2156 pr_warning("Could not create debugfs 'trace' entry\n");
2158 entry
= debugfs_create_file("tracing_max_latency", 0644, d_tracer
,
2159 &tracing_max_latency
,
2160 &tracing_max_lat_fops
);
2162 pr_warning("Could not create debugfs "
2163 "'tracing_max_latency' entry\n");
2165 entry
= debugfs_create_file("tracing_thresh", 0644, d_tracer
,
2166 &tracing_thresh
, &tracing_max_lat_fops
);
2168 pr_warning("Could not create debugfs "
2169 "'tracing_threash' entry\n");
2170 entry
= debugfs_create_file("README", 0644, d_tracer
,
2171 NULL
, &tracing_readme_fops
);
2173 pr_warning("Could not create debugfs 'README' entry\n");
2175 entry
= debugfs_create_file("trace_pipe", 0644, d_tracer
,
2176 NULL
, &tracing_pipe_fops
);
2178 pr_warning("Could not create debugfs "
2179 "'tracing_threash' entry\n");
2181 #ifdef CONFIG_DYNAMIC_FTRACE
2182 entry
= debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
2183 &ftrace_update_tot_cnt
,
2184 &tracing_read_long_fops
);
2186 pr_warning("Could not create debugfs "
2187 "'dyn_ftrace_total_info' entry\n");
2191 /* dummy trace to disable tracing */
2192 static struct tracer no_tracer __read_mostly
=
2197 static int trace_alloc_page(void)
2199 struct trace_array_cpu
*data
;
2200 struct page
*page
, *tmp
;
2205 /* first allocate a page for each CPU */
2206 for_each_possible_cpu(i
) {
2207 array
= (void *)__get_free_page(GFP_KERNEL
);
2208 if (array
== NULL
) {
2209 printk(KERN_ERR
"tracer: failed to allocate page"
2210 "for trace buffer!\n");
2214 page
= virt_to_page(array
);
2215 list_add(&page
->lru
, &pages
);
2217 /* Only allocate if we are actually using the max trace */
2218 #ifdef CONFIG_TRACER_MAX_TRACE
2219 array
= (void *)__get_free_page(GFP_KERNEL
);
2220 if (array
== NULL
) {
2221 printk(KERN_ERR
"tracer: failed to allocate page"
2222 "for trace buffer!\n");
2225 page
= virt_to_page(array
);
2226 list_add(&page
->lru
, &pages
);
2230 /* Now that we successfully allocate a page per CPU, add them */
2231 for_each_possible_cpu(i
) {
2232 data
= global_trace
.data
[i
];
2233 spin_lock_init(&data
->lock
);
2234 lockdep_set_class(&data
->lock
, &data
->lock_key
);
2235 page
= list_entry(pages
.next
, struct page
, lru
);
2236 list_del_init(&page
->lru
);
2237 list_add_tail(&page
->lru
, &data
->trace_pages
);
2240 #ifdef CONFIG_TRACER_MAX_TRACE
2241 data
= max_tr
.data
[i
];
2242 spin_lock_init(&data
->lock
);
2243 lockdep_set_class(&data
->lock
, &data
->lock_key
);
2244 page
= list_entry(pages
.next
, struct page
, lru
);
2245 list_del_init(&page
->lru
);
2246 list_add_tail(&page
->lru
, &data
->trace_pages
);
2250 global_trace
.entries
+= ENTRIES_PER_PAGE
;
2255 list_for_each_entry_safe(page
, tmp
, &pages
, lru
) {
2256 list_del_init(&page
->lru
);
2262 __init
static int tracer_alloc_buffers(void)
2264 struct trace_array_cpu
*data
;
2271 global_trace
.ctrl
= tracer_enabled
;
2273 /* Allocate the first page for all buffers */
2274 for_each_possible_cpu(i
) {
2275 data
= global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
2276 max_tr
.data
[i
] = &per_cpu(max_data
, i
);
2278 array
= (void *)__get_free_page(GFP_KERNEL
);
2279 if (array
== NULL
) {
2280 printk(KERN_ERR
"tracer: failed to allocate page"
2281 "for trace buffer!\n");
2285 /* set the array to the list */
2286 INIT_LIST_HEAD(&data
->trace_pages
);
2287 page
= virt_to_page(array
);
2288 list_add(&page
->lru
, &data
->trace_pages
);
2289 /* use the LRU flag to differentiate the two buffers */
2292 /* Only allocate if we are actually using the max trace */
2293 #ifdef CONFIG_TRACER_MAX_TRACE
2294 array
= (void *)__get_free_page(GFP_KERNEL
);
2295 if (array
== NULL
) {
2296 printk(KERN_ERR
"tracer: failed to allocate page"
2297 "for trace buffer!\n");
2301 INIT_LIST_HEAD(&max_tr
.data
[i
]->trace_pages
);
2302 page
= virt_to_page(array
);
2303 list_add(&page
->lru
, &max_tr
.data
[i
]->trace_pages
);
2309 * Since we allocate by orders of pages, we may be able to
2312 global_trace
.entries
= ENTRIES_PER_PAGE
;
2315 while (global_trace
.entries
< trace_nr_entries
) {
2316 if (trace_alloc_page())
2320 max_tr
.entries
= global_trace
.entries
;
2322 pr_info("tracer: %d pages allocated for %ld",
2323 pages
, trace_nr_entries
);
2324 pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE
);
2325 pr_info(" actual entries %ld\n", global_trace
.entries
);
2327 tracer_init_debugfs();
2329 trace_init_cmdlines();
2331 register_tracer(&no_tracer
);
2332 current_trace
= &no_tracer
;
2334 /* All seems OK, enable tracing */
2335 tracing_disabled
= 0;
2340 for (i
-- ; i
>= 0; i
--) {
2341 struct page
*page
, *tmp
;
2342 struct trace_array_cpu
*data
= global_trace
.data
[i
];
2345 list_for_each_entry_safe(page
, tmp
,
2346 &data
->trace_pages
, lru
) {
2347 list_del_init(&page
->lru
);
2352 #ifdef CONFIG_TRACER_MAX_TRACE
2353 data
= max_tr
.data
[i
];
2355 list_for_each_entry_safe(page
, tmp
,
2356 &data
->trace_pages
, lru
) {
2357 list_del_init(&page
->lru
);
2365 fs_initcall(tracer_alloc_buffers
);