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#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);
1/*
2 *
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>
7 *
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>
14
15#include "trace.h"
16#include "trace_output.h"
17
18/* When set, irq functions will be ignored */
19static int ftrace_graph_skip_irqs;
20
21struct fgraph_cpu_data {
22 pid_t last_pid;
23 int depth;
24 int depth_irq;
25 int ignore;
26 unsigned long enter_funcs[FTRACE_RETFUNC_DEPTH];
27};
28
29struct fgraph_data {
30 struct fgraph_cpu_data __percpu *cpu_data;
31
32 /* Place to preserve last processed entry. */
33 struct ftrace_graph_ent_entry ent;
34 struct ftrace_graph_ret_entry ret;
35 int failed;
36 int cpu;
37};
38
39#define TRACE_GRAPH_INDENT 2
40
41/* Flag options */
42#define TRACE_GRAPH_PRINT_OVERRUN 0x1
43#define TRACE_GRAPH_PRINT_CPU 0x2
44#define TRACE_GRAPH_PRINT_OVERHEAD 0x4
45#define TRACE_GRAPH_PRINT_PROC 0x8
46#define TRACE_GRAPH_PRINT_DURATION 0x10
47#define TRACE_GRAPH_PRINT_ABS_TIME 0x20
48#define TRACE_GRAPH_PRINT_IRQS 0x40
49
50static struct tracer_opt trace_opts[] = {
51 /* Display overruns? (for self-debug purpose) */
52 { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
53 /* Display CPU ? */
54 { TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
55 /* Display Overhead ? */
56 { TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
57 /* Display proc name/pid */
58 { TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
59 /* Display duration of execution */
60 { TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
61 /* Display absolute time of an entry */
62 { TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
63 /* Display interrupts */
64 { TRACER_OPT(funcgraph-irqs, TRACE_GRAPH_PRINT_IRQS) },
65 { } /* Empty entry */
66};
67
68static struct tracer_flags tracer_flags = {
69 /* Don't display overruns and proc by default */
70 .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
71 TRACE_GRAPH_PRINT_DURATION | TRACE_GRAPH_PRINT_IRQS,
72 .opts = trace_opts
73};
74
75static struct trace_array *graph_array;
76
77/*
78 * DURATION column is being also used to display IRQ signs,
79 * following values are used by print_graph_irq and others
80 * to fill in space into DURATION column.
81 */
82enum {
83 DURATION_FILL_FULL = -1,
84 DURATION_FILL_START = -2,
85 DURATION_FILL_END = -3,
86};
87
88static enum print_line_t
89print_graph_duration(unsigned long long duration, struct trace_seq *s,
90 u32 flags);
91
92/* Add a function return address to the trace stack on thread info.*/
93int
94ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
95 unsigned long frame_pointer)
96{
97 unsigned long long calltime;
98 int index;
99
100 if (!current->ret_stack)
101 return -EBUSY;
102
103 /*
104 * We must make sure the ret_stack is tested before we read
105 * anything else.
106 */
107 smp_rmb();
108
109 /* The return trace stack is full */
110 if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
111 atomic_inc(¤t->trace_overrun);
112 return -EBUSY;
113 }
114
115 calltime = trace_clock_local();
116
117 index = ++current->curr_ret_stack;
118 barrier();
119 current->ret_stack[index].ret = ret;
120 current->ret_stack[index].func = func;
121 current->ret_stack[index].calltime = calltime;
122 current->ret_stack[index].subtime = 0;
123 current->ret_stack[index].fp = frame_pointer;
124 *depth = index;
125
126 return 0;
127}
128
129/* Retrieve a function return address to the trace stack on thread info.*/
130static void
131ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
132 unsigned long frame_pointer)
133{
134 int index;
135
136 index = current->curr_ret_stack;
137
138 if (unlikely(index < 0)) {
139 ftrace_graph_stop();
140 WARN_ON(1);
141 /* Might as well panic, otherwise we have no where to go */
142 *ret = (unsigned long)panic;
143 return;
144 }
145
146#ifdef CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST
147 /*
148 * The arch may choose to record the frame pointer used
149 * and check it here to make sure that it is what we expect it
150 * to be. If gcc does not set the place holder of the return
151 * address in the frame pointer, and does a copy instead, then
152 * the function graph trace will fail. This test detects this
153 * case.
154 *
155 * Currently, x86_32 with optimize for size (-Os) makes the latest
156 * gcc do the above.
157 */
158 if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
159 ftrace_graph_stop();
160 WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
161 " from func %ps return to %lx\n",
162 current->ret_stack[index].fp,
163 frame_pointer,
164 (void *)current->ret_stack[index].func,
165 current->ret_stack[index].ret);
166 *ret = (unsigned long)panic;
167 return;
168 }
169#endif
170
171 *ret = current->ret_stack[index].ret;
172 trace->func = current->ret_stack[index].func;
173 trace->calltime = current->ret_stack[index].calltime;
174 trace->overrun = atomic_read(¤t->trace_overrun);
175 trace->depth = index;
176}
177
178/*
179 * Send the trace to the ring-buffer.
180 * @return the original return address.
181 */
182unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
183{
184 struct ftrace_graph_ret trace;
185 unsigned long ret;
186
187 ftrace_pop_return_trace(&trace, &ret, frame_pointer);
188 trace.rettime = trace_clock_local();
189 ftrace_graph_return(&trace);
190 barrier();
191 current->curr_ret_stack--;
192
193 if (unlikely(!ret)) {
194 ftrace_graph_stop();
195 WARN_ON(1);
196 /* Might as well panic. What else to do? */
197 ret = (unsigned long)panic;
198 }
199
200 return ret;
201}
202
203int __trace_graph_entry(struct trace_array *tr,
204 struct ftrace_graph_ent *trace,
205 unsigned long flags,
206 int pc)
207{
208 struct ftrace_event_call *call = &event_funcgraph_entry;
209 struct ring_buffer_event *event;
210 struct ring_buffer *buffer = tr->buffer;
211 struct ftrace_graph_ent_entry *entry;
212
213 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
214 return 0;
215
216 event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
217 sizeof(*entry), flags, pc);
218 if (!event)
219 return 0;
220 entry = ring_buffer_event_data(event);
221 entry->graph_ent = *trace;
222 if (!filter_current_check_discard(buffer, call, entry, event))
223 ring_buffer_unlock_commit(buffer, event);
224
225 return 1;
226}
227
228static inline int ftrace_graph_ignore_irqs(void)
229{
230 if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
231 return 0;
232
233 return in_irq();
234}
235
236int trace_graph_entry(struct ftrace_graph_ent *trace)
237{
238 struct trace_array *tr = graph_array;
239 struct trace_array_cpu *data;
240 unsigned long flags;
241 long disabled;
242 int ret;
243 int cpu;
244 int pc;
245
246 if (!ftrace_trace_task(current))
247 return 0;
248
249 /* trace it when it is-nested-in or is a function enabled. */
250 if (!(trace->depth || ftrace_graph_addr(trace->func)) ||
251 ftrace_graph_ignore_irqs())
252 return 0;
253
254 local_irq_save(flags);
255 cpu = raw_smp_processor_id();
256 data = tr->data[cpu];
257 disabled = atomic_inc_return(&data->disabled);
258 if (likely(disabled == 1)) {
259 pc = preempt_count();
260 ret = __trace_graph_entry(tr, trace, flags, pc);
261 } else {
262 ret = 0;
263 }
264
265 atomic_dec(&data->disabled);
266 local_irq_restore(flags);
267
268 return ret;
269}
270
271int trace_graph_thresh_entry(struct ftrace_graph_ent *trace)
272{
273 if (tracing_thresh)
274 return 1;
275 else
276 return trace_graph_entry(trace);
277}
278
279static void
280__trace_graph_function(struct trace_array *tr,
281 unsigned long ip, unsigned long flags, int pc)
282{
283 u64 time = trace_clock_local();
284 struct ftrace_graph_ent ent = {
285 .func = ip,
286 .depth = 0,
287 };
288 struct ftrace_graph_ret ret = {
289 .func = ip,
290 .depth = 0,
291 .calltime = time,
292 .rettime = time,
293 };
294
295 __trace_graph_entry(tr, &ent, flags, pc);
296 __trace_graph_return(tr, &ret, flags, pc);
297}
298
299void
300trace_graph_function(struct trace_array *tr,
301 unsigned long ip, unsigned long parent_ip,
302 unsigned long flags, int pc)
303{
304 __trace_graph_function(tr, ip, flags, pc);
305}
306
307void __trace_graph_return(struct trace_array *tr,
308 struct ftrace_graph_ret *trace,
309 unsigned long flags,
310 int pc)
311{
312 struct ftrace_event_call *call = &event_funcgraph_exit;
313 struct ring_buffer_event *event;
314 struct ring_buffer *buffer = tr->buffer;
315 struct ftrace_graph_ret_entry *entry;
316
317 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
318 return;
319
320 event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
321 sizeof(*entry), flags, pc);
322 if (!event)
323 return;
324 entry = ring_buffer_event_data(event);
325 entry->ret = *trace;
326 if (!filter_current_check_discard(buffer, call, entry, event))
327 ring_buffer_unlock_commit(buffer, event);
328}
329
330void trace_graph_return(struct ftrace_graph_ret *trace)
331{
332 struct trace_array *tr = graph_array;
333 struct trace_array_cpu *data;
334 unsigned long flags;
335 long disabled;
336 int cpu;
337 int pc;
338
339 local_irq_save(flags);
340 cpu = raw_smp_processor_id();
341 data = tr->data[cpu];
342 disabled = atomic_inc_return(&data->disabled);
343 if (likely(disabled == 1)) {
344 pc = preempt_count();
345 __trace_graph_return(tr, trace, flags, pc);
346 }
347 atomic_dec(&data->disabled);
348 local_irq_restore(flags);
349}
350
351void set_graph_array(struct trace_array *tr)
352{
353 graph_array = tr;
354
355 /* Make graph_array visible before we start tracing */
356
357 smp_mb();
358}
359
360void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
361{
362 if (tracing_thresh &&
363 (trace->rettime - trace->calltime < tracing_thresh))
364 return;
365 else
366 trace_graph_return(trace);
367}
368
369static int graph_trace_init(struct trace_array *tr)
370{
371 int ret;
372
373 set_graph_array(tr);
374 if (tracing_thresh)
375 ret = register_ftrace_graph(&trace_graph_thresh_return,
376 &trace_graph_thresh_entry);
377 else
378 ret = register_ftrace_graph(&trace_graph_return,
379 &trace_graph_entry);
380 if (ret)
381 return ret;
382 tracing_start_cmdline_record();
383
384 return 0;
385}
386
387static void graph_trace_reset(struct trace_array *tr)
388{
389 tracing_stop_cmdline_record();
390 unregister_ftrace_graph();
391}
392
393static int max_bytes_for_cpu;
394
395static enum print_line_t
396print_graph_cpu(struct trace_seq *s, int cpu)
397{
398 int ret;
399
400 /*
401 * Start with a space character - to make it stand out
402 * to the right a bit when trace output is pasted into
403 * email:
404 */
405 ret = trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
406 if (!ret)
407 return TRACE_TYPE_PARTIAL_LINE;
408
409 return TRACE_TYPE_HANDLED;
410}
411
412#define TRACE_GRAPH_PROCINFO_LENGTH 14
413
414static enum print_line_t
415print_graph_proc(struct trace_seq *s, pid_t pid)
416{
417 char comm[TASK_COMM_LEN];
418 /* sign + log10(MAX_INT) + '\0' */
419 char pid_str[11];
420 int spaces = 0;
421 int ret;
422 int len;
423 int i;
424
425 trace_find_cmdline(pid, comm);
426 comm[7] = '\0';
427 sprintf(pid_str, "%d", pid);
428
429 /* 1 stands for the "-" character */
430 len = strlen(comm) + strlen(pid_str) + 1;
431
432 if (len < TRACE_GRAPH_PROCINFO_LENGTH)
433 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
434
435 /* First spaces to align center */
436 for (i = 0; i < spaces / 2; i++) {
437 ret = trace_seq_printf(s, " ");
438 if (!ret)
439 return TRACE_TYPE_PARTIAL_LINE;
440 }
441
442 ret = trace_seq_printf(s, "%s-%s", comm, pid_str);
443 if (!ret)
444 return TRACE_TYPE_PARTIAL_LINE;
445
446 /* Last spaces to align center */
447 for (i = 0; i < spaces - (spaces / 2); i++) {
448 ret = trace_seq_printf(s, " ");
449 if (!ret)
450 return TRACE_TYPE_PARTIAL_LINE;
451 }
452 return TRACE_TYPE_HANDLED;
453}
454
455
456static enum print_line_t
457print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
458{
459 if (!trace_seq_putc(s, ' '))
460 return 0;
461
462 return trace_print_lat_fmt(s, entry);
463}
464
465/* If the pid changed since the last trace, output this event */
466static enum print_line_t
467verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
468{
469 pid_t prev_pid;
470 pid_t *last_pid;
471 int ret;
472
473 if (!data)
474 return TRACE_TYPE_HANDLED;
475
476 last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
477
478 if (*last_pid == pid)
479 return TRACE_TYPE_HANDLED;
480
481 prev_pid = *last_pid;
482 *last_pid = pid;
483
484 if (prev_pid == -1)
485 return TRACE_TYPE_HANDLED;
486/*
487 * Context-switch trace line:
488
489 ------------------------------------------
490 | 1) migration/0--1 => sshd-1755
491 ------------------------------------------
492
493 */
494 ret = trace_seq_printf(s,
495 " ------------------------------------------\n");
496 if (!ret)
497 return TRACE_TYPE_PARTIAL_LINE;
498
499 ret = print_graph_cpu(s, cpu);
500 if (ret == TRACE_TYPE_PARTIAL_LINE)
501 return TRACE_TYPE_PARTIAL_LINE;
502
503 ret = print_graph_proc(s, prev_pid);
504 if (ret == TRACE_TYPE_PARTIAL_LINE)
505 return TRACE_TYPE_PARTIAL_LINE;
506
507 ret = trace_seq_printf(s, " => ");
508 if (!ret)
509 return TRACE_TYPE_PARTIAL_LINE;
510
511 ret = print_graph_proc(s, pid);
512 if (ret == TRACE_TYPE_PARTIAL_LINE)
513 return TRACE_TYPE_PARTIAL_LINE;
514
515 ret = trace_seq_printf(s,
516 "\n ------------------------------------------\n\n");
517 if (!ret)
518 return TRACE_TYPE_PARTIAL_LINE;
519
520 return TRACE_TYPE_HANDLED;
521}
522
523static struct ftrace_graph_ret_entry *
524get_return_for_leaf(struct trace_iterator *iter,
525 struct ftrace_graph_ent_entry *curr)
526{
527 struct fgraph_data *data = iter->private;
528 struct ring_buffer_iter *ring_iter = NULL;
529 struct ring_buffer_event *event;
530 struct ftrace_graph_ret_entry *next;
531
532 /*
533 * If the previous output failed to write to the seq buffer,
534 * then we just reuse the data from before.
535 */
536 if (data && data->failed) {
537 curr = &data->ent;
538 next = &data->ret;
539 } else {
540
541 ring_iter = iter->buffer_iter[iter->cpu];
542
543 /* First peek to compare current entry and the next one */
544 if (ring_iter)
545 event = ring_buffer_iter_peek(ring_iter, NULL);
546 else {
547 /*
548 * We need to consume the current entry to see
549 * the next one.
550 */
551 ring_buffer_consume(iter->tr->buffer, iter->cpu,
552 NULL, NULL);
553 event = ring_buffer_peek(iter->tr->buffer, iter->cpu,
554 NULL, NULL);
555 }
556
557 if (!event)
558 return NULL;
559
560 next = ring_buffer_event_data(event);
561
562 if (data) {
563 /*
564 * Save current and next entries for later reference
565 * if the output fails.
566 */
567 data->ent = *curr;
568 /*
569 * If the next event is not a return type, then
570 * we only care about what type it is. Otherwise we can
571 * safely copy the entire event.
572 */
573 if (next->ent.type == TRACE_GRAPH_RET)
574 data->ret = *next;
575 else
576 data->ret.ent.type = next->ent.type;
577 }
578 }
579
580 if (next->ent.type != TRACE_GRAPH_RET)
581 return NULL;
582
583 if (curr->ent.pid != next->ent.pid ||
584 curr->graph_ent.func != next->ret.func)
585 return NULL;
586
587 /* this is a leaf, now advance the iterator */
588 if (ring_iter)
589 ring_buffer_read(ring_iter, NULL);
590
591 return next;
592}
593
594static int print_graph_abs_time(u64 t, struct trace_seq *s)
595{
596 unsigned long usecs_rem;
597
598 usecs_rem = do_div(t, NSEC_PER_SEC);
599 usecs_rem /= 1000;
600
601 return trace_seq_printf(s, "%5lu.%06lu | ",
602 (unsigned long)t, usecs_rem);
603}
604
605static enum print_line_t
606print_graph_irq(struct trace_iterator *iter, unsigned long addr,
607 enum trace_type type, int cpu, pid_t pid, u32 flags)
608{
609 int ret;
610 struct trace_seq *s = &iter->seq;
611
612 if (addr < (unsigned long)__irqentry_text_start ||
613 addr >= (unsigned long)__irqentry_text_end)
614 return TRACE_TYPE_UNHANDLED;
615
616 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
617 /* Absolute time */
618 if (flags & TRACE_GRAPH_PRINT_ABS_TIME) {
619 ret = print_graph_abs_time(iter->ts, s);
620 if (!ret)
621 return TRACE_TYPE_PARTIAL_LINE;
622 }
623
624 /* Cpu */
625 if (flags & TRACE_GRAPH_PRINT_CPU) {
626 ret = print_graph_cpu(s, cpu);
627 if (ret == TRACE_TYPE_PARTIAL_LINE)
628 return TRACE_TYPE_PARTIAL_LINE;
629 }
630
631 /* Proc */
632 if (flags & TRACE_GRAPH_PRINT_PROC) {
633 ret = print_graph_proc(s, pid);
634 if (ret == TRACE_TYPE_PARTIAL_LINE)
635 return TRACE_TYPE_PARTIAL_LINE;
636 ret = trace_seq_printf(s, " | ");
637 if (!ret)
638 return TRACE_TYPE_PARTIAL_LINE;
639 }
640 }
641
642 /* No overhead */
643 ret = print_graph_duration(DURATION_FILL_START, s, flags);
644 if (ret != TRACE_TYPE_HANDLED)
645 return ret;
646
647 if (type == TRACE_GRAPH_ENT)
648 ret = trace_seq_printf(s, "==========>");
649 else
650 ret = trace_seq_printf(s, "<==========");
651
652 if (!ret)
653 return TRACE_TYPE_PARTIAL_LINE;
654
655 ret = print_graph_duration(DURATION_FILL_END, s, flags);
656 if (ret != TRACE_TYPE_HANDLED)
657 return ret;
658
659 ret = trace_seq_printf(s, "\n");
660
661 if (!ret)
662 return TRACE_TYPE_PARTIAL_LINE;
663 return TRACE_TYPE_HANDLED;
664}
665
666enum print_line_t
667trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
668{
669 unsigned long nsecs_rem = do_div(duration, 1000);
670 /* log10(ULONG_MAX) + '\0' */
671 char msecs_str[21];
672 char nsecs_str[5];
673 int ret, len;
674 int i;
675
676 sprintf(msecs_str, "%lu", (unsigned long) duration);
677
678 /* Print msecs */
679 ret = trace_seq_printf(s, "%s", msecs_str);
680 if (!ret)
681 return TRACE_TYPE_PARTIAL_LINE;
682
683 len = strlen(msecs_str);
684
685 /* Print nsecs (we don't want to exceed 7 numbers) */
686 if (len < 7) {
687 size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
688
689 snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
690 ret = trace_seq_printf(s, ".%s", nsecs_str);
691 if (!ret)
692 return TRACE_TYPE_PARTIAL_LINE;
693 len += strlen(nsecs_str);
694 }
695
696 ret = trace_seq_printf(s, " us ");
697 if (!ret)
698 return TRACE_TYPE_PARTIAL_LINE;
699
700 /* Print remaining spaces to fit the row's width */
701 for (i = len; i < 7; i++) {
702 ret = trace_seq_printf(s, " ");
703 if (!ret)
704 return TRACE_TYPE_PARTIAL_LINE;
705 }
706 return TRACE_TYPE_HANDLED;
707}
708
709static enum print_line_t
710print_graph_duration(unsigned long long duration, struct trace_seq *s,
711 u32 flags)
712{
713 int ret = -1;
714
715 if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
716 !(trace_flags & TRACE_ITER_CONTEXT_INFO))
717 return TRACE_TYPE_HANDLED;
718
719 /* No real adata, just filling the column with spaces */
720 switch (duration) {
721 case DURATION_FILL_FULL:
722 ret = trace_seq_printf(s, " | ");
723 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
724 case DURATION_FILL_START:
725 ret = trace_seq_printf(s, " ");
726 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
727 case DURATION_FILL_END:
728 ret = trace_seq_printf(s, " |");
729 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
730 }
731
732 /* Signal a overhead of time execution to the output */
733 if (flags & TRACE_GRAPH_PRINT_OVERHEAD) {
734 /* Duration exceeded 100 msecs */
735 if (duration > 100000ULL)
736 ret = trace_seq_printf(s, "! ");
737 /* Duration exceeded 10 msecs */
738 else if (duration > 10000ULL)
739 ret = trace_seq_printf(s, "+ ");
740 }
741
742 /*
743 * The -1 means we either did not exceed the duration tresholds
744 * or we dont want to print out the overhead. Either way we need
745 * to fill out the space.
746 */
747 if (ret == -1)
748 ret = trace_seq_printf(s, " ");
749
750 /* Catching here any failure happenned above */
751 if (!ret)
752 return TRACE_TYPE_PARTIAL_LINE;
753
754 ret = trace_print_graph_duration(duration, s);
755 if (ret != TRACE_TYPE_HANDLED)
756 return ret;
757
758 ret = trace_seq_printf(s, "| ");
759 if (!ret)
760 return TRACE_TYPE_PARTIAL_LINE;
761
762 return TRACE_TYPE_HANDLED;
763}
764
765/* Case of a leaf function on its call entry */
766static enum print_line_t
767print_graph_entry_leaf(struct trace_iterator *iter,
768 struct ftrace_graph_ent_entry *entry,
769 struct ftrace_graph_ret_entry *ret_entry,
770 struct trace_seq *s, u32 flags)
771{
772 struct fgraph_data *data = iter->private;
773 struct ftrace_graph_ret *graph_ret;
774 struct ftrace_graph_ent *call;
775 unsigned long long duration;
776 int ret;
777 int i;
778
779 graph_ret = &ret_entry->ret;
780 call = &entry->graph_ent;
781 duration = graph_ret->rettime - graph_ret->calltime;
782
783 if (data) {
784 struct fgraph_cpu_data *cpu_data;
785 int cpu = iter->cpu;
786
787 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
788
789 /*
790 * Comments display at + 1 to depth. Since
791 * this is a leaf function, keep the comments
792 * equal to this depth.
793 */
794 cpu_data->depth = call->depth - 1;
795
796 /* No need to keep this function around for this depth */
797 if (call->depth < FTRACE_RETFUNC_DEPTH)
798 cpu_data->enter_funcs[call->depth] = 0;
799 }
800
801 /* Overhead and duration */
802 ret = print_graph_duration(duration, s, flags);
803 if (ret == TRACE_TYPE_PARTIAL_LINE)
804 return TRACE_TYPE_PARTIAL_LINE;
805
806 /* Function */
807 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
808 ret = trace_seq_printf(s, " ");
809 if (!ret)
810 return TRACE_TYPE_PARTIAL_LINE;
811 }
812
813 ret = trace_seq_printf(s, "%ps();\n", (void *)call->func);
814 if (!ret)
815 return TRACE_TYPE_PARTIAL_LINE;
816
817 return TRACE_TYPE_HANDLED;
818}
819
820static enum print_line_t
821print_graph_entry_nested(struct trace_iterator *iter,
822 struct ftrace_graph_ent_entry *entry,
823 struct trace_seq *s, int cpu, u32 flags)
824{
825 struct ftrace_graph_ent *call = &entry->graph_ent;
826 struct fgraph_data *data = iter->private;
827 int ret;
828 int i;
829
830 if (data) {
831 struct fgraph_cpu_data *cpu_data;
832 int cpu = iter->cpu;
833
834 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
835 cpu_data->depth = call->depth;
836
837 /* Save this function pointer to see if the exit matches */
838 if (call->depth < FTRACE_RETFUNC_DEPTH)
839 cpu_data->enter_funcs[call->depth] = call->func;
840 }
841
842 /* No time */
843 ret = print_graph_duration(DURATION_FILL_FULL, s, flags);
844 if (ret != TRACE_TYPE_HANDLED)
845 return ret;
846
847 /* Function */
848 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
849 ret = trace_seq_printf(s, " ");
850 if (!ret)
851 return TRACE_TYPE_PARTIAL_LINE;
852 }
853
854 ret = trace_seq_printf(s, "%ps() {\n", (void *)call->func);
855 if (!ret)
856 return TRACE_TYPE_PARTIAL_LINE;
857
858 /*
859 * we already consumed the current entry to check the next one
860 * and see if this is a leaf.
861 */
862 return TRACE_TYPE_NO_CONSUME;
863}
864
865static enum print_line_t
866print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
867 int type, unsigned long addr, u32 flags)
868{
869 struct fgraph_data *data = iter->private;
870 struct trace_entry *ent = iter->ent;
871 int cpu = iter->cpu;
872 int ret;
873
874 /* Pid */
875 if (verif_pid(s, ent->pid, cpu, data) == TRACE_TYPE_PARTIAL_LINE)
876 return TRACE_TYPE_PARTIAL_LINE;
877
878 if (type) {
879 /* Interrupt */
880 ret = print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
881 if (ret == TRACE_TYPE_PARTIAL_LINE)
882 return TRACE_TYPE_PARTIAL_LINE;
883 }
884
885 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
886 return 0;
887
888 /* Absolute time */
889 if (flags & TRACE_GRAPH_PRINT_ABS_TIME) {
890 ret = print_graph_abs_time(iter->ts, s);
891 if (!ret)
892 return TRACE_TYPE_PARTIAL_LINE;
893 }
894
895 /* Cpu */
896 if (flags & TRACE_GRAPH_PRINT_CPU) {
897 ret = print_graph_cpu(s, cpu);
898 if (ret == TRACE_TYPE_PARTIAL_LINE)
899 return TRACE_TYPE_PARTIAL_LINE;
900 }
901
902 /* Proc */
903 if (flags & TRACE_GRAPH_PRINT_PROC) {
904 ret = print_graph_proc(s, ent->pid);
905 if (ret == TRACE_TYPE_PARTIAL_LINE)
906 return TRACE_TYPE_PARTIAL_LINE;
907
908 ret = trace_seq_printf(s, " | ");
909 if (!ret)
910 return TRACE_TYPE_PARTIAL_LINE;
911 }
912
913 /* Latency format */
914 if (trace_flags & TRACE_ITER_LATENCY_FMT) {
915 ret = print_graph_lat_fmt(s, ent);
916 if (ret == TRACE_TYPE_PARTIAL_LINE)
917 return TRACE_TYPE_PARTIAL_LINE;
918 }
919
920 return 0;
921}
922
923/*
924 * Entry check for irq code
925 *
926 * returns 1 if
927 * - we are inside irq code
928 * - we just entered irq code
929 *
930 * retunns 0 if
931 * - funcgraph-interrupts option is set
932 * - we are not inside irq code
933 */
934static int
935check_irq_entry(struct trace_iterator *iter, u32 flags,
936 unsigned long addr, int depth)
937{
938 int cpu = iter->cpu;
939 int *depth_irq;
940 struct fgraph_data *data = iter->private;
941
942 /*
943 * If we are either displaying irqs, or we got called as
944 * a graph event and private data does not exist,
945 * then we bypass the irq check.
946 */
947 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
948 (!data))
949 return 0;
950
951 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
952
953 /*
954 * We are inside the irq code
955 */
956 if (*depth_irq >= 0)
957 return 1;
958
959 if ((addr < (unsigned long)__irqentry_text_start) ||
960 (addr >= (unsigned long)__irqentry_text_end))
961 return 0;
962
963 /*
964 * We are entering irq code.
965 */
966 *depth_irq = depth;
967 return 1;
968}
969
970/*
971 * Return check for irq code
972 *
973 * returns 1 if
974 * - we are inside irq code
975 * - we just left irq code
976 *
977 * returns 0 if
978 * - funcgraph-interrupts option is set
979 * - we are not inside irq code
980 */
981static int
982check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
983{
984 int cpu = iter->cpu;
985 int *depth_irq;
986 struct fgraph_data *data = iter->private;
987
988 /*
989 * If we are either displaying irqs, or we got called as
990 * a graph event and private data does not exist,
991 * then we bypass the irq check.
992 */
993 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
994 (!data))
995 return 0;
996
997 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
998
999 /*
1000 * We are not inside the irq code.
1001 */
1002 if (*depth_irq == -1)
1003 return 0;
1004
1005 /*
1006 * We are inside the irq code, and this is returning entry.
1007 * Let's not trace it and clear the entry depth, since
1008 * we are out of irq code.
1009 *
1010 * This condition ensures that we 'leave the irq code' once
1011 * we are out of the entry depth. Thus protecting us from
1012 * the RETURN entry loss.
1013 */
1014 if (*depth_irq >= depth) {
1015 *depth_irq = -1;
1016 return 1;
1017 }
1018
1019 /*
1020 * We are inside the irq code, and this is not the entry.
1021 */
1022 return 1;
1023}
1024
1025static enum print_line_t
1026print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
1027 struct trace_iterator *iter, u32 flags)
1028{
1029 struct fgraph_data *data = iter->private;
1030 struct ftrace_graph_ent *call = &field->graph_ent;
1031 struct ftrace_graph_ret_entry *leaf_ret;
1032 static enum print_line_t ret;
1033 int cpu = iter->cpu;
1034
1035 if (check_irq_entry(iter, flags, call->func, call->depth))
1036 return TRACE_TYPE_HANDLED;
1037
1038 if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags))
1039 return TRACE_TYPE_PARTIAL_LINE;
1040
1041 leaf_ret = get_return_for_leaf(iter, field);
1042 if (leaf_ret)
1043 ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
1044 else
1045 ret = print_graph_entry_nested(iter, field, s, cpu, flags);
1046
1047 if (data) {
1048 /*
1049 * If we failed to write our output, then we need to make
1050 * note of it. Because we already consumed our entry.
1051 */
1052 if (s->full) {
1053 data->failed = 1;
1054 data->cpu = cpu;
1055 } else
1056 data->failed = 0;
1057 }
1058
1059 return ret;
1060}
1061
1062static enum print_line_t
1063print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
1064 struct trace_entry *ent, struct trace_iterator *iter,
1065 u32 flags)
1066{
1067 unsigned long long duration = trace->rettime - trace->calltime;
1068 struct fgraph_data *data = iter->private;
1069 pid_t pid = ent->pid;
1070 int cpu = iter->cpu;
1071 int func_match = 1;
1072 int ret;
1073 int i;
1074
1075 if (check_irq_return(iter, flags, trace->depth))
1076 return TRACE_TYPE_HANDLED;
1077
1078 if (data) {
1079 struct fgraph_cpu_data *cpu_data;
1080 int cpu = iter->cpu;
1081
1082 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
1083
1084 /*
1085 * Comments display at + 1 to depth. This is the
1086 * return from a function, we now want the comments
1087 * to display at the same level of the bracket.
1088 */
1089 cpu_data->depth = trace->depth - 1;
1090
1091 if (trace->depth < FTRACE_RETFUNC_DEPTH) {
1092 if (cpu_data->enter_funcs[trace->depth] != trace->func)
1093 func_match = 0;
1094 cpu_data->enter_funcs[trace->depth] = 0;
1095 }
1096 }
1097
1098 if (print_graph_prologue(iter, s, 0, 0, flags))
1099 return TRACE_TYPE_PARTIAL_LINE;
1100
1101 /* Overhead and duration */
1102 ret = print_graph_duration(duration, s, flags);
1103 if (ret == TRACE_TYPE_PARTIAL_LINE)
1104 return TRACE_TYPE_PARTIAL_LINE;
1105
1106 /* Closing brace */
1107 for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
1108 ret = trace_seq_printf(s, " ");
1109 if (!ret)
1110 return TRACE_TYPE_PARTIAL_LINE;
1111 }
1112
1113 /*
1114 * If the return function does not have a matching entry,
1115 * then the entry was lost. Instead of just printing
1116 * the '}' and letting the user guess what function this
1117 * belongs to, write out the function name.
1118 */
1119 if (func_match) {
1120 ret = trace_seq_printf(s, "}\n");
1121 if (!ret)
1122 return TRACE_TYPE_PARTIAL_LINE;
1123 } else {
1124 ret = trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
1125 if (!ret)
1126 return TRACE_TYPE_PARTIAL_LINE;
1127 }
1128
1129 /* Overrun */
1130 if (flags & TRACE_GRAPH_PRINT_OVERRUN) {
1131 ret = trace_seq_printf(s, " (Overruns: %lu)\n",
1132 trace->overrun);
1133 if (!ret)
1134 return TRACE_TYPE_PARTIAL_LINE;
1135 }
1136
1137 ret = print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
1138 cpu, pid, flags);
1139 if (ret == TRACE_TYPE_PARTIAL_LINE)
1140 return TRACE_TYPE_PARTIAL_LINE;
1141
1142 return TRACE_TYPE_HANDLED;
1143}
1144
1145static enum print_line_t
1146print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1147 struct trace_iterator *iter, u32 flags)
1148{
1149 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1150 struct fgraph_data *data = iter->private;
1151 struct trace_event *event;
1152 int depth = 0;
1153 int ret;
1154 int i;
1155
1156 if (data)
1157 depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1158
1159 if (print_graph_prologue(iter, s, 0, 0, flags))
1160 return TRACE_TYPE_PARTIAL_LINE;
1161
1162 /* No time */
1163 ret = print_graph_duration(DURATION_FILL_FULL, s, flags);
1164 if (ret != TRACE_TYPE_HANDLED)
1165 return ret;
1166
1167 /* Indentation */
1168 if (depth > 0)
1169 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++) {
1170 ret = trace_seq_printf(s, " ");
1171 if (!ret)
1172 return TRACE_TYPE_PARTIAL_LINE;
1173 }
1174
1175 /* The comment */
1176 ret = trace_seq_printf(s, "/* ");
1177 if (!ret)
1178 return TRACE_TYPE_PARTIAL_LINE;
1179
1180 switch (iter->ent->type) {
1181 case TRACE_BPRINT:
1182 ret = trace_print_bprintk_msg_only(iter);
1183 if (ret != TRACE_TYPE_HANDLED)
1184 return ret;
1185 break;
1186 case TRACE_PRINT:
1187 ret = trace_print_printk_msg_only(iter);
1188 if (ret != TRACE_TYPE_HANDLED)
1189 return ret;
1190 break;
1191 default:
1192 event = ftrace_find_event(ent->type);
1193 if (!event)
1194 return TRACE_TYPE_UNHANDLED;
1195
1196 ret = event->funcs->trace(iter, sym_flags, event);
1197 if (ret != TRACE_TYPE_HANDLED)
1198 return ret;
1199 }
1200
1201 /* Strip ending newline */
1202 if (s->buffer[s->len - 1] == '\n') {
1203 s->buffer[s->len - 1] = '\0';
1204 s->len--;
1205 }
1206
1207 ret = trace_seq_printf(s, " */\n");
1208 if (!ret)
1209 return TRACE_TYPE_PARTIAL_LINE;
1210
1211 return TRACE_TYPE_HANDLED;
1212}
1213
1214
1215enum print_line_t
1216print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1217{
1218 struct ftrace_graph_ent_entry *field;
1219 struct fgraph_data *data = iter->private;
1220 struct trace_entry *entry = iter->ent;
1221 struct trace_seq *s = &iter->seq;
1222 int cpu = iter->cpu;
1223 int ret;
1224
1225 if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1226 per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1227 return TRACE_TYPE_HANDLED;
1228 }
1229
1230 /*
1231 * If the last output failed, there's a possibility we need
1232 * to print out the missing entry which would never go out.
1233 */
1234 if (data && data->failed) {
1235 field = &data->ent;
1236 iter->cpu = data->cpu;
1237 ret = print_graph_entry(field, s, iter, flags);
1238 if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1239 per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1240 ret = TRACE_TYPE_NO_CONSUME;
1241 }
1242 iter->cpu = cpu;
1243 return ret;
1244 }
1245
1246 switch (entry->type) {
1247 case TRACE_GRAPH_ENT: {
1248 /*
1249 * print_graph_entry() may consume the current event,
1250 * thus @field may become invalid, so we need to save it.
1251 * sizeof(struct ftrace_graph_ent_entry) is very small,
1252 * it can be safely saved at the stack.
1253 */
1254 struct ftrace_graph_ent_entry saved;
1255 trace_assign_type(field, entry);
1256 saved = *field;
1257 return print_graph_entry(&saved, s, iter, flags);
1258 }
1259 case TRACE_GRAPH_RET: {
1260 struct ftrace_graph_ret_entry *field;
1261 trace_assign_type(field, entry);
1262 return print_graph_return(&field->ret, s, entry, iter, flags);
1263 }
1264 case TRACE_STACK:
1265 case TRACE_FN:
1266 /* dont trace stack and functions as comments */
1267 return TRACE_TYPE_UNHANDLED;
1268
1269 default:
1270 return print_graph_comment(s, entry, iter, flags);
1271 }
1272
1273 return TRACE_TYPE_HANDLED;
1274}
1275
1276static enum print_line_t
1277print_graph_function(struct trace_iterator *iter)
1278{
1279 return print_graph_function_flags(iter, tracer_flags.val);
1280}
1281
1282static enum print_line_t
1283print_graph_function_event(struct trace_iterator *iter, int flags,
1284 struct trace_event *event)
1285{
1286 return print_graph_function(iter);
1287}
1288
1289static void print_lat_header(struct seq_file *s, u32 flags)
1290{
1291 static const char spaces[] = " " /* 16 spaces */
1292 " " /* 4 spaces */
1293 " "; /* 17 spaces */
1294 int size = 0;
1295
1296 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1297 size += 16;
1298 if (flags & TRACE_GRAPH_PRINT_CPU)
1299 size += 4;
1300 if (flags & TRACE_GRAPH_PRINT_PROC)
1301 size += 17;
1302
1303 seq_printf(s, "#%.*s _-----=> irqs-off \n", size, spaces);
1304 seq_printf(s, "#%.*s / _----=> need-resched \n", size, spaces);
1305 seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1306 seq_printf(s, "#%.*s|| / _--=> preempt-depth \n", size, spaces);
1307 seq_printf(s, "#%.*s||| / \n", size, spaces);
1308}
1309
1310static void __print_graph_headers_flags(struct seq_file *s, u32 flags)
1311{
1312 int lat = trace_flags & TRACE_ITER_LATENCY_FMT;
1313
1314 if (lat)
1315 print_lat_header(s, flags);
1316
1317 /* 1st line */
1318 seq_printf(s, "#");
1319 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1320 seq_printf(s, " TIME ");
1321 if (flags & TRACE_GRAPH_PRINT_CPU)
1322 seq_printf(s, " CPU");
1323 if (flags & TRACE_GRAPH_PRINT_PROC)
1324 seq_printf(s, " TASK/PID ");
1325 if (lat)
1326 seq_printf(s, "||||");
1327 if (flags & TRACE_GRAPH_PRINT_DURATION)
1328 seq_printf(s, " DURATION ");
1329 seq_printf(s, " FUNCTION CALLS\n");
1330
1331 /* 2nd line */
1332 seq_printf(s, "#");
1333 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1334 seq_printf(s, " | ");
1335 if (flags & TRACE_GRAPH_PRINT_CPU)
1336 seq_printf(s, " | ");
1337 if (flags & TRACE_GRAPH_PRINT_PROC)
1338 seq_printf(s, " | | ");
1339 if (lat)
1340 seq_printf(s, "||||");
1341 if (flags & TRACE_GRAPH_PRINT_DURATION)
1342 seq_printf(s, " | | ");
1343 seq_printf(s, " | | | |\n");
1344}
1345
1346void print_graph_headers(struct seq_file *s)
1347{
1348 print_graph_headers_flags(s, tracer_flags.val);
1349}
1350
1351void print_graph_headers_flags(struct seq_file *s, u32 flags)
1352{
1353 struct trace_iterator *iter = s->private;
1354
1355 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
1356 return;
1357
1358 if (trace_flags & TRACE_ITER_LATENCY_FMT) {
1359 /* print nothing if the buffers are empty */
1360 if (trace_empty(iter))
1361 return;
1362
1363 print_trace_header(s, iter);
1364 }
1365
1366 __print_graph_headers_flags(s, flags);
1367}
1368
1369void graph_trace_open(struct trace_iterator *iter)
1370{
1371 /* pid and depth on the last trace processed */
1372 struct fgraph_data *data;
1373 int cpu;
1374
1375 iter->private = NULL;
1376
1377 data = kzalloc(sizeof(*data), GFP_KERNEL);
1378 if (!data)
1379 goto out_err;
1380
1381 data->cpu_data = alloc_percpu(struct fgraph_cpu_data);
1382 if (!data->cpu_data)
1383 goto out_err_free;
1384
1385 for_each_possible_cpu(cpu) {
1386 pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1387 int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1388 int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1389 int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1390
1391 *pid = -1;
1392 *depth = 0;
1393 *ignore = 0;
1394 *depth_irq = -1;
1395 }
1396
1397 iter->private = data;
1398
1399 return;
1400
1401 out_err_free:
1402 kfree(data);
1403 out_err:
1404 pr_warning("function graph tracer: not enough memory\n");
1405}
1406
1407void graph_trace_close(struct trace_iterator *iter)
1408{
1409 struct fgraph_data *data = iter->private;
1410
1411 if (data) {
1412 free_percpu(data->cpu_data);
1413 kfree(data);
1414 }
1415}
1416
1417static int func_graph_set_flag(u32 old_flags, u32 bit, int set)
1418{
1419 if (bit == TRACE_GRAPH_PRINT_IRQS)
1420 ftrace_graph_skip_irqs = !set;
1421
1422 return 0;
1423}
1424
1425static struct trace_event_functions graph_functions = {
1426 .trace = print_graph_function_event,
1427};
1428
1429static struct trace_event graph_trace_entry_event = {
1430 .type = TRACE_GRAPH_ENT,
1431 .funcs = &graph_functions,
1432};
1433
1434static struct trace_event graph_trace_ret_event = {
1435 .type = TRACE_GRAPH_RET,
1436 .funcs = &graph_functions
1437};
1438
1439static struct tracer graph_trace __read_mostly = {
1440 .name = "function_graph",
1441 .open = graph_trace_open,
1442 .pipe_open = graph_trace_open,
1443 .close = graph_trace_close,
1444 .pipe_close = graph_trace_close,
1445 .wait_pipe = poll_wait_pipe,
1446 .init = graph_trace_init,
1447 .reset = graph_trace_reset,
1448 .print_line = print_graph_function,
1449 .print_header = print_graph_headers,
1450 .flags = &tracer_flags,
1451 .set_flag = func_graph_set_flag,
1452#ifdef CONFIG_FTRACE_SELFTEST
1453 .selftest = trace_selftest_startup_function_graph,
1454#endif
1455};
1456
1457static __init int init_graph_trace(void)
1458{
1459 max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1460
1461 if (!register_ftrace_event(&graph_trace_entry_event)) {
1462 pr_warning("Warning: could not register graph trace events\n");
1463 return 1;
1464 }
1465
1466 if (!register_ftrace_event(&graph_trace_ret_event)) {
1467 pr_warning("Warning: could not register graph trace events\n");
1468 return 1;
1469 }
1470
1471 return register_tracer(&graph_trace);
1472}
1473
1474device_initcall(init_graph_trace);