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/poll.h>
28 #include <linux/gfp.h>
31 #include <linux/stacktrace.h>
35 unsigned long __read_mostly tracing_max_latency
= (cycle_t
)ULONG_MAX
;
36 unsigned long __read_mostly tracing_thresh
;
38 static int tracing_disabled
= 1;
41 ns2usecs(cycle_t nsec
)
48 cycle_t
ftrace_now(int cpu
)
50 return cpu_clock(cpu
);
53 static struct trace_array global_trace
;
55 static DEFINE_PER_CPU(struct trace_array_cpu
, global_trace_cpu
);
57 static struct trace_array max_tr
;
59 static DEFINE_PER_CPU(struct trace_array_cpu
, max_data
);
61 static int tracer_enabled
= 1;
62 static unsigned long trace_nr_entries
= 65536UL;
64 static struct tracer
*trace_types __read_mostly
;
65 static struct tracer
*current_trace __read_mostly
;
66 static int max_tracer_type_len
;
68 static DEFINE_MUTEX(trace_types_lock
);
69 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
71 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
;
73 void trace_wake_up(void)
76 * The runqueue_is_locked() can fail, but this is the best we
79 if (!(trace_flags
& TRACE_ITER_BLOCK
) && !runqueue_is_locked())
83 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
85 static int __init
set_nr_entries(char *str
)
89 trace_nr_entries
= simple_strtoul(str
, &str
, 0);
92 __setup("trace_entries=", set_nr_entries
);
94 unsigned long nsecs_to_usecs(unsigned long nsecs
)
99 enum trace_flag_type
{
100 TRACE_FLAG_IRQS_OFF
= 0x01,
101 TRACE_FLAG_NEED_RESCHED
= 0x02,
102 TRACE_FLAG_HARDIRQ
= 0x04,
103 TRACE_FLAG_SOFTIRQ
= 0x08,
106 #define TRACE_ITER_SYM_MASK \
107 (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
109 /* These must match the bit postions above */
110 static const char *trace_options
[] = {
124 static raw_spinlock_t ftrace_max_lock
=
125 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
128 * Copy the new maximum trace into the separate maximum-trace
129 * structure. (this way the maximum trace is permanently saved,
130 * for later retrieval via /debugfs/tracing/latency_trace)
133 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
135 struct trace_array_cpu
*data
= tr
->data
[cpu
];
138 max_tr
.time_start
= data
->preempt_timestamp
;
140 data
= max_tr
.data
[cpu
];
141 data
->saved_latency
= tracing_max_latency
;
143 memcpy(data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
144 data
->pid
= tsk
->pid
;
145 data
->uid
= tsk
->uid
;
146 data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
147 data
->policy
= tsk
->policy
;
148 data
->rt_priority
= tsk
->rt_priority
;
150 /* record this tasks comm */
151 tracing_record_cmdline(current
);
154 void check_pages(struct trace_array_cpu
*data
)
156 struct page
*page
, *tmp
;
158 BUG_ON(data
->trace_pages
.next
->prev
!= &data
->trace_pages
);
159 BUG_ON(data
->trace_pages
.prev
->next
!= &data
->trace_pages
);
161 list_for_each_entry_safe(page
, tmp
, &data
->trace_pages
, lru
) {
162 BUG_ON(page
->lru
.next
->prev
!= &page
->lru
);
163 BUG_ON(page
->lru
.prev
->next
!= &page
->lru
);
167 void *head_page(struct trace_array_cpu
*data
)
172 if (list_empty(&data
->trace_pages
))
175 page
= list_entry(data
->trace_pages
.next
, struct page
, lru
);
176 BUG_ON(&page
->lru
== &data
->trace_pages
);
178 return page_address(page
);
182 trace_seq_printf(struct trace_seq
*s
, const char *fmt
, ...)
184 int len
= (PAGE_SIZE
- 1) - s
->len
;
192 ret
= vsnprintf(s
->buffer
+ s
->len
, len
, fmt
, ap
);
195 /* If we can't write it all, don't bother writing anything */
205 trace_seq_puts(struct trace_seq
*s
, const char *str
)
207 int len
= strlen(str
);
209 if (len
> ((PAGE_SIZE
- 1) - s
->len
))
212 memcpy(s
->buffer
+ s
->len
, str
, len
);
219 trace_seq_putc(struct trace_seq
*s
, unsigned char c
)
221 if (s
->len
>= (PAGE_SIZE
- 1))
224 s
->buffer
[s
->len
++] = c
;
230 trace_seq_putmem(struct trace_seq
*s
, void *mem
, size_t len
)
232 if (len
> ((PAGE_SIZE
- 1) - s
->len
))
235 memcpy(s
->buffer
+ s
->len
, mem
, len
);
244 trace_seq_putmem_hex(struct trace_seq
*s
, void *mem
, size_t len
)
246 unsigned char hex
[HEX_CHARS
];
251 BUG_ON(len
>= HEX_CHARS
);
256 for (i
= 0, j
= 0; i
< len
; i
++) {
258 for (i
= len
-1, j
= 0; i
>= 0; i
--) {
262 hex
[j
] = byte
& 0x0f;
279 return trace_seq_putmem(s
, hex
, j
);
283 trace_seq_reset(struct trace_seq
*s
)
289 trace_print_seq(struct seq_file
*m
, struct trace_seq
*s
)
291 int len
= s
->len
>= PAGE_SIZE
? PAGE_SIZE
- 1 : s
->len
;
294 seq_puts(m
, s
->buffer
);
300 flip_trace(struct trace_array_cpu
*tr1
, struct trace_array_cpu
*tr2
)
302 struct list_head flip_pages
;
304 INIT_LIST_HEAD(&flip_pages
);
306 memcpy(&tr1
->trace_head_idx
, &tr2
->trace_head_idx
,
307 sizeof(struct trace_array_cpu
) -
308 offsetof(struct trace_array_cpu
, trace_head_idx
));
312 list_splice_init(&tr1
->trace_pages
, &flip_pages
);
313 list_splice_init(&tr2
->trace_pages
, &tr1
->trace_pages
);
314 list_splice_init(&flip_pages
, &tr2
->trace_pages
);
315 BUG_ON(!list_empty(&flip_pages
));
321 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
323 struct trace_array_cpu
*data
;
326 WARN_ON_ONCE(!irqs_disabled());
327 __raw_spin_lock(&ftrace_max_lock
);
328 /* clear out all the previous traces */
329 for_each_possible_cpu(i
) {
331 flip_trace(max_tr
.data
[i
], data
);
335 __update_max_tr(tr
, tsk
, cpu
);
336 __raw_spin_unlock(&ftrace_max_lock
);
340 * update_max_tr_single - only copy one trace over, and reset the rest
342 * @tsk - task with the latency
343 * @cpu - the cpu of the buffer to copy.
346 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
348 struct trace_array_cpu
*data
= tr
->data
[cpu
];
351 WARN_ON_ONCE(!irqs_disabled());
352 __raw_spin_lock(&ftrace_max_lock
);
353 for_each_possible_cpu(i
)
354 tracing_reset(max_tr
.data
[i
]);
356 flip_trace(max_tr
.data
[cpu
], data
);
359 __update_max_tr(tr
, tsk
, cpu
);
360 __raw_spin_unlock(&ftrace_max_lock
);
363 int register_tracer(struct tracer
*type
)
370 pr_info("Tracer must have a name\n");
374 mutex_lock(&trace_types_lock
);
375 for (t
= trace_types
; t
; t
= t
->next
) {
376 if (strcmp(type
->name
, t
->name
) == 0) {
378 pr_info("Trace %s already registered\n",
385 #ifdef CONFIG_FTRACE_STARTUP_TEST
386 if (type
->selftest
) {
387 struct tracer
*saved_tracer
= current_trace
;
388 struct trace_array_cpu
*data
;
389 struct trace_array
*tr
= &global_trace
;
390 int saved_ctrl
= tr
->ctrl
;
393 * Run a selftest on this tracer.
394 * Here we reset the trace buffer, and set the current
395 * tracer to be this tracer. The tracer can then run some
396 * internal tracing to verify that everything is in order.
397 * If we fail, we do not register this tracer.
399 for_each_possible_cpu(i
) {
401 if (!head_page(data
))
405 current_trace
= type
;
407 /* the test is responsible for initializing and enabling */
408 pr_info("Testing tracer %s: ", type
->name
);
409 ret
= type
->selftest(type
, tr
);
410 /* the test is responsible for resetting too */
411 current_trace
= saved_tracer
;
412 tr
->ctrl
= saved_ctrl
;
414 printk(KERN_CONT
"FAILED!\n");
417 /* Only reset on passing, to avoid touching corrupted buffers */
418 for_each_possible_cpu(i
) {
420 if (!head_page(data
))
424 printk(KERN_CONT
"PASSED\n");
428 type
->next
= trace_types
;
430 len
= strlen(type
->name
);
431 if (len
> max_tracer_type_len
)
432 max_tracer_type_len
= len
;
435 mutex_unlock(&trace_types_lock
);
440 void unregister_tracer(struct tracer
*type
)
445 mutex_lock(&trace_types_lock
);
446 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
450 pr_info("Trace %s not registered\n", type
->name
);
455 if (strlen(type
->name
) != max_tracer_type_len
)
458 max_tracer_type_len
= 0;
459 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
460 len
= strlen((*t
)->name
);
461 if (len
> max_tracer_type_len
)
462 max_tracer_type_len
= len
;
465 mutex_unlock(&trace_types_lock
);
468 void tracing_reset(struct trace_array_cpu
*data
)
471 data
->trace_head
= data
->trace_tail
= head_page(data
);
472 data
->trace_head_idx
= 0;
473 data
->trace_tail_idx
= 0;
476 #define SAVED_CMDLINES 128
477 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
478 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
479 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
480 static int cmdline_idx
;
481 static DEFINE_SPINLOCK(trace_cmdline_lock
);
482 atomic_t trace_record_cmdline_disabled
;
484 static void trace_init_cmdlines(void)
486 memset(&map_pid_to_cmdline
, -1, sizeof(map_pid_to_cmdline
));
487 memset(&map_cmdline_to_pid
, -1, sizeof(map_cmdline_to_pid
));
491 void trace_stop_cmdline_recording(void);
493 static void trace_save_cmdline(struct task_struct
*tsk
)
498 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
502 * It's not the end of the world if we don't get
503 * the lock, but we also don't want to spin
504 * nor do we want to disable interrupts,
505 * so if we miss here, then better luck next time.
507 if (!spin_trylock(&trace_cmdline_lock
))
510 idx
= map_pid_to_cmdline
[tsk
->pid
];
511 if (idx
>= SAVED_CMDLINES
) {
512 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
514 map
= map_cmdline_to_pid
[idx
];
515 if (map
<= PID_MAX_DEFAULT
)
516 map_pid_to_cmdline
[map
] = (unsigned)-1;
518 map_pid_to_cmdline
[tsk
->pid
] = idx
;
523 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
525 spin_unlock(&trace_cmdline_lock
);
528 static char *trace_find_cmdline(int pid
)
530 char *cmdline
= "<...>";
536 if (pid
> PID_MAX_DEFAULT
)
539 map
= map_pid_to_cmdline
[pid
];
540 if (map
>= SAVED_CMDLINES
)
543 cmdline
= saved_cmdlines
[map
];
549 void tracing_record_cmdline(struct task_struct
*tsk
)
551 if (atomic_read(&trace_record_cmdline_disabled
))
554 trace_save_cmdline(tsk
);
557 static inline struct list_head
*
558 trace_next_list(struct trace_array_cpu
*data
, struct list_head
*next
)
561 * Roundrobin - but skip the head (which is not a real page):
564 if (unlikely(next
== &data
->trace_pages
))
566 BUG_ON(next
== &data
->trace_pages
);
572 trace_next_page(struct trace_array_cpu
*data
, void *addr
)
574 struct list_head
*next
;
577 page
= virt_to_page(addr
);
579 next
= trace_next_list(data
, &page
->lru
);
580 page
= list_entry(next
, struct page
, lru
);
582 return page_address(page
);
585 static inline struct trace_entry
*
586 tracing_get_trace_entry(struct trace_array
*tr
, struct trace_array_cpu
*data
)
588 unsigned long idx
, idx_next
;
589 struct trace_entry
*entry
;
592 idx
= data
->trace_head_idx
;
595 BUG_ON(idx
* TRACE_ENTRY_SIZE
>= PAGE_SIZE
);
597 entry
= data
->trace_head
+ idx
* TRACE_ENTRY_SIZE
;
599 if (unlikely(idx_next
>= ENTRIES_PER_PAGE
)) {
600 data
->trace_head
= trace_next_page(data
, data
->trace_head
);
604 if (data
->trace_head
== data
->trace_tail
&&
605 idx_next
== data
->trace_tail_idx
) {
607 data
->trace_tail_idx
++;
608 if (data
->trace_tail_idx
>= ENTRIES_PER_PAGE
) {
610 trace_next_page(data
, data
->trace_tail
);
611 data
->trace_tail_idx
= 0;
615 data
->trace_head_idx
= idx_next
;
621 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
)
623 struct task_struct
*tsk
= current
;
626 pc
= preempt_count();
628 entry
->preempt_count
= pc
& 0xff;
629 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
630 entry
->t
= ftrace_now(raw_smp_processor_id());
631 entry
->flags
= (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
632 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
633 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
634 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
638 trace_function(struct trace_array
*tr
, struct trace_array_cpu
*data
,
639 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
)
641 struct trace_entry
*entry
;
642 unsigned long irq_flags
;
644 raw_local_irq_save(irq_flags
);
645 __raw_spin_lock(&data
->lock
);
646 entry
= tracing_get_trace_entry(tr
, data
);
647 tracing_generic_entry_update(entry
, flags
);
648 entry
->type
= TRACE_FN
;
650 entry
->fn
.parent_ip
= parent_ip
;
651 __raw_spin_unlock(&data
->lock
);
652 raw_local_irq_restore(irq_flags
);
656 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
657 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
)
659 if (likely(!atomic_read(&data
->disabled
)))
660 trace_function(tr
, data
, ip
, parent_ip
, flags
);
664 __trace_special(void *__tr
, void *__data
,
665 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
667 struct trace_array_cpu
*data
= __data
;
668 struct trace_array
*tr
= __tr
;
669 struct trace_entry
*entry
;
670 unsigned long irq_flags
;
672 raw_local_irq_save(irq_flags
);
673 __raw_spin_lock(&data
->lock
);
674 entry
= tracing_get_trace_entry(tr
, data
);
675 tracing_generic_entry_update(entry
, 0);
676 entry
->type
= TRACE_SPECIAL
;
677 entry
->special
.arg1
= arg1
;
678 entry
->special
.arg2
= arg2
;
679 entry
->special
.arg3
= arg3
;
680 __raw_spin_unlock(&data
->lock
);
681 raw_local_irq_restore(irq_flags
);
686 void __trace_stack(struct trace_array
*tr
,
687 struct trace_array_cpu
*data
,
691 struct trace_entry
*entry
;
692 struct stack_trace trace
;
694 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
697 entry
= tracing_get_trace_entry(tr
, data
);
698 tracing_generic_entry_update(entry
, flags
);
699 entry
->type
= TRACE_STACK
;
701 memset(&entry
->stack
, 0, sizeof(entry
->stack
));
703 trace
.nr_entries
= 0;
704 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
706 trace
.entries
= entry
->stack
.caller
;
708 save_stack_trace(&trace
);
712 tracing_sched_switch_trace(struct trace_array
*tr
,
713 struct trace_array_cpu
*data
,
714 struct task_struct
*prev
,
715 struct task_struct
*next
,
718 struct trace_entry
*entry
;
719 unsigned long irq_flags
;
721 raw_local_irq_save(irq_flags
);
722 __raw_spin_lock(&data
->lock
);
723 entry
= tracing_get_trace_entry(tr
, data
);
724 tracing_generic_entry_update(entry
, flags
);
725 entry
->type
= TRACE_CTX
;
726 entry
->ctx
.prev_pid
= prev
->pid
;
727 entry
->ctx
.prev_prio
= prev
->prio
;
728 entry
->ctx
.prev_state
= prev
->state
;
729 entry
->ctx
.next_pid
= next
->pid
;
730 entry
->ctx
.next_prio
= next
->prio
;
731 entry
->ctx
.next_state
= next
->state
;
732 __trace_stack(tr
, data
, flags
, 4);
733 __raw_spin_unlock(&data
->lock
);
734 raw_local_irq_restore(irq_flags
);
738 tracing_sched_wakeup_trace(struct trace_array
*tr
,
739 struct trace_array_cpu
*data
,
740 struct task_struct
*wakee
,
741 struct task_struct
*curr
,
744 struct trace_entry
*entry
;
745 unsigned long irq_flags
;
747 raw_local_irq_save(irq_flags
);
748 __raw_spin_lock(&data
->lock
);
749 entry
= tracing_get_trace_entry(tr
, data
);
750 tracing_generic_entry_update(entry
, flags
);
751 entry
->type
= TRACE_WAKE
;
752 entry
->ctx
.prev_pid
= curr
->pid
;
753 entry
->ctx
.prev_prio
= curr
->prio
;
754 entry
->ctx
.prev_state
= curr
->state
;
755 entry
->ctx
.next_pid
= wakee
->pid
;
756 entry
->ctx
.next_prio
= wakee
->prio
;
757 entry
->ctx
.next_state
= wakee
->state
;
758 __trace_stack(tr
, data
, flags
, 5);
759 __raw_spin_unlock(&data
->lock
);
760 raw_local_irq_restore(irq_flags
);
767 function_trace_call(unsigned long ip
, unsigned long parent_ip
)
769 struct trace_array
*tr
= &global_trace
;
770 struct trace_array_cpu
*data
;
775 if (unlikely(!tracer_enabled
))
778 local_irq_save(flags
);
779 cpu
= raw_smp_processor_id();
780 data
= tr
->data
[cpu
];
781 disabled
= atomic_inc_return(&data
->disabled
);
783 if (likely(disabled
== 1))
784 trace_function(tr
, data
, ip
, parent_ip
, flags
);
786 atomic_dec(&data
->disabled
);
787 local_irq_restore(flags
);
790 static struct ftrace_ops trace_ops __read_mostly
=
792 .func
= function_trace_call
,
795 void tracing_start_function_trace(void)
797 register_ftrace_function(&trace_ops
);
800 void tracing_stop_function_trace(void)
802 unregister_ftrace_function(&trace_ops
);
806 enum trace_file_type
{
807 TRACE_FILE_LAT_FMT
= 1,
810 static struct trace_entry
*
811 trace_entry_idx(struct trace_array
*tr
, struct trace_array_cpu
*data
,
812 struct trace_iterator
*iter
, int cpu
)
815 struct trace_entry
*array
;
817 if (iter
->next_idx
[cpu
] >= tr
->entries
||
818 iter
->next_idx
[cpu
] >= data
->trace_idx
||
819 (data
->trace_head
== data
->trace_tail
&&
820 data
->trace_head_idx
== data
->trace_tail_idx
))
823 if (!iter
->next_page
[cpu
]) {
824 /* Initialize the iterator for this cpu trace buffer */
825 WARN_ON(!data
->trace_tail
);
826 page
= virt_to_page(data
->trace_tail
);
827 iter
->next_page
[cpu
] = &page
->lru
;
828 iter
->next_page_idx
[cpu
] = data
->trace_tail_idx
;
831 page
= list_entry(iter
->next_page
[cpu
], struct page
, lru
);
832 BUG_ON(&data
->trace_pages
== &page
->lru
);
834 array
= page_address(page
);
836 WARN_ON(iter
->next_page_idx
[cpu
] >= ENTRIES_PER_PAGE
);
837 return &array
[iter
->next_page_idx
[cpu
]];
840 static struct trace_entry
*
841 find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
)
843 struct trace_array
*tr
= iter
->tr
;
844 struct trace_entry
*ent
, *next
= NULL
;
848 for_each_possible_cpu(cpu
) {
849 if (!head_page(tr
->data
[cpu
]))
851 ent
= trace_entry_idx(tr
, tr
->data
[cpu
], iter
, cpu
);
853 * Pick the entry with the smallest timestamp:
855 if (ent
&& (!next
|| ent
->t
< next
->t
)) {
867 static void trace_iterator_increment(struct trace_iterator
*iter
)
870 iter
->next_idx
[iter
->cpu
]++;
871 iter
->next_page_idx
[iter
->cpu
]++;
873 if (iter
->next_page_idx
[iter
->cpu
] >= ENTRIES_PER_PAGE
) {
874 struct trace_array_cpu
*data
= iter
->tr
->data
[iter
->cpu
];
876 iter
->next_page_idx
[iter
->cpu
] = 0;
877 iter
->next_page
[iter
->cpu
] =
878 trace_next_list(data
, iter
->next_page
[iter
->cpu
]);
882 static void trace_consume(struct trace_iterator
*iter
)
884 struct trace_array_cpu
*data
= iter
->tr
->data
[iter
->cpu
];
886 data
->trace_tail_idx
++;
887 if (data
->trace_tail_idx
>= ENTRIES_PER_PAGE
) {
888 data
->trace_tail
= trace_next_page(data
, data
->trace_tail
);
889 data
->trace_tail_idx
= 0;
892 /* Check if we empty it, then reset the index */
893 if (data
->trace_head
== data
->trace_tail
&&
894 data
->trace_head_idx
== data
->trace_tail_idx
)
898 static void *find_next_entry_inc(struct trace_iterator
*iter
)
900 struct trace_entry
*next
;
903 next
= find_next_entry(iter
, &next_cpu
);
905 iter
->prev_ent
= iter
->ent
;
906 iter
->prev_cpu
= iter
->cpu
;
909 iter
->cpu
= next_cpu
;
912 trace_iterator_increment(iter
);
914 return next
? iter
: NULL
;
917 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
919 struct trace_iterator
*iter
= m
->private;
920 void *last_ent
= iter
->ent
;
926 /* can't go backwards */
931 ent
= find_next_entry_inc(iter
);
935 while (ent
&& iter
->idx
< i
)
936 ent
= find_next_entry_inc(iter
);
940 if (last_ent
&& !ent
)
941 seq_puts(m
, "\n\nvim:ft=help\n");
946 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
948 struct trace_iterator
*iter
= m
->private;
953 mutex_lock(&trace_types_lock
);
955 if (!current_trace
|| current_trace
!= iter
->trace
) {
956 mutex_unlock(&trace_types_lock
);
960 atomic_inc(&trace_record_cmdline_disabled
);
962 /* let the tracer grab locks here if needed */
963 if (current_trace
->start
)
964 current_trace
->start(iter
);
966 if (*pos
!= iter
->pos
) {
970 iter
->prev_ent
= NULL
;
973 for_each_possible_cpu(i
) {
974 iter
->next_idx
[i
] = 0;
975 iter
->next_page
[i
] = NULL
;
978 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
983 p
= s_next(m
, p
, &l
);
989 static void s_stop(struct seq_file
*m
, void *p
)
991 struct trace_iterator
*iter
= m
->private;
993 atomic_dec(&trace_record_cmdline_disabled
);
995 /* let the tracer release locks here if needed */
996 if (current_trace
&& current_trace
== iter
->trace
&& iter
->trace
->stop
)
997 iter
->trace
->stop(iter
);
999 mutex_unlock(&trace_types_lock
);
1003 seq_print_sym_short(struct trace_seq
*s
, const char *fmt
, unsigned long address
)
1005 #ifdef CONFIG_KALLSYMS
1006 char str
[KSYM_SYMBOL_LEN
];
1008 kallsyms_lookup(address
, NULL
, NULL
, NULL
, str
);
1010 return trace_seq_printf(s
, fmt
, str
);
1016 seq_print_sym_offset(struct trace_seq
*s
, const char *fmt
,
1017 unsigned long address
)
1019 #ifdef CONFIG_KALLSYMS
1020 char str
[KSYM_SYMBOL_LEN
];
1022 sprint_symbol(str
, address
);
1023 return trace_seq_printf(s
, fmt
, str
);
1028 #ifndef CONFIG_64BIT
1029 # define IP_FMT "%08lx"
1031 # define IP_FMT "%016lx"
1035 seq_print_ip_sym(struct trace_seq
*s
, unsigned long ip
, unsigned long sym_flags
)
1040 return trace_seq_printf(s
, "0");
1042 if (sym_flags
& TRACE_ITER_SYM_OFFSET
)
1043 ret
= seq_print_sym_offset(s
, "%s", ip
);
1045 ret
= seq_print_sym_short(s
, "%s", ip
);
1050 if (sym_flags
& TRACE_ITER_SYM_ADDR
)
1051 ret
= trace_seq_printf(s
, " <" IP_FMT
">", ip
);
1055 static void print_lat_help_header(struct seq_file
*m
)
1057 seq_puts(m
, "# _------=> CPU# \n");
1058 seq_puts(m
, "# / _-----=> irqs-off \n");
1059 seq_puts(m
, "# | / _----=> need-resched \n");
1060 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1061 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1062 seq_puts(m
, "# |||| / \n");
1063 seq_puts(m
, "# ||||| delay \n");
1064 seq_puts(m
, "# cmd pid ||||| time | caller \n");
1065 seq_puts(m
, "# \\ / ||||| \\ | / \n");
1068 static void print_func_help_header(struct seq_file
*m
)
1070 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1071 seq_puts(m
, "# | | | | |\n");
1076 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1078 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1079 struct trace_array
*tr
= iter
->tr
;
1080 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1081 struct tracer
*type
= current_trace
;
1082 unsigned long total
= 0;
1083 unsigned long entries
= 0;
1085 const char *name
= "preemption";
1090 for_each_possible_cpu(cpu
) {
1091 if (head_page(tr
->data
[cpu
])) {
1092 total
+= tr
->data
[cpu
]->trace_idx
;
1093 if (tr
->data
[cpu
]->trace_idx
> tr
->entries
)
1094 entries
+= tr
->entries
;
1096 entries
+= tr
->data
[cpu
]->trace_idx
;
1100 seq_printf(m
, "%s latency trace v1.1.5 on %s\n",
1102 seq_puts(m
, "-----------------------------------"
1103 "---------------------------------\n");
1104 seq_printf(m
, " latency: %lu us, #%lu/%lu, CPU#%d |"
1105 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1106 nsecs_to_usecs(data
->saved_latency
),
1110 #if defined(CONFIG_PREEMPT_NONE)
1112 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1114 #elif defined(CONFIG_PREEMPT_DESKTOP)
1119 /* These are reserved for later use */
1122 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1126 seq_puts(m
, " -----------------\n");
1127 seq_printf(m
, " | task: %.16s-%d "
1128 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1129 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1130 data
->policy
, data
->rt_priority
);
1131 seq_puts(m
, " -----------------\n");
1133 if (data
->critical_start
) {
1134 seq_puts(m
, " => started at: ");
1135 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1136 trace_print_seq(m
, &iter
->seq
);
1137 seq_puts(m
, "\n => ended at: ");
1138 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1139 trace_print_seq(m
, &iter
->seq
);
1147 lat_print_generic(struct trace_seq
*s
, struct trace_entry
*entry
, int cpu
)
1149 int hardirq
, softirq
;
1152 comm
= trace_find_cmdline(entry
->pid
);
1154 trace_seq_printf(s
, "%8.8s-%-5d ", comm
, entry
->pid
);
1155 trace_seq_printf(s
, "%d", cpu
);
1156 trace_seq_printf(s
, "%c%c",
1157 (entry
->flags
& TRACE_FLAG_IRQS_OFF
) ? 'd' : '.',
1158 ((entry
->flags
& TRACE_FLAG_NEED_RESCHED
) ? 'N' : '.'));
1160 hardirq
= entry
->flags
& TRACE_FLAG_HARDIRQ
;
1161 softirq
= entry
->flags
& TRACE_FLAG_SOFTIRQ
;
1162 if (hardirq
&& softirq
)
1163 trace_seq_putc(s
, 'H');
1166 trace_seq_putc(s
, 'h');
1169 trace_seq_putc(s
, 's');
1171 trace_seq_putc(s
, '.');
1175 if (entry
->preempt_count
)
1176 trace_seq_printf(s
, "%x", entry
->preempt_count
);
1178 trace_seq_puts(s
, ".");
1181 unsigned long preempt_mark_thresh
= 100;
1184 lat_print_timestamp(struct trace_seq
*s
, unsigned long long abs_usecs
,
1185 unsigned long rel_usecs
)
1187 trace_seq_printf(s
, " %4lldus", abs_usecs
);
1188 if (rel_usecs
> preempt_mark_thresh
)
1189 trace_seq_puts(s
, "!: ");
1190 else if (rel_usecs
> 1)
1191 trace_seq_puts(s
, "+: ");
1193 trace_seq_puts(s
, " : ");
1196 static const char state_to_char
[] = TASK_STATE_TO_CHAR_STR
;
1199 print_lat_fmt(struct trace_iterator
*iter
, unsigned int trace_idx
, int cpu
)
1201 struct trace_seq
*s
= &iter
->seq
;
1202 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1203 struct trace_entry
*next_entry
= find_next_entry(iter
, NULL
);
1204 unsigned long verbose
= (trace_flags
& TRACE_ITER_VERBOSE
);
1205 struct trace_entry
*entry
= iter
->ent
;
1206 unsigned long abs_usecs
;
1207 unsigned long rel_usecs
;
1215 rel_usecs
= ns2usecs(next_entry
->t
- entry
->t
);
1216 abs_usecs
= ns2usecs(entry
->t
- iter
->tr
->time_start
);
1219 comm
= trace_find_cmdline(entry
->pid
);
1220 trace_seq_printf(s
, "%16s %5d %d %d %08x %08x [%08lx]"
1221 " %ld.%03ldms (+%ld.%03ldms): ",
1223 entry
->pid
, cpu
, entry
->flags
,
1224 entry
->preempt_count
, trace_idx
,
1227 abs_usecs
% 1000, rel_usecs
/1000,
1230 lat_print_generic(s
, entry
, cpu
);
1231 lat_print_timestamp(s
, abs_usecs
, rel_usecs
);
1233 switch (entry
->type
) {
1235 seq_print_ip_sym(s
, entry
->fn
.ip
, sym_flags
);
1236 trace_seq_puts(s
, " (");
1237 seq_print_ip_sym(s
, entry
->fn
.parent_ip
, sym_flags
);
1238 trace_seq_puts(s
, ")\n");
1242 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1243 state_to_char
[entry
->ctx
.next_state
] : 'X';
1245 state
= entry
->ctx
.prev_state
? __ffs(entry
->ctx
.prev_state
) + 1 : 0;
1246 S
= state
< sizeof(state_to_char
) - 1 ? state_to_char
[state
] : 'X';
1247 comm
= trace_find_cmdline(entry
->ctx
.next_pid
);
1248 trace_seq_printf(s
, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1249 entry
->ctx
.prev_pid
,
1250 entry
->ctx
.prev_prio
,
1251 S
, entry
->type
== TRACE_CTX
? "==>" : " +",
1252 entry
->ctx
.next_pid
,
1253 entry
->ctx
.next_prio
,
1257 trace_seq_printf(s
, "# %ld %ld %ld\n",
1258 entry
->special
.arg1
,
1259 entry
->special
.arg2
,
1260 entry
->special
.arg3
);
1263 for (i
= 0; i
< FTRACE_STACK_ENTRIES
; i
++) {
1265 trace_seq_puts(s
, " <= ");
1266 seq_print_ip_sym(s
, entry
->stack
.caller
[i
], sym_flags
);
1268 trace_seq_puts(s
, "\n");
1271 trace_seq_printf(s
, "Unknown type %d\n", entry
->type
);
1276 static int print_trace_fmt(struct trace_iterator
*iter
)
1278 struct trace_seq
*s
= &iter
->seq
;
1279 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1280 struct trace_entry
*entry
;
1281 unsigned long usec_rem
;
1282 unsigned long long t
;
1291 comm
= trace_find_cmdline(iter
->ent
->pid
);
1293 t
= ns2usecs(entry
->t
);
1294 usec_rem
= do_div(t
, 1000000ULL);
1295 secs
= (unsigned long)t
;
1297 ret
= trace_seq_printf(s
, "%16s-%-5d ", comm
, entry
->pid
);
1300 ret
= trace_seq_printf(s
, "[%02d] ", iter
->cpu
);
1303 ret
= trace_seq_printf(s
, "%5lu.%06lu: ", secs
, usec_rem
);
1307 switch (entry
->type
) {
1309 ret
= seq_print_ip_sym(s
, entry
->fn
.ip
, sym_flags
);
1312 if ((sym_flags
& TRACE_ITER_PRINT_PARENT
) &&
1313 entry
->fn
.parent_ip
) {
1314 ret
= trace_seq_printf(s
, " <-");
1317 ret
= seq_print_ip_sym(s
, entry
->fn
.parent_ip
,
1322 ret
= trace_seq_printf(s
, "\n");
1328 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1329 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1330 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1331 state_to_char
[entry
->ctx
.next_state
] : 'X';
1332 ret
= trace_seq_printf(s
, " %5d:%3d:%c %s %5d:%3d:%c\n",
1333 entry
->ctx
.prev_pid
,
1334 entry
->ctx
.prev_prio
,
1336 entry
->type
== TRACE_CTX
? "==>" : " +",
1337 entry
->ctx
.next_pid
,
1338 entry
->ctx
.next_prio
,
1344 ret
= trace_seq_printf(s
, "# %ld %ld %ld\n",
1345 entry
->special
.arg1
,
1346 entry
->special
.arg2
,
1347 entry
->special
.arg3
);
1352 for (i
= 0; i
< FTRACE_STACK_ENTRIES
; i
++) {
1354 ret
= trace_seq_puts(s
, " <= ");
1358 ret
= seq_print_ip_sym(s
, entry
->stack
.caller
[i
],
1363 ret
= trace_seq_puts(s
, "\n");
1371 static int print_raw_fmt(struct trace_iterator
*iter
)
1373 struct trace_seq
*s
= &iter
->seq
;
1374 struct trace_entry
*entry
;
1380 ret
= trace_seq_printf(s
, "%d %d %llu ",
1381 entry
->pid
, iter
->cpu
, entry
->t
);
1385 switch (entry
->type
) {
1387 ret
= trace_seq_printf(s
, "%x %x\n",
1388 entry
->fn
.ip
, entry
->fn
.parent_ip
);
1394 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1395 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1396 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1397 state_to_char
[entry
->ctx
.next_state
] : 'X';
1398 if (entry
->type
== TRACE_WAKE
)
1400 ret
= trace_seq_printf(s
, "%d %d %c %d %d %c\n",
1401 entry
->ctx
.prev_pid
,
1402 entry
->ctx
.prev_prio
,
1404 entry
->ctx
.next_pid
,
1405 entry
->ctx
.next_prio
,
1412 ret
= trace_seq_printf(s
, "# %ld %ld %ld\n",
1413 entry
->special
.arg1
,
1414 entry
->special
.arg2
,
1415 entry
->special
.arg3
);
1423 #define SEQ_PUT_FIELD_RET(s, x) \
1425 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1429 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1431 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1435 static int print_hex_fmt(struct trace_iterator
*iter
)
1437 struct trace_seq
*s
= &iter
->seq
;
1438 unsigned char newline
= '\n';
1439 struct trace_entry
*entry
;
1444 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1445 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1446 SEQ_PUT_HEX_FIELD_RET(s
, entry
->t
);
1448 switch (entry
->type
) {
1450 SEQ_PUT_HEX_FIELD_RET(s
, entry
->fn
.ip
);
1451 SEQ_PUT_HEX_FIELD_RET(s
, entry
->fn
.parent_ip
);
1455 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1456 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1457 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1458 state_to_char
[entry
->ctx
.next_state
] : 'X';
1459 if (entry
->type
== TRACE_WAKE
)
1461 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.prev_pid
);
1462 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.prev_prio
);
1463 SEQ_PUT_HEX_FIELD_RET(s
, S
);
1464 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.next_pid
);
1465 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.next_prio
);
1466 SEQ_PUT_HEX_FIELD_RET(s
, entry
->fn
.parent_ip
);
1467 SEQ_PUT_HEX_FIELD_RET(s
, T
);
1471 SEQ_PUT_HEX_FIELD_RET(s
, entry
->special
.arg1
);
1472 SEQ_PUT_HEX_FIELD_RET(s
, entry
->special
.arg2
);
1473 SEQ_PUT_HEX_FIELD_RET(s
, entry
->special
.arg3
);
1476 SEQ_PUT_FIELD_RET(s
, newline
);
1481 static int print_bin_fmt(struct trace_iterator
*iter
)
1483 struct trace_seq
*s
= &iter
->seq
;
1484 struct trace_entry
*entry
;
1488 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1489 SEQ_PUT_FIELD_RET(s
, entry
->cpu
);
1490 SEQ_PUT_FIELD_RET(s
, entry
->t
);
1492 switch (entry
->type
) {
1494 SEQ_PUT_FIELD_RET(s
, entry
->fn
.ip
);
1495 SEQ_PUT_FIELD_RET(s
, entry
->fn
.parent_ip
);
1498 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_pid
);
1499 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_prio
);
1500 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_state
);
1501 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_pid
);
1502 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_prio
);
1503 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_state
);
1507 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg1
);
1508 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg2
);
1509 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg3
);
1515 static int trace_empty(struct trace_iterator
*iter
)
1517 struct trace_array_cpu
*data
;
1520 for_each_possible_cpu(cpu
) {
1521 data
= iter
->tr
->data
[cpu
];
1523 if (head_page(data
) && data
->trace_idx
&&
1524 (data
->trace_tail
!= data
->trace_head
||
1525 data
->trace_tail_idx
!= data
->trace_head_idx
))
1531 static int print_trace_line(struct trace_iterator
*iter
)
1533 if (iter
->trace
&& iter
->trace
->print_line
)
1534 return iter
->trace
->print_line(iter
);
1536 if (trace_flags
& TRACE_ITER_BIN
)
1537 return print_bin_fmt(iter
);
1539 if (trace_flags
& TRACE_ITER_HEX
)
1540 return print_hex_fmt(iter
);
1542 if (trace_flags
& TRACE_ITER_RAW
)
1543 return print_raw_fmt(iter
);
1545 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
1546 return print_lat_fmt(iter
, iter
->idx
, iter
->cpu
);
1548 return print_trace_fmt(iter
);
1551 static int s_show(struct seq_file
*m
, void *v
)
1553 struct trace_iterator
*iter
= v
;
1555 if (iter
->ent
== NULL
) {
1557 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
1560 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1561 /* print nothing if the buffers are empty */
1562 if (trace_empty(iter
))
1564 print_trace_header(m
, iter
);
1565 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1566 print_lat_help_header(m
);
1568 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1569 print_func_help_header(m
);
1572 print_trace_line(iter
);
1573 trace_print_seq(m
, &iter
->seq
);
1579 static struct seq_operations tracer_seq_ops
= {
1586 static struct trace_iterator
*
1587 __tracing_open(struct inode
*inode
, struct file
*file
, int *ret
)
1589 struct trace_iterator
*iter
;
1591 if (tracing_disabled
) {
1596 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
1602 mutex_lock(&trace_types_lock
);
1603 if (current_trace
&& current_trace
->print_max
)
1606 iter
->tr
= inode
->i_private
;
1607 iter
->trace
= current_trace
;
1610 /* TODO stop tracer */
1611 *ret
= seq_open(file
, &tracer_seq_ops
);
1613 struct seq_file
*m
= file
->private_data
;
1616 /* stop the trace while dumping */
1620 if (iter
->trace
&& iter
->trace
->open
)
1621 iter
->trace
->open(iter
);
1626 mutex_unlock(&trace_types_lock
);
1632 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
1634 if (tracing_disabled
)
1637 filp
->private_data
= inode
->i_private
;
1641 int tracing_release(struct inode
*inode
, struct file
*file
)
1643 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
1644 struct trace_iterator
*iter
= m
->private;
1646 mutex_lock(&trace_types_lock
);
1647 if (iter
->trace
&& iter
->trace
->close
)
1648 iter
->trace
->close(iter
);
1650 /* reenable tracing if it was previously enabled */
1653 mutex_unlock(&trace_types_lock
);
1655 seq_release(inode
, file
);
1660 static int tracing_open(struct inode
*inode
, struct file
*file
)
1664 __tracing_open(inode
, file
, &ret
);
1669 static int tracing_lt_open(struct inode
*inode
, struct file
*file
)
1671 struct trace_iterator
*iter
;
1674 iter
= __tracing_open(inode
, file
, &ret
);
1677 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
1684 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1686 struct tracer
*t
= m
->private;
1698 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
1700 struct tracer
*t
= m
->private;
1703 mutex_lock(&trace_types_lock
);
1704 for (; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
1710 static void t_stop(struct seq_file
*m
, void *p
)
1712 mutex_unlock(&trace_types_lock
);
1715 static int t_show(struct seq_file
*m
, void *v
)
1717 struct tracer
*t
= v
;
1722 seq_printf(m
, "%s", t
->name
);
1731 static struct seq_operations show_traces_seq_ops
= {
1738 static int show_traces_open(struct inode
*inode
, struct file
*file
)
1742 if (tracing_disabled
)
1745 ret
= seq_open(file
, &show_traces_seq_ops
);
1747 struct seq_file
*m
= file
->private_data
;
1748 m
->private = trace_types
;
1754 static struct file_operations tracing_fops
= {
1755 .open
= tracing_open
,
1757 .llseek
= seq_lseek
,
1758 .release
= tracing_release
,
1761 static struct file_operations tracing_lt_fops
= {
1762 .open
= tracing_lt_open
,
1764 .llseek
= seq_lseek
,
1765 .release
= tracing_release
,
1768 static struct file_operations show_traces_fops
= {
1769 .open
= show_traces_open
,
1771 .release
= seq_release
,
1775 * Only trace on a CPU if the bitmask is set:
1777 static cpumask_t tracing_cpumask
= CPU_MASK_ALL
;
1780 * When tracing/tracing_cpu_mask is modified then this holds
1781 * the new bitmask we are about to install:
1783 static cpumask_t tracing_cpumask_new
;
1786 * The tracer itself will not take this lock, but still we want
1787 * to provide a consistent cpumask to user-space:
1789 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
1792 * Temporary storage for the character representation of the
1793 * CPU bitmask (and one more byte for the newline):
1795 static char mask_str
[NR_CPUS
+ 1];
1798 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
1799 size_t count
, loff_t
*ppos
)
1803 mutex_lock(&tracing_cpumask_update_lock
);
1805 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
1806 if (count
- len
< 2) {
1810 len
+= sprintf(mask_str
+ len
, "\n");
1811 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
1814 mutex_unlock(&tracing_cpumask_update_lock
);
1820 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
1821 size_t count
, loff_t
*ppos
)
1825 mutex_lock(&tracing_cpumask_update_lock
);
1826 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
1830 raw_local_irq_disable();
1831 __raw_spin_lock(&ftrace_max_lock
);
1832 for_each_possible_cpu(cpu
) {
1834 * Increase/decrease the disabled counter if we are
1835 * about to flip a bit in the cpumask:
1837 if (cpu_isset(cpu
, tracing_cpumask
) &&
1838 !cpu_isset(cpu
, tracing_cpumask_new
)) {
1839 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
1841 if (!cpu_isset(cpu
, tracing_cpumask
) &&
1842 cpu_isset(cpu
, tracing_cpumask_new
)) {
1843 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
1846 __raw_spin_unlock(&ftrace_max_lock
);
1847 raw_local_irq_enable();
1849 tracing_cpumask
= tracing_cpumask_new
;
1851 mutex_unlock(&tracing_cpumask_update_lock
);
1856 mutex_unlock(&tracing_cpumask_update_lock
);
1861 static struct file_operations tracing_cpumask_fops
= {
1862 .open
= tracing_open_generic
,
1863 .read
= tracing_cpumask_read
,
1864 .write
= tracing_cpumask_write
,
1868 tracing_iter_ctrl_read(struct file
*filp
, char __user
*ubuf
,
1869 size_t cnt
, loff_t
*ppos
)
1876 /* calulate max size */
1877 for (i
= 0; trace_options
[i
]; i
++) {
1878 len
+= strlen(trace_options
[i
]);
1879 len
+= 3; /* "no" and space */
1882 /* +2 for \n and \0 */
1883 buf
= kmalloc(len
+ 2, GFP_KERNEL
);
1887 for (i
= 0; trace_options
[i
]; i
++) {
1888 if (trace_flags
& (1 << i
))
1889 r
+= sprintf(buf
+ r
, "%s ", trace_options
[i
]);
1891 r
+= sprintf(buf
+ r
, "no%s ", trace_options
[i
]);
1894 r
+= sprintf(buf
+ r
, "\n");
1895 WARN_ON(r
>= len
+ 2);
1897 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
1905 tracing_iter_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
1906 size_t cnt
, loff_t
*ppos
)
1916 if (copy_from_user(&buf
, ubuf
, cnt
))
1921 if (strncmp(buf
, "no", 2) == 0) {
1926 for (i
= 0; trace_options
[i
]; i
++) {
1927 int len
= strlen(trace_options
[i
]);
1929 if (strncmp(cmp
, trace_options
[i
], len
) == 0) {
1931 trace_flags
&= ~(1 << i
);
1933 trace_flags
|= (1 << i
);
1938 * If no option could be set, return an error:
1940 if (!trace_options
[i
])
1948 static struct file_operations tracing_iter_fops
= {
1949 .open
= tracing_open_generic
,
1950 .read
= tracing_iter_ctrl_read
,
1951 .write
= tracing_iter_ctrl_write
,
1954 static const char readme_msg
[] =
1955 "tracing mini-HOWTO:\n\n"
1957 "# mount -t debugfs nodev /debug\n\n"
1958 "# cat /debug/tracing/available_tracers\n"
1959 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1960 "# cat /debug/tracing/current_tracer\n"
1962 "# echo sched_switch > /debug/tracing/current_tracer\n"
1963 "# cat /debug/tracing/current_tracer\n"
1965 "# cat /debug/tracing/iter_ctrl\n"
1966 "noprint-parent nosym-offset nosym-addr noverbose\n"
1967 "# echo print-parent > /debug/tracing/iter_ctrl\n"
1968 "# echo 1 > /debug/tracing/tracing_enabled\n"
1969 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1970 "echo 0 > /debug/tracing/tracing_enabled\n"
1974 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
1975 size_t cnt
, loff_t
*ppos
)
1977 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
1978 readme_msg
, strlen(readme_msg
));
1981 static struct file_operations tracing_readme_fops
= {
1982 .open
= tracing_open_generic
,
1983 .read
= tracing_readme_read
,
1987 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
1988 size_t cnt
, loff_t
*ppos
)
1990 struct trace_array
*tr
= filp
->private_data
;
1994 r
= sprintf(buf
, "%ld\n", tr
->ctrl
);
1995 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
1999 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2000 size_t cnt
, loff_t
*ppos
)
2002 struct trace_array
*tr
= filp
->private_data
;
2009 if (copy_from_user(&buf
, ubuf
, cnt
))
2014 val
= simple_strtoul(buf
, NULL
, 10);
2018 mutex_lock(&trace_types_lock
);
2019 if (tr
->ctrl
^ val
) {
2027 if (current_trace
&& current_trace
->ctrl_update
)
2028 current_trace
->ctrl_update(tr
);
2030 mutex_unlock(&trace_types_lock
);
2038 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2039 size_t cnt
, loff_t
*ppos
)
2041 char buf
[max_tracer_type_len
+2];
2044 mutex_lock(&trace_types_lock
);
2046 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2048 r
= sprintf(buf
, "\n");
2049 mutex_unlock(&trace_types_lock
);
2051 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2055 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2056 size_t cnt
, loff_t
*ppos
)
2058 struct trace_array
*tr
= &global_trace
;
2060 char buf
[max_tracer_type_len
+1];
2063 if (cnt
> max_tracer_type_len
)
2064 cnt
= max_tracer_type_len
;
2066 if (copy_from_user(&buf
, ubuf
, cnt
))
2071 /* strip ending whitespace. */
2072 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2075 mutex_lock(&trace_types_lock
);
2076 for (t
= trace_types
; t
; t
= t
->next
) {
2077 if (strcmp(t
->name
, buf
) == 0)
2080 if (!t
|| t
== current_trace
)
2083 if (current_trace
&& current_trace
->reset
)
2084 current_trace
->reset(tr
);
2091 mutex_unlock(&trace_types_lock
);
2099 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2100 size_t cnt
, loff_t
*ppos
)
2102 unsigned long *ptr
= filp
->private_data
;
2106 r
= snprintf(buf
, 64, "%ld\n",
2107 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2110 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2114 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2115 size_t cnt
, loff_t
*ppos
)
2117 long *ptr
= filp
->private_data
;
2124 if (copy_from_user(&buf
, ubuf
, cnt
))
2129 val
= simple_strtoul(buf
, NULL
, 10);
2136 static atomic_t tracing_reader
;
2138 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2140 struct trace_iterator
*iter
;
2142 if (tracing_disabled
)
2145 /* We only allow for reader of the pipe */
2146 if (atomic_inc_return(&tracing_reader
) != 1) {
2147 atomic_dec(&tracing_reader
);
2151 /* create a buffer to store the information to pass to userspace */
2152 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2156 iter
->tr
= &global_trace
;
2157 iter
->trace
= current_trace
;
2159 filp
->private_data
= iter
;
2164 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
2166 struct trace_iterator
*iter
= file
->private_data
;
2169 atomic_dec(&tracing_reader
);
2175 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
2177 struct trace_iterator
*iter
= filp
->private_data
;
2179 if (trace_flags
& TRACE_ITER_BLOCK
) {
2181 * Always select as readable when in blocking mode
2183 return POLLIN
| POLLRDNORM
;
2186 if (!trace_empty(iter
))
2187 return POLLIN
| POLLRDNORM
;
2188 poll_wait(filp
, &trace_wait
, poll_table
);
2189 if (!trace_empty(iter
))
2190 return POLLIN
| POLLRDNORM
;
2200 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
2201 size_t cnt
, loff_t
*ppos
)
2203 struct trace_iterator
*iter
= filp
->private_data
;
2204 struct trace_array_cpu
*data
;
2205 struct trace_array
*tr
= iter
->tr
;
2206 struct tracer
*tracer
= iter
->trace
;
2207 static cpumask_t mask
;
2209 unsigned long flags
;
2210 #ifdef CONFIG_FTRACE
2218 /* return any leftover data */
2219 if (iter
->seq
.len
> start
) {
2220 len
= iter
->seq
.len
- start
;
2223 ret
= copy_to_user(ubuf
, iter
->seq
.buffer
+ start
, cnt
);
2232 trace_seq_reset(&iter
->seq
);
2235 while (trace_empty(iter
)) {
2237 * This is a make-shift waitqueue. The reason we don't use
2238 * an actual wait queue is because:
2239 * 1) we only ever have one waiter
2240 * 2) the tracing, traces all functions, we don't want
2241 * the overhead of calling wake_up and friends
2242 * (and tracing them too)
2243 * Anyway, this is really very primitive wakeup.
2245 set_current_state(TASK_INTERRUPTIBLE
);
2246 iter
->tr
->waiter
= current
;
2248 /* sleep for one second, and try again. */
2249 schedule_timeout(HZ
);
2251 iter
->tr
->waiter
= NULL
;
2253 if (signal_pending(current
))
2257 * We block until we read something and tracing is disabled.
2258 * We still block if tracing is disabled, but we have never
2259 * read anything. This allows a user to cat this file, and
2260 * then enable tracing. But after we have read something,
2261 * we give an EOF when tracing is again disabled.
2263 * iter->pos will be 0 if we haven't read anything.
2265 if (!tracer_enabled
&& iter
->pos
)
2271 /* stop when tracing is finished */
2272 if (trace_empty(iter
))
2275 if (cnt
>= PAGE_SIZE
)
2276 cnt
= PAGE_SIZE
- 1;
2278 memset(iter
, 0, sizeof(*iter
));
2280 iter
->trace
= tracer
;
2284 * We need to stop all tracing on all CPUS to read the
2285 * the next buffer. This is a bit expensive, but is
2286 * not done often. We fill all what we can read,
2287 * and then release the locks again.
2291 local_irq_save(flags
);
2292 #ifdef CONFIG_FTRACE
2293 ftrace_save
= ftrace_enabled
;
2297 for_each_possible_cpu(cpu
) {
2298 data
= iter
->tr
->data
[cpu
];
2300 if (!head_page(data
) || !data
->trace_idx
)
2303 atomic_inc(&data
->disabled
);
2307 for_each_cpu_mask(cpu
, mask
) {
2308 data
= iter
->tr
->data
[cpu
];
2309 __raw_spin_lock(&data
->lock
);
2312 while (find_next_entry_inc(iter
) != NULL
) {
2313 int len
= iter
->seq
.len
;
2315 ret
= print_trace_line(iter
);
2317 /* don't print partial lines */
2318 iter
->seq
.len
= len
;
2322 trace_consume(iter
);
2324 if (iter
->seq
.len
>= cnt
)
2328 for_each_cpu_mask(cpu
, mask
) {
2329 data
= iter
->tr
->data
[cpu
];
2330 __raw_spin_unlock(&data
->lock
);
2333 for_each_cpu_mask(cpu
, mask
) {
2334 data
= iter
->tr
->data
[cpu
];
2335 atomic_dec(&data
->disabled
);
2337 #ifdef CONFIG_FTRACE
2338 ftrace_enabled
= ftrace_save
;
2340 local_irq_restore(flags
);
2342 /* Now copy what we have to the user */
2343 read
= iter
->seq
.len
;
2347 ret
= copy_to_user(ubuf
, iter
->seq
.buffer
, read
);
2349 if (read
< iter
->seq
.len
)
2352 trace_seq_reset(&iter
->seq
);
2360 static struct file_operations tracing_max_lat_fops
= {
2361 .open
= tracing_open_generic
,
2362 .read
= tracing_max_lat_read
,
2363 .write
= tracing_max_lat_write
,
2366 static struct file_operations tracing_ctrl_fops
= {
2367 .open
= tracing_open_generic
,
2368 .read
= tracing_ctrl_read
,
2369 .write
= tracing_ctrl_write
,
2372 static struct file_operations set_tracer_fops
= {
2373 .open
= tracing_open_generic
,
2374 .read
= tracing_set_trace_read
,
2375 .write
= tracing_set_trace_write
,
2378 static struct file_operations tracing_pipe_fops
= {
2379 .open
= tracing_open_pipe
,
2380 .poll
= tracing_poll_pipe
,
2381 .read
= tracing_read_pipe
,
2382 .release
= tracing_release_pipe
,
2385 #ifdef CONFIG_DYNAMIC_FTRACE
2388 tracing_read_long(struct file
*filp
, char __user
*ubuf
,
2389 size_t cnt
, loff_t
*ppos
)
2391 unsigned long *p
= filp
->private_data
;
2395 r
= sprintf(buf
, "%ld\n", *p
);
2397 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2400 static struct file_operations tracing_read_long_fops
= {
2401 .open
= tracing_open_generic
,
2402 .read
= tracing_read_long
,
2406 static struct dentry
*d_tracer
;
2408 struct dentry
*tracing_init_dentry(void)
2415 d_tracer
= debugfs_create_dir("tracing", NULL
);
2417 if (!d_tracer
&& !once
) {
2419 pr_warning("Could not create debugfs directory 'tracing'\n");
2426 #ifdef CONFIG_FTRACE_SELFTEST
2427 /* Let selftest have access to static functions in this file */
2428 #include "trace_selftest.c"
2431 static __init
void tracer_init_debugfs(void)
2433 struct dentry
*d_tracer
;
2434 struct dentry
*entry
;
2436 d_tracer
= tracing_init_dentry();
2438 entry
= debugfs_create_file("tracing_enabled", 0644, d_tracer
,
2439 &global_trace
, &tracing_ctrl_fops
);
2441 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2443 entry
= debugfs_create_file("iter_ctrl", 0644, d_tracer
,
2444 NULL
, &tracing_iter_fops
);
2446 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2448 entry
= debugfs_create_file("tracing_cpumask", 0644, d_tracer
,
2449 NULL
, &tracing_cpumask_fops
);
2451 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2453 entry
= debugfs_create_file("latency_trace", 0444, d_tracer
,
2454 &global_trace
, &tracing_lt_fops
);
2456 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2458 entry
= debugfs_create_file("trace", 0444, d_tracer
,
2459 &global_trace
, &tracing_fops
);
2461 pr_warning("Could not create debugfs 'trace' entry\n");
2463 entry
= debugfs_create_file("available_tracers", 0444, d_tracer
,
2464 &global_trace
, &show_traces_fops
);
2466 pr_warning("Could not create debugfs 'trace' entry\n");
2468 entry
= debugfs_create_file("current_tracer", 0444, d_tracer
,
2469 &global_trace
, &set_tracer_fops
);
2471 pr_warning("Could not create debugfs 'trace' entry\n");
2473 entry
= debugfs_create_file("tracing_max_latency", 0644, d_tracer
,
2474 &tracing_max_latency
,
2475 &tracing_max_lat_fops
);
2477 pr_warning("Could not create debugfs "
2478 "'tracing_max_latency' entry\n");
2480 entry
= debugfs_create_file("tracing_thresh", 0644, d_tracer
,
2481 &tracing_thresh
, &tracing_max_lat_fops
);
2483 pr_warning("Could not create debugfs "
2484 "'tracing_threash' entry\n");
2485 entry
= debugfs_create_file("README", 0644, d_tracer
,
2486 NULL
, &tracing_readme_fops
);
2488 pr_warning("Could not create debugfs 'README' entry\n");
2490 entry
= debugfs_create_file("trace_pipe", 0644, d_tracer
,
2491 NULL
, &tracing_pipe_fops
);
2493 pr_warning("Could not create debugfs "
2494 "'tracing_threash' entry\n");
2496 #ifdef CONFIG_DYNAMIC_FTRACE
2497 entry
= debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
2498 &ftrace_update_tot_cnt
,
2499 &tracing_read_long_fops
);
2501 pr_warning("Could not create debugfs "
2502 "'dyn_ftrace_total_info' entry\n");
2506 /* dummy trace to disable tracing */
2507 static struct tracer no_tracer __read_mostly
=
2512 static int trace_alloc_page(void)
2514 struct trace_array_cpu
*data
;
2515 struct page
*page
, *tmp
;
2520 /* first allocate a page for each CPU */
2521 for_each_possible_cpu(i
) {
2522 array
= (void *)__get_free_page(GFP_KERNEL
);
2523 if (array
== NULL
) {
2524 printk(KERN_ERR
"tracer: failed to allocate page"
2525 "for trace buffer!\n");
2529 page
= virt_to_page(array
);
2530 list_add(&page
->lru
, &pages
);
2532 /* Only allocate if we are actually using the max trace */
2533 #ifdef CONFIG_TRACER_MAX_TRACE
2534 array
= (void *)__get_free_page(GFP_KERNEL
);
2535 if (array
== NULL
) {
2536 printk(KERN_ERR
"tracer: failed to allocate page"
2537 "for trace buffer!\n");
2540 page
= virt_to_page(array
);
2541 list_add(&page
->lru
, &pages
);
2545 /* Now that we successfully allocate a page per CPU, add them */
2546 for_each_possible_cpu(i
) {
2547 data
= global_trace
.data
[i
];
2548 data
->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
2549 page
= list_entry(pages
.next
, struct page
, lru
);
2550 list_del_init(&page
->lru
);
2551 list_add_tail(&page
->lru
, &data
->trace_pages
);
2554 #ifdef CONFIG_TRACER_MAX_TRACE
2555 data
= max_tr
.data
[i
];
2556 data
->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
2557 page
= list_entry(pages
.next
, struct page
, lru
);
2558 list_del_init(&page
->lru
);
2559 list_add_tail(&page
->lru
, &data
->trace_pages
);
2563 global_trace
.entries
+= ENTRIES_PER_PAGE
;
2568 list_for_each_entry_safe(page
, tmp
, &pages
, lru
) {
2569 list_del_init(&page
->lru
);
2575 __init
static int tracer_alloc_buffers(void)
2577 struct trace_array_cpu
*data
;
2584 global_trace
.ctrl
= tracer_enabled
;
2586 /* Allocate the first page for all buffers */
2587 for_each_possible_cpu(i
) {
2588 data
= global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
2589 max_tr
.data
[i
] = &per_cpu(max_data
, i
);
2591 array
= (void *)__get_free_page(GFP_KERNEL
);
2592 if (array
== NULL
) {
2593 printk(KERN_ERR
"tracer: failed to allocate page"
2594 "for trace buffer!\n");
2598 /* set the array to the list */
2599 INIT_LIST_HEAD(&data
->trace_pages
);
2600 page
= virt_to_page(array
);
2601 list_add(&page
->lru
, &data
->trace_pages
);
2602 /* use the LRU flag to differentiate the two buffers */
2605 /* Only allocate if we are actually using the max trace */
2606 #ifdef CONFIG_TRACER_MAX_TRACE
2607 array
= (void *)__get_free_page(GFP_KERNEL
);
2608 if (array
== NULL
) {
2609 printk(KERN_ERR
"tracer: failed to allocate page"
2610 "for trace buffer!\n");
2614 INIT_LIST_HEAD(&max_tr
.data
[i
]->trace_pages
);
2615 page
= virt_to_page(array
);
2616 list_add(&page
->lru
, &max_tr
.data
[i
]->trace_pages
);
2622 * Since we allocate by orders of pages, we may be able to
2625 global_trace
.entries
= ENTRIES_PER_PAGE
;
2628 while (global_trace
.entries
< trace_nr_entries
) {
2629 if (trace_alloc_page())
2633 max_tr
.entries
= global_trace
.entries
;
2635 pr_info("tracer: %d pages allocated for %ld",
2636 pages
, trace_nr_entries
);
2637 pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE
);
2638 pr_info(" actual entries %ld\n", global_trace
.entries
);
2640 tracer_init_debugfs();
2642 trace_init_cmdlines();
2644 register_tracer(&no_tracer
);
2645 current_trace
= &no_tracer
;
2647 /* All seems OK, enable tracing */
2648 tracing_disabled
= 0;
2653 for (i
-- ; i
>= 0; i
--) {
2654 struct page
*page
, *tmp
;
2655 struct trace_array_cpu
*data
= global_trace
.data
[i
];
2658 list_for_each_entry_safe(page
, tmp
,
2659 &data
->trace_pages
, lru
) {
2660 list_del_init(&page
->lru
);
2665 #ifdef CONFIG_TRACER_MAX_TRACE
2666 data
= max_tr
.data
[i
];
2668 list_for_each_entry_safe(page
, tmp
,
2669 &data
->trace_pages
, lru
) {
2670 list_del_init(&page
->lru
);
2678 fs_initcall(tracer_alloc_buffers
);