Merge tag 'fixes-non-critical-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6.git] / kernel / trace / trace_functions_graph.c
blob8388bc99f2eee165efacbef7ddde4f7f21d49ea3
1 /*
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>
8 */
9 #include <linux/debugfs.h>
10 #include <linux/uaccess.h>
11 #include <linux/ftrace.h>
12 #include <linux/slab.h>
13 #include <linux/fs.h>
15 #include "trace.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 {
22 pid_t last_pid;
23 int depth;
24 int depth_irq;
25 int ignore;
26 unsigned long enter_funcs[FTRACE_RETFUNC_DEPTH];
29 struct fgraph_data {
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;
35 int failed;
36 int cpu;
39 #define TRACE_GRAPH_INDENT 2
41 /* Flag options */
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) },
55 /* Display CPU ? */
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) },
67 { } /* Empty entry */
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,
74 .opts = trace_opts
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.
84 enum {
85 DURATION_FILL_FULL = -1,
86 DURATION_FILL_START = -2,
87 DURATION_FILL_END = -3,
90 static enum print_line_t
91 print_graph_duration(unsigned long long duration, struct trace_seq *s,
92 u32 flags);
94 /* Add a function return address to the trace stack on thread info.*/
95 int
96 ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
97 unsigned long frame_pointer)
99 unsigned long long calltime;
100 int index;
102 if (!current->ret_stack)
103 return -EBUSY;
106 * We must make sure the ret_stack is tested before we read
107 * anything else.
109 smp_rmb();
111 /* The return trace stack is full */
112 if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
113 atomic_inc(&current->trace_overrun);
114 return -EBUSY;
117 calltime = trace_clock_local();
119 index = ++current->curr_ret_stack;
120 barrier();
121 current->ret_stack[index].ret = ret;
122 current->ret_stack[index].func = func;
123 current->ret_stack[index].calltime = calltime;
124 current->ret_stack[index].subtime = 0;
125 current->ret_stack[index].fp = frame_pointer;
126 *depth = index;
128 return 0;
131 /* Retrieve a function return address to the trace stack on thread info.*/
132 static void
133 ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
134 unsigned long frame_pointer)
136 int index;
138 index = current->curr_ret_stack;
140 if (unlikely(index < 0)) {
141 ftrace_graph_stop();
142 WARN_ON(1);
143 /* Might as well panic, otherwise we have no where to go */
144 *ret = (unsigned long)panic;
145 return;
148 #if defined(CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST) && !defined(CC_USING_FENTRY)
150 * The arch may choose to record the frame pointer used
151 * and check it here to make sure that it is what we expect it
152 * to be. If gcc does not set the place holder of the return
153 * address in the frame pointer, and does a copy instead, then
154 * the function graph trace will fail. This test detects this
155 * case.
157 * Currently, x86_32 with optimize for size (-Os) makes the latest
158 * gcc do the above.
160 * Note, -mfentry does not use frame pointers, and this test
161 * is not needed if CC_USING_FENTRY is set.
163 if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
164 ftrace_graph_stop();
165 WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
166 " from func %ps return to %lx\n",
167 current->ret_stack[index].fp,
168 frame_pointer,
169 (void *)current->ret_stack[index].func,
170 current->ret_stack[index].ret);
171 *ret = (unsigned long)panic;
172 return;
174 #endif
176 *ret = current->ret_stack[index].ret;
177 trace->func = current->ret_stack[index].func;
178 trace->calltime = current->ret_stack[index].calltime;
179 trace->overrun = atomic_read(&current->trace_overrun);
180 trace->depth = index;
184 * Send the trace to the ring-buffer.
185 * @return the original return address.
187 unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
189 struct ftrace_graph_ret trace;
190 unsigned long ret;
192 ftrace_pop_return_trace(&trace, &ret, frame_pointer);
193 trace.rettime = trace_clock_local();
194 barrier();
195 current->curr_ret_stack--;
198 * The trace should run after decrementing the ret counter
199 * in case an interrupt were to come in. We don't want to
200 * lose the interrupt if max_depth is set.
202 ftrace_graph_return(&trace);
204 if (unlikely(!ret)) {
205 ftrace_graph_stop();
206 WARN_ON(1);
207 /* Might as well panic. What else to do? */
208 ret = (unsigned long)panic;
211 return ret;
214 int __trace_graph_entry(struct trace_array *tr,
215 struct ftrace_graph_ent *trace,
216 unsigned long flags,
217 int pc)
219 struct ftrace_event_call *call = &event_funcgraph_entry;
220 struct ring_buffer_event *event;
221 struct ring_buffer *buffer = tr->trace_buffer.buffer;
222 struct ftrace_graph_ent_entry *entry;
224 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
225 return 0;
227 event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
228 sizeof(*entry), flags, pc);
229 if (!event)
230 return 0;
231 entry = ring_buffer_event_data(event);
232 entry->graph_ent = *trace;
233 if (!filter_current_check_discard(buffer, call, entry, event))
234 __buffer_unlock_commit(buffer, event);
236 return 1;
239 static inline int ftrace_graph_ignore_irqs(void)
241 if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
242 return 0;
244 return in_irq();
247 int trace_graph_entry(struct ftrace_graph_ent *trace)
249 struct trace_array *tr = graph_array;
250 struct trace_array_cpu *data;
251 unsigned long flags;
252 long disabled;
253 int ret;
254 int cpu;
255 int pc;
257 if (!ftrace_trace_task(current))
258 return 0;
260 /* trace it when it is-nested-in or is a function enabled. */
261 if ((!(trace->depth || ftrace_graph_addr(trace->func)) ||
262 ftrace_graph_ignore_irqs()) ||
263 (max_depth && trace->depth >= max_depth))
264 return 0;
266 local_irq_save(flags);
267 cpu = raw_smp_processor_id();
268 data = per_cpu_ptr(tr->trace_buffer.data, cpu);
269 disabled = atomic_inc_return(&data->disabled);
270 if (likely(disabled == 1)) {
271 pc = preempt_count();
272 ret = __trace_graph_entry(tr, trace, flags, pc);
273 } else {
274 ret = 0;
277 atomic_dec(&data->disabled);
278 local_irq_restore(flags);
280 return ret;
283 int trace_graph_thresh_entry(struct ftrace_graph_ent *trace)
285 if (tracing_thresh)
286 return 1;
287 else
288 return trace_graph_entry(trace);
291 static void
292 __trace_graph_function(struct trace_array *tr,
293 unsigned long ip, unsigned long flags, int pc)
295 u64 time = trace_clock_local();
296 struct ftrace_graph_ent ent = {
297 .func = ip,
298 .depth = 0,
300 struct ftrace_graph_ret ret = {
301 .func = ip,
302 .depth = 0,
303 .calltime = time,
304 .rettime = time,
307 __trace_graph_entry(tr, &ent, flags, pc);
308 __trace_graph_return(tr, &ret, flags, pc);
311 void
312 trace_graph_function(struct trace_array *tr,
313 unsigned long ip, unsigned long parent_ip,
314 unsigned long flags, int pc)
316 __trace_graph_function(tr, ip, flags, pc);
319 void __trace_graph_return(struct trace_array *tr,
320 struct ftrace_graph_ret *trace,
321 unsigned long flags,
322 int pc)
324 struct ftrace_event_call *call = &event_funcgraph_exit;
325 struct ring_buffer_event *event;
326 struct ring_buffer *buffer = tr->trace_buffer.buffer;
327 struct ftrace_graph_ret_entry *entry;
329 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
330 return;
332 event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
333 sizeof(*entry), flags, pc);
334 if (!event)
335 return;
336 entry = ring_buffer_event_data(event);
337 entry->ret = *trace;
338 if (!filter_current_check_discard(buffer, call, entry, event))
339 __buffer_unlock_commit(buffer, event);
342 void trace_graph_return(struct ftrace_graph_ret *trace)
344 struct trace_array *tr = graph_array;
345 struct trace_array_cpu *data;
346 unsigned long flags;
347 long disabled;
348 int cpu;
349 int pc;
351 local_irq_save(flags);
352 cpu = raw_smp_processor_id();
353 data = per_cpu_ptr(tr->trace_buffer.data, cpu);
354 disabled = atomic_inc_return(&data->disabled);
355 if (likely(disabled == 1)) {
356 pc = preempt_count();
357 __trace_graph_return(tr, trace, flags, pc);
359 atomic_dec(&data->disabled);
360 local_irq_restore(flags);
363 void set_graph_array(struct trace_array *tr)
365 graph_array = tr;
367 /* Make graph_array visible before we start tracing */
369 smp_mb();
372 void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
374 if (tracing_thresh &&
375 (trace->rettime - trace->calltime < tracing_thresh))
376 return;
377 else
378 trace_graph_return(trace);
381 static int graph_trace_init(struct trace_array *tr)
383 int ret;
385 set_graph_array(tr);
386 if (tracing_thresh)
387 ret = register_ftrace_graph(&trace_graph_thresh_return,
388 &trace_graph_thresh_entry);
389 else
390 ret = register_ftrace_graph(&trace_graph_return,
391 &trace_graph_entry);
392 if (ret)
393 return ret;
394 tracing_start_cmdline_record();
396 return 0;
399 static void graph_trace_reset(struct trace_array *tr)
401 tracing_stop_cmdline_record();
402 unregister_ftrace_graph();
405 static int max_bytes_for_cpu;
407 static enum print_line_t
408 print_graph_cpu(struct trace_seq *s, int cpu)
410 int ret;
413 * Start with a space character - to make it stand out
414 * to the right a bit when trace output is pasted into
415 * email:
417 ret = trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
418 if (!ret)
419 return TRACE_TYPE_PARTIAL_LINE;
421 return TRACE_TYPE_HANDLED;
424 #define TRACE_GRAPH_PROCINFO_LENGTH 14
426 static enum print_line_t
427 print_graph_proc(struct trace_seq *s, pid_t pid)
429 char comm[TASK_COMM_LEN];
430 /* sign + log10(MAX_INT) + '\0' */
431 char pid_str[11];
432 int spaces = 0;
433 int ret;
434 int len;
435 int i;
437 trace_find_cmdline(pid, comm);
438 comm[7] = '\0';
439 sprintf(pid_str, "%d", pid);
441 /* 1 stands for the "-" character */
442 len = strlen(comm) + strlen(pid_str) + 1;
444 if (len < TRACE_GRAPH_PROCINFO_LENGTH)
445 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
447 /* First spaces to align center */
448 for (i = 0; i < spaces / 2; i++) {
449 ret = trace_seq_printf(s, " ");
450 if (!ret)
451 return TRACE_TYPE_PARTIAL_LINE;
454 ret = trace_seq_printf(s, "%s-%s", comm, pid_str);
455 if (!ret)
456 return TRACE_TYPE_PARTIAL_LINE;
458 /* Last spaces to align center */
459 for (i = 0; i < spaces - (spaces / 2); i++) {
460 ret = trace_seq_printf(s, " ");
461 if (!ret)
462 return TRACE_TYPE_PARTIAL_LINE;
464 return TRACE_TYPE_HANDLED;
468 static enum print_line_t
469 print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
471 if (!trace_seq_putc(s, ' '))
472 return 0;
474 return trace_print_lat_fmt(s, entry);
477 /* If the pid changed since the last trace, output this event */
478 static enum print_line_t
479 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
481 pid_t prev_pid;
482 pid_t *last_pid;
483 int ret;
485 if (!data)
486 return TRACE_TYPE_HANDLED;
488 last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
490 if (*last_pid == pid)
491 return TRACE_TYPE_HANDLED;
493 prev_pid = *last_pid;
494 *last_pid = pid;
496 if (prev_pid == -1)
497 return TRACE_TYPE_HANDLED;
499 * Context-switch trace line:
501 ------------------------------------------
502 | 1) migration/0--1 => sshd-1755
503 ------------------------------------------
506 ret = trace_seq_printf(s,
507 " ------------------------------------------\n");
508 if (!ret)
509 return TRACE_TYPE_PARTIAL_LINE;
511 ret = print_graph_cpu(s, cpu);
512 if (ret == TRACE_TYPE_PARTIAL_LINE)
513 return TRACE_TYPE_PARTIAL_LINE;
515 ret = print_graph_proc(s, prev_pid);
516 if (ret == TRACE_TYPE_PARTIAL_LINE)
517 return TRACE_TYPE_PARTIAL_LINE;
519 ret = trace_seq_printf(s, " => ");
520 if (!ret)
521 return TRACE_TYPE_PARTIAL_LINE;
523 ret = print_graph_proc(s, pid);
524 if (ret == TRACE_TYPE_PARTIAL_LINE)
525 return TRACE_TYPE_PARTIAL_LINE;
527 ret = trace_seq_printf(s,
528 "\n ------------------------------------------\n\n");
529 if (!ret)
530 return TRACE_TYPE_PARTIAL_LINE;
532 return TRACE_TYPE_HANDLED;
535 static struct ftrace_graph_ret_entry *
536 get_return_for_leaf(struct trace_iterator *iter,
537 struct ftrace_graph_ent_entry *curr)
539 struct fgraph_data *data = iter->private;
540 struct ring_buffer_iter *ring_iter = NULL;
541 struct ring_buffer_event *event;
542 struct ftrace_graph_ret_entry *next;
545 * If the previous output failed to write to the seq buffer,
546 * then we just reuse the data from before.
548 if (data && data->failed) {
549 curr = &data->ent;
550 next = &data->ret;
551 } else {
553 ring_iter = trace_buffer_iter(iter, iter->cpu);
555 /* First peek to compare current entry and the next one */
556 if (ring_iter)
557 event = ring_buffer_iter_peek(ring_iter, NULL);
558 else {
560 * We need to consume the current entry to see
561 * the next one.
563 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu,
564 NULL, NULL);
565 event = ring_buffer_peek(iter->trace_buffer->buffer, iter->cpu,
566 NULL, NULL);
569 if (!event)
570 return NULL;
572 next = ring_buffer_event_data(event);
574 if (data) {
576 * Save current and next entries for later reference
577 * if the output fails.
579 data->ent = *curr;
581 * If the next event is not a return type, then
582 * we only care about what type it is. Otherwise we can
583 * safely copy the entire event.
585 if (next->ent.type == TRACE_GRAPH_RET)
586 data->ret = *next;
587 else
588 data->ret.ent.type = next->ent.type;
592 if (next->ent.type != TRACE_GRAPH_RET)
593 return NULL;
595 if (curr->ent.pid != next->ent.pid ||
596 curr->graph_ent.func != next->ret.func)
597 return NULL;
599 /* this is a leaf, now advance the iterator */
600 if (ring_iter)
601 ring_buffer_read(ring_iter, NULL);
603 return next;
606 static int print_graph_abs_time(u64 t, struct trace_seq *s)
608 unsigned long usecs_rem;
610 usecs_rem = do_div(t, NSEC_PER_SEC);
611 usecs_rem /= 1000;
613 return trace_seq_printf(s, "%5lu.%06lu | ",
614 (unsigned long)t, usecs_rem);
617 static enum print_line_t
618 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
619 enum trace_type type, int cpu, pid_t pid, u32 flags)
621 int ret;
622 struct trace_seq *s = &iter->seq;
624 if (addr < (unsigned long)__irqentry_text_start ||
625 addr >= (unsigned long)__irqentry_text_end)
626 return TRACE_TYPE_UNHANDLED;
628 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
629 /* Absolute time */
630 if (flags & TRACE_GRAPH_PRINT_ABS_TIME) {
631 ret = print_graph_abs_time(iter->ts, s);
632 if (!ret)
633 return TRACE_TYPE_PARTIAL_LINE;
636 /* Cpu */
637 if (flags & TRACE_GRAPH_PRINT_CPU) {
638 ret = print_graph_cpu(s, cpu);
639 if (ret == TRACE_TYPE_PARTIAL_LINE)
640 return TRACE_TYPE_PARTIAL_LINE;
643 /* Proc */
644 if (flags & TRACE_GRAPH_PRINT_PROC) {
645 ret = print_graph_proc(s, pid);
646 if (ret == TRACE_TYPE_PARTIAL_LINE)
647 return TRACE_TYPE_PARTIAL_LINE;
648 ret = trace_seq_printf(s, " | ");
649 if (!ret)
650 return TRACE_TYPE_PARTIAL_LINE;
654 /* No overhead */
655 ret = print_graph_duration(DURATION_FILL_START, s, flags);
656 if (ret != TRACE_TYPE_HANDLED)
657 return ret;
659 if (type == TRACE_GRAPH_ENT)
660 ret = trace_seq_printf(s, "==========>");
661 else
662 ret = trace_seq_printf(s, "<==========");
664 if (!ret)
665 return TRACE_TYPE_PARTIAL_LINE;
667 ret = print_graph_duration(DURATION_FILL_END, s, flags);
668 if (ret != TRACE_TYPE_HANDLED)
669 return ret;
671 ret = trace_seq_printf(s, "\n");
673 if (!ret)
674 return TRACE_TYPE_PARTIAL_LINE;
675 return TRACE_TYPE_HANDLED;
678 enum print_line_t
679 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
681 unsigned long nsecs_rem = do_div(duration, 1000);
682 /* log10(ULONG_MAX) + '\0' */
683 char msecs_str[21];
684 char nsecs_str[5];
685 int ret, len;
686 int i;
688 sprintf(msecs_str, "%lu", (unsigned long) duration);
690 /* Print msecs */
691 ret = trace_seq_printf(s, "%s", msecs_str);
692 if (!ret)
693 return TRACE_TYPE_PARTIAL_LINE;
695 len = strlen(msecs_str);
697 /* Print nsecs (we don't want to exceed 7 numbers) */
698 if (len < 7) {
699 size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
701 snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
702 ret = trace_seq_printf(s, ".%s", nsecs_str);
703 if (!ret)
704 return TRACE_TYPE_PARTIAL_LINE;
705 len += strlen(nsecs_str);
708 ret = trace_seq_printf(s, " us ");
709 if (!ret)
710 return TRACE_TYPE_PARTIAL_LINE;
712 /* Print remaining spaces to fit the row's width */
713 for (i = len; i < 7; i++) {
714 ret = trace_seq_printf(s, " ");
715 if (!ret)
716 return TRACE_TYPE_PARTIAL_LINE;
718 return TRACE_TYPE_HANDLED;
721 static enum print_line_t
722 print_graph_duration(unsigned long long duration, struct trace_seq *s,
723 u32 flags)
725 int ret = -1;
727 if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
728 !(trace_flags & TRACE_ITER_CONTEXT_INFO))
729 return TRACE_TYPE_HANDLED;
731 /* No real adata, just filling the column with spaces */
732 switch (duration) {
733 case DURATION_FILL_FULL:
734 ret = trace_seq_printf(s, " | ");
735 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
736 case DURATION_FILL_START:
737 ret = trace_seq_printf(s, " ");
738 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
739 case DURATION_FILL_END:
740 ret = trace_seq_printf(s, " |");
741 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
744 /* Signal a overhead of time execution to the output */
745 if (flags & TRACE_GRAPH_PRINT_OVERHEAD) {
746 /* Duration exceeded 100 msecs */
747 if (duration > 100000ULL)
748 ret = trace_seq_printf(s, "! ");
749 /* Duration exceeded 10 msecs */
750 else if (duration > 10000ULL)
751 ret = trace_seq_printf(s, "+ ");
755 * The -1 means we either did not exceed the duration tresholds
756 * or we dont want to print out the overhead. Either way we need
757 * to fill out the space.
759 if (ret == -1)
760 ret = trace_seq_printf(s, " ");
762 /* Catching here any failure happenned above */
763 if (!ret)
764 return TRACE_TYPE_PARTIAL_LINE;
766 ret = trace_print_graph_duration(duration, s);
767 if (ret != TRACE_TYPE_HANDLED)
768 return ret;
770 ret = trace_seq_printf(s, "| ");
771 if (!ret)
772 return TRACE_TYPE_PARTIAL_LINE;
774 return TRACE_TYPE_HANDLED;
777 /* Case of a leaf function on its call entry */
778 static enum print_line_t
779 print_graph_entry_leaf(struct trace_iterator *iter,
780 struct ftrace_graph_ent_entry *entry,
781 struct ftrace_graph_ret_entry *ret_entry,
782 struct trace_seq *s, u32 flags)
784 struct fgraph_data *data = iter->private;
785 struct ftrace_graph_ret *graph_ret;
786 struct ftrace_graph_ent *call;
787 unsigned long long duration;
788 int ret;
789 int i;
791 graph_ret = &ret_entry->ret;
792 call = &entry->graph_ent;
793 duration = graph_ret->rettime - graph_ret->calltime;
795 if (data) {
796 struct fgraph_cpu_data *cpu_data;
797 int cpu = iter->cpu;
799 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
802 * Comments display at + 1 to depth. Since
803 * this is a leaf function, keep the comments
804 * equal to this depth.
806 cpu_data->depth = call->depth - 1;
808 /* No need to keep this function around for this depth */
809 if (call->depth < FTRACE_RETFUNC_DEPTH)
810 cpu_data->enter_funcs[call->depth] = 0;
813 /* Overhead and duration */
814 ret = print_graph_duration(duration, s, flags);
815 if (ret == TRACE_TYPE_PARTIAL_LINE)
816 return TRACE_TYPE_PARTIAL_LINE;
818 /* Function */
819 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
820 ret = trace_seq_printf(s, " ");
821 if (!ret)
822 return TRACE_TYPE_PARTIAL_LINE;
825 ret = trace_seq_printf(s, "%ps();\n", (void *)call->func);
826 if (!ret)
827 return TRACE_TYPE_PARTIAL_LINE;
829 return TRACE_TYPE_HANDLED;
832 static enum print_line_t
833 print_graph_entry_nested(struct trace_iterator *iter,
834 struct ftrace_graph_ent_entry *entry,
835 struct trace_seq *s, int cpu, u32 flags)
837 struct ftrace_graph_ent *call = &entry->graph_ent;
838 struct fgraph_data *data = iter->private;
839 int ret;
840 int i;
842 if (data) {
843 struct fgraph_cpu_data *cpu_data;
844 int cpu = iter->cpu;
846 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
847 cpu_data->depth = call->depth;
849 /* Save this function pointer to see if the exit matches */
850 if (call->depth < FTRACE_RETFUNC_DEPTH)
851 cpu_data->enter_funcs[call->depth] = call->func;
854 /* No time */
855 ret = print_graph_duration(DURATION_FILL_FULL, s, flags);
856 if (ret != TRACE_TYPE_HANDLED)
857 return ret;
859 /* Function */
860 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
861 ret = trace_seq_printf(s, " ");
862 if (!ret)
863 return TRACE_TYPE_PARTIAL_LINE;
866 ret = trace_seq_printf(s, "%ps() {\n", (void *)call->func);
867 if (!ret)
868 return TRACE_TYPE_PARTIAL_LINE;
871 * we already consumed the current entry to check the next one
872 * and see if this is a leaf.
874 return TRACE_TYPE_NO_CONSUME;
877 static enum print_line_t
878 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
879 int type, unsigned long addr, u32 flags)
881 struct fgraph_data *data = iter->private;
882 struct trace_entry *ent = iter->ent;
883 int cpu = iter->cpu;
884 int ret;
886 /* Pid */
887 if (verif_pid(s, ent->pid, cpu, data) == TRACE_TYPE_PARTIAL_LINE)
888 return TRACE_TYPE_PARTIAL_LINE;
890 if (type) {
891 /* Interrupt */
892 ret = print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
893 if (ret == TRACE_TYPE_PARTIAL_LINE)
894 return TRACE_TYPE_PARTIAL_LINE;
897 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
898 return 0;
900 /* Absolute time */
901 if (flags & TRACE_GRAPH_PRINT_ABS_TIME) {
902 ret = print_graph_abs_time(iter->ts, s);
903 if (!ret)
904 return TRACE_TYPE_PARTIAL_LINE;
907 /* Cpu */
908 if (flags & TRACE_GRAPH_PRINT_CPU) {
909 ret = print_graph_cpu(s, cpu);
910 if (ret == TRACE_TYPE_PARTIAL_LINE)
911 return TRACE_TYPE_PARTIAL_LINE;
914 /* Proc */
915 if (flags & TRACE_GRAPH_PRINT_PROC) {
916 ret = print_graph_proc(s, ent->pid);
917 if (ret == TRACE_TYPE_PARTIAL_LINE)
918 return TRACE_TYPE_PARTIAL_LINE;
920 ret = trace_seq_printf(s, " | ");
921 if (!ret)
922 return TRACE_TYPE_PARTIAL_LINE;
925 /* Latency format */
926 if (trace_flags & TRACE_ITER_LATENCY_FMT) {
927 ret = print_graph_lat_fmt(s, ent);
928 if (ret == TRACE_TYPE_PARTIAL_LINE)
929 return TRACE_TYPE_PARTIAL_LINE;
932 return 0;
936 * Entry check for irq code
938 * returns 1 if
939 * - we are inside irq code
940 * - we just entered irq code
942 * retunns 0 if
943 * - funcgraph-interrupts option is set
944 * - we are not inside irq code
946 static int
947 check_irq_entry(struct trace_iterator *iter, u32 flags,
948 unsigned long addr, int depth)
950 int cpu = iter->cpu;
951 int *depth_irq;
952 struct fgraph_data *data = iter->private;
955 * If we are either displaying irqs, or we got called as
956 * a graph event and private data does not exist,
957 * then we bypass the irq check.
959 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
960 (!data))
961 return 0;
963 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
966 * We are inside the irq code
968 if (*depth_irq >= 0)
969 return 1;
971 if ((addr < (unsigned long)__irqentry_text_start) ||
972 (addr >= (unsigned long)__irqentry_text_end))
973 return 0;
976 * We are entering irq code.
978 *depth_irq = depth;
979 return 1;
983 * Return check for irq code
985 * returns 1 if
986 * - we are inside irq code
987 * - we just left irq code
989 * returns 0 if
990 * - funcgraph-interrupts option is set
991 * - we are not inside irq code
993 static int
994 check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
996 int cpu = iter->cpu;
997 int *depth_irq;
998 struct fgraph_data *data = iter->private;
1001 * If we are either displaying irqs, or we got called as
1002 * a graph event and private data does not exist,
1003 * then we bypass the irq check.
1005 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
1006 (!data))
1007 return 0;
1009 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1012 * We are not inside the irq code.
1014 if (*depth_irq == -1)
1015 return 0;
1018 * We are inside the irq code, and this is returning entry.
1019 * Let's not trace it and clear the entry depth, since
1020 * we are out of irq code.
1022 * This condition ensures that we 'leave the irq code' once
1023 * we are out of the entry depth. Thus protecting us from
1024 * the RETURN entry loss.
1026 if (*depth_irq >= depth) {
1027 *depth_irq = -1;
1028 return 1;
1032 * We are inside the irq code, and this is not the entry.
1034 return 1;
1037 static enum print_line_t
1038 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
1039 struct trace_iterator *iter, u32 flags)
1041 struct fgraph_data *data = iter->private;
1042 struct ftrace_graph_ent *call = &field->graph_ent;
1043 struct ftrace_graph_ret_entry *leaf_ret;
1044 static enum print_line_t ret;
1045 int cpu = iter->cpu;
1047 if (check_irq_entry(iter, flags, call->func, call->depth))
1048 return TRACE_TYPE_HANDLED;
1050 if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags))
1051 return TRACE_TYPE_PARTIAL_LINE;
1053 leaf_ret = get_return_for_leaf(iter, field);
1054 if (leaf_ret)
1055 ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
1056 else
1057 ret = print_graph_entry_nested(iter, field, s, cpu, flags);
1059 if (data) {
1061 * If we failed to write our output, then we need to make
1062 * note of it. Because we already consumed our entry.
1064 if (s->full) {
1065 data->failed = 1;
1066 data->cpu = cpu;
1067 } else
1068 data->failed = 0;
1071 return ret;
1074 static enum print_line_t
1075 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
1076 struct trace_entry *ent, struct trace_iterator *iter,
1077 u32 flags)
1079 unsigned long long duration = trace->rettime - trace->calltime;
1080 struct fgraph_data *data = iter->private;
1081 pid_t pid = ent->pid;
1082 int cpu = iter->cpu;
1083 int func_match = 1;
1084 int ret;
1085 int i;
1087 if (check_irq_return(iter, flags, trace->depth))
1088 return TRACE_TYPE_HANDLED;
1090 if (data) {
1091 struct fgraph_cpu_data *cpu_data;
1092 int cpu = iter->cpu;
1094 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
1097 * Comments display at + 1 to depth. This is the
1098 * return from a function, we now want the comments
1099 * to display at the same level of the bracket.
1101 cpu_data->depth = trace->depth - 1;
1103 if (trace->depth < FTRACE_RETFUNC_DEPTH) {
1104 if (cpu_data->enter_funcs[trace->depth] != trace->func)
1105 func_match = 0;
1106 cpu_data->enter_funcs[trace->depth] = 0;
1110 if (print_graph_prologue(iter, s, 0, 0, flags))
1111 return TRACE_TYPE_PARTIAL_LINE;
1113 /* Overhead and duration */
1114 ret = print_graph_duration(duration, s, flags);
1115 if (ret == TRACE_TYPE_PARTIAL_LINE)
1116 return TRACE_TYPE_PARTIAL_LINE;
1118 /* Closing brace */
1119 for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
1120 ret = trace_seq_printf(s, " ");
1121 if (!ret)
1122 return TRACE_TYPE_PARTIAL_LINE;
1126 * If the return function does not have a matching entry,
1127 * then the entry was lost. Instead of just printing
1128 * the '}' and letting the user guess what function this
1129 * belongs to, write out the function name.
1131 if (func_match) {
1132 ret = trace_seq_printf(s, "}\n");
1133 if (!ret)
1134 return TRACE_TYPE_PARTIAL_LINE;
1135 } else {
1136 ret = trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
1137 if (!ret)
1138 return TRACE_TYPE_PARTIAL_LINE;
1141 /* Overrun */
1142 if (flags & TRACE_GRAPH_PRINT_OVERRUN) {
1143 ret = trace_seq_printf(s, " (Overruns: %lu)\n",
1144 trace->overrun);
1145 if (!ret)
1146 return TRACE_TYPE_PARTIAL_LINE;
1149 ret = print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
1150 cpu, pid, flags);
1151 if (ret == TRACE_TYPE_PARTIAL_LINE)
1152 return TRACE_TYPE_PARTIAL_LINE;
1154 return TRACE_TYPE_HANDLED;
1157 static enum print_line_t
1158 print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1159 struct trace_iterator *iter, u32 flags)
1161 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1162 struct fgraph_data *data = iter->private;
1163 struct trace_event *event;
1164 int depth = 0;
1165 int ret;
1166 int i;
1168 if (data)
1169 depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1171 if (print_graph_prologue(iter, s, 0, 0, flags))
1172 return TRACE_TYPE_PARTIAL_LINE;
1174 /* No time */
1175 ret = print_graph_duration(DURATION_FILL_FULL, s, flags);
1176 if (ret != TRACE_TYPE_HANDLED)
1177 return ret;
1179 /* Indentation */
1180 if (depth > 0)
1181 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++) {
1182 ret = trace_seq_printf(s, " ");
1183 if (!ret)
1184 return TRACE_TYPE_PARTIAL_LINE;
1187 /* The comment */
1188 ret = trace_seq_printf(s, "/* ");
1189 if (!ret)
1190 return TRACE_TYPE_PARTIAL_LINE;
1192 switch (iter->ent->type) {
1193 case TRACE_BPRINT:
1194 ret = trace_print_bprintk_msg_only(iter);
1195 if (ret != TRACE_TYPE_HANDLED)
1196 return ret;
1197 break;
1198 case TRACE_PRINT:
1199 ret = trace_print_printk_msg_only(iter);
1200 if (ret != TRACE_TYPE_HANDLED)
1201 return ret;
1202 break;
1203 default:
1204 event = ftrace_find_event(ent->type);
1205 if (!event)
1206 return TRACE_TYPE_UNHANDLED;
1208 ret = event->funcs->trace(iter, sym_flags, event);
1209 if (ret != TRACE_TYPE_HANDLED)
1210 return ret;
1213 /* Strip ending newline */
1214 if (s->buffer[s->len - 1] == '\n') {
1215 s->buffer[s->len - 1] = '\0';
1216 s->len--;
1219 ret = trace_seq_printf(s, " */\n");
1220 if (!ret)
1221 return TRACE_TYPE_PARTIAL_LINE;
1223 return TRACE_TYPE_HANDLED;
1227 enum print_line_t
1228 print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1230 struct ftrace_graph_ent_entry *field;
1231 struct fgraph_data *data = iter->private;
1232 struct trace_entry *entry = iter->ent;
1233 struct trace_seq *s = &iter->seq;
1234 int cpu = iter->cpu;
1235 int ret;
1237 if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1238 per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1239 return TRACE_TYPE_HANDLED;
1243 * If the last output failed, there's a possibility we need
1244 * to print out the missing entry which would never go out.
1246 if (data && data->failed) {
1247 field = &data->ent;
1248 iter->cpu = data->cpu;
1249 ret = print_graph_entry(field, s, iter, flags);
1250 if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1251 per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1252 ret = TRACE_TYPE_NO_CONSUME;
1254 iter->cpu = cpu;
1255 return ret;
1258 switch (entry->type) {
1259 case TRACE_GRAPH_ENT: {
1261 * print_graph_entry() may consume the current event,
1262 * thus @field may become invalid, so we need to save it.
1263 * sizeof(struct ftrace_graph_ent_entry) is very small,
1264 * it can be safely saved at the stack.
1266 struct ftrace_graph_ent_entry saved;
1267 trace_assign_type(field, entry);
1268 saved = *field;
1269 return print_graph_entry(&saved, s, iter, flags);
1271 case TRACE_GRAPH_RET: {
1272 struct ftrace_graph_ret_entry *field;
1273 trace_assign_type(field, entry);
1274 return print_graph_return(&field->ret, s, entry, iter, flags);
1276 case TRACE_STACK:
1277 case TRACE_FN:
1278 /* dont trace stack and functions as comments */
1279 return TRACE_TYPE_UNHANDLED;
1281 default:
1282 return print_graph_comment(s, entry, iter, flags);
1285 return TRACE_TYPE_HANDLED;
1288 static enum print_line_t
1289 print_graph_function(struct trace_iterator *iter)
1291 return print_graph_function_flags(iter, tracer_flags.val);
1294 static enum print_line_t
1295 print_graph_function_event(struct trace_iterator *iter, int flags,
1296 struct trace_event *event)
1298 return print_graph_function(iter);
1301 static void print_lat_header(struct seq_file *s, u32 flags)
1303 static const char spaces[] = " " /* 16 spaces */
1304 " " /* 4 spaces */
1305 " "; /* 17 spaces */
1306 int size = 0;
1308 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1309 size += 16;
1310 if (flags & TRACE_GRAPH_PRINT_CPU)
1311 size += 4;
1312 if (flags & TRACE_GRAPH_PRINT_PROC)
1313 size += 17;
1315 seq_printf(s, "#%.*s _-----=> irqs-off \n", size, spaces);
1316 seq_printf(s, "#%.*s / _----=> need-resched \n", size, spaces);
1317 seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1318 seq_printf(s, "#%.*s|| / _--=> preempt-depth \n", size, spaces);
1319 seq_printf(s, "#%.*s||| / \n", size, spaces);
1322 static void __print_graph_headers_flags(struct seq_file *s, u32 flags)
1324 int lat = trace_flags & TRACE_ITER_LATENCY_FMT;
1326 if (lat)
1327 print_lat_header(s, flags);
1329 /* 1st line */
1330 seq_printf(s, "#");
1331 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1332 seq_printf(s, " TIME ");
1333 if (flags & TRACE_GRAPH_PRINT_CPU)
1334 seq_printf(s, " CPU");
1335 if (flags & TRACE_GRAPH_PRINT_PROC)
1336 seq_printf(s, " TASK/PID ");
1337 if (lat)
1338 seq_printf(s, "||||");
1339 if (flags & TRACE_GRAPH_PRINT_DURATION)
1340 seq_printf(s, " DURATION ");
1341 seq_printf(s, " FUNCTION CALLS\n");
1343 /* 2nd line */
1344 seq_printf(s, "#");
1345 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1346 seq_printf(s, " | ");
1347 if (flags & TRACE_GRAPH_PRINT_CPU)
1348 seq_printf(s, " | ");
1349 if (flags & TRACE_GRAPH_PRINT_PROC)
1350 seq_printf(s, " | | ");
1351 if (lat)
1352 seq_printf(s, "||||");
1353 if (flags & TRACE_GRAPH_PRINT_DURATION)
1354 seq_printf(s, " | | ");
1355 seq_printf(s, " | | | |\n");
1358 void print_graph_headers(struct seq_file *s)
1360 print_graph_headers_flags(s, tracer_flags.val);
1363 void print_graph_headers_flags(struct seq_file *s, u32 flags)
1365 struct trace_iterator *iter = s->private;
1367 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
1368 return;
1370 if (trace_flags & TRACE_ITER_LATENCY_FMT) {
1371 /* print nothing if the buffers are empty */
1372 if (trace_empty(iter))
1373 return;
1375 print_trace_header(s, iter);
1378 __print_graph_headers_flags(s, flags);
1381 void graph_trace_open(struct trace_iterator *iter)
1383 /* pid and depth on the last trace processed */
1384 struct fgraph_data *data;
1385 int cpu;
1387 iter->private = NULL;
1389 data = kzalloc(sizeof(*data), GFP_KERNEL);
1390 if (!data)
1391 goto out_err;
1393 data->cpu_data = alloc_percpu(struct fgraph_cpu_data);
1394 if (!data->cpu_data)
1395 goto out_err_free;
1397 for_each_possible_cpu(cpu) {
1398 pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1399 int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1400 int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1401 int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1403 *pid = -1;
1404 *depth = 0;
1405 *ignore = 0;
1406 *depth_irq = -1;
1409 iter->private = data;
1411 return;
1413 out_err_free:
1414 kfree(data);
1415 out_err:
1416 pr_warning("function graph tracer: not enough memory\n");
1419 void graph_trace_close(struct trace_iterator *iter)
1421 struct fgraph_data *data = iter->private;
1423 if (data) {
1424 free_percpu(data->cpu_data);
1425 kfree(data);
1429 static int func_graph_set_flag(u32 old_flags, u32 bit, int set)
1431 if (bit == TRACE_GRAPH_PRINT_IRQS)
1432 ftrace_graph_skip_irqs = !set;
1434 return 0;
1437 static struct trace_event_functions graph_functions = {
1438 .trace = print_graph_function_event,
1441 static struct trace_event graph_trace_entry_event = {
1442 .type = TRACE_GRAPH_ENT,
1443 .funcs = &graph_functions,
1446 static struct trace_event graph_trace_ret_event = {
1447 .type = TRACE_GRAPH_RET,
1448 .funcs = &graph_functions
1451 static struct tracer graph_trace __read_mostly = {
1452 .name = "function_graph",
1453 .open = graph_trace_open,
1454 .pipe_open = graph_trace_open,
1455 .close = graph_trace_close,
1456 .pipe_close = graph_trace_close,
1457 .wait_pipe = poll_wait_pipe,
1458 .init = graph_trace_init,
1459 .reset = graph_trace_reset,
1460 .print_line = print_graph_function,
1461 .print_header = print_graph_headers,
1462 .flags = &tracer_flags,
1463 .set_flag = func_graph_set_flag,
1464 #ifdef CONFIG_FTRACE_SELFTEST
1465 .selftest = trace_selftest_startup_function_graph,
1466 #endif
1470 static ssize_t
1471 graph_depth_write(struct file *filp, const char __user *ubuf, size_t cnt,
1472 loff_t *ppos)
1474 unsigned long val;
1475 int ret;
1477 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1478 if (ret)
1479 return ret;
1481 max_depth = val;
1483 *ppos += cnt;
1485 return cnt;
1488 static ssize_t
1489 graph_depth_read(struct file *filp, char __user *ubuf, size_t cnt,
1490 loff_t *ppos)
1492 char buf[15]; /* More than enough to hold UINT_MAX + "\n"*/
1493 int n;
1495 n = sprintf(buf, "%d\n", max_depth);
1497 return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
1500 static const struct file_operations graph_depth_fops = {
1501 .open = tracing_open_generic,
1502 .write = graph_depth_write,
1503 .read = graph_depth_read,
1504 .llseek = generic_file_llseek,
1507 static __init int init_graph_debugfs(void)
1509 struct dentry *d_tracer;
1511 d_tracer = tracing_init_dentry();
1512 if (!d_tracer)
1513 return 0;
1515 trace_create_file("max_graph_depth", 0644, d_tracer,
1516 NULL, &graph_depth_fops);
1518 return 0;
1520 fs_initcall(init_graph_debugfs);
1522 static __init int init_graph_trace(void)
1524 max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1526 if (!register_ftrace_event(&graph_trace_entry_event)) {
1527 pr_warning("Warning: could not register graph trace events\n");
1528 return 1;
1531 if (!register_ftrace_event(&graph_trace_ret_event)) {
1532 pr_warning("Warning: could not register graph trace events\n");
1533 return 1;
1536 return register_tracer(&graph_trace);
1539 core_initcall(init_graph_trace);