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;
99 /* function tracing enabled */
100 int ftrace_function_enabled
;
103 * trace_nr_entries is the number of entries that is allocated
104 * for a buffer. Note, the number of entries is always rounded
105 * to ENTRIES_PER_PAGE.
107 static unsigned long trace_nr_entries
= 65536UL;
109 /* trace_types holds a link list of available tracers. */
110 static struct tracer
*trace_types __read_mostly
;
112 /* current_trace points to the tracer that is currently active */
113 static struct tracer
*current_trace __read_mostly
;
116 * max_tracer_type_len is used to simplify the allocating of
117 * buffers to read userspace tracer names. We keep track of
118 * the longest tracer name registered.
120 static int max_tracer_type_len
;
123 * trace_types_lock is used to protect the trace_types list.
124 * This lock is also used to keep user access serialized.
125 * Accesses from userspace will grab this lock while userspace
126 * activities happen inside the kernel.
128 static DEFINE_MUTEX(trace_types_lock
);
130 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
131 static DECLARE_WAIT_QUEUE_HEAD(trace_wait
);
133 /* trace_flags holds iter_ctrl options */
134 unsigned long trace_flags
= TRACE_ITER_PRINT_PARENT
;
136 static notrace
void no_trace_init(struct trace_array
*tr
)
140 ftrace_function_enabled
= 0;
142 for_each_online_cpu(cpu
)
143 tracing_reset(tr
->data
[cpu
]);
147 /* dummy trace to disable tracing */
148 static struct tracer no_tracer __read_mostly
= {
150 .init
= no_trace_init
155 * trace_wake_up - wake up tasks waiting for trace input
157 * Simply wakes up any task that is blocked on the trace_wait
158 * queue. These is used with trace_poll for tasks polling the trace.
160 void trace_wake_up(void)
163 * The runqueue_is_locked() can fail, but this is the best we
166 if (!(trace_flags
& TRACE_ITER_BLOCK
) && !runqueue_is_locked())
167 wake_up(&trace_wait
);
170 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
172 static int __init
set_nr_entries(char *str
)
174 unsigned long nr_entries
;
179 ret
= strict_strtoul(str
, 0, &nr_entries
);
180 /* nr_entries can not be zero */
181 if (ret
< 0 || nr_entries
== 0)
183 trace_nr_entries
= nr_entries
;
186 __setup("trace_entries=", set_nr_entries
);
188 unsigned long nsecs_to_usecs(unsigned long nsecs
)
194 * trace_flag_type is an enumeration that holds different
195 * states when a trace occurs. These are:
196 * IRQS_OFF - interrupts were disabled
197 * NEED_RESCED - reschedule is requested
198 * HARDIRQ - inside an interrupt handler
199 * SOFTIRQ - inside a softirq handler
201 enum trace_flag_type
{
202 TRACE_FLAG_IRQS_OFF
= 0x01,
203 TRACE_FLAG_NEED_RESCHED
= 0x02,
204 TRACE_FLAG_HARDIRQ
= 0x04,
205 TRACE_FLAG_SOFTIRQ
= 0x08,
209 * TRACE_ITER_SYM_MASK masks the options in trace_flags that
210 * control the output of kernel symbols.
212 #define TRACE_ITER_SYM_MASK \
213 (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
215 /* These must match the bit postions in trace_iterator_flags */
216 static const char *trace_options
[] = {
231 * ftrace_max_lock is used to protect the swapping of buffers
232 * when taking a max snapshot. The buffers themselves are
233 * protected by per_cpu spinlocks. But the action of the swap
234 * needs its own lock.
236 * This is defined as a raw_spinlock_t in order to help
237 * with performance when lockdep debugging is enabled.
239 static raw_spinlock_t ftrace_max_lock
=
240 (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
243 * Copy the new maximum trace into the separate maximum-trace
244 * structure. (this way the maximum trace is permanently saved,
245 * for later retrieval via /debugfs/tracing/latency_trace)
248 __update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
250 struct trace_array_cpu
*data
= tr
->data
[cpu
];
253 max_tr
.time_start
= data
->preempt_timestamp
;
255 data
= max_tr
.data
[cpu
];
256 data
->saved_latency
= tracing_max_latency
;
258 memcpy(data
->comm
, tsk
->comm
, TASK_COMM_LEN
);
259 data
->pid
= tsk
->pid
;
260 data
->uid
= tsk
->uid
;
261 data
->nice
= tsk
->static_prio
- 20 - MAX_RT_PRIO
;
262 data
->policy
= tsk
->policy
;
263 data
->rt_priority
= tsk
->rt_priority
;
265 /* record this tasks comm */
266 tracing_record_cmdline(current
);
269 #define CHECK_COND(cond) \
270 if (unlikely(cond)) { \
271 tracing_disabled = 1; \
277 * check_pages - integrity check of trace buffers
279 * As a safty measure we check to make sure the data pages have not
282 int check_pages(struct trace_array_cpu
*data
)
284 struct page
*page
, *tmp
;
286 CHECK_COND(data
->trace_pages
.next
->prev
!= &data
->trace_pages
);
287 CHECK_COND(data
->trace_pages
.prev
->next
!= &data
->trace_pages
);
289 list_for_each_entry_safe(page
, tmp
, &data
->trace_pages
, lru
) {
290 CHECK_COND(page
->lru
.next
->prev
!= &page
->lru
);
291 CHECK_COND(page
->lru
.prev
->next
!= &page
->lru
);
298 * head_page - page address of the first page in per_cpu buffer.
300 * head_page returns the page address of the first page in
301 * a per_cpu buffer. This also preforms various consistency
302 * checks to make sure the buffer has not been corrupted.
304 void *head_page(struct trace_array_cpu
*data
)
308 if (list_empty(&data
->trace_pages
))
311 page
= list_entry(data
->trace_pages
.next
, struct page
, lru
);
312 BUG_ON(&page
->lru
== &data
->trace_pages
);
314 return page_address(page
);
318 * trace_seq_printf - sequence printing of trace information
319 * @s: trace sequence descriptor
320 * @fmt: printf format string
322 * The tracer may use either sequence operations or its own
323 * copy to user routines. To simplify formating of a trace
324 * trace_seq_printf is used to store strings into a special
325 * buffer (@s). Then the output may be either used by
326 * the sequencer or pulled into another buffer.
329 trace_seq_printf(struct trace_seq
*s
, const char *fmt
, ...)
331 int len
= (PAGE_SIZE
- 1) - s
->len
;
339 ret
= vsnprintf(s
->buffer
+ s
->len
, len
, fmt
, ap
);
342 /* If we can't write it all, don't bother writing anything */
352 * trace_seq_puts - trace sequence printing of simple string
353 * @s: trace sequence descriptor
354 * @str: simple string to record
356 * The tracer may use either the sequence operations or its own
357 * copy to user routines. This function records a simple string
358 * into a special buffer (@s) for later retrieval by a sequencer
359 * or other mechanism.
362 trace_seq_puts(struct trace_seq
*s
, const char *str
)
364 int len
= strlen(str
);
366 if (len
> ((PAGE_SIZE
- 1) - s
->len
))
369 memcpy(s
->buffer
+ s
->len
, str
, len
);
376 trace_seq_putc(struct trace_seq
*s
, unsigned char c
)
378 if (s
->len
>= (PAGE_SIZE
- 1))
381 s
->buffer
[s
->len
++] = c
;
387 trace_seq_putmem(struct trace_seq
*s
, void *mem
, size_t len
)
389 if (len
> ((PAGE_SIZE
- 1) - s
->len
))
392 memcpy(s
->buffer
+ s
->len
, mem
, len
);
399 static const char hex2asc
[] = "0123456789abcdef";
402 trace_seq_putmem_hex(struct trace_seq
*s
, void *mem
, size_t len
)
404 unsigned char hex
[HEX_CHARS
];
405 unsigned char *data
= mem
;
409 BUG_ON(len
>= HEX_CHARS
);
412 for (i
= 0, j
= 0; i
< len
; i
++) {
414 for (i
= len
-1, j
= 0; i
>= 0; i
--) {
418 hex
[j
++] = hex2asc
[byte
& 0x0f];
419 hex
[j
++] = hex2asc
[byte
>> 4];
423 return trace_seq_putmem(s
, hex
, j
);
427 trace_seq_reset(struct trace_seq
*s
)
433 ssize_t
trace_seq_to_user(struct trace_seq
*s
, char __user
*ubuf
, size_t cnt
)
438 if (s
->len
<= s
->readpos
)
441 len
= s
->len
- s
->readpos
;
444 ret
= copy_to_user(ubuf
, s
->buffer
+ s
->readpos
, cnt
);
453 trace_print_seq(struct seq_file
*m
, struct trace_seq
*s
)
455 int len
= s
->len
>= PAGE_SIZE
? PAGE_SIZE
- 1 : s
->len
;
458 seq_puts(m
, s
->buffer
);
464 * flip the trace buffers between two trace descriptors.
465 * This usually is the buffers between the global_trace and
466 * the max_tr to record a snapshot of a current trace.
468 * The ftrace_max_lock must be held.
471 flip_trace(struct trace_array_cpu
*tr1
, struct trace_array_cpu
*tr2
)
473 struct list_head flip_pages
;
475 INIT_LIST_HEAD(&flip_pages
);
477 memcpy(&tr1
->trace_head_idx
, &tr2
->trace_head_idx
,
478 sizeof(struct trace_array_cpu
) -
479 offsetof(struct trace_array_cpu
, trace_head_idx
));
483 list_splice_init(&tr1
->trace_pages
, &flip_pages
);
484 list_splice_init(&tr2
->trace_pages
, &tr1
->trace_pages
);
485 list_splice_init(&flip_pages
, &tr2
->trace_pages
);
486 BUG_ON(!list_empty(&flip_pages
));
492 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
494 * @tsk: the task with the latency
495 * @cpu: The cpu that initiated the trace.
497 * Flip the buffers between the @tr and the max_tr and record information
498 * about which task was the cause of this latency.
501 update_max_tr(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
503 struct trace_array_cpu
*data
;
506 WARN_ON_ONCE(!irqs_disabled());
507 __raw_spin_lock(&ftrace_max_lock
);
508 /* clear out all the previous traces */
509 for_each_tracing_cpu(i
) {
511 flip_trace(max_tr
.data
[i
], data
);
515 __update_max_tr(tr
, tsk
, cpu
);
516 __raw_spin_unlock(&ftrace_max_lock
);
520 * update_max_tr_single - only copy one trace over, and reset the rest
522 * @tsk - task with the latency
523 * @cpu - the cpu of the buffer to copy.
525 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
528 update_max_tr_single(struct trace_array
*tr
, struct task_struct
*tsk
, int cpu
)
530 struct trace_array_cpu
*data
= tr
->data
[cpu
];
533 WARN_ON_ONCE(!irqs_disabled());
534 __raw_spin_lock(&ftrace_max_lock
);
535 for_each_tracing_cpu(i
)
536 tracing_reset(max_tr
.data
[i
]);
538 flip_trace(max_tr
.data
[cpu
], data
);
541 __update_max_tr(tr
, tsk
, cpu
);
542 __raw_spin_unlock(&ftrace_max_lock
);
546 * register_tracer - register a tracer with the ftrace system.
547 * @type - the plugin for the tracer
549 * Register a new plugin tracer.
551 int register_tracer(struct tracer
*type
)
558 pr_info("Tracer must have a name\n");
562 mutex_lock(&trace_types_lock
);
563 for (t
= trace_types
; t
; t
= t
->next
) {
564 if (strcmp(type
->name
, t
->name
) == 0) {
566 pr_info("Trace %s already registered\n",
573 #ifdef CONFIG_FTRACE_STARTUP_TEST
574 if (type
->selftest
) {
575 struct tracer
*saved_tracer
= current_trace
;
576 struct trace_array_cpu
*data
;
577 struct trace_array
*tr
= &global_trace
;
578 int saved_ctrl
= tr
->ctrl
;
581 * Run a selftest on this tracer.
582 * Here we reset the trace buffer, and set the current
583 * tracer to be this tracer. The tracer can then run some
584 * internal tracing to verify that everything is in order.
585 * If we fail, we do not register this tracer.
587 for_each_tracing_cpu(i
) {
589 if (!head_page(data
))
593 current_trace
= type
;
595 /* the test is responsible for initializing and enabling */
596 pr_info("Testing tracer %s: ", type
->name
);
597 ret
= type
->selftest(type
, tr
);
598 /* the test is responsible for resetting too */
599 current_trace
= saved_tracer
;
600 tr
->ctrl
= saved_ctrl
;
602 printk(KERN_CONT
"FAILED!\n");
605 /* Only reset on passing, to avoid touching corrupted buffers */
606 for_each_tracing_cpu(i
) {
608 if (!head_page(data
))
612 printk(KERN_CONT
"PASSED\n");
616 type
->next
= trace_types
;
618 len
= strlen(type
->name
);
619 if (len
> max_tracer_type_len
)
620 max_tracer_type_len
= len
;
623 mutex_unlock(&trace_types_lock
);
628 void unregister_tracer(struct tracer
*type
)
633 mutex_lock(&trace_types_lock
);
634 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
638 pr_info("Trace %s not registered\n", type
->name
);
643 if (strlen(type
->name
) != max_tracer_type_len
)
646 max_tracer_type_len
= 0;
647 for (t
= &trace_types
; *t
; t
= &(*t
)->next
) {
648 len
= strlen((*t
)->name
);
649 if (len
> max_tracer_type_len
)
650 max_tracer_type_len
= len
;
653 mutex_unlock(&trace_types_lock
);
656 void tracing_reset(struct trace_array_cpu
*data
)
660 data
->trace_head
= data
->trace_tail
= head_page(data
);
661 data
->trace_head_idx
= 0;
662 data
->trace_tail_idx
= 0;
665 #define SAVED_CMDLINES 128
666 static unsigned map_pid_to_cmdline
[PID_MAX_DEFAULT
+1];
667 static unsigned map_cmdline_to_pid
[SAVED_CMDLINES
];
668 static char saved_cmdlines
[SAVED_CMDLINES
][TASK_COMM_LEN
];
669 static int cmdline_idx
;
670 static DEFINE_SPINLOCK(trace_cmdline_lock
);
672 /* temporary disable recording */
673 atomic_t trace_record_cmdline_disabled __read_mostly
;
675 static void trace_init_cmdlines(void)
677 memset(&map_pid_to_cmdline
, -1, sizeof(map_pid_to_cmdline
));
678 memset(&map_cmdline_to_pid
, -1, sizeof(map_cmdline_to_pid
));
682 void trace_stop_cmdline_recording(void);
684 static void trace_save_cmdline(struct task_struct
*tsk
)
689 if (!tsk
->pid
|| unlikely(tsk
->pid
> PID_MAX_DEFAULT
))
693 * It's not the end of the world if we don't get
694 * the lock, but we also don't want to spin
695 * nor do we want to disable interrupts,
696 * so if we miss here, then better luck next time.
698 if (!spin_trylock(&trace_cmdline_lock
))
701 idx
= map_pid_to_cmdline
[tsk
->pid
];
702 if (idx
>= SAVED_CMDLINES
) {
703 idx
= (cmdline_idx
+ 1) % SAVED_CMDLINES
;
705 map
= map_cmdline_to_pid
[idx
];
706 if (map
<= PID_MAX_DEFAULT
)
707 map_pid_to_cmdline
[map
] = (unsigned)-1;
709 map_pid_to_cmdline
[tsk
->pid
] = idx
;
714 memcpy(&saved_cmdlines
[idx
], tsk
->comm
, TASK_COMM_LEN
);
716 spin_unlock(&trace_cmdline_lock
);
719 static char *trace_find_cmdline(int pid
)
721 char *cmdline
= "<...>";
727 if (pid
> PID_MAX_DEFAULT
)
730 map
= map_pid_to_cmdline
[pid
];
731 if (map
>= SAVED_CMDLINES
)
734 cmdline
= saved_cmdlines
[map
];
740 void tracing_record_cmdline(struct task_struct
*tsk
)
742 if (atomic_read(&trace_record_cmdline_disabled
))
745 trace_save_cmdline(tsk
);
748 static inline struct list_head
*
749 trace_next_list(struct trace_array_cpu
*data
, struct list_head
*next
)
752 * Roundrobin - but skip the head (which is not a real page):
755 if (unlikely(next
== &data
->trace_pages
))
757 BUG_ON(next
== &data
->trace_pages
);
763 trace_next_page(struct trace_array_cpu
*data
, void *addr
)
765 struct list_head
*next
;
768 page
= virt_to_page(addr
);
770 next
= trace_next_list(data
, &page
->lru
);
771 page
= list_entry(next
, struct page
, lru
);
773 return page_address(page
);
776 static inline struct trace_entry
*
777 tracing_get_trace_entry(struct trace_array
*tr
, struct trace_array_cpu
*data
)
779 unsigned long idx
, idx_next
;
780 struct trace_entry
*entry
;
783 idx
= data
->trace_head_idx
;
786 BUG_ON(idx
* TRACE_ENTRY_SIZE
>= PAGE_SIZE
);
788 entry
= data
->trace_head
+ idx
* TRACE_ENTRY_SIZE
;
790 if (unlikely(idx_next
>= ENTRIES_PER_PAGE
)) {
791 data
->trace_head
= trace_next_page(data
, data
->trace_head
);
795 if (data
->trace_head
== data
->trace_tail
&&
796 idx_next
== data
->trace_tail_idx
) {
799 data
->trace_tail_idx
++;
800 if (data
->trace_tail_idx
>= ENTRIES_PER_PAGE
) {
802 trace_next_page(data
, data
->trace_tail
);
803 data
->trace_tail_idx
= 0;
807 data
->trace_head_idx
= idx_next
;
813 tracing_generic_entry_update(struct trace_entry
*entry
, unsigned long flags
)
815 struct task_struct
*tsk
= current
;
818 pc
= preempt_count();
820 entry
->preempt_count
= pc
& 0xff;
821 entry
->pid
= (tsk
) ? tsk
->pid
: 0;
822 entry
->t
= ftrace_now(raw_smp_processor_id());
823 entry
->flags
= (irqs_disabled_flags(flags
) ? TRACE_FLAG_IRQS_OFF
: 0) |
824 ((pc
& HARDIRQ_MASK
) ? TRACE_FLAG_HARDIRQ
: 0) |
825 ((pc
& SOFTIRQ_MASK
) ? TRACE_FLAG_SOFTIRQ
: 0) |
826 (need_resched() ? TRACE_FLAG_NEED_RESCHED
: 0);
830 trace_function(struct trace_array
*tr
, struct trace_array_cpu
*data
,
831 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
)
833 struct trace_entry
*entry
;
834 unsigned long irq_flags
;
836 raw_local_irq_save(irq_flags
);
837 __raw_spin_lock(&data
->lock
);
838 entry
= tracing_get_trace_entry(tr
, data
);
839 tracing_generic_entry_update(entry
, flags
);
840 entry
->type
= TRACE_FN
;
842 entry
->fn
.parent_ip
= parent_ip
;
843 __raw_spin_unlock(&data
->lock
);
844 raw_local_irq_restore(irq_flags
);
848 ftrace(struct trace_array
*tr
, struct trace_array_cpu
*data
,
849 unsigned long ip
, unsigned long parent_ip
, unsigned long flags
)
851 if (likely(!atomic_read(&data
->disabled
)))
852 trace_function(tr
, data
, ip
, parent_ip
, flags
);
855 #ifdef CONFIG_MMIOTRACE
856 void __trace_mmiotrace_rw(struct trace_array
*tr
, struct trace_array_cpu
*data
,
857 struct mmiotrace_rw
*rw
)
859 struct trace_entry
*entry
;
860 unsigned long irq_flags
;
862 raw_local_irq_save(irq_flags
);
863 __raw_spin_lock(&data
->lock
);
865 entry
= tracing_get_trace_entry(tr
, data
);
866 tracing_generic_entry_update(entry
, 0);
867 entry
->type
= TRACE_MMIO_RW
;
870 __raw_spin_unlock(&data
->lock
);
871 raw_local_irq_restore(irq_flags
);
876 void __trace_mmiotrace_map(struct trace_array
*tr
, struct trace_array_cpu
*data
,
877 struct mmiotrace_map
*map
)
879 struct trace_entry
*entry
;
880 unsigned long irq_flags
;
882 raw_local_irq_save(irq_flags
);
883 __raw_spin_lock(&data
->lock
);
885 entry
= tracing_get_trace_entry(tr
, data
);
886 tracing_generic_entry_update(entry
, 0);
887 entry
->type
= TRACE_MMIO_MAP
;
888 entry
->mmiomap
= *map
;
890 __raw_spin_unlock(&data
->lock
);
891 raw_local_irq_restore(irq_flags
);
897 void __trace_stack(struct trace_array
*tr
,
898 struct trace_array_cpu
*data
,
902 struct trace_entry
*entry
;
903 struct stack_trace trace
;
905 if (!(trace_flags
& TRACE_ITER_STACKTRACE
))
908 entry
= tracing_get_trace_entry(tr
, data
);
909 tracing_generic_entry_update(entry
, flags
);
910 entry
->type
= TRACE_STACK
;
912 memset(&entry
->stack
, 0, sizeof(entry
->stack
));
914 trace
.nr_entries
= 0;
915 trace
.max_entries
= FTRACE_STACK_ENTRIES
;
917 trace
.entries
= entry
->stack
.caller
;
919 save_stack_trace(&trace
);
923 __trace_special(void *__tr
, void *__data
,
924 unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
926 struct trace_array_cpu
*data
= __data
;
927 struct trace_array
*tr
= __tr
;
928 struct trace_entry
*entry
;
929 unsigned long irq_flags
;
931 raw_local_irq_save(irq_flags
);
932 __raw_spin_lock(&data
->lock
);
933 entry
= tracing_get_trace_entry(tr
, data
);
934 tracing_generic_entry_update(entry
, 0);
935 entry
->type
= TRACE_SPECIAL
;
936 entry
->special
.arg1
= arg1
;
937 entry
->special
.arg2
= arg2
;
938 entry
->special
.arg3
= arg3
;
939 __trace_stack(tr
, data
, irq_flags
, 4);
940 __raw_spin_unlock(&data
->lock
);
941 raw_local_irq_restore(irq_flags
);
947 tracing_sched_switch_trace(struct trace_array
*tr
,
948 struct trace_array_cpu
*data
,
949 struct task_struct
*prev
,
950 struct task_struct
*next
,
953 struct trace_entry
*entry
;
954 unsigned long irq_flags
;
956 raw_local_irq_save(irq_flags
);
957 __raw_spin_lock(&data
->lock
);
958 entry
= tracing_get_trace_entry(tr
, data
);
959 tracing_generic_entry_update(entry
, flags
);
960 entry
->type
= TRACE_CTX
;
961 entry
->ctx
.prev_pid
= prev
->pid
;
962 entry
->ctx
.prev_prio
= prev
->prio
;
963 entry
->ctx
.prev_state
= prev
->state
;
964 entry
->ctx
.next_pid
= next
->pid
;
965 entry
->ctx
.next_prio
= next
->prio
;
966 entry
->ctx
.next_state
= next
->state
;
967 __trace_stack(tr
, data
, flags
, 5);
968 __raw_spin_unlock(&data
->lock
);
969 raw_local_irq_restore(irq_flags
);
973 tracing_sched_wakeup_trace(struct trace_array
*tr
,
974 struct trace_array_cpu
*data
,
975 struct task_struct
*wakee
,
976 struct task_struct
*curr
,
979 struct trace_entry
*entry
;
980 unsigned long irq_flags
;
982 raw_local_irq_save(irq_flags
);
983 __raw_spin_lock(&data
->lock
);
984 entry
= tracing_get_trace_entry(tr
, data
);
985 tracing_generic_entry_update(entry
, flags
);
986 entry
->type
= TRACE_WAKE
;
987 entry
->ctx
.prev_pid
= curr
->pid
;
988 entry
->ctx
.prev_prio
= curr
->prio
;
989 entry
->ctx
.prev_state
= curr
->state
;
990 entry
->ctx
.next_pid
= wakee
->pid
;
991 entry
->ctx
.next_prio
= wakee
->prio
;
992 entry
->ctx
.next_state
= wakee
->state
;
993 __trace_stack(tr
, data
, flags
, 6);
994 __raw_spin_unlock(&data
->lock
);
995 raw_local_irq_restore(irq_flags
);
1001 ftrace_special(unsigned long arg1
, unsigned long arg2
, unsigned long arg3
)
1003 struct trace_array
*tr
= &global_trace
;
1004 struct trace_array_cpu
*data
;
1005 unsigned long flags
;
1009 if (tracing_disabled
|| current_trace
== &no_tracer
|| !tr
->ctrl
)
1012 local_irq_save(flags
);
1013 cpu
= raw_smp_processor_id();
1014 data
= tr
->data
[cpu
];
1015 disabled
= atomic_inc_return(&data
->disabled
);
1017 if (likely(disabled
== 1))
1018 __trace_special(tr
, data
, arg1
, arg2
, arg3
);
1020 atomic_dec(&data
->disabled
);
1021 local_irq_restore(flags
);
1024 #ifdef CONFIG_FTRACE
1026 function_trace_call(unsigned long ip
, unsigned long parent_ip
)
1028 struct trace_array
*tr
= &global_trace
;
1029 struct trace_array_cpu
*data
;
1030 unsigned long flags
;
1034 if (unlikely(!ftrace_function_enabled
))
1040 local_irq_save(flags
);
1041 cpu
= raw_smp_processor_id();
1042 data
= tr
->data
[cpu
];
1043 disabled
= atomic_inc_return(&data
->disabled
);
1045 if (likely(disabled
== 1))
1046 trace_function(tr
, data
, ip
, parent_ip
, flags
);
1048 atomic_dec(&data
->disabled
);
1049 local_irq_restore(flags
);
1052 static struct ftrace_ops trace_ops __read_mostly
=
1054 .func
= function_trace_call
,
1057 void tracing_start_function_trace(void)
1059 ftrace_function_enabled
= 0;
1060 register_ftrace_function(&trace_ops
);
1062 ftrace_function_enabled
= 1;
1065 void tracing_stop_function_trace(void)
1067 ftrace_function_enabled
= 0;
1068 unregister_ftrace_function(&trace_ops
);
1072 enum trace_file_type
{
1073 TRACE_FILE_LAT_FMT
= 1,
1076 static struct trace_entry
*
1077 trace_entry_idx(struct trace_array
*tr
, struct trace_array_cpu
*data
,
1078 struct trace_iterator
*iter
, int cpu
)
1081 struct trace_entry
*array
;
1083 if (iter
->next_idx
[cpu
] >= tr
->entries
||
1084 iter
->next_idx
[cpu
] >= data
->trace_idx
||
1085 (data
->trace_head
== data
->trace_tail
&&
1086 data
->trace_head_idx
== data
->trace_tail_idx
))
1089 if (!iter
->next_page
[cpu
]) {
1090 /* Initialize the iterator for this cpu trace buffer */
1091 WARN_ON(!data
->trace_tail
);
1092 page
= virt_to_page(data
->trace_tail
);
1093 iter
->next_page
[cpu
] = &page
->lru
;
1094 iter
->next_page_idx
[cpu
] = data
->trace_tail_idx
;
1097 page
= list_entry(iter
->next_page
[cpu
], struct page
, lru
);
1098 BUG_ON(&data
->trace_pages
== &page
->lru
);
1100 array
= page_address(page
);
1102 WARN_ON(iter
->next_page_idx
[cpu
] >= ENTRIES_PER_PAGE
);
1103 return &array
[iter
->next_page_idx
[cpu
]];
1106 static struct trace_entry
*
1107 find_next_entry(struct trace_iterator
*iter
, int *ent_cpu
)
1109 struct trace_array
*tr
= iter
->tr
;
1110 struct trace_entry
*ent
, *next
= NULL
;
1114 for_each_tracing_cpu(cpu
) {
1115 if (!head_page(tr
->data
[cpu
]))
1117 ent
= trace_entry_idx(tr
, tr
->data
[cpu
], iter
, cpu
);
1119 * Pick the entry with the smallest timestamp:
1121 if (ent
&& (!next
|| ent
->t
< next
->t
)) {
1128 *ent_cpu
= next_cpu
;
1133 static void trace_iterator_increment(struct trace_iterator
*iter
)
1136 iter
->next_idx
[iter
->cpu
]++;
1137 iter
->next_page_idx
[iter
->cpu
]++;
1139 if (iter
->next_page_idx
[iter
->cpu
] >= ENTRIES_PER_PAGE
) {
1140 struct trace_array_cpu
*data
= iter
->tr
->data
[iter
->cpu
];
1142 iter
->next_page_idx
[iter
->cpu
] = 0;
1143 iter
->next_page
[iter
->cpu
] =
1144 trace_next_list(data
, iter
->next_page
[iter
->cpu
]);
1148 static void trace_consume(struct trace_iterator
*iter
)
1150 struct trace_array_cpu
*data
= iter
->tr
->data
[iter
->cpu
];
1152 data
->trace_tail_idx
++;
1153 if (data
->trace_tail_idx
>= ENTRIES_PER_PAGE
) {
1154 data
->trace_tail
= trace_next_page(data
, data
->trace_tail
);
1155 data
->trace_tail_idx
= 0;
1158 /* Check if we empty it, then reset the index */
1159 if (data
->trace_head
== data
->trace_tail
&&
1160 data
->trace_head_idx
== data
->trace_tail_idx
)
1161 data
->trace_idx
= 0;
1164 static void *find_next_entry_inc(struct trace_iterator
*iter
)
1166 struct trace_entry
*next
;
1169 next
= find_next_entry(iter
, &next_cpu
);
1171 iter
->prev_ent
= iter
->ent
;
1172 iter
->prev_cpu
= iter
->cpu
;
1175 iter
->cpu
= next_cpu
;
1178 trace_iterator_increment(iter
);
1180 return next
? iter
: NULL
;
1183 static void *s_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1185 struct trace_iterator
*iter
= m
->private;
1186 void *last_ent
= iter
->ent
;
1192 /* can't go backwards */
1197 ent
= find_next_entry_inc(iter
);
1201 while (ent
&& iter
->idx
< i
)
1202 ent
= find_next_entry_inc(iter
);
1206 if (last_ent
&& !ent
)
1207 seq_puts(m
, "\n\nvim:ft=help\n");
1212 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
1214 struct trace_iterator
*iter
= m
->private;
1219 mutex_lock(&trace_types_lock
);
1221 if (!current_trace
|| current_trace
!= iter
->trace
) {
1222 mutex_unlock(&trace_types_lock
);
1226 atomic_inc(&trace_record_cmdline_disabled
);
1228 /* let the tracer grab locks here if needed */
1229 if (current_trace
->start
)
1230 current_trace
->start(iter
);
1232 if (*pos
!= iter
->pos
) {
1236 iter
->prev_ent
= NULL
;
1237 iter
->prev_cpu
= -1;
1239 for_each_tracing_cpu(i
) {
1240 iter
->next_idx
[i
] = 0;
1241 iter
->next_page
[i
] = NULL
;
1244 for (p
= iter
; p
&& l
< *pos
; p
= s_next(m
, p
, &l
))
1249 p
= s_next(m
, p
, &l
);
1255 static void s_stop(struct seq_file
*m
, void *p
)
1257 struct trace_iterator
*iter
= m
->private;
1259 atomic_dec(&trace_record_cmdline_disabled
);
1261 /* let the tracer release locks here if needed */
1262 if (current_trace
&& current_trace
== iter
->trace
&& iter
->trace
->stop
)
1263 iter
->trace
->stop(iter
);
1265 mutex_unlock(&trace_types_lock
);
1268 #define KRETPROBE_MSG "[unknown/kretprobe'd]"
1270 #ifdef CONFIG_KRETPROBES
1271 static inline int kretprobed(unsigned long addr
)
1273 return addr
== (unsigned long)kretprobe_trampoline
;
1276 static inline int kretprobed(unsigned long addr
)
1280 #endif /* CONFIG_KRETPROBES */
1283 seq_print_sym_short(struct trace_seq
*s
, const char *fmt
, unsigned long address
)
1285 #ifdef CONFIG_KALLSYMS
1286 char str
[KSYM_SYMBOL_LEN
];
1288 kallsyms_lookup(address
, NULL
, NULL
, NULL
, str
);
1290 return trace_seq_printf(s
, fmt
, str
);
1296 seq_print_sym_offset(struct trace_seq
*s
, const char *fmt
,
1297 unsigned long address
)
1299 #ifdef CONFIG_KALLSYMS
1300 char str
[KSYM_SYMBOL_LEN
];
1302 sprint_symbol(str
, address
);
1303 return trace_seq_printf(s
, fmt
, str
);
1308 #ifndef CONFIG_64BIT
1309 # define IP_FMT "%08lx"
1311 # define IP_FMT "%016lx"
1315 seq_print_ip_sym(struct trace_seq
*s
, unsigned long ip
, unsigned long sym_flags
)
1320 return trace_seq_printf(s
, "0");
1322 if (sym_flags
& TRACE_ITER_SYM_OFFSET
)
1323 ret
= seq_print_sym_offset(s
, "%s", ip
);
1325 ret
= seq_print_sym_short(s
, "%s", ip
);
1330 if (sym_flags
& TRACE_ITER_SYM_ADDR
)
1331 ret
= trace_seq_printf(s
, " <" IP_FMT
">", ip
);
1335 static void print_lat_help_header(struct seq_file
*m
)
1337 seq_puts(m
, "# _------=> CPU# \n");
1338 seq_puts(m
, "# / _-----=> irqs-off \n");
1339 seq_puts(m
, "# | / _----=> need-resched \n");
1340 seq_puts(m
, "# || / _---=> hardirq/softirq \n");
1341 seq_puts(m
, "# ||| / _--=> preempt-depth \n");
1342 seq_puts(m
, "# |||| / \n");
1343 seq_puts(m
, "# ||||| delay \n");
1344 seq_puts(m
, "# cmd pid ||||| time | caller \n");
1345 seq_puts(m
, "# \\ / ||||| \\ | / \n");
1348 static void print_func_help_header(struct seq_file
*m
)
1350 seq_puts(m
, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1351 seq_puts(m
, "# | | | | |\n");
1356 print_trace_header(struct seq_file
*m
, struct trace_iterator
*iter
)
1358 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1359 struct trace_array
*tr
= iter
->tr
;
1360 struct trace_array_cpu
*data
= tr
->data
[tr
->cpu
];
1361 struct tracer
*type
= current_trace
;
1362 unsigned long total
= 0;
1363 unsigned long entries
= 0;
1365 const char *name
= "preemption";
1370 for_each_tracing_cpu(cpu
) {
1371 if (head_page(tr
->data
[cpu
])) {
1372 total
+= tr
->data
[cpu
]->trace_idx
;
1373 if (tr
->data
[cpu
]->trace_idx
> tr
->entries
)
1374 entries
+= tr
->entries
;
1376 entries
+= tr
->data
[cpu
]->trace_idx
;
1380 seq_printf(m
, "%s latency trace v1.1.5 on %s\n",
1382 seq_puts(m
, "-----------------------------------"
1383 "---------------------------------\n");
1384 seq_printf(m
, " latency: %lu us, #%lu/%lu, CPU#%d |"
1385 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1386 nsecs_to_usecs(data
->saved_latency
),
1390 #if defined(CONFIG_PREEMPT_NONE)
1392 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1394 #elif defined(CONFIG_PREEMPT)
1399 /* These are reserved for later use */
1402 seq_printf(m
, " #P:%d)\n", num_online_cpus());
1406 seq_puts(m
, " -----------------\n");
1407 seq_printf(m
, " | task: %.16s-%d "
1408 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1409 data
->comm
, data
->pid
, data
->uid
, data
->nice
,
1410 data
->policy
, data
->rt_priority
);
1411 seq_puts(m
, " -----------------\n");
1413 if (data
->critical_start
) {
1414 seq_puts(m
, " => started at: ");
1415 seq_print_ip_sym(&iter
->seq
, data
->critical_start
, sym_flags
);
1416 trace_print_seq(m
, &iter
->seq
);
1417 seq_puts(m
, "\n => ended at: ");
1418 seq_print_ip_sym(&iter
->seq
, data
->critical_end
, sym_flags
);
1419 trace_print_seq(m
, &iter
->seq
);
1427 lat_print_generic(struct trace_seq
*s
, struct trace_entry
*entry
, int cpu
)
1429 int hardirq
, softirq
;
1432 comm
= trace_find_cmdline(entry
->pid
);
1434 trace_seq_printf(s
, "%8.8s-%-5d ", comm
, entry
->pid
);
1435 trace_seq_printf(s
, "%d", cpu
);
1436 trace_seq_printf(s
, "%c%c",
1437 (entry
->flags
& TRACE_FLAG_IRQS_OFF
) ? 'd' : '.',
1438 ((entry
->flags
& TRACE_FLAG_NEED_RESCHED
) ? 'N' : '.'));
1440 hardirq
= entry
->flags
& TRACE_FLAG_HARDIRQ
;
1441 softirq
= entry
->flags
& TRACE_FLAG_SOFTIRQ
;
1442 if (hardirq
&& softirq
) {
1443 trace_seq_putc(s
, 'H');
1446 trace_seq_putc(s
, 'h');
1449 trace_seq_putc(s
, 's');
1451 trace_seq_putc(s
, '.');
1455 if (entry
->preempt_count
)
1456 trace_seq_printf(s
, "%x", entry
->preempt_count
);
1458 trace_seq_puts(s
, ".");
1461 unsigned long preempt_mark_thresh
= 100;
1464 lat_print_timestamp(struct trace_seq
*s
, unsigned long long abs_usecs
,
1465 unsigned long rel_usecs
)
1467 trace_seq_printf(s
, " %4lldus", abs_usecs
);
1468 if (rel_usecs
> preempt_mark_thresh
)
1469 trace_seq_puts(s
, "!: ");
1470 else if (rel_usecs
> 1)
1471 trace_seq_puts(s
, "+: ");
1473 trace_seq_puts(s
, " : ");
1476 static const char state_to_char
[] = TASK_STATE_TO_CHAR_STR
;
1479 print_lat_fmt(struct trace_iterator
*iter
, unsigned int trace_idx
, int cpu
)
1481 struct trace_seq
*s
= &iter
->seq
;
1482 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1483 struct trace_entry
*next_entry
= find_next_entry(iter
, NULL
);
1484 unsigned long verbose
= (trace_flags
& TRACE_ITER_VERBOSE
);
1485 struct trace_entry
*entry
= iter
->ent
;
1486 unsigned long abs_usecs
;
1487 unsigned long rel_usecs
;
1495 rel_usecs
= ns2usecs(next_entry
->t
- entry
->t
);
1496 abs_usecs
= ns2usecs(entry
->t
- iter
->tr
->time_start
);
1499 comm
= trace_find_cmdline(entry
->pid
);
1500 trace_seq_printf(s
, "%16s %5d %d %d %08x %08x [%08lx]"
1501 " %ld.%03ldms (+%ld.%03ldms): ",
1503 entry
->pid
, cpu
, entry
->flags
,
1504 entry
->preempt_count
, trace_idx
,
1507 abs_usecs
% 1000, rel_usecs
/1000,
1510 lat_print_generic(s
, entry
, cpu
);
1511 lat_print_timestamp(s
, abs_usecs
, rel_usecs
);
1513 switch (entry
->type
) {
1515 seq_print_ip_sym(s
, entry
->fn
.ip
, sym_flags
);
1516 trace_seq_puts(s
, " (");
1517 if (kretprobed(entry
->fn
.parent_ip
))
1518 trace_seq_puts(s
, KRETPROBE_MSG
);
1520 seq_print_ip_sym(s
, entry
->fn
.parent_ip
, sym_flags
);
1521 trace_seq_puts(s
, ")\n");
1525 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1526 state_to_char
[entry
->ctx
.next_state
] : 'X';
1528 state
= entry
->ctx
.prev_state
? __ffs(entry
->ctx
.prev_state
) + 1 : 0;
1529 S
= state
< sizeof(state_to_char
) - 1 ? state_to_char
[state
] : 'X';
1530 comm
= trace_find_cmdline(entry
->ctx
.next_pid
);
1531 trace_seq_printf(s
, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1532 entry
->ctx
.prev_pid
,
1533 entry
->ctx
.prev_prio
,
1534 S
, entry
->type
== TRACE_CTX
? "==>" : " +",
1535 entry
->ctx
.next_pid
,
1536 entry
->ctx
.next_prio
,
1540 trace_seq_printf(s
, "# %ld %ld %ld\n",
1541 entry
->special
.arg1
,
1542 entry
->special
.arg2
,
1543 entry
->special
.arg3
);
1546 for (i
= 0; i
< FTRACE_STACK_ENTRIES
; i
++) {
1548 trace_seq_puts(s
, " <= ");
1549 seq_print_ip_sym(s
, entry
->stack
.caller
[i
], sym_flags
);
1551 trace_seq_puts(s
, "\n");
1554 trace_seq_printf(s
, "Unknown type %d\n", entry
->type
);
1559 static int print_trace_fmt(struct trace_iterator
*iter
)
1561 struct trace_seq
*s
= &iter
->seq
;
1562 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1563 struct trace_entry
*entry
;
1564 unsigned long usec_rem
;
1565 unsigned long long t
;
1574 comm
= trace_find_cmdline(iter
->ent
->pid
);
1576 t
= ns2usecs(entry
->t
);
1577 usec_rem
= do_div(t
, 1000000ULL);
1578 secs
= (unsigned long)t
;
1580 ret
= trace_seq_printf(s
, "%16s-%-5d ", comm
, entry
->pid
);
1583 ret
= trace_seq_printf(s
, "[%02d] ", iter
->cpu
);
1586 ret
= trace_seq_printf(s
, "%5lu.%06lu: ", secs
, usec_rem
);
1590 switch (entry
->type
) {
1592 ret
= seq_print_ip_sym(s
, entry
->fn
.ip
, sym_flags
);
1595 if ((sym_flags
& TRACE_ITER_PRINT_PARENT
) &&
1596 entry
->fn
.parent_ip
) {
1597 ret
= trace_seq_printf(s
, " <-");
1600 if (kretprobed(entry
->fn
.parent_ip
))
1601 ret
= trace_seq_puts(s
, KRETPROBE_MSG
);
1603 ret
= seq_print_ip_sym(s
, entry
->fn
.parent_ip
,
1608 ret
= trace_seq_printf(s
, "\n");
1614 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1615 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1616 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1617 state_to_char
[entry
->ctx
.next_state
] : 'X';
1618 ret
= trace_seq_printf(s
, " %5d:%3d:%c %s %5d:%3d:%c\n",
1619 entry
->ctx
.prev_pid
,
1620 entry
->ctx
.prev_prio
,
1622 entry
->type
== TRACE_CTX
? "==>" : " +",
1623 entry
->ctx
.next_pid
,
1624 entry
->ctx
.next_prio
,
1630 ret
= trace_seq_printf(s
, "# %ld %ld %ld\n",
1631 entry
->special
.arg1
,
1632 entry
->special
.arg2
,
1633 entry
->special
.arg3
);
1638 for (i
= 0; i
< FTRACE_STACK_ENTRIES
; i
++) {
1640 ret
= trace_seq_puts(s
, " <= ");
1644 ret
= seq_print_ip_sym(s
, entry
->stack
.caller
[i
],
1649 ret
= trace_seq_puts(s
, "\n");
1657 static int print_raw_fmt(struct trace_iterator
*iter
)
1659 struct trace_seq
*s
= &iter
->seq
;
1660 struct trace_entry
*entry
;
1666 ret
= trace_seq_printf(s
, "%d %d %llu ",
1667 entry
->pid
, iter
->cpu
, entry
->t
);
1671 switch (entry
->type
) {
1673 ret
= trace_seq_printf(s
, "%x %x\n",
1674 entry
->fn
.ip
, entry
->fn
.parent_ip
);
1680 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1681 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1682 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1683 state_to_char
[entry
->ctx
.next_state
] : 'X';
1684 if (entry
->type
== TRACE_WAKE
)
1686 ret
= trace_seq_printf(s
, "%d %d %c %d %d %c\n",
1687 entry
->ctx
.prev_pid
,
1688 entry
->ctx
.prev_prio
,
1690 entry
->ctx
.next_pid
,
1691 entry
->ctx
.next_prio
,
1698 ret
= trace_seq_printf(s
, "# %ld %ld %ld\n",
1699 entry
->special
.arg1
,
1700 entry
->special
.arg2
,
1701 entry
->special
.arg3
);
1709 #define SEQ_PUT_FIELD_RET(s, x) \
1711 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1715 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1717 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1721 static int print_hex_fmt(struct trace_iterator
*iter
)
1723 struct trace_seq
*s
= &iter
->seq
;
1724 unsigned char newline
= '\n';
1725 struct trace_entry
*entry
;
1730 SEQ_PUT_HEX_FIELD_RET(s
, entry
->pid
);
1731 SEQ_PUT_HEX_FIELD_RET(s
, iter
->cpu
);
1732 SEQ_PUT_HEX_FIELD_RET(s
, entry
->t
);
1734 switch (entry
->type
) {
1736 SEQ_PUT_HEX_FIELD_RET(s
, entry
->fn
.ip
);
1737 SEQ_PUT_HEX_FIELD_RET(s
, entry
->fn
.parent_ip
);
1741 S
= entry
->ctx
.prev_state
< sizeof(state_to_char
) ?
1742 state_to_char
[entry
->ctx
.prev_state
] : 'X';
1743 T
= entry
->ctx
.next_state
< sizeof(state_to_char
) ?
1744 state_to_char
[entry
->ctx
.next_state
] : 'X';
1745 if (entry
->type
== TRACE_WAKE
)
1747 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.prev_pid
);
1748 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.prev_prio
);
1749 SEQ_PUT_HEX_FIELD_RET(s
, S
);
1750 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.next_pid
);
1751 SEQ_PUT_HEX_FIELD_RET(s
, entry
->ctx
.next_prio
);
1752 SEQ_PUT_HEX_FIELD_RET(s
, entry
->fn
.parent_ip
);
1753 SEQ_PUT_HEX_FIELD_RET(s
, T
);
1757 SEQ_PUT_HEX_FIELD_RET(s
, entry
->special
.arg1
);
1758 SEQ_PUT_HEX_FIELD_RET(s
, entry
->special
.arg2
);
1759 SEQ_PUT_HEX_FIELD_RET(s
, entry
->special
.arg3
);
1762 SEQ_PUT_FIELD_RET(s
, newline
);
1767 static int print_bin_fmt(struct trace_iterator
*iter
)
1769 struct trace_seq
*s
= &iter
->seq
;
1770 struct trace_entry
*entry
;
1774 SEQ_PUT_FIELD_RET(s
, entry
->pid
);
1775 SEQ_PUT_FIELD_RET(s
, entry
->cpu
);
1776 SEQ_PUT_FIELD_RET(s
, entry
->t
);
1778 switch (entry
->type
) {
1780 SEQ_PUT_FIELD_RET(s
, entry
->fn
.ip
);
1781 SEQ_PUT_FIELD_RET(s
, entry
->fn
.parent_ip
);
1784 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_pid
);
1785 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_prio
);
1786 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.prev_state
);
1787 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_pid
);
1788 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_prio
);
1789 SEQ_PUT_FIELD_RET(s
, entry
->ctx
.next_state
);
1793 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg1
);
1794 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg2
);
1795 SEQ_PUT_FIELD_RET(s
, entry
->special
.arg3
);
1801 static int trace_empty(struct trace_iterator
*iter
)
1803 struct trace_array_cpu
*data
;
1806 for_each_tracing_cpu(cpu
) {
1807 data
= iter
->tr
->data
[cpu
];
1809 if (head_page(data
) && data
->trace_idx
&&
1810 (data
->trace_tail
!= data
->trace_head
||
1811 data
->trace_tail_idx
!= data
->trace_head_idx
))
1817 static int print_trace_line(struct trace_iterator
*iter
)
1819 if (iter
->trace
&& iter
->trace
->print_line
)
1820 return iter
->trace
->print_line(iter
);
1822 if (trace_flags
& TRACE_ITER_BIN
)
1823 return print_bin_fmt(iter
);
1825 if (trace_flags
& TRACE_ITER_HEX
)
1826 return print_hex_fmt(iter
);
1828 if (trace_flags
& TRACE_ITER_RAW
)
1829 return print_raw_fmt(iter
);
1831 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
)
1832 return print_lat_fmt(iter
, iter
->idx
, iter
->cpu
);
1834 return print_trace_fmt(iter
);
1837 static int s_show(struct seq_file
*m
, void *v
)
1839 struct trace_iterator
*iter
= v
;
1841 if (iter
->ent
== NULL
) {
1843 seq_printf(m
, "# tracer: %s\n", iter
->trace
->name
);
1846 if (iter
->iter_flags
& TRACE_FILE_LAT_FMT
) {
1847 /* print nothing if the buffers are empty */
1848 if (trace_empty(iter
))
1850 print_trace_header(m
, iter
);
1851 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1852 print_lat_help_header(m
);
1854 if (!(trace_flags
& TRACE_ITER_VERBOSE
))
1855 print_func_help_header(m
);
1858 print_trace_line(iter
);
1859 trace_print_seq(m
, &iter
->seq
);
1865 static struct seq_operations tracer_seq_ops
= {
1872 static struct trace_iterator
*
1873 __tracing_open(struct inode
*inode
, struct file
*file
, int *ret
)
1875 struct trace_iterator
*iter
;
1877 if (tracing_disabled
) {
1882 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
1888 mutex_lock(&trace_types_lock
);
1889 if (current_trace
&& current_trace
->print_max
)
1892 iter
->tr
= inode
->i_private
;
1893 iter
->trace
= current_trace
;
1896 /* TODO stop tracer */
1897 *ret
= seq_open(file
, &tracer_seq_ops
);
1899 struct seq_file
*m
= file
->private_data
;
1902 /* stop the trace while dumping */
1903 if (iter
->tr
->ctrl
) {
1905 ftrace_function_enabled
= 0;
1908 if (iter
->trace
&& iter
->trace
->open
)
1909 iter
->trace
->open(iter
);
1914 mutex_unlock(&trace_types_lock
);
1920 int tracing_open_generic(struct inode
*inode
, struct file
*filp
)
1922 if (tracing_disabled
)
1925 filp
->private_data
= inode
->i_private
;
1929 int tracing_release(struct inode
*inode
, struct file
*file
)
1931 struct seq_file
*m
= (struct seq_file
*)file
->private_data
;
1932 struct trace_iterator
*iter
= m
->private;
1934 mutex_lock(&trace_types_lock
);
1935 if (iter
->trace
&& iter
->trace
->close
)
1936 iter
->trace
->close(iter
);
1938 /* reenable tracing if it was previously enabled */
1939 if (iter
->tr
->ctrl
) {
1942 * It is safe to enable function tracing even if it
1945 ftrace_function_enabled
= 1;
1947 mutex_unlock(&trace_types_lock
);
1949 seq_release(inode
, file
);
1954 static int tracing_open(struct inode
*inode
, struct file
*file
)
1958 __tracing_open(inode
, file
, &ret
);
1963 static int tracing_lt_open(struct inode
*inode
, struct file
*file
)
1965 struct trace_iterator
*iter
;
1968 iter
= __tracing_open(inode
, file
, &ret
);
1971 iter
->iter_flags
|= TRACE_FILE_LAT_FMT
;
1978 t_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1980 struct tracer
*t
= m
->private;
1992 static void *t_start(struct seq_file
*m
, loff_t
*pos
)
1994 struct tracer
*t
= m
->private;
1997 mutex_lock(&trace_types_lock
);
1998 for (; t
&& l
< *pos
; t
= t_next(m
, t
, &l
))
2004 static void t_stop(struct seq_file
*m
, void *p
)
2006 mutex_unlock(&trace_types_lock
);
2009 static int t_show(struct seq_file
*m
, void *v
)
2011 struct tracer
*t
= v
;
2016 seq_printf(m
, "%s", t
->name
);
2025 static struct seq_operations show_traces_seq_ops
= {
2032 static int show_traces_open(struct inode
*inode
, struct file
*file
)
2036 if (tracing_disabled
)
2039 ret
= seq_open(file
, &show_traces_seq_ops
);
2041 struct seq_file
*m
= file
->private_data
;
2042 m
->private = trace_types
;
2048 static struct file_operations tracing_fops
= {
2049 .open
= tracing_open
,
2051 .llseek
= seq_lseek
,
2052 .release
= tracing_release
,
2055 static struct file_operations tracing_lt_fops
= {
2056 .open
= tracing_lt_open
,
2058 .llseek
= seq_lseek
,
2059 .release
= tracing_release
,
2062 static struct file_operations show_traces_fops
= {
2063 .open
= show_traces_open
,
2065 .release
= seq_release
,
2069 * Only trace on a CPU if the bitmask is set:
2071 static cpumask_t tracing_cpumask
= CPU_MASK_ALL
;
2074 * When tracing/tracing_cpu_mask is modified then this holds
2075 * the new bitmask we are about to install:
2077 static cpumask_t tracing_cpumask_new
;
2080 * The tracer itself will not take this lock, but still we want
2081 * to provide a consistent cpumask to user-space:
2083 static DEFINE_MUTEX(tracing_cpumask_update_lock
);
2086 * Temporary storage for the character representation of the
2087 * CPU bitmask (and one more byte for the newline):
2089 static char mask_str
[NR_CPUS
+ 1];
2092 tracing_cpumask_read(struct file
*filp
, char __user
*ubuf
,
2093 size_t count
, loff_t
*ppos
)
2097 mutex_lock(&tracing_cpumask_update_lock
);
2099 len
= cpumask_scnprintf(mask_str
, count
, tracing_cpumask
);
2100 if (count
- len
< 2) {
2104 len
+= sprintf(mask_str
+ len
, "\n");
2105 count
= simple_read_from_buffer(ubuf
, count
, ppos
, mask_str
, NR_CPUS
+1);
2108 mutex_unlock(&tracing_cpumask_update_lock
);
2114 tracing_cpumask_write(struct file
*filp
, const char __user
*ubuf
,
2115 size_t count
, loff_t
*ppos
)
2119 mutex_lock(&tracing_cpumask_update_lock
);
2120 err
= cpumask_parse_user(ubuf
, count
, tracing_cpumask_new
);
2124 raw_local_irq_disable();
2125 __raw_spin_lock(&ftrace_max_lock
);
2126 for_each_tracing_cpu(cpu
) {
2128 * Increase/decrease the disabled counter if we are
2129 * about to flip a bit in the cpumask:
2131 if (cpu_isset(cpu
, tracing_cpumask
) &&
2132 !cpu_isset(cpu
, tracing_cpumask_new
)) {
2133 atomic_inc(&global_trace
.data
[cpu
]->disabled
);
2135 if (!cpu_isset(cpu
, tracing_cpumask
) &&
2136 cpu_isset(cpu
, tracing_cpumask_new
)) {
2137 atomic_dec(&global_trace
.data
[cpu
]->disabled
);
2140 __raw_spin_unlock(&ftrace_max_lock
);
2141 raw_local_irq_enable();
2143 tracing_cpumask
= tracing_cpumask_new
;
2145 mutex_unlock(&tracing_cpumask_update_lock
);
2150 mutex_unlock(&tracing_cpumask_update_lock
);
2155 static struct file_operations tracing_cpumask_fops
= {
2156 .open
= tracing_open_generic
,
2157 .read
= tracing_cpumask_read
,
2158 .write
= tracing_cpumask_write
,
2162 tracing_iter_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2163 size_t cnt
, loff_t
*ppos
)
2170 /* calulate max size */
2171 for (i
= 0; trace_options
[i
]; i
++) {
2172 len
+= strlen(trace_options
[i
]);
2173 len
+= 3; /* "no" and space */
2176 /* +2 for \n and \0 */
2177 buf
= kmalloc(len
+ 2, GFP_KERNEL
);
2181 for (i
= 0; trace_options
[i
]; i
++) {
2182 if (trace_flags
& (1 << i
))
2183 r
+= sprintf(buf
+ r
, "%s ", trace_options
[i
]);
2185 r
+= sprintf(buf
+ r
, "no%s ", trace_options
[i
]);
2188 r
+= sprintf(buf
+ r
, "\n");
2189 WARN_ON(r
>= len
+ 2);
2191 r
= simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2199 tracing_iter_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2200 size_t cnt
, loff_t
*ppos
)
2207 if (cnt
>= sizeof(buf
))
2210 if (copy_from_user(&buf
, ubuf
, cnt
))
2215 if (strncmp(buf
, "no", 2) == 0) {
2220 for (i
= 0; trace_options
[i
]; i
++) {
2221 int len
= strlen(trace_options
[i
]);
2223 if (strncmp(cmp
, trace_options
[i
], len
) == 0) {
2225 trace_flags
&= ~(1 << i
);
2227 trace_flags
|= (1 << i
);
2232 * If no option could be set, return an error:
2234 if (!trace_options
[i
])
2242 static struct file_operations tracing_iter_fops
= {
2243 .open
= tracing_open_generic
,
2244 .read
= tracing_iter_ctrl_read
,
2245 .write
= tracing_iter_ctrl_write
,
2248 static const char readme_msg
[] =
2249 "tracing mini-HOWTO:\n\n"
2251 "# mount -t debugfs nodev /debug\n\n"
2252 "# cat /debug/tracing/available_tracers\n"
2253 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2254 "# cat /debug/tracing/current_tracer\n"
2256 "# echo sched_switch > /debug/tracing/current_tracer\n"
2257 "# cat /debug/tracing/current_tracer\n"
2259 "# cat /debug/tracing/iter_ctrl\n"
2260 "noprint-parent nosym-offset nosym-addr noverbose\n"
2261 "# echo print-parent > /debug/tracing/iter_ctrl\n"
2262 "# echo 1 > /debug/tracing/tracing_enabled\n"
2263 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2264 "echo 0 > /debug/tracing/tracing_enabled\n"
2268 tracing_readme_read(struct file
*filp
, char __user
*ubuf
,
2269 size_t cnt
, loff_t
*ppos
)
2271 return simple_read_from_buffer(ubuf
, cnt
, ppos
,
2272 readme_msg
, strlen(readme_msg
));
2275 static struct file_operations tracing_readme_fops
= {
2276 .open
= tracing_open_generic
,
2277 .read
= tracing_readme_read
,
2281 tracing_ctrl_read(struct file
*filp
, char __user
*ubuf
,
2282 size_t cnt
, loff_t
*ppos
)
2284 struct trace_array
*tr
= filp
->private_data
;
2288 r
= sprintf(buf
, "%ld\n", tr
->ctrl
);
2289 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2293 tracing_ctrl_write(struct file
*filp
, const char __user
*ubuf
,
2294 size_t cnt
, loff_t
*ppos
)
2296 struct trace_array
*tr
= filp
->private_data
;
2301 if (cnt
>= sizeof(buf
))
2304 if (copy_from_user(&buf
, ubuf
, cnt
))
2309 ret
= strict_strtoul(buf
, 10, &val
);
2315 mutex_lock(&trace_types_lock
);
2316 if (tr
->ctrl
^ val
) {
2324 if (current_trace
&& current_trace
->ctrl_update
)
2325 current_trace
->ctrl_update(tr
);
2327 mutex_unlock(&trace_types_lock
);
2335 tracing_set_trace_read(struct file
*filp
, char __user
*ubuf
,
2336 size_t cnt
, loff_t
*ppos
)
2338 char buf
[max_tracer_type_len
+2];
2341 mutex_lock(&trace_types_lock
);
2343 r
= sprintf(buf
, "%s\n", current_trace
->name
);
2345 r
= sprintf(buf
, "\n");
2346 mutex_unlock(&trace_types_lock
);
2348 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2352 tracing_set_trace_write(struct file
*filp
, const char __user
*ubuf
,
2353 size_t cnt
, loff_t
*ppos
)
2355 struct trace_array
*tr
= &global_trace
;
2357 char buf
[max_tracer_type_len
+1];
2360 if (cnt
> max_tracer_type_len
)
2361 cnt
= max_tracer_type_len
;
2363 if (copy_from_user(&buf
, ubuf
, cnt
))
2368 /* strip ending whitespace. */
2369 for (i
= cnt
- 1; i
> 0 && isspace(buf
[i
]); i
--)
2372 mutex_lock(&trace_types_lock
);
2373 for (t
= trace_types
; t
; t
= t
->next
) {
2374 if (strcmp(t
->name
, buf
) == 0)
2377 if (!t
|| t
== current_trace
)
2380 if (current_trace
&& current_trace
->reset
)
2381 current_trace
->reset(tr
);
2388 mutex_unlock(&trace_types_lock
);
2396 tracing_max_lat_read(struct file
*filp
, char __user
*ubuf
,
2397 size_t cnt
, loff_t
*ppos
)
2399 unsigned long *ptr
= filp
->private_data
;
2403 r
= snprintf(buf
, sizeof(buf
), "%ld\n",
2404 *ptr
== (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr
));
2405 if (r
> sizeof(buf
))
2407 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2411 tracing_max_lat_write(struct file
*filp
, const char __user
*ubuf
,
2412 size_t cnt
, loff_t
*ppos
)
2414 long *ptr
= filp
->private_data
;
2419 if (cnt
>= sizeof(buf
))
2422 if (copy_from_user(&buf
, ubuf
, cnt
))
2427 ret
= strict_strtoul(buf
, 10, &val
);
2436 static atomic_t tracing_reader
;
2438 static int tracing_open_pipe(struct inode
*inode
, struct file
*filp
)
2440 struct trace_iterator
*iter
;
2442 if (tracing_disabled
)
2445 /* We only allow for reader of the pipe */
2446 if (atomic_inc_return(&tracing_reader
) != 1) {
2447 atomic_dec(&tracing_reader
);
2451 /* create a buffer to store the information to pass to userspace */
2452 iter
= kzalloc(sizeof(*iter
), GFP_KERNEL
);
2456 mutex_lock(&trace_types_lock
);
2457 iter
->tr
= &global_trace
;
2458 iter
->trace
= current_trace
;
2459 filp
->private_data
= iter
;
2461 if (iter
->trace
->pipe_open
)
2462 iter
->trace
->pipe_open(iter
);
2463 mutex_unlock(&trace_types_lock
);
2468 static int tracing_release_pipe(struct inode
*inode
, struct file
*file
)
2470 struct trace_iterator
*iter
= file
->private_data
;
2473 atomic_dec(&tracing_reader
);
2479 tracing_poll_pipe(struct file
*filp
, poll_table
*poll_table
)
2481 struct trace_iterator
*iter
= filp
->private_data
;
2483 if (trace_flags
& TRACE_ITER_BLOCK
) {
2485 * Always select as readable when in blocking mode
2487 return POLLIN
| POLLRDNORM
;
2489 if (!trace_empty(iter
))
2490 return POLLIN
| POLLRDNORM
;
2491 poll_wait(filp
, &trace_wait
, poll_table
);
2492 if (!trace_empty(iter
))
2493 return POLLIN
| POLLRDNORM
;
2503 tracing_read_pipe(struct file
*filp
, char __user
*ubuf
,
2504 size_t cnt
, loff_t
*ppos
)
2506 struct trace_iterator
*iter
= filp
->private_data
;
2507 struct trace_array_cpu
*data
;
2508 static cpumask_t mask
;
2509 unsigned long flags
;
2510 #ifdef CONFIG_FTRACE
2516 /* return any leftover data */
2517 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
2522 trace_seq_reset(&iter
->seq
);
2524 mutex_lock(&trace_types_lock
);
2525 if (iter
->trace
->read
) {
2526 sret
= iter
->trace
->read(iter
, filp
, ubuf
, cnt
, ppos
);
2531 while (trace_empty(iter
)) {
2533 if ((filp
->f_flags
& O_NONBLOCK
)) {
2539 * This is a make-shift waitqueue. The reason we don't use
2540 * an actual wait queue is because:
2541 * 1) we only ever have one waiter
2542 * 2) the tracing, traces all functions, we don't want
2543 * the overhead of calling wake_up and friends
2544 * (and tracing them too)
2545 * Anyway, this is really very primitive wakeup.
2547 set_current_state(TASK_INTERRUPTIBLE
);
2548 iter
->tr
->waiter
= current
;
2550 mutex_unlock(&trace_types_lock
);
2552 /* sleep for 100 msecs, and try again. */
2553 schedule_timeout(HZ
/10);
2555 mutex_lock(&trace_types_lock
);
2557 iter
->tr
->waiter
= NULL
;
2559 if (signal_pending(current
)) {
2564 if (iter
->trace
!= current_trace
)
2568 * We block until we read something and tracing is disabled.
2569 * We still block if tracing is disabled, but we have never
2570 * read anything. This allows a user to cat this file, and
2571 * then enable tracing. But after we have read something,
2572 * we give an EOF when tracing is again disabled.
2574 * iter->pos will be 0 if we haven't read anything.
2576 if (!tracer_enabled
&& iter
->pos
)
2582 /* stop when tracing is finished */
2583 if (trace_empty(iter
))
2586 if (cnt
>= PAGE_SIZE
)
2587 cnt
= PAGE_SIZE
- 1;
2589 /* reset all but tr, trace, and overruns */
2590 memset(&iter
->seq
, 0,
2591 sizeof(struct trace_iterator
) -
2592 offsetof(struct trace_iterator
, seq
));
2596 * We need to stop all tracing on all CPUS to read the
2597 * the next buffer. This is a bit expensive, but is
2598 * not done often. We fill all what we can read,
2599 * and then release the locks again.
2603 local_irq_save(flags
);
2604 #ifdef CONFIG_FTRACE
2605 ftrace_save
= ftrace_enabled
;
2609 for_each_tracing_cpu(cpu
) {
2610 data
= iter
->tr
->data
[cpu
];
2612 if (!head_page(data
) || !data
->trace_idx
)
2615 atomic_inc(&data
->disabled
);
2619 for_each_cpu_mask(cpu
, mask
) {
2620 data
= iter
->tr
->data
[cpu
];
2621 __raw_spin_lock(&data
->lock
);
2623 if (data
->overrun
> iter
->last_overrun
[cpu
])
2624 iter
->overrun
[cpu
] +=
2625 data
->overrun
- iter
->last_overrun
[cpu
];
2626 iter
->last_overrun
[cpu
] = data
->overrun
;
2629 while (find_next_entry_inc(iter
) != NULL
) {
2631 int len
= iter
->seq
.len
;
2633 ret
= print_trace_line(iter
);
2635 /* don't print partial lines */
2636 iter
->seq
.len
= len
;
2640 trace_consume(iter
);
2642 if (iter
->seq
.len
>= cnt
)
2646 for_each_cpu_mask(cpu
, mask
) {
2647 data
= iter
->tr
->data
[cpu
];
2648 __raw_spin_unlock(&data
->lock
);
2651 for_each_cpu_mask(cpu
, mask
) {
2652 data
= iter
->tr
->data
[cpu
];
2653 atomic_dec(&data
->disabled
);
2655 #ifdef CONFIG_FTRACE
2656 ftrace_enabled
= ftrace_save
;
2658 local_irq_restore(flags
);
2660 /* Now copy what we have to the user */
2661 sret
= trace_seq_to_user(&iter
->seq
, ubuf
, cnt
);
2662 if (iter
->seq
.readpos
>= iter
->seq
.len
)
2663 trace_seq_reset(&iter
->seq
);
2668 mutex_unlock(&trace_types_lock
);
2674 tracing_entries_read(struct file
*filp
, char __user
*ubuf
,
2675 size_t cnt
, loff_t
*ppos
)
2677 struct trace_array
*tr
= filp
->private_data
;
2681 r
= sprintf(buf
, "%lu\n", tr
->entries
);
2682 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2686 tracing_entries_write(struct file
*filp
, const char __user
*ubuf
,
2687 size_t cnt
, loff_t
*ppos
)
2693 if (cnt
>= sizeof(buf
))
2696 if (copy_from_user(&buf
, ubuf
, cnt
))
2701 ret
= strict_strtoul(buf
, 10, &val
);
2705 /* must have at least 1 entry */
2709 mutex_lock(&trace_types_lock
);
2711 if (current_trace
!= &no_tracer
) {
2713 pr_info("ftrace: set current_tracer to none"
2714 " before modifying buffer size\n");
2718 if (val
> global_trace
.entries
) {
2719 long pages_requested
;
2720 unsigned long freeable_pages
;
2722 /* make sure we have enough memory before mapping */
2724 (val
+ (ENTRIES_PER_PAGE
-1)) / ENTRIES_PER_PAGE
;
2726 /* account for each buffer (and max_tr) */
2727 pages_requested
*= tracing_nr_buffers
* 2;
2729 /* Check for overflow */
2730 if (pages_requested
< 0) {
2735 freeable_pages
= determine_dirtyable_memory();
2737 /* we only allow to request 1/4 of useable memory */
2738 if (pages_requested
>
2739 ((freeable_pages
+ tracing_pages_allocated
) / 4)) {
2744 while (global_trace
.entries
< val
) {
2745 if (trace_alloc_page()) {
2749 /* double check that we don't go over the known pages */
2750 if (tracing_pages_allocated
> pages_requested
)
2755 /* include the number of entries in val (inc of page entries) */
2756 while (global_trace
.entries
> val
+ (ENTRIES_PER_PAGE
- 1))
2760 /* check integrity */
2761 for_each_tracing_cpu(i
)
2762 check_pages(global_trace
.data
[i
]);
2766 /* If check pages failed, return ENOMEM */
2767 if (tracing_disabled
)
2770 max_tr
.entries
= global_trace
.entries
;
2771 mutex_unlock(&trace_types_lock
);
2776 static struct file_operations tracing_max_lat_fops
= {
2777 .open
= tracing_open_generic
,
2778 .read
= tracing_max_lat_read
,
2779 .write
= tracing_max_lat_write
,
2782 static struct file_operations tracing_ctrl_fops
= {
2783 .open
= tracing_open_generic
,
2784 .read
= tracing_ctrl_read
,
2785 .write
= tracing_ctrl_write
,
2788 static struct file_operations set_tracer_fops
= {
2789 .open
= tracing_open_generic
,
2790 .read
= tracing_set_trace_read
,
2791 .write
= tracing_set_trace_write
,
2794 static struct file_operations tracing_pipe_fops
= {
2795 .open
= tracing_open_pipe
,
2796 .poll
= tracing_poll_pipe
,
2797 .read
= tracing_read_pipe
,
2798 .release
= tracing_release_pipe
,
2801 static struct file_operations tracing_entries_fops
= {
2802 .open
= tracing_open_generic
,
2803 .read
= tracing_entries_read
,
2804 .write
= tracing_entries_write
,
2807 #ifdef CONFIG_DYNAMIC_FTRACE
2810 tracing_read_long(struct file
*filp
, char __user
*ubuf
,
2811 size_t cnt
, loff_t
*ppos
)
2813 unsigned long *p
= filp
->private_data
;
2817 r
= sprintf(buf
, "%ld\n", *p
);
2819 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2822 static struct file_operations tracing_read_long_fops
= {
2823 .open
= tracing_open_generic
,
2824 .read
= tracing_read_long
,
2828 static struct dentry
*d_tracer
;
2830 struct dentry
*tracing_init_dentry(void)
2837 d_tracer
= debugfs_create_dir("tracing", NULL
);
2839 if (!d_tracer
&& !once
) {
2841 pr_warning("Could not create debugfs directory 'tracing'\n");
2848 #ifdef CONFIG_FTRACE_SELFTEST
2849 /* Let selftest have access to static functions in this file */
2850 #include "trace_selftest.c"
2853 static __init
void tracer_init_debugfs(void)
2855 struct dentry
*d_tracer
;
2856 struct dentry
*entry
;
2858 d_tracer
= tracing_init_dentry();
2860 entry
= debugfs_create_file("tracing_enabled", 0644, d_tracer
,
2861 &global_trace
, &tracing_ctrl_fops
);
2863 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2865 entry
= debugfs_create_file("iter_ctrl", 0644, d_tracer
,
2866 NULL
, &tracing_iter_fops
);
2868 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2870 entry
= debugfs_create_file("tracing_cpumask", 0644, d_tracer
,
2871 NULL
, &tracing_cpumask_fops
);
2873 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2875 entry
= debugfs_create_file("latency_trace", 0444, d_tracer
,
2876 &global_trace
, &tracing_lt_fops
);
2878 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2880 entry
= debugfs_create_file("trace", 0444, d_tracer
,
2881 &global_trace
, &tracing_fops
);
2883 pr_warning("Could not create debugfs 'trace' entry\n");
2885 entry
= debugfs_create_file("available_tracers", 0444, d_tracer
,
2886 &global_trace
, &show_traces_fops
);
2888 pr_warning("Could not create debugfs 'trace' entry\n");
2890 entry
= debugfs_create_file("current_tracer", 0444, d_tracer
,
2891 &global_trace
, &set_tracer_fops
);
2893 pr_warning("Could not create debugfs 'trace' entry\n");
2895 entry
= debugfs_create_file("tracing_max_latency", 0644, d_tracer
,
2896 &tracing_max_latency
,
2897 &tracing_max_lat_fops
);
2899 pr_warning("Could not create debugfs "
2900 "'tracing_max_latency' entry\n");
2902 entry
= debugfs_create_file("tracing_thresh", 0644, d_tracer
,
2903 &tracing_thresh
, &tracing_max_lat_fops
);
2905 pr_warning("Could not create debugfs "
2906 "'tracing_threash' entry\n");
2907 entry
= debugfs_create_file("README", 0644, d_tracer
,
2908 NULL
, &tracing_readme_fops
);
2910 pr_warning("Could not create debugfs 'README' entry\n");
2912 entry
= debugfs_create_file("trace_pipe", 0644, d_tracer
,
2913 NULL
, &tracing_pipe_fops
);
2915 pr_warning("Could not create debugfs "
2916 "'tracing_threash' entry\n");
2918 entry
= debugfs_create_file("trace_entries", 0644, d_tracer
,
2919 &global_trace
, &tracing_entries_fops
);
2921 pr_warning("Could not create debugfs "
2922 "'tracing_threash' entry\n");
2924 #ifdef CONFIG_DYNAMIC_FTRACE
2925 entry
= debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer
,
2926 &ftrace_update_tot_cnt
,
2927 &tracing_read_long_fops
);
2929 pr_warning("Could not create debugfs "
2930 "'dyn_ftrace_total_info' entry\n");
2932 #ifdef CONFIG_SYSPROF_TRACER
2933 init_tracer_sysprof_debugfs(d_tracer
);
2937 static int trace_alloc_page(void)
2939 struct trace_array_cpu
*data
;
2940 struct page
*page
, *tmp
;
2943 unsigned pages_allocated
= 0;
2946 /* first allocate a page for each CPU */
2947 for_each_tracing_cpu(i
) {
2948 array
= (void *)__get_free_page(GFP_KERNEL
);
2949 if (array
== NULL
) {
2950 printk(KERN_ERR
"tracer: failed to allocate page"
2951 "for trace buffer!\n");
2956 page
= virt_to_page(array
);
2957 list_add(&page
->lru
, &pages
);
2959 /* Only allocate if we are actually using the max trace */
2960 #ifdef CONFIG_TRACER_MAX_TRACE
2961 array
= (void *)__get_free_page(GFP_KERNEL
);
2962 if (array
== NULL
) {
2963 printk(KERN_ERR
"tracer: failed to allocate page"
2964 "for trace buffer!\n");
2968 page
= virt_to_page(array
);
2969 list_add(&page
->lru
, &pages
);
2973 /* Now that we successfully allocate a page per CPU, add them */
2974 for_each_tracing_cpu(i
) {
2975 data
= global_trace
.data
[i
];
2976 page
= list_entry(pages
.next
, struct page
, lru
);
2977 list_del_init(&page
->lru
);
2978 list_add_tail(&page
->lru
, &data
->trace_pages
);
2981 #ifdef CONFIG_TRACER_MAX_TRACE
2982 data
= max_tr
.data
[i
];
2983 page
= list_entry(pages
.next
, struct page
, lru
);
2984 list_del_init(&page
->lru
);
2985 list_add_tail(&page
->lru
, &data
->trace_pages
);
2989 tracing_pages_allocated
+= pages_allocated
;
2990 global_trace
.entries
+= ENTRIES_PER_PAGE
;
2995 list_for_each_entry_safe(page
, tmp
, &pages
, lru
) {
2996 list_del_init(&page
->lru
);
3002 static int trace_free_page(void)
3004 struct trace_array_cpu
*data
;
3006 struct list_head
*p
;
3010 /* free one page from each buffer */
3011 for_each_tracing_cpu(i
) {
3012 data
= global_trace
.data
[i
];
3013 p
= data
->trace_pages
.next
;
3014 if (p
== &data
->trace_pages
) {
3015 /* should never happen */
3017 tracing_disabled
= 1;
3021 page
= list_entry(p
, struct page
, lru
);
3023 list_del(&page
->lru
);
3024 tracing_pages_allocated
--;
3025 tracing_pages_allocated
--;
3028 tracing_reset(data
);
3030 #ifdef CONFIG_TRACER_MAX_TRACE
3031 data
= max_tr
.data
[i
];
3032 p
= data
->trace_pages
.next
;
3033 if (p
== &data
->trace_pages
) {
3034 /* should never happen */
3036 tracing_disabled
= 1;
3040 page
= list_entry(p
, struct page
, lru
);
3042 list_del(&page
->lru
);
3045 tracing_reset(data
);
3048 global_trace
.entries
-= ENTRIES_PER_PAGE
;
3053 __init
static int tracer_alloc_buffers(void)
3055 struct trace_array_cpu
*data
;
3062 /* TODO: make the number of buffers hot pluggable with CPUS */
3063 tracing_nr_buffers
= num_possible_cpus();
3064 tracing_buffer_mask
= cpu_possible_map
;
3066 /* Allocate the first page for all buffers */
3067 for_each_tracing_cpu(i
) {
3068 data
= global_trace
.data
[i
] = &per_cpu(global_trace_cpu
, i
);
3069 max_tr
.data
[i
] = &per_cpu(max_data
, i
);
3071 array
= (void *)__get_free_page(GFP_KERNEL
);
3072 if (array
== NULL
) {
3073 printk(KERN_ERR
"tracer: failed to allocate page"
3074 "for trace buffer!\n");
3078 /* set the array to the list */
3079 INIT_LIST_HEAD(&data
->trace_pages
);
3080 page
= virt_to_page(array
);
3081 list_add(&page
->lru
, &data
->trace_pages
);
3082 /* use the LRU flag to differentiate the two buffers */
3085 data
->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
3086 max_tr
.data
[i
]->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
3088 /* Only allocate if we are actually using the max trace */
3089 #ifdef CONFIG_TRACER_MAX_TRACE
3090 array
= (void *)__get_free_page(GFP_KERNEL
);
3091 if (array
== NULL
) {
3092 printk(KERN_ERR
"tracer: failed to allocate page"
3093 "for trace buffer!\n");
3097 INIT_LIST_HEAD(&max_tr
.data
[i
]->trace_pages
);
3098 page
= virt_to_page(array
);
3099 list_add(&page
->lru
, &max_tr
.data
[i
]->trace_pages
);
3105 * Since we allocate by orders of pages, we may be able to
3108 global_trace
.entries
= ENTRIES_PER_PAGE
;
3111 while (global_trace
.entries
< trace_nr_entries
) {
3112 if (trace_alloc_page())
3116 max_tr
.entries
= global_trace
.entries
;
3118 pr_info("tracer: %d pages allocated for %ld entries of %ld bytes\n",
3119 pages
, trace_nr_entries
, (long)TRACE_ENTRY_SIZE
);
3120 pr_info(" actual entries %ld\n", global_trace
.entries
);
3122 tracer_init_debugfs();
3124 trace_init_cmdlines();
3126 register_tracer(&no_tracer
);
3127 current_trace
= &no_tracer
;
3129 /* All seems OK, enable tracing */
3130 global_trace
.ctrl
= tracer_enabled
;
3131 tracing_disabled
= 0;
3136 for (i
-- ; i
>= 0; i
--) {
3137 struct page
*page
, *tmp
;
3138 struct trace_array_cpu
*data
= global_trace
.data
[i
];
3141 list_for_each_entry_safe(page
, tmp
,
3142 &data
->trace_pages
, lru
) {
3143 list_del_init(&page
->lru
);
3148 #ifdef CONFIG_TRACER_MAX_TRACE
3149 data
= max_tr
.data
[i
];
3151 list_for_each_entry_safe(page
, tmp
,
3152 &data
->trace_pages
, lru
) {
3153 list_del_init(&page
->lru
);
3161 fs_initcall(tracer_alloc_buffers
);