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>
30 #include <linux/kprobes.h>
31 #include <linux/writeback.h>
33 #include <linux/stacktrace.h>
37 unsigned long __read_mostly tracing_max_latency
= (cycle_t
)ULONG_MAX
;
38 unsigned long __read_mostly tracing_thresh
;
40 static unsigned long __read_mostly tracing_nr_buffers
;
41 static cpumask_t __read_mostly tracing_buffer_mask
;
43 #define for_each_tracing_cpu(cpu) \
44 for_each_cpu_mask(cpu, tracing_buffer_mask)
46 static int trace_alloc_page(void);
47 static int trace_free_page(void);
49 static int tracing_disabled
= 1;
51 static unsigned long tracing_pages_allocated
;
54 ns2usecs(cycle_t nsec
)
61 cycle_t
ftrace_now(int cpu
)
63 return cpu_clock(cpu
);
67 * The global_trace is the descriptor that holds the tracing
68 * buffers for the live tracing. For each CPU, it contains
69 * a link list of pages that will store trace entries. The
70 * page descriptor of the pages in the memory is used to hold
71 * the link list by linking the lru item in the page descriptor
72 * to each of the pages in the buffer per CPU.
74 * For each active CPU there is a data field that holds the
75 * pages for the buffer for that CPU. Each CPU has the same number
76 * of pages allocated for its buffer.
78 static struct trace_array global_trace
;
80 static DEFINE_PER_CPU(struct trace_array_cpu
, global_trace_cpu
);
83 * The max_tr is used to snapshot the global_trace when a maximum
84 * latency is reached. Some tracers will use this to store a maximum
85 * trace while it continues examining live traces.
87 * The buffers for the max_tr are set up the same as the global_trace.
88 * When a snapshot is taken, the link list of the max_tr is swapped
89 * with the link list of the global_trace and the buffers are reset for
90 * the global_trace so the tracing can continue.
92 static struct trace_array max_tr
;
94 static DEFINE_PER_CPU(struct trace_array_cpu
, max_data
);
96 /* tracer_enabled is used to toggle activation of a tracer */
97 static int tracer_enabled
= 1;
100 * trace_nr_entries is the number of entries that is allocated
101 * for a buffer. Note, the number of entries is always rounded
102 * to ENTRIES_PER_PAGE.
104 static unsigned long trace_nr_entries
= 65536UL;
106 /* trace_types holds a link list of available tracers. */
107 static struct tracer
*trace_types __read_mostly
;
109 /* current_trace points to the tracer that is currently active */
110 static struct tracer
*current_trace __read_mostly
;
113 * max_tracer_type_len is used to simplify the allocating of
114 * buffers to read userspace tracer names. We keep track of
115 * the longest tracer name registered.
117 static int max_tracer_type_len
;
120 * trace_types_lock is used to protect the trace_types list.
121 * This lock is also used to keep user access serialized.
122 * Accesses from userspace will grab this lock while userspace
123 * activities happen inside the kernel.
125 static DEFINE_MUTEX(trace_types_lock
);
127 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
128 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
130 /* trace_flags holds iter_ctrl options */
131 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
;
133 static notrace
void no_trace_init(struct trace_array
*tr
)
138 for_each_online_cpu(cpu
)
139 tracing_reset(tr
->data
[cpu
]);
143 /* dummy trace to disable tracing */
144 static struct tracer no_tracer __read_mostly
= {
146 .init
= no_trace_init
151 * trace_wake_up - wake up tasks waiting for trace input
153 * Simply wakes up any task that is blocked on the trace_wait
154 * queue. These is used with trace_poll for tasks polling the trace.
156 void trace_wake_up(void)
159 * The runqueue_is_locked() can fail, but this is the best we
162 if (!(trace_flags
& TRACE_ITER_BLOCK
) && !runqueue_is_locked())
163 wake_up(&trace_wait
);
166 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
168 static int __init
set_nr_entries(char *str
)
170 unsigned long nr_entries
;
175 ret
= strict_strtoul(str
, 0, &nr_entries
);
176 /* nr_entries can not be zero */
177 if (ret
< 0 || nr_entries
== 0)
179 trace_nr_entries
= nr_entries
;
182 __setup("trace_entries=", set_nr_entries
);
184 unsigned long nsecs_to_usecs(unsigned long nsecs
)
190 * trace_flag_type is an enumeration that holds different
191 * states when a trace occurs. These are:
192 * IRQS_OFF - interrupts were disabled
193 * NEED_RESCED - reschedule is requested
194 * HARDIRQ - inside an interrupt handler
195 * SOFTIRQ - inside a softirq handler
197 enum trace_flag_type
{
198 TRACE_FLAG_IRQS_OFF
= 0x01,
199 TRACE_FLAG_NEED_RESCHED
= 0x02,
200 TRACE_FLAG_HARDIRQ
= 0x04,
201 TRACE_FLAG_SOFTIRQ
= 0x08,
205 * TRACE_ITER_SYM_MASK masks the options in trace_flags that
206 * control the output of kernel symbols.
208 #define TRACE_ITER_SYM_MASK \
209 (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
211 /* These must match the bit postions in trace_iterator_flags */
212 static const char *trace_options
[] = {
227 * ftrace_max_lock is used to protect the swapping of buffers
228 * when taking a max snapshot. The buffers themselves are
229 * protected by per_cpu spinlocks. But the action of the swap
230 * needs its own lock.
232 * This is defined as a raw_spinlock_t in order to help
233 * with performance when lockdep debugging is enabled.
235 static raw_spinlock_t ftrace_max_lock
=
236 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
239 * Copy the new maximum trace into the separate maximum-trace
240 * structure. (this way the maximum trace is permanently saved,
241 * for later retrieval via /debugfs/tracing/latency_trace)
244 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
246 struct trace_array_cpu
*data
= tr
->data
[cpu
];
249 max_tr
.time_start
= data
->preempt_timestamp
;
251 data
= max_tr
.data
[cpu
];
252 data
->saved_latency
= tracing_max_latency
;
254 memcpy(data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
255 data
->pid
= tsk
->pid
;
256 data
->uid
= tsk
->uid
;
257 data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
258 data
->policy
= tsk
->policy
;
259 data
->rt_priority
= tsk
->rt_priority
;
261 /* record this tasks comm */
262 tracing_record_cmdline(current
);
265 #define CHECK_COND(cond) \
266 if (unlikely(cond)) { \
267 tracing_disabled = 1; \
273 * check_pages - integrity check of trace buffers
275 * As a safty measure we check to make sure the data pages have not
278 int check_pages(struct trace_array_cpu
*data
)
280 struct page
*page
, *tmp
;
282 CHECK_COND(data
->trace_pages
.next
->prev
!= &data
->trace_pages
);
283 CHECK_COND(data
->trace_pages
.prev
->next
!= &data
->trace_pages
);
285 list_for_each_entry_safe(page
, tmp
, &data
->trace_pages
, lru
) {
286 CHECK_COND(page
->lru
.next
->prev
!= &page
->lru
);
287 CHECK_COND(page
->lru
.prev
->next
!= &page
->lru
);
294 * head_page - page address of the first page in per_cpu buffer.
296 * head_page returns the page address of the first page in
297 * a per_cpu buffer. This also preforms various consistency
298 * checks to make sure the buffer has not been corrupted.
300 void *head_page(struct trace_array_cpu
*data
)
304 if (list_empty(&data
->trace_pages
))
307 page
= list_entry(data
->trace_pages
.next
, struct page
, lru
);
308 BUG_ON(&page
->lru
== &data
->trace_pages
);
310 return page_address(page
);
314 * trace_seq_printf - sequence printing of trace information
315 * @s: trace sequence descriptor
316 * @fmt: printf format string
318 * The tracer may use either sequence operations or its own
319 * copy to user routines. To simplify formating of a trace
320 * trace_seq_printf is used to store strings into a special
321 * buffer (@s). Then the output may be either used by
322 * the sequencer or pulled into another buffer.
325 trace_seq_printf(struct trace_seq
*s
, const char *fmt
, ...)
327 int len
= (PAGE_SIZE
- 1) - s
->len
;
335 ret
= vsnprintf(s
->buffer
+ s
->len
, len
, fmt
, ap
);
338 /* If we can't write it all, don't bother writing anything */
348 * trace_seq_puts - trace sequence printing of simple string
349 * @s: trace sequence descriptor
350 * @str: simple string to record
352 * The tracer may use either the sequence operations or its own
353 * copy to user routines. This function records a simple string
354 * into a special buffer (@s) for later retrieval by a sequencer
355 * or other mechanism.
358 trace_seq_puts(struct trace_seq
*s
, const char *str
)
360 int len
= strlen(str
);
362 if (len
> ((PAGE_SIZE
- 1) - s
->len
))
365 memcpy(s
->buffer
+ s
->len
, str
, len
);
372 trace_seq_putc(struct trace_seq
*s
, unsigned char c
)
374 if (s
->len
>= (PAGE_SIZE
- 1))
377 s
->buffer
[s
->len
++] = c
;
383 trace_seq_putmem(struct trace_seq
*s
, void *mem
, size_t len
)
385 if (len
> ((PAGE_SIZE
- 1) - s
->len
))
388 memcpy(s
->buffer
+ s
->len
, mem
, len
);
395 static const char hex2asc
[] = "0123456789abcdef";
398 trace_seq_putmem_hex(struct trace_seq
*s
, void *mem
, size_t len
)
400 unsigned char hex
[HEX_CHARS
];
401 unsigned char *data
= mem
;
405 BUG_ON(len
>= HEX_CHARS
);
408 for (i
= 0, j
= 0; i
< len
; i
++) {
410 for (i
= len
-1, j
= 0; i
>= 0; i
--) {
414 hex
[j
++] = hex2asc
[byte
& 0x0f];
415 hex
[j
++] = hex2asc
[byte
>> 4];
419 return trace_seq_putmem(s
, hex
, j
);
423 trace_seq_reset(struct trace_seq
*s
)
429 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
434 if (s
->len
<= s
->readpos
)
437 len
= s
->len
- s
->readpos
;
440 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
449 trace_print_seq(struct seq_file
*m
, struct trace_seq
*s
)
451 int len
= s
->len
>= PAGE_SIZE
? PAGE_SIZE
- 1 : s
->len
;
454 seq_puts(m
, s
->buffer
);
460 * flip the trace buffers between two trace descriptors.
461 * This usually is the buffers between the global_trace and
462 * the max_tr to record a snapshot of a current trace.
464 * The ftrace_max_lock must be held.
467 flip_trace(struct trace_array_cpu
*tr1
, struct trace_array_cpu
*tr2
)
469 struct list_head flip_pages
;
471 INIT_LIST_HEAD(&flip_pages
);
473 memcpy(&tr1
->trace_head_idx
, &tr2
->trace_head_idx
,
474 sizeof(struct trace_array_cpu
) -
475 offsetof(struct trace_array_cpu
, trace_head_idx
));
479 list_splice_init(&tr1
->trace_pages
, &flip_pages
);
480 list_splice_init(&tr2
->trace_pages
, &tr1
->trace_pages
);
481 list_splice_init(&flip_pages
, &tr2
->trace_pages
);
482 BUG_ON(!list_empty(&flip_pages
));
488 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
490 * @tsk: the task with the latency
491 * @cpu: The cpu that initiated the trace.
493 * Flip the buffers between the @tr and the max_tr and record information
494 * about which task was the cause of this latency.
497 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
499 struct trace_array_cpu
*data
;
502 WARN_ON_ONCE(!irqs_disabled());
503 __raw_spin_lock(&ftrace_max_lock
);
504 /* clear out all the previous traces */
505 for_each_tracing_cpu(i
) {
507 flip_trace(max_tr
.data
[i
], data
);
511 __update_max_tr(tr
, tsk
, cpu
);
512 __raw_spin_unlock(&ftrace_max_lock
);
516 * update_max_tr_single - only copy one trace over, and reset the rest
518 * @tsk - task with the latency
519 * @cpu - the cpu of the buffer to copy.
521 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
524 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
526 struct trace_array_cpu
*data
= tr
->data
[cpu
];
529 WARN_ON_ONCE(!irqs_disabled());
530 __raw_spin_lock(&ftrace_max_lock
);
531 for_each_tracing_cpu(i
)
532 tracing_reset(max_tr
.data
[i
]);
534 flip_trace(max_tr
.data
[cpu
], data
);
537 __update_max_tr(tr
, tsk
, cpu
);
538 __raw_spin_unlock(&ftrace_max_lock
);
542 * register_tracer - register a tracer with the ftrace system.
543 * @type - the plugin for the tracer
545 * Register a new plugin tracer.
547 int register_tracer(struct tracer
*type
)
554 pr_info("Tracer must have a name\n");
558 mutex_lock(&trace_types_lock
);
559 for (t
= trace_types
; t
; t
= t
->next
) {
560 if (strcmp(type
->name
, t
->name
) == 0) {
562 pr_info("Trace %s already registered\n",
569 #ifdef CONFIG_FTRACE_STARTUP_TEST
570 if (type
->selftest
) {
571 struct tracer
*saved_tracer
= current_trace
;
572 struct trace_array_cpu
*data
;
573 struct trace_array
*tr
= &global_trace
;
574 int saved_ctrl
= tr
->ctrl
;
577 * Run a selftest on this tracer.
578 * Here we reset the trace buffer, and set the current
579 * tracer to be this tracer. The tracer can then run some
580 * internal tracing to verify that everything is in order.
581 * If we fail, we do not register this tracer.
583 for_each_tracing_cpu(i
) {
585 if (!head_page(data
))
589 current_trace
= type
;
591 /* the test is responsible for initializing and enabling */
592 pr_info("Testing tracer %s: ", type
->name
);
593 ret
= type
->selftest(type
, tr
);
594 /* the test is responsible for resetting too */
595 current_trace
= saved_tracer
;
596 tr
->ctrl
= saved_ctrl
;
598 printk(KERN_CONT
"FAILED!\n");
601 /* Only reset on passing, to avoid touching corrupted buffers */
602 for_each_tracing_cpu(i
) {
604 if (!head_page(data
))
608 printk(KERN_CONT
"PASSED\n");
612 type
->next
= trace_types
;
614 len
= strlen(type
->name
);
615 if (len
> max_tracer_type_len
)
616 max_tracer_type_len
= len
;
619 mutex_unlock(&trace_types_lock
);
624 void unregister_tracer(struct tracer
*type
)
629 mutex_lock(&trace_types_lock
);
630 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
634 pr_info("Trace %s not registered\n", type
->name
);
639 if (strlen(type
->name
) != max_tracer_type_len
)
642 max_tracer_type_len
= 0;
643 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
644 len
= strlen((*t
)->name
);
645 if (len
> max_tracer_type_len
)
646 max_tracer_type_len
= len
;
649 mutex_unlock(&trace_types_lock
);
652 void tracing_reset(struct trace_array_cpu
*data
)
656 data
->trace_head
= data
->trace_tail
= head_page(data
);
657 data
->trace_head_idx
= 0;
658 data
->trace_tail_idx
= 0;
661 #define SAVED_CMDLINES 128
662 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
663 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
664 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
665 static int cmdline_idx
;
666 static DEFINE_SPINLOCK(trace_cmdline_lock
);
668 /* temporary disable recording */
669 atomic_t trace_record_cmdline_disabled __read_mostly
;
671 static void trace_init_cmdlines(void)
673 memset(&map_pid_to_cmdline
, -1, sizeof(map_pid_to_cmdline
));
674 memset(&map_cmdline_to_pid
, -1, sizeof(map_cmdline_to_pid
));
678 void trace_stop_cmdline_recording(void);
680 static void trace_save_cmdline(struct task_struct
*tsk
)
685 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
689 * It's not the end of the world if we don't get
690 * the lock, but we also don't want to spin
691 * nor do we want to disable interrupts,
692 * so if we miss here, then better luck next time.
694 if (!spin_trylock(&trace_cmdline_lock
))
697 idx
= map_pid_to_cmdline
[tsk
->pid
];
698 if (idx
>= SAVED_CMDLINES
) {
699 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
701 map
= map_cmdline_to_pid
[idx
];
702 if (map
<= PID_MAX_DEFAULT
)
703 map_pid_to_cmdline
[map
] = (unsigned)-1;
705 map_pid_to_cmdline
[tsk
->pid
] = idx
;
710 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
712 spin_unlock(&trace_cmdline_lock
);
715 static char *trace_find_cmdline(int pid
)
717 char *cmdline
= "<...>";
723 if (pid
> PID_MAX_DEFAULT
)
726 map
= map_pid_to_cmdline
[pid
];
727 if (map
>= SAVED_CMDLINES
)
730 cmdline
= saved_cmdlines
[map
];
736 void tracing_record_cmdline(struct task_struct
*tsk
)
738 if (atomic_read(&trace_record_cmdline_disabled
))
741 trace_save_cmdline(tsk
);
744 static inline struct list_head
*
745 trace_next_list(struct trace_array_cpu
*data
, struct list_head
*next
)
748 * Roundrobin - but skip the head (which is not a real page):
751 if (unlikely(next
== &data
->trace_pages
))
753 BUG_ON(next
== &data
->trace_pages
);
759 trace_next_page(struct trace_array_cpu
*data
, void *addr
)
761 struct list_head
*next
;
764 page
= virt_to_page(addr
);
766 next
= trace_next_list(data
, &page
->lru
);
767 page
= list_entry(next
, struct page
, lru
);
769 return page_address(page
);
772 static inline struct trace_entry
*
773 tracing_get_trace_entry(struct trace_array
*tr
, struct trace_array_cpu
*data
)
775 unsigned long idx
, idx_next
;
776 struct trace_entry
*entry
;
779 idx
= data
->trace_head_idx
;
782 BUG_ON(idx
* TRACE_ENTRY_SIZE
>= PAGE_SIZE
);
784 entry
= data
->trace_head
+ idx
* TRACE_ENTRY_SIZE
;
786 if (unlikely(idx_next
>= ENTRIES_PER_PAGE
)) {
787 data
->trace_head
= trace_next_page(data
, data
->trace_head
);
791 if (data
->trace_head
== data
->trace_tail
&&
792 idx_next
== data
->trace_tail_idx
) {
795 data
->trace_tail_idx
++;
796 if (data
->trace_tail_idx
>= ENTRIES_PER_PAGE
) {
798 trace_next_page(data
, data
->trace_tail
);
799 data
->trace_tail_idx
= 0;
803 data
->trace_head_idx
= idx_next
;
809 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
)
811 struct task_struct
*tsk
= current
;
814 pc
= preempt_count();
816 entry
->preempt_count
= pc
& 0xff;
817 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
818 entry
->t
= ftrace_now(raw_smp_processor_id());
819 entry
->flags
= (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
820 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
821 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
822 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
826 trace_function(struct trace_array
*tr
, struct trace_array_cpu
*data
,
827 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
)
829 struct trace_entry
*entry
;
830 unsigned long irq_flags
;
832 raw_local_irq_save(irq_flags
);
833 __raw_spin_lock(&data
->lock
);
834 entry
= tracing_get_trace_entry(tr
, data
);
835 tracing_generic_entry_update(entry
, flags
);
836 entry
->type
= TRACE_FN
;
838 entry
->fn
.parent_ip
= parent_ip
;
839 __raw_spin_unlock(&data
->lock
);
840 raw_local_irq_restore(irq_flags
);
844 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
845 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
)
847 if (likely(!atomic_read(&data
->disabled
)))
848 trace_function(tr
, data
, ip
, parent_ip
, flags
);
851 void __trace_stack(struct trace_array
*tr
,
852 struct trace_array_cpu
*data
,
856 struct trace_entry
*entry
;
857 struct stack_trace trace
;
859 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
862 entry
= tracing_get_trace_entry(tr
, data
);
863 tracing_generic_entry_update(entry
, flags
);
864 entry
->type
= TRACE_STACK
;
866 memset(&entry
->stack
, 0, sizeof(entry
->stack
));
868 trace
.nr_entries
= 0;
869 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
871 trace
.entries
= entry
->stack
.caller
;
873 save_stack_trace(&trace
);
877 __trace_special(void *__tr
, void *__data
,
878 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
880 struct trace_array_cpu
*data
= __data
;
881 struct trace_array
*tr
= __tr
;
882 struct trace_entry
*entry
;
883 unsigned long irq_flags
;
885 raw_local_irq_save(irq_flags
);
886 __raw_spin_lock(&data
->lock
);
887 entry
= tracing_get_trace_entry(tr
, data
);
888 tracing_generic_entry_update(entry
, 0);
889 entry
->type
= TRACE_SPECIAL
;
890 entry
->special
.arg1
= arg1
;
891 entry
->special
.arg2
= arg2
;
892 entry
->special
.arg3
= arg3
;
893 __trace_stack(tr
, data
, irq_flags
, 4);
894 __raw_spin_unlock(&data
->lock
);
895 raw_local_irq_restore(irq_flags
);
901 tracing_sched_switch_trace(struct trace_array
*tr
,
902 struct trace_array_cpu
*data
,
903 struct task_struct
*prev
,
904 struct task_struct
*next
,
907 struct trace_entry
*entry
;
908 unsigned long irq_flags
;
910 raw_local_irq_save(irq_flags
);
911 __raw_spin_lock(&data
->lock
);
912 entry
= tracing_get_trace_entry(tr
, data
);
913 tracing_generic_entry_update(entry
, flags
);
914 entry
->type
= TRACE_CTX
;
915 entry
->ctx
.prev_pid
= prev
->pid
;
916 entry
->ctx
.prev_prio
= prev
->prio
;
917 entry
->ctx
.prev_state
= prev
->state
;
918 entry
->ctx
.next_pid
= next
->pid
;
919 entry
->ctx
.next_prio
= next
->prio
;
920 entry
->ctx
.next_state
= next
->state
;
921 __trace_stack(tr
, data
, flags
, 5);
922 __raw_spin_unlock(&data
->lock
);
923 raw_local_irq_restore(irq_flags
);
927 tracing_sched_wakeup_trace(struct trace_array
*tr
,
928 struct trace_array_cpu
*data
,
929 struct task_struct
*wakee
,
930 struct task_struct
*curr
,
933 struct trace_entry
*entry
;
934 unsigned long irq_flags
;
936 raw_local_irq_save(irq_flags
);
937 __raw_spin_lock(&data
->lock
);
938 entry
= tracing_get_trace_entry(tr
, data
);
939 tracing_generic_entry_update(entry
, flags
);
940 entry
->type
= TRACE_WAKE
;
941 entry
->ctx
.prev_pid
= curr
->pid
;
942 entry
->ctx
.prev_prio
= curr
->prio
;
943 entry
->ctx
.prev_state
= curr
->state
;
944 entry
->ctx
.next_pid
= wakee
->pid
;
945 entry
->ctx
.next_prio
= wakee
->prio
;
946 entry
->ctx
.next_state
= wakee
->state
;
947 __trace_stack(tr
, data
, flags
, 6);
948 __raw_spin_unlock(&data
->lock
);
949 raw_local_irq_restore(irq_flags
);
955 ftrace_special(unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
957 struct trace_array
*tr
= &global_trace
;
958 struct trace_array_cpu
*data
;
963 if (tracing_disabled
|| current_trace
== &no_tracer
|| !tr
->ctrl
)
966 local_irq_save(flags
);
967 cpu
= raw_smp_processor_id();
968 data
= tr
->data
[cpu
];
969 disabled
= atomic_inc_return(&data
->disabled
);
971 if (likely(disabled
== 1))
972 __trace_special(tr
, data
, arg1
, arg2
, arg3
);
974 atomic_dec(&data
->disabled
);
975 local_irq_restore(flags
);
980 function_trace_call(unsigned long ip
, unsigned long parent_ip
)
982 struct trace_array
*tr
= &global_trace
;
983 struct trace_array_cpu
*data
;
988 if (unlikely(!tracer_enabled
))
994 local_irq_save(flags
);
995 cpu
= raw_smp_processor_id();
996 data
= tr
->data
[cpu
];
997 disabled
= atomic_inc_return(&data
->disabled
);
999 if (likely(disabled
== 1))
1000 trace_function(tr
, data
, ip
, parent_ip
, flags
);
1002 atomic_dec(&data
->disabled
);
1003 local_irq_restore(flags
);
1006 static struct ftrace_ops trace_ops __read_mostly
=
1008 .func
= function_trace_call
,
1011 void tracing_start_function_trace(void)
1013 register_ftrace_function(&trace_ops
);
1016 void tracing_stop_function_trace(void)
1018 unregister_ftrace_function(&trace_ops
);
1022 enum trace_file_type
{
1023 TRACE_FILE_LAT_FMT
= 1,
1026 static struct trace_entry
*
1027 trace_entry_idx(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1028 struct trace_iterator
*iter
, int cpu
)
1031 struct trace_entry
*array
;
1033 if (iter
->next_idx
[cpu
] >= tr
->entries
||
1034 iter
->next_idx
[cpu
] >= data
->trace_idx
||
1035 (data
->trace_head
== data
->trace_tail
&&
1036 data
->trace_head_idx
== data
->trace_tail_idx
))
1039 if (!iter
->next_page
[cpu
]) {
1040 /* Initialize the iterator for this cpu trace buffer */
1041 WARN_ON(!data
->trace_tail
);
1042 page
= virt_to_page(data
->trace_tail
);
1043 iter
->next_page
[cpu
] = &page
->lru
;
1044 iter
->next_page_idx
[cpu
] = data
->trace_tail_idx
;
1047 page
= list_entry(iter
->next_page
[cpu
], struct page
, lru
);
1048 BUG_ON(&data
->trace_pages
== &page
->lru
);
1050 array
= page_address(page
);
1052 WARN_ON(iter
->next_page_idx
[cpu
] >= ENTRIES_PER_PAGE
);
1053 return &array
[iter
->next_page_idx
[cpu
]];
1056 static struct trace_entry
*
1057 find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
)
1059 struct trace_array
*tr
= iter
->tr
;
1060 struct trace_entry
*ent
, *next
= NULL
;
1064 for_each_tracing_cpu(cpu
) {
1065 if (!head_page(tr
->data
[cpu
]))
1067 ent
= trace_entry_idx(tr
, tr
->data
[cpu
], iter
, cpu
);
1069 * Pick the entry with the smallest timestamp:
1071 if (ent
&& (!next
|| ent
->t
< next
->t
)) {
1078 *ent_cpu
= next_cpu
;
1083 static void trace_iterator_increment(struct trace_iterator
*iter
)
1086 iter
->next_idx
[iter
->cpu
]++;
1087 iter
->next_page_idx
[iter
->cpu
]++;
1089 if (iter
->next_page_idx
[iter
->cpu
] >= ENTRIES_PER_PAGE
) {
1090 struct trace_array_cpu
*data
= iter
->tr
->data
[iter
->cpu
];
1092 iter
->next_page_idx
[iter
->cpu
] = 0;
1093 iter
->next_page
[iter
->cpu
] =
1094 trace_next_list(data
, iter
->next_page
[iter
->cpu
]);
1098 static void trace_consume(struct trace_iterator
*iter
)
1100 struct trace_array_cpu
*data
= iter
->tr
->data
[iter
->cpu
];
1102 data
->trace_tail_idx
++;
1103 if (data
->trace_tail_idx
>= ENTRIES_PER_PAGE
) {
1104 data
->trace_tail
= trace_next_page(data
, data
->trace_tail
);
1105 data
->trace_tail_idx
= 0;
1108 /* Check if we empty it, then reset the index */
1109 if (data
->trace_head
== data
->trace_tail
&&
1110 data
->trace_head_idx
== data
->trace_tail_idx
)
1111 data
->trace_idx
= 0;
1114 static void *find_next_entry_inc(struct trace_iterator
*iter
)
1116 struct trace_entry
*next
;
1119 next
= find_next_entry(iter
, &next_cpu
);
1121 iter
->prev_ent
= iter
->ent
;
1122 iter
->prev_cpu
= iter
->cpu
;
1125 iter
->cpu
= next_cpu
;
1128 trace_iterator_increment(iter
);
1130 return next
? iter
: NULL
;
1133 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1135 struct trace_iterator
*iter
= m
->private;
1136 void *last_ent
= iter
->ent
;
1142 /* can't go backwards */
1147 ent
= find_next_entry_inc(iter
);
1151 while (ent
&& iter
->idx
< i
)
1152 ent
= find_next_entry_inc(iter
);
1156 if (last_ent
&& !ent
)
1157 seq_puts(m
, "\n\nvim:ft=help\n");
1162 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1164 struct trace_iterator
*iter
= m
->private;
1169 mutex_lock(&trace_types_lock
);
1171 if (!current_trace
|| current_trace
!= iter
->trace
) {
1172 mutex_unlock(&trace_types_lock
);
1176 atomic_inc(&trace_record_cmdline_disabled
);
1178 /* let the tracer grab locks here if needed */
1179 if (current_trace
->start
)
1180 current_trace
->start(iter
);
1182 if (*pos
!= iter
->pos
) {
1186 iter
->prev_ent
= NULL
;
1187 iter
->prev_cpu
= -1;
1189 for_each_tracing_cpu(i
) {
1190 iter
->next_idx
[i
] = 0;
1191 iter
->next_page
[i
] = NULL
;
1194 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1199 p
= s_next(m
, p
, &l
);
1205 static void s_stop(struct seq_file
*m
, void *p
)
1207 struct trace_iterator
*iter
= m
->private;
1209 atomic_dec(&trace_record_cmdline_disabled
);
1211 /* let the tracer release locks here if needed */
1212 if (current_trace
&& current_trace
== iter
->trace
&& iter
->trace
->stop
)
1213 iter
->trace
->stop(iter
);
1215 mutex_unlock(&trace_types_lock
);
1218 #define KRETPROBE_MSG "[unknown/kretprobe'd]"
1220 #ifdef CONFIG_KRETPROBES
1221 static inline int kretprobed(unsigned long addr
)
1223 return addr
== (unsigned long)kretprobe_trampoline
;
1226 static inline int kretprobed(unsigned long addr
)
1230 #endif /* CONFIG_KRETPROBES */
1233 seq_print_sym_short(struct trace_seq
*s
, const char *fmt
, unsigned long address
)
1235 #ifdef CONFIG_KALLSYMS
1236 char str
[KSYM_SYMBOL_LEN
];
1238 kallsyms_lookup(address
, NULL
, NULL
, NULL
, str
);
1240 return trace_seq_printf(s
, fmt
, str
);
1246 seq_print_sym_offset(struct trace_seq
*s
, const char *fmt
,
1247 unsigned long address
)
1249 #ifdef CONFIG_KALLSYMS
1250 char str
[KSYM_SYMBOL_LEN
];
1252 sprint_symbol(str
, address
);
1253 return trace_seq_printf(s
, fmt
, str
);
1258 #ifndef CONFIG_64BIT
1259 # define IP_FMT "%08lx"
1261 # define IP_FMT "%016lx"
1265 seq_print_ip_sym(struct trace_seq
*s
, unsigned long ip
, unsigned long sym_flags
)
1270 return trace_seq_printf(s
, "0");
1272 if (sym_flags
& TRACE_ITER_SYM_OFFSET
)
1273 ret
= seq_print_sym_offset(s
, "%s", ip
);
1275 ret
= seq_print_sym_short(s
, "%s", ip
);
1280 if (sym_flags
& TRACE_ITER_SYM_ADDR
)
1281 ret
= trace_seq_printf(s
, " <" IP_FMT
">", ip
);
1285 static void print_lat_help_header(struct seq_file
*m
)
1287 seq_puts(m
, "# _------=> CPU# \n");
1288 seq_puts(m
, "# / _-----=> irqs-off \n");
1289 seq_puts(m
, "# | / _----=> need-resched \n");
1290 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1291 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1292 seq_puts(m
, "# |||| / \n");
1293 seq_puts(m
, "# ||||| delay \n");
1294 seq_puts(m
, "# cmd pid ||||| time | caller \n");
1295 seq_puts(m
, "# \\ / ||||| \\ | / \n");
1298 static void print_func_help_header(struct seq_file
*m
)
1300 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1301 seq_puts(m
, "# | | | | |\n");
1306 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1308 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1309 struct trace_array
*tr
= iter
->tr
;
1310 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1311 struct tracer
*type
= current_trace
;
1312 unsigned long total
= 0;
1313 unsigned long entries
= 0;
1315 const char *name
= "preemption";
1320 for_each_tracing_cpu(cpu
) {
1321 if (head_page(tr
->data
[cpu
])) {
1322 total
+= tr
->data
[cpu
]->trace_idx
;
1323 if (tr
->data
[cpu
]->trace_idx
> tr
->entries
)
1324 entries
+= tr
->entries
;
1326 entries
+= tr
->data
[cpu
]->trace_idx
;
1330 seq_printf(m
, "%s latency trace v1.1.5 on %s\n",
1332 seq_puts(m
, "-----------------------------------"
1333 "---------------------------------\n");
1334 seq_printf(m
, " latency: %lu us, #%lu/%lu, CPU#%d |"
1335 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1336 nsecs_to_usecs(data
->saved_latency
),
1340 #if defined(CONFIG_PREEMPT_NONE)
1342 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1344 #elif defined(CONFIG_PREEMPT)
1349 /* These are reserved for later use */
1352 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1356 seq_puts(m
, " -----------------\n");
1357 seq_printf(m
, " | task: %.16s-%d "
1358 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1359 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1360 data
->policy
, data
->rt_priority
);
1361 seq_puts(m
, " -----------------\n");
1363 if (data
->critical_start
) {
1364 seq_puts(m
, " => started at: ");
1365 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1366 trace_print_seq(m
, &iter
->seq
);
1367 seq_puts(m
, "\n => ended at: ");
1368 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1369 trace_print_seq(m
, &iter
->seq
);
1377 lat_print_generic(struct trace_seq
*s
, struct trace_entry
*entry
, int cpu
)
1379 int hardirq
, softirq
;
1382 comm
= trace_find_cmdline(entry
->pid
);
1384 trace_seq_printf(s
, "%8.8s-%-5d ", comm
, entry
->pid
);
1385 trace_seq_printf(s
, "%d", cpu
);
1386 trace_seq_printf(s
, "%c%c",
1387 (entry
->flags
& TRACE_FLAG_IRQS_OFF
) ? 'd' : '.',
1388 ((entry
->flags
& TRACE_FLAG_NEED_RESCHED
) ? 'N' : '.'));
1390 hardirq
= entry
->flags
& TRACE_FLAG_HARDIRQ
;
1391 softirq
= entry
->flags
& TRACE_FLAG_SOFTIRQ
;
1392 if (hardirq
&& softirq
) {
1393 trace_seq_putc(s
, 'H');
1396 trace_seq_putc(s
, 'h');
1399 trace_seq_putc(s
, 's');
1401 trace_seq_putc(s
, '.');
1405 if (entry
->preempt_count
)
1406 trace_seq_printf(s
, "%x", entry
->preempt_count
);
1408 trace_seq_puts(s
, ".");
1411 unsigned long preempt_mark_thresh
= 100;
1414 lat_print_timestamp(struct trace_seq
*s
, unsigned long long abs_usecs
,
1415 unsigned long rel_usecs
)
1417 trace_seq_printf(s
, " %4lldus", abs_usecs
);
1418 if (rel_usecs
> preempt_mark_thresh
)
1419 trace_seq_puts(s
, "!: ");
1420 else if (rel_usecs
> 1)
1421 trace_seq_puts(s
, "+: ");
1423 trace_seq_puts(s
, " : ");
1426 static const char state_to_char
[] = TASK_STATE_TO_CHAR_STR
;
1429 print_lat_fmt(struct trace_iterator
*iter
, unsigned int trace_idx
, int cpu
)
1431 struct trace_seq
*s
= &iter
->seq
;
1432 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1433 struct trace_entry
*next_entry
= find_next_entry(iter
, NULL
);
1434 unsigned long verbose
= (trace_flags
& TRACE_ITER_VERBOSE
);
1435 struct trace_entry
*entry
= iter
->ent
;
1436 unsigned long abs_usecs
;
1437 unsigned long rel_usecs
;
1445 rel_usecs
= ns2usecs(next_entry
->t
- entry
->t
);
1446 abs_usecs
= ns2usecs(entry
->t
- iter
->tr
->time_start
);
1449 comm
= trace_find_cmdline(entry
->pid
);
1450 trace_seq_printf(s
, "%16s %5d %d %d %08x %08x [%08lx]"
1451 " %ld.%03ldms (+%ld.%03ldms): ",
1453 entry
->pid
, cpu
, entry
->flags
,
1454 entry
->preempt_count
, trace_idx
,
1457 abs_usecs
% 1000, rel_usecs
/1000,
1460 lat_print_generic(s
, entry
, cpu
);
1461 lat_print_timestamp(s
, abs_usecs
, rel_usecs
);
1463 switch (entry
->type
) {
1465 seq_print_ip_sym(s
, entry
->fn
.ip
, sym_flags
);
1466 trace_seq_puts(s
, " (");
1467 if (kretprobed(entry
->fn
.parent_ip
))
1468 trace_seq_puts(s
, KRETPROBE_MSG
);
1470 seq_print_ip_sym(s
, entry
->fn
.parent_ip
, sym_flags
);
1471 trace_seq_puts(s
, ")\n");
1475 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1476 state_to_char
[entry
->ctx
.next_state
] : 'X';
1478 state
= entry
->ctx
.prev_state
? __ffs(entry
->ctx
.prev_state
) + 1 : 0;
1479 S
= state
< sizeof(state_to_char
) - 1 ? state_to_char
[state
] : 'X';
1480 comm
= trace_find_cmdline(entry
->ctx
.next_pid
);
1481 trace_seq_printf(s
, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1482 entry
->ctx
.prev_pid
,
1483 entry
->ctx
.prev_prio
,
1484 S
, entry
->type
== TRACE_CTX
? "==>" : " +",
1485 entry
->ctx
.next_pid
,
1486 entry
->ctx
.next_prio
,
1490 trace_seq_printf(s
, "# %ld %ld %ld\n",
1491 entry
->special
.arg1
,
1492 entry
->special
.arg2
,
1493 entry
->special
.arg3
);
1496 for (i
= 0; i
< FTRACE_STACK_ENTRIES
; i
++) {
1498 trace_seq_puts(s
, " <= ");
1499 seq_print_ip_sym(s
, entry
->stack
.caller
[i
], sym_flags
);
1501 trace_seq_puts(s
, "\n");
1504 trace_seq_printf(s
, "Unknown type %d\n", entry
->type
);
1509 static int print_trace_fmt(struct trace_iterator
*iter
)
1511 struct trace_seq
*s
= &iter
->seq
;
1512 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1513 struct trace_entry
*entry
;
1514 unsigned long usec_rem
;
1515 unsigned long long t
;
1524 comm
= trace_find_cmdline(iter
->ent
->pid
);
1526 t
= ns2usecs(entry
->t
);
1527 usec_rem
= do_div(t
, 1000000ULL);
1528 secs
= (unsigned long)t
;
1530 ret
= trace_seq_printf(s
, "%16s-%-5d ", comm
, entry
->pid
);
1533 ret
= trace_seq_printf(s
, "[%02d] ", iter
->cpu
);
1536 ret
= trace_seq_printf(s
, "%5lu.%06lu: ", secs
, usec_rem
);
1540 switch (entry
->type
) {
1542 ret
= seq_print_ip_sym(s
, entry
->fn
.ip
, sym_flags
);
1545 if ((sym_flags
& TRACE_ITER_PRINT_PARENT
) &&
1546 entry
->fn
.parent_ip
) {
1547 ret
= trace_seq_printf(s
, " <-");
1550 if (kretprobed(entry
->fn
.parent_ip
))
1551 ret
= trace_seq_puts(s
, KRETPROBE_MSG
);
1553 ret
= seq_print_ip_sym(s
, entry
->fn
.parent_ip
,
1558 ret
= trace_seq_printf(s
, "\n");
1564 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1565 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1566 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1567 state_to_char
[entry
->ctx
.next_state
] : 'X';
1568 ret
= trace_seq_printf(s
, " %5d:%3d:%c %s %5d:%3d:%c\n",
1569 entry
->ctx
.prev_pid
,
1570 entry
->ctx
.prev_prio
,
1572 entry
->type
== TRACE_CTX
? "==>" : " +",
1573 entry
->ctx
.next_pid
,
1574 entry
->ctx
.next_prio
,
1580 ret
= trace_seq_printf(s
, "# %ld %ld %ld\n",
1581 entry
->special
.arg1
,
1582 entry
->special
.arg2
,
1583 entry
->special
.arg3
);
1588 for (i
= 0; i
< FTRACE_STACK_ENTRIES
; i
++) {
1590 ret
= trace_seq_puts(s
, " <= ");
1594 ret
= seq_print_ip_sym(s
, entry
->stack
.caller
[i
],
1599 ret
= trace_seq_puts(s
, "\n");
1607 static int print_raw_fmt(struct trace_iterator
*iter
)
1609 struct trace_seq
*s
= &iter
->seq
;
1610 struct trace_entry
*entry
;
1616 ret
= trace_seq_printf(s
, "%d %d %llu ",
1617 entry
->pid
, iter
->cpu
, entry
->t
);
1621 switch (entry
->type
) {
1623 ret
= trace_seq_printf(s
, "%x %x\n",
1624 entry
->fn
.ip
, entry
->fn
.parent_ip
);
1630 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1631 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1632 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1633 state_to_char
[entry
->ctx
.next_state
] : 'X';
1634 if (entry
->type
== TRACE_WAKE
)
1636 ret
= trace_seq_printf(s
, "%d %d %c %d %d %c\n",
1637 entry
->ctx
.prev_pid
,
1638 entry
->ctx
.prev_prio
,
1640 entry
->ctx
.next_pid
,
1641 entry
->ctx
.next_prio
,
1648 ret
= trace_seq_printf(s
, "# %ld %ld %ld\n",
1649 entry
->special
.arg1
,
1650 entry
->special
.arg2
,
1651 entry
->special
.arg3
);
1659 #define SEQ_PUT_FIELD_RET(s, x) \
1661 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1665 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1667 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1671 static int print_hex_fmt(struct trace_iterator
*iter
)
1673 struct trace_seq
*s
= &iter
->seq
;
1674 unsigned char newline
= '\n';
1675 struct trace_entry
*entry
;
1680 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1681 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1682 SEQ_PUT_HEX_FIELD_RET(s
, entry
->t
);
1684 switch (entry
->type
) {
1686 SEQ_PUT_HEX_FIELD_RET(s
, entry
->fn
.ip
);
1687 SEQ_PUT_HEX_FIELD_RET(s
, entry
->fn
.parent_ip
);
1691 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1692 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1693 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1694 state_to_char
[entry
->ctx
.next_state
] : 'X';
1695 if (entry
->type
== TRACE_WAKE
)
1697 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.prev_pid
);
1698 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.prev_prio
);
1699 SEQ_PUT_HEX_FIELD_RET(s
, S
);
1700 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.next_pid
);
1701 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.next_prio
);
1702 SEQ_PUT_HEX_FIELD_RET(s
, entry
->fn
.parent_ip
);
1703 SEQ_PUT_HEX_FIELD_RET(s
, T
);
1707 SEQ_PUT_HEX_FIELD_RET(s
, entry
->special
.arg1
);
1708 SEQ_PUT_HEX_FIELD_RET(s
, entry
->special
.arg2
);
1709 SEQ_PUT_HEX_FIELD_RET(s
, entry
->special
.arg3
);
1712 SEQ_PUT_FIELD_RET(s
, newline
);
1717 static int print_bin_fmt(struct trace_iterator
*iter
)
1719 struct trace_seq
*s
= &iter
->seq
;
1720 struct trace_entry
*entry
;
1724 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1725 SEQ_PUT_FIELD_RET(s
, entry
->cpu
);
1726 SEQ_PUT_FIELD_RET(s
, entry
->t
);
1728 switch (entry
->type
) {
1730 SEQ_PUT_FIELD_RET(s
, entry
->fn
.ip
);
1731 SEQ_PUT_FIELD_RET(s
, entry
->fn
.parent_ip
);
1734 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_pid
);
1735 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_prio
);
1736 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_state
);
1737 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_pid
);
1738 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_prio
);
1739 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_state
);
1743 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg1
);
1744 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg2
);
1745 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg3
);
1751 static int trace_empty(struct trace_iterator
*iter
)
1753 struct trace_array_cpu
*data
;
1756 for_each_tracing_cpu(cpu
) {
1757 data
= iter
->tr
->data
[cpu
];
1759 if (head_page(data
) && data
->trace_idx
&&
1760 (data
->trace_tail
!= data
->trace_head
||
1761 data
->trace_tail_idx
!= data
->trace_head_idx
))
1767 static int print_trace_line(struct trace_iterator
*iter
)
1769 if (iter
->trace
&& iter
->trace
->print_line
)
1770 return iter
->trace
->print_line(iter
);
1772 if (trace_flags
& TRACE_ITER_BIN
)
1773 return print_bin_fmt(iter
);
1775 if (trace_flags
& TRACE_ITER_HEX
)
1776 return print_hex_fmt(iter
);
1778 if (trace_flags
& TRACE_ITER_RAW
)
1779 return print_raw_fmt(iter
);
1781 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
1782 return print_lat_fmt(iter
, iter
->idx
, iter
->cpu
);
1784 return print_trace_fmt(iter
);
1787 static int s_show(struct seq_file
*m
, void *v
)
1789 struct trace_iterator
*iter
= v
;
1791 if (iter
->ent
== NULL
) {
1793 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
1796 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1797 /* print nothing if the buffers are empty */
1798 if (trace_empty(iter
))
1800 print_trace_header(m
, iter
);
1801 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1802 print_lat_help_header(m
);
1804 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1805 print_func_help_header(m
);
1808 print_trace_line(iter
);
1809 trace_print_seq(m
, &iter
->seq
);
1815 static struct seq_operations tracer_seq_ops
= {
1822 static struct trace_iterator
*
1823 __tracing_open(struct inode
*inode
, struct file
*file
, int *ret
)
1825 struct trace_iterator
*iter
;
1827 if (tracing_disabled
) {
1832 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
1838 mutex_lock(&trace_types_lock
);
1839 if (current_trace
&& current_trace
->print_max
)
1842 iter
->tr
= inode
->i_private
;
1843 iter
->trace
= current_trace
;
1846 /* TODO stop tracer */
1847 *ret
= seq_open(file
, &tracer_seq_ops
);
1849 struct seq_file
*m
= file
->private_data
;
1852 /* stop the trace while dumping */
1856 if (iter
->trace
&& iter
->trace
->open
)
1857 iter
->trace
->open(iter
);
1862 mutex_unlock(&trace_types_lock
);
1868 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
1870 if (tracing_disabled
)
1873 filp
->private_data
= inode
->i_private
;
1877 int tracing_release(struct inode
*inode
, struct file
*file
)
1879 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
1880 struct trace_iterator
*iter
= m
->private;
1882 mutex_lock(&trace_types_lock
);
1883 if (iter
->trace
&& iter
->trace
->close
)
1884 iter
->trace
->close(iter
);
1886 /* reenable tracing if it was previously enabled */
1889 mutex_unlock(&trace_types_lock
);
1891 seq_release(inode
, file
);
1896 static int tracing_open(struct inode
*inode
, struct file
*file
)
1900 __tracing_open(inode
, file
, &ret
);
1905 static int tracing_lt_open(struct inode
*inode
, struct file
*file
)
1907 struct trace_iterator
*iter
;
1910 iter
= __tracing_open(inode
, file
, &ret
);
1913 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
1920 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1922 struct tracer
*t
= m
->private;
1934 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
1936 struct tracer
*t
= m
->private;
1939 mutex_lock(&trace_types_lock
);
1940 for (; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
1946 static void t_stop(struct seq_file
*m
, void *p
)
1948 mutex_unlock(&trace_types_lock
);
1951 static int t_show(struct seq_file
*m
, void *v
)
1953 struct tracer
*t
= v
;
1958 seq_printf(m
, "%s", t
->name
);
1967 static struct seq_operations show_traces_seq_ops
= {
1974 static int show_traces_open(struct inode
*inode
, struct file
*file
)
1978 if (tracing_disabled
)
1981 ret
= seq_open(file
, &show_traces_seq_ops
);
1983 struct seq_file
*m
= file
->private_data
;
1984 m
->private = trace_types
;
1990 static struct file_operations tracing_fops
= {
1991 .open
= tracing_open
,
1993 .llseek
= seq_lseek
,
1994 .release
= tracing_release
,
1997 static struct file_operations tracing_lt_fops
= {
1998 .open
= tracing_lt_open
,
2000 .llseek
= seq_lseek
,
2001 .release
= tracing_release
,
2004 static struct file_operations show_traces_fops
= {
2005 .open
= show_traces_open
,
2007 .release
= seq_release
,
2011 * Only trace on a CPU if the bitmask is set:
2013 static cpumask_t tracing_cpumask
= CPU_MASK_ALL
;
2016 * When tracing/tracing_cpu_mask is modified then this holds
2017 * the new bitmask we are about to install:
2019 static cpumask_t tracing_cpumask_new
;
2022 * The tracer itself will not take this lock, but still we want
2023 * to provide a consistent cpumask to user-space:
2025 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2028 * Temporary storage for the character representation of the
2029 * CPU bitmask (and one more byte for the newline):
2031 static char mask_str
[NR_CPUS
+ 1];
2034 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2035 size_t count
, loff_t
*ppos
)
2039 mutex_lock(&tracing_cpumask_update_lock
);
2041 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2042 if (count
- len
< 2) {
2046 len
+= sprintf(mask_str
+ len
, "\n");
2047 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2050 mutex_unlock(&tracing_cpumask_update_lock
);
2056 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2057 size_t count
, loff_t
*ppos
)
2061 mutex_lock(&tracing_cpumask_update_lock
);
2062 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2066 raw_local_irq_disable();
2067 __raw_spin_lock(&ftrace_max_lock
);
2068 for_each_tracing_cpu(cpu
) {
2070 * Increase/decrease the disabled counter if we are
2071 * about to flip a bit in the cpumask:
2073 if (cpu_isset(cpu
, tracing_cpumask
) &&
2074 !cpu_isset(cpu
, tracing_cpumask_new
)) {
2075 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2077 if (!cpu_isset(cpu
, tracing_cpumask
) &&
2078 cpu_isset(cpu
, tracing_cpumask_new
)) {
2079 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2082 __raw_spin_unlock(&ftrace_max_lock
);
2083 raw_local_irq_enable();
2085 tracing_cpumask
= tracing_cpumask_new
;
2087 mutex_unlock(&tracing_cpumask_update_lock
);
2092 mutex_unlock(&tracing_cpumask_update_lock
);
2097 static struct file_operations tracing_cpumask_fops
= {
2098 .open
= tracing_open_generic
,
2099 .read
= tracing_cpumask_read
,
2100 .write
= tracing_cpumask_write
,
2104 tracing_iter_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2105 size_t cnt
, loff_t
*ppos
)
2112 /* calulate max size */
2113 for (i
= 0; trace_options
[i
]; i
++) {
2114 len
+= strlen(trace_options
[i
]);
2115 len
+= 3; /* "no" and space */
2118 /* +2 for \n and \0 */
2119 buf
= kmalloc(len
+ 2, GFP_KERNEL
);
2123 for (i
= 0; trace_options
[i
]; i
++) {
2124 if (trace_flags
& (1 << i
))
2125 r
+= sprintf(buf
+ r
, "%s ", trace_options
[i
]);
2127 r
+= sprintf(buf
+ r
, "no%s ", trace_options
[i
]);
2130 r
+= sprintf(buf
+ r
, "\n");
2131 WARN_ON(r
>= len
+ 2);
2133 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2141 tracing_iter_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2142 size_t cnt
, loff_t
*ppos
)
2149 if (cnt
>= sizeof(buf
))
2152 if (copy_from_user(&buf
, ubuf
, cnt
))
2157 if (strncmp(buf
, "no", 2) == 0) {
2162 for (i
= 0; trace_options
[i
]; i
++) {
2163 int len
= strlen(trace_options
[i
]);
2165 if (strncmp(cmp
, trace_options
[i
], len
) == 0) {
2167 trace_flags
&= ~(1 << i
);
2169 trace_flags
|= (1 << i
);
2174 * If no option could be set, return an error:
2176 if (!trace_options
[i
])
2184 static struct file_operations tracing_iter_fops
= {
2185 .open
= tracing_open_generic
,
2186 .read
= tracing_iter_ctrl_read
,
2187 .write
= tracing_iter_ctrl_write
,
2190 static const char readme_msg
[] =
2191 "tracing mini-HOWTO:\n\n"
2193 "# mount -t debugfs nodev /debug\n\n"
2194 "# cat /debug/tracing/available_tracers\n"
2195 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2196 "# cat /debug/tracing/current_tracer\n"
2198 "# echo sched_switch > /debug/tracing/current_tracer\n"
2199 "# cat /debug/tracing/current_tracer\n"
2201 "# cat /debug/tracing/iter_ctrl\n"
2202 "noprint-parent nosym-offset nosym-addr noverbose\n"
2203 "# echo print-parent > /debug/tracing/iter_ctrl\n"
2204 "# echo 1 > /debug/tracing/tracing_enabled\n"
2205 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2206 "echo 0 > /debug/tracing/tracing_enabled\n"
2210 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2211 size_t cnt
, loff_t
*ppos
)
2213 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2214 readme_msg
, strlen(readme_msg
));
2217 static struct file_operations tracing_readme_fops
= {
2218 .open
= tracing_open_generic
,
2219 .read
= tracing_readme_read
,
2223 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2224 size_t cnt
, loff_t
*ppos
)
2226 struct trace_array
*tr
= filp
->private_data
;
2230 r
= sprintf(buf
, "%ld\n", tr
->ctrl
);
2231 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2235 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2236 size_t cnt
, loff_t
*ppos
)
2238 struct trace_array
*tr
= filp
->private_data
;
2243 if (cnt
>= sizeof(buf
))
2246 if (copy_from_user(&buf
, ubuf
, cnt
))
2251 ret
= strict_strtoul(buf
, 10, &val
);
2257 mutex_lock(&trace_types_lock
);
2258 if (tr
->ctrl
^ val
) {
2266 if (current_trace
&& current_trace
->ctrl_update
)
2267 current_trace
->ctrl_update(tr
);
2269 mutex_unlock(&trace_types_lock
);
2277 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2278 size_t cnt
, loff_t
*ppos
)
2280 char buf
[max_tracer_type_len
+2];
2283 mutex_lock(&trace_types_lock
);
2285 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2287 r
= sprintf(buf
, "\n");
2288 mutex_unlock(&trace_types_lock
);
2290 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2294 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2295 size_t cnt
, loff_t
*ppos
)
2297 struct trace_array
*tr
= &global_trace
;
2299 char buf
[max_tracer_type_len
+1];
2302 if (cnt
> max_tracer_type_len
)
2303 cnt
= max_tracer_type_len
;
2305 if (copy_from_user(&buf
, ubuf
, cnt
))
2310 /* strip ending whitespace. */
2311 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2314 mutex_lock(&trace_types_lock
);
2315 for (t
= trace_types
; t
; t
= t
->next
) {
2316 if (strcmp(t
->name
, buf
) == 0)
2319 if (!t
|| t
== current_trace
)
2322 if (current_trace
&& current_trace
->reset
)
2323 current_trace
->reset(tr
);
2330 mutex_unlock(&trace_types_lock
);
2338 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2339 size_t cnt
, loff_t
*ppos
)
2341 unsigned long *ptr
= filp
->private_data
;
2345 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
2346 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2347 if (r
> sizeof(buf
))
2349 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2353 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2354 size_t cnt
, loff_t
*ppos
)
2356 long *ptr
= filp
->private_data
;
2361 if (cnt
>= sizeof(buf
))
2364 if (copy_from_user(&buf
, ubuf
, cnt
))
2369 ret
= strict_strtoul(buf
, 10, &val
);
2378 static atomic_t tracing_reader
;
2380 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2382 struct trace_iterator
*iter
;
2384 if (tracing_disabled
)
2387 /* We only allow for reader of the pipe */
2388 if (atomic_inc_return(&tracing_reader
) != 1) {
2389 atomic_dec(&tracing_reader
);
2393 /* create a buffer to store the information to pass to userspace */
2394 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2398 mutex_lock(&trace_types_lock
);
2399 iter
->tr
= &global_trace
;
2400 iter
->trace
= current_trace
;
2401 filp
->private_data
= iter
;
2403 if (iter
->trace
->pipe_open
)
2404 iter
->trace
->pipe_open(iter
);
2405 mutex_unlock(&trace_types_lock
);
2410 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
2412 struct trace_iterator
*iter
= file
->private_data
;
2415 atomic_dec(&tracing_reader
);
2421 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
2423 struct trace_iterator
*iter
= filp
->private_data
;
2425 if (trace_flags
& TRACE_ITER_BLOCK
) {
2427 * Always select as readable when in blocking mode
2429 return POLLIN
| POLLRDNORM
;
2431 if (!trace_empty(iter
))
2432 return POLLIN
| POLLRDNORM
;
2433 poll_wait(filp
, &trace_wait
, poll_table
);
2434 if (!trace_empty(iter
))
2435 return POLLIN
| POLLRDNORM
;
2445 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
2446 size_t cnt
, loff_t
*ppos
)
2448 struct trace_iterator
*iter
= filp
->private_data
;
2449 struct trace_array_cpu
*data
;
2450 static cpumask_t mask
;
2451 unsigned long flags
;
2452 #ifdef CONFIG_FTRACE
2458 /* return any leftover data */
2459 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
2464 trace_seq_reset(&iter
->seq
);
2466 mutex_lock(&trace_types_lock
);
2467 if (iter
->trace
->read
) {
2468 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
2473 while (trace_empty(iter
)) {
2475 if ((filp
->f_flags
& O_NONBLOCK
)) {
2481 * This is a make-shift waitqueue. The reason we don't use
2482 * an actual wait queue is because:
2483 * 1) we only ever have one waiter
2484 * 2) the tracing, traces all functions, we don't want
2485 * the overhead of calling wake_up and friends
2486 * (and tracing them too)
2487 * Anyway, this is really very primitive wakeup.
2489 set_current_state(TASK_INTERRUPTIBLE
);
2490 iter
->tr
->waiter
= current
;
2492 mutex_unlock(&trace_types_lock
);
2494 /* sleep for 100 msecs, and try again. */
2495 schedule_timeout(HZ
/10);
2497 mutex_lock(&trace_types_lock
);
2499 iter
->tr
->waiter
= NULL
;
2501 if (signal_pending(current
)) {
2506 if (iter
->trace
!= current_trace
)
2510 * We block until we read something and tracing is disabled.
2511 * We still block if tracing is disabled, but we have never
2512 * read anything. This allows a user to cat this file, and
2513 * then enable tracing. But after we have read something,
2514 * we give an EOF when tracing is again disabled.
2516 * iter->pos will be 0 if we haven't read anything.
2518 if (!tracer_enabled
&& iter
->pos
)
2524 /* stop when tracing is finished */
2525 if (trace_empty(iter
))
2528 if (cnt
>= PAGE_SIZE
)
2529 cnt
= PAGE_SIZE
- 1;
2531 /* reset all but tr, trace, and overruns */
2532 memset(&iter
->seq
, 0,
2533 sizeof(struct trace_iterator
) -
2534 offsetof(struct trace_iterator
, seq
));
2538 * We need to stop all tracing on all CPUS to read the
2539 * the next buffer. This is a bit expensive, but is
2540 * not done often. We fill all what we can read,
2541 * and then release the locks again.
2545 local_irq_save(flags
);
2546 #ifdef CONFIG_FTRACE
2547 ftrace_save
= ftrace_enabled
;
2551 for_each_tracing_cpu(cpu
) {
2552 data
= iter
->tr
->data
[cpu
];
2554 if (!head_page(data
) || !data
->trace_idx
)
2557 atomic_inc(&data
->disabled
);
2561 for_each_cpu_mask(cpu
, mask
) {
2562 data
= iter
->tr
->data
[cpu
];
2563 __raw_spin_lock(&data
->lock
);
2565 if (data
->overrun
> iter
->last_overrun
[cpu
])
2566 iter
->overrun
[cpu
] +=
2567 data
->overrun
- iter
->last_overrun
[cpu
];
2568 iter
->last_overrun
[cpu
] = data
->overrun
;
2571 while (find_next_entry_inc(iter
) != NULL
) {
2573 int len
= iter
->seq
.len
;
2575 ret
= print_trace_line(iter
);
2577 /* don't print partial lines */
2578 iter
->seq
.len
= len
;
2582 trace_consume(iter
);
2584 if (iter
->seq
.len
>= cnt
)
2588 for_each_cpu_mask(cpu
, mask
) {
2589 data
= iter
->tr
->data
[cpu
];
2590 __raw_spin_unlock(&data
->lock
);
2593 for_each_cpu_mask(cpu
, mask
) {
2594 data
= iter
->tr
->data
[cpu
];
2595 atomic_dec(&data
->disabled
);
2597 #ifdef CONFIG_FTRACE
2598 ftrace_enabled
= ftrace_save
;
2600 local_irq_restore(flags
);
2602 /* Now copy what we have to the user */
2603 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
2604 if (iter
->seq
.readpos
>= iter
->seq
.len
)
2605 trace_seq_reset(&iter
->seq
);
2610 mutex_unlock(&trace_types_lock
);
2616 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
2617 size_t cnt
, loff_t
*ppos
)
2619 struct trace_array
*tr
= filp
->private_data
;
2623 r
= sprintf(buf
, "%lu\n", tr
->entries
);
2624 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2628 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
2629 size_t cnt
, loff_t
*ppos
)
2635 if (cnt
>= sizeof(buf
))
2638 if (copy_from_user(&buf
, ubuf
, cnt
))
2643 ret
= strict_strtoul(buf
, 10, &val
);
2647 /* must have at least 1 entry */
2651 mutex_lock(&trace_types_lock
);
2653 if (current_trace
!= &no_tracer
) {
2655 pr_info("ftrace: set current_tracer to none"
2656 " before modifying buffer size\n");
2660 if (val
> global_trace
.entries
) {
2661 long pages_requested
;
2662 unsigned long freeable_pages
;
2664 /* make sure we have enough memory before mapping */
2666 (val
+ (ENTRIES_PER_PAGE
-1)) / ENTRIES_PER_PAGE
;
2668 /* account for each buffer (and max_tr) */
2669 pages_requested
*= tracing_nr_buffers
* 2;
2671 /* Check for overflow */
2672 if (pages_requested
< 0) {
2677 freeable_pages
= determine_dirtyable_memory();
2679 /* we only allow to request 1/4 of useable memory */
2680 if (pages_requested
>
2681 ((freeable_pages
+ tracing_pages_allocated
) / 4)) {
2686 while (global_trace
.entries
< val
) {
2687 if (trace_alloc_page()) {
2691 /* double check that we don't go over the known pages */
2692 if (tracing_pages_allocated
> pages_requested
)
2697 /* include the number of entries in val (inc of page entries) */
2698 while (global_trace
.entries
> val
+ (ENTRIES_PER_PAGE
- 1))
2702 /* check integrity */
2703 for_each_tracing_cpu(i
)
2704 check_pages(global_trace
.data
[i
]);
2708 /* If check pages failed, return ENOMEM */
2709 if (tracing_disabled
)
2712 max_tr
.entries
= global_trace
.entries
;
2713 mutex_unlock(&trace_types_lock
);
2718 static struct file_operations tracing_max_lat_fops
= {
2719 .open
= tracing_open_generic
,
2720 .read
= tracing_max_lat_read
,
2721 .write
= tracing_max_lat_write
,
2724 static struct file_operations tracing_ctrl_fops
= {
2725 .open
= tracing_open_generic
,
2726 .read
= tracing_ctrl_read
,
2727 .write
= tracing_ctrl_write
,
2730 static struct file_operations set_tracer_fops
= {
2731 .open
= tracing_open_generic
,
2732 .read
= tracing_set_trace_read
,
2733 .write
= tracing_set_trace_write
,
2736 static struct file_operations tracing_pipe_fops
= {
2737 .open
= tracing_open_pipe
,
2738 .poll
= tracing_poll_pipe
,
2739 .read
= tracing_read_pipe
,
2740 .release
= tracing_release_pipe
,
2743 static struct file_operations tracing_entries_fops
= {
2744 .open
= tracing_open_generic
,
2745 .read
= tracing_entries_read
,
2746 .write
= tracing_entries_write
,
2749 #ifdef CONFIG_DYNAMIC_FTRACE
2752 tracing_read_long(struct file
*filp
, char __user
*ubuf
,
2753 size_t cnt
, loff_t
*ppos
)
2755 unsigned long *p
= filp
->private_data
;
2759 r
= sprintf(buf
, "%ld\n", *p
);
2761 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2764 static struct file_operations tracing_read_long_fops
= {
2765 .open
= tracing_open_generic
,
2766 .read
= tracing_read_long
,
2770 static struct dentry
*d_tracer
;
2772 struct dentry
*tracing_init_dentry(void)
2779 d_tracer
= debugfs_create_dir("tracing", NULL
);
2781 if (!d_tracer
&& !once
) {
2783 pr_warning("Could not create debugfs directory 'tracing'\n");
2790 #ifdef CONFIG_FTRACE_SELFTEST
2791 /* Let selftest have access to static functions in this file */
2792 #include "trace_selftest.c"
2795 static __init
void tracer_init_debugfs(void)
2797 struct dentry
*d_tracer
;
2798 struct dentry
*entry
;
2800 d_tracer
= tracing_init_dentry();
2802 entry
= debugfs_create_file("tracing_enabled", 0644, d_tracer
,
2803 &global_trace
, &tracing_ctrl_fops
);
2805 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2807 entry
= debugfs_create_file("iter_ctrl", 0644, d_tracer
,
2808 NULL
, &tracing_iter_fops
);
2810 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2812 entry
= debugfs_create_file("tracing_cpumask", 0644, d_tracer
,
2813 NULL
, &tracing_cpumask_fops
);
2815 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2817 entry
= debugfs_create_file("latency_trace", 0444, d_tracer
,
2818 &global_trace
, &tracing_lt_fops
);
2820 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2822 entry
= debugfs_create_file("trace", 0444, d_tracer
,
2823 &global_trace
, &tracing_fops
);
2825 pr_warning("Could not create debugfs 'trace' entry\n");
2827 entry
= debugfs_create_file("available_tracers", 0444, d_tracer
,
2828 &global_trace
, &show_traces_fops
);
2830 pr_warning("Could not create debugfs 'trace' entry\n");
2832 entry
= debugfs_create_file("current_tracer", 0444, d_tracer
,
2833 &global_trace
, &set_tracer_fops
);
2835 pr_warning("Could not create debugfs 'trace' entry\n");
2837 entry
= debugfs_create_file("tracing_max_latency", 0644, d_tracer
,
2838 &tracing_max_latency
,
2839 &tracing_max_lat_fops
);
2841 pr_warning("Could not create debugfs "
2842 "'tracing_max_latency' entry\n");
2844 entry
= debugfs_create_file("tracing_thresh", 0644, d_tracer
,
2845 &tracing_thresh
, &tracing_max_lat_fops
);
2847 pr_warning("Could not create debugfs "
2848 "'tracing_threash' entry\n");
2849 entry
= debugfs_create_file("README", 0644, d_tracer
,
2850 NULL
, &tracing_readme_fops
);
2852 pr_warning("Could not create debugfs 'README' entry\n");
2854 entry
= debugfs_create_file("trace_pipe", 0644, d_tracer
,
2855 NULL
, &tracing_pipe_fops
);
2857 pr_warning("Could not create debugfs "
2858 "'tracing_threash' entry\n");
2860 entry
= debugfs_create_file("trace_entries", 0644, d_tracer
,
2861 &global_trace
, &tracing_entries_fops
);
2863 pr_warning("Could not create debugfs "
2864 "'tracing_threash' entry\n");
2866 #ifdef CONFIG_DYNAMIC_FTRACE
2867 entry
= debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
2868 &ftrace_update_tot_cnt
,
2869 &tracing_read_long_fops
);
2871 pr_warning("Could not create debugfs "
2872 "'dyn_ftrace_total_info' entry\n");
2876 static int trace_alloc_page(void)
2878 struct trace_array_cpu
*data
;
2879 struct page
*page
, *tmp
;
2882 unsigned pages_allocated
= 0;
2885 /* first allocate a page for each CPU */
2886 for_each_tracing_cpu(i
) {
2887 array
= (void *)__get_free_page(GFP_KERNEL
);
2888 if (array
== NULL
) {
2889 printk(KERN_ERR
"tracer: failed to allocate page"
2890 "for trace buffer!\n");
2895 page
= virt_to_page(array
);
2896 list_add(&page
->lru
, &pages
);
2898 /* Only allocate if we are actually using the max trace */
2899 #ifdef CONFIG_TRACER_MAX_TRACE
2900 array
= (void *)__get_free_page(GFP_KERNEL
);
2901 if (array
== NULL
) {
2902 printk(KERN_ERR
"tracer: failed to allocate page"
2903 "for trace buffer!\n");
2907 page
= virt_to_page(array
);
2908 list_add(&page
->lru
, &pages
);
2912 /* Now that we successfully allocate a page per CPU, add them */
2913 for_each_tracing_cpu(i
) {
2914 data
= global_trace
.data
[i
];
2915 page
= list_entry(pages
.next
, struct page
, lru
);
2916 list_del_init(&page
->lru
);
2917 list_add_tail(&page
->lru
, &data
->trace_pages
);
2920 #ifdef CONFIG_TRACER_MAX_TRACE
2921 data
= max_tr
.data
[i
];
2922 page
= list_entry(pages
.next
, struct page
, lru
);
2923 list_del_init(&page
->lru
);
2924 list_add_tail(&page
->lru
, &data
->trace_pages
);
2928 tracing_pages_allocated
+= pages_allocated
;
2929 global_trace
.entries
+= ENTRIES_PER_PAGE
;
2934 list_for_each_entry_safe(page
, tmp
, &pages
, lru
) {
2935 list_del_init(&page
->lru
);
2941 static int trace_free_page(void)
2943 struct trace_array_cpu
*data
;
2945 struct list_head
*p
;
2949 /* free one page from each buffer */
2950 for_each_tracing_cpu(i
) {
2951 data
= global_trace
.data
[i
];
2952 p
= data
->trace_pages
.next
;
2953 if (p
== &data
->trace_pages
) {
2954 /* should never happen */
2956 tracing_disabled
= 1;
2960 page
= list_entry(p
, struct page
, lru
);
2962 list_del(&page
->lru
);
2963 tracing_pages_allocated
--;
2964 tracing_pages_allocated
--;
2967 tracing_reset(data
);
2969 #ifdef CONFIG_TRACER_MAX_TRACE
2970 data
= max_tr
.data
[i
];
2971 p
= data
->trace_pages
.next
;
2972 if (p
== &data
->trace_pages
) {
2973 /* should never happen */
2975 tracing_disabled
= 1;
2979 page
= list_entry(p
, struct page
, lru
);
2981 list_del(&page
->lru
);
2984 tracing_reset(data
);
2987 global_trace
.entries
-= ENTRIES_PER_PAGE
;
2992 __init
static int tracer_alloc_buffers(void)
2994 struct trace_array_cpu
*data
;
3001 /* TODO: make the number of buffers hot pluggable with CPUS */
3002 tracing_nr_buffers
= num_possible_cpus();
3003 tracing_buffer_mask
= cpu_possible_map
;
3005 /* Allocate the first page for all buffers */
3006 for_each_tracing_cpu(i
) {
3007 data
= global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
3008 max_tr
.data
[i
] = &per_cpu(max_data
, i
);
3010 array
= (void *)__get_free_page(GFP_KERNEL
);
3011 if (array
== NULL
) {
3012 printk(KERN_ERR
"tracer: failed to allocate page"
3013 "for trace buffer!\n");
3017 /* set the array to the list */
3018 INIT_LIST_HEAD(&data
->trace_pages
);
3019 page
= virt_to_page(array
);
3020 list_add(&page
->lru
, &data
->trace_pages
);
3021 /* use the LRU flag to differentiate the two buffers */
3024 data
->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
3025 max_tr
.data
[i
]->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
3027 /* Only allocate if we are actually using the max trace */
3028 #ifdef CONFIG_TRACER_MAX_TRACE
3029 array
= (void *)__get_free_page(GFP_KERNEL
);
3030 if (array
== NULL
) {
3031 printk(KERN_ERR
"tracer: failed to allocate page"
3032 "for trace buffer!\n");
3036 INIT_LIST_HEAD(&max_tr
.data
[i
]->trace_pages
);
3037 page
= virt_to_page(array
);
3038 list_add(&page
->lru
, &max_tr
.data
[i
]->trace_pages
);
3044 * Since we allocate by orders of pages, we may be able to
3047 global_trace
.entries
= ENTRIES_PER_PAGE
;
3050 while (global_trace
.entries
< trace_nr_entries
) {
3051 if (trace_alloc_page())
3055 max_tr
.entries
= global_trace
.entries
;
3057 pr_info("tracer: %d pages allocated for %ld entries of %ld bytes\n",
3058 pages
, trace_nr_entries
, (long)TRACE_ENTRY_SIZE
);
3059 pr_info(" actual entries %ld\n", global_trace
.entries
);
3061 tracer_init_debugfs();
3063 trace_init_cmdlines();
3065 register_tracer(&no_tracer
);
3066 current_trace
= &no_tracer
;
3068 /* All seems OK, enable tracing */
3069 global_trace
.ctrl
= tracer_enabled
;
3070 tracing_disabled
= 0;
3075 for (i
-- ; i
>= 0; i
--) {
3076 struct page
*page
, *tmp
;
3077 struct trace_array_cpu
*data
= global_trace
.data
[i
];
3080 list_for_each_entry_safe(page
, tmp
,
3081 &data
->trace_pages
, lru
) {
3082 list_del_init(&page
->lru
);
3087 #ifdef CONFIG_TRACER_MAX_TRACE
3088 data
= max_tr
.data
[i
];
3090 list_for_each_entry_safe(page
, tmp
,
3091 &data
->trace_pages
, lru
) {
3092 list_del_init(&page
->lru
);
3100 fs_initcall(tracer_alloc_buffers
);