net_sched: sfb: optimize enqueue on full queue
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / trace / trace_functions_graph.c
bloba7d2a4c653d8d893f51652c8cd599bf780359f45
1 /*
3 * Function graph tracer.
4 * Copyright (c) 2008-2009 Frederic Weisbecker <fweisbec@gmail.com>
5 * Mostly borrowed from function tracer which
6 * is Copyright (c) Steven Rostedt <srostedt@redhat.com>
8 */
9 #include <linux/debugfs.h>
10 #include <linux/uaccess.h>
11 #include <linux/ftrace.h>
12 #include <linux/slab.h>
13 #include <linux/fs.h>
15 #include "trace.h"
16 #include "trace_output.h"
18 /* When set, irq functions will be ignored */
19 static int ftrace_graph_skip_irqs;
21 struct fgraph_cpu_data {
22 pid_t last_pid;
23 int depth;
24 int depth_irq;
25 int ignore;
26 unsigned long enter_funcs[FTRACE_RETFUNC_DEPTH];
29 struct fgraph_data {
30 struct fgraph_cpu_data __percpu *cpu_data;
32 /* Place to preserve last processed entry. */
33 struct ftrace_graph_ent_entry ent;
34 struct ftrace_graph_ret_entry ret;
35 int failed;
36 int cpu;
39 #define TRACE_GRAPH_INDENT 2
41 /* Flag options */
42 #define TRACE_GRAPH_PRINT_OVERRUN 0x1
43 #define TRACE_GRAPH_PRINT_CPU 0x2
44 #define TRACE_GRAPH_PRINT_OVERHEAD 0x4
45 #define TRACE_GRAPH_PRINT_PROC 0x8
46 #define TRACE_GRAPH_PRINT_DURATION 0x10
47 #define TRACE_GRAPH_PRINT_ABS_TIME 0x20
48 #define TRACE_GRAPH_PRINT_IRQS 0x40
50 static 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 */
68 static 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
75 static struct trace_array *graph_array;
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.
82 enum {
83 DURATION_FILL_FULL = -1,
84 DURATION_FILL_START = -2,
85 DURATION_FILL_END = -3,
88 static enum print_line_t
89 print_graph_duration(unsigned long long duration, struct trace_seq *s,
90 u32 flags);
92 /* Add a function return address to the trace stack on thread info.*/
93 int
94 ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
95 unsigned long frame_pointer)
97 unsigned long long calltime;
98 int index;
100 if (!current->ret_stack)
101 return -EBUSY;
104 * We must make sure the ret_stack is tested before we read
105 * anything else.
107 smp_rmb();
109 /* The return trace stack is full */
110 if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
111 atomic_inc(&current->trace_overrun);
112 return -EBUSY;
115 calltime = trace_clock_local();
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;
126 return 0;
129 /* Retrieve a function return address to the trace stack on thread info.*/
130 static void
131 ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
132 unsigned long frame_pointer)
134 int index;
136 index = current->curr_ret_stack;
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;
146 #ifdef CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST
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.
155 * Currently, x86_32 with optimize for size (-Os) makes the latest
156 * gcc do the above.
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;
169 #endif
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(&current->trace_overrun);
175 trace->depth = index;
179 * Send the trace to the ring-buffer.
180 * @return the original return address.
182 unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
184 struct ftrace_graph_ret trace;
185 unsigned long ret;
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--;
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;
200 return ret;
203 int __trace_graph_entry(struct trace_array *tr,
204 struct ftrace_graph_ent *trace,
205 unsigned long flags,
206 int pc)
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;
213 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
214 return 0;
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);
225 return 1;
228 static inline int ftrace_graph_ignore_irqs(void)
230 if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
231 return 0;
233 return in_irq();
236 int trace_graph_entry(struct ftrace_graph_ent *trace)
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;
246 if (!ftrace_trace_task(current))
247 return 0;
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;
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;
265 atomic_dec(&data->disabled);
266 local_irq_restore(flags);
268 return ret;
271 int trace_graph_thresh_entry(struct ftrace_graph_ent *trace)
273 if (tracing_thresh)
274 return 1;
275 else
276 return trace_graph_entry(trace);
279 static void
280 __trace_graph_function(struct trace_array *tr,
281 unsigned long ip, unsigned long flags, int pc)
283 u64 time = trace_clock_local();
284 struct ftrace_graph_ent ent = {
285 .func = ip,
286 .depth = 0,
288 struct ftrace_graph_ret ret = {
289 .func = ip,
290 .depth = 0,
291 .calltime = time,
292 .rettime = time,
295 __trace_graph_entry(tr, &ent, flags, pc);
296 __trace_graph_return(tr, &ret, flags, pc);
299 void
300 trace_graph_function(struct trace_array *tr,
301 unsigned long ip, unsigned long parent_ip,
302 unsigned long flags, int pc)
304 __trace_graph_function(tr, ip, flags, pc);
307 void __trace_graph_return(struct trace_array *tr,
308 struct ftrace_graph_ret *trace,
309 unsigned long flags,
310 int pc)
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;
317 if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
318 return;
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);
330 void trace_graph_return(struct ftrace_graph_ret *trace)
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;
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);
347 atomic_dec(&data->disabled);
348 local_irq_restore(flags);
351 void set_graph_array(struct trace_array *tr)
353 graph_array = tr;
355 /* Make graph_array visible before we start tracing */
357 smp_mb();
360 void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
362 if (tracing_thresh &&
363 (trace->rettime - trace->calltime < tracing_thresh))
364 return;
365 else
366 trace_graph_return(trace);
369 static int graph_trace_init(struct trace_array *tr)
371 int ret;
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();
384 return 0;
387 static void graph_trace_reset(struct trace_array *tr)
389 tracing_stop_cmdline_record();
390 unregister_ftrace_graph();
393 static int max_bytes_for_cpu;
395 static enum print_line_t
396 print_graph_cpu(struct trace_seq *s, int cpu)
398 int ret;
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:
405 ret = trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
406 if (!ret)
407 return TRACE_TYPE_PARTIAL_LINE;
409 return TRACE_TYPE_HANDLED;
412 #define TRACE_GRAPH_PROCINFO_LENGTH 14
414 static enum print_line_t
415 print_graph_proc(struct trace_seq *s, pid_t pid)
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;
425 trace_find_cmdline(pid, comm);
426 comm[7] = '\0';
427 sprintf(pid_str, "%d", pid);
429 /* 1 stands for the "-" character */
430 len = strlen(comm) + strlen(pid_str) + 1;
432 if (len < TRACE_GRAPH_PROCINFO_LENGTH)
433 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
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;
442 ret = trace_seq_printf(s, "%s-%s", comm, pid_str);
443 if (!ret)
444 return TRACE_TYPE_PARTIAL_LINE;
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;
452 return TRACE_TYPE_HANDLED;
456 static enum print_line_t
457 print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
459 if (!trace_seq_putc(s, ' '))
460 return 0;
462 return trace_print_lat_fmt(s, entry);
465 /* If the pid changed since the last trace, output this event */
466 static enum print_line_t
467 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
469 pid_t prev_pid;
470 pid_t *last_pid;
471 int ret;
473 if (!data)
474 return TRACE_TYPE_HANDLED;
476 last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
478 if (*last_pid == pid)
479 return TRACE_TYPE_HANDLED;
481 prev_pid = *last_pid;
482 *last_pid = pid;
484 if (prev_pid == -1)
485 return TRACE_TYPE_HANDLED;
487 * Context-switch trace line:
489 ------------------------------------------
490 | 1) migration/0--1 => sshd-1755
491 ------------------------------------------
494 ret = trace_seq_printf(s,
495 " ------------------------------------------\n");
496 if (!ret)
497 return TRACE_TYPE_PARTIAL_LINE;
499 ret = print_graph_cpu(s, cpu);
500 if (ret == TRACE_TYPE_PARTIAL_LINE)
501 return TRACE_TYPE_PARTIAL_LINE;
503 ret = print_graph_proc(s, prev_pid);
504 if (ret == TRACE_TYPE_PARTIAL_LINE)
505 return TRACE_TYPE_PARTIAL_LINE;
507 ret = trace_seq_printf(s, " => ");
508 if (!ret)
509 return TRACE_TYPE_PARTIAL_LINE;
511 ret = print_graph_proc(s, pid);
512 if (ret == TRACE_TYPE_PARTIAL_LINE)
513 return TRACE_TYPE_PARTIAL_LINE;
515 ret = trace_seq_printf(s,
516 "\n ------------------------------------------\n\n");
517 if (!ret)
518 return TRACE_TYPE_PARTIAL_LINE;
520 return TRACE_TYPE_HANDLED;
523 static struct ftrace_graph_ret_entry *
524 get_return_for_leaf(struct trace_iterator *iter,
525 struct ftrace_graph_ent_entry *curr)
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;
533 * If the previous output failed to write to the seq buffer,
534 * then we just reuse the data from before.
536 if (data && data->failed) {
537 curr = &data->ent;
538 next = &data->ret;
539 } else {
541 ring_iter = iter->buffer_iter[iter->cpu];
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 {
548 * We need to consume the current entry to see
549 * the next one.
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);
557 if (!event)
558 return NULL;
560 next = ring_buffer_event_data(event);
562 if (data) {
564 * Save current and next entries for later reference
565 * if the output fails.
567 data->ent = *curr;
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.
573 if (next->ent.type == TRACE_GRAPH_RET)
574 data->ret = *next;
575 else
576 data->ret.ent.type = next->ent.type;
580 if (next->ent.type != TRACE_GRAPH_RET)
581 return NULL;
583 if (curr->ent.pid != next->ent.pid ||
584 curr->graph_ent.func != next->ret.func)
585 return NULL;
587 /* this is a leaf, now advance the iterator */
588 if (ring_iter)
589 ring_buffer_read(ring_iter, NULL);
591 return next;
594 static int print_graph_abs_time(u64 t, struct trace_seq *s)
596 unsigned long usecs_rem;
598 usecs_rem = do_div(t, NSEC_PER_SEC);
599 usecs_rem /= 1000;
601 return trace_seq_printf(s, "%5lu.%06lu | ",
602 (unsigned long)t, usecs_rem);
605 static enum print_line_t
606 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
607 enum trace_type type, int cpu, pid_t pid, u32 flags)
609 int ret;
610 struct trace_seq *s = &iter->seq;
612 if (addr < (unsigned long)__irqentry_text_start ||
613 addr >= (unsigned long)__irqentry_text_end)
614 return TRACE_TYPE_UNHANDLED;
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;
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;
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;
642 /* No overhead */
643 ret = print_graph_duration(DURATION_FILL_START, s, flags);
644 if (ret != TRACE_TYPE_HANDLED)
645 return ret;
647 if (type == TRACE_GRAPH_ENT)
648 ret = trace_seq_printf(s, "==========>");
649 else
650 ret = trace_seq_printf(s, "<==========");
652 if (!ret)
653 return TRACE_TYPE_PARTIAL_LINE;
655 ret = print_graph_duration(DURATION_FILL_END, s, flags);
656 if (ret != TRACE_TYPE_HANDLED)
657 return ret;
659 ret = trace_seq_printf(s, "\n");
661 if (!ret)
662 return TRACE_TYPE_PARTIAL_LINE;
663 return TRACE_TYPE_HANDLED;
666 enum print_line_t
667 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
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;
676 sprintf(msecs_str, "%lu", (unsigned long) duration);
678 /* Print msecs */
679 ret = trace_seq_printf(s, "%s", msecs_str);
680 if (!ret)
681 return TRACE_TYPE_PARTIAL_LINE;
683 len = strlen(msecs_str);
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);
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);
696 ret = trace_seq_printf(s, " us ");
697 if (!ret)
698 return TRACE_TYPE_PARTIAL_LINE;
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;
706 return TRACE_TYPE_HANDLED;
709 static enum print_line_t
710 print_graph_duration(unsigned long long duration, struct trace_seq *s,
711 u32 flags)
713 int ret = -1;
715 if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
716 !(trace_flags & TRACE_ITER_CONTEXT_INFO))
717 return TRACE_TYPE_HANDLED;
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;
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, "+ ");
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.
747 if (ret == -1)
748 ret = trace_seq_printf(s, " ");
750 /* Catching here any failure happenned above */
751 if (!ret)
752 return TRACE_TYPE_PARTIAL_LINE;
754 ret = trace_print_graph_duration(duration, s);
755 if (ret != TRACE_TYPE_HANDLED)
756 return ret;
758 ret = trace_seq_printf(s, "| ");
759 if (!ret)
760 return TRACE_TYPE_PARTIAL_LINE;
762 return TRACE_TYPE_HANDLED;
765 /* Case of a leaf function on its call entry */
766 static enum print_line_t
767 print_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)
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;
779 graph_ret = &ret_entry->ret;
780 call = &entry->graph_ent;
781 duration = graph_ret->rettime - graph_ret->calltime;
783 if (data) {
784 struct fgraph_cpu_data *cpu_data;
785 int cpu = iter->cpu;
787 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
790 * Comments display at + 1 to depth. Since
791 * this is a leaf function, keep the comments
792 * equal to this depth.
794 cpu_data->depth = call->depth - 1;
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;
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;
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;
813 ret = trace_seq_printf(s, "%ps();\n", (void *)call->func);
814 if (!ret)
815 return TRACE_TYPE_PARTIAL_LINE;
817 return TRACE_TYPE_HANDLED;
820 static enum print_line_t
821 print_graph_entry_nested(struct trace_iterator *iter,
822 struct ftrace_graph_ent_entry *entry,
823 struct trace_seq *s, int cpu, u32 flags)
825 struct ftrace_graph_ent *call = &entry->graph_ent;
826 struct fgraph_data *data = iter->private;
827 int ret;
828 int i;
830 if (data) {
831 struct fgraph_cpu_data *cpu_data;
832 int cpu = iter->cpu;
834 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
835 cpu_data->depth = call->depth;
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;
842 /* No time */
843 ret = print_graph_duration(DURATION_FILL_FULL, s, flags);
844 if (ret != TRACE_TYPE_HANDLED)
845 return ret;
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;
854 ret = trace_seq_printf(s, "%ps() {\n", (void *)call->func);
855 if (!ret)
856 return TRACE_TYPE_PARTIAL_LINE;
859 * we already consumed the current entry to check the next one
860 * and see if this is a leaf.
862 return TRACE_TYPE_NO_CONSUME;
865 static enum print_line_t
866 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
867 int type, unsigned long addr, u32 flags)
869 struct fgraph_data *data = iter->private;
870 struct trace_entry *ent = iter->ent;
871 int cpu = iter->cpu;
872 int ret;
874 /* Pid */
875 if (verif_pid(s, ent->pid, cpu, data) == TRACE_TYPE_PARTIAL_LINE)
876 return TRACE_TYPE_PARTIAL_LINE;
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;
885 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
886 return 0;
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;
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;
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;
908 ret = trace_seq_printf(s, " | ");
909 if (!ret)
910 return TRACE_TYPE_PARTIAL_LINE;
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;
920 return 0;
924 * Entry check for irq code
926 * returns 1 if
927 * - we are inside irq code
928 * - we just entered irq code
930 * retunns 0 if
931 * - funcgraph-interrupts option is set
932 * - we are not inside irq code
934 static int
935 check_irq_entry(struct trace_iterator *iter, u32 flags,
936 unsigned long addr, int depth)
938 int cpu = iter->cpu;
939 int *depth_irq;
940 struct fgraph_data *data = iter->private;
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.
947 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
948 (!data))
949 return 0;
951 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
954 * We are inside the irq code
956 if (*depth_irq >= 0)
957 return 1;
959 if ((addr < (unsigned long)__irqentry_text_start) ||
960 (addr >= (unsigned long)__irqentry_text_end))
961 return 0;
964 * We are entering irq code.
966 *depth_irq = depth;
967 return 1;
971 * Return check for irq code
973 * returns 1 if
974 * - we are inside irq code
975 * - we just left irq code
977 * returns 0 if
978 * - funcgraph-interrupts option is set
979 * - we are not inside irq code
981 static int
982 check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
984 int cpu = iter->cpu;
985 int *depth_irq;
986 struct fgraph_data *data = iter->private;
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.
993 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
994 (!data))
995 return 0;
997 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1000 * We are not inside the irq code.
1002 if (*depth_irq == -1)
1003 return 0;
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.
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.
1014 if (*depth_irq >= depth) {
1015 *depth_irq = -1;
1016 return 1;
1020 * We are inside the irq code, and this is not the entry.
1022 return 1;
1025 static enum print_line_t
1026 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
1027 struct trace_iterator *iter, u32 flags)
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;
1035 if (check_irq_entry(iter, flags, call->func, call->depth))
1036 return TRACE_TYPE_HANDLED;
1038 if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags))
1039 return TRACE_TYPE_PARTIAL_LINE;
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);
1047 if (data) {
1049 * If we failed to write our output, then we need to make
1050 * note of it. Because we already consumed our entry.
1052 if (s->full) {
1053 data->failed = 1;
1054 data->cpu = cpu;
1055 } else
1056 data->failed = 0;
1059 return ret;
1062 static enum print_line_t
1063 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
1064 struct trace_entry *ent, struct trace_iterator *iter,
1065 u32 flags)
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;
1075 if (check_irq_return(iter, flags, trace->depth))
1076 return TRACE_TYPE_HANDLED;
1078 if (data) {
1079 struct fgraph_cpu_data *cpu_data;
1080 int cpu = iter->cpu;
1082 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
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.
1089 cpu_data->depth = trace->depth - 1;
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;
1098 if (print_graph_prologue(iter, s, 0, 0, flags))
1099 return TRACE_TYPE_PARTIAL_LINE;
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;
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;
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.
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;
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;
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;
1142 return TRACE_TYPE_HANDLED;
1145 static enum print_line_t
1146 print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1147 struct trace_iterator *iter, u32 flags)
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;
1156 if (data)
1157 depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1159 if (print_graph_prologue(iter, s, 0, 0, flags))
1160 return TRACE_TYPE_PARTIAL_LINE;
1162 /* No time */
1163 ret = print_graph_duration(DURATION_FILL_FULL, s, flags);
1164 if (ret != TRACE_TYPE_HANDLED)
1165 return ret;
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;
1175 /* The comment */
1176 ret = trace_seq_printf(s, "/* ");
1177 if (!ret)
1178 return TRACE_TYPE_PARTIAL_LINE;
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;
1196 ret = event->funcs->trace(iter, sym_flags, event);
1197 if (ret != TRACE_TYPE_HANDLED)
1198 return ret;
1201 /* Strip ending newline */
1202 if (s->buffer[s->len - 1] == '\n') {
1203 s->buffer[s->len - 1] = '\0';
1204 s->len--;
1207 ret = trace_seq_printf(s, " */\n");
1208 if (!ret)
1209 return TRACE_TYPE_PARTIAL_LINE;
1211 return TRACE_TYPE_HANDLED;
1215 enum print_line_t
1216 print_graph_function_flags(struct trace_iterator *iter, u32 flags)
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;
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;
1231 * If the last output failed, there's a possibility we need
1232 * to print out the missing entry which would never go out.
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;
1242 iter->cpu = cpu;
1243 return ret;
1246 switch (entry->type) {
1247 case TRACE_GRAPH_ENT: {
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.
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);
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);
1264 case TRACE_STACK:
1265 case TRACE_FN:
1266 /* dont trace stack and functions as comments */
1267 return TRACE_TYPE_UNHANDLED;
1269 default:
1270 return print_graph_comment(s, entry, iter, flags);
1273 return TRACE_TYPE_HANDLED;
1276 static enum print_line_t
1277 print_graph_function(struct trace_iterator *iter)
1279 return print_graph_function_flags(iter, tracer_flags.val);
1282 static enum print_line_t
1283 print_graph_function_event(struct trace_iterator *iter, int flags,
1284 struct trace_event *event)
1286 return print_graph_function(iter);
1289 static void print_lat_header(struct seq_file *s, u32 flags)
1291 static const char spaces[] = " " /* 16 spaces */
1292 " " /* 4 spaces */
1293 " "; /* 17 spaces */
1294 int size = 0;
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;
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);
1310 static void __print_graph_headers_flags(struct seq_file *s, u32 flags)
1312 int lat = trace_flags & TRACE_ITER_LATENCY_FMT;
1314 if (lat)
1315 print_lat_header(s, flags);
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");
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");
1346 void print_graph_headers(struct seq_file *s)
1348 print_graph_headers_flags(s, tracer_flags.val);
1351 void print_graph_headers_flags(struct seq_file *s, u32 flags)
1353 struct trace_iterator *iter = s->private;
1355 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
1356 return;
1358 if (trace_flags & TRACE_ITER_LATENCY_FMT) {
1359 /* print nothing if the buffers are empty */
1360 if (trace_empty(iter))
1361 return;
1363 print_trace_header(s, iter);
1366 __print_graph_headers_flags(s, flags);
1369 void graph_trace_open(struct trace_iterator *iter)
1371 /* pid and depth on the last trace processed */
1372 struct fgraph_data *data;
1373 int cpu;
1375 iter->private = NULL;
1377 data = kzalloc(sizeof(*data), GFP_KERNEL);
1378 if (!data)
1379 goto out_err;
1381 data->cpu_data = alloc_percpu(struct fgraph_cpu_data);
1382 if (!data->cpu_data)
1383 goto out_err_free;
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);
1391 *pid = -1;
1392 *depth = 0;
1393 *ignore = 0;
1394 *depth_irq = -1;
1397 iter->private = data;
1399 return;
1401 out_err_free:
1402 kfree(data);
1403 out_err:
1404 pr_warning("function graph tracer: not enough memory\n");
1407 void graph_trace_close(struct trace_iterator *iter)
1409 struct fgraph_data *data = iter->private;
1411 if (data) {
1412 free_percpu(data->cpu_data);
1413 kfree(data);
1417 static int func_graph_set_flag(u32 old_flags, u32 bit, int set)
1419 if (bit == TRACE_GRAPH_PRINT_IRQS)
1420 ftrace_graph_skip_irqs = !set;
1422 return 0;
1425 static struct trace_event_functions graph_functions = {
1426 .trace = print_graph_function_event,
1429 static struct trace_event graph_trace_entry_event = {
1430 .type = TRACE_GRAPH_ENT,
1431 .funcs = &graph_functions,
1434 static struct trace_event graph_trace_ret_event = {
1435 .type = TRACE_GRAPH_RET,
1436 .funcs = &graph_functions
1439 static 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
1457 static __init int init_graph_trace(void)
1459 max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1461 if (!register_ftrace_event(&graph_trace_entry_event)) {
1462 pr_warning("Warning: could not register graph trace events\n");
1463 return 1;
1466 if (!register_ftrace_event(&graph_trace_ret_event)) {
1467 pr_warning("Warning: could not register graph trace events\n");
1468 return 1;
1471 return register_tracer(&graph_trace);
1474 device_initcall(init_graph_trace);