1 // SPDX-License-Identifier: GPL-2.0
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>
10 #include <linux/uaccess.h>
11 #include <linux/ftrace.h>
12 #include <linux/interrupt.h>
13 #include <linux/slab.h>
17 #include "trace_output.h"
19 /* When set, irq functions will be ignored */
20 static int ftrace_graph_skip_irqs;
22 struct fgraph_cpu_data {
27 unsigned long enter_funcs[FTRACE_RETFUNC_DEPTH];
31 struct fgraph_cpu_data __percpu *cpu_data;
33 /* Place to preserve last processed entry. */
34 struct ftrace_graph_ent_entry ent;
35 struct ftrace_graph_ret_entry ret;
40 #define TRACE_GRAPH_INDENT 2
42 unsigned int fgraph_max_depth;
44 static struct tracer_opt trace_opts[] = {
45 /* Display overruns? (for self-debug purpose) */
46 { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
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) },
67 /* Include sleep time (scheduled out) between entry and return */
68 { TRACER_OPT(sleep-time, TRACE_GRAPH_SLEEP_TIME) },
70 #ifdef CONFIG_FUNCTION_PROFILER
71 /* Include time within nested functions */
72 { TRACER_OPT(graph-time, TRACE_GRAPH_GRAPH_TIME) },
78 static 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,
86 static struct trace_array *graph_array;
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.
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,
100 print_graph_duration(struct trace_array *tr, unsigned long long duration,
101 struct trace_seq *s, u32 flags);
103 int __trace_graph_entry(struct trace_array *tr,
104 struct ftrace_graph_ent *trace,
105 unsigned int trace_ctx)
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;
112 event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
113 sizeof(*entry), trace_ctx);
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);
124 static inline int ftrace_graph_ignore_irqs(void)
126 if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
132 int trace_graph_entry(struct ftrace_graph_ent *trace)
134 struct trace_array *tr = graph_array;
135 struct trace_array_cpu *data;
137 unsigned int trace_ctx;
142 if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT))
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.
152 if (ftrace_graph_notrace_addr(trace->func)) {
153 trace_recursion_set(TRACE_GRAPH_NOTRACE_BIT);
155 * Need to return 1 to have the return called
156 * that will clear the NOTRACE bit.
161 if (!ftrace_trace_task(tr))
164 if (ftrace_graph_ignore_func(trace))
167 if (ftrace_graph_ignore_irqs())
171 * Stop here if tracing_threshold is set. We only write function return
172 * events to the ring buffer.
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);
188 atomic_dec(&data->disabled);
189 local_irq_restore(flags);
195 __trace_graph_function(struct trace_array *tr,
196 unsigned long ip, unsigned int trace_ctx)
198 u64 time = trace_clock_local();
199 struct ftrace_graph_ent ent = {
203 struct ftrace_graph_ret ret = {
210 __trace_graph_entry(tr, &ent, trace_ctx);
211 __trace_graph_return(tr, &ret, trace_ctx);
215 trace_graph_function(struct trace_array *tr,
216 unsigned long ip, unsigned long parent_ip,
217 unsigned int trace_ctx)
219 __trace_graph_function(tr, ip, trace_ctx);
222 void __trace_graph_return(struct trace_array *tr,
223 struct ftrace_graph_ret *trace,
224 unsigned int trace_ctx)
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;
231 event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
232 sizeof(*entry), trace_ctx);
235 entry = ring_buffer_event_data(event);
237 if (!call_filter_check_discard(call, entry, buffer, event))
238 trace_buffer_unlock_commit_nostack(buffer, event);
241 void trace_graph_return(struct ftrace_graph_ret *trace)
243 struct trace_array *tr = graph_array;
244 struct trace_array_cpu *data;
246 unsigned int trace_ctx;
250 ftrace_graph_addr_finish(trace);
252 if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT)) {
253 trace_recursion_clear(TRACE_GRAPH_NOTRACE_BIT);
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);
265 atomic_dec(&data->disabled);
266 local_irq_restore(flags);
269 void set_graph_array(struct trace_array *tr)
273 /* Make graph_array visible before we start tracing */
278 static void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
280 ftrace_graph_addr_finish(trace);
282 if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT)) {
283 trace_recursion_clear(TRACE_GRAPH_NOTRACE_BIT);
287 if (tracing_thresh &&
288 (trace->rettime - trace->calltime < tracing_thresh))
291 trace_graph_return(trace);
294 static struct fgraph_ops funcgraph_thresh_ops = {
295 .entryfunc = &trace_graph_entry,
296 .retfunc = &trace_graph_thresh_return,
299 static struct fgraph_ops funcgraph_ops = {
300 .entryfunc = &trace_graph_entry,
301 .retfunc = &trace_graph_return,
304 static int graph_trace_init(struct trace_array *tr)
310 ret = register_ftrace_graph(&funcgraph_thresh_ops);
312 ret = register_ftrace_graph(&funcgraph_ops);
315 tracing_start_cmdline_record();
320 static void graph_trace_reset(struct trace_array *tr)
322 tracing_stop_cmdline_record();
324 unregister_ftrace_graph(&funcgraph_thresh_ops);
326 unregister_ftrace_graph(&funcgraph_ops);
329 static int graph_trace_update_thresh(struct trace_array *tr)
331 graph_trace_reset(tr);
332 return graph_trace_init(tr);
335 static int max_bytes_for_cpu;
337 static void print_graph_cpu(struct trace_seq *s, int cpu)
340 * Start with a space character - to make it stand out
341 * to the right a bit when trace output is pasted into
344 trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
347 #define TRACE_GRAPH_PROCINFO_LENGTH 14
349 static void print_graph_proc(struct trace_seq *s, pid_t pid)
351 char comm[TASK_COMM_LEN];
352 /* sign + log10(MAX_INT) + '\0' */
358 trace_find_cmdline(pid, comm);
360 sprintf(pid_str, "%d", pid);
362 /* 1 stands for the "-" character */
363 len = strlen(comm) + strlen(pid_str) + 1;
365 if (len < TRACE_GRAPH_PROCINFO_LENGTH)
366 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
368 /* First spaces to align center */
369 for (i = 0; i < spaces / 2; i++)
370 trace_seq_putc(s, ' ');
372 trace_seq_printf(s, "%s-%s", comm, pid_str);
374 /* Last spaces to align center */
375 for (i = 0; i < spaces - (spaces / 2); i++)
376 trace_seq_putc(s, ' ');
380 static void print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
382 trace_seq_putc(s, ' ');
383 trace_print_lat_fmt(s, entry);
384 trace_seq_puts(s, " | ");
387 /* If the pid changed since the last trace, output this event */
389 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
397 last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
399 if (*last_pid == pid)
402 prev_pid = *last_pid;
408 * Context-switch trace line:
410 ------------------------------------------
411 | 1) migration/0--1 => sshd-1755
412 ------------------------------------------
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");
423 static struct ftrace_graph_ret_entry *
424 get_return_for_leaf(struct trace_iterator *iter,
425 struct ftrace_graph_ent_entry *curr)
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;
433 * If the previous output failed to write to the seq buffer,
434 * then we just reuse the data from before.
436 if (data && data->failed) {
441 ring_iter = trace_buffer_iter(iter, iter->cpu);
443 /* First peek to compare current entry and the next one */
445 event = ring_buffer_iter_peek(ring_iter, NULL);
448 * We need to consume the current entry to see
451 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu,
453 event = ring_buffer_peek(iter->array_buffer->buffer, iter->cpu,
460 next = ring_buffer_event_data(event);
464 * Save current and next entries for later reference
465 * if the output fails.
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.
473 if (next->ent.type == TRACE_GRAPH_RET)
476 data->ret.ent.type = next->ent.type;
480 if (next->ent.type != TRACE_GRAPH_RET)
483 if (curr->ent.pid != next->ent.pid ||
484 curr->graph_ent.func != next->ret.func)
487 /* this is a leaf, now advance the iterator */
489 ring_buffer_iter_advance(ring_iter);
494 static void print_graph_abs_time(u64 t, struct trace_seq *s)
496 unsigned long usecs_rem;
498 usecs_rem = do_div(t, NSEC_PER_SEC);
501 trace_seq_printf(s, "%5lu.%06lu | ",
502 (unsigned long)t, usecs_rem);
506 print_graph_rel_time(struct trace_iterator *iter, struct trace_seq *s)
508 unsigned long long usecs;
510 usecs = iter->ts - iter->array_buffer->time_start;
511 do_div(usecs, NSEC_PER_USEC);
513 trace_seq_printf(s, "%9llu us | ", usecs);
517 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
518 enum trace_type type, int cpu, pid_t pid, u32 flags)
520 struct trace_array *tr = iter->tr;
521 struct trace_seq *s = &iter->seq;
522 struct trace_entry *ent = iter->ent;
524 if (addr < (unsigned long)__irqentry_text_start ||
525 addr >= (unsigned long)__irqentry_text_end)
528 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
530 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
531 print_graph_abs_time(iter->ts, s);
534 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
535 print_graph_rel_time(iter, s);
538 if (flags & TRACE_GRAPH_PRINT_CPU)
539 print_graph_cpu(s, cpu);
542 if (flags & TRACE_GRAPH_PRINT_PROC) {
543 print_graph_proc(s, pid);
544 trace_seq_puts(s, " | ");
548 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
549 print_graph_lat_fmt(s, ent);
553 print_graph_duration(tr, 0, s, flags | FLAGS_FILL_START);
555 if (type == TRACE_GRAPH_ENT)
556 trace_seq_puts(s, "==========>");
558 trace_seq_puts(s, "<==========");
560 print_graph_duration(tr, 0, s, flags | FLAGS_FILL_END);
561 trace_seq_putc(s, '\n');
565 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
567 unsigned long nsecs_rem = do_div(duration, 1000);
568 /* log10(ULONG_MAX) + '\0' */
574 sprintf(usecs_str, "%lu", (unsigned long) duration);
577 trace_seq_printf(s, "%s", usecs_str);
579 len = strlen(usecs_str);
581 /* Print nsecs (we don't want to exceed 7 numbers) */
583 size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
585 snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
586 trace_seq_printf(s, ".%s", nsecs_str);
587 len += strlen(nsecs_str) + 1;
590 trace_seq_puts(s, " us ");
592 /* Print remaining spaces to fit the row's width */
593 for (i = len; i < 8; i++)
594 trace_seq_putc(s, ' ');
598 print_graph_duration(struct trace_array *tr, unsigned long long duration,
599 struct trace_seq *s, u32 flags)
601 if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
602 !(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
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, " | ");
610 case FLAGS_FILL_START:
611 trace_seq_puts(s, " ");
614 trace_seq_puts(s, " |");
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));
622 trace_seq_puts(s, " ");
624 trace_print_graph_duration(duration, s);
625 trace_seq_puts(s, "| ");
628 #ifdef CONFIG_FUNCTION_GRAPH_RETVAL
630 #define __TRACE_GRAPH_PRINT_RETVAL TRACE_GRAPH_PRINT_RETVAL
632 static void print_graph_retval(struct trace_seq *s, unsigned long retval,
633 bool leaf, void *func, bool hex_format)
635 unsigned long err_code = 0;
637 if (retval == 0 || hex_format)
640 /* Check if the return value matches the negative format */
641 if (IS_ENABLED(CONFIG_64BIT) && (retval & BIT(31)) &&
642 (((u64)retval) >> 32) == 0) {
644 err_code = (unsigned long)(s32)retval;
649 if (!IS_ERR_VALUE(err_code))
654 if (hex_format || (err_code == 0))
655 trace_seq_printf(s, "%ps(); /* = 0x%lx */\n",
658 trace_seq_printf(s, "%ps(); /* = %ld */\n",
661 if (hex_format || (err_code == 0))
662 trace_seq_printf(s, "} /* %ps = 0x%lx */\n",
665 trace_seq_printf(s, "} /* %ps = %ld */\n",
672 #define __TRACE_GRAPH_PRINT_RETVAL 0
674 #define print_graph_retval(_seq, _retval, _leaf, _func, _format) do {} while (0)
678 /* Case of a leaf function on its call entry */
679 static enum print_line_t
680 print_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)
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;
693 graph_ret = &ret_entry->ret;
694 call = &entry->graph_ent;
695 duration = graph_ret->rettime - graph_ret->calltime;
698 struct fgraph_cpu_data *cpu_data;
700 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
703 * Comments display at + 1 to depth. Since
704 * this is a leaf function, keep the comments
705 * equal to this depth.
707 cpu_data->depth = call->depth - 1;
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;
715 /* Overhead and duration */
716 print_graph_duration(tr, duration, s, flags);
719 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
720 trace_seq_putc(s, ' ');
723 * Write out the function return value if the option function-retval is
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));
730 trace_seq_printf(s, "%ps();\n", (void *)call->func);
732 print_graph_irq(iter, graph_ret->func, TRACE_GRAPH_RET,
733 cpu, iter->ent->pid, flags);
735 return trace_handle_return(s);
738 static enum print_line_t
739 print_graph_entry_nested(struct trace_iterator *iter,
740 struct ftrace_graph_ent_entry *entry,
741 struct trace_seq *s, int cpu, u32 flags)
743 struct ftrace_graph_ent *call = &entry->graph_ent;
744 struct fgraph_data *data = iter->private;
745 struct trace_array *tr = iter->tr;
749 struct fgraph_cpu_data *cpu_data;
752 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
753 cpu_data->depth = call->depth;
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;
762 print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
765 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
766 trace_seq_putc(s, ' ');
768 trace_seq_printf(s, "%ps() {\n", (void *)call->func);
770 if (trace_seq_has_overflowed(s))
771 return TRACE_TYPE_PARTIAL_LINE;
774 * we already consumed the current entry to check the next one
775 * and see if this is a leaf.
777 return TRACE_TYPE_NO_CONSUME;
781 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
782 int type, unsigned long addr, u32 flags)
784 struct fgraph_data *data = iter->private;
785 struct trace_entry *ent = iter->ent;
786 struct trace_array *tr = iter->tr;
790 verif_pid(s, ent->pid, cpu, data);
794 print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
796 if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
800 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
801 print_graph_abs_time(iter->ts, s);
804 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
805 print_graph_rel_time(iter, s);
808 if (flags & TRACE_GRAPH_PRINT_CPU)
809 print_graph_cpu(s, cpu);
812 if (flags & TRACE_GRAPH_PRINT_PROC) {
813 print_graph_proc(s, ent->pid);
814 trace_seq_puts(s, " | ");
818 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
819 print_graph_lat_fmt(s, ent);
825 * Entry check for irq code
828 * - we are inside irq code
829 * - we just entered irq code
832 * - funcgraph-interrupts option is set
833 * - we are not inside irq code
836 check_irq_entry(struct trace_iterator *iter, u32 flags,
837 unsigned long addr, int depth)
841 struct fgraph_data *data = iter->private;
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.
848 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
852 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
855 * We are inside the irq code
860 if ((addr < (unsigned long)__irqentry_text_start) ||
861 (addr >= (unsigned long)__irqentry_text_end))
865 * We are entering irq code.
872 * Return check for irq code
875 * - we are inside irq code
876 * - we just left irq code
879 * - funcgraph-interrupts option is set
880 * - we are not inside irq code
883 check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
887 struct fgraph_data *data = iter->private;
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.
894 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
898 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
901 * We are not inside the irq code.
903 if (*depth_irq == -1)
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.
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.
915 if (*depth_irq >= depth) {
921 * We are inside the irq code, and this is not the entry.
926 static enum print_line_t
927 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
928 struct trace_iterator *iter, u32 flags)
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;
936 if (check_irq_entry(iter, flags, call->func, call->depth))
937 return TRACE_TYPE_HANDLED;
939 print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags);
941 leaf_ret = get_return_for_leaf(iter, field);
943 ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
945 ret = print_graph_entry_nested(iter, field, s, cpu, flags);
949 * If we failed to write our output, then we need to make
950 * note of it. Because we already consumed our entry.
962 static enum print_line_t
963 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
964 struct trace_entry *ent, struct trace_iterator *iter,
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;
975 if (check_irq_return(iter, flags, trace->depth))
976 return TRACE_TYPE_HANDLED;
979 struct fgraph_cpu_data *cpu_data;
982 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
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.
989 cpu_data->depth = trace->depth - 1;
991 if (trace->depth < FTRACE_RETFUNC_DEPTH &&
992 !WARN_ON_ONCE(trace->depth < 0)) {
993 if (cpu_data->enter_funcs[trace->depth] != trace->func)
995 cpu_data->enter_funcs[trace->depth] = 0;
999 print_graph_prologue(iter, s, 0, 0, flags);
1001 /* Overhead and duration */
1002 print_graph_duration(tr, duration, s, flags);
1005 for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++)
1006 trace_seq_putc(s, ' ');
1009 * Always write out the function name and its return value if the
1010 * function-retval option is enabled.
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));
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.
1023 if (func_match && !(flags & TRACE_GRAPH_PRINT_TAIL))
1024 trace_seq_puts(s, "}\n");
1026 trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
1030 if (flags & TRACE_GRAPH_PRINT_OVERRUN)
1031 trace_seq_printf(s, " (Overruns: %u)\n",
1034 print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
1037 return trace_handle_return(s);
1040 static enum print_line_t
1041 print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1042 struct trace_iterator *iter, u32 flags)
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;
1053 depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1055 print_graph_prologue(iter, s, 0, 0, flags);
1058 print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
1062 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++)
1063 trace_seq_putc(s, ' ');
1066 trace_seq_puts(s, "/* ");
1068 switch (iter->ent->type) {
1070 ret = trace_print_bputs_msg_only(iter);
1071 if (ret != TRACE_TYPE_HANDLED)
1075 ret = trace_print_bprintk_msg_only(iter);
1076 if (ret != TRACE_TYPE_HANDLED)
1080 ret = trace_print_printk_msg_only(iter);
1081 if (ret != TRACE_TYPE_HANDLED)
1085 event = ftrace_find_event(ent->type);
1087 return TRACE_TYPE_UNHANDLED;
1089 ret = event->funcs->trace(iter, sym_flags, event);
1090 if (ret != TRACE_TYPE_HANDLED)
1094 if (trace_seq_has_overflowed(s))
1097 /* Strip ending newline */
1098 if (s->buffer[s->seq.len - 1] == '\n') {
1099 s->buffer[s->seq.len - 1] = '\0';
1103 trace_seq_puts(s, " */\n");
1105 return trace_handle_return(s);
1110 print_graph_function_flags(struct trace_iterator *iter, u32 flags)
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;
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;
1125 * If the last output failed, there's a possibility we need
1126 * to print out the missing entry which would never go out.
1128 if (data && data->failed) {
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;
1140 switch (entry->type) {
1141 case TRACE_GRAPH_ENT: {
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.
1148 struct ftrace_graph_ent_entry saved;
1149 trace_assign_type(field, entry);
1151 return print_graph_entry(&saved, s, iter, flags);
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);
1160 /* dont trace stack and functions as comments */
1161 return TRACE_TYPE_UNHANDLED;
1164 return print_graph_comment(s, entry, iter, flags);
1167 return TRACE_TYPE_HANDLED;
1170 static enum print_line_t
1171 print_graph_function(struct trace_iterator *iter)
1173 return print_graph_function_flags(iter, tracer_flags.val);
1176 static enum print_line_t
1177 print_graph_function_event(struct trace_iterator *iter, int flags,
1178 struct trace_event *event)
1180 return print_graph_function(iter);
1183 static void print_lat_header(struct seq_file *s, u32 flags)
1185 static const char spaces[] = " " /* 16 spaces */
1187 " "; /* 17 spaces */
1190 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1192 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1194 if (flags & TRACE_GRAPH_PRINT_CPU)
1196 if (flags & TRACE_GRAPH_PRINT_PROC)
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);
1206 static void __print_graph_headers_flags(struct trace_array *tr,
1207 struct seq_file *s, u32 flags)
1209 int lat = tr->trace_flags & TRACE_ITER_LATENCY_FMT;
1212 print_lat_header(s, flags);
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 ");
1225 seq_puts(s, "|||| ");
1226 if (flags & TRACE_GRAPH_PRINT_DURATION)
1227 seq_puts(s, " DURATION ");
1228 seq_puts(s, " FUNCTION CALLS\n");
1232 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1234 if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1236 if (flags & TRACE_GRAPH_PRINT_CPU)
1238 if (flags & TRACE_GRAPH_PRINT_PROC)
1239 seq_puts(s, " | | ");
1241 seq_puts(s, "|||| ");
1242 if (flags & TRACE_GRAPH_PRINT_DURATION)
1243 seq_puts(s, " | | ");
1244 seq_puts(s, " | | | |\n");
1247 static void print_graph_headers(struct seq_file *s)
1249 print_graph_headers_flags(s, tracer_flags.val);
1252 void print_graph_headers_flags(struct seq_file *s, u32 flags)
1254 struct trace_iterator *iter = s->private;
1255 struct trace_array *tr = iter->tr;
1257 if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
1260 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT) {
1261 /* print nothing if the buffers are empty */
1262 if (trace_empty(iter))
1265 print_trace_header(s, iter);
1268 __print_graph_headers_flags(tr, s, flags);
1271 void graph_trace_open(struct trace_iterator *iter)
1273 /* pid and depth on the last trace processed */
1274 struct fgraph_data *data;
1278 iter->private = NULL;
1280 /* We can be called in atomic context via ftrace_dump() */
1281 gfpflags = (in_atomic() || irqs_disabled()) ? GFP_ATOMIC : GFP_KERNEL;
1283 data = kzalloc(sizeof(*data), gfpflags);
1287 data->cpu_data = alloc_percpu_gfp(struct fgraph_cpu_data, gfpflags);
1288 if (!data->cpu_data)
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);
1303 iter->private = data;
1310 pr_warn("function graph tracer: not enough memory\n");
1313 void graph_trace_close(struct trace_iterator *iter)
1315 struct fgraph_data *data = iter->private;
1318 free_percpu(data->cpu_data);
1324 func_graph_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1326 if (bit == TRACE_GRAPH_PRINT_IRQS)
1327 ftrace_graph_skip_irqs = !set;
1329 if (bit == TRACE_GRAPH_SLEEP_TIME)
1330 ftrace_graph_sleep_time_control(set);
1332 if (bit == TRACE_GRAPH_GRAPH_TIME)
1333 ftrace_graph_graph_time_control(set);
1338 static struct trace_event_functions graph_functions = {
1339 .trace = print_graph_function_event,
1342 static struct trace_event graph_trace_entry_event = {
1343 .type = TRACE_GRAPH_ENT,
1344 .funcs = &graph_functions,
1347 static struct trace_event graph_trace_ret_event = {
1348 .type = TRACE_GRAPH_RET,
1349 .funcs = &graph_functions
1352 static 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,
1372 graph_depth_write(struct file *filp, const char __user *ubuf, size_t cnt,
1378 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1382 fgraph_max_depth = val;
1390 graph_depth_read(struct file *filp, char __user *ubuf, size_t cnt,
1393 char buf[15]; /* More than enough to hold UINT_MAX + "\n"*/
1396 n = sprintf(buf, "%d\n", fgraph_max_depth);
1398 return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
1401 static 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,
1408 static __init int init_graph_tracefs(void)
1412 ret = tracing_init_dentry();
1416 trace_create_file("max_graph_depth", TRACE_MODE_WRITE, NULL,
1417 NULL, &graph_depth_fops);
1421 fs_initcall(init_graph_tracefs);
1423 static __init int init_graph_trace(void)
1425 max_bytes_for_cpu = snprintf(NULL, 0, "%u", nr_cpu_ids - 1);
1427 if (!register_trace_event(&graph_trace_entry_event)) {
1428 pr_warn("Warning: could not register graph trace events\n");
1432 if (!register_trace_event(&graph_trace_ret_event)) {
1433 pr_warn("Warning: could not register graph trace events\n");
1437 return register_tracer(&graph_trace);
1440 core_initcall(init_graph_trace);