PCI: PCIe: Move PCIe PME code to the pcie directory
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / trace / trace_functions_graph.c
blob6bff2362578115f4a087195362a5257c3b838d42
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 struct fgraph_cpu_data {
19 pid_t last_pid;
20 int depth;
21 int ignore;
22 unsigned long enter_funcs[FTRACE_RETFUNC_DEPTH];
25 struct fgraph_data {
26 struct fgraph_cpu_data *cpu_data;
28 /* Place to preserve last processed entry. */
29 struct ftrace_graph_ent_entry ent;
30 struct ftrace_graph_ret_entry ret;
31 int failed;
32 int cpu;
35 #define TRACE_GRAPH_INDENT 2
37 /* Flag options */
38 #define TRACE_GRAPH_PRINT_OVERRUN 0x1
39 #define TRACE_GRAPH_PRINT_CPU 0x2
40 #define TRACE_GRAPH_PRINT_OVERHEAD 0x4
41 #define TRACE_GRAPH_PRINT_PROC 0x8
42 #define TRACE_GRAPH_PRINT_DURATION 0x10
43 #define TRACE_GRAPH_PRINT_ABS_TIME 0x20
45 static struct tracer_opt trace_opts[] = {
46 /* Display overruns? (for self-debug purpose) */
47 { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
48 /* Display CPU ? */
49 { TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
50 /* Display Overhead ? */
51 { TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
52 /* Display proc name/pid */
53 { TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
54 /* Display duration of execution */
55 { TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
56 /* Display absolute time of an entry */
57 { TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
58 { } /* Empty entry */
61 static struct tracer_flags tracer_flags = {
62 /* Don't display overruns and proc by default */
63 .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
64 TRACE_GRAPH_PRINT_DURATION,
65 .opts = trace_opts
68 static struct trace_array *graph_array;
71 /* Add a function return address to the trace stack on thread info.*/
72 int
73 ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
74 unsigned long frame_pointer)
76 unsigned long long calltime;
77 int index;
79 if (!current->ret_stack)
80 return -EBUSY;
83 * We must make sure the ret_stack is tested before we read
84 * anything else.
86 smp_rmb();
88 /* The return trace stack is full */
89 if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
90 atomic_inc(&current->trace_overrun);
91 return -EBUSY;
94 calltime = trace_clock_local();
96 index = ++current->curr_ret_stack;
97 barrier();
98 current->ret_stack[index].ret = ret;
99 current->ret_stack[index].func = func;
100 current->ret_stack[index].calltime = calltime;
101 current->ret_stack[index].subtime = 0;
102 current->ret_stack[index].fp = frame_pointer;
103 *depth = index;
105 return 0;
108 /* Retrieve a function return address to the trace stack on thread info.*/
109 static void
110 ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
111 unsigned long frame_pointer)
113 int index;
115 index = current->curr_ret_stack;
117 if (unlikely(index < 0)) {
118 ftrace_graph_stop();
119 WARN_ON(1);
120 /* Might as well panic, otherwise we have no where to go */
121 *ret = (unsigned long)panic;
122 return;
125 #ifdef CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST
127 * The arch may choose to record the frame pointer used
128 * and check it here to make sure that it is what we expect it
129 * to be. If gcc does not set the place holder of the return
130 * address in the frame pointer, and does a copy instead, then
131 * the function graph trace will fail. This test detects this
132 * case.
134 * Currently, x86_32 with optimize for size (-Os) makes the latest
135 * gcc do the above.
137 if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
138 ftrace_graph_stop();
139 WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
140 " from func %ps return to %lx\n",
141 current->ret_stack[index].fp,
142 frame_pointer,
143 (void *)current->ret_stack[index].func,
144 current->ret_stack[index].ret);
145 *ret = (unsigned long)panic;
146 return;
148 #endif
150 *ret = current->ret_stack[index].ret;
151 trace->func = current->ret_stack[index].func;
152 trace->calltime = current->ret_stack[index].calltime;
153 trace->overrun = atomic_read(&current->trace_overrun);
154 trace->depth = index;
158 * Send the trace to the ring-buffer.
159 * @return the original return address.
161 unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
163 struct ftrace_graph_ret trace;
164 unsigned long ret;
166 ftrace_pop_return_trace(&trace, &ret, frame_pointer);
167 trace.rettime = trace_clock_local();
168 ftrace_graph_return(&trace);
169 barrier();
170 current->curr_ret_stack--;
172 if (unlikely(!ret)) {
173 ftrace_graph_stop();
174 WARN_ON(1);
175 /* Might as well panic. What else to do? */
176 ret = (unsigned long)panic;
179 return ret;
182 int __trace_graph_entry(struct trace_array *tr,
183 struct ftrace_graph_ent *trace,
184 unsigned long flags,
185 int pc)
187 struct ftrace_event_call *call = &event_funcgraph_entry;
188 struct ring_buffer_event *event;
189 struct ring_buffer *buffer = tr->buffer;
190 struct ftrace_graph_ent_entry *entry;
192 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
193 return 0;
195 event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
196 sizeof(*entry), flags, pc);
197 if (!event)
198 return 0;
199 entry = ring_buffer_event_data(event);
200 entry->graph_ent = *trace;
201 if (!filter_current_check_discard(buffer, call, entry, event))
202 ring_buffer_unlock_commit(buffer, event);
204 return 1;
207 int trace_graph_entry(struct ftrace_graph_ent *trace)
209 struct trace_array *tr = graph_array;
210 struct trace_array_cpu *data;
211 unsigned long flags;
212 long disabled;
213 int ret;
214 int cpu;
215 int pc;
217 if (!ftrace_trace_task(current))
218 return 0;
220 /* trace it when it is-nested-in or is a function enabled. */
221 if (!(trace->depth || ftrace_graph_addr(trace->func)))
222 return 0;
224 local_irq_save(flags);
225 cpu = raw_smp_processor_id();
226 data = tr->data[cpu];
227 disabled = atomic_inc_return(&data->disabled);
228 if (likely(disabled == 1)) {
229 pc = preempt_count();
230 ret = __trace_graph_entry(tr, trace, flags, pc);
231 } else {
232 ret = 0;
235 atomic_dec(&data->disabled);
236 local_irq_restore(flags);
238 return ret;
241 int trace_graph_thresh_entry(struct ftrace_graph_ent *trace)
243 if (tracing_thresh)
244 return 1;
245 else
246 return trace_graph_entry(trace);
249 void __trace_graph_return(struct trace_array *tr,
250 struct ftrace_graph_ret *trace,
251 unsigned long flags,
252 int pc)
254 struct ftrace_event_call *call = &event_funcgraph_exit;
255 struct ring_buffer_event *event;
256 struct ring_buffer *buffer = tr->buffer;
257 struct ftrace_graph_ret_entry *entry;
259 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
260 return;
262 event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
263 sizeof(*entry), flags, pc);
264 if (!event)
265 return;
266 entry = ring_buffer_event_data(event);
267 entry->ret = *trace;
268 if (!filter_current_check_discard(buffer, call, entry, event))
269 ring_buffer_unlock_commit(buffer, event);
272 void trace_graph_return(struct ftrace_graph_ret *trace)
274 struct trace_array *tr = graph_array;
275 struct trace_array_cpu *data;
276 unsigned long flags;
277 long disabled;
278 int cpu;
279 int pc;
281 local_irq_save(flags);
282 cpu = raw_smp_processor_id();
283 data = tr->data[cpu];
284 disabled = atomic_inc_return(&data->disabled);
285 if (likely(disabled == 1)) {
286 pc = preempt_count();
287 __trace_graph_return(tr, trace, flags, pc);
289 atomic_dec(&data->disabled);
290 local_irq_restore(flags);
293 void set_graph_array(struct trace_array *tr)
295 graph_array = tr;
297 /* Make graph_array visible before we start tracing */
299 smp_mb();
302 void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
304 if (tracing_thresh &&
305 (trace->rettime - trace->calltime < tracing_thresh))
306 return;
307 else
308 trace_graph_return(trace);
311 static int graph_trace_init(struct trace_array *tr)
313 int ret;
315 set_graph_array(tr);
316 if (tracing_thresh)
317 ret = register_ftrace_graph(&trace_graph_thresh_return,
318 &trace_graph_thresh_entry);
319 else
320 ret = register_ftrace_graph(&trace_graph_return,
321 &trace_graph_entry);
322 if (ret)
323 return ret;
324 tracing_start_cmdline_record();
326 return 0;
329 static void graph_trace_reset(struct trace_array *tr)
331 tracing_stop_cmdline_record();
332 unregister_ftrace_graph();
335 static int max_bytes_for_cpu;
337 static enum print_line_t
338 print_graph_cpu(struct trace_seq *s, int cpu)
340 int ret;
343 * Start with a space character - to make it stand out
344 * to the right a bit when trace output is pasted into
345 * email:
347 ret = trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
348 if (!ret)
349 return TRACE_TYPE_PARTIAL_LINE;
351 return TRACE_TYPE_HANDLED;
354 #define TRACE_GRAPH_PROCINFO_LENGTH 14
356 static enum print_line_t
357 print_graph_proc(struct trace_seq *s, pid_t pid)
359 char comm[TASK_COMM_LEN];
360 /* sign + log10(MAX_INT) + '\0' */
361 char pid_str[11];
362 int spaces = 0;
363 int ret;
364 int len;
365 int i;
367 trace_find_cmdline(pid, comm);
368 comm[7] = '\0';
369 sprintf(pid_str, "%d", pid);
371 /* 1 stands for the "-" character */
372 len = strlen(comm) + strlen(pid_str) + 1;
374 if (len < TRACE_GRAPH_PROCINFO_LENGTH)
375 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
377 /* First spaces to align center */
378 for (i = 0; i < spaces / 2; i++) {
379 ret = trace_seq_printf(s, " ");
380 if (!ret)
381 return TRACE_TYPE_PARTIAL_LINE;
384 ret = trace_seq_printf(s, "%s-%s", comm, pid_str);
385 if (!ret)
386 return TRACE_TYPE_PARTIAL_LINE;
388 /* Last spaces to align center */
389 for (i = 0; i < spaces - (spaces / 2); i++) {
390 ret = trace_seq_printf(s, " ");
391 if (!ret)
392 return TRACE_TYPE_PARTIAL_LINE;
394 return TRACE_TYPE_HANDLED;
398 static enum print_line_t
399 print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
401 if (!trace_seq_putc(s, ' '))
402 return 0;
404 return trace_print_lat_fmt(s, entry);
407 /* If the pid changed since the last trace, output this event */
408 static enum print_line_t
409 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
411 pid_t prev_pid;
412 pid_t *last_pid;
413 int ret;
415 if (!data)
416 return TRACE_TYPE_HANDLED;
418 last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
420 if (*last_pid == pid)
421 return TRACE_TYPE_HANDLED;
423 prev_pid = *last_pid;
424 *last_pid = pid;
426 if (prev_pid == -1)
427 return TRACE_TYPE_HANDLED;
429 * Context-switch trace line:
431 ------------------------------------------
432 | 1) migration/0--1 => sshd-1755
433 ------------------------------------------
436 ret = trace_seq_printf(s,
437 " ------------------------------------------\n");
438 if (!ret)
439 return TRACE_TYPE_PARTIAL_LINE;
441 ret = print_graph_cpu(s, cpu);
442 if (ret == TRACE_TYPE_PARTIAL_LINE)
443 return TRACE_TYPE_PARTIAL_LINE;
445 ret = print_graph_proc(s, prev_pid);
446 if (ret == TRACE_TYPE_PARTIAL_LINE)
447 return TRACE_TYPE_PARTIAL_LINE;
449 ret = trace_seq_printf(s, " => ");
450 if (!ret)
451 return TRACE_TYPE_PARTIAL_LINE;
453 ret = print_graph_proc(s, pid);
454 if (ret == TRACE_TYPE_PARTIAL_LINE)
455 return TRACE_TYPE_PARTIAL_LINE;
457 ret = trace_seq_printf(s,
458 "\n ------------------------------------------\n\n");
459 if (!ret)
460 return TRACE_TYPE_PARTIAL_LINE;
462 return TRACE_TYPE_HANDLED;
465 static struct ftrace_graph_ret_entry *
466 get_return_for_leaf(struct trace_iterator *iter,
467 struct ftrace_graph_ent_entry *curr)
469 struct fgraph_data *data = iter->private;
470 struct ring_buffer_iter *ring_iter = NULL;
471 struct ring_buffer_event *event;
472 struct ftrace_graph_ret_entry *next;
475 * If the previous output failed to write to the seq buffer,
476 * then we just reuse the data from before.
478 if (data && data->failed) {
479 curr = &data->ent;
480 next = &data->ret;
481 } else {
483 ring_iter = iter->buffer_iter[iter->cpu];
485 /* First peek to compare current entry and the next one */
486 if (ring_iter)
487 event = ring_buffer_iter_peek(ring_iter, NULL);
488 else {
490 * We need to consume the current entry to see
491 * the next one.
493 ring_buffer_consume(iter->tr->buffer, iter->cpu,
494 NULL, NULL);
495 event = ring_buffer_peek(iter->tr->buffer, iter->cpu,
496 NULL, NULL);
499 if (!event)
500 return NULL;
502 next = ring_buffer_event_data(event);
504 if (data) {
506 * Save current and next entries for later reference
507 * if the output fails.
509 data->ent = *curr;
510 data->ret = *next;
514 if (next->ent.type != TRACE_GRAPH_RET)
515 return NULL;
517 if (curr->ent.pid != next->ent.pid ||
518 curr->graph_ent.func != next->ret.func)
519 return NULL;
521 /* this is a leaf, now advance the iterator */
522 if (ring_iter)
523 ring_buffer_read(ring_iter, NULL);
525 return next;
528 /* Signal a overhead of time execution to the output */
529 static int
530 print_graph_overhead(unsigned long long duration, struct trace_seq *s,
531 u32 flags)
533 /* If duration disappear, we don't need anything */
534 if (!(flags & TRACE_GRAPH_PRINT_DURATION))
535 return 1;
537 /* Non nested entry or return */
538 if (duration == -1)
539 return trace_seq_printf(s, " ");
541 if (flags & TRACE_GRAPH_PRINT_OVERHEAD) {
542 /* Duration exceeded 100 msecs */
543 if (duration > 100000ULL)
544 return trace_seq_printf(s, "! ");
546 /* Duration exceeded 10 msecs */
547 if (duration > 10000ULL)
548 return trace_seq_printf(s, "+ ");
551 return trace_seq_printf(s, " ");
554 static int print_graph_abs_time(u64 t, struct trace_seq *s)
556 unsigned long usecs_rem;
558 usecs_rem = do_div(t, NSEC_PER_SEC);
559 usecs_rem /= 1000;
561 return trace_seq_printf(s, "%5lu.%06lu | ",
562 (unsigned long)t, usecs_rem);
565 static enum print_line_t
566 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
567 enum trace_type type, int cpu, pid_t pid, u32 flags)
569 int ret;
570 struct trace_seq *s = &iter->seq;
572 if (addr < (unsigned long)__irqentry_text_start ||
573 addr >= (unsigned long)__irqentry_text_end)
574 return TRACE_TYPE_UNHANDLED;
576 /* Absolute time */
577 if (flags & TRACE_GRAPH_PRINT_ABS_TIME) {
578 ret = print_graph_abs_time(iter->ts, s);
579 if (!ret)
580 return TRACE_TYPE_PARTIAL_LINE;
583 /* Cpu */
584 if (flags & TRACE_GRAPH_PRINT_CPU) {
585 ret = print_graph_cpu(s, cpu);
586 if (ret == TRACE_TYPE_PARTIAL_LINE)
587 return TRACE_TYPE_PARTIAL_LINE;
590 /* Proc */
591 if (flags & TRACE_GRAPH_PRINT_PROC) {
592 ret = print_graph_proc(s, pid);
593 if (ret == TRACE_TYPE_PARTIAL_LINE)
594 return TRACE_TYPE_PARTIAL_LINE;
595 ret = trace_seq_printf(s, " | ");
596 if (!ret)
597 return TRACE_TYPE_PARTIAL_LINE;
600 /* No overhead */
601 ret = print_graph_overhead(-1, s, flags);
602 if (!ret)
603 return TRACE_TYPE_PARTIAL_LINE;
605 if (type == TRACE_GRAPH_ENT)
606 ret = trace_seq_printf(s, "==========>");
607 else
608 ret = trace_seq_printf(s, "<==========");
610 if (!ret)
611 return TRACE_TYPE_PARTIAL_LINE;
613 /* Don't close the duration column if haven't one */
614 if (flags & TRACE_GRAPH_PRINT_DURATION)
615 trace_seq_printf(s, " |");
616 ret = trace_seq_printf(s, "\n");
618 if (!ret)
619 return TRACE_TYPE_PARTIAL_LINE;
620 return TRACE_TYPE_HANDLED;
623 enum print_line_t
624 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
626 unsigned long nsecs_rem = do_div(duration, 1000);
627 /* log10(ULONG_MAX) + '\0' */
628 char msecs_str[21];
629 char nsecs_str[5];
630 int ret, len;
631 int i;
633 sprintf(msecs_str, "%lu", (unsigned long) duration);
635 /* Print msecs */
636 ret = trace_seq_printf(s, "%s", msecs_str);
637 if (!ret)
638 return TRACE_TYPE_PARTIAL_LINE;
640 len = strlen(msecs_str);
642 /* Print nsecs (we don't want to exceed 7 numbers) */
643 if (len < 7) {
644 snprintf(nsecs_str, min(sizeof(nsecs_str), 8UL - len), "%03lu",
645 nsecs_rem);
646 ret = trace_seq_printf(s, ".%s", nsecs_str);
647 if (!ret)
648 return TRACE_TYPE_PARTIAL_LINE;
649 len += strlen(nsecs_str);
652 ret = trace_seq_printf(s, " us ");
653 if (!ret)
654 return TRACE_TYPE_PARTIAL_LINE;
656 /* Print remaining spaces to fit the row's width */
657 for (i = len; i < 7; i++) {
658 ret = trace_seq_printf(s, " ");
659 if (!ret)
660 return TRACE_TYPE_PARTIAL_LINE;
662 return TRACE_TYPE_HANDLED;
665 static enum print_line_t
666 print_graph_duration(unsigned long long duration, struct trace_seq *s)
668 int ret;
670 ret = trace_print_graph_duration(duration, s);
671 if (ret != TRACE_TYPE_HANDLED)
672 return ret;
674 ret = trace_seq_printf(s, "| ");
675 if (!ret)
676 return TRACE_TYPE_PARTIAL_LINE;
678 return TRACE_TYPE_HANDLED;
681 /* Case of a leaf function on its call entry */
682 static enum print_line_t
683 print_graph_entry_leaf(struct trace_iterator *iter,
684 struct ftrace_graph_ent_entry *entry,
685 struct ftrace_graph_ret_entry *ret_entry,
686 struct trace_seq *s, u32 flags)
688 struct fgraph_data *data = iter->private;
689 struct ftrace_graph_ret *graph_ret;
690 struct ftrace_graph_ent *call;
691 unsigned long long duration;
692 int ret;
693 int i;
695 graph_ret = &ret_entry->ret;
696 call = &entry->graph_ent;
697 duration = graph_ret->rettime - graph_ret->calltime;
699 if (data) {
700 struct fgraph_cpu_data *cpu_data;
701 int cpu = iter->cpu;
703 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
706 * Comments display at + 1 to depth. Since
707 * this is a leaf function, keep the comments
708 * equal to this depth.
710 cpu_data->depth = call->depth - 1;
712 /* No need to keep this function around for this depth */
713 if (call->depth < FTRACE_RETFUNC_DEPTH)
714 cpu_data->enter_funcs[call->depth] = 0;
717 /* Overhead */
718 ret = print_graph_overhead(duration, s, flags);
719 if (!ret)
720 return TRACE_TYPE_PARTIAL_LINE;
722 /* Duration */
723 if (flags & TRACE_GRAPH_PRINT_DURATION) {
724 ret = print_graph_duration(duration, s);
725 if (ret == TRACE_TYPE_PARTIAL_LINE)
726 return TRACE_TYPE_PARTIAL_LINE;
729 /* Function */
730 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
731 ret = trace_seq_printf(s, " ");
732 if (!ret)
733 return TRACE_TYPE_PARTIAL_LINE;
736 ret = trace_seq_printf(s, "%ps();\n", (void *)call->func);
737 if (!ret)
738 return TRACE_TYPE_PARTIAL_LINE;
740 return TRACE_TYPE_HANDLED;
743 static enum print_line_t
744 print_graph_entry_nested(struct trace_iterator *iter,
745 struct ftrace_graph_ent_entry *entry,
746 struct trace_seq *s, int cpu, u32 flags)
748 struct ftrace_graph_ent *call = &entry->graph_ent;
749 struct fgraph_data *data = iter->private;
750 int ret;
751 int i;
753 if (data) {
754 struct fgraph_cpu_data *cpu_data;
755 int cpu = iter->cpu;
757 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
758 cpu_data->depth = call->depth;
760 /* Save this function pointer to see if the exit matches */
761 if (call->depth < FTRACE_RETFUNC_DEPTH)
762 cpu_data->enter_funcs[call->depth] = call->func;
765 /* No overhead */
766 ret = print_graph_overhead(-1, s, flags);
767 if (!ret)
768 return TRACE_TYPE_PARTIAL_LINE;
770 /* No time */
771 if (flags & TRACE_GRAPH_PRINT_DURATION) {
772 ret = trace_seq_printf(s, " | ");
773 if (!ret)
774 return TRACE_TYPE_PARTIAL_LINE;
777 /* Function */
778 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
779 ret = trace_seq_printf(s, " ");
780 if (!ret)
781 return TRACE_TYPE_PARTIAL_LINE;
784 ret = trace_seq_printf(s, "%ps() {\n", (void *)call->func);
785 if (!ret)
786 return TRACE_TYPE_PARTIAL_LINE;
789 * we already consumed the current entry to check the next one
790 * and see if this is a leaf.
792 return TRACE_TYPE_NO_CONSUME;
795 static enum print_line_t
796 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
797 int type, unsigned long addr, u32 flags)
799 struct fgraph_data *data = iter->private;
800 struct trace_entry *ent = iter->ent;
801 int cpu = iter->cpu;
802 int ret;
804 /* Pid */
805 if (verif_pid(s, ent->pid, cpu, data) == TRACE_TYPE_PARTIAL_LINE)
806 return TRACE_TYPE_PARTIAL_LINE;
808 if (type) {
809 /* Interrupt */
810 ret = print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
811 if (ret == TRACE_TYPE_PARTIAL_LINE)
812 return TRACE_TYPE_PARTIAL_LINE;
815 /* Absolute time */
816 if (flags & TRACE_GRAPH_PRINT_ABS_TIME) {
817 ret = print_graph_abs_time(iter->ts, s);
818 if (!ret)
819 return TRACE_TYPE_PARTIAL_LINE;
822 /* Cpu */
823 if (flags & TRACE_GRAPH_PRINT_CPU) {
824 ret = print_graph_cpu(s, cpu);
825 if (ret == TRACE_TYPE_PARTIAL_LINE)
826 return TRACE_TYPE_PARTIAL_LINE;
829 /* Proc */
830 if (flags & TRACE_GRAPH_PRINT_PROC) {
831 ret = print_graph_proc(s, ent->pid);
832 if (ret == TRACE_TYPE_PARTIAL_LINE)
833 return TRACE_TYPE_PARTIAL_LINE;
835 ret = trace_seq_printf(s, " | ");
836 if (!ret)
837 return TRACE_TYPE_PARTIAL_LINE;
840 /* Latency format */
841 if (trace_flags & TRACE_ITER_LATENCY_FMT) {
842 ret = print_graph_lat_fmt(s, ent);
843 if (ret == TRACE_TYPE_PARTIAL_LINE)
844 return TRACE_TYPE_PARTIAL_LINE;
847 return 0;
850 static enum print_line_t
851 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
852 struct trace_iterator *iter, u32 flags)
854 struct fgraph_data *data = iter->private;
855 struct ftrace_graph_ent *call = &field->graph_ent;
856 struct ftrace_graph_ret_entry *leaf_ret;
857 static enum print_line_t ret;
858 int cpu = iter->cpu;
860 if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags))
861 return TRACE_TYPE_PARTIAL_LINE;
863 leaf_ret = get_return_for_leaf(iter, field);
864 if (leaf_ret)
865 ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
866 else
867 ret = print_graph_entry_nested(iter, field, s, cpu, flags);
869 if (data) {
871 * If we failed to write our output, then we need to make
872 * note of it. Because we already consumed our entry.
874 if (s->full) {
875 data->failed = 1;
876 data->cpu = cpu;
877 } else
878 data->failed = 0;
881 return ret;
884 static enum print_line_t
885 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
886 struct trace_entry *ent, struct trace_iterator *iter,
887 u32 flags)
889 unsigned long long duration = trace->rettime - trace->calltime;
890 struct fgraph_data *data = iter->private;
891 pid_t pid = ent->pid;
892 int cpu = iter->cpu;
893 int func_match = 1;
894 int ret;
895 int i;
897 if (data) {
898 struct fgraph_cpu_data *cpu_data;
899 int cpu = iter->cpu;
901 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
904 * Comments display at + 1 to depth. This is the
905 * return from a function, we now want the comments
906 * to display at the same level of the bracket.
908 cpu_data->depth = trace->depth - 1;
910 if (trace->depth < FTRACE_RETFUNC_DEPTH) {
911 if (cpu_data->enter_funcs[trace->depth] != trace->func)
912 func_match = 0;
913 cpu_data->enter_funcs[trace->depth] = 0;
917 if (print_graph_prologue(iter, s, 0, 0, flags))
918 return TRACE_TYPE_PARTIAL_LINE;
920 /* Overhead */
921 ret = print_graph_overhead(duration, s, flags);
922 if (!ret)
923 return TRACE_TYPE_PARTIAL_LINE;
925 /* Duration */
926 if (flags & TRACE_GRAPH_PRINT_DURATION) {
927 ret = print_graph_duration(duration, s);
928 if (ret == TRACE_TYPE_PARTIAL_LINE)
929 return TRACE_TYPE_PARTIAL_LINE;
932 /* Closing brace */
933 for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
934 ret = trace_seq_printf(s, " ");
935 if (!ret)
936 return TRACE_TYPE_PARTIAL_LINE;
940 * If the return function does not have a matching entry,
941 * then the entry was lost. Instead of just printing
942 * the '}' and letting the user guess what function this
943 * belongs to, write out the function name.
945 if (func_match) {
946 ret = trace_seq_printf(s, "}\n");
947 if (!ret)
948 return TRACE_TYPE_PARTIAL_LINE;
949 } else {
950 ret = trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
951 if (!ret)
952 return TRACE_TYPE_PARTIAL_LINE;
955 /* Overrun */
956 if (flags & TRACE_GRAPH_PRINT_OVERRUN) {
957 ret = trace_seq_printf(s, " (Overruns: %lu)\n",
958 trace->overrun);
959 if (!ret)
960 return TRACE_TYPE_PARTIAL_LINE;
963 ret = print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
964 cpu, pid, flags);
965 if (ret == TRACE_TYPE_PARTIAL_LINE)
966 return TRACE_TYPE_PARTIAL_LINE;
968 return TRACE_TYPE_HANDLED;
971 static enum print_line_t
972 print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
973 struct trace_iterator *iter, u32 flags)
975 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
976 struct fgraph_data *data = iter->private;
977 struct trace_event *event;
978 int depth = 0;
979 int ret;
980 int i;
982 if (data)
983 depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
985 if (print_graph_prologue(iter, s, 0, 0, flags))
986 return TRACE_TYPE_PARTIAL_LINE;
988 /* No overhead */
989 ret = print_graph_overhead(-1, s, flags);
990 if (!ret)
991 return TRACE_TYPE_PARTIAL_LINE;
993 /* No time */
994 if (flags & TRACE_GRAPH_PRINT_DURATION) {
995 ret = trace_seq_printf(s, " | ");
996 if (!ret)
997 return TRACE_TYPE_PARTIAL_LINE;
1000 /* Indentation */
1001 if (depth > 0)
1002 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++) {
1003 ret = trace_seq_printf(s, " ");
1004 if (!ret)
1005 return TRACE_TYPE_PARTIAL_LINE;
1008 /* The comment */
1009 ret = trace_seq_printf(s, "/* ");
1010 if (!ret)
1011 return TRACE_TYPE_PARTIAL_LINE;
1013 switch (iter->ent->type) {
1014 case TRACE_BPRINT:
1015 ret = trace_print_bprintk_msg_only(iter);
1016 if (ret != TRACE_TYPE_HANDLED)
1017 return ret;
1018 break;
1019 case TRACE_PRINT:
1020 ret = trace_print_printk_msg_only(iter);
1021 if (ret != TRACE_TYPE_HANDLED)
1022 return ret;
1023 break;
1024 default:
1025 event = ftrace_find_event(ent->type);
1026 if (!event)
1027 return TRACE_TYPE_UNHANDLED;
1029 ret = event->funcs->trace(iter, sym_flags, event);
1030 if (ret != TRACE_TYPE_HANDLED)
1031 return ret;
1034 /* Strip ending newline */
1035 if (s->buffer[s->len - 1] == '\n') {
1036 s->buffer[s->len - 1] = '\0';
1037 s->len--;
1040 ret = trace_seq_printf(s, " */\n");
1041 if (!ret)
1042 return TRACE_TYPE_PARTIAL_LINE;
1044 return TRACE_TYPE_HANDLED;
1048 enum print_line_t
1049 print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1051 struct ftrace_graph_ent_entry *field;
1052 struct fgraph_data *data = iter->private;
1053 struct trace_entry *entry = iter->ent;
1054 struct trace_seq *s = &iter->seq;
1055 int cpu = iter->cpu;
1056 int ret;
1058 if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1059 per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1060 return TRACE_TYPE_HANDLED;
1064 * If the last output failed, there's a possibility we need
1065 * to print out the missing entry which would never go out.
1067 if (data && data->failed) {
1068 field = &data->ent;
1069 iter->cpu = data->cpu;
1070 ret = print_graph_entry(field, s, iter, flags);
1071 if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1072 per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1073 ret = TRACE_TYPE_NO_CONSUME;
1075 iter->cpu = cpu;
1076 return ret;
1079 switch (entry->type) {
1080 case TRACE_GRAPH_ENT: {
1082 * print_graph_entry() may consume the current event,
1083 * thus @field may become invalid, so we need to save it.
1084 * sizeof(struct ftrace_graph_ent_entry) is very small,
1085 * it can be safely saved at the stack.
1087 struct ftrace_graph_ent_entry saved;
1088 trace_assign_type(field, entry);
1089 saved = *field;
1090 return print_graph_entry(&saved, s, iter, flags);
1092 case TRACE_GRAPH_RET: {
1093 struct ftrace_graph_ret_entry *field;
1094 trace_assign_type(field, entry);
1095 return print_graph_return(&field->ret, s, entry, iter, flags);
1097 case TRACE_STACK:
1098 case TRACE_FN:
1099 /* dont trace stack and functions as comments */
1100 return TRACE_TYPE_UNHANDLED;
1102 default:
1103 return print_graph_comment(s, entry, iter, flags);
1106 return TRACE_TYPE_HANDLED;
1109 static enum print_line_t
1110 print_graph_function(struct trace_iterator *iter)
1112 return print_graph_function_flags(iter, tracer_flags.val);
1115 static enum print_line_t
1116 print_graph_function_event(struct trace_iterator *iter, int flags,
1117 struct trace_event *event)
1119 return print_graph_function(iter);
1122 static void print_lat_header(struct seq_file *s, u32 flags)
1124 static const char spaces[] = " " /* 16 spaces */
1125 " " /* 4 spaces */
1126 " "; /* 17 spaces */
1127 int size = 0;
1129 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1130 size += 16;
1131 if (flags & TRACE_GRAPH_PRINT_CPU)
1132 size += 4;
1133 if (flags & TRACE_GRAPH_PRINT_PROC)
1134 size += 17;
1136 seq_printf(s, "#%.*s _-----=> irqs-off \n", size, spaces);
1137 seq_printf(s, "#%.*s / _----=> need-resched \n", size, spaces);
1138 seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1139 seq_printf(s, "#%.*s|| / _--=> preempt-depth \n", size, spaces);
1140 seq_printf(s, "#%.*s||| / _-=> lock-depth \n", size, spaces);
1141 seq_printf(s, "#%.*s|||| / \n", size, spaces);
1144 void print_graph_headers_flags(struct seq_file *s, u32 flags)
1146 int lat = trace_flags & TRACE_ITER_LATENCY_FMT;
1148 if (lat)
1149 print_lat_header(s, flags);
1151 /* 1st line */
1152 seq_printf(s, "#");
1153 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1154 seq_printf(s, " TIME ");
1155 if (flags & TRACE_GRAPH_PRINT_CPU)
1156 seq_printf(s, " CPU");
1157 if (flags & TRACE_GRAPH_PRINT_PROC)
1158 seq_printf(s, " TASK/PID ");
1159 if (lat)
1160 seq_printf(s, "|||||");
1161 if (flags & TRACE_GRAPH_PRINT_DURATION)
1162 seq_printf(s, " DURATION ");
1163 seq_printf(s, " FUNCTION CALLS\n");
1165 /* 2nd line */
1166 seq_printf(s, "#");
1167 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1168 seq_printf(s, " | ");
1169 if (flags & TRACE_GRAPH_PRINT_CPU)
1170 seq_printf(s, " | ");
1171 if (flags & TRACE_GRAPH_PRINT_PROC)
1172 seq_printf(s, " | | ");
1173 if (lat)
1174 seq_printf(s, "|||||");
1175 if (flags & TRACE_GRAPH_PRINT_DURATION)
1176 seq_printf(s, " | | ");
1177 seq_printf(s, " | | | |\n");
1180 void print_graph_headers(struct seq_file *s)
1182 print_graph_headers_flags(s, tracer_flags.val);
1185 void graph_trace_open(struct trace_iterator *iter)
1187 /* pid and depth on the last trace processed */
1188 struct fgraph_data *data;
1189 int cpu;
1191 iter->private = NULL;
1193 data = kzalloc(sizeof(*data), GFP_KERNEL);
1194 if (!data)
1195 goto out_err;
1197 data->cpu_data = alloc_percpu(struct fgraph_cpu_data);
1198 if (!data->cpu_data)
1199 goto out_err_free;
1201 for_each_possible_cpu(cpu) {
1202 pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1203 int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1204 int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1205 *pid = -1;
1206 *depth = 0;
1207 *ignore = 0;
1210 iter->private = data;
1212 return;
1214 out_err_free:
1215 kfree(data);
1216 out_err:
1217 pr_warning("function graph tracer: not enough memory\n");
1220 void graph_trace_close(struct trace_iterator *iter)
1222 struct fgraph_data *data = iter->private;
1224 if (data) {
1225 free_percpu(data->cpu_data);
1226 kfree(data);
1230 static struct trace_event_functions graph_functions = {
1231 .trace = print_graph_function_event,
1234 static struct trace_event graph_trace_entry_event = {
1235 .type = TRACE_GRAPH_ENT,
1236 .funcs = &graph_functions,
1239 static struct trace_event graph_trace_ret_event = {
1240 .type = TRACE_GRAPH_RET,
1241 .funcs = &graph_functions
1244 static struct tracer graph_trace __read_mostly = {
1245 .name = "function_graph",
1246 .open = graph_trace_open,
1247 .pipe_open = graph_trace_open,
1248 .close = graph_trace_close,
1249 .pipe_close = graph_trace_close,
1250 .wait_pipe = poll_wait_pipe,
1251 .init = graph_trace_init,
1252 .reset = graph_trace_reset,
1253 .print_line = print_graph_function,
1254 .print_header = print_graph_headers,
1255 .flags = &tracer_flags,
1256 #ifdef CONFIG_FTRACE_SELFTEST
1257 .selftest = trace_selftest_startup_function_graph,
1258 #endif
1261 static __init int init_graph_trace(void)
1263 max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1265 if (!register_ftrace_event(&graph_trace_entry_event)) {
1266 pr_warning("Warning: could not register graph trace events\n");
1267 return 1;
1270 if (!register_ftrace_event(&graph_trace_ret_event)) {
1271 pr_warning("Warning: could not register graph trace events\n");
1272 return 1;
1275 return register_tracer(&graph_trace);
1278 device_initcall(init_graph_trace);