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
)
100 __TRACE_FIRST_TYPE
= 0,
111 enum trace_flag_type
{
112 TRACE_FLAG_IRQS_OFF
= 0x01,
113 TRACE_FLAG_NEED_RESCHED
= 0x02,
114 TRACE_FLAG_HARDIRQ
= 0x04,
115 TRACE_FLAG_SOFTIRQ
= 0x08,
118 #define TRACE_ITER_SYM_MASK \
119 (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
121 /* These must match the bit postions above */
122 static const char *trace_options
[] = {
136 static raw_spinlock_t ftrace_max_lock
=
137 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
140 * Copy the new maximum trace into the separate maximum-trace
141 * structure. (this way the maximum trace is permanently saved,
142 * for later retrieval via /debugfs/tracing/latency_trace)
145 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
147 struct trace_array_cpu
*data
= tr
->data
[cpu
];
150 max_tr
.time_start
= data
->preempt_timestamp
;
152 data
= max_tr
.data
[cpu
];
153 data
->saved_latency
= tracing_max_latency
;
155 memcpy(data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
156 data
->pid
= tsk
->pid
;
157 data
->uid
= tsk
->uid
;
158 data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
159 data
->policy
= tsk
->policy
;
160 data
->rt_priority
= tsk
->rt_priority
;
162 /* record this tasks comm */
163 tracing_record_cmdline(current
);
166 void check_pages(struct trace_array_cpu
*data
)
168 struct page
*page
, *tmp
;
170 BUG_ON(data
->trace_pages
.next
->prev
!= &data
->trace_pages
);
171 BUG_ON(data
->trace_pages
.prev
->next
!= &data
->trace_pages
);
173 list_for_each_entry_safe(page
, tmp
, &data
->trace_pages
, lru
) {
174 BUG_ON(page
->lru
.next
->prev
!= &page
->lru
);
175 BUG_ON(page
->lru
.prev
->next
!= &page
->lru
);
179 void *head_page(struct trace_array_cpu
*data
)
184 if (list_empty(&data
->trace_pages
))
187 page
= list_entry(data
->trace_pages
.next
, struct page
, lru
);
188 BUG_ON(&page
->lru
== &data
->trace_pages
);
190 return page_address(page
);
194 trace_seq_printf(struct trace_seq
*s
, const char *fmt
, ...)
196 int len
= (PAGE_SIZE
- 1) - s
->len
;
204 ret
= vsnprintf(s
->buffer
+ s
->len
, len
, fmt
, ap
);
207 /* If we can't write it all, don't bother writing anything */
217 trace_seq_puts(struct trace_seq
*s
, const char *str
)
219 int len
= strlen(str
);
221 if (len
> ((PAGE_SIZE
- 1) - s
->len
))
224 memcpy(s
->buffer
+ s
->len
, str
, len
);
231 trace_seq_putc(struct trace_seq
*s
, unsigned char c
)
233 if (s
->len
>= (PAGE_SIZE
- 1))
236 s
->buffer
[s
->len
++] = c
;
242 trace_seq_putmem(struct trace_seq
*s
, void *mem
, size_t len
)
244 if (len
> ((PAGE_SIZE
- 1) - s
->len
))
247 memcpy(s
->buffer
+ s
->len
, mem
, len
);
256 trace_seq_putmem_hex(struct trace_seq
*s
, void *mem
, size_t len
)
258 unsigned char hex
[HEX_CHARS
];
263 BUG_ON(len
>= HEX_CHARS
);
268 for (i
= 0, j
= 0; i
< len
; i
++) {
270 for (i
= len
-1, j
= 0; i
>= 0; i
--) {
274 hex
[j
] = byte
& 0x0f;
291 return trace_seq_putmem(s
, hex
, j
);
295 trace_seq_reset(struct trace_seq
*s
)
301 trace_print_seq(struct seq_file
*m
, struct trace_seq
*s
)
303 int len
= s
->len
>= PAGE_SIZE
? PAGE_SIZE
- 1 : s
->len
;
306 seq_puts(m
, s
->buffer
);
312 flip_trace(struct trace_array_cpu
*tr1
, struct trace_array_cpu
*tr2
)
314 struct list_head flip_pages
;
316 INIT_LIST_HEAD(&flip_pages
);
318 memcpy(&tr1
->trace_head_idx
, &tr2
->trace_head_idx
,
319 sizeof(struct trace_array_cpu
) -
320 offsetof(struct trace_array_cpu
, trace_head_idx
));
324 list_splice_init(&tr1
->trace_pages
, &flip_pages
);
325 list_splice_init(&tr2
->trace_pages
, &tr1
->trace_pages
);
326 list_splice_init(&flip_pages
, &tr2
->trace_pages
);
327 BUG_ON(!list_empty(&flip_pages
));
333 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
335 struct trace_array_cpu
*data
;
338 WARN_ON_ONCE(!irqs_disabled());
339 __raw_spin_lock(&ftrace_max_lock
);
340 /* clear out all the previous traces */
341 for_each_possible_cpu(i
) {
343 flip_trace(max_tr
.data
[i
], data
);
347 __update_max_tr(tr
, tsk
, cpu
);
348 __raw_spin_unlock(&ftrace_max_lock
);
352 * update_max_tr_single - only copy one trace over, and reset the rest
354 * @tsk - task with the latency
355 * @cpu - the cpu of the buffer to copy.
358 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
360 struct trace_array_cpu
*data
= tr
->data
[cpu
];
363 WARN_ON_ONCE(!irqs_disabled());
364 __raw_spin_lock(&ftrace_max_lock
);
365 for_each_possible_cpu(i
)
366 tracing_reset(max_tr
.data
[i
]);
368 flip_trace(max_tr
.data
[cpu
], data
);
371 __update_max_tr(tr
, tsk
, cpu
);
372 __raw_spin_unlock(&ftrace_max_lock
);
375 int register_tracer(struct tracer
*type
)
382 pr_info("Tracer must have a name\n");
386 mutex_lock(&trace_types_lock
);
387 for (t
= trace_types
; t
; t
= t
->next
) {
388 if (strcmp(type
->name
, t
->name
) == 0) {
390 pr_info("Trace %s already registered\n",
397 #ifdef CONFIG_FTRACE_STARTUP_TEST
398 if (type
->selftest
) {
399 struct tracer
*saved_tracer
= current_trace
;
400 struct trace_array_cpu
*data
;
401 struct trace_array
*tr
= &global_trace
;
402 int saved_ctrl
= tr
->ctrl
;
405 * Run a selftest on this tracer.
406 * Here we reset the trace buffer, and set the current
407 * tracer to be this tracer. The tracer can then run some
408 * internal tracing to verify that everything is in order.
409 * If we fail, we do not register this tracer.
411 for_each_possible_cpu(i
) {
413 if (!head_page(data
))
417 current_trace
= type
;
419 /* the test is responsible for initializing and enabling */
420 pr_info("Testing tracer %s: ", type
->name
);
421 ret
= type
->selftest(type
, tr
);
422 /* the test is responsible for resetting too */
423 current_trace
= saved_tracer
;
424 tr
->ctrl
= saved_ctrl
;
426 printk(KERN_CONT
"FAILED!\n");
429 /* Only reset on passing, to avoid touching corrupted buffers */
430 for_each_possible_cpu(i
) {
432 if (!head_page(data
))
436 printk(KERN_CONT
"PASSED\n");
440 type
->next
= trace_types
;
442 len
= strlen(type
->name
);
443 if (len
> max_tracer_type_len
)
444 max_tracer_type_len
= len
;
447 mutex_unlock(&trace_types_lock
);
452 void unregister_tracer(struct tracer
*type
)
457 mutex_lock(&trace_types_lock
);
458 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
462 pr_info("Trace %s not registered\n", type
->name
);
467 if (strlen(type
->name
) != max_tracer_type_len
)
470 max_tracer_type_len
= 0;
471 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
472 len
= strlen((*t
)->name
);
473 if (len
> max_tracer_type_len
)
474 max_tracer_type_len
= len
;
477 mutex_unlock(&trace_types_lock
);
480 void tracing_reset(struct trace_array_cpu
*data
)
483 data
->trace_head
= data
->trace_tail
= head_page(data
);
484 data
->trace_head_idx
= 0;
485 data
->trace_tail_idx
= 0;
488 #define SAVED_CMDLINES 128
489 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
490 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
491 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
492 static int cmdline_idx
;
493 static DEFINE_SPINLOCK(trace_cmdline_lock
);
494 atomic_t trace_record_cmdline_disabled
;
496 static void trace_init_cmdlines(void)
498 memset(&map_pid_to_cmdline
, -1, sizeof(map_pid_to_cmdline
));
499 memset(&map_cmdline_to_pid
, -1, sizeof(map_cmdline_to_pid
));
503 void trace_stop_cmdline_recording(void);
505 static void trace_save_cmdline(struct task_struct
*tsk
)
510 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
514 * It's not the end of the world if we don't get
515 * the lock, but we also don't want to spin
516 * nor do we want to disable interrupts,
517 * so if we miss here, then better luck next time.
519 if (!spin_trylock(&trace_cmdline_lock
))
522 idx
= map_pid_to_cmdline
[tsk
->pid
];
523 if (idx
>= SAVED_CMDLINES
) {
524 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
526 map
= map_cmdline_to_pid
[idx
];
527 if (map
<= PID_MAX_DEFAULT
)
528 map_pid_to_cmdline
[map
] = (unsigned)-1;
530 map_pid_to_cmdline
[tsk
->pid
] = idx
;
535 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
537 spin_unlock(&trace_cmdline_lock
);
540 static char *trace_find_cmdline(int pid
)
542 char *cmdline
= "<...>";
548 if (pid
> PID_MAX_DEFAULT
)
551 map
= map_pid_to_cmdline
[pid
];
552 if (map
>= SAVED_CMDLINES
)
555 cmdline
= saved_cmdlines
[map
];
561 void tracing_record_cmdline(struct task_struct
*tsk
)
563 if (atomic_read(&trace_record_cmdline_disabled
))
566 trace_save_cmdline(tsk
);
569 static inline struct list_head
*
570 trace_next_list(struct trace_array_cpu
*data
, struct list_head
*next
)
573 * Roundrobin - but skip the head (which is not a real page):
576 if (unlikely(next
== &data
->trace_pages
))
578 BUG_ON(next
== &data
->trace_pages
);
584 trace_next_page(struct trace_array_cpu
*data
, void *addr
)
586 struct list_head
*next
;
589 page
= virt_to_page(addr
);
591 next
= trace_next_list(data
, &page
->lru
);
592 page
= list_entry(next
, struct page
, lru
);
594 return page_address(page
);
597 static inline struct trace_entry
*
598 tracing_get_trace_entry(struct trace_array
*tr
, struct trace_array_cpu
*data
)
600 unsigned long idx
, idx_next
;
601 struct trace_entry
*entry
;
604 idx
= data
->trace_head_idx
;
607 BUG_ON(idx
* TRACE_ENTRY_SIZE
>= PAGE_SIZE
);
609 entry
= data
->trace_head
+ idx
* TRACE_ENTRY_SIZE
;
611 if (unlikely(idx_next
>= ENTRIES_PER_PAGE
)) {
612 data
->trace_head
= trace_next_page(data
, data
->trace_head
);
616 if (data
->trace_head
== data
->trace_tail
&&
617 idx_next
== data
->trace_tail_idx
) {
619 data
->trace_tail_idx
++;
620 if (data
->trace_tail_idx
>= ENTRIES_PER_PAGE
) {
622 trace_next_page(data
, data
->trace_tail
);
623 data
->trace_tail_idx
= 0;
627 data
->trace_head_idx
= idx_next
;
633 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
)
635 struct task_struct
*tsk
= current
;
638 pc
= preempt_count();
640 entry
->preempt_count
= pc
& 0xff;
641 entry
->pid
= tsk
->pid
;
642 entry
->t
= ftrace_now(raw_smp_processor_id());
643 entry
->flags
= (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
644 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
645 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
646 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
650 trace_function(struct trace_array
*tr
, struct trace_array_cpu
*data
,
651 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
)
653 struct trace_entry
*entry
;
654 unsigned long irq_flags
;
656 raw_local_irq_save(irq_flags
);
657 __raw_spin_lock(&data
->lock
);
658 entry
= tracing_get_trace_entry(tr
, data
);
659 tracing_generic_entry_update(entry
, flags
);
660 entry
->type
= TRACE_FN
;
662 entry
->fn
.parent_ip
= parent_ip
;
663 __raw_spin_unlock(&data
->lock
);
664 raw_local_irq_restore(irq_flags
);
668 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
669 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
)
671 if (likely(!atomic_read(&data
->disabled
)))
672 trace_function(tr
, data
, ip
, parent_ip
, flags
);
676 __trace_special(void *__tr
, void *__data
,
677 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
679 struct trace_array_cpu
*data
= __data
;
680 struct trace_array
*tr
= __tr
;
681 struct trace_entry
*entry
;
682 unsigned long irq_flags
;
684 raw_local_irq_save(irq_flags
);
685 __raw_spin_lock(&data
->lock
);
686 entry
= tracing_get_trace_entry(tr
, data
);
687 tracing_generic_entry_update(entry
, 0);
688 entry
->type
= TRACE_SPECIAL
;
689 entry
->special
.arg1
= arg1
;
690 entry
->special
.arg2
= arg2
;
691 entry
->special
.arg3
= arg3
;
692 __raw_spin_unlock(&data
->lock
);
693 raw_local_irq_restore(irq_flags
);
698 void __trace_stack(struct trace_array
*tr
,
699 struct trace_array_cpu
*data
,
703 struct trace_entry
*entry
;
704 struct stack_trace trace
;
706 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
709 entry
= tracing_get_trace_entry(tr
, data
);
710 tracing_generic_entry_update(entry
, flags
);
711 entry
->type
= TRACE_STACK
;
713 memset(&entry
->stack
, 0, sizeof(entry
->stack
));
715 trace
.nr_entries
= 0;
716 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
718 trace
.entries
= entry
->stack
.caller
;
720 save_stack_trace(&trace
);
724 tracing_sched_switch_trace(struct trace_array
*tr
,
725 struct trace_array_cpu
*data
,
726 struct task_struct
*prev
,
727 struct task_struct
*next
,
730 struct trace_entry
*entry
;
731 unsigned long irq_flags
;
733 raw_local_irq_save(irq_flags
);
734 __raw_spin_lock(&data
->lock
);
735 entry
= tracing_get_trace_entry(tr
, data
);
736 tracing_generic_entry_update(entry
, flags
);
737 entry
->type
= TRACE_CTX
;
738 entry
->ctx
.prev_pid
= prev
->pid
;
739 entry
->ctx
.prev_prio
= prev
->prio
;
740 entry
->ctx
.prev_state
= prev
->state
;
741 entry
->ctx
.next_pid
= next
->pid
;
742 entry
->ctx
.next_prio
= next
->prio
;
743 entry
->ctx
.next_state
= next
->state
;
744 __trace_stack(tr
, data
, flags
, 4);
745 __raw_spin_unlock(&data
->lock
);
746 raw_local_irq_restore(irq_flags
);
750 tracing_sched_wakeup_trace(struct trace_array
*tr
,
751 struct trace_array_cpu
*data
,
752 struct task_struct
*wakee
,
753 struct task_struct
*curr
,
756 struct trace_entry
*entry
;
757 unsigned long irq_flags
;
759 raw_local_irq_save(irq_flags
);
760 __raw_spin_lock(&data
->lock
);
761 entry
= tracing_get_trace_entry(tr
, data
);
762 tracing_generic_entry_update(entry
, flags
);
763 entry
->type
= TRACE_WAKE
;
764 entry
->ctx
.prev_pid
= curr
->pid
;
765 entry
->ctx
.prev_prio
= curr
->prio
;
766 entry
->ctx
.prev_state
= curr
->state
;
767 entry
->ctx
.next_pid
= wakee
->pid
;
768 entry
->ctx
.next_prio
= wakee
->prio
;
769 entry
->ctx
.next_state
= wakee
->state
;
770 __trace_stack(tr
, data
, flags
, 5);
771 __raw_spin_unlock(&data
->lock
);
772 raw_local_irq_restore(irq_flags
);
779 function_trace_call(unsigned long ip
, unsigned long parent_ip
)
781 struct trace_array
*tr
= &global_trace
;
782 struct trace_array_cpu
*data
;
787 if (unlikely(!tracer_enabled
))
790 local_irq_save(flags
);
791 cpu
= raw_smp_processor_id();
792 data
= tr
->data
[cpu
];
793 disabled
= atomic_inc_return(&data
->disabled
);
795 if (likely(disabled
== 1))
796 trace_function(tr
, data
, ip
, parent_ip
, flags
);
798 atomic_dec(&data
->disabled
);
799 local_irq_restore(flags
);
802 static struct ftrace_ops trace_ops __read_mostly
=
804 .func
= function_trace_call
,
807 void tracing_start_function_trace(void)
809 register_ftrace_function(&trace_ops
);
812 void tracing_stop_function_trace(void)
814 unregister_ftrace_function(&trace_ops
);
818 enum trace_file_type
{
819 TRACE_FILE_LAT_FMT
= 1,
822 static struct trace_entry
*
823 trace_entry_idx(struct trace_array
*tr
, struct trace_array_cpu
*data
,
824 struct trace_iterator
*iter
, int cpu
)
827 struct trace_entry
*array
;
829 if (iter
->next_idx
[cpu
] >= tr
->entries
||
830 iter
->next_idx
[cpu
] >= data
->trace_idx
||
831 (data
->trace_head
== data
->trace_tail
&&
832 data
->trace_head_idx
== data
->trace_tail_idx
))
835 if (!iter
->next_page
[cpu
]) {
836 /* Initialize the iterator for this cpu trace buffer */
837 WARN_ON(!data
->trace_tail
);
838 page
= virt_to_page(data
->trace_tail
);
839 iter
->next_page
[cpu
] = &page
->lru
;
840 iter
->next_page_idx
[cpu
] = data
->trace_tail_idx
;
843 page
= list_entry(iter
->next_page
[cpu
], struct page
, lru
);
844 BUG_ON(&data
->trace_pages
== &page
->lru
);
846 array
= page_address(page
);
848 WARN_ON(iter
->next_page_idx
[cpu
] >= ENTRIES_PER_PAGE
);
849 return &array
[iter
->next_page_idx
[cpu
]];
852 static struct trace_entry
*
853 find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
)
855 struct trace_array
*tr
= iter
->tr
;
856 struct trace_entry
*ent
, *next
= NULL
;
860 for_each_possible_cpu(cpu
) {
861 if (!head_page(tr
->data
[cpu
]))
863 ent
= trace_entry_idx(tr
, tr
->data
[cpu
], iter
, cpu
);
865 * Pick the entry with the smallest timestamp:
867 if (ent
&& (!next
|| ent
->t
< next
->t
)) {
879 static void trace_iterator_increment(struct trace_iterator
*iter
)
882 iter
->next_idx
[iter
->cpu
]++;
883 iter
->next_page_idx
[iter
->cpu
]++;
885 if (iter
->next_page_idx
[iter
->cpu
] >= ENTRIES_PER_PAGE
) {
886 struct trace_array_cpu
*data
= iter
->tr
->data
[iter
->cpu
];
888 iter
->next_page_idx
[iter
->cpu
] = 0;
889 iter
->next_page
[iter
->cpu
] =
890 trace_next_list(data
, iter
->next_page
[iter
->cpu
]);
894 static void trace_consume(struct trace_iterator
*iter
)
896 struct trace_array_cpu
*data
= iter
->tr
->data
[iter
->cpu
];
898 data
->trace_tail_idx
++;
899 if (data
->trace_tail_idx
>= ENTRIES_PER_PAGE
) {
900 data
->trace_tail
= trace_next_page(data
, data
->trace_tail
);
901 data
->trace_tail_idx
= 0;
904 /* Check if we empty it, then reset the index */
905 if (data
->trace_head
== data
->trace_tail
&&
906 data
->trace_head_idx
== data
->trace_tail_idx
)
910 static void *find_next_entry_inc(struct trace_iterator
*iter
)
912 struct trace_entry
*next
;
915 next
= find_next_entry(iter
, &next_cpu
);
917 iter
->prev_ent
= iter
->ent
;
918 iter
->prev_cpu
= iter
->cpu
;
921 iter
->cpu
= next_cpu
;
924 trace_iterator_increment(iter
);
926 return next
? iter
: NULL
;
929 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
931 struct trace_iterator
*iter
= m
->private;
932 void *last_ent
= iter
->ent
;
938 /* can't go backwards */
943 ent
= find_next_entry_inc(iter
);
947 while (ent
&& iter
->idx
< i
)
948 ent
= find_next_entry_inc(iter
);
952 if (last_ent
&& !ent
)
953 seq_puts(m
, "\n\nvim:ft=help\n");
958 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
960 struct trace_iterator
*iter
= m
->private;
965 mutex_lock(&trace_types_lock
);
967 if (!current_trace
|| current_trace
!= iter
->trace
)
970 atomic_inc(&trace_record_cmdline_disabled
);
972 /* let the tracer grab locks here if needed */
973 if (current_trace
->start
)
974 current_trace
->start(iter
);
976 if (*pos
!= iter
->pos
) {
980 iter
->prev_ent
= NULL
;
983 for_each_possible_cpu(i
) {
984 iter
->next_idx
[i
] = 0;
985 iter
->next_page
[i
] = NULL
;
988 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
993 p
= s_next(m
, p
, &l
);
999 static void s_stop(struct seq_file
*m
, void *p
)
1001 struct trace_iterator
*iter
= m
->private;
1003 atomic_dec(&trace_record_cmdline_disabled
);
1005 /* let the tracer release locks here if needed */
1006 if (current_trace
&& current_trace
== iter
->trace
&& iter
->trace
->stop
)
1007 iter
->trace
->stop(iter
);
1009 mutex_unlock(&trace_types_lock
);
1013 seq_print_sym_short(struct trace_seq
*s
, const char *fmt
, unsigned long address
)
1015 #ifdef CONFIG_KALLSYMS
1016 char str
[KSYM_SYMBOL_LEN
];
1018 kallsyms_lookup(address
, NULL
, NULL
, NULL
, str
);
1020 return trace_seq_printf(s
, fmt
, str
);
1026 seq_print_sym_offset(struct trace_seq
*s
, const char *fmt
,
1027 unsigned long address
)
1029 #ifdef CONFIG_KALLSYMS
1030 char str
[KSYM_SYMBOL_LEN
];
1032 sprint_symbol(str
, address
);
1033 return trace_seq_printf(s
, fmt
, str
);
1038 #ifndef CONFIG_64BIT
1039 # define IP_FMT "%08lx"
1041 # define IP_FMT "%016lx"
1045 seq_print_ip_sym(struct trace_seq
*s
, unsigned long ip
, unsigned long sym_flags
)
1050 return trace_seq_printf(s
, "0");
1052 if (sym_flags
& TRACE_ITER_SYM_OFFSET
)
1053 ret
= seq_print_sym_offset(s
, "%s", ip
);
1055 ret
= seq_print_sym_short(s
, "%s", ip
);
1060 if (sym_flags
& TRACE_ITER_SYM_ADDR
)
1061 ret
= trace_seq_printf(s
, " <" IP_FMT
">", ip
);
1065 static void print_lat_help_header(struct seq_file
*m
)
1067 seq_puts(m
, "# _------=> CPU# \n");
1068 seq_puts(m
, "# / _-----=> irqs-off \n");
1069 seq_puts(m
, "# | / _----=> need-resched \n");
1070 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1071 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1072 seq_puts(m
, "# |||| / \n");
1073 seq_puts(m
, "# ||||| delay \n");
1074 seq_puts(m
, "# cmd pid ||||| time | caller \n");
1075 seq_puts(m
, "# \\ / ||||| \\ | / \n");
1078 static void print_func_help_header(struct seq_file
*m
)
1080 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1081 seq_puts(m
, "# | | | | |\n");
1086 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1088 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1089 struct trace_array
*tr
= iter
->tr
;
1090 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1091 struct tracer
*type
= current_trace
;
1092 unsigned long total
= 0;
1093 unsigned long entries
= 0;
1095 const char *name
= "preemption";
1100 for_each_possible_cpu(cpu
) {
1101 if (head_page(tr
->data
[cpu
])) {
1102 total
+= tr
->data
[cpu
]->trace_idx
;
1103 if (tr
->data
[cpu
]->trace_idx
> tr
->entries
)
1104 entries
+= tr
->entries
;
1106 entries
+= tr
->data
[cpu
]->trace_idx
;
1110 seq_printf(m
, "%s latency trace v1.1.5 on %s\n",
1112 seq_puts(m
, "-----------------------------------"
1113 "---------------------------------\n");
1114 seq_printf(m
, " latency: %lu us, #%lu/%lu, CPU#%d |"
1115 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1116 nsecs_to_usecs(data
->saved_latency
),
1120 #if defined(CONFIG_PREEMPT_NONE)
1122 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1124 #elif defined(CONFIG_PREEMPT_DESKTOP)
1129 /* These are reserved for later use */
1132 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1136 seq_puts(m
, " -----------------\n");
1137 seq_printf(m
, " | task: %.16s-%d "
1138 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1139 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1140 data
->policy
, data
->rt_priority
);
1141 seq_puts(m
, " -----------------\n");
1143 if (data
->critical_start
) {
1144 seq_puts(m
, " => started at: ");
1145 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1146 trace_print_seq(m
, &iter
->seq
);
1147 seq_puts(m
, "\n => ended at: ");
1148 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1149 trace_print_seq(m
, &iter
->seq
);
1157 lat_print_generic(struct trace_seq
*s
, struct trace_entry
*entry
, int cpu
)
1159 int hardirq
, softirq
;
1162 comm
= trace_find_cmdline(entry
->pid
);
1164 trace_seq_printf(s
, "%8.8s-%-5d ", comm
, entry
->pid
);
1165 trace_seq_printf(s
, "%d", cpu
);
1166 trace_seq_printf(s
, "%c%c",
1167 (entry
->flags
& TRACE_FLAG_IRQS_OFF
) ? 'd' : '.',
1168 ((entry
->flags
& TRACE_FLAG_NEED_RESCHED
) ? 'N' : '.'));
1170 hardirq
= entry
->flags
& TRACE_FLAG_HARDIRQ
;
1171 softirq
= entry
->flags
& TRACE_FLAG_SOFTIRQ
;
1172 if (hardirq
&& softirq
)
1173 trace_seq_putc(s
, 'H');
1176 trace_seq_putc(s
, 'h');
1179 trace_seq_putc(s
, 's');
1181 trace_seq_putc(s
, '.');
1185 if (entry
->preempt_count
)
1186 trace_seq_printf(s
, "%x", entry
->preempt_count
);
1188 trace_seq_puts(s
, ".");
1191 unsigned long preempt_mark_thresh
= 100;
1194 lat_print_timestamp(struct trace_seq
*s
, unsigned long long abs_usecs
,
1195 unsigned long rel_usecs
)
1197 trace_seq_printf(s
, " %4lldus", abs_usecs
);
1198 if (rel_usecs
> preempt_mark_thresh
)
1199 trace_seq_puts(s
, "!: ");
1200 else if (rel_usecs
> 1)
1201 trace_seq_puts(s
, "+: ");
1203 trace_seq_puts(s
, " : ");
1206 static const char state_to_char
[] = TASK_STATE_TO_CHAR_STR
;
1209 print_lat_fmt(struct trace_iterator
*iter
, unsigned int trace_idx
, int cpu
)
1211 struct trace_seq
*s
= &iter
->seq
;
1212 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1213 struct trace_entry
*next_entry
= find_next_entry(iter
, NULL
);
1214 unsigned long verbose
= (trace_flags
& TRACE_ITER_VERBOSE
);
1215 struct trace_entry
*entry
= iter
->ent
;
1216 unsigned long abs_usecs
;
1217 unsigned long rel_usecs
;
1224 rel_usecs
= ns2usecs(next_entry
->t
- entry
->t
);
1225 abs_usecs
= ns2usecs(entry
->t
- iter
->tr
->time_start
);
1228 comm
= trace_find_cmdline(entry
->pid
);
1229 trace_seq_printf(s
, "%16s %5d %d %d %08x %08x [%08lx]"
1230 " %ld.%03ldms (+%ld.%03ldms): ",
1232 entry
->pid
, cpu
, entry
->flags
,
1233 entry
->preempt_count
, trace_idx
,
1236 abs_usecs
% 1000, rel_usecs
/1000,
1239 lat_print_generic(s
, entry
, cpu
);
1240 lat_print_timestamp(s
, abs_usecs
, rel_usecs
);
1242 switch (entry
->type
) {
1244 seq_print_ip_sym(s
, entry
->fn
.ip
, sym_flags
);
1245 trace_seq_puts(s
, " (");
1246 seq_print_ip_sym(s
, entry
->fn
.parent_ip
, sym_flags
);
1247 trace_seq_puts(s
, ")\n");
1251 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1252 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1253 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1254 state_to_char
[entry
->ctx
.next_state
] : 'X';
1256 comm
= trace_find_cmdline(entry
->ctx
.next_pid
);
1257 trace_seq_printf(s
, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1258 entry
->ctx
.prev_pid
,
1259 entry
->ctx
.prev_prio
,
1260 S
, entry
->type
== TRACE_CTX
? "==>" : " +",
1261 entry
->ctx
.next_pid
,
1262 entry
->ctx
.next_prio
,
1266 trace_seq_printf(s
, "# %ld %ld %ld\n",
1267 entry
->special
.arg1
,
1268 entry
->special
.arg2
,
1269 entry
->special
.arg3
);
1272 for (i
= 0; i
< FTRACE_STACK_ENTRIES
; i
++) {
1274 trace_seq_puts(s
, " <= ");
1275 seq_print_ip_sym(s
, entry
->stack
.caller
[i
], sym_flags
);
1277 trace_seq_puts(s
, "\n");
1280 trace_seq_printf(s
, "Unknown type %d\n", entry
->type
);
1285 static int print_trace_fmt(struct trace_iterator
*iter
)
1287 struct trace_seq
*s
= &iter
->seq
;
1288 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1289 struct trace_entry
*entry
;
1290 unsigned long usec_rem
;
1291 unsigned long long t
;
1300 comm
= trace_find_cmdline(iter
->ent
->pid
);
1302 t
= ns2usecs(entry
->t
);
1303 usec_rem
= do_div(t
, 1000000ULL);
1304 secs
= (unsigned long)t
;
1306 ret
= trace_seq_printf(s
, "%16s-%-5d ", comm
, entry
->pid
);
1309 ret
= trace_seq_printf(s
, "[%02d] ", iter
->cpu
);
1312 ret
= trace_seq_printf(s
, "%5lu.%06lu: ", secs
, usec_rem
);
1316 switch (entry
->type
) {
1318 ret
= seq_print_ip_sym(s
, entry
->fn
.ip
, sym_flags
);
1321 if ((sym_flags
& TRACE_ITER_PRINT_PARENT
) &&
1322 entry
->fn
.parent_ip
) {
1323 ret
= trace_seq_printf(s
, " <-");
1326 ret
= seq_print_ip_sym(s
, entry
->fn
.parent_ip
,
1331 ret
= trace_seq_printf(s
, "\n");
1337 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1338 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1339 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1340 state_to_char
[entry
->ctx
.next_state
] : 'X';
1341 ret
= trace_seq_printf(s
, " %5d:%3d:%c %s %5d:%3d:%c\n",
1342 entry
->ctx
.prev_pid
,
1343 entry
->ctx
.prev_prio
,
1345 entry
->type
== TRACE_CTX
? "==>" : " +",
1346 entry
->ctx
.next_pid
,
1347 entry
->ctx
.next_prio
,
1353 ret
= trace_seq_printf(s
, "# %ld %ld %ld\n",
1354 entry
->special
.arg1
,
1355 entry
->special
.arg2
,
1356 entry
->special
.arg3
);
1361 for (i
= 0; i
< FTRACE_STACK_ENTRIES
; i
++) {
1363 ret
= trace_seq_puts(s
, " <= ");
1367 ret
= seq_print_ip_sym(s
, entry
->stack
.caller
[i
],
1372 ret
= trace_seq_puts(s
, "\n");
1380 static int print_raw_fmt(struct trace_iterator
*iter
)
1382 struct trace_seq
*s
= &iter
->seq
;
1383 struct trace_entry
*entry
;
1389 ret
= trace_seq_printf(s
, "%d %d %llu ",
1390 entry
->pid
, iter
->cpu
, entry
->t
);
1394 switch (entry
->type
) {
1396 ret
= trace_seq_printf(s
, "%x %x\n",
1397 entry
->fn
.ip
, entry
->fn
.parent_ip
);
1403 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1404 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1405 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1406 state_to_char
[entry
->ctx
.next_state
] : 'X';
1407 if (entry
->type
== TRACE_WAKE
)
1409 ret
= trace_seq_printf(s
, "%d %d %c %d %d %c\n",
1410 entry
->ctx
.prev_pid
,
1411 entry
->ctx
.prev_prio
,
1413 entry
->ctx
.next_pid
,
1414 entry
->ctx
.next_prio
,
1421 ret
= trace_seq_printf(s
, "# %ld %ld %ld\n",
1422 entry
->special
.arg1
,
1423 entry
->special
.arg2
,
1424 entry
->special
.arg3
);
1432 #define SEQ_PUT_FIELD_RET(s, x) \
1434 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1438 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1440 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1444 static int print_hex_fmt(struct trace_iterator
*iter
)
1446 struct trace_seq
*s
= &iter
->seq
;
1447 unsigned char newline
= '\n';
1448 struct trace_entry
*entry
;
1453 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1454 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1455 SEQ_PUT_HEX_FIELD_RET(s
, entry
->t
);
1457 switch (entry
->type
) {
1459 SEQ_PUT_HEX_FIELD_RET(s
, entry
->fn
.ip
);
1460 SEQ_PUT_HEX_FIELD_RET(s
, entry
->fn
.parent_ip
);
1464 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1465 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1466 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1467 state_to_char
[entry
->ctx
.next_state
] : 'X';
1468 if (entry
->type
== TRACE_WAKE
)
1470 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.prev_pid
);
1471 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.prev_prio
);
1472 SEQ_PUT_HEX_FIELD_RET(s
, S
);
1473 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.next_pid
);
1474 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.next_prio
);
1475 SEQ_PUT_HEX_FIELD_RET(s
, entry
->fn
.parent_ip
);
1476 SEQ_PUT_HEX_FIELD_RET(s
, T
);
1480 SEQ_PUT_HEX_FIELD_RET(s
, entry
->special
.arg1
);
1481 SEQ_PUT_HEX_FIELD_RET(s
, entry
->special
.arg2
);
1482 SEQ_PUT_HEX_FIELD_RET(s
, entry
->special
.arg3
);
1485 SEQ_PUT_FIELD_RET(s
, newline
);
1490 static int print_bin_fmt(struct trace_iterator
*iter
)
1492 struct trace_seq
*s
= &iter
->seq
;
1493 struct trace_entry
*entry
;
1497 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1498 SEQ_PUT_FIELD_RET(s
, entry
->cpu
);
1499 SEQ_PUT_FIELD_RET(s
, entry
->t
);
1501 switch (entry
->type
) {
1503 SEQ_PUT_FIELD_RET(s
, entry
->fn
.ip
);
1504 SEQ_PUT_FIELD_RET(s
, entry
->fn
.parent_ip
);
1507 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_pid
);
1508 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_prio
);
1509 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_state
);
1510 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_pid
);
1511 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_prio
);
1512 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_state
);
1516 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg1
);
1517 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg2
);
1518 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg3
);
1524 static int trace_empty(struct trace_iterator
*iter
)
1526 struct trace_array_cpu
*data
;
1529 for_each_possible_cpu(cpu
) {
1530 data
= iter
->tr
->data
[cpu
];
1532 if (head_page(data
) && data
->trace_idx
&&
1533 (data
->trace_tail
!= data
->trace_head
||
1534 data
->trace_tail_idx
!= data
->trace_head_idx
))
1540 static int print_trace_line(struct trace_iterator
*iter
)
1542 if (trace_flags
& TRACE_ITER_BIN
)
1543 return print_bin_fmt(iter
);
1545 if (trace_flags
& TRACE_ITER_HEX
)
1546 return print_hex_fmt(iter
);
1548 if (trace_flags
& TRACE_ITER_RAW
)
1549 return print_raw_fmt(iter
);
1551 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
1552 return print_lat_fmt(iter
, iter
->idx
, iter
->cpu
);
1554 return print_trace_fmt(iter
);
1557 static int s_show(struct seq_file
*m
, void *v
)
1559 struct trace_iterator
*iter
= v
;
1561 if (iter
->ent
== NULL
) {
1563 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
1566 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1567 /* print nothing if the buffers are empty */
1568 if (trace_empty(iter
))
1570 print_trace_header(m
, iter
);
1571 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1572 print_lat_help_header(m
);
1574 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1575 print_func_help_header(m
);
1578 print_trace_line(iter
);
1579 trace_print_seq(m
, &iter
->seq
);
1585 static struct seq_operations tracer_seq_ops
= {
1592 static struct trace_iterator
*
1593 __tracing_open(struct inode
*inode
, struct file
*file
, int *ret
)
1595 struct trace_iterator
*iter
;
1597 if (tracing_disabled
) {
1602 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
1608 mutex_lock(&trace_types_lock
);
1609 if (current_trace
&& current_trace
->print_max
)
1612 iter
->tr
= inode
->i_private
;
1613 iter
->trace
= current_trace
;
1616 /* TODO stop tracer */
1617 *ret
= seq_open(file
, &tracer_seq_ops
);
1619 struct seq_file
*m
= file
->private_data
;
1622 /* stop the trace while dumping */
1626 if (iter
->trace
&& iter
->trace
->open
)
1627 iter
->trace
->open(iter
);
1632 mutex_unlock(&trace_types_lock
);
1638 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
1640 if (tracing_disabled
)
1643 filp
->private_data
= inode
->i_private
;
1647 int tracing_release(struct inode
*inode
, struct file
*file
)
1649 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
1650 struct trace_iterator
*iter
= m
->private;
1652 mutex_lock(&trace_types_lock
);
1653 if (iter
->trace
&& iter
->trace
->close
)
1654 iter
->trace
->close(iter
);
1656 /* reenable tracing if it was previously enabled */
1659 mutex_unlock(&trace_types_lock
);
1661 seq_release(inode
, file
);
1666 static int tracing_open(struct inode
*inode
, struct file
*file
)
1670 __tracing_open(inode
, file
, &ret
);
1675 static int tracing_lt_open(struct inode
*inode
, struct file
*file
)
1677 struct trace_iterator
*iter
;
1680 iter
= __tracing_open(inode
, file
, &ret
);
1683 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
1690 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1692 struct tracer
*t
= m
->private;
1704 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
1706 struct tracer
*t
= m
->private;
1709 mutex_lock(&trace_types_lock
);
1710 for (; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
1716 static void t_stop(struct seq_file
*m
, void *p
)
1718 mutex_unlock(&trace_types_lock
);
1721 static int t_show(struct seq_file
*m
, void *v
)
1723 struct tracer
*t
= v
;
1728 seq_printf(m
, "%s", t
->name
);
1737 static struct seq_operations show_traces_seq_ops
= {
1744 static int show_traces_open(struct inode
*inode
, struct file
*file
)
1748 if (tracing_disabled
)
1751 ret
= seq_open(file
, &show_traces_seq_ops
);
1753 struct seq_file
*m
= file
->private_data
;
1754 m
->private = trace_types
;
1760 static struct file_operations tracing_fops
= {
1761 .open
= tracing_open
,
1763 .llseek
= seq_lseek
,
1764 .release
= tracing_release
,
1767 static struct file_operations tracing_lt_fops
= {
1768 .open
= tracing_lt_open
,
1770 .llseek
= seq_lseek
,
1771 .release
= tracing_release
,
1774 static struct file_operations show_traces_fops
= {
1775 .open
= show_traces_open
,
1777 .release
= seq_release
,
1781 * Only trace on a CPU if the bitmask is set:
1783 static cpumask_t tracing_cpumask
= CPU_MASK_ALL
;
1786 * When tracing/tracing_cpu_mask is modified then this holds
1787 * the new bitmask we are about to install:
1789 static cpumask_t tracing_cpumask_new
;
1792 * The tracer itself will not take this lock, but still we want
1793 * to provide a consistent cpumask to user-space:
1795 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
1798 * Temporary storage for the character representation of the
1799 * CPU bitmask (and one more byte for the newline):
1801 static char mask_str
[NR_CPUS
+ 1];
1804 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
1805 size_t count
, loff_t
*ppos
)
1809 mutex_lock(&tracing_cpumask_update_lock
);
1811 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
1812 if (count
- len
< 2) {
1816 len
+= sprintf(mask_str
+ len
, "\n");
1817 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
1820 mutex_unlock(&tracing_cpumask_update_lock
);
1826 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
1827 size_t count
, loff_t
*ppos
)
1831 mutex_lock(&tracing_cpumask_update_lock
);
1832 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
1836 raw_local_irq_disable();
1837 __raw_spin_lock(&ftrace_max_lock
);
1838 for_each_possible_cpu(cpu
) {
1840 * Increase/decrease the disabled counter if we are
1841 * about to flip a bit in the cpumask:
1843 if (cpu_isset(cpu
, tracing_cpumask
) &&
1844 !cpu_isset(cpu
, tracing_cpumask_new
)) {
1845 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
1847 if (!cpu_isset(cpu
, tracing_cpumask
) &&
1848 cpu_isset(cpu
, tracing_cpumask_new
)) {
1849 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
1852 __raw_spin_unlock(&ftrace_max_lock
);
1853 raw_local_irq_enable();
1855 tracing_cpumask
= tracing_cpumask_new
;
1857 mutex_unlock(&tracing_cpumask_update_lock
);
1862 mutex_unlock(&tracing_cpumask_update_lock
);
1867 static struct file_operations tracing_cpumask_fops
= {
1868 .open
= tracing_open_generic
,
1869 .read
= tracing_cpumask_read
,
1870 .write
= tracing_cpumask_write
,
1874 tracing_iter_ctrl_read(struct file
*filp
, char __user
*ubuf
,
1875 size_t cnt
, loff_t
*ppos
)
1882 /* calulate max size */
1883 for (i
= 0; trace_options
[i
]; i
++) {
1884 len
+= strlen(trace_options
[i
]);
1885 len
+= 3; /* "no" and space */
1888 /* +2 for \n and \0 */
1889 buf
= kmalloc(len
+ 2, GFP_KERNEL
);
1893 for (i
= 0; trace_options
[i
]; i
++) {
1894 if (trace_flags
& (1 << i
))
1895 r
+= sprintf(buf
+ r
, "%s ", trace_options
[i
]);
1897 r
+= sprintf(buf
+ r
, "no%s ", trace_options
[i
]);
1900 r
+= sprintf(buf
+ r
, "\n");
1901 WARN_ON(r
>= len
+ 2);
1903 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
1911 tracing_iter_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
1912 size_t cnt
, loff_t
*ppos
)
1922 if (copy_from_user(&buf
, ubuf
, cnt
))
1927 if (strncmp(buf
, "no", 2) == 0) {
1932 for (i
= 0; trace_options
[i
]; i
++) {
1933 int len
= strlen(trace_options
[i
]);
1935 if (strncmp(cmp
, trace_options
[i
], len
) == 0) {
1937 trace_flags
&= ~(1 << i
);
1939 trace_flags
|= (1 << i
);
1944 * If no option could be set, return an error:
1946 if (!trace_options
[i
])
1954 static struct file_operations tracing_iter_fops
= {
1955 .open
= tracing_open_generic
,
1956 .read
= tracing_iter_ctrl_read
,
1957 .write
= tracing_iter_ctrl_write
,
1960 static const char readme_msg
[] =
1961 "tracing mini-HOWTO:\n\n"
1963 "# mount -t debugfs nodev /debug\n\n"
1964 "# cat /debug/tracing/available_tracers\n"
1965 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
1966 "# cat /debug/tracing/current_tracer\n"
1968 "# echo sched_switch > /debug/tracing/current_tracer\n"
1969 "# cat /debug/tracing/current_tracer\n"
1971 "# cat /debug/tracing/iter_ctrl\n"
1972 "noprint-parent nosym-offset nosym-addr noverbose\n"
1973 "# echo print-parent > /debug/tracing/iter_ctrl\n"
1974 "# echo 1 > /debug/tracing/tracing_enabled\n"
1975 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
1976 "echo 0 > /debug/tracing/tracing_enabled\n"
1980 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
1981 size_t cnt
, loff_t
*ppos
)
1983 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
1984 readme_msg
, strlen(readme_msg
));
1987 static struct file_operations tracing_readme_fops
= {
1988 .open
= tracing_open_generic
,
1989 .read
= tracing_readme_read
,
1993 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
1994 size_t cnt
, loff_t
*ppos
)
1996 struct trace_array
*tr
= filp
->private_data
;
2000 r
= sprintf(buf
, "%ld\n", tr
->ctrl
);
2001 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2005 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2006 size_t cnt
, loff_t
*ppos
)
2008 struct trace_array
*tr
= filp
->private_data
;
2015 if (copy_from_user(&buf
, ubuf
, cnt
))
2020 val
= simple_strtoul(buf
, NULL
, 10);
2024 mutex_lock(&trace_types_lock
);
2025 if (tr
->ctrl
^ val
) {
2033 if (current_trace
&& current_trace
->ctrl_update
)
2034 current_trace
->ctrl_update(tr
);
2036 mutex_unlock(&trace_types_lock
);
2044 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2045 size_t cnt
, loff_t
*ppos
)
2047 char buf
[max_tracer_type_len
+2];
2050 mutex_lock(&trace_types_lock
);
2052 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2054 r
= sprintf(buf
, "\n");
2055 mutex_unlock(&trace_types_lock
);
2057 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2061 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2062 size_t cnt
, loff_t
*ppos
)
2064 struct trace_array
*tr
= &global_trace
;
2066 char buf
[max_tracer_type_len
+1];
2069 if (cnt
> max_tracer_type_len
)
2070 cnt
= max_tracer_type_len
;
2072 if (copy_from_user(&buf
, ubuf
, cnt
))
2077 /* strip ending whitespace. */
2078 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2081 mutex_lock(&trace_types_lock
);
2082 for (t
= trace_types
; t
; t
= t
->next
) {
2083 if (strcmp(t
->name
, buf
) == 0)
2086 if (!t
|| t
== current_trace
)
2089 if (current_trace
&& current_trace
->reset
)
2090 current_trace
->reset(tr
);
2097 mutex_unlock(&trace_types_lock
);
2105 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2106 size_t cnt
, loff_t
*ppos
)
2108 unsigned long *ptr
= filp
->private_data
;
2112 r
= snprintf(buf
, 64, "%ld\n",
2113 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2116 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2120 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2121 size_t cnt
, loff_t
*ppos
)
2123 long *ptr
= filp
->private_data
;
2130 if (copy_from_user(&buf
, ubuf
, cnt
))
2135 val
= simple_strtoul(buf
, NULL
, 10);
2142 static atomic_t tracing_reader
;
2144 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2146 struct trace_iterator
*iter
;
2148 if (tracing_disabled
)
2151 /* We only allow for reader of the pipe */
2152 if (atomic_inc_return(&tracing_reader
) != 1) {
2153 atomic_dec(&tracing_reader
);
2157 /* create a buffer to store the information to pass to userspace */
2158 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2162 iter
->tr
= &global_trace
;
2164 filp
->private_data
= iter
;
2169 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
2171 struct trace_iterator
*iter
= file
->private_data
;
2174 atomic_dec(&tracing_reader
);
2180 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
2182 struct trace_iterator
*iter
= filp
->private_data
;
2184 if (trace_flags
& TRACE_ITER_BLOCK
) {
2186 * Always select as readable when in blocking mode
2188 return POLLIN
| POLLRDNORM
;
2191 if (!trace_empty(iter
))
2192 return POLLIN
| POLLRDNORM
;
2193 poll_wait(filp
, &trace_wait
, poll_table
);
2194 if (!trace_empty(iter
))
2195 return POLLIN
| POLLRDNORM
;
2205 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
2206 size_t cnt
, loff_t
*ppos
)
2208 struct trace_iterator
*iter
= filp
->private_data
;
2209 struct trace_array_cpu
*data
;
2210 static cpumask_t mask
;
2212 unsigned long flags
;
2213 #ifdef CONFIG_FTRACE
2221 /* return any leftover data */
2222 if (iter
->seq
.len
> start
) {
2223 len
= iter
->seq
.len
- start
;
2226 ret
= copy_to_user(ubuf
, iter
->seq
.buffer
+ start
, cnt
);
2235 trace_seq_reset(&iter
->seq
);
2238 while (trace_empty(iter
)) {
2239 if (!(trace_flags
& TRACE_ITER_BLOCK
))
2240 return -EWOULDBLOCK
;
2242 * This is a make-shift waitqueue. The reason we don't use
2243 * an actual wait queue is because:
2244 * 1) we only ever have one waiter
2245 * 2) the tracing, traces all functions, we don't want
2246 * the overhead of calling wake_up and friends
2247 * (and tracing them too)
2248 * Anyway, this is really very primitive wakeup.
2250 set_current_state(TASK_INTERRUPTIBLE
);
2251 iter
->tr
->waiter
= current
;
2253 /* sleep for one second, and try again. */
2254 schedule_timeout(HZ
);
2256 iter
->tr
->waiter
= NULL
;
2258 if (signal_pending(current
))
2262 * We block until we read something and tracing is disabled.
2263 * We still block if tracing is disabled, but we have never
2264 * read anything. This allows a user to cat this file, and
2265 * then enable tracing. But after we have read something,
2266 * we give an EOF when tracing is again disabled.
2268 * iter->pos will be 0 if we haven't read anything.
2270 if (!tracer_enabled
&& iter
->pos
)
2276 /* stop when tracing is finished */
2277 if (trace_empty(iter
))
2280 if (cnt
>= PAGE_SIZE
)
2281 cnt
= PAGE_SIZE
- 1;
2283 memset(iter
, 0, sizeof(*iter
));
2284 iter
->tr
= &global_trace
;
2288 * We need to stop all tracing on all CPUS to read the
2289 * the next buffer. This is a bit expensive, but is
2290 * not done often. We fill all what we can read,
2291 * and then release the locks again.
2295 local_irq_save(flags
);
2296 #ifdef CONFIG_FTRACE
2297 ftrace_save
= ftrace_enabled
;
2301 for_each_possible_cpu(cpu
) {
2302 data
= iter
->tr
->data
[cpu
];
2304 if (!head_page(data
) || !data
->trace_idx
)
2307 atomic_inc(&data
->disabled
);
2311 for_each_cpu_mask(cpu
, mask
) {
2312 data
= iter
->tr
->data
[cpu
];
2313 __raw_spin_lock(&data
->lock
);
2316 while (find_next_entry_inc(iter
) != NULL
) {
2317 int len
= iter
->seq
.len
;
2319 ret
= print_trace_line(iter
);
2321 /* don't print partial lines */
2322 iter
->seq
.len
= len
;
2326 trace_consume(iter
);
2328 if (iter
->seq
.len
>= cnt
)
2332 for_each_cpu_mask(cpu
, mask
) {
2333 data
= iter
->tr
->data
[cpu
];
2334 __raw_spin_unlock(&data
->lock
);
2337 for_each_cpu_mask(cpu
, mask
) {
2338 data
= iter
->tr
->data
[cpu
];
2339 atomic_dec(&data
->disabled
);
2341 #ifdef CONFIG_FTRACE
2342 ftrace_enabled
= ftrace_save
;
2344 local_irq_restore(flags
);
2346 /* Now copy what we have to the user */
2347 read
= iter
->seq
.len
;
2351 ret
= copy_to_user(ubuf
, iter
->seq
.buffer
, read
);
2353 if (read
< iter
->seq
.len
)
2356 trace_seq_reset(&iter
->seq
);
2364 static struct file_operations tracing_max_lat_fops
= {
2365 .open
= tracing_open_generic
,
2366 .read
= tracing_max_lat_read
,
2367 .write
= tracing_max_lat_write
,
2370 static struct file_operations tracing_ctrl_fops
= {
2371 .open
= tracing_open_generic
,
2372 .read
= tracing_ctrl_read
,
2373 .write
= tracing_ctrl_write
,
2376 static struct file_operations set_tracer_fops
= {
2377 .open
= tracing_open_generic
,
2378 .read
= tracing_set_trace_read
,
2379 .write
= tracing_set_trace_write
,
2382 static struct file_operations tracing_pipe_fops
= {
2383 .open
= tracing_open_pipe
,
2384 .poll
= tracing_poll_pipe
,
2385 .read
= tracing_read_pipe
,
2386 .release
= tracing_release_pipe
,
2389 #ifdef CONFIG_DYNAMIC_FTRACE
2392 tracing_read_long(struct file
*filp
, char __user
*ubuf
,
2393 size_t cnt
, loff_t
*ppos
)
2395 unsigned long *p
= filp
->private_data
;
2399 r
= sprintf(buf
, "%ld\n", *p
);
2401 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2404 static struct file_operations tracing_read_long_fops
= {
2405 .open
= tracing_open_generic
,
2406 .read
= tracing_read_long
,
2410 static struct dentry
*d_tracer
;
2412 struct dentry
*tracing_init_dentry(void)
2419 d_tracer
= debugfs_create_dir("tracing", NULL
);
2421 if (!d_tracer
&& !once
) {
2423 pr_warning("Could not create debugfs directory 'tracing'\n");
2430 #ifdef CONFIG_FTRACE_SELFTEST
2431 /* Let selftest have access to static functions in this file */
2432 #include "trace_selftest.c"
2435 static __init
void tracer_init_debugfs(void)
2437 struct dentry
*d_tracer
;
2438 struct dentry
*entry
;
2440 d_tracer
= tracing_init_dentry();
2442 entry
= debugfs_create_file("tracing_enabled", 0644, d_tracer
,
2443 &global_trace
, &tracing_ctrl_fops
);
2445 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2447 entry
= debugfs_create_file("iter_ctrl", 0644, d_tracer
,
2448 NULL
, &tracing_iter_fops
);
2450 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2452 entry
= debugfs_create_file("tracing_cpumask", 0644, d_tracer
,
2453 NULL
, &tracing_cpumask_fops
);
2455 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2457 entry
= debugfs_create_file("latency_trace", 0444, d_tracer
,
2458 &global_trace
, &tracing_lt_fops
);
2460 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2462 entry
= debugfs_create_file("trace", 0444, d_tracer
,
2463 &global_trace
, &tracing_fops
);
2465 pr_warning("Could not create debugfs 'trace' entry\n");
2467 entry
= debugfs_create_file("available_tracers", 0444, d_tracer
,
2468 &global_trace
, &show_traces_fops
);
2470 pr_warning("Could not create debugfs 'trace' entry\n");
2472 entry
= debugfs_create_file("current_tracer", 0444, d_tracer
,
2473 &global_trace
, &set_tracer_fops
);
2475 pr_warning("Could not create debugfs 'trace' entry\n");
2477 entry
= debugfs_create_file("tracing_max_latency", 0644, d_tracer
,
2478 &tracing_max_latency
,
2479 &tracing_max_lat_fops
);
2481 pr_warning("Could not create debugfs "
2482 "'tracing_max_latency' entry\n");
2484 entry
= debugfs_create_file("tracing_thresh", 0644, d_tracer
,
2485 &tracing_thresh
, &tracing_max_lat_fops
);
2487 pr_warning("Could not create debugfs "
2488 "'tracing_threash' entry\n");
2489 entry
= debugfs_create_file("README", 0644, d_tracer
,
2490 NULL
, &tracing_readme_fops
);
2492 pr_warning("Could not create debugfs 'README' entry\n");
2494 entry
= debugfs_create_file("trace_pipe", 0644, d_tracer
,
2495 NULL
, &tracing_pipe_fops
);
2497 pr_warning("Could not create debugfs "
2498 "'tracing_threash' entry\n");
2500 #ifdef CONFIG_DYNAMIC_FTRACE
2501 entry
= debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
2502 &ftrace_update_tot_cnt
,
2503 &tracing_read_long_fops
);
2505 pr_warning("Could not create debugfs "
2506 "'dyn_ftrace_total_info' entry\n");
2510 /* dummy trace to disable tracing */
2511 static struct tracer no_tracer __read_mostly
=
2516 static int trace_alloc_page(void)
2518 struct trace_array_cpu
*data
;
2519 struct page
*page
, *tmp
;
2524 /* first allocate a page for each CPU */
2525 for_each_possible_cpu(i
) {
2526 array
= (void *)__get_free_page(GFP_KERNEL
);
2527 if (array
== NULL
) {
2528 printk(KERN_ERR
"tracer: failed to allocate page"
2529 "for trace buffer!\n");
2533 page
= virt_to_page(array
);
2534 list_add(&page
->lru
, &pages
);
2536 /* Only allocate if we are actually using the max trace */
2537 #ifdef CONFIG_TRACER_MAX_TRACE
2538 array
= (void *)__get_free_page(GFP_KERNEL
);
2539 if (array
== NULL
) {
2540 printk(KERN_ERR
"tracer: failed to allocate page"
2541 "for trace buffer!\n");
2544 page
= virt_to_page(array
);
2545 list_add(&page
->lru
, &pages
);
2549 /* Now that we successfully allocate a page per CPU, add them */
2550 for_each_possible_cpu(i
) {
2551 data
= global_trace
.data
[i
];
2552 data
->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
2553 page
= list_entry(pages
.next
, struct page
, lru
);
2554 list_del_init(&page
->lru
);
2555 list_add_tail(&page
->lru
, &data
->trace_pages
);
2558 #ifdef CONFIG_TRACER_MAX_TRACE
2559 data
= max_tr
.data
[i
];
2560 data
->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
2561 page
= list_entry(pages
.next
, struct page
, lru
);
2562 list_del_init(&page
->lru
);
2563 list_add_tail(&page
->lru
, &data
->trace_pages
);
2567 global_trace
.entries
+= ENTRIES_PER_PAGE
;
2572 list_for_each_entry_safe(page
, tmp
, &pages
, lru
) {
2573 list_del_init(&page
->lru
);
2579 __init
static int tracer_alloc_buffers(void)
2581 struct trace_array_cpu
*data
;
2588 global_trace
.ctrl
= tracer_enabled
;
2590 /* Allocate the first page for all buffers */
2591 for_each_possible_cpu(i
) {
2592 data
= global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
2593 max_tr
.data
[i
] = &per_cpu(max_data
, i
);
2595 array
= (void *)__get_free_page(GFP_KERNEL
);
2596 if (array
== NULL
) {
2597 printk(KERN_ERR
"tracer: failed to allocate page"
2598 "for trace buffer!\n");
2602 /* set the array to the list */
2603 INIT_LIST_HEAD(&data
->trace_pages
);
2604 page
= virt_to_page(array
);
2605 list_add(&page
->lru
, &data
->trace_pages
);
2606 /* use the LRU flag to differentiate the two buffers */
2609 /* Only allocate if we are actually using the max trace */
2610 #ifdef CONFIG_TRACER_MAX_TRACE
2611 array
= (void *)__get_free_page(GFP_KERNEL
);
2612 if (array
== NULL
) {
2613 printk(KERN_ERR
"tracer: failed to allocate page"
2614 "for trace buffer!\n");
2618 INIT_LIST_HEAD(&max_tr
.data
[i
]->trace_pages
);
2619 page
= virt_to_page(array
);
2620 list_add(&page
->lru
, &max_tr
.data
[i
]->trace_pages
);
2626 * Since we allocate by orders of pages, we may be able to
2629 global_trace
.entries
= ENTRIES_PER_PAGE
;
2632 while (global_trace
.entries
< trace_nr_entries
) {
2633 if (trace_alloc_page())
2637 max_tr
.entries
= global_trace
.entries
;
2639 pr_info("tracer: %d pages allocated for %ld",
2640 pages
, trace_nr_entries
);
2641 pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE
);
2642 pr_info(" actual entries %ld\n", global_trace
.entries
);
2644 tracer_init_debugfs();
2646 trace_init_cmdlines();
2648 register_tracer(&no_tracer
);
2649 current_trace
= &no_tracer
;
2651 /* All seems OK, enable tracing */
2652 tracing_disabled
= 0;
2657 for (i
-- ; i
>= 0; i
--) {
2658 struct page
*page
, *tmp
;
2659 struct trace_array_cpu
*data
= global_trace
.data
[i
];
2662 list_for_each_entry_safe(page
, tmp
,
2663 &data
->trace_pages
, lru
) {
2664 list_del_init(&page
->lru
);
2669 #ifdef CONFIG_TRACER_MAX_TRACE
2670 data
= max_tr
.data
[i
];
2672 list_for_each_entry_safe(page
, tmp
,
2673 &data
->trace_pages
, lru
) {
2674 list_del_init(&page
->lru
);
2682 fs_initcall(tracer_alloc_buffers
);