Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 *
4 * Function graph tracer.
5 * Copyright (c) 2008-2009 Frederic Weisbecker <fweisbec@gmail.com>
6 * Mostly borrowed from function tracer which
7 * is Copyright (c) Steven Rostedt <srostedt@redhat.com>
8 *
9 */
10#include <linux/uaccess.h>
11#include <linux/ftrace.h>
12#include <linux/interrupt.h>
13#include <linux/slab.h>
14#include <linux/fs.h>
15
16#include "trace.h"
17#include "trace_output.h"
18
19/* When set, irq functions will be ignored */
20static int ftrace_graph_skip_irqs;
21
22struct fgraph_cpu_data {
23 pid_t last_pid;
24 int depth;
25 int depth_irq;
26 int ignore;
27 unsigned long enter_funcs[FTRACE_RETFUNC_DEPTH];
28};
29
30struct fgraph_data {
31 struct fgraph_cpu_data __percpu *cpu_data;
32
33 /* Place to preserve last processed entry. */
34 struct ftrace_graph_ent_entry ent;
35 struct ftrace_graph_ret_entry ret;
36 int failed;
37 int cpu;
38};
39
40#define TRACE_GRAPH_INDENT 2
41
42unsigned int fgraph_max_depth;
43
44static struct tracer_opt trace_opts[] = {
45 /* Display overruns? (for self-debug purpose) */
46 { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
47 /* Display CPU ? */
48 { TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
49 /* Display Overhead ? */
50 { TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
51 /* Display proc name/pid */
52 { TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
53 /* Display duration of execution */
54 { TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
55 /* Display absolute time of an entry */
56 { TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
57 /* Display interrupts */
58 { TRACER_OPT(funcgraph-irqs, TRACE_GRAPH_PRINT_IRQS) },
59 /* Display function name after trailing } */
60 { TRACER_OPT(funcgraph-tail, TRACE_GRAPH_PRINT_TAIL) },
61 /* Include sleep time (scheduled out) between entry and return */
62 { TRACER_OPT(sleep-time, TRACE_GRAPH_SLEEP_TIME) },
63
64#ifdef CONFIG_FUNCTION_PROFILER
65 /* Include time within nested functions */
66 { TRACER_OPT(graph-time, TRACE_GRAPH_GRAPH_TIME) },
67#endif
68
69 { } /* Empty entry */
70};
71
72static struct tracer_flags tracer_flags = {
73 /* Don't display overruns, proc, or tail by default */
74 .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
75 TRACE_GRAPH_PRINT_DURATION | TRACE_GRAPH_PRINT_IRQS |
76 TRACE_GRAPH_SLEEP_TIME | TRACE_GRAPH_GRAPH_TIME,
77 .opts = trace_opts
78};
79
80static struct trace_array *graph_array;
81
82/*
83 * DURATION column is being also used to display IRQ signs,
84 * following values are used by print_graph_irq and others
85 * to fill in space into DURATION column.
86 */
87enum {
88 FLAGS_FILL_FULL = 1 << TRACE_GRAPH_PRINT_FILL_SHIFT,
89 FLAGS_FILL_START = 2 << TRACE_GRAPH_PRINT_FILL_SHIFT,
90 FLAGS_FILL_END = 3 << TRACE_GRAPH_PRINT_FILL_SHIFT,
91};
92
93static void
94print_graph_duration(struct trace_array *tr, unsigned long long duration,
95 struct trace_seq *s, u32 flags);
96
97int __trace_graph_entry(struct trace_array *tr,
98 struct ftrace_graph_ent *trace,
99 unsigned long flags,
100 int pc)
101{
102 struct trace_event_call *call = &event_funcgraph_entry;
103 struct ring_buffer_event *event;
104 struct ring_buffer *buffer = tr->trace_buffer.buffer;
105 struct ftrace_graph_ent_entry *entry;
106
107 event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
108 sizeof(*entry), flags, pc);
109 if (!event)
110 return 0;
111 entry = ring_buffer_event_data(event);
112 entry->graph_ent = *trace;
113 if (!call_filter_check_discard(call, entry, buffer, event))
114 trace_buffer_unlock_commit_nostack(buffer, event);
115
116 return 1;
117}
118
119static inline int ftrace_graph_ignore_irqs(void)
120{
121 if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
122 return 0;
123
124 return in_irq();
125}
126
127int trace_graph_entry(struct ftrace_graph_ent *trace)
128{
129 struct trace_array *tr = graph_array;
130 struct trace_array_cpu *data;
131 unsigned long flags;
132 long disabled;
133 int ret;
134 int cpu;
135 int pc;
136
137 if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT))
138 return 0;
139
140 /*
141 * Do not trace a function if it's filtered by set_graph_notrace.
142 * Make the index of ret stack negative to indicate that it should
143 * ignore further functions. But it needs its own ret stack entry
144 * to recover the original index in order to continue tracing after
145 * returning from the function.
146 */
147 if (ftrace_graph_notrace_addr(trace->func)) {
148 trace_recursion_set(TRACE_GRAPH_NOTRACE_BIT);
149 /*
150 * Need to return 1 to have the return called
151 * that will clear the NOTRACE bit.
152 */
153 return 1;
154 }
155
156 if (!ftrace_trace_task(tr))
157 return 0;
158
159 if (ftrace_graph_ignore_func(trace))
160 return 0;
161
162 if (ftrace_graph_ignore_irqs())
163 return 0;
164
165 /*
166 * Stop here if tracing_threshold is set. We only write function return
167 * events to the ring buffer.
168 */
169 if (tracing_thresh)
170 return 1;
171
172 local_irq_save(flags);
173 cpu = raw_smp_processor_id();
174 data = per_cpu_ptr(tr->trace_buffer.data, cpu);
175 disabled = atomic_inc_return(&data->disabled);
176 if (likely(disabled == 1)) {
177 pc = preempt_count();
178 ret = __trace_graph_entry(tr, trace, flags, pc);
179 } else {
180 ret = 0;
181 }
182
183 atomic_dec(&data->disabled);
184 local_irq_restore(flags);
185
186 return ret;
187}
188
189static void
190__trace_graph_function(struct trace_array *tr,
191 unsigned long ip, unsigned long flags, int pc)
192{
193 u64 time = trace_clock_local();
194 struct ftrace_graph_ent ent = {
195 .func = ip,
196 .depth = 0,
197 };
198 struct ftrace_graph_ret ret = {
199 .func = ip,
200 .depth = 0,
201 .calltime = time,
202 .rettime = time,
203 };
204
205 __trace_graph_entry(tr, &ent, flags, pc);
206 __trace_graph_return(tr, &ret, flags, pc);
207}
208
209void
210trace_graph_function(struct trace_array *tr,
211 unsigned long ip, unsigned long parent_ip,
212 unsigned long flags, int pc)
213{
214 __trace_graph_function(tr, ip, flags, pc);
215}
216
217void __trace_graph_return(struct trace_array *tr,
218 struct ftrace_graph_ret *trace,
219 unsigned long flags,
220 int pc)
221{
222 struct trace_event_call *call = &event_funcgraph_exit;
223 struct ring_buffer_event *event;
224 struct ring_buffer *buffer = tr->trace_buffer.buffer;
225 struct ftrace_graph_ret_entry *entry;
226
227 event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
228 sizeof(*entry), flags, pc);
229 if (!event)
230 return;
231 entry = ring_buffer_event_data(event);
232 entry->ret = *trace;
233 if (!call_filter_check_discard(call, entry, buffer, event))
234 trace_buffer_unlock_commit_nostack(buffer, event);
235}
236
237void trace_graph_return(struct ftrace_graph_ret *trace)
238{
239 struct trace_array *tr = graph_array;
240 struct trace_array_cpu *data;
241 unsigned long flags;
242 long disabled;
243 int cpu;
244 int pc;
245
246 ftrace_graph_addr_finish(trace);
247
248 if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT)) {
249 trace_recursion_clear(TRACE_GRAPH_NOTRACE_BIT);
250 return;
251 }
252
253 local_irq_save(flags);
254 cpu = raw_smp_processor_id();
255 data = per_cpu_ptr(tr->trace_buffer.data, cpu);
256 disabled = atomic_inc_return(&data->disabled);
257 if (likely(disabled == 1)) {
258 pc = preempt_count();
259 __trace_graph_return(tr, trace, flags, pc);
260 }
261 atomic_dec(&data->disabled);
262 local_irq_restore(flags);
263}
264
265void set_graph_array(struct trace_array *tr)
266{
267 graph_array = tr;
268
269 /* Make graph_array visible before we start tracing */
270
271 smp_mb();
272}
273
274static void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
275{
276 ftrace_graph_addr_finish(trace);
277
278 if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT)) {
279 trace_recursion_clear(TRACE_GRAPH_NOTRACE_BIT);
280 return;
281 }
282
283 if (tracing_thresh &&
284 (trace->rettime - trace->calltime < tracing_thresh))
285 return;
286 else
287 trace_graph_return(trace);
288}
289
290static struct fgraph_ops funcgraph_thresh_ops = {
291 .entryfunc = &trace_graph_entry,
292 .retfunc = &trace_graph_thresh_return,
293};
294
295static struct fgraph_ops funcgraph_ops = {
296 .entryfunc = &trace_graph_entry,
297 .retfunc = &trace_graph_return,
298};
299
300static int graph_trace_init(struct trace_array *tr)
301{
302 int ret;
303
304 set_graph_array(tr);
305 if (tracing_thresh)
306 ret = register_ftrace_graph(&funcgraph_thresh_ops);
307 else
308 ret = register_ftrace_graph(&funcgraph_ops);
309 if (ret)
310 return ret;
311 tracing_start_cmdline_record();
312
313 return 0;
314}
315
316static void graph_trace_reset(struct trace_array *tr)
317{
318 tracing_stop_cmdline_record();
319 if (tracing_thresh)
320 unregister_ftrace_graph(&funcgraph_thresh_ops);
321 else
322 unregister_ftrace_graph(&funcgraph_ops);
323}
324
325static int graph_trace_update_thresh(struct trace_array *tr)
326{
327 graph_trace_reset(tr);
328 return graph_trace_init(tr);
329}
330
331static int max_bytes_for_cpu;
332
333static void print_graph_cpu(struct trace_seq *s, int cpu)
334{
335 /*
336 * Start with a space character - to make it stand out
337 * to the right a bit when trace output is pasted into
338 * email:
339 */
340 trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
341}
342
343#define TRACE_GRAPH_PROCINFO_LENGTH 14
344
345static void print_graph_proc(struct trace_seq *s, pid_t pid)
346{
347 char comm[TASK_COMM_LEN];
348 /* sign + log10(MAX_INT) + '\0' */
349 char pid_str[11];
350 int spaces = 0;
351 int len;
352 int i;
353
354 trace_find_cmdline(pid, comm);
355 comm[7] = '\0';
356 sprintf(pid_str, "%d", pid);
357
358 /* 1 stands for the "-" character */
359 len = strlen(comm) + strlen(pid_str) + 1;
360
361 if (len < TRACE_GRAPH_PROCINFO_LENGTH)
362 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
363
364 /* First spaces to align center */
365 for (i = 0; i < spaces / 2; i++)
366 trace_seq_putc(s, ' ');
367
368 trace_seq_printf(s, "%s-%s", comm, pid_str);
369
370 /* Last spaces to align center */
371 for (i = 0; i < spaces - (spaces / 2); i++)
372 trace_seq_putc(s, ' ');
373}
374
375
376static void print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
377{
378 trace_seq_putc(s, ' ');
379 trace_print_lat_fmt(s, entry);
380 trace_seq_puts(s, " | ");
381}
382
383/* If the pid changed since the last trace, output this event */
384static void
385verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
386{
387 pid_t prev_pid;
388 pid_t *last_pid;
389
390 if (!data)
391 return;
392
393 last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
394
395 if (*last_pid == pid)
396 return;
397
398 prev_pid = *last_pid;
399 *last_pid = pid;
400
401 if (prev_pid == -1)
402 return;
403/*
404 * Context-switch trace line:
405
406 ------------------------------------------
407 | 1) migration/0--1 => sshd-1755
408 ------------------------------------------
409
410 */
411 trace_seq_puts(s, " ------------------------------------------\n");
412 print_graph_cpu(s, cpu);
413 print_graph_proc(s, prev_pid);
414 trace_seq_puts(s, " => ");
415 print_graph_proc(s, pid);
416 trace_seq_puts(s, "\n ------------------------------------------\n\n");
417}
418
419static struct ftrace_graph_ret_entry *
420get_return_for_leaf(struct trace_iterator *iter,
421 struct ftrace_graph_ent_entry *curr)
422{
423 struct fgraph_data *data = iter->private;
424 struct ring_buffer_iter *ring_iter = NULL;
425 struct ring_buffer_event *event;
426 struct ftrace_graph_ret_entry *next;
427
428 /*
429 * If the previous output failed to write to the seq buffer,
430 * then we just reuse the data from before.
431 */
432 if (data && data->failed) {
433 curr = &data->ent;
434 next = &data->ret;
435 } else {
436
437 ring_iter = trace_buffer_iter(iter, iter->cpu);
438
439 /* First peek to compare current entry and the next one */
440 if (ring_iter)
441 event = ring_buffer_iter_peek(ring_iter, NULL);
442 else {
443 /*
444 * We need to consume the current entry to see
445 * the next one.
446 */
447 ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu,
448 NULL, NULL);
449 event = ring_buffer_peek(iter->trace_buffer->buffer, iter->cpu,
450 NULL, NULL);
451 }
452
453 if (!event)
454 return NULL;
455
456 next = ring_buffer_event_data(event);
457
458 if (data) {
459 /*
460 * Save current and next entries for later reference
461 * if the output fails.
462 */
463 data->ent = *curr;
464 /*
465 * If the next event is not a return type, then
466 * we only care about what type it is. Otherwise we can
467 * safely copy the entire event.
468 */
469 if (next->ent.type == TRACE_GRAPH_RET)
470 data->ret = *next;
471 else
472 data->ret.ent.type = next->ent.type;
473 }
474 }
475
476 if (next->ent.type != TRACE_GRAPH_RET)
477 return NULL;
478
479 if (curr->ent.pid != next->ent.pid ||
480 curr->graph_ent.func != next->ret.func)
481 return NULL;
482
483 /* this is a leaf, now advance the iterator */
484 if (ring_iter)
485 ring_buffer_read(ring_iter, NULL);
486
487 return next;
488}
489
490static void print_graph_abs_time(u64 t, struct trace_seq *s)
491{
492 unsigned long usecs_rem;
493
494 usecs_rem = do_div(t, NSEC_PER_SEC);
495 usecs_rem /= 1000;
496
497 trace_seq_printf(s, "%5lu.%06lu | ",
498 (unsigned long)t, usecs_rem);
499}
500
501static void
502print_graph_rel_time(struct trace_iterator *iter, struct trace_seq *s)
503{
504 unsigned long long usecs;
505
506 usecs = iter->ts - iter->trace_buffer->time_start;
507 do_div(usecs, NSEC_PER_USEC);
508
509 trace_seq_printf(s, "%9llu us | ", usecs);
510}
511
512static void
513print_graph_irq(struct trace_iterator *iter, unsigned long addr,
514 enum trace_type type, int cpu, pid_t pid, u32 flags)
515{
516 struct trace_array *tr = iter->tr;
517 struct trace_seq *s = &iter->seq;
518 struct trace_entry *ent = iter->ent;
519
520 if (addr < (unsigned long)__irqentry_text_start ||
521 addr >= (unsigned long)__irqentry_text_end)
522 return;
523
524 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
525 /* Absolute time */
526 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
527 print_graph_abs_time(iter->ts, s);
528
529 /* Relative time */
530 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
531 print_graph_rel_time(iter, s);
532
533 /* Cpu */
534 if (flags & TRACE_GRAPH_PRINT_CPU)
535 print_graph_cpu(s, cpu);
536
537 /* Proc */
538 if (flags & TRACE_GRAPH_PRINT_PROC) {
539 print_graph_proc(s, pid);
540 trace_seq_puts(s, " | ");
541 }
542
543 /* Latency format */
544 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
545 print_graph_lat_fmt(s, ent);
546 }
547
548 /* No overhead */
549 print_graph_duration(tr, 0, s, flags | FLAGS_FILL_START);
550
551 if (type == TRACE_GRAPH_ENT)
552 trace_seq_puts(s, "==========>");
553 else
554 trace_seq_puts(s, "<==========");
555
556 print_graph_duration(tr, 0, s, flags | FLAGS_FILL_END);
557 trace_seq_putc(s, '\n');
558}
559
560void
561trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
562{
563 unsigned long nsecs_rem = do_div(duration, 1000);
564 /* log10(ULONG_MAX) + '\0' */
565 char usecs_str[21];
566 char nsecs_str[5];
567 int len;
568 int i;
569
570 sprintf(usecs_str, "%lu", (unsigned long) duration);
571
572 /* Print msecs */
573 trace_seq_printf(s, "%s", usecs_str);
574
575 len = strlen(usecs_str);
576
577 /* Print nsecs (we don't want to exceed 7 numbers) */
578 if (len < 7) {
579 size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
580
581 snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
582 trace_seq_printf(s, ".%s", nsecs_str);
583 len += strlen(nsecs_str) + 1;
584 }
585
586 trace_seq_puts(s, " us ");
587
588 /* Print remaining spaces to fit the row's width */
589 for (i = len; i < 8; i++)
590 trace_seq_putc(s, ' ');
591}
592
593static void
594print_graph_duration(struct trace_array *tr, unsigned long long duration,
595 struct trace_seq *s, u32 flags)
596{
597 if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
598 !(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
599 return;
600
601 /* No real adata, just filling the column with spaces */
602 switch (flags & TRACE_GRAPH_PRINT_FILL_MASK) {
603 case FLAGS_FILL_FULL:
604 trace_seq_puts(s, " | ");
605 return;
606 case FLAGS_FILL_START:
607 trace_seq_puts(s, " ");
608 return;
609 case FLAGS_FILL_END:
610 trace_seq_puts(s, " |");
611 return;
612 }
613
614 /* Signal a overhead of time execution to the output */
615 if (flags & TRACE_GRAPH_PRINT_OVERHEAD)
616 trace_seq_printf(s, "%c ", trace_find_mark(duration));
617 else
618 trace_seq_puts(s, " ");
619
620 trace_print_graph_duration(duration, s);
621 trace_seq_puts(s, "| ");
622}
623
624/* Case of a leaf function on its call entry */
625static enum print_line_t
626print_graph_entry_leaf(struct trace_iterator *iter,
627 struct ftrace_graph_ent_entry *entry,
628 struct ftrace_graph_ret_entry *ret_entry,
629 struct trace_seq *s, u32 flags)
630{
631 struct fgraph_data *data = iter->private;
632 struct trace_array *tr = iter->tr;
633 struct ftrace_graph_ret *graph_ret;
634 struct ftrace_graph_ent *call;
635 unsigned long long duration;
636 int cpu = iter->cpu;
637 int i;
638
639 graph_ret = &ret_entry->ret;
640 call = &entry->graph_ent;
641 duration = graph_ret->rettime - graph_ret->calltime;
642
643 if (data) {
644 struct fgraph_cpu_data *cpu_data;
645
646 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
647
648 /*
649 * Comments display at + 1 to depth. Since
650 * this is a leaf function, keep the comments
651 * equal to this depth.
652 */
653 cpu_data->depth = call->depth - 1;
654
655 /* No need to keep this function around for this depth */
656 if (call->depth < FTRACE_RETFUNC_DEPTH &&
657 !WARN_ON_ONCE(call->depth < 0))
658 cpu_data->enter_funcs[call->depth] = 0;
659 }
660
661 /* Overhead and duration */
662 print_graph_duration(tr, duration, s, flags);
663
664 /* Function */
665 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
666 trace_seq_putc(s, ' ');
667
668 trace_seq_printf(s, "%ps();\n", (void *)call->func);
669
670 print_graph_irq(iter, graph_ret->func, TRACE_GRAPH_RET,
671 cpu, iter->ent->pid, flags);
672
673 return trace_handle_return(s);
674}
675
676static enum print_line_t
677print_graph_entry_nested(struct trace_iterator *iter,
678 struct ftrace_graph_ent_entry *entry,
679 struct trace_seq *s, int cpu, u32 flags)
680{
681 struct ftrace_graph_ent *call = &entry->graph_ent;
682 struct fgraph_data *data = iter->private;
683 struct trace_array *tr = iter->tr;
684 int i;
685
686 if (data) {
687 struct fgraph_cpu_data *cpu_data;
688 int cpu = iter->cpu;
689
690 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
691 cpu_data->depth = call->depth;
692
693 /* Save this function pointer to see if the exit matches */
694 if (call->depth < FTRACE_RETFUNC_DEPTH &&
695 !WARN_ON_ONCE(call->depth < 0))
696 cpu_data->enter_funcs[call->depth] = call->func;
697 }
698
699 /* No time */
700 print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
701
702 /* Function */
703 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
704 trace_seq_putc(s, ' ');
705
706 trace_seq_printf(s, "%ps() {\n", (void *)call->func);
707
708 if (trace_seq_has_overflowed(s))
709 return TRACE_TYPE_PARTIAL_LINE;
710
711 /*
712 * we already consumed the current entry to check the next one
713 * and see if this is a leaf.
714 */
715 return TRACE_TYPE_NO_CONSUME;
716}
717
718static void
719print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
720 int type, unsigned long addr, u32 flags)
721{
722 struct fgraph_data *data = iter->private;
723 struct trace_entry *ent = iter->ent;
724 struct trace_array *tr = iter->tr;
725 int cpu = iter->cpu;
726
727 /* Pid */
728 verif_pid(s, ent->pid, cpu, data);
729
730 if (type)
731 /* Interrupt */
732 print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
733
734 if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
735 return;
736
737 /* Absolute time */
738 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
739 print_graph_abs_time(iter->ts, s);
740
741 /* Relative time */
742 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
743 print_graph_rel_time(iter, s);
744
745 /* Cpu */
746 if (flags & TRACE_GRAPH_PRINT_CPU)
747 print_graph_cpu(s, cpu);
748
749 /* Proc */
750 if (flags & TRACE_GRAPH_PRINT_PROC) {
751 print_graph_proc(s, ent->pid);
752 trace_seq_puts(s, " | ");
753 }
754
755 /* Latency format */
756 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
757 print_graph_lat_fmt(s, ent);
758
759 return;
760}
761
762/*
763 * Entry check for irq code
764 *
765 * returns 1 if
766 * - we are inside irq code
767 * - we just entered irq code
768 *
769 * retunns 0 if
770 * - funcgraph-interrupts option is set
771 * - we are not inside irq code
772 */
773static int
774check_irq_entry(struct trace_iterator *iter, u32 flags,
775 unsigned long addr, int depth)
776{
777 int cpu = iter->cpu;
778 int *depth_irq;
779 struct fgraph_data *data = iter->private;
780
781 /*
782 * If we are either displaying irqs, or we got called as
783 * a graph event and private data does not exist,
784 * then we bypass the irq check.
785 */
786 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
787 (!data))
788 return 0;
789
790 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
791
792 /*
793 * We are inside the irq code
794 */
795 if (*depth_irq >= 0)
796 return 1;
797
798 if ((addr < (unsigned long)__irqentry_text_start) ||
799 (addr >= (unsigned long)__irqentry_text_end))
800 return 0;
801
802 /*
803 * We are entering irq code.
804 */
805 *depth_irq = depth;
806 return 1;
807}
808
809/*
810 * Return check for irq code
811 *
812 * returns 1 if
813 * - we are inside irq code
814 * - we just left irq code
815 *
816 * returns 0 if
817 * - funcgraph-interrupts option is set
818 * - we are not inside irq code
819 */
820static int
821check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
822{
823 int cpu = iter->cpu;
824 int *depth_irq;
825 struct fgraph_data *data = iter->private;
826
827 /*
828 * If we are either displaying irqs, or we got called as
829 * a graph event and private data does not exist,
830 * then we bypass the irq check.
831 */
832 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
833 (!data))
834 return 0;
835
836 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
837
838 /*
839 * We are not inside the irq code.
840 */
841 if (*depth_irq == -1)
842 return 0;
843
844 /*
845 * We are inside the irq code, and this is returning entry.
846 * Let's not trace it and clear the entry depth, since
847 * we are out of irq code.
848 *
849 * This condition ensures that we 'leave the irq code' once
850 * we are out of the entry depth. Thus protecting us from
851 * the RETURN entry loss.
852 */
853 if (*depth_irq >= depth) {
854 *depth_irq = -1;
855 return 1;
856 }
857
858 /*
859 * We are inside the irq code, and this is not the entry.
860 */
861 return 1;
862}
863
864static enum print_line_t
865print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
866 struct trace_iterator *iter, u32 flags)
867{
868 struct fgraph_data *data = iter->private;
869 struct ftrace_graph_ent *call = &field->graph_ent;
870 struct ftrace_graph_ret_entry *leaf_ret;
871 static enum print_line_t ret;
872 int cpu = iter->cpu;
873
874 if (check_irq_entry(iter, flags, call->func, call->depth))
875 return TRACE_TYPE_HANDLED;
876
877 print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags);
878
879 leaf_ret = get_return_for_leaf(iter, field);
880 if (leaf_ret)
881 ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
882 else
883 ret = print_graph_entry_nested(iter, field, s, cpu, flags);
884
885 if (data) {
886 /*
887 * If we failed to write our output, then we need to make
888 * note of it. Because we already consumed our entry.
889 */
890 if (s->full) {
891 data->failed = 1;
892 data->cpu = cpu;
893 } else
894 data->failed = 0;
895 }
896
897 return ret;
898}
899
900static enum print_line_t
901print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
902 struct trace_entry *ent, struct trace_iterator *iter,
903 u32 flags)
904{
905 unsigned long long duration = trace->rettime - trace->calltime;
906 struct fgraph_data *data = iter->private;
907 struct trace_array *tr = iter->tr;
908 pid_t pid = ent->pid;
909 int cpu = iter->cpu;
910 int func_match = 1;
911 int i;
912
913 if (check_irq_return(iter, flags, trace->depth))
914 return TRACE_TYPE_HANDLED;
915
916 if (data) {
917 struct fgraph_cpu_data *cpu_data;
918 int cpu = iter->cpu;
919
920 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
921
922 /*
923 * Comments display at + 1 to depth. This is the
924 * return from a function, we now want the comments
925 * to display at the same level of the bracket.
926 */
927 cpu_data->depth = trace->depth - 1;
928
929 if (trace->depth < FTRACE_RETFUNC_DEPTH &&
930 !WARN_ON_ONCE(trace->depth < 0)) {
931 if (cpu_data->enter_funcs[trace->depth] != trace->func)
932 func_match = 0;
933 cpu_data->enter_funcs[trace->depth] = 0;
934 }
935 }
936
937 print_graph_prologue(iter, s, 0, 0, flags);
938
939 /* Overhead and duration */
940 print_graph_duration(tr, duration, s, flags);
941
942 /* Closing brace */
943 for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++)
944 trace_seq_putc(s, ' ');
945
946 /*
947 * If the return function does not have a matching entry,
948 * then the entry was lost. Instead of just printing
949 * the '}' and letting the user guess what function this
950 * belongs to, write out the function name. Always do
951 * that if the funcgraph-tail option is enabled.
952 */
953 if (func_match && !(flags & TRACE_GRAPH_PRINT_TAIL))
954 trace_seq_puts(s, "}\n");
955 else
956 trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
957
958 /* Overrun */
959 if (flags & TRACE_GRAPH_PRINT_OVERRUN)
960 trace_seq_printf(s, " (Overruns: %lu)\n",
961 trace->overrun);
962
963 print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
964 cpu, pid, flags);
965
966 return trace_handle_return(s);
967}
968
969static enum print_line_t
970print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
971 struct trace_iterator *iter, u32 flags)
972{
973 struct trace_array *tr = iter->tr;
974 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
975 struct fgraph_data *data = iter->private;
976 struct trace_event *event;
977 int depth = 0;
978 int ret;
979 int i;
980
981 if (data)
982 depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
983
984 print_graph_prologue(iter, s, 0, 0, flags);
985
986 /* No time */
987 print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
988
989 /* Indentation */
990 if (depth > 0)
991 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++)
992 trace_seq_putc(s, ' ');
993
994 /* The comment */
995 trace_seq_puts(s, "/* ");
996
997 switch (iter->ent->type) {
998 case TRACE_BPUTS:
999 ret = trace_print_bputs_msg_only(iter);
1000 if (ret != TRACE_TYPE_HANDLED)
1001 return ret;
1002 break;
1003 case TRACE_BPRINT:
1004 ret = trace_print_bprintk_msg_only(iter);
1005 if (ret != TRACE_TYPE_HANDLED)
1006 return ret;
1007 break;
1008 case TRACE_PRINT:
1009 ret = trace_print_printk_msg_only(iter);
1010 if (ret != TRACE_TYPE_HANDLED)
1011 return ret;
1012 break;
1013 default:
1014 event = ftrace_find_event(ent->type);
1015 if (!event)
1016 return TRACE_TYPE_UNHANDLED;
1017
1018 ret = event->funcs->trace(iter, sym_flags, event);
1019 if (ret != TRACE_TYPE_HANDLED)
1020 return ret;
1021 }
1022
1023 if (trace_seq_has_overflowed(s))
1024 goto out;
1025
1026 /* Strip ending newline */
1027 if (s->buffer[s->seq.len - 1] == '\n') {
1028 s->buffer[s->seq.len - 1] = '\0';
1029 s->seq.len--;
1030 }
1031
1032 trace_seq_puts(s, " */\n");
1033 out:
1034 return trace_handle_return(s);
1035}
1036
1037
1038enum print_line_t
1039print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1040{
1041 struct ftrace_graph_ent_entry *field;
1042 struct fgraph_data *data = iter->private;
1043 struct trace_entry *entry = iter->ent;
1044 struct trace_seq *s = &iter->seq;
1045 int cpu = iter->cpu;
1046 int ret;
1047
1048 if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1049 per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1050 return TRACE_TYPE_HANDLED;
1051 }
1052
1053 /*
1054 * If the last output failed, there's a possibility we need
1055 * to print out the missing entry which would never go out.
1056 */
1057 if (data && data->failed) {
1058 field = &data->ent;
1059 iter->cpu = data->cpu;
1060 ret = print_graph_entry(field, s, iter, flags);
1061 if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1062 per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1063 ret = TRACE_TYPE_NO_CONSUME;
1064 }
1065 iter->cpu = cpu;
1066 return ret;
1067 }
1068
1069 switch (entry->type) {
1070 case TRACE_GRAPH_ENT: {
1071 /*
1072 * print_graph_entry() may consume the current event,
1073 * thus @field may become invalid, so we need to save it.
1074 * sizeof(struct ftrace_graph_ent_entry) is very small,
1075 * it can be safely saved at the stack.
1076 */
1077 struct ftrace_graph_ent_entry saved;
1078 trace_assign_type(field, entry);
1079 saved = *field;
1080 return print_graph_entry(&saved, s, iter, flags);
1081 }
1082 case TRACE_GRAPH_RET: {
1083 struct ftrace_graph_ret_entry *field;
1084 trace_assign_type(field, entry);
1085 return print_graph_return(&field->ret, s, entry, iter, flags);
1086 }
1087 case TRACE_STACK:
1088 case TRACE_FN:
1089 /* dont trace stack and functions as comments */
1090 return TRACE_TYPE_UNHANDLED;
1091
1092 default:
1093 return print_graph_comment(s, entry, iter, flags);
1094 }
1095
1096 return TRACE_TYPE_HANDLED;
1097}
1098
1099static enum print_line_t
1100print_graph_function(struct trace_iterator *iter)
1101{
1102 return print_graph_function_flags(iter, tracer_flags.val);
1103}
1104
1105static enum print_line_t
1106print_graph_function_event(struct trace_iterator *iter, int flags,
1107 struct trace_event *event)
1108{
1109 return print_graph_function(iter);
1110}
1111
1112static void print_lat_header(struct seq_file *s, u32 flags)
1113{
1114 static const char spaces[] = " " /* 16 spaces */
1115 " " /* 4 spaces */
1116 " "; /* 17 spaces */
1117 int size = 0;
1118
1119 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1120 size += 16;
1121 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1122 size += 16;
1123 if (flags & TRACE_GRAPH_PRINT_CPU)
1124 size += 4;
1125 if (flags & TRACE_GRAPH_PRINT_PROC)
1126 size += 17;
1127
1128 seq_printf(s, "#%.*s _-----=> irqs-off \n", size, spaces);
1129 seq_printf(s, "#%.*s / _----=> need-resched \n", size, spaces);
1130 seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1131 seq_printf(s, "#%.*s|| / _--=> preempt-depth \n", size, spaces);
1132 seq_printf(s, "#%.*s||| / \n", size, spaces);
1133}
1134
1135static void __print_graph_headers_flags(struct trace_array *tr,
1136 struct seq_file *s, u32 flags)
1137{
1138 int lat = tr->trace_flags & TRACE_ITER_LATENCY_FMT;
1139
1140 if (lat)
1141 print_lat_header(s, flags);
1142
1143 /* 1st line */
1144 seq_putc(s, '#');
1145 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1146 seq_puts(s, " TIME ");
1147 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1148 seq_puts(s, " REL TIME ");
1149 if (flags & TRACE_GRAPH_PRINT_CPU)
1150 seq_puts(s, " CPU");
1151 if (flags & TRACE_GRAPH_PRINT_PROC)
1152 seq_puts(s, " TASK/PID ");
1153 if (lat)
1154 seq_puts(s, "|||| ");
1155 if (flags & TRACE_GRAPH_PRINT_DURATION)
1156 seq_puts(s, " DURATION ");
1157 seq_puts(s, " FUNCTION CALLS\n");
1158
1159 /* 2nd line */
1160 seq_putc(s, '#');
1161 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1162 seq_puts(s, " | ");
1163 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1164 seq_puts(s, " | ");
1165 if (flags & TRACE_GRAPH_PRINT_CPU)
1166 seq_puts(s, " | ");
1167 if (flags & TRACE_GRAPH_PRINT_PROC)
1168 seq_puts(s, " | | ");
1169 if (lat)
1170 seq_puts(s, "|||| ");
1171 if (flags & TRACE_GRAPH_PRINT_DURATION)
1172 seq_puts(s, " | | ");
1173 seq_puts(s, " | | | |\n");
1174}
1175
1176static void print_graph_headers(struct seq_file *s)
1177{
1178 print_graph_headers_flags(s, tracer_flags.val);
1179}
1180
1181void print_graph_headers_flags(struct seq_file *s, u32 flags)
1182{
1183 struct trace_iterator *iter = s->private;
1184 struct trace_array *tr = iter->tr;
1185
1186 if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
1187 return;
1188
1189 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT) {
1190 /* print nothing if the buffers are empty */
1191 if (trace_empty(iter))
1192 return;
1193
1194 print_trace_header(s, iter);
1195 }
1196
1197 __print_graph_headers_flags(tr, s, flags);
1198}
1199
1200void graph_trace_open(struct trace_iterator *iter)
1201{
1202 /* pid and depth on the last trace processed */
1203 struct fgraph_data *data;
1204 gfp_t gfpflags;
1205 int cpu;
1206
1207 iter->private = NULL;
1208
1209 /* We can be called in atomic context via ftrace_dump() */
1210 gfpflags = (in_atomic() || irqs_disabled()) ? GFP_ATOMIC : GFP_KERNEL;
1211
1212 data = kzalloc(sizeof(*data), gfpflags);
1213 if (!data)
1214 goto out_err;
1215
1216 data->cpu_data = alloc_percpu_gfp(struct fgraph_cpu_data, gfpflags);
1217 if (!data->cpu_data)
1218 goto out_err_free;
1219
1220 for_each_possible_cpu(cpu) {
1221 pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1222 int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1223 int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1224 int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1225
1226 *pid = -1;
1227 *depth = 0;
1228 *ignore = 0;
1229 *depth_irq = -1;
1230 }
1231
1232 iter->private = data;
1233
1234 return;
1235
1236 out_err_free:
1237 kfree(data);
1238 out_err:
1239 pr_warn("function graph tracer: not enough memory\n");
1240}
1241
1242void graph_trace_close(struct trace_iterator *iter)
1243{
1244 struct fgraph_data *data = iter->private;
1245
1246 if (data) {
1247 free_percpu(data->cpu_data);
1248 kfree(data);
1249 }
1250}
1251
1252static int
1253func_graph_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1254{
1255 if (bit == TRACE_GRAPH_PRINT_IRQS)
1256 ftrace_graph_skip_irqs = !set;
1257
1258 if (bit == TRACE_GRAPH_SLEEP_TIME)
1259 ftrace_graph_sleep_time_control(set);
1260
1261 if (bit == TRACE_GRAPH_GRAPH_TIME)
1262 ftrace_graph_graph_time_control(set);
1263
1264 return 0;
1265}
1266
1267static struct trace_event_functions graph_functions = {
1268 .trace = print_graph_function_event,
1269};
1270
1271static struct trace_event graph_trace_entry_event = {
1272 .type = TRACE_GRAPH_ENT,
1273 .funcs = &graph_functions,
1274};
1275
1276static struct trace_event graph_trace_ret_event = {
1277 .type = TRACE_GRAPH_RET,
1278 .funcs = &graph_functions
1279};
1280
1281static struct tracer graph_trace __tracer_data = {
1282 .name = "function_graph",
1283 .update_thresh = graph_trace_update_thresh,
1284 .open = graph_trace_open,
1285 .pipe_open = graph_trace_open,
1286 .close = graph_trace_close,
1287 .pipe_close = graph_trace_close,
1288 .init = graph_trace_init,
1289 .reset = graph_trace_reset,
1290 .print_line = print_graph_function,
1291 .print_header = print_graph_headers,
1292 .flags = &tracer_flags,
1293 .set_flag = func_graph_set_flag,
1294#ifdef CONFIG_FTRACE_SELFTEST
1295 .selftest = trace_selftest_startup_function_graph,
1296#endif
1297};
1298
1299
1300static ssize_t
1301graph_depth_write(struct file *filp, const char __user *ubuf, size_t cnt,
1302 loff_t *ppos)
1303{
1304 unsigned long val;
1305 int ret;
1306
1307 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1308 if (ret)
1309 return ret;
1310
1311 fgraph_max_depth = val;
1312
1313 *ppos += cnt;
1314
1315 return cnt;
1316}
1317
1318static ssize_t
1319graph_depth_read(struct file *filp, char __user *ubuf, size_t cnt,
1320 loff_t *ppos)
1321{
1322 char buf[15]; /* More than enough to hold UINT_MAX + "\n"*/
1323 int n;
1324
1325 n = sprintf(buf, "%d\n", fgraph_max_depth);
1326
1327 return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
1328}
1329
1330static const struct file_operations graph_depth_fops = {
1331 .open = tracing_open_generic,
1332 .write = graph_depth_write,
1333 .read = graph_depth_read,
1334 .llseek = generic_file_llseek,
1335};
1336
1337static __init int init_graph_tracefs(void)
1338{
1339 struct dentry *d_tracer;
1340
1341 d_tracer = tracing_init_dentry();
1342 if (IS_ERR(d_tracer))
1343 return 0;
1344
1345 trace_create_file("max_graph_depth", 0644, d_tracer,
1346 NULL, &graph_depth_fops);
1347
1348 return 0;
1349}
1350fs_initcall(init_graph_tracefs);
1351
1352static __init int init_graph_trace(void)
1353{
1354 max_bytes_for_cpu = snprintf(NULL, 0, "%u", nr_cpu_ids - 1);
1355
1356 if (!register_trace_event(&graph_trace_entry_event)) {
1357 pr_warn("Warning: could not register graph trace events\n");
1358 return 1;
1359 }
1360
1361 if (!register_trace_event(&graph_trace_ret_event)) {
1362 pr_warn("Warning: could not register graph trace events\n");
1363 return 1;
1364 }
1365
1366 return register_tracer(&graph_trace);
1367}
1368
1369core_initcall(init_graph_trace);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 *
4 * Function graph tracer.
5 * Copyright (c) 2008-2009 Frederic Weisbecker <fweisbec@gmail.com>
6 * Mostly borrowed from function tracer which
7 * is Copyright (c) Steven Rostedt <srostedt@redhat.com>
8 *
9 */
10#include <linux/uaccess.h>
11#include <linux/ftrace.h>
12#include <linux/interrupt.h>
13#include <linux/slab.h>
14#include <linux/fs.h>
15
16#include "trace.h"
17#include "trace_output.h"
18
19/* When set, irq functions will be ignored */
20static int ftrace_graph_skip_irqs;
21
22struct fgraph_cpu_data {
23 pid_t last_pid;
24 int depth;
25 int depth_irq;
26 int ignore;
27 unsigned long enter_funcs[FTRACE_RETFUNC_DEPTH];
28};
29
30struct fgraph_data {
31 struct fgraph_cpu_data __percpu *cpu_data;
32
33 /* Place to preserve last processed entry. */
34 struct ftrace_graph_ent_entry ent;
35 struct ftrace_graph_ret_entry ret;
36 int failed;
37 int cpu;
38};
39
40#define TRACE_GRAPH_INDENT 2
41
42unsigned int fgraph_max_depth;
43
44static struct tracer_opt trace_opts[] = {
45 /* Display overruns? (for self-debug purpose) */
46 { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
47 /* Display CPU ? */
48 { TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
49 /* Display Overhead ? */
50 { TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
51 /* Display proc name/pid */
52 { TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
53 /* Display duration of execution */
54 { TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
55 /* Display absolute time of an entry */
56 { TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
57 /* Display interrupts */
58 { TRACER_OPT(funcgraph-irqs, TRACE_GRAPH_PRINT_IRQS) },
59 /* Display function name after trailing } */
60 { TRACER_OPT(funcgraph-tail, TRACE_GRAPH_PRINT_TAIL) },
61#ifdef CONFIG_FUNCTION_GRAPH_RETVAL
62 /* Display function return value ? */
63 { TRACER_OPT(funcgraph-retval, TRACE_GRAPH_PRINT_RETVAL) },
64 /* Display function return value in hexadecimal format ? */
65 { TRACER_OPT(funcgraph-retval-hex, TRACE_GRAPH_PRINT_RETVAL_HEX) },
66#endif
67 /* Include sleep time (scheduled out) between entry and return */
68 { TRACER_OPT(sleep-time, TRACE_GRAPH_SLEEP_TIME) },
69
70#ifdef CONFIG_FUNCTION_PROFILER
71 /* Include time within nested functions */
72 { TRACER_OPT(graph-time, TRACE_GRAPH_GRAPH_TIME) },
73#endif
74
75 { } /* Empty entry */
76};
77
78static struct tracer_flags tracer_flags = {
79 /* Don't display overruns, proc, or tail by default */
80 .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
81 TRACE_GRAPH_PRINT_DURATION | TRACE_GRAPH_PRINT_IRQS |
82 TRACE_GRAPH_SLEEP_TIME | TRACE_GRAPH_GRAPH_TIME,
83 .opts = trace_opts
84};
85
86static struct trace_array *graph_array;
87
88/*
89 * DURATION column is being also used to display IRQ signs,
90 * following values are used by print_graph_irq and others
91 * to fill in space into DURATION column.
92 */
93enum {
94 FLAGS_FILL_FULL = 1 << TRACE_GRAPH_PRINT_FILL_SHIFT,
95 FLAGS_FILL_START = 2 << TRACE_GRAPH_PRINT_FILL_SHIFT,
96 FLAGS_FILL_END = 3 << TRACE_GRAPH_PRINT_FILL_SHIFT,
97};
98
99static void
100print_graph_duration(struct trace_array *tr, unsigned long long duration,
101 struct trace_seq *s, u32 flags);
102
103int __trace_graph_entry(struct trace_array *tr,
104 struct ftrace_graph_ent *trace,
105 unsigned int trace_ctx)
106{
107 struct trace_event_call *call = &event_funcgraph_entry;
108 struct ring_buffer_event *event;
109 struct trace_buffer *buffer = tr->array_buffer.buffer;
110 struct ftrace_graph_ent_entry *entry;
111
112 event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
113 sizeof(*entry), trace_ctx);
114 if (!event)
115 return 0;
116 entry = ring_buffer_event_data(event);
117 entry->graph_ent = *trace;
118 if (!call_filter_check_discard(call, entry, buffer, event))
119 trace_buffer_unlock_commit_nostack(buffer, event);
120
121 return 1;
122}
123
124static inline int ftrace_graph_ignore_irqs(void)
125{
126 if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
127 return 0;
128
129 return in_hardirq();
130}
131
132int trace_graph_entry(struct ftrace_graph_ent *trace)
133{
134 struct trace_array *tr = graph_array;
135 struct trace_array_cpu *data;
136 unsigned long flags;
137 unsigned int trace_ctx;
138 long disabled;
139 int ret;
140 int cpu;
141
142 if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT))
143 return 0;
144
145 /*
146 * Do not trace a function if it's filtered by set_graph_notrace.
147 * Make the index of ret stack negative to indicate that it should
148 * ignore further functions. But it needs its own ret stack entry
149 * to recover the original index in order to continue tracing after
150 * returning from the function.
151 */
152 if (ftrace_graph_notrace_addr(trace->func)) {
153 trace_recursion_set(TRACE_GRAPH_NOTRACE_BIT);
154 /*
155 * Need to return 1 to have the return called
156 * that will clear the NOTRACE bit.
157 */
158 return 1;
159 }
160
161 if (!ftrace_trace_task(tr))
162 return 0;
163
164 if (ftrace_graph_ignore_func(trace))
165 return 0;
166
167 if (ftrace_graph_ignore_irqs())
168 return 0;
169
170 /*
171 * Stop here if tracing_threshold is set. We only write function return
172 * events to the ring buffer.
173 */
174 if (tracing_thresh)
175 return 1;
176
177 local_irq_save(flags);
178 cpu = raw_smp_processor_id();
179 data = per_cpu_ptr(tr->array_buffer.data, cpu);
180 disabled = atomic_inc_return(&data->disabled);
181 if (likely(disabled == 1)) {
182 trace_ctx = tracing_gen_ctx_flags(flags);
183 ret = __trace_graph_entry(tr, trace, trace_ctx);
184 } else {
185 ret = 0;
186 }
187
188 atomic_dec(&data->disabled);
189 local_irq_restore(flags);
190
191 return ret;
192}
193
194static void
195__trace_graph_function(struct trace_array *tr,
196 unsigned long ip, unsigned int trace_ctx)
197{
198 u64 time = trace_clock_local();
199 struct ftrace_graph_ent ent = {
200 .func = ip,
201 .depth = 0,
202 };
203 struct ftrace_graph_ret ret = {
204 .func = ip,
205 .depth = 0,
206 .calltime = time,
207 .rettime = time,
208 };
209
210 __trace_graph_entry(tr, &ent, trace_ctx);
211 __trace_graph_return(tr, &ret, trace_ctx);
212}
213
214void
215trace_graph_function(struct trace_array *tr,
216 unsigned long ip, unsigned long parent_ip,
217 unsigned int trace_ctx)
218{
219 __trace_graph_function(tr, ip, trace_ctx);
220}
221
222void __trace_graph_return(struct trace_array *tr,
223 struct ftrace_graph_ret *trace,
224 unsigned int trace_ctx)
225{
226 struct trace_event_call *call = &event_funcgraph_exit;
227 struct ring_buffer_event *event;
228 struct trace_buffer *buffer = tr->array_buffer.buffer;
229 struct ftrace_graph_ret_entry *entry;
230
231 event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
232 sizeof(*entry), trace_ctx);
233 if (!event)
234 return;
235 entry = ring_buffer_event_data(event);
236 entry->ret = *trace;
237 if (!call_filter_check_discard(call, entry, buffer, event))
238 trace_buffer_unlock_commit_nostack(buffer, event);
239}
240
241void trace_graph_return(struct ftrace_graph_ret *trace)
242{
243 struct trace_array *tr = graph_array;
244 struct trace_array_cpu *data;
245 unsigned long flags;
246 unsigned int trace_ctx;
247 long disabled;
248 int cpu;
249
250 ftrace_graph_addr_finish(trace);
251
252 if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT)) {
253 trace_recursion_clear(TRACE_GRAPH_NOTRACE_BIT);
254 return;
255 }
256
257 local_irq_save(flags);
258 cpu = raw_smp_processor_id();
259 data = per_cpu_ptr(tr->array_buffer.data, cpu);
260 disabled = atomic_inc_return(&data->disabled);
261 if (likely(disabled == 1)) {
262 trace_ctx = tracing_gen_ctx_flags(flags);
263 __trace_graph_return(tr, trace, trace_ctx);
264 }
265 atomic_dec(&data->disabled);
266 local_irq_restore(flags);
267}
268
269void set_graph_array(struct trace_array *tr)
270{
271 graph_array = tr;
272
273 /* Make graph_array visible before we start tracing */
274
275 smp_mb();
276}
277
278static void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
279{
280 ftrace_graph_addr_finish(trace);
281
282 if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT)) {
283 trace_recursion_clear(TRACE_GRAPH_NOTRACE_BIT);
284 return;
285 }
286
287 if (tracing_thresh &&
288 (trace->rettime - trace->calltime < tracing_thresh))
289 return;
290 else
291 trace_graph_return(trace);
292}
293
294static struct fgraph_ops funcgraph_thresh_ops = {
295 .entryfunc = &trace_graph_entry,
296 .retfunc = &trace_graph_thresh_return,
297};
298
299static struct fgraph_ops funcgraph_ops = {
300 .entryfunc = &trace_graph_entry,
301 .retfunc = &trace_graph_return,
302};
303
304static int graph_trace_init(struct trace_array *tr)
305{
306 int ret;
307
308 set_graph_array(tr);
309 if (tracing_thresh)
310 ret = register_ftrace_graph(&funcgraph_thresh_ops);
311 else
312 ret = register_ftrace_graph(&funcgraph_ops);
313 if (ret)
314 return ret;
315 tracing_start_cmdline_record();
316
317 return 0;
318}
319
320static void graph_trace_reset(struct trace_array *tr)
321{
322 tracing_stop_cmdline_record();
323 if (tracing_thresh)
324 unregister_ftrace_graph(&funcgraph_thresh_ops);
325 else
326 unregister_ftrace_graph(&funcgraph_ops);
327}
328
329static int graph_trace_update_thresh(struct trace_array *tr)
330{
331 graph_trace_reset(tr);
332 return graph_trace_init(tr);
333}
334
335static int max_bytes_for_cpu;
336
337static void print_graph_cpu(struct trace_seq *s, int cpu)
338{
339 /*
340 * Start with a space character - to make it stand out
341 * to the right a bit when trace output is pasted into
342 * email:
343 */
344 trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
345}
346
347#define TRACE_GRAPH_PROCINFO_LENGTH 14
348
349static void print_graph_proc(struct trace_seq *s, pid_t pid)
350{
351 char comm[TASK_COMM_LEN];
352 /* sign + log10(MAX_INT) + '\0' */
353 char pid_str[11];
354 int spaces = 0;
355 int len;
356 int i;
357
358 trace_find_cmdline(pid, comm);
359 comm[7] = '\0';
360 sprintf(pid_str, "%d", pid);
361
362 /* 1 stands for the "-" character */
363 len = strlen(comm) + strlen(pid_str) + 1;
364
365 if (len < TRACE_GRAPH_PROCINFO_LENGTH)
366 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
367
368 /* First spaces to align center */
369 for (i = 0; i < spaces / 2; i++)
370 trace_seq_putc(s, ' ');
371
372 trace_seq_printf(s, "%s-%s", comm, pid_str);
373
374 /* Last spaces to align center */
375 for (i = 0; i < spaces - (spaces / 2); i++)
376 trace_seq_putc(s, ' ');
377}
378
379
380static void print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
381{
382 trace_seq_putc(s, ' ');
383 trace_print_lat_fmt(s, entry);
384 trace_seq_puts(s, " | ");
385}
386
387/* If the pid changed since the last trace, output this event */
388static void
389verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
390{
391 pid_t prev_pid;
392 pid_t *last_pid;
393
394 if (!data)
395 return;
396
397 last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
398
399 if (*last_pid == pid)
400 return;
401
402 prev_pid = *last_pid;
403 *last_pid = pid;
404
405 if (prev_pid == -1)
406 return;
407/*
408 * Context-switch trace line:
409
410 ------------------------------------------
411 | 1) migration/0--1 => sshd-1755
412 ------------------------------------------
413
414 */
415 trace_seq_puts(s, " ------------------------------------------\n");
416 print_graph_cpu(s, cpu);
417 print_graph_proc(s, prev_pid);
418 trace_seq_puts(s, " => ");
419 print_graph_proc(s, pid);
420 trace_seq_puts(s, "\n ------------------------------------------\n\n");
421}
422
423static struct ftrace_graph_ret_entry *
424get_return_for_leaf(struct trace_iterator *iter,
425 struct ftrace_graph_ent_entry *curr)
426{
427 struct fgraph_data *data = iter->private;
428 struct ring_buffer_iter *ring_iter = NULL;
429 struct ring_buffer_event *event;
430 struct ftrace_graph_ret_entry *next;
431
432 /*
433 * If the previous output failed to write to the seq buffer,
434 * then we just reuse the data from before.
435 */
436 if (data && data->failed) {
437 curr = &data->ent;
438 next = &data->ret;
439 } else {
440
441 ring_iter = trace_buffer_iter(iter, iter->cpu);
442
443 /* First peek to compare current entry and the next one */
444 if (ring_iter)
445 event = ring_buffer_iter_peek(ring_iter, NULL);
446 else {
447 /*
448 * We need to consume the current entry to see
449 * the next one.
450 */
451 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu,
452 NULL, NULL);
453 event = ring_buffer_peek(iter->array_buffer->buffer, iter->cpu,
454 NULL, NULL);
455 }
456
457 if (!event)
458 return NULL;
459
460 next = ring_buffer_event_data(event);
461
462 if (data) {
463 /*
464 * Save current and next entries for later reference
465 * if the output fails.
466 */
467 data->ent = *curr;
468 /*
469 * If the next event is not a return type, then
470 * we only care about what type it is. Otherwise we can
471 * safely copy the entire event.
472 */
473 if (next->ent.type == TRACE_GRAPH_RET)
474 data->ret = *next;
475 else
476 data->ret.ent.type = next->ent.type;
477 }
478 }
479
480 if (next->ent.type != TRACE_GRAPH_RET)
481 return NULL;
482
483 if (curr->ent.pid != next->ent.pid ||
484 curr->graph_ent.func != next->ret.func)
485 return NULL;
486
487 /* this is a leaf, now advance the iterator */
488 if (ring_iter)
489 ring_buffer_iter_advance(ring_iter);
490
491 return next;
492}
493
494static void print_graph_abs_time(u64 t, struct trace_seq *s)
495{
496 unsigned long usecs_rem;
497
498 usecs_rem = do_div(t, NSEC_PER_SEC);
499 usecs_rem /= 1000;
500
501 trace_seq_printf(s, "%5lu.%06lu | ",
502 (unsigned long)t, usecs_rem);
503}
504
505static void
506print_graph_rel_time(struct trace_iterator *iter, struct trace_seq *s)
507{
508 unsigned long long usecs;
509
510 usecs = iter->ts - iter->array_buffer->time_start;
511 do_div(usecs, NSEC_PER_USEC);
512
513 trace_seq_printf(s, "%9llu us | ", usecs);
514}
515
516static void
517print_graph_irq(struct trace_iterator *iter, unsigned long addr,
518 enum trace_type type, int cpu, pid_t pid, u32 flags)
519{
520 struct trace_array *tr = iter->tr;
521 struct trace_seq *s = &iter->seq;
522 struct trace_entry *ent = iter->ent;
523
524 if (addr < (unsigned long)__irqentry_text_start ||
525 addr >= (unsigned long)__irqentry_text_end)
526 return;
527
528 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
529 /* Absolute time */
530 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
531 print_graph_abs_time(iter->ts, s);
532
533 /* Relative time */
534 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
535 print_graph_rel_time(iter, s);
536
537 /* Cpu */
538 if (flags & TRACE_GRAPH_PRINT_CPU)
539 print_graph_cpu(s, cpu);
540
541 /* Proc */
542 if (flags & TRACE_GRAPH_PRINT_PROC) {
543 print_graph_proc(s, pid);
544 trace_seq_puts(s, " | ");
545 }
546
547 /* Latency format */
548 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
549 print_graph_lat_fmt(s, ent);
550 }
551
552 /* No overhead */
553 print_graph_duration(tr, 0, s, flags | FLAGS_FILL_START);
554
555 if (type == TRACE_GRAPH_ENT)
556 trace_seq_puts(s, "==========>");
557 else
558 trace_seq_puts(s, "<==========");
559
560 print_graph_duration(tr, 0, s, flags | FLAGS_FILL_END);
561 trace_seq_putc(s, '\n');
562}
563
564void
565trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
566{
567 unsigned long nsecs_rem = do_div(duration, 1000);
568 /* log10(ULONG_MAX) + '\0' */
569 char usecs_str[21];
570 char nsecs_str[5];
571 int len;
572 int i;
573
574 sprintf(usecs_str, "%lu", (unsigned long) duration);
575
576 /* Print msecs */
577 trace_seq_printf(s, "%s", usecs_str);
578
579 len = strlen(usecs_str);
580
581 /* Print nsecs (we don't want to exceed 7 numbers) */
582 if (len < 7) {
583 size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
584
585 snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
586 trace_seq_printf(s, ".%s", nsecs_str);
587 len += strlen(nsecs_str) + 1;
588 }
589
590 trace_seq_puts(s, " us ");
591
592 /* Print remaining spaces to fit the row's width */
593 for (i = len; i < 8; i++)
594 trace_seq_putc(s, ' ');
595}
596
597static void
598print_graph_duration(struct trace_array *tr, unsigned long long duration,
599 struct trace_seq *s, u32 flags)
600{
601 if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
602 !(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
603 return;
604
605 /* No real adata, just filling the column with spaces */
606 switch (flags & TRACE_GRAPH_PRINT_FILL_MASK) {
607 case FLAGS_FILL_FULL:
608 trace_seq_puts(s, " | ");
609 return;
610 case FLAGS_FILL_START:
611 trace_seq_puts(s, " ");
612 return;
613 case FLAGS_FILL_END:
614 trace_seq_puts(s, " |");
615 return;
616 }
617
618 /* Signal a overhead of time execution to the output */
619 if (flags & TRACE_GRAPH_PRINT_OVERHEAD)
620 trace_seq_printf(s, "%c ", trace_find_mark(duration));
621 else
622 trace_seq_puts(s, " ");
623
624 trace_print_graph_duration(duration, s);
625 trace_seq_puts(s, "| ");
626}
627
628#ifdef CONFIG_FUNCTION_GRAPH_RETVAL
629
630#define __TRACE_GRAPH_PRINT_RETVAL TRACE_GRAPH_PRINT_RETVAL
631
632static void print_graph_retval(struct trace_seq *s, unsigned long retval,
633 bool leaf, void *func, bool hex_format)
634{
635 unsigned long err_code = 0;
636
637 if (retval == 0 || hex_format)
638 goto done;
639
640 /* Check if the return value matches the negative format */
641 if (IS_ENABLED(CONFIG_64BIT) && (retval & BIT(31)) &&
642 (((u64)retval) >> 32) == 0) {
643 /* sign extension */
644 err_code = (unsigned long)(s32)retval;
645 } else {
646 err_code = retval;
647 }
648
649 if (!IS_ERR_VALUE(err_code))
650 err_code = 0;
651
652done:
653 if (leaf) {
654 if (hex_format || (err_code == 0))
655 trace_seq_printf(s, "%ps(); /* = 0x%lx */\n",
656 func, retval);
657 else
658 trace_seq_printf(s, "%ps(); /* = %ld */\n",
659 func, err_code);
660 } else {
661 if (hex_format || (err_code == 0))
662 trace_seq_printf(s, "} /* %ps = 0x%lx */\n",
663 func, retval);
664 else
665 trace_seq_printf(s, "} /* %ps = %ld */\n",
666 func, err_code);
667 }
668}
669
670#else
671
672#define __TRACE_GRAPH_PRINT_RETVAL 0
673
674#define print_graph_retval(_seq, _retval, _leaf, _func, _format) do {} while (0)
675
676#endif
677
678/* Case of a leaf function on its call entry */
679static enum print_line_t
680print_graph_entry_leaf(struct trace_iterator *iter,
681 struct ftrace_graph_ent_entry *entry,
682 struct ftrace_graph_ret_entry *ret_entry,
683 struct trace_seq *s, u32 flags)
684{
685 struct fgraph_data *data = iter->private;
686 struct trace_array *tr = iter->tr;
687 struct ftrace_graph_ret *graph_ret;
688 struct ftrace_graph_ent *call;
689 unsigned long long duration;
690 int cpu = iter->cpu;
691 int i;
692
693 graph_ret = &ret_entry->ret;
694 call = &entry->graph_ent;
695 duration = graph_ret->rettime - graph_ret->calltime;
696
697 if (data) {
698 struct fgraph_cpu_data *cpu_data;
699
700 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
701
702 /*
703 * Comments display at + 1 to depth. Since
704 * this is a leaf function, keep the comments
705 * equal to this depth.
706 */
707 cpu_data->depth = call->depth - 1;
708
709 /* No need to keep this function around for this depth */
710 if (call->depth < FTRACE_RETFUNC_DEPTH &&
711 !WARN_ON_ONCE(call->depth < 0))
712 cpu_data->enter_funcs[call->depth] = 0;
713 }
714
715 /* Overhead and duration */
716 print_graph_duration(tr, duration, s, flags);
717
718 /* Function */
719 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
720 trace_seq_putc(s, ' ');
721
722 /*
723 * Write out the function return value if the option function-retval is
724 * enabled.
725 */
726 if (flags & __TRACE_GRAPH_PRINT_RETVAL)
727 print_graph_retval(s, graph_ret->retval, true, (void *)call->func,
728 !!(flags & TRACE_GRAPH_PRINT_RETVAL_HEX));
729 else
730 trace_seq_printf(s, "%ps();\n", (void *)call->func);
731
732 print_graph_irq(iter, graph_ret->func, TRACE_GRAPH_RET,
733 cpu, iter->ent->pid, flags);
734
735 return trace_handle_return(s);
736}
737
738static enum print_line_t
739print_graph_entry_nested(struct trace_iterator *iter,
740 struct ftrace_graph_ent_entry *entry,
741 struct trace_seq *s, int cpu, u32 flags)
742{
743 struct ftrace_graph_ent *call = &entry->graph_ent;
744 struct fgraph_data *data = iter->private;
745 struct trace_array *tr = iter->tr;
746 int i;
747
748 if (data) {
749 struct fgraph_cpu_data *cpu_data;
750 int cpu = iter->cpu;
751
752 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
753 cpu_data->depth = call->depth;
754
755 /* Save this function pointer to see if the exit matches */
756 if (call->depth < FTRACE_RETFUNC_DEPTH &&
757 !WARN_ON_ONCE(call->depth < 0))
758 cpu_data->enter_funcs[call->depth] = call->func;
759 }
760
761 /* No time */
762 print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
763
764 /* Function */
765 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
766 trace_seq_putc(s, ' ');
767
768 trace_seq_printf(s, "%ps() {\n", (void *)call->func);
769
770 if (trace_seq_has_overflowed(s))
771 return TRACE_TYPE_PARTIAL_LINE;
772
773 /*
774 * we already consumed the current entry to check the next one
775 * and see if this is a leaf.
776 */
777 return TRACE_TYPE_NO_CONSUME;
778}
779
780static void
781print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
782 int type, unsigned long addr, u32 flags)
783{
784 struct fgraph_data *data = iter->private;
785 struct trace_entry *ent = iter->ent;
786 struct trace_array *tr = iter->tr;
787 int cpu = iter->cpu;
788
789 /* Pid */
790 verif_pid(s, ent->pid, cpu, data);
791
792 if (type)
793 /* Interrupt */
794 print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
795
796 if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
797 return;
798
799 /* Absolute time */
800 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
801 print_graph_abs_time(iter->ts, s);
802
803 /* Relative time */
804 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
805 print_graph_rel_time(iter, s);
806
807 /* Cpu */
808 if (flags & TRACE_GRAPH_PRINT_CPU)
809 print_graph_cpu(s, cpu);
810
811 /* Proc */
812 if (flags & TRACE_GRAPH_PRINT_PROC) {
813 print_graph_proc(s, ent->pid);
814 trace_seq_puts(s, " | ");
815 }
816
817 /* Latency format */
818 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
819 print_graph_lat_fmt(s, ent);
820
821 return;
822}
823
824/*
825 * Entry check for irq code
826 *
827 * returns 1 if
828 * - we are inside irq code
829 * - we just entered irq code
830 *
831 * returns 0 if
832 * - funcgraph-interrupts option is set
833 * - we are not inside irq code
834 */
835static int
836check_irq_entry(struct trace_iterator *iter, u32 flags,
837 unsigned long addr, int depth)
838{
839 int cpu = iter->cpu;
840 int *depth_irq;
841 struct fgraph_data *data = iter->private;
842
843 /*
844 * If we are either displaying irqs, or we got called as
845 * a graph event and private data does not exist,
846 * then we bypass the irq check.
847 */
848 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
849 (!data))
850 return 0;
851
852 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
853
854 /*
855 * We are inside the irq code
856 */
857 if (*depth_irq >= 0)
858 return 1;
859
860 if ((addr < (unsigned long)__irqentry_text_start) ||
861 (addr >= (unsigned long)__irqentry_text_end))
862 return 0;
863
864 /*
865 * We are entering irq code.
866 */
867 *depth_irq = depth;
868 return 1;
869}
870
871/*
872 * Return check for irq code
873 *
874 * returns 1 if
875 * - we are inside irq code
876 * - we just left irq code
877 *
878 * returns 0 if
879 * - funcgraph-interrupts option is set
880 * - we are not inside irq code
881 */
882static int
883check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
884{
885 int cpu = iter->cpu;
886 int *depth_irq;
887 struct fgraph_data *data = iter->private;
888
889 /*
890 * If we are either displaying irqs, or we got called as
891 * a graph event and private data does not exist,
892 * then we bypass the irq check.
893 */
894 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
895 (!data))
896 return 0;
897
898 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
899
900 /*
901 * We are not inside the irq code.
902 */
903 if (*depth_irq == -1)
904 return 0;
905
906 /*
907 * We are inside the irq code, and this is returning entry.
908 * Let's not trace it and clear the entry depth, since
909 * we are out of irq code.
910 *
911 * This condition ensures that we 'leave the irq code' once
912 * we are out of the entry depth. Thus protecting us from
913 * the RETURN entry loss.
914 */
915 if (*depth_irq >= depth) {
916 *depth_irq = -1;
917 return 1;
918 }
919
920 /*
921 * We are inside the irq code, and this is not the entry.
922 */
923 return 1;
924}
925
926static enum print_line_t
927print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
928 struct trace_iterator *iter, u32 flags)
929{
930 struct fgraph_data *data = iter->private;
931 struct ftrace_graph_ent *call = &field->graph_ent;
932 struct ftrace_graph_ret_entry *leaf_ret;
933 static enum print_line_t ret;
934 int cpu = iter->cpu;
935
936 if (check_irq_entry(iter, flags, call->func, call->depth))
937 return TRACE_TYPE_HANDLED;
938
939 print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags);
940
941 leaf_ret = get_return_for_leaf(iter, field);
942 if (leaf_ret)
943 ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
944 else
945 ret = print_graph_entry_nested(iter, field, s, cpu, flags);
946
947 if (data) {
948 /*
949 * If we failed to write our output, then we need to make
950 * note of it. Because we already consumed our entry.
951 */
952 if (s->full) {
953 data->failed = 1;
954 data->cpu = cpu;
955 } else
956 data->failed = 0;
957 }
958
959 return ret;
960}
961
962static enum print_line_t
963print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
964 struct trace_entry *ent, struct trace_iterator *iter,
965 u32 flags)
966{
967 unsigned long long duration = trace->rettime - trace->calltime;
968 struct fgraph_data *data = iter->private;
969 struct trace_array *tr = iter->tr;
970 pid_t pid = ent->pid;
971 int cpu = iter->cpu;
972 int func_match = 1;
973 int i;
974
975 if (check_irq_return(iter, flags, trace->depth))
976 return TRACE_TYPE_HANDLED;
977
978 if (data) {
979 struct fgraph_cpu_data *cpu_data;
980 int cpu = iter->cpu;
981
982 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
983
984 /*
985 * Comments display at + 1 to depth. This is the
986 * return from a function, we now want the comments
987 * to display at the same level of the bracket.
988 */
989 cpu_data->depth = trace->depth - 1;
990
991 if (trace->depth < FTRACE_RETFUNC_DEPTH &&
992 !WARN_ON_ONCE(trace->depth < 0)) {
993 if (cpu_data->enter_funcs[trace->depth] != trace->func)
994 func_match = 0;
995 cpu_data->enter_funcs[trace->depth] = 0;
996 }
997 }
998
999 print_graph_prologue(iter, s, 0, 0, flags);
1000
1001 /* Overhead and duration */
1002 print_graph_duration(tr, duration, s, flags);
1003
1004 /* Closing brace */
1005 for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++)
1006 trace_seq_putc(s, ' ');
1007
1008 /*
1009 * Always write out the function name and its return value if the
1010 * function-retval option is enabled.
1011 */
1012 if (flags & __TRACE_GRAPH_PRINT_RETVAL) {
1013 print_graph_retval(s, trace->retval, false, (void *)trace->func,
1014 !!(flags & TRACE_GRAPH_PRINT_RETVAL_HEX));
1015 } else {
1016 /*
1017 * If the return function does not have a matching entry,
1018 * then the entry was lost. Instead of just printing
1019 * the '}' and letting the user guess what function this
1020 * belongs to, write out the function name. Always do
1021 * that if the funcgraph-tail option is enabled.
1022 */
1023 if (func_match && !(flags & TRACE_GRAPH_PRINT_TAIL))
1024 trace_seq_puts(s, "}\n");
1025 else
1026 trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
1027 }
1028
1029 /* Overrun */
1030 if (flags & TRACE_GRAPH_PRINT_OVERRUN)
1031 trace_seq_printf(s, " (Overruns: %u)\n",
1032 trace->overrun);
1033
1034 print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
1035 cpu, pid, flags);
1036
1037 return trace_handle_return(s);
1038}
1039
1040static enum print_line_t
1041print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1042 struct trace_iterator *iter, u32 flags)
1043{
1044 struct trace_array *tr = iter->tr;
1045 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
1046 struct fgraph_data *data = iter->private;
1047 struct trace_event *event;
1048 int depth = 0;
1049 int ret;
1050 int i;
1051
1052 if (data)
1053 depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1054
1055 print_graph_prologue(iter, s, 0, 0, flags);
1056
1057 /* No time */
1058 print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
1059
1060 /* Indentation */
1061 if (depth > 0)
1062 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++)
1063 trace_seq_putc(s, ' ');
1064
1065 /* The comment */
1066 trace_seq_puts(s, "/* ");
1067
1068 switch (iter->ent->type) {
1069 case TRACE_BPUTS:
1070 ret = trace_print_bputs_msg_only(iter);
1071 if (ret != TRACE_TYPE_HANDLED)
1072 return ret;
1073 break;
1074 case TRACE_BPRINT:
1075 ret = trace_print_bprintk_msg_only(iter);
1076 if (ret != TRACE_TYPE_HANDLED)
1077 return ret;
1078 break;
1079 case TRACE_PRINT:
1080 ret = trace_print_printk_msg_only(iter);
1081 if (ret != TRACE_TYPE_HANDLED)
1082 return ret;
1083 break;
1084 default:
1085 event = ftrace_find_event(ent->type);
1086 if (!event)
1087 return TRACE_TYPE_UNHANDLED;
1088
1089 ret = event->funcs->trace(iter, sym_flags, event);
1090 if (ret != TRACE_TYPE_HANDLED)
1091 return ret;
1092 }
1093
1094 if (trace_seq_has_overflowed(s))
1095 goto out;
1096
1097 /* Strip ending newline */
1098 if (s->buffer[s->seq.len - 1] == '\n') {
1099 s->buffer[s->seq.len - 1] = '\0';
1100 s->seq.len--;
1101 }
1102
1103 trace_seq_puts(s, " */\n");
1104 out:
1105 return trace_handle_return(s);
1106}
1107
1108
1109enum print_line_t
1110print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1111{
1112 struct ftrace_graph_ent_entry *field;
1113 struct fgraph_data *data = iter->private;
1114 struct trace_entry *entry = iter->ent;
1115 struct trace_seq *s = &iter->seq;
1116 int cpu = iter->cpu;
1117 int ret;
1118
1119 if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1120 per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1121 return TRACE_TYPE_HANDLED;
1122 }
1123
1124 /*
1125 * If the last output failed, there's a possibility we need
1126 * to print out the missing entry which would never go out.
1127 */
1128 if (data && data->failed) {
1129 field = &data->ent;
1130 iter->cpu = data->cpu;
1131 ret = print_graph_entry(field, s, iter, flags);
1132 if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1133 per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1134 ret = TRACE_TYPE_NO_CONSUME;
1135 }
1136 iter->cpu = cpu;
1137 return ret;
1138 }
1139
1140 switch (entry->type) {
1141 case TRACE_GRAPH_ENT: {
1142 /*
1143 * print_graph_entry() may consume the current event,
1144 * thus @field may become invalid, so we need to save it.
1145 * sizeof(struct ftrace_graph_ent_entry) is very small,
1146 * it can be safely saved at the stack.
1147 */
1148 struct ftrace_graph_ent_entry saved;
1149 trace_assign_type(field, entry);
1150 saved = *field;
1151 return print_graph_entry(&saved, s, iter, flags);
1152 }
1153 case TRACE_GRAPH_RET: {
1154 struct ftrace_graph_ret_entry *field;
1155 trace_assign_type(field, entry);
1156 return print_graph_return(&field->ret, s, entry, iter, flags);
1157 }
1158 case TRACE_STACK:
1159 case TRACE_FN:
1160 /* dont trace stack and functions as comments */
1161 return TRACE_TYPE_UNHANDLED;
1162
1163 default:
1164 return print_graph_comment(s, entry, iter, flags);
1165 }
1166
1167 return TRACE_TYPE_HANDLED;
1168}
1169
1170static enum print_line_t
1171print_graph_function(struct trace_iterator *iter)
1172{
1173 return print_graph_function_flags(iter, tracer_flags.val);
1174}
1175
1176static enum print_line_t
1177print_graph_function_event(struct trace_iterator *iter, int flags,
1178 struct trace_event *event)
1179{
1180 return print_graph_function(iter);
1181}
1182
1183static void print_lat_header(struct seq_file *s, u32 flags)
1184{
1185 static const char spaces[] = " " /* 16 spaces */
1186 " " /* 4 spaces */
1187 " "; /* 17 spaces */
1188 int size = 0;
1189
1190 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1191 size += 16;
1192 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1193 size += 16;
1194 if (flags & TRACE_GRAPH_PRINT_CPU)
1195 size += 4;
1196 if (flags & TRACE_GRAPH_PRINT_PROC)
1197 size += 17;
1198
1199 seq_printf(s, "#%.*s _-----=> irqs-off \n", size, spaces);
1200 seq_printf(s, "#%.*s / _----=> need-resched \n", size, spaces);
1201 seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1202 seq_printf(s, "#%.*s|| / _--=> preempt-depth \n", size, spaces);
1203 seq_printf(s, "#%.*s||| / \n", size, spaces);
1204}
1205
1206static void __print_graph_headers_flags(struct trace_array *tr,
1207 struct seq_file *s, u32 flags)
1208{
1209 int lat = tr->trace_flags & TRACE_ITER_LATENCY_FMT;
1210
1211 if (lat)
1212 print_lat_header(s, flags);
1213
1214 /* 1st line */
1215 seq_putc(s, '#');
1216 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1217 seq_puts(s, " TIME ");
1218 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1219 seq_puts(s, " REL TIME ");
1220 if (flags & TRACE_GRAPH_PRINT_CPU)
1221 seq_puts(s, " CPU");
1222 if (flags & TRACE_GRAPH_PRINT_PROC)
1223 seq_puts(s, " TASK/PID ");
1224 if (lat)
1225 seq_puts(s, "|||| ");
1226 if (flags & TRACE_GRAPH_PRINT_DURATION)
1227 seq_puts(s, " DURATION ");
1228 seq_puts(s, " FUNCTION CALLS\n");
1229
1230 /* 2nd line */
1231 seq_putc(s, '#');
1232 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1233 seq_puts(s, " | ");
1234 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1235 seq_puts(s, " | ");
1236 if (flags & TRACE_GRAPH_PRINT_CPU)
1237 seq_puts(s, " | ");
1238 if (flags & TRACE_GRAPH_PRINT_PROC)
1239 seq_puts(s, " | | ");
1240 if (lat)
1241 seq_puts(s, "|||| ");
1242 if (flags & TRACE_GRAPH_PRINT_DURATION)
1243 seq_puts(s, " | | ");
1244 seq_puts(s, " | | | |\n");
1245}
1246
1247static void print_graph_headers(struct seq_file *s)
1248{
1249 print_graph_headers_flags(s, tracer_flags.val);
1250}
1251
1252void print_graph_headers_flags(struct seq_file *s, u32 flags)
1253{
1254 struct trace_iterator *iter = s->private;
1255 struct trace_array *tr = iter->tr;
1256
1257 if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
1258 return;
1259
1260 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT) {
1261 /* print nothing if the buffers are empty */
1262 if (trace_empty(iter))
1263 return;
1264
1265 print_trace_header(s, iter);
1266 }
1267
1268 __print_graph_headers_flags(tr, s, flags);
1269}
1270
1271void graph_trace_open(struct trace_iterator *iter)
1272{
1273 /* pid and depth on the last trace processed */
1274 struct fgraph_data *data;
1275 gfp_t gfpflags;
1276 int cpu;
1277
1278 iter->private = NULL;
1279
1280 /* We can be called in atomic context via ftrace_dump() */
1281 gfpflags = (in_atomic() || irqs_disabled()) ? GFP_ATOMIC : GFP_KERNEL;
1282
1283 data = kzalloc(sizeof(*data), gfpflags);
1284 if (!data)
1285 goto out_err;
1286
1287 data->cpu_data = alloc_percpu_gfp(struct fgraph_cpu_data, gfpflags);
1288 if (!data->cpu_data)
1289 goto out_err_free;
1290
1291 for_each_possible_cpu(cpu) {
1292 pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1293 int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1294 int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1295 int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1296
1297 *pid = -1;
1298 *depth = 0;
1299 *ignore = 0;
1300 *depth_irq = -1;
1301 }
1302
1303 iter->private = data;
1304
1305 return;
1306
1307 out_err_free:
1308 kfree(data);
1309 out_err:
1310 pr_warn("function graph tracer: not enough memory\n");
1311}
1312
1313void graph_trace_close(struct trace_iterator *iter)
1314{
1315 struct fgraph_data *data = iter->private;
1316
1317 if (data) {
1318 free_percpu(data->cpu_data);
1319 kfree(data);
1320 }
1321}
1322
1323static int
1324func_graph_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1325{
1326 if (bit == TRACE_GRAPH_PRINT_IRQS)
1327 ftrace_graph_skip_irqs = !set;
1328
1329 if (bit == TRACE_GRAPH_SLEEP_TIME)
1330 ftrace_graph_sleep_time_control(set);
1331
1332 if (bit == TRACE_GRAPH_GRAPH_TIME)
1333 ftrace_graph_graph_time_control(set);
1334
1335 return 0;
1336}
1337
1338static struct trace_event_functions graph_functions = {
1339 .trace = print_graph_function_event,
1340};
1341
1342static struct trace_event graph_trace_entry_event = {
1343 .type = TRACE_GRAPH_ENT,
1344 .funcs = &graph_functions,
1345};
1346
1347static struct trace_event graph_trace_ret_event = {
1348 .type = TRACE_GRAPH_RET,
1349 .funcs = &graph_functions
1350};
1351
1352static struct tracer graph_trace __tracer_data = {
1353 .name = "function_graph",
1354 .update_thresh = graph_trace_update_thresh,
1355 .open = graph_trace_open,
1356 .pipe_open = graph_trace_open,
1357 .close = graph_trace_close,
1358 .pipe_close = graph_trace_close,
1359 .init = graph_trace_init,
1360 .reset = graph_trace_reset,
1361 .print_line = print_graph_function,
1362 .print_header = print_graph_headers,
1363 .flags = &tracer_flags,
1364 .set_flag = func_graph_set_flag,
1365#ifdef CONFIG_FTRACE_SELFTEST
1366 .selftest = trace_selftest_startup_function_graph,
1367#endif
1368};
1369
1370
1371static ssize_t
1372graph_depth_write(struct file *filp, const char __user *ubuf, size_t cnt,
1373 loff_t *ppos)
1374{
1375 unsigned long val;
1376 int ret;
1377
1378 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1379 if (ret)
1380 return ret;
1381
1382 fgraph_max_depth = val;
1383
1384 *ppos += cnt;
1385
1386 return cnt;
1387}
1388
1389static ssize_t
1390graph_depth_read(struct file *filp, char __user *ubuf, size_t cnt,
1391 loff_t *ppos)
1392{
1393 char buf[15]; /* More than enough to hold UINT_MAX + "\n"*/
1394 int n;
1395
1396 n = sprintf(buf, "%d\n", fgraph_max_depth);
1397
1398 return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
1399}
1400
1401static const struct file_operations graph_depth_fops = {
1402 .open = tracing_open_generic,
1403 .write = graph_depth_write,
1404 .read = graph_depth_read,
1405 .llseek = generic_file_llseek,
1406};
1407
1408static __init int init_graph_tracefs(void)
1409{
1410 int ret;
1411
1412 ret = tracing_init_dentry();
1413 if (ret)
1414 return 0;
1415
1416 trace_create_file("max_graph_depth", TRACE_MODE_WRITE, NULL,
1417 NULL, &graph_depth_fops);
1418
1419 return 0;
1420}
1421fs_initcall(init_graph_tracefs);
1422
1423static __init int init_graph_trace(void)
1424{
1425 max_bytes_for_cpu = snprintf(NULL, 0, "%u", nr_cpu_ids - 1);
1426
1427 if (!register_trace_event(&graph_trace_entry_event)) {
1428 pr_warn("Warning: could not register graph trace events\n");
1429 return 1;
1430 }
1431
1432 if (!register_trace_event(&graph_trace_ret_event)) {
1433 pr_warn("Warning: could not register graph trace events\n");
1434 return 1;
1435 }
1436
1437 return register_tracer(&graph_trace);
1438}
1439
1440core_initcall(init_graph_trace);