3 * Function graph tracer.
4 * Copyright (c) 2008-2009 Frederic Weisbecker <fweisbec@gmail.com>
5 * Mostly borrowed from function tracer which
6 * is Copyright (c) Steven Rostedt <srostedt@redhat.com>
9 #include <linux/debugfs.h>
10 #include <linux/uaccess.h>
11 #include <linux/ftrace.h>
12 #include <linux/slab.h>
16 #include "trace_output.h"
18 /* When set, irq functions will be ignored */
19 static int ftrace_graph_skip_irqs
;
21 struct fgraph_cpu_data
{
26 unsigned long enter_funcs
[FTRACE_RETFUNC_DEPTH
];
30 struct fgraph_cpu_data __percpu
*cpu_data
;
32 /* Place to preserve last processed entry. */
33 struct ftrace_graph_ent_entry ent
;
34 struct ftrace_graph_ret_entry ret
;
39 #define TRACE_GRAPH_INDENT 2
42 #define TRACE_GRAPH_PRINT_OVERRUN 0x1
43 #define TRACE_GRAPH_PRINT_CPU 0x2
44 #define TRACE_GRAPH_PRINT_OVERHEAD 0x4
45 #define TRACE_GRAPH_PRINT_PROC 0x8
46 #define TRACE_GRAPH_PRINT_DURATION 0x10
47 #define TRACE_GRAPH_PRINT_ABS_TIME 0x20
48 #define TRACE_GRAPH_PRINT_IRQS 0x40
50 static unsigned int max_depth
;
52 static struct tracer_opt trace_opts
[] = {
53 /* Display overruns? (for self-debug purpose) */
54 { TRACER_OPT(funcgraph
-overrun
, TRACE_GRAPH_PRINT_OVERRUN
) },
56 { TRACER_OPT(funcgraph
-cpu
, TRACE_GRAPH_PRINT_CPU
) },
57 /* Display Overhead ? */
58 { TRACER_OPT(funcgraph
-overhead
, TRACE_GRAPH_PRINT_OVERHEAD
) },
59 /* Display proc name/pid */
60 { TRACER_OPT(funcgraph
-proc
, TRACE_GRAPH_PRINT_PROC
) },
61 /* Display duration of execution */
62 { TRACER_OPT(funcgraph
-duration
, TRACE_GRAPH_PRINT_DURATION
) },
63 /* Display absolute time of an entry */
64 { TRACER_OPT(funcgraph
-abstime
, TRACE_GRAPH_PRINT_ABS_TIME
) },
65 /* Display interrupts */
66 { TRACER_OPT(funcgraph
-irqs
, TRACE_GRAPH_PRINT_IRQS
) },
70 static struct tracer_flags tracer_flags
= {
71 /* Don't display overruns and proc by default */
72 .val
= TRACE_GRAPH_PRINT_CPU
| TRACE_GRAPH_PRINT_OVERHEAD
|
73 TRACE_GRAPH_PRINT_DURATION
| TRACE_GRAPH_PRINT_IRQS
,
77 static struct trace_array
*graph_array
;
80 * DURATION column is being also used to display IRQ signs,
81 * following values are used by print_graph_irq and others
82 * to fill in space into DURATION column.
85 FLAGS_FILL_FULL
= 1 << TRACE_GRAPH_PRINT_FILL_SHIFT
,
86 FLAGS_FILL_START
= 2 << TRACE_GRAPH_PRINT_FILL_SHIFT
,
87 FLAGS_FILL_END
= 3 << TRACE_GRAPH_PRINT_FILL_SHIFT
,
90 static enum print_line_t
91 print_graph_duration(unsigned long long duration
, struct trace_seq
*s
,
94 /* Add a function return address to the trace stack on thread info.*/
96 ftrace_push_return_trace(unsigned long ret
, unsigned long func
, int *depth
,
97 unsigned long frame_pointer
)
99 unsigned long long calltime
;
102 if (!current
->ret_stack
)
106 * We must make sure the ret_stack is tested before we read
111 /* The return trace stack is full */
112 if (current
->curr_ret_stack
== FTRACE_RETFUNC_DEPTH
- 1) {
113 atomic_inc(¤t
->trace_overrun
);
118 * The curr_ret_stack is an index to ftrace return stack of
119 * current task. Its value should be in [0, FTRACE_RETFUNC_
120 * DEPTH) when the function graph tracer is used. To support
121 * filtering out specific functions, it makes the index
122 * negative by subtracting huge value (FTRACE_NOTRACE_DEPTH)
123 * so when it sees a negative index the ftrace will ignore
124 * the record. And the index gets recovered when returning
125 * from the filtered function by adding the FTRACE_NOTRACE_
126 * DEPTH and then it'll continue to record functions normally.
128 * The curr_ret_stack is initialized to -1 and get increased
129 * in this function. So it can be less than -1 only if it was
130 * filtered out via ftrace_graph_notrace_addr() which can be
131 * set from set_graph_notrace file in debugfs by user.
133 if (current
->curr_ret_stack
< -1)
136 calltime
= trace_clock_local();
138 index
= ++current
->curr_ret_stack
;
139 if (ftrace_graph_notrace_addr(func
))
140 current
->curr_ret_stack
-= FTRACE_NOTRACE_DEPTH
;
142 current
->ret_stack
[index
].ret
= ret
;
143 current
->ret_stack
[index
].func
= func
;
144 current
->ret_stack
[index
].calltime
= calltime
;
145 current
->ret_stack
[index
].subtime
= 0;
146 current
->ret_stack
[index
].fp
= frame_pointer
;
147 *depth
= current
->curr_ret_stack
;
152 /* Retrieve a function return address to the trace stack on thread info.*/
154 ftrace_pop_return_trace(struct ftrace_graph_ret
*trace
, unsigned long *ret
,
155 unsigned long frame_pointer
)
159 index
= current
->curr_ret_stack
;
162 * A negative index here means that it's just returned from a
163 * notrace'd function. Recover index to get an original
164 * return address. See ftrace_push_return_trace().
166 * TODO: Need to check whether the stack gets corrupted.
169 index
+= FTRACE_NOTRACE_DEPTH
;
171 if (unlikely(index
< 0 || index
>= FTRACE_RETFUNC_DEPTH
)) {
174 /* Might as well panic, otherwise we have no where to go */
175 *ret
= (unsigned long)panic
;
179 #if defined(CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST) && !defined(CC_USING_FENTRY)
181 * The arch may choose to record the frame pointer used
182 * and check it here to make sure that it is what we expect it
183 * to be. If gcc does not set the place holder of the return
184 * address in the frame pointer, and does a copy instead, then
185 * the function graph trace will fail. This test detects this
188 * Currently, x86_32 with optimize for size (-Os) makes the latest
191 * Note, -mfentry does not use frame pointers, and this test
192 * is not needed if CC_USING_FENTRY is set.
194 if (unlikely(current
->ret_stack
[index
].fp
!= frame_pointer
)) {
196 WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
197 " from func %ps return to %lx\n",
198 current
->ret_stack
[index
].fp
,
200 (void *)current
->ret_stack
[index
].func
,
201 current
->ret_stack
[index
].ret
);
202 *ret
= (unsigned long)panic
;
207 *ret
= current
->ret_stack
[index
].ret
;
208 trace
->func
= current
->ret_stack
[index
].func
;
209 trace
->calltime
= current
->ret_stack
[index
].calltime
;
210 trace
->overrun
= atomic_read(¤t
->trace_overrun
);
211 trace
->depth
= index
;
215 * Send the trace to the ring-buffer.
216 * @return the original return address.
218 unsigned long ftrace_return_to_handler(unsigned long frame_pointer
)
220 struct ftrace_graph_ret trace
;
223 ftrace_pop_return_trace(&trace
, &ret
, frame_pointer
);
224 trace
.rettime
= trace_clock_local();
226 current
->curr_ret_stack
--;
228 * The curr_ret_stack can be less than -1 only if it was
229 * filtered out and it's about to return from the function.
230 * Recover the index and continue to trace normal functions.
232 if (current
->curr_ret_stack
< -1) {
233 current
->curr_ret_stack
+= FTRACE_NOTRACE_DEPTH
;
238 * The trace should run after decrementing the ret counter
239 * in case an interrupt were to come in. We don't want to
240 * lose the interrupt if max_depth is set.
242 ftrace_graph_return(&trace
);
244 if (unlikely(!ret
)) {
247 /* Might as well panic. What else to do? */
248 ret
= (unsigned long)panic
;
254 int __trace_graph_entry(struct trace_array
*tr
,
255 struct ftrace_graph_ent
*trace
,
259 struct ftrace_event_call
*call
= &event_funcgraph_entry
;
260 struct ring_buffer_event
*event
;
261 struct ring_buffer
*buffer
= tr
->trace_buffer
.buffer
;
262 struct ftrace_graph_ent_entry
*entry
;
264 if (unlikely(__this_cpu_read(ftrace_cpu_disabled
)))
267 event
= trace_buffer_lock_reserve(buffer
, TRACE_GRAPH_ENT
,
268 sizeof(*entry
), flags
, pc
);
271 entry
= ring_buffer_event_data(event
);
272 entry
->graph_ent
= *trace
;
273 if (!call_filter_check_discard(call
, entry
, buffer
, event
))
274 __buffer_unlock_commit(buffer
, event
);
279 static inline int ftrace_graph_ignore_irqs(void)
281 if (!ftrace_graph_skip_irqs
|| trace_recursion_test(TRACE_IRQ_BIT
))
287 int trace_graph_entry(struct ftrace_graph_ent
*trace
)
289 struct trace_array
*tr
= graph_array
;
290 struct trace_array_cpu
*data
;
297 if (!ftrace_trace_task(current
))
300 /* trace it when it is-nested-in or is a function enabled. */
301 if ((!(trace
->depth
|| ftrace_graph_addr(trace
->func
)) ||
302 ftrace_graph_ignore_irqs()) || (trace
->depth
< 0) ||
303 (max_depth
&& trace
->depth
>= max_depth
))
307 * Do not trace a function if it's filtered by set_graph_notrace.
308 * Make the index of ret stack negative to indicate that it should
309 * ignore further functions. But it needs its own ret stack entry
310 * to recover the original index in order to continue tracing after
311 * returning from the function.
313 if (ftrace_graph_notrace_addr(trace
->func
))
316 local_irq_save(flags
);
317 cpu
= raw_smp_processor_id();
318 data
= per_cpu_ptr(tr
->trace_buffer
.data
, cpu
);
319 disabled
= atomic_inc_return(&data
->disabled
);
320 if (likely(disabled
== 1)) {
321 pc
= preempt_count();
322 ret
= __trace_graph_entry(tr
, trace
, flags
, pc
);
327 atomic_dec(&data
->disabled
);
328 local_irq_restore(flags
);
333 int trace_graph_thresh_entry(struct ftrace_graph_ent
*trace
)
338 return trace_graph_entry(trace
);
342 __trace_graph_function(struct trace_array
*tr
,
343 unsigned long ip
, unsigned long flags
, int pc
)
345 u64 time
= trace_clock_local();
346 struct ftrace_graph_ent ent
= {
350 struct ftrace_graph_ret ret
= {
357 __trace_graph_entry(tr
, &ent
, flags
, pc
);
358 __trace_graph_return(tr
, &ret
, flags
, pc
);
362 trace_graph_function(struct trace_array
*tr
,
363 unsigned long ip
, unsigned long parent_ip
,
364 unsigned long flags
, int pc
)
366 __trace_graph_function(tr
, ip
, flags
, pc
);
369 void __trace_graph_return(struct trace_array
*tr
,
370 struct ftrace_graph_ret
*trace
,
374 struct ftrace_event_call
*call
= &event_funcgraph_exit
;
375 struct ring_buffer_event
*event
;
376 struct ring_buffer
*buffer
= tr
->trace_buffer
.buffer
;
377 struct ftrace_graph_ret_entry
*entry
;
379 if (unlikely(__this_cpu_read(ftrace_cpu_disabled
)))
382 event
= trace_buffer_lock_reserve(buffer
, TRACE_GRAPH_RET
,
383 sizeof(*entry
), flags
, pc
);
386 entry
= ring_buffer_event_data(event
);
388 if (!call_filter_check_discard(call
, entry
, buffer
, event
))
389 __buffer_unlock_commit(buffer
, event
);
392 void trace_graph_return(struct ftrace_graph_ret
*trace
)
394 struct trace_array
*tr
= graph_array
;
395 struct trace_array_cpu
*data
;
401 local_irq_save(flags
);
402 cpu
= raw_smp_processor_id();
403 data
= per_cpu_ptr(tr
->trace_buffer
.data
, cpu
);
404 disabled
= atomic_inc_return(&data
->disabled
);
405 if (likely(disabled
== 1)) {
406 pc
= preempt_count();
407 __trace_graph_return(tr
, trace
, flags
, pc
);
409 atomic_dec(&data
->disabled
);
410 local_irq_restore(flags
);
413 void set_graph_array(struct trace_array
*tr
)
417 /* Make graph_array visible before we start tracing */
422 void trace_graph_thresh_return(struct ftrace_graph_ret
*trace
)
424 if (tracing_thresh
&&
425 (trace
->rettime
- trace
->calltime
< tracing_thresh
))
428 trace_graph_return(trace
);
431 static int graph_trace_init(struct trace_array
*tr
)
437 ret
= register_ftrace_graph(&trace_graph_thresh_return
,
438 &trace_graph_thresh_entry
);
440 ret
= register_ftrace_graph(&trace_graph_return
,
444 tracing_start_cmdline_record();
449 static void graph_trace_reset(struct trace_array
*tr
)
451 tracing_stop_cmdline_record();
452 unregister_ftrace_graph();
455 static int max_bytes_for_cpu
;
457 static enum print_line_t
458 print_graph_cpu(struct trace_seq
*s
, int cpu
)
463 * Start with a space character - to make it stand out
464 * to the right a bit when trace output is pasted into
467 ret
= trace_seq_printf(s
, " %*d) ", max_bytes_for_cpu
, cpu
);
469 return TRACE_TYPE_PARTIAL_LINE
;
471 return TRACE_TYPE_HANDLED
;
474 #define TRACE_GRAPH_PROCINFO_LENGTH 14
476 static enum print_line_t
477 print_graph_proc(struct trace_seq
*s
, pid_t pid
)
479 char comm
[TASK_COMM_LEN
];
480 /* sign + log10(MAX_INT) + '\0' */
487 trace_find_cmdline(pid
, comm
);
489 sprintf(pid_str
, "%d", pid
);
491 /* 1 stands for the "-" character */
492 len
= strlen(comm
) + strlen(pid_str
) + 1;
494 if (len
< TRACE_GRAPH_PROCINFO_LENGTH
)
495 spaces
= TRACE_GRAPH_PROCINFO_LENGTH
- len
;
497 /* First spaces to align center */
498 for (i
= 0; i
< spaces
/ 2; i
++) {
499 ret
= trace_seq_putc(s
, ' ');
501 return TRACE_TYPE_PARTIAL_LINE
;
504 ret
= trace_seq_printf(s
, "%s-%s", comm
, pid_str
);
506 return TRACE_TYPE_PARTIAL_LINE
;
508 /* Last spaces to align center */
509 for (i
= 0; i
< spaces
- (spaces
/ 2); i
++) {
510 ret
= trace_seq_putc(s
, ' ');
512 return TRACE_TYPE_PARTIAL_LINE
;
514 return TRACE_TYPE_HANDLED
;
518 static enum print_line_t
519 print_graph_lat_fmt(struct trace_seq
*s
, struct trace_entry
*entry
)
521 if (!trace_seq_putc(s
, ' '))
524 return trace_print_lat_fmt(s
, entry
);
527 /* If the pid changed since the last trace, output this event */
528 static enum print_line_t
529 verif_pid(struct trace_seq
*s
, pid_t pid
, int cpu
, struct fgraph_data
*data
)
536 return TRACE_TYPE_HANDLED
;
538 last_pid
= &(per_cpu_ptr(data
->cpu_data
, cpu
)->last_pid
);
540 if (*last_pid
== pid
)
541 return TRACE_TYPE_HANDLED
;
543 prev_pid
= *last_pid
;
547 return TRACE_TYPE_HANDLED
;
549 * Context-switch trace line:
551 ------------------------------------------
552 | 1) migration/0--1 => sshd-1755
553 ------------------------------------------
556 ret
= trace_seq_puts(s
,
557 " ------------------------------------------\n");
559 return TRACE_TYPE_PARTIAL_LINE
;
561 ret
= print_graph_cpu(s
, cpu
);
562 if (ret
== TRACE_TYPE_PARTIAL_LINE
)
563 return TRACE_TYPE_PARTIAL_LINE
;
565 ret
= print_graph_proc(s
, prev_pid
);
566 if (ret
== TRACE_TYPE_PARTIAL_LINE
)
567 return TRACE_TYPE_PARTIAL_LINE
;
569 ret
= trace_seq_puts(s
, " => ");
571 return TRACE_TYPE_PARTIAL_LINE
;
573 ret
= print_graph_proc(s
, pid
);
574 if (ret
== TRACE_TYPE_PARTIAL_LINE
)
575 return TRACE_TYPE_PARTIAL_LINE
;
577 ret
= trace_seq_puts(s
,
578 "\n ------------------------------------------\n\n");
580 return TRACE_TYPE_PARTIAL_LINE
;
582 return TRACE_TYPE_HANDLED
;
585 static struct ftrace_graph_ret_entry
*
586 get_return_for_leaf(struct trace_iterator
*iter
,
587 struct ftrace_graph_ent_entry
*curr
)
589 struct fgraph_data
*data
= iter
->private;
590 struct ring_buffer_iter
*ring_iter
= NULL
;
591 struct ring_buffer_event
*event
;
592 struct ftrace_graph_ret_entry
*next
;
595 * If the previous output failed to write to the seq buffer,
596 * then we just reuse the data from before.
598 if (data
&& data
->failed
) {
603 ring_iter
= trace_buffer_iter(iter
, iter
->cpu
);
605 /* First peek to compare current entry and the next one */
607 event
= ring_buffer_iter_peek(ring_iter
, NULL
);
610 * We need to consume the current entry to see
613 ring_buffer_consume(iter
->trace_buffer
->buffer
, iter
->cpu
,
615 event
= ring_buffer_peek(iter
->trace_buffer
->buffer
, iter
->cpu
,
622 next
= ring_buffer_event_data(event
);
626 * Save current and next entries for later reference
627 * if the output fails.
631 * If the next event is not a return type, then
632 * we only care about what type it is. Otherwise we can
633 * safely copy the entire event.
635 if (next
->ent
.type
== TRACE_GRAPH_RET
)
638 data
->ret
.ent
.type
= next
->ent
.type
;
642 if (next
->ent
.type
!= TRACE_GRAPH_RET
)
645 if (curr
->ent
.pid
!= next
->ent
.pid
||
646 curr
->graph_ent
.func
!= next
->ret
.func
)
649 /* this is a leaf, now advance the iterator */
651 ring_buffer_read(ring_iter
, NULL
);
656 static int print_graph_abs_time(u64 t
, struct trace_seq
*s
)
658 unsigned long usecs_rem
;
660 usecs_rem
= do_div(t
, NSEC_PER_SEC
);
663 return trace_seq_printf(s
, "%5lu.%06lu | ",
664 (unsigned long)t
, usecs_rem
);
667 static enum print_line_t
668 print_graph_irq(struct trace_iterator
*iter
, unsigned long addr
,
669 enum trace_type type
, int cpu
, pid_t pid
, u32 flags
)
672 struct trace_seq
*s
= &iter
->seq
;
674 if (addr
< (unsigned long)__irqentry_text_start
||
675 addr
>= (unsigned long)__irqentry_text_end
)
676 return TRACE_TYPE_UNHANDLED
;
678 if (trace_flags
& TRACE_ITER_CONTEXT_INFO
) {
680 if (flags
& TRACE_GRAPH_PRINT_ABS_TIME
) {
681 ret
= print_graph_abs_time(iter
->ts
, s
);
683 return TRACE_TYPE_PARTIAL_LINE
;
687 if (flags
& TRACE_GRAPH_PRINT_CPU
) {
688 ret
= print_graph_cpu(s
, cpu
);
689 if (ret
== TRACE_TYPE_PARTIAL_LINE
)
690 return TRACE_TYPE_PARTIAL_LINE
;
694 if (flags
& TRACE_GRAPH_PRINT_PROC
) {
695 ret
= print_graph_proc(s
, pid
);
696 if (ret
== TRACE_TYPE_PARTIAL_LINE
)
697 return TRACE_TYPE_PARTIAL_LINE
;
698 ret
= trace_seq_puts(s
, " | ");
700 return TRACE_TYPE_PARTIAL_LINE
;
705 ret
= print_graph_duration(0, s
, flags
| FLAGS_FILL_START
);
706 if (ret
!= TRACE_TYPE_HANDLED
)
709 if (type
== TRACE_GRAPH_ENT
)
710 ret
= trace_seq_puts(s
, "==========>");
712 ret
= trace_seq_puts(s
, "<==========");
715 return TRACE_TYPE_PARTIAL_LINE
;
717 ret
= print_graph_duration(0, s
, flags
| FLAGS_FILL_END
);
718 if (ret
!= TRACE_TYPE_HANDLED
)
721 ret
= trace_seq_putc(s
, '\n');
724 return TRACE_TYPE_PARTIAL_LINE
;
725 return TRACE_TYPE_HANDLED
;
729 trace_print_graph_duration(unsigned long long duration
, struct trace_seq
*s
)
731 unsigned long nsecs_rem
= do_div(duration
, 1000);
732 /* log10(ULONG_MAX) + '\0' */
738 sprintf(msecs_str
, "%lu", (unsigned long) duration
);
741 ret
= trace_seq_printf(s
, "%s", msecs_str
);
743 return TRACE_TYPE_PARTIAL_LINE
;
745 len
= strlen(msecs_str
);
747 /* Print nsecs (we don't want to exceed 7 numbers) */
749 size_t slen
= min_t(size_t, sizeof(nsecs_str
), 8UL - len
);
751 snprintf(nsecs_str
, slen
, "%03lu", nsecs_rem
);
752 ret
= trace_seq_printf(s
, ".%s", nsecs_str
);
754 return TRACE_TYPE_PARTIAL_LINE
;
755 len
+= strlen(nsecs_str
);
758 ret
= trace_seq_puts(s
, " us ");
760 return TRACE_TYPE_PARTIAL_LINE
;
762 /* Print remaining spaces to fit the row's width */
763 for (i
= len
; i
< 7; i
++) {
764 ret
= trace_seq_putc(s
, ' ');
766 return TRACE_TYPE_PARTIAL_LINE
;
768 return TRACE_TYPE_HANDLED
;
771 static enum print_line_t
772 print_graph_duration(unsigned long long duration
, struct trace_seq
*s
,
777 if (!(flags
& TRACE_GRAPH_PRINT_DURATION
) ||
778 !(trace_flags
& TRACE_ITER_CONTEXT_INFO
))
779 return TRACE_TYPE_HANDLED
;
781 /* No real adata, just filling the column with spaces */
782 switch (flags
& TRACE_GRAPH_PRINT_FILL_MASK
) {
783 case FLAGS_FILL_FULL
:
784 ret
= trace_seq_puts(s
, " | ");
785 return ret
? TRACE_TYPE_HANDLED
: TRACE_TYPE_PARTIAL_LINE
;
786 case FLAGS_FILL_START
:
787 ret
= trace_seq_puts(s
, " ");
788 return ret
? TRACE_TYPE_HANDLED
: TRACE_TYPE_PARTIAL_LINE
;
790 ret
= trace_seq_puts(s
, " |");
791 return ret
? TRACE_TYPE_HANDLED
: TRACE_TYPE_PARTIAL_LINE
;
794 /* Signal a overhead of time execution to the output */
795 if (flags
& TRACE_GRAPH_PRINT_OVERHEAD
) {
796 /* Duration exceeded 100 msecs */
797 if (duration
> 100000ULL)
798 ret
= trace_seq_puts(s
, "! ");
799 /* Duration exceeded 10 msecs */
800 else if (duration
> 10000ULL)
801 ret
= trace_seq_puts(s
, "+ ");
805 * The -1 means we either did not exceed the duration tresholds
806 * or we dont want to print out the overhead. Either way we need
807 * to fill out the space.
810 ret
= trace_seq_puts(s
, " ");
812 /* Catching here any failure happenned above */
814 return TRACE_TYPE_PARTIAL_LINE
;
816 ret
= trace_print_graph_duration(duration
, s
);
817 if (ret
!= TRACE_TYPE_HANDLED
)
820 ret
= trace_seq_puts(s
, "| ");
822 return TRACE_TYPE_PARTIAL_LINE
;
824 return TRACE_TYPE_HANDLED
;
827 /* Case of a leaf function on its call entry */
828 static enum print_line_t
829 print_graph_entry_leaf(struct trace_iterator
*iter
,
830 struct ftrace_graph_ent_entry
*entry
,
831 struct ftrace_graph_ret_entry
*ret_entry
,
832 struct trace_seq
*s
, u32 flags
)
834 struct fgraph_data
*data
= iter
->private;
835 struct ftrace_graph_ret
*graph_ret
;
836 struct ftrace_graph_ent
*call
;
837 unsigned long long duration
;
841 graph_ret
= &ret_entry
->ret
;
842 call
= &entry
->graph_ent
;
843 duration
= graph_ret
->rettime
- graph_ret
->calltime
;
846 struct fgraph_cpu_data
*cpu_data
;
849 cpu_data
= per_cpu_ptr(data
->cpu_data
, cpu
);
852 * Comments display at + 1 to depth. Since
853 * this is a leaf function, keep the comments
854 * equal to this depth.
856 cpu_data
->depth
= call
->depth
- 1;
858 /* No need to keep this function around for this depth */
859 if (call
->depth
< FTRACE_RETFUNC_DEPTH
)
860 cpu_data
->enter_funcs
[call
->depth
] = 0;
863 /* Overhead and duration */
864 ret
= print_graph_duration(duration
, s
, flags
);
865 if (ret
== TRACE_TYPE_PARTIAL_LINE
)
866 return TRACE_TYPE_PARTIAL_LINE
;
869 for (i
= 0; i
< call
->depth
* TRACE_GRAPH_INDENT
; i
++) {
870 ret
= trace_seq_putc(s
, ' ');
872 return TRACE_TYPE_PARTIAL_LINE
;
875 ret
= trace_seq_printf(s
, "%ps();\n", (void *)call
->func
);
877 return TRACE_TYPE_PARTIAL_LINE
;
879 return TRACE_TYPE_HANDLED
;
882 static enum print_line_t
883 print_graph_entry_nested(struct trace_iterator
*iter
,
884 struct ftrace_graph_ent_entry
*entry
,
885 struct trace_seq
*s
, int cpu
, u32 flags
)
887 struct ftrace_graph_ent
*call
= &entry
->graph_ent
;
888 struct fgraph_data
*data
= iter
->private;
893 struct fgraph_cpu_data
*cpu_data
;
896 cpu_data
= per_cpu_ptr(data
->cpu_data
, cpu
);
897 cpu_data
->depth
= call
->depth
;
899 /* Save this function pointer to see if the exit matches */
900 if (call
->depth
< FTRACE_RETFUNC_DEPTH
)
901 cpu_data
->enter_funcs
[call
->depth
] = call
->func
;
905 ret
= print_graph_duration(0, s
, flags
| FLAGS_FILL_FULL
);
906 if (ret
!= TRACE_TYPE_HANDLED
)
910 for (i
= 0; i
< call
->depth
* TRACE_GRAPH_INDENT
; i
++) {
911 ret
= trace_seq_putc(s
, ' ');
913 return TRACE_TYPE_PARTIAL_LINE
;
916 ret
= trace_seq_printf(s
, "%ps() {\n", (void *)call
->func
);
918 return TRACE_TYPE_PARTIAL_LINE
;
921 * we already consumed the current entry to check the next one
922 * and see if this is a leaf.
924 return TRACE_TYPE_NO_CONSUME
;
927 static enum print_line_t
928 print_graph_prologue(struct trace_iterator
*iter
, struct trace_seq
*s
,
929 int type
, unsigned long addr
, u32 flags
)
931 struct fgraph_data
*data
= iter
->private;
932 struct trace_entry
*ent
= iter
->ent
;
937 if (verif_pid(s
, ent
->pid
, cpu
, data
) == TRACE_TYPE_PARTIAL_LINE
)
938 return TRACE_TYPE_PARTIAL_LINE
;
942 ret
= print_graph_irq(iter
, addr
, type
, cpu
, ent
->pid
, flags
);
943 if (ret
== TRACE_TYPE_PARTIAL_LINE
)
944 return TRACE_TYPE_PARTIAL_LINE
;
947 if (!(trace_flags
& TRACE_ITER_CONTEXT_INFO
))
951 if (flags
& TRACE_GRAPH_PRINT_ABS_TIME
) {
952 ret
= print_graph_abs_time(iter
->ts
, s
);
954 return TRACE_TYPE_PARTIAL_LINE
;
958 if (flags
& TRACE_GRAPH_PRINT_CPU
) {
959 ret
= print_graph_cpu(s
, cpu
);
960 if (ret
== TRACE_TYPE_PARTIAL_LINE
)
961 return TRACE_TYPE_PARTIAL_LINE
;
965 if (flags
& TRACE_GRAPH_PRINT_PROC
) {
966 ret
= print_graph_proc(s
, ent
->pid
);
967 if (ret
== TRACE_TYPE_PARTIAL_LINE
)
968 return TRACE_TYPE_PARTIAL_LINE
;
970 ret
= trace_seq_puts(s
, " | ");
972 return TRACE_TYPE_PARTIAL_LINE
;
976 if (trace_flags
& TRACE_ITER_LATENCY_FMT
) {
977 ret
= print_graph_lat_fmt(s
, ent
);
978 if (ret
== TRACE_TYPE_PARTIAL_LINE
)
979 return TRACE_TYPE_PARTIAL_LINE
;
986 * Entry check for irq code
989 * - we are inside irq code
990 * - we just entered irq code
993 * - funcgraph-interrupts option is set
994 * - we are not inside irq code
997 check_irq_entry(struct trace_iterator
*iter
, u32 flags
,
998 unsigned long addr
, int depth
)
1000 int cpu
= iter
->cpu
;
1002 struct fgraph_data
*data
= iter
->private;
1005 * If we are either displaying irqs, or we got called as
1006 * a graph event and private data does not exist,
1007 * then we bypass the irq check.
1009 if ((flags
& TRACE_GRAPH_PRINT_IRQS
) ||
1013 depth_irq
= &(per_cpu_ptr(data
->cpu_data
, cpu
)->depth_irq
);
1016 * We are inside the irq code
1018 if (*depth_irq
>= 0)
1021 if ((addr
< (unsigned long)__irqentry_text_start
) ||
1022 (addr
>= (unsigned long)__irqentry_text_end
))
1026 * We are entering irq code.
1033 * Return check for irq code
1036 * - we are inside irq code
1037 * - we just left irq code
1040 * - funcgraph-interrupts option is set
1041 * - we are not inside irq code
1044 check_irq_return(struct trace_iterator
*iter
, u32 flags
, int depth
)
1046 int cpu
= iter
->cpu
;
1048 struct fgraph_data
*data
= iter
->private;
1051 * If we are either displaying irqs, or we got called as
1052 * a graph event and private data does not exist,
1053 * then we bypass the irq check.
1055 if ((flags
& TRACE_GRAPH_PRINT_IRQS
) ||
1059 depth_irq
= &(per_cpu_ptr(data
->cpu_data
, cpu
)->depth_irq
);
1062 * We are not inside the irq code.
1064 if (*depth_irq
== -1)
1068 * We are inside the irq code, and this is returning entry.
1069 * Let's not trace it and clear the entry depth, since
1070 * we are out of irq code.
1072 * This condition ensures that we 'leave the irq code' once
1073 * we are out of the entry depth. Thus protecting us from
1074 * the RETURN entry loss.
1076 if (*depth_irq
>= depth
) {
1082 * We are inside the irq code, and this is not the entry.
1087 static enum print_line_t
1088 print_graph_entry(struct ftrace_graph_ent_entry
*field
, struct trace_seq
*s
,
1089 struct trace_iterator
*iter
, u32 flags
)
1091 struct fgraph_data
*data
= iter
->private;
1092 struct ftrace_graph_ent
*call
= &field
->graph_ent
;
1093 struct ftrace_graph_ret_entry
*leaf_ret
;
1094 static enum print_line_t ret
;
1095 int cpu
= iter
->cpu
;
1097 if (check_irq_entry(iter
, flags
, call
->func
, call
->depth
))
1098 return TRACE_TYPE_HANDLED
;
1100 if (print_graph_prologue(iter
, s
, TRACE_GRAPH_ENT
, call
->func
, flags
))
1101 return TRACE_TYPE_PARTIAL_LINE
;
1103 leaf_ret
= get_return_for_leaf(iter
, field
);
1105 ret
= print_graph_entry_leaf(iter
, field
, leaf_ret
, s
, flags
);
1107 ret
= print_graph_entry_nested(iter
, field
, s
, cpu
, flags
);
1111 * If we failed to write our output, then we need to make
1112 * note of it. Because we already consumed our entry.
1124 static enum print_line_t
1125 print_graph_return(struct ftrace_graph_ret
*trace
, struct trace_seq
*s
,
1126 struct trace_entry
*ent
, struct trace_iterator
*iter
,
1129 unsigned long long duration
= trace
->rettime
- trace
->calltime
;
1130 struct fgraph_data
*data
= iter
->private;
1131 pid_t pid
= ent
->pid
;
1132 int cpu
= iter
->cpu
;
1137 if (check_irq_return(iter
, flags
, trace
->depth
))
1138 return TRACE_TYPE_HANDLED
;
1141 struct fgraph_cpu_data
*cpu_data
;
1142 int cpu
= iter
->cpu
;
1144 cpu_data
= per_cpu_ptr(data
->cpu_data
, cpu
);
1147 * Comments display at + 1 to depth. This is the
1148 * return from a function, we now want the comments
1149 * to display at the same level of the bracket.
1151 cpu_data
->depth
= trace
->depth
- 1;
1153 if (trace
->depth
< FTRACE_RETFUNC_DEPTH
) {
1154 if (cpu_data
->enter_funcs
[trace
->depth
] != trace
->func
)
1156 cpu_data
->enter_funcs
[trace
->depth
] = 0;
1160 if (print_graph_prologue(iter
, s
, 0, 0, flags
))
1161 return TRACE_TYPE_PARTIAL_LINE
;
1163 /* Overhead and duration */
1164 ret
= print_graph_duration(duration
, s
, flags
);
1165 if (ret
== TRACE_TYPE_PARTIAL_LINE
)
1166 return TRACE_TYPE_PARTIAL_LINE
;
1169 for (i
= 0; i
< trace
->depth
* TRACE_GRAPH_INDENT
; i
++) {
1170 ret
= trace_seq_putc(s
, ' ');
1172 return TRACE_TYPE_PARTIAL_LINE
;
1176 * If the return function does not have a matching entry,
1177 * then the entry was lost. Instead of just printing
1178 * the '}' and letting the user guess what function this
1179 * belongs to, write out the function name.
1182 ret
= trace_seq_puts(s
, "}\n");
1184 return TRACE_TYPE_PARTIAL_LINE
;
1186 ret
= trace_seq_printf(s
, "} /* %ps */\n", (void *)trace
->func
);
1188 return TRACE_TYPE_PARTIAL_LINE
;
1192 if (flags
& TRACE_GRAPH_PRINT_OVERRUN
) {
1193 ret
= trace_seq_printf(s
, " (Overruns: %lu)\n",
1196 return TRACE_TYPE_PARTIAL_LINE
;
1199 ret
= print_graph_irq(iter
, trace
->func
, TRACE_GRAPH_RET
,
1201 if (ret
== TRACE_TYPE_PARTIAL_LINE
)
1202 return TRACE_TYPE_PARTIAL_LINE
;
1204 return TRACE_TYPE_HANDLED
;
1207 static enum print_line_t
1208 print_graph_comment(struct trace_seq
*s
, struct trace_entry
*ent
,
1209 struct trace_iterator
*iter
, u32 flags
)
1211 unsigned long sym_flags
= (trace_flags
& TRACE_ITER_SYM_MASK
);
1212 struct fgraph_data
*data
= iter
->private;
1213 struct trace_event
*event
;
1219 depth
= per_cpu_ptr(data
->cpu_data
, iter
->cpu
)->depth
;
1221 if (print_graph_prologue(iter
, s
, 0, 0, flags
))
1222 return TRACE_TYPE_PARTIAL_LINE
;
1225 ret
= print_graph_duration(0, s
, flags
| FLAGS_FILL_FULL
);
1226 if (ret
!= TRACE_TYPE_HANDLED
)
1231 for (i
= 0; i
< (depth
+ 1) * TRACE_GRAPH_INDENT
; i
++) {
1232 ret
= trace_seq_putc(s
, ' ');
1234 return TRACE_TYPE_PARTIAL_LINE
;
1238 ret
= trace_seq_puts(s
, "/* ");
1240 return TRACE_TYPE_PARTIAL_LINE
;
1242 switch (iter
->ent
->type
) {
1244 ret
= trace_print_bprintk_msg_only(iter
);
1245 if (ret
!= TRACE_TYPE_HANDLED
)
1249 ret
= trace_print_printk_msg_only(iter
);
1250 if (ret
!= TRACE_TYPE_HANDLED
)
1254 event
= ftrace_find_event(ent
->type
);
1256 return TRACE_TYPE_UNHANDLED
;
1258 ret
= event
->funcs
->trace(iter
, sym_flags
, event
);
1259 if (ret
!= TRACE_TYPE_HANDLED
)
1263 /* Strip ending newline */
1264 if (s
->buffer
[s
->len
- 1] == '\n') {
1265 s
->buffer
[s
->len
- 1] = '\0';
1269 ret
= trace_seq_puts(s
, " */\n");
1271 return TRACE_TYPE_PARTIAL_LINE
;
1273 return TRACE_TYPE_HANDLED
;
1278 print_graph_function_flags(struct trace_iterator
*iter
, u32 flags
)
1280 struct ftrace_graph_ent_entry
*field
;
1281 struct fgraph_data
*data
= iter
->private;
1282 struct trace_entry
*entry
= iter
->ent
;
1283 struct trace_seq
*s
= &iter
->seq
;
1284 int cpu
= iter
->cpu
;
1287 if (data
&& per_cpu_ptr(data
->cpu_data
, cpu
)->ignore
) {
1288 per_cpu_ptr(data
->cpu_data
, cpu
)->ignore
= 0;
1289 return TRACE_TYPE_HANDLED
;
1293 * If the last output failed, there's a possibility we need
1294 * to print out the missing entry which would never go out.
1296 if (data
&& data
->failed
) {
1298 iter
->cpu
= data
->cpu
;
1299 ret
= print_graph_entry(field
, s
, iter
, flags
);
1300 if (ret
== TRACE_TYPE_HANDLED
&& iter
->cpu
!= cpu
) {
1301 per_cpu_ptr(data
->cpu_data
, iter
->cpu
)->ignore
= 1;
1302 ret
= TRACE_TYPE_NO_CONSUME
;
1308 switch (entry
->type
) {
1309 case TRACE_GRAPH_ENT
: {
1311 * print_graph_entry() may consume the current event,
1312 * thus @field may become invalid, so we need to save it.
1313 * sizeof(struct ftrace_graph_ent_entry) is very small,
1314 * it can be safely saved at the stack.
1316 struct ftrace_graph_ent_entry saved
;
1317 trace_assign_type(field
, entry
);
1319 return print_graph_entry(&saved
, s
, iter
, flags
);
1321 case TRACE_GRAPH_RET
: {
1322 struct ftrace_graph_ret_entry
*field
;
1323 trace_assign_type(field
, entry
);
1324 return print_graph_return(&field
->ret
, s
, entry
, iter
, flags
);
1328 /* dont trace stack and functions as comments */
1329 return TRACE_TYPE_UNHANDLED
;
1332 return print_graph_comment(s
, entry
, iter
, flags
);
1335 return TRACE_TYPE_HANDLED
;
1338 static enum print_line_t
1339 print_graph_function(struct trace_iterator
*iter
)
1341 return print_graph_function_flags(iter
, tracer_flags
.val
);
1344 static enum print_line_t
1345 print_graph_function_event(struct trace_iterator
*iter
, int flags
,
1346 struct trace_event
*event
)
1348 return print_graph_function(iter
);
1351 static void print_lat_header(struct seq_file
*s
, u32 flags
)
1353 static const char spaces
[] = " " /* 16 spaces */
1355 " "; /* 17 spaces */
1358 if (flags
& TRACE_GRAPH_PRINT_ABS_TIME
)
1360 if (flags
& TRACE_GRAPH_PRINT_CPU
)
1362 if (flags
& TRACE_GRAPH_PRINT_PROC
)
1365 seq_printf(s
, "#%.*s _-----=> irqs-off \n", size
, spaces
);
1366 seq_printf(s
, "#%.*s / _----=> need-resched \n", size
, spaces
);
1367 seq_printf(s
, "#%.*s| / _---=> hardirq/softirq \n", size
, spaces
);
1368 seq_printf(s
, "#%.*s|| / _--=> preempt-depth \n", size
, spaces
);
1369 seq_printf(s
, "#%.*s||| / \n", size
, spaces
);
1372 static void __print_graph_headers_flags(struct seq_file
*s
, u32 flags
)
1374 int lat
= trace_flags
& TRACE_ITER_LATENCY_FMT
;
1377 print_lat_header(s
, flags
);
1381 if (flags
& TRACE_GRAPH_PRINT_ABS_TIME
)
1382 seq_printf(s
, " TIME ");
1383 if (flags
& TRACE_GRAPH_PRINT_CPU
)
1384 seq_printf(s
, " CPU");
1385 if (flags
& TRACE_GRAPH_PRINT_PROC
)
1386 seq_printf(s
, " TASK/PID ");
1388 seq_printf(s
, "||||");
1389 if (flags
& TRACE_GRAPH_PRINT_DURATION
)
1390 seq_printf(s
, " DURATION ");
1391 seq_printf(s
, " FUNCTION CALLS\n");
1395 if (flags
& TRACE_GRAPH_PRINT_ABS_TIME
)
1396 seq_printf(s
, " | ");
1397 if (flags
& TRACE_GRAPH_PRINT_CPU
)
1398 seq_printf(s
, " | ");
1399 if (flags
& TRACE_GRAPH_PRINT_PROC
)
1400 seq_printf(s
, " | | ");
1402 seq_printf(s
, "||||");
1403 if (flags
& TRACE_GRAPH_PRINT_DURATION
)
1404 seq_printf(s
, " | | ");
1405 seq_printf(s
, " | | | |\n");
1408 void print_graph_headers(struct seq_file
*s
)
1410 print_graph_headers_flags(s
, tracer_flags
.val
);
1413 void print_graph_headers_flags(struct seq_file
*s
, u32 flags
)
1415 struct trace_iterator
*iter
= s
->private;
1417 if (!(trace_flags
& TRACE_ITER_CONTEXT_INFO
))
1420 if (trace_flags
& TRACE_ITER_LATENCY_FMT
) {
1421 /* print nothing if the buffers are empty */
1422 if (trace_empty(iter
))
1425 print_trace_header(s
, iter
);
1428 __print_graph_headers_flags(s
, flags
);
1431 void graph_trace_open(struct trace_iterator
*iter
)
1433 /* pid and depth on the last trace processed */
1434 struct fgraph_data
*data
;
1437 iter
->private = NULL
;
1439 data
= kzalloc(sizeof(*data
), GFP_KERNEL
);
1443 data
->cpu_data
= alloc_percpu(struct fgraph_cpu_data
);
1444 if (!data
->cpu_data
)
1447 for_each_possible_cpu(cpu
) {
1448 pid_t
*pid
= &(per_cpu_ptr(data
->cpu_data
, cpu
)->last_pid
);
1449 int *depth
= &(per_cpu_ptr(data
->cpu_data
, cpu
)->depth
);
1450 int *ignore
= &(per_cpu_ptr(data
->cpu_data
, cpu
)->ignore
);
1451 int *depth_irq
= &(per_cpu_ptr(data
->cpu_data
, cpu
)->depth_irq
);
1459 iter
->private = data
;
1466 pr_warning("function graph tracer: not enough memory\n");
1469 void graph_trace_close(struct trace_iterator
*iter
)
1471 struct fgraph_data
*data
= iter
->private;
1474 free_percpu(data
->cpu_data
);
1479 static int func_graph_set_flag(u32 old_flags
, u32 bit
, int set
)
1481 if (bit
== TRACE_GRAPH_PRINT_IRQS
)
1482 ftrace_graph_skip_irqs
= !set
;
1487 static struct trace_event_functions graph_functions
= {
1488 .trace
= print_graph_function_event
,
1491 static struct trace_event graph_trace_entry_event
= {
1492 .type
= TRACE_GRAPH_ENT
,
1493 .funcs
= &graph_functions
,
1496 static struct trace_event graph_trace_ret_event
= {
1497 .type
= TRACE_GRAPH_RET
,
1498 .funcs
= &graph_functions
1501 static struct tracer graph_trace __tracer_data
= {
1502 .name
= "function_graph",
1503 .open
= graph_trace_open
,
1504 .pipe_open
= graph_trace_open
,
1505 .close
= graph_trace_close
,
1506 .pipe_close
= graph_trace_close
,
1507 .wait_pipe
= poll_wait_pipe
,
1508 .init
= graph_trace_init
,
1509 .reset
= graph_trace_reset
,
1510 .print_line
= print_graph_function
,
1511 .print_header
= print_graph_headers
,
1512 .flags
= &tracer_flags
,
1513 .set_flag
= func_graph_set_flag
,
1514 #ifdef CONFIG_FTRACE_SELFTEST
1515 .selftest
= trace_selftest_startup_function_graph
,
1521 graph_depth_write(struct file
*filp
, const char __user
*ubuf
, size_t cnt
,
1527 ret
= kstrtoul_from_user(ubuf
, cnt
, 10, &val
);
1539 graph_depth_read(struct file
*filp
, char __user
*ubuf
, size_t cnt
,
1542 char buf
[15]; /* More than enough to hold UINT_MAX + "\n"*/
1545 n
= sprintf(buf
, "%d\n", max_depth
);
1547 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, n
);
1550 static const struct file_operations graph_depth_fops
= {
1551 .open
= tracing_open_generic
,
1552 .write
= graph_depth_write
,
1553 .read
= graph_depth_read
,
1554 .llseek
= generic_file_llseek
,
1557 static __init
int init_graph_debugfs(void)
1559 struct dentry
*d_tracer
;
1561 d_tracer
= tracing_init_dentry();
1565 trace_create_file("max_graph_depth", 0644, d_tracer
,
1566 NULL
, &graph_depth_fops
);
1570 fs_initcall(init_graph_debugfs
);
1572 static __init
int init_graph_trace(void)
1574 max_bytes_for_cpu
= snprintf(NULL
, 0, "%d", nr_cpu_ids
- 1);
1576 if (!register_ftrace_event(&graph_trace_entry_event
)) {
1577 pr_warning("Warning: could not register graph trace events\n");
1581 if (!register_ftrace_event(&graph_trace_ret_event
)) {
1582 pr_warning("Warning: could not register graph trace events\n");
1586 return register_tracer(&graph_trace
);
1589 core_initcall(init_graph_trace
);