Merge branch 'misc.namei' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux-2.6-microblaze.git] / kernel / trace / trace_functions_graph.c
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 */
20 static int ftrace_graph_skip_irqs;
21
22 struct 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
30 struct 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
42 unsigned int fgraph_max_depth;
43
44 static 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
72 static 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
80 static 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  */
87 enum {
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
93 static void
94 print_graph_duration(struct trace_array *tr, unsigned long long duration,
95                      struct trace_seq *s, u32 flags);
96
97 int __trace_graph_entry(struct trace_array *tr,
98                                 struct ftrace_graph_ent *trace,
99                                 unsigned int trace_ctx)
100 {
101         struct trace_event_call *call = &event_funcgraph_entry;
102         struct ring_buffer_event *event;
103         struct trace_buffer *buffer = tr->array_buffer.buffer;
104         struct ftrace_graph_ent_entry *entry;
105
106         event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
107                                           sizeof(*entry), trace_ctx);
108         if (!event)
109                 return 0;
110         entry   = ring_buffer_event_data(event);
111         entry->graph_ent                        = *trace;
112         if (!call_filter_check_discard(call, entry, buffer, event))
113                 trace_buffer_unlock_commit_nostack(buffer, event);
114
115         return 1;
116 }
117
118 static inline int ftrace_graph_ignore_irqs(void)
119 {
120         if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
121                 return 0;
122
123         return in_irq();
124 }
125
126 int trace_graph_entry(struct ftrace_graph_ent *trace)
127 {
128         struct trace_array *tr = graph_array;
129         struct trace_array_cpu *data;
130         unsigned long flags;
131         unsigned int trace_ctx;
132         long disabled;
133         int ret;
134         int cpu;
135
136         if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT))
137                 return 0;
138
139         /*
140          * Do not trace a function if it's filtered by set_graph_notrace.
141          * Make the index of ret stack negative to indicate that it should
142          * ignore further functions.  But it needs its own ret stack entry
143          * to recover the original index in order to continue tracing after
144          * returning from the function.
145          */
146         if (ftrace_graph_notrace_addr(trace->func)) {
147                 trace_recursion_set(TRACE_GRAPH_NOTRACE_BIT);
148                 /*
149                  * Need to return 1 to have the return called
150                  * that will clear the NOTRACE bit.
151                  */
152                 return 1;
153         }
154
155         if (!ftrace_trace_task(tr))
156                 return 0;
157
158         if (ftrace_graph_ignore_func(trace))
159                 return 0;
160
161         if (ftrace_graph_ignore_irqs())
162                 return 0;
163
164         /*
165          * Stop here if tracing_threshold is set. We only write function return
166          * events to the ring buffer.
167          */
168         if (tracing_thresh)
169                 return 1;
170
171         local_irq_save(flags);
172         cpu = raw_smp_processor_id();
173         data = per_cpu_ptr(tr->array_buffer.data, cpu);
174         disabled = atomic_inc_return(&data->disabled);
175         if (likely(disabled == 1)) {
176                 trace_ctx = tracing_gen_ctx_flags(flags);
177                 ret = __trace_graph_entry(tr, trace, trace_ctx);
178         } else {
179                 ret = 0;
180         }
181
182         atomic_dec(&data->disabled);
183         local_irq_restore(flags);
184
185         return ret;
186 }
187
188 static void
189 __trace_graph_function(struct trace_array *tr,
190                 unsigned long ip, unsigned int trace_ctx)
191 {
192         u64 time = trace_clock_local();
193         struct ftrace_graph_ent ent = {
194                 .func  = ip,
195                 .depth = 0,
196         };
197         struct ftrace_graph_ret ret = {
198                 .func     = ip,
199                 .depth    = 0,
200                 .calltime = time,
201                 .rettime  = time,
202         };
203
204         __trace_graph_entry(tr, &ent, trace_ctx);
205         __trace_graph_return(tr, &ret, trace_ctx);
206 }
207
208 void
209 trace_graph_function(struct trace_array *tr,
210                 unsigned long ip, unsigned long parent_ip,
211                 unsigned int trace_ctx)
212 {
213         __trace_graph_function(tr, ip, trace_ctx);
214 }
215
216 void __trace_graph_return(struct trace_array *tr,
217                                 struct ftrace_graph_ret *trace,
218                                 unsigned int trace_ctx)
219 {
220         struct trace_event_call *call = &event_funcgraph_exit;
221         struct ring_buffer_event *event;
222         struct trace_buffer *buffer = tr->array_buffer.buffer;
223         struct ftrace_graph_ret_entry *entry;
224
225         event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
226                                           sizeof(*entry), trace_ctx);
227         if (!event)
228                 return;
229         entry   = ring_buffer_event_data(event);
230         entry->ret                              = *trace;
231         if (!call_filter_check_discard(call, entry, buffer, event))
232                 trace_buffer_unlock_commit_nostack(buffer, event);
233 }
234
235 void trace_graph_return(struct ftrace_graph_ret *trace)
236 {
237         struct trace_array *tr = graph_array;
238         struct trace_array_cpu *data;
239         unsigned long flags;
240         unsigned int trace_ctx;
241         long disabled;
242         int cpu;
243
244         ftrace_graph_addr_finish(trace);
245
246         if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT)) {
247                 trace_recursion_clear(TRACE_GRAPH_NOTRACE_BIT);
248                 return;
249         }
250
251         local_irq_save(flags);
252         cpu = raw_smp_processor_id();
253         data = per_cpu_ptr(tr->array_buffer.data, cpu);
254         disabled = atomic_inc_return(&data->disabled);
255         if (likely(disabled == 1)) {
256                 trace_ctx = tracing_gen_ctx_flags(flags);
257                 __trace_graph_return(tr, trace, trace_ctx);
258         }
259         atomic_dec(&data->disabled);
260         local_irq_restore(flags);
261 }
262
263 void set_graph_array(struct trace_array *tr)
264 {
265         graph_array = tr;
266
267         /* Make graph_array visible before we start tracing */
268
269         smp_mb();
270 }
271
272 static void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
273 {
274         ftrace_graph_addr_finish(trace);
275
276         if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT)) {
277                 trace_recursion_clear(TRACE_GRAPH_NOTRACE_BIT);
278                 return;
279         }
280
281         if (tracing_thresh &&
282             (trace->rettime - trace->calltime < tracing_thresh))
283                 return;
284         else
285                 trace_graph_return(trace);
286 }
287
288 static struct fgraph_ops funcgraph_thresh_ops = {
289         .entryfunc = &trace_graph_entry,
290         .retfunc = &trace_graph_thresh_return,
291 };
292
293 static struct fgraph_ops funcgraph_ops = {
294         .entryfunc = &trace_graph_entry,
295         .retfunc = &trace_graph_return,
296 };
297
298 static int graph_trace_init(struct trace_array *tr)
299 {
300         int ret;
301
302         set_graph_array(tr);
303         if (tracing_thresh)
304                 ret = register_ftrace_graph(&funcgraph_thresh_ops);
305         else
306                 ret = register_ftrace_graph(&funcgraph_ops);
307         if (ret)
308                 return ret;
309         tracing_start_cmdline_record();
310
311         return 0;
312 }
313
314 static void graph_trace_reset(struct trace_array *tr)
315 {
316         tracing_stop_cmdline_record();
317         if (tracing_thresh)
318                 unregister_ftrace_graph(&funcgraph_thresh_ops);
319         else
320                 unregister_ftrace_graph(&funcgraph_ops);
321 }
322
323 static int graph_trace_update_thresh(struct trace_array *tr)
324 {
325         graph_trace_reset(tr);
326         return graph_trace_init(tr);
327 }
328
329 static int max_bytes_for_cpu;
330
331 static void print_graph_cpu(struct trace_seq *s, int cpu)
332 {
333         /*
334          * Start with a space character - to make it stand out
335          * to the right a bit when trace output is pasted into
336          * email:
337          */
338         trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
339 }
340
341 #define TRACE_GRAPH_PROCINFO_LENGTH     14
342
343 static void print_graph_proc(struct trace_seq *s, pid_t pid)
344 {
345         char comm[TASK_COMM_LEN];
346         /* sign + log10(MAX_INT) + '\0' */
347         char pid_str[11];
348         int spaces = 0;
349         int len;
350         int i;
351
352         trace_find_cmdline(pid, comm);
353         comm[7] = '\0';
354         sprintf(pid_str, "%d", pid);
355
356         /* 1 stands for the "-" character */
357         len = strlen(comm) + strlen(pid_str) + 1;
358
359         if (len < TRACE_GRAPH_PROCINFO_LENGTH)
360                 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
361
362         /* First spaces to align center */
363         for (i = 0; i < spaces / 2; i++)
364                 trace_seq_putc(s, ' ');
365
366         trace_seq_printf(s, "%s-%s", comm, pid_str);
367
368         /* Last spaces to align center */
369         for (i = 0; i < spaces - (spaces / 2); i++)
370                 trace_seq_putc(s, ' ');
371 }
372
373
374 static void print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
375 {
376         trace_seq_putc(s, ' ');
377         trace_print_lat_fmt(s, entry);
378         trace_seq_puts(s, " | ");
379 }
380
381 /* If the pid changed since the last trace, output this event */
382 static void
383 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
384 {
385         pid_t prev_pid;
386         pid_t *last_pid;
387
388         if (!data)
389                 return;
390
391         last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
392
393         if (*last_pid == pid)
394                 return;
395
396         prev_pid = *last_pid;
397         *last_pid = pid;
398
399         if (prev_pid == -1)
400                 return;
401 /*
402  * Context-switch trace line:
403
404  ------------------------------------------
405  | 1)  migration/0--1  =>  sshd-1755
406  ------------------------------------------
407
408  */
409         trace_seq_puts(s, " ------------------------------------------\n");
410         print_graph_cpu(s, cpu);
411         print_graph_proc(s, prev_pid);
412         trace_seq_puts(s, " => ");
413         print_graph_proc(s, pid);
414         trace_seq_puts(s, "\n ------------------------------------------\n\n");
415 }
416
417 static struct ftrace_graph_ret_entry *
418 get_return_for_leaf(struct trace_iterator *iter,
419                 struct ftrace_graph_ent_entry *curr)
420 {
421         struct fgraph_data *data = iter->private;
422         struct ring_buffer_iter *ring_iter = NULL;
423         struct ring_buffer_event *event;
424         struct ftrace_graph_ret_entry *next;
425
426         /*
427          * If the previous output failed to write to the seq buffer,
428          * then we just reuse the data from before.
429          */
430         if (data && data->failed) {
431                 curr = &data->ent;
432                 next = &data->ret;
433         } else {
434
435                 ring_iter = trace_buffer_iter(iter, iter->cpu);
436
437                 /* First peek to compare current entry and the next one */
438                 if (ring_iter)
439                         event = ring_buffer_iter_peek(ring_iter, NULL);
440                 else {
441                         /*
442                          * We need to consume the current entry to see
443                          * the next one.
444                          */
445                         ring_buffer_consume(iter->array_buffer->buffer, iter->cpu,
446                                             NULL, NULL);
447                         event = ring_buffer_peek(iter->array_buffer->buffer, iter->cpu,
448                                                  NULL, NULL);
449                 }
450
451                 if (!event)
452                         return NULL;
453
454                 next = ring_buffer_event_data(event);
455
456                 if (data) {
457                         /*
458                          * Save current and next entries for later reference
459                          * if the output fails.
460                          */
461                         data->ent = *curr;
462                         /*
463                          * If the next event is not a return type, then
464                          * we only care about what type it is. Otherwise we can
465                          * safely copy the entire event.
466                          */
467                         if (next->ent.type == TRACE_GRAPH_RET)
468                                 data->ret = *next;
469                         else
470                                 data->ret.ent.type = next->ent.type;
471                 }
472         }
473
474         if (next->ent.type != TRACE_GRAPH_RET)
475                 return NULL;
476
477         if (curr->ent.pid != next->ent.pid ||
478                         curr->graph_ent.func != next->ret.func)
479                 return NULL;
480
481         /* this is a leaf, now advance the iterator */
482         if (ring_iter)
483                 ring_buffer_iter_advance(ring_iter);
484
485         return next;
486 }
487
488 static void print_graph_abs_time(u64 t, struct trace_seq *s)
489 {
490         unsigned long usecs_rem;
491
492         usecs_rem = do_div(t, NSEC_PER_SEC);
493         usecs_rem /= 1000;
494
495         trace_seq_printf(s, "%5lu.%06lu |  ",
496                          (unsigned long)t, usecs_rem);
497 }
498
499 static void
500 print_graph_rel_time(struct trace_iterator *iter, struct trace_seq *s)
501 {
502         unsigned long long usecs;
503
504         usecs = iter->ts - iter->array_buffer->time_start;
505         do_div(usecs, NSEC_PER_USEC);
506
507         trace_seq_printf(s, "%9llu us |  ", usecs);
508 }
509
510 static void
511 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
512                 enum trace_type type, int cpu, pid_t pid, u32 flags)
513 {
514         struct trace_array *tr = iter->tr;
515         struct trace_seq *s = &iter->seq;
516         struct trace_entry *ent = iter->ent;
517
518         if (addr < (unsigned long)__irqentry_text_start ||
519                 addr >= (unsigned long)__irqentry_text_end)
520                 return;
521
522         if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
523                 /* Absolute time */
524                 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
525                         print_graph_abs_time(iter->ts, s);
526
527                 /* Relative time */
528                 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
529                         print_graph_rel_time(iter, s);
530
531                 /* Cpu */
532                 if (flags & TRACE_GRAPH_PRINT_CPU)
533                         print_graph_cpu(s, cpu);
534
535                 /* Proc */
536                 if (flags & TRACE_GRAPH_PRINT_PROC) {
537                         print_graph_proc(s, pid);
538                         trace_seq_puts(s, " | ");
539                 }
540
541                 /* Latency format */
542                 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
543                         print_graph_lat_fmt(s, ent);
544         }
545
546         /* No overhead */
547         print_graph_duration(tr, 0, s, flags | FLAGS_FILL_START);
548
549         if (type == TRACE_GRAPH_ENT)
550                 trace_seq_puts(s, "==========>");
551         else
552                 trace_seq_puts(s, "<==========");
553
554         print_graph_duration(tr, 0, s, flags | FLAGS_FILL_END);
555         trace_seq_putc(s, '\n');
556 }
557
558 void
559 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
560 {
561         unsigned long nsecs_rem = do_div(duration, 1000);
562         /* log10(ULONG_MAX) + '\0' */
563         char usecs_str[21];
564         char nsecs_str[5];
565         int len;
566         int i;
567
568         sprintf(usecs_str, "%lu", (unsigned long) duration);
569
570         /* Print msecs */
571         trace_seq_printf(s, "%s", usecs_str);
572
573         len = strlen(usecs_str);
574
575         /* Print nsecs (we don't want to exceed 7 numbers) */
576         if (len < 7) {
577                 size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
578
579                 snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
580                 trace_seq_printf(s, ".%s", nsecs_str);
581                 len += strlen(nsecs_str) + 1;
582         }
583
584         trace_seq_puts(s, " us ");
585
586         /* Print remaining spaces to fit the row's width */
587         for (i = len; i < 8; i++)
588                 trace_seq_putc(s, ' ');
589 }
590
591 static void
592 print_graph_duration(struct trace_array *tr, unsigned long long duration,
593                      struct trace_seq *s, u32 flags)
594 {
595         if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
596             !(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
597                 return;
598
599         /* No real adata, just filling the column with spaces */
600         switch (flags & TRACE_GRAPH_PRINT_FILL_MASK) {
601         case FLAGS_FILL_FULL:
602                 trace_seq_puts(s, "              |  ");
603                 return;
604         case FLAGS_FILL_START:
605                 trace_seq_puts(s, "  ");
606                 return;
607         case FLAGS_FILL_END:
608                 trace_seq_puts(s, " |");
609                 return;
610         }
611
612         /* Signal a overhead of time execution to the output */
613         if (flags & TRACE_GRAPH_PRINT_OVERHEAD)
614                 trace_seq_printf(s, "%c ", trace_find_mark(duration));
615         else
616                 trace_seq_puts(s, "  ");
617
618         trace_print_graph_duration(duration, s);
619         trace_seq_puts(s, "|  ");
620 }
621
622 /* Case of a leaf function on its call entry */
623 static enum print_line_t
624 print_graph_entry_leaf(struct trace_iterator *iter,
625                 struct ftrace_graph_ent_entry *entry,
626                 struct ftrace_graph_ret_entry *ret_entry,
627                 struct trace_seq *s, u32 flags)
628 {
629         struct fgraph_data *data = iter->private;
630         struct trace_array *tr = iter->tr;
631         struct ftrace_graph_ret *graph_ret;
632         struct ftrace_graph_ent *call;
633         unsigned long long duration;
634         int cpu = iter->cpu;
635         int i;
636
637         graph_ret = &ret_entry->ret;
638         call = &entry->graph_ent;
639         duration = graph_ret->rettime - graph_ret->calltime;
640
641         if (data) {
642                 struct fgraph_cpu_data *cpu_data;
643
644                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
645
646                 /*
647                  * Comments display at + 1 to depth. Since
648                  * this is a leaf function, keep the comments
649                  * equal to this depth.
650                  */
651                 cpu_data->depth = call->depth - 1;
652
653                 /* No need to keep this function around for this depth */
654                 if (call->depth < FTRACE_RETFUNC_DEPTH &&
655                     !WARN_ON_ONCE(call->depth < 0))
656                         cpu_data->enter_funcs[call->depth] = 0;
657         }
658
659         /* Overhead and duration */
660         print_graph_duration(tr, duration, s, flags);
661
662         /* Function */
663         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
664                 trace_seq_putc(s, ' ');
665
666         trace_seq_printf(s, "%ps();\n", (void *)call->func);
667
668         print_graph_irq(iter, graph_ret->func, TRACE_GRAPH_RET,
669                         cpu, iter->ent->pid, flags);
670
671         return trace_handle_return(s);
672 }
673
674 static enum print_line_t
675 print_graph_entry_nested(struct trace_iterator *iter,
676                          struct ftrace_graph_ent_entry *entry,
677                          struct trace_seq *s, int cpu, u32 flags)
678 {
679         struct ftrace_graph_ent *call = &entry->graph_ent;
680         struct fgraph_data *data = iter->private;
681         struct trace_array *tr = iter->tr;
682         int i;
683
684         if (data) {
685                 struct fgraph_cpu_data *cpu_data;
686                 int cpu = iter->cpu;
687
688                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
689                 cpu_data->depth = call->depth;
690
691                 /* Save this function pointer to see if the exit matches */
692                 if (call->depth < FTRACE_RETFUNC_DEPTH &&
693                     !WARN_ON_ONCE(call->depth < 0))
694                         cpu_data->enter_funcs[call->depth] = call->func;
695         }
696
697         /* No time */
698         print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
699
700         /* Function */
701         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
702                 trace_seq_putc(s, ' ');
703
704         trace_seq_printf(s, "%ps() {\n", (void *)call->func);
705
706         if (trace_seq_has_overflowed(s))
707                 return TRACE_TYPE_PARTIAL_LINE;
708
709         /*
710          * we already consumed the current entry to check the next one
711          * and see if this is a leaf.
712          */
713         return TRACE_TYPE_NO_CONSUME;
714 }
715
716 static void
717 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
718                      int type, unsigned long addr, u32 flags)
719 {
720         struct fgraph_data *data = iter->private;
721         struct trace_entry *ent = iter->ent;
722         struct trace_array *tr = iter->tr;
723         int cpu = iter->cpu;
724
725         /* Pid */
726         verif_pid(s, ent->pid, cpu, data);
727
728         if (type)
729                 /* Interrupt */
730                 print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
731
732         if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
733                 return;
734
735         /* Absolute time */
736         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
737                 print_graph_abs_time(iter->ts, s);
738
739         /* Relative time */
740         if (flags & TRACE_GRAPH_PRINT_REL_TIME)
741                 print_graph_rel_time(iter, s);
742
743         /* Cpu */
744         if (flags & TRACE_GRAPH_PRINT_CPU)
745                 print_graph_cpu(s, cpu);
746
747         /* Proc */
748         if (flags & TRACE_GRAPH_PRINT_PROC) {
749                 print_graph_proc(s, ent->pid);
750                 trace_seq_puts(s, " | ");
751         }
752
753         /* Latency format */
754         if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
755                 print_graph_lat_fmt(s, ent);
756
757         return;
758 }
759
760 /*
761  * Entry check for irq code
762  *
763  * returns 1 if
764  *  - we are inside irq code
765  *  - we just entered irq code
766  *
767  * returns 0 if
768  *  - funcgraph-interrupts option is set
769  *  - we are not inside irq code
770  */
771 static int
772 check_irq_entry(struct trace_iterator *iter, u32 flags,
773                 unsigned long addr, int depth)
774 {
775         int cpu = iter->cpu;
776         int *depth_irq;
777         struct fgraph_data *data = iter->private;
778
779         /*
780          * If we are either displaying irqs, or we got called as
781          * a graph event and private data does not exist,
782          * then we bypass the irq check.
783          */
784         if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
785             (!data))
786                 return 0;
787
788         depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
789
790         /*
791          * We are inside the irq code
792          */
793         if (*depth_irq >= 0)
794                 return 1;
795
796         if ((addr < (unsigned long)__irqentry_text_start) ||
797             (addr >= (unsigned long)__irqentry_text_end))
798                 return 0;
799
800         /*
801          * We are entering irq code.
802          */
803         *depth_irq = depth;
804         return 1;
805 }
806
807 /*
808  * Return check for irq code
809  *
810  * returns 1 if
811  *  - we are inside irq code
812  *  - we just left irq code
813  *
814  * returns 0 if
815  *  - funcgraph-interrupts option is set
816  *  - we are not inside irq code
817  */
818 static int
819 check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
820 {
821         int cpu = iter->cpu;
822         int *depth_irq;
823         struct fgraph_data *data = iter->private;
824
825         /*
826          * If we are either displaying irqs, or we got called as
827          * a graph event and private data does not exist,
828          * then we bypass the irq check.
829          */
830         if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
831             (!data))
832                 return 0;
833
834         depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
835
836         /*
837          * We are not inside the irq code.
838          */
839         if (*depth_irq == -1)
840                 return 0;
841
842         /*
843          * We are inside the irq code, and this is returning entry.
844          * Let's not trace it and clear the entry depth, since
845          * we are out of irq code.
846          *
847          * This condition ensures that we 'leave the irq code' once
848          * we are out of the entry depth. Thus protecting us from
849          * the RETURN entry loss.
850          */
851         if (*depth_irq >= depth) {
852                 *depth_irq = -1;
853                 return 1;
854         }
855
856         /*
857          * We are inside the irq code, and this is not the entry.
858          */
859         return 1;
860 }
861
862 static enum print_line_t
863 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
864                         struct trace_iterator *iter, u32 flags)
865 {
866         struct fgraph_data *data = iter->private;
867         struct ftrace_graph_ent *call = &field->graph_ent;
868         struct ftrace_graph_ret_entry *leaf_ret;
869         static enum print_line_t ret;
870         int cpu = iter->cpu;
871
872         if (check_irq_entry(iter, flags, call->func, call->depth))
873                 return TRACE_TYPE_HANDLED;
874
875         print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags);
876
877         leaf_ret = get_return_for_leaf(iter, field);
878         if (leaf_ret)
879                 ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
880         else
881                 ret = print_graph_entry_nested(iter, field, s, cpu, flags);
882
883         if (data) {
884                 /*
885                  * If we failed to write our output, then we need to make
886                  * note of it. Because we already consumed our entry.
887                  */
888                 if (s->full) {
889                         data->failed = 1;
890                         data->cpu = cpu;
891                 } else
892                         data->failed = 0;
893         }
894
895         return ret;
896 }
897
898 static enum print_line_t
899 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
900                    struct trace_entry *ent, struct trace_iterator *iter,
901                    u32 flags)
902 {
903         unsigned long long duration = trace->rettime - trace->calltime;
904         struct fgraph_data *data = iter->private;
905         struct trace_array *tr = iter->tr;
906         pid_t pid = ent->pid;
907         int cpu = iter->cpu;
908         int func_match = 1;
909         int i;
910
911         if (check_irq_return(iter, flags, trace->depth))
912                 return TRACE_TYPE_HANDLED;
913
914         if (data) {
915                 struct fgraph_cpu_data *cpu_data;
916                 int cpu = iter->cpu;
917
918                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
919
920                 /*
921                  * Comments display at + 1 to depth. This is the
922                  * return from a function, we now want the comments
923                  * to display at the same level of the bracket.
924                  */
925                 cpu_data->depth = trace->depth - 1;
926
927                 if (trace->depth < FTRACE_RETFUNC_DEPTH &&
928                     !WARN_ON_ONCE(trace->depth < 0)) {
929                         if (cpu_data->enter_funcs[trace->depth] != trace->func)
930                                 func_match = 0;
931                         cpu_data->enter_funcs[trace->depth] = 0;
932                 }
933         }
934
935         print_graph_prologue(iter, s, 0, 0, flags);
936
937         /* Overhead and duration */
938         print_graph_duration(tr, duration, s, flags);
939
940         /* Closing brace */
941         for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++)
942                 trace_seq_putc(s, ' ');
943
944         /*
945          * If the return function does not have a matching entry,
946          * then the entry was lost. Instead of just printing
947          * the '}' and letting the user guess what function this
948          * belongs to, write out the function name. Always do
949          * that if the funcgraph-tail option is enabled.
950          */
951         if (func_match && !(flags & TRACE_GRAPH_PRINT_TAIL))
952                 trace_seq_puts(s, "}\n");
953         else
954                 trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
955
956         /* Overrun */
957         if (flags & TRACE_GRAPH_PRINT_OVERRUN)
958                 trace_seq_printf(s, " (Overruns: %u)\n",
959                                  trace->overrun);
960
961         print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
962                         cpu, pid, flags);
963
964         return trace_handle_return(s);
965 }
966
967 static enum print_line_t
968 print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
969                     struct trace_iterator *iter, u32 flags)
970 {
971         struct trace_array *tr = iter->tr;
972         unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
973         struct fgraph_data *data = iter->private;
974         struct trace_event *event;
975         int depth = 0;
976         int ret;
977         int i;
978
979         if (data)
980                 depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
981
982         print_graph_prologue(iter, s, 0, 0, flags);
983
984         /* No time */
985         print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
986
987         /* Indentation */
988         if (depth > 0)
989                 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++)
990                         trace_seq_putc(s, ' ');
991
992         /* The comment */
993         trace_seq_puts(s, "/* ");
994
995         switch (iter->ent->type) {
996         case TRACE_BPUTS:
997                 ret = trace_print_bputs_msg_only(iter);
998                 if (ret != TRACE_TYPE_HANDLED)
999                         return ret;
1000                 break;
1001         case TRACE_BPRINT:
1002                 ret = trace_print_bprintk_msg_only(iter);
1003                 if (ret != TRACE_TYPE_HANDLED)
1004                         return ret;
1005                 break;
1006         case TRACE_PRINT:
1007                 ret = trace_print_printk_msg_only(iter);
1008                 if (ret != TRACE_TYPE_HANDLED)
1009                         return ret;
1010                 break;
1011         default:
1012                 event = ftrace_find_event(ent->type);
1013                 if (!event)
1014                         return TRACE_TYPE_UNHANDLED;
1015
1016                 ret = event->funcs->trace(iter, sym_flags, event);
1017                 if (ret != TRACE_TYPE_HANDLED)
1018                         return ret;
1019         }
1020
1021         if (trace_seq_has_overflowed(s))
1022                 goto out;
1023
1024         /* Strip ending newline */
1025         if (s->buffer[s->seq.len - 1] == '\n') {
1026                 s->buffer[s->seq.len - 1] = '\0';
1027                 s->seq.len--;
1028         }
1029
1030         trace_seq_puts(s, " */\n");
1031  out:
1032         return trace_handle_return(s);
1033 }
1034
1035
1036 enum print_line_t
1037 print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1038 {
1039         struct ftrace_graph_ent_entry *field;
1040         struct fgraph_data *data = iter->private;
1041         struct trace_entry *entry = iter->ent;
1042         struct trace_seq *s = &iter->seq;
1043         int cpu = iter->cpu;
1044         int ret;
1045
1046         if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1047                 per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1048                 return TRACE_TYPE_HANDLED;
1049         }
1050
1051         /*
1052          * If the last output failed, there's a possibility we need
1053          * to print out the missing entry which would never go out.
1054          */
1055         if (data && data->failed) {
1056                 field = &data->ent;
1057                 iter->cpu = data->cpu;
1058                 ret = print_graph_entry(field, s, iter, flags);
1059                 if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1060                         per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1061                         ret = TRACE_TYPE_NO_CONSUME;
1062                 }
1063                 iter->cpu = cpu;
1064                 return ret;
1065         }
1066
1067         switch (entry->type) {
1068         case TRACE_GRAPH_ENT: {
1069                 /*
1070                  * print_graph_entry() may consume the current event,
1071                  * thus @field may become invalid, so we need to save it.
1072                  * sizeof(struct ftrace_graph_ent_entry) is very small,
1073                  * it can be safely saved at the stack.
1074                  */
1075                 struct ftrace_graph_ent_entry saved;
1076                 trace_assign_type(field, entry);
1077                 saved = *field;
1078                 return print_graph_entry(&saved, s, iter, flags);
1079         }
1080         case TRACE_GRAPH_RET: {
1081                 struct ftrace_graph_ret_entry *field;
1082                 trace_assign_type(field, entry);
1083                 return print_graph_return(&field->ret, s, entry, iter, flags);
1084         }
1085         case TRACE_STACK:
1086         case TRACE_FN:
1087                 /* dont trace stack and functions as comments */
1088                 return TRACE_TYPE_UNHANDLED;
1089
1090         default:
1091                 return print_graph_comment(s, entry, iter, flags);
1092         }
1093
1094         return TRACE_TYPE_HANDLED;
1095 }
1096
1097 static enum print_line_t
1098 print_graph_function(struct trace_iterator *iter)
1099 {
1100         return print_graph_function_flags(iter, tracer_flags.val);
1101 }
1102
1103 static enum print_line_t
1104 print_graph_function_event(struct trace_iterator *iter, int flags,
1105                            struct trace_event *event)
1106 {
1107         return print_graph_function(iter);
1108 }
1109
1110 static void print_lat_header(struct seq_file *s, u32 flags)
1111 {
1112         static const char spaces[] = "                " /* 16 spaces */
1113                 "    "                                  /* 4 spaces */
1114                 "                 ";                    /* 17 spaces */
1115         int size = 0;
1116
1117         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1118                 size += 16;
1119         if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1120                 size += 16;
1121         if (flags & TRACE_GRAPH_PRINT_CPU)
1122                 size += 4;
1123         if (flags & TRACE_GRAPH_PRINT_PROC)
1124                 size += 17;
1125
1126         seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
1127         seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
1128         seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1129         seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
1130         seq_printf(s, "#%.*s||| /                      \n", size, spaces);
1131 }
1132
1133 static void __print_graph_headers_flags(struct trace_array *tr,
1134                                         struct seq_file *s, u32 flags)
1135 {
1136         int lat = tr->trace_flags & TRACE_ITER_LATENCY_FMT;
1137
1138         if (lat)
1139                 print_lat_header(s, flags);
1140
1141         /* 1st line */
1142         seq_putc(s, '#');
1143         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1144                 seq_puts(s, "     TIME       ");
1145         if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1146                 seq_puts(s, "   REL TIME     ");
1147         if (flags & TRACE_GRAPH_PRINT_CPU)
1148                 seq_puts(s, " CPU");
1149         if (flags & TRACE_GRAPH_PRINT_PROC)
1150                 seq_puts(s, "  TASK/PID       ");
1151         if (lat)
1152                 seq_puts(s, "||||   ");
1153         if (flags & TRACE_GRAPH_PRINT_DURATION)
1154                 seq_puts(s, "  DURATION   ");
1155         seq_puts(s, "               FUNCTION CALLS\n");
1156
1157         /* 2nd line */
1158         seq_putc(s, '#');
1159         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1160                 seq_puts(s, "      |         ");
1161         if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1162                 seq_puts(s, "      |         ");
1163         if (flags & TRACE_GRAPH_PRINT_CPU)
1164                 seq_puts(s, " |  ");
1165         if (flags & TRACE_GRAPH_PRINT_PROC)
1166                 seq_puts(s, "   |    |        ");
1167         if (lat)
1168                 seq_puts(s, "||||   ");
1169         if (flags & TRACE_GRAPH_PRINT_DURATION)
1170                 seq_puts(s, "   |   |      ");
1171         seq_puts(s, "               |   |   |   |\n");
1172 }
1173
1174 static void print_graph_headers(struct seq_file *s)
1175 {
1176         print_graph_headers_flags(s, tracer_flags.val);
1177 }
1178
1179 void print_graph_headers_flags(struct seq_file *s, u32 flags)
1180 {
1181         struct trace_iterator *iter = s->private;
1182         struct trace_array *tr = iter->tr;
1183
1184         if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
1185                 return;
1186
1187         if (tr->trace_flags & TRACE_ITER_LATENCY_FMT) {
1188                 /* print nothing if the buffers are empty */
1189                 if (trace_empty(iter))
1190                         return;
1191
1192                 print_trace_header(s, iter);
1193         }
1194
1195         __print_graph_headers_flags(tr, s, flags);
1196 }
1197
1198 void graph_trace_open(struct trace_iterator *iter)
1199 {
1200         /* pid and depth on the last trace processed */
1201         struct fgraph_data *data;
1202         gfp_t gfpflags;
1203         int cpu;
1204
1205         iter->private = NULL;
1206
1207         /* We can be called in atomic context via ftrace_dump() */
1208         gfpflags = (in_atomic() || irqs_disabled()) ? GFP_ATOMIC : GFP_KERNEL;
1209
1210         data = kzalloc(sizeof(*data), gfpflags);
1211         if (!data)
1212                 goto out_err;
1213
1214         data->cpu_data = alloc_percpu_gfp(struct fgraph_cpu_data, gfpflags);
1215         if (!data->cpu_data)
1216                 goto out_err_free;
1217
1218         for_each_possible_cpu(cpu) {
1219                 pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1220                 int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1221                 int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1222                 int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1223
1224                 *pid = -1;
1225                 *depth = 0;
1226                 *ignore = 0;
1227                 *depth_irq = -1;
1228         }
1229
1230         iter->private = data;
1231
1232         return;
1233
1234  out_err_free:
1235         kfree(data);
1236  out_err:
1237         pr_warn("function graph tracer: not enough memory\n");
1238 }
1239
1240 void graph_trace_close(struct trace_iterator *iter)
1241 {
1242         struct fgraph_data *data = iter->private;
1243
1244         if (data) {
1245                 free_percpu(data->cpu_data);
1246                 kfree(data);
1247         }
1248 }
1249
1250 static int
1251 func_graph_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1252 {
1253         if (bit == TRACE_GRAPH_PRINT_IRQS)
1254                 ftrace_graph_skip_irqs = !set;
1255
1256         if (bit == TRACE_GRAPH_SLEEP_TIME)
1257                 ftrace_graph_sleep_time_control(set);
1258
1259         if (bit == TRACE_GRAPH_GRAPH_TIME)
1260                 ftrace_graph_graph_time_control(set);
1261
1262         return 0;
1263 }
1264
1265 static struct trace_event_functions graph_functions = {
1266         .trace          = print_graph_function_event,
1267 };
1268
1269 static struct trace_event graph_trace_entry_event = {
1270         .type           = TRACE_GRAPH_ENT,
1271         .funcs          = &graph_functions,
1272 };
1273
1274 static struct trace_event graph_trace_ret_event = {
1275         .type           = TRACE_GRAPH_RET,
1276         .funcs          = &graph_functions
1277 };
1278
1279 static struct tracer graph_trace __tracer_data = {
1280         .name           = "function_graph",
1281         .update_thresh  = graph_trace_update_thresh,
1282         .open           = graph_trace_open,
1283         .pipe_open      = graph_trace_open,
1284         .close          = graph_trace_close,
1285         .pipe_close     = graph_trace_close,
1286         .init           = graph_trace_init,
1287         .reset          = graph_trace_reset,
1288         .print_line     = print_graph_function,
1289         .print_header   = print_graph_headers,
1290         .flags          = &tracer_flags,
1291         .set_flag       = func_graph_set_flag,
1292 #ifdef CONFIG_FTRACE_SELFTEST
1293         .selftest       = trace_selftest_startup_function_graph,
1294 #endif
1295 };
1296
1297
1298 static ssize_t
1299 graph_depth_write(struct file *filp, const char __user *ubuf, size_t cnt,
1300                   loff_t *ppos)
1301 {
1302         unsigned long val;
1303         int ret;
1304
1305         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1306         if (ret)
1307                 return ret;
1308
1309         fgraph_max_depth = val;
1310
1311         *ppos += cnt;
1312
1313         return cnt;
1314 }
1315
1316 static ssize_t
1317 graph_depth_read(struct file *filp, char __user *ubuf, size_t cnt,
1318                  loff_t *ppos)
1319 {
1320         char buf[15]; /* More than enough to hold UINT_MAX + "\n"*/
1321         int n;
1322
1323         n = sprintf(buf, "%d\n", fgraph_max_depth);
1324
1325         return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
1326 }
1327
1328 static const struct file_operations graph_depth_fops = {
1329         .open           = tracing_open_generic,
1330         .write          = graph_depth_write,
1331         .read           = graph_depth_read,
1332         .llseek         = generic_file_llseek,
1333 };
1334
1335 static __init int init_graph_tracefs(void)
1336 {
1337         int ret;
1338
1339         ret = tracing_init_dentry();
1340         if (ret)
1341                 return 0;
1342
1343         trace_create_file("max_graph_depth", 0644, NULL,
1344                           NULL, &graph_depth_fops);
1345
1346         return 0;
1347 }
1348 fs_initcall(init_graph_tracefs);
1349
1350 static __init int init_graph_trace(void)
1351 {
1352         max_bytes_for_cpu = snprintf(NULL, 0, "%u", nr_cpu_ids - 1);
1353
1354         if (!register_trace_event(&graph_trace_entry_event)) {
1355                 pr_warn("Warning: could not register graph trace events\n");
1356                 return 1;
1357         }
1358
1359         if (!register_trace_event(&graph_trace_ret_event)) {
1360                 pr_warn("Warning: could not register graph trace events\n");
1361                 return 1;
1362         }
1363
1364         return register_tracer(&graph_trace);
1365 }
1366
1367 core_initcall(init_graph_trace);