1 ========================
2 ftrace - Function Tracer
3 ========================
4
5 Copyright 2008 Red Hat Inc.
6
7 :Author: Steven Rostedt <srostedt@redhat.com>
8 :License: The GNU Free Documentation License,
9 (dual licensed under the GPL v2)
10 :Original Reviewers: Elias Oltmanns, Randy Du
11 John Kacur, and David Te
12
13 - Written for: 2.6.28-rc2
14 - Updated for: 3.10
15 - Updated for: 4.13 - Copyright 2017 VMware In
16 - Converted to rst format - Changbin Du <changb
17
18 Introduction
19 ------------
20
21 Ftrace is an internal tracer designed to help
22 designers of systems to find what is going on
23 It can be used for debugging or analyzing late
24 performance issues that take place outside of
25
26 Although ftrace is typically considered the fu
27 is really a framework of several assorted trac
28 There's latency tracing to examine what occurs
29 disabled and enabled, as well as for preemptio
30 a task is woken to the task is actually schedu
31
32 One of the most common uses of ftrace is the e
33 Throughout the kernel is hundreds of static ev
34 can be enabled via the tracefs file system to
35 going on in certain parts of the kernel.
36
37 See events.rst for more information.
38
39
40 Implementation Details
41 ----------------------
42
43 See Documentation/trace/ftrace-design.rst for
44
45
46 The File System
47 ---------------
48
49 Ftrace uses the tracefs file system to hold th
50 well as the files to display output.
51
52 When tracefs is configured into the kernel (wh
53 option will do) the directory /sys/kernel/trac
54 this directory, you can add to your /etc/fstab
55
56 tracefs /sys/kernel/tracing trace
57
58 Or you can mount it at run time with::
59
60 mount -t tracefs nodev /sys/kernel/tracing
61
62 For quicker access to that directory you may w
63 it::
64
65 ln -s /sys/kernel/tracing /tracing
66
67 .. attention::
68
69 Before 4.1, all ftrace tracing control files
70 file system, which is typically located at /
71 For backward compatibility, when mounting th
72 the tracefs file system will be automaticall
73
74 /sys/kernel/debug/tracing
75
76 All files located in the tracefs file system
77 debugfs file system directory as well.
78
79 .. attention::
80
81 Any selected ftrace option will also create
82 The rest of the document will assume that yo
83 (cd /sys/kernel/tracing) and will only conce
84 directory and not distract from the content
85 "/sys/kernel/tracing" path name.
86
87 That's it! (assuming that you have ftrace conf
88
89 After mounting tracefs you will have access to
90 of ftrace. Here is a list of some of the key f
91
92
93 Note: all time values are in microseconds.
94
95 current_tracer:
96
97 This is used to set or display the cur
98 that is configured. Changing the curre
99 the ring buffer content as well as the
100
101 available_tracers:
102
103 This holds the different types of trac
104 have been compiled into the kernel. Th
105 tracers listed here can be configured
106 echoing their name into current_tracer
107
108 tracing_on:
109
110 This sets or displays whether writing
111 ring buffer is enabled. Echo 0 into th
112 the tracer or 1 to enable it. Note, th
113 writing to the ring buffer, the tracin
114 still be occurring.
115
116 The kernel function tracing_off() can
117 kernel to disable writing to the ring
118 set this file to "0". User space can r
119 echoing "1" into the file.
120
121 Note, the function and event trigger "
122 set this file to zero and stop tracing
123 be re-enabled by user space using this
124
125 trace:
126
127 This file holds the output of the trac
128 readable format (described below). Ope
129 writing with the O_TRUNC flag clears t
130 Note, this file is not a consumer. If
131 (no tracer running, or tracing_on is z
132 the same output each time it is read.
133 it may produce inconsistent results as
134 the entire buffer without consuming it
135
136 trace_pipe:
137
138 The output is the same as the "trace"
139 file is meant to be streamed with live
140 Reads from this file will block until
141 retrieved. Unlike the "trace" file, t
142 consumer. This means reading from this
143 sequential reads to display more curre
144 data is read from this file, it is con
145 will not be read again with a sequenti
146 "trace" file is static, and if the tra
147 adding more data, it will display the
148 information every time it is read.
149
150 trace_options:
151
152 This file lets the user control the am
153 that is displayed in one of the above
154 files. Options also exist to modify ho
155 or events work (stack traces, timestam
156
157 options:
158
159 This is a directory that has a file fo
160 trace option (also in trace_options).
161 or cleared by writing a "1" or "0" res
162 corresponding file with the option nam
163
164 tracing_max_latency:
165
166 Some of the tracers record the max lat
167 For example, the maximum time that int
168 The maximum time is saved in this file
169 stored, and displayed by "trace". A ne
170 recorded if the latency is greater tha
171 (in microseconds).
172
173 By echoing in a time into this file, n
174 unless it is greater than the time in
175
176 tracing_thresh:
177
178 Some latency tracers will record a tra
179 latency is greater than the number in
180 Only active when the file contains a n
181 (in microseconds)
182
183 buffer_percent:
184
185 This is the watermark for how much the
186 before a waiter is woken up. That is,
187 blocking read syscall on one of the pe
188 will block until the given amount of d
189 is in the ring buffer before it wakes
190 controls how the splice system calls a
191
192 0 - means to wake up as soon as th
193 50 - means to wake up when roughly
194 are full.
195 100 - means to block until the ring
196 about to start overwriting the
197
198 buffer_size_kb:
199
200 This sets or displays the number of ki
201 buffer holds. By default, the trace bu
202 for each CPU. The displayed number is
203 CPU buffer and not total size of all b
204 trace buffers are allocated in pages (
205 that the kernel uses for allocation, u
206 A few extra pages may be allocated to
207 meta-data. If the last page allocated
208 than requested, the rest of the page w
209 making the actual allocation bigger th
210 ( Note, the size may not be a multiple
211 due to buffer management meta-data. )
212
213 Buffer sizes for individual CPUs may v
214 (see "per_cpu/cpu0/buffer_size_kb" bel
215 this file will show "X".
216
217 buffer_total_size_kb:
218
219 This displays the total combined size
220
221 buffer_subbuf_size_kb:
222
223 This sets or displays the sub buffer s
224 into several same size "sub buffers".
225 the size of the sub buffer. Normally,
226 architecture's page (4K on x86). The s
227 at the start which also limits the siz
228 the sub buffer is a page size, no even
229 size minus the sub buffer meta data.
230
231 Note, the buffer_subbuf_size_kb is a w
232 minimum size of the subbuffer. The ker
233 implementation details, or simply fail
234 not handle the request.
235
236 Changing the sub buffer size allows fo
237 page size.
238
239 Note: When changing the sub-buffer siz
240 data in the ring buffer and the snapsh
241
242 free_buffer:
243
244 If a process is performing tracing, an
245 shrunk "freed" when the process is fin
246 killed by a signal, this file can be u
247 of this file, the ring buffer will be
248 Having a process that is tracing also
249 exits its file descriptor for this fil
250 the ring buffer will be "freed".
251
252 It may also stop tracing if disable_on
253
254 tracing_cpumask:
255
256 This is a mask that lets the user only
257 The format is a hex string representin
258
259 set_ftrace_filter:
260
261 When dynamic ftrace is configured in (
262 section below "dynamic ftrace"), the c
263 modified (code text rewrite) to disabl
264 function profiler (mcount). This lets
265 in with practically no overhead in per
266 has a side effect of enabling or disab
267 to be traced. Echoing names of functio
268 will limit the trace to only those fun
269 This influences the tracers "function"
270 and thus also function profiling (see
271
272 The functions listed in "available_fil
273 can be written into this file.
274
275 This interface also allows for command
276 "Filter commands" section for more det
277
278 As a speed up, since processing string
279 and requires a check of all functions
280 an index can be written into this file
281 written will instead select the same c
282 of the "available_filter_functions" fi
283
284 set_ftrace_notrace:
285
286 This has an effect opposite to that of
287 set_ftrace_filter. Any function that i
288 be traced. If a function exists in bot
289 and set_ftrace_notrace, the function w
290
291 set_ftrace_pid:
292
293 Have the function tracer only trace th
294 listed in this file.
295
296 If the "function-fork" option is set,
297 PID is listed in this file forks, the
298 automatically be added to this file, a
299 traced by the function tracer as well.
300 cause PIDs of tasks that exit to be re
301
302 set_ftrace_notrace_pid:
303
304 Have the function tracer ignore thread
305 this file.
306
307 If the "function-fork" option is set,
308 PID is listed in this file forks, the
309 automatically be added to this file, a
310 traced by the function tracer as well.
311 cause PIDs of tasks that exit to be re
312
313 If a PID is in both this file and "set
314 file takes precedence, and the thread
315
316 set_event_pid:
317
318 Have the events only trace a task with
319 Note, sched_switch and sched_wake_up w
320 listed in this file.
321
322 To have the PIDs of children of tasks
323 added on fork, enable the "event-fork"
324 cause the PIDs of tasks to be removed
325 exits.
326
327 set_event_notrace_pid:
328
329 Have the events not trace a task with
330 Note, sched_switch and sched_wakeup wi
331 in this file, even if a thread's PID i
332 sched_switch or sched_wakeup events al
333 be traced.
334
335 To have the PIDs of children of tasks
336 added on fork, enable the "event-fork"
337 cause the PIDs of tasks to be removed
338 exits.
339
340 set_graph_function:
341
342 Functions listed in this file will cau
343 tracer to only trace these functions a
344 they call. (See the section "dynamic f
345 Note, set_ftrace_filter and set_ftrace
346 what functions are being traced.
347
348 set_graph_notrace:
349
350 Similar to set_graph_function, but wil
351 tracing when the function is hit until
352 This makes it possible to ignore traci
353 by a specific function.
354
355 available_filter_functions:
356
357 This lists the functions that ftrace h
358 These are the function names that you
359 "set_ftrace_filter", "set_ftrace_notra
360 "set_graph_function", or "set_graph_no
361 (See the section "dynamic ftrace" belo
362
363 available_filter_functions_addrs:
364
365 Similar to available_filter_functions,
366 for each function. The displayed addre
367 and can differ from /proc/kallsyms add
368
369 dyn_ftrace_total_info:
370
371 This file is for debugging purposes. T
372 have been converted to nops and are av
373
374 enabled_functions:
375
376 This file is more for debugging ftrace
377 in seeing if any function has a callba
378 Not only does the trace infrastructure
379 trace utility, but other subsystems mi
380 displays all functions that have a cal
381 as well as the number of callbacks tha
382 Note, a callback may also call multipl
383 not be listed in this count.
384
385 If the callback registered to be trace
386 the "save regs" attribute (thus even m
387 will be displayed on the same line as
388 is returning registers.
389
390 If the callback registered to be trace
391 the "ip modify" attribute (thus the re
392 an 'I' will be displayed on the same l
393 can be overridden.
394
395 If a non ftrace trampoline is attached
396 Note, normal ftrace trampolines can al
397 "direct" trampoline can be attached to
398
399 Some architectures can not call direct
400 the ftrace ops function located above
401 such cases an 'O' will be displayed.
402
403 If a function had either the "ip modif
404 it in the past, a 'M' will be shown. T
405 used to know if a function was every m
406 and can be used for debugging.
407
408 If the architecture supports it, it wi
409 is being directly called by the functi
410 than 1 it most likely will be ftrace_o
411
412 If the callback of a function jumps to
413 specific to the callback and which is
414 its address will be printed as well as
415 trampoline calls.
416
417 touched_functions:
418
419 This file contains all the functions t
420 to it via the ftrace infrastructure. I
421 enabled_functions but shows all functi
422 traced.
423
424 To see any function that has every bee
425 direct trampoline, one can perform the
426
427 grep ' M ' /sys/kernel/tracing/touched
428
429 function_profile_enabled:
430
431 When set it will enable all functions
432 tracer, or if configured, the function
433 keep a histogram of the number of func
434 and if the function graph tracer was c
435 track of the time spent in those funct
436 content can be displayed in the files:
437
438 trace_stat/function<cpu> ( function0,
439
440 trace_stat:
441
442 A directory that holds different traci
443
444 kprobe_events:
445
446 Enable dynamic trace points. See kprob
447
448 kprobe_profile:
449
450 Dynamic trace points stats. See kprobe
451
452 max_graph_depth:
453
454 Used with the function graph tracer. T
455 it will trace into a function. Setting
456 one will show only the first kernel fu
457 from user space.
458
459 printk_formats:
460
461 This is for tools that read the raw fo
462 the ring buffer references a string, o
463 is recorded into the buffer and not th
464 tools from knowing what that string wa
465 and address for the string allowing to
466 the strings were.
467
468 saved_cmdlines:
469
470 Only the pid of the task is recorded i
471 the event specifically saves the task
472 makes a cache of pid mappings to comms
473 comms for events. If a pid for a comm
474 "<...>" is displayed in the output.
475
476 If the option "record-cmd" is set to "
477 will not be saved during recording. By
478
479 saved_cmdlines_size:
480
481 By default, 128 comms are saved (see "
482 increase or decrease the amount of com
483 the number of comms to cache into this
484
485 saved_tgids:
486
487 If the option "record-tgid" is set, on
488 the Task Group ID of a task is saved i
489 the thread to its TGID. By default, th
490 disabled.
491
492 snapshot:
493
494 This displays the "snapshot" buffer an
495 take a snapshot of the current running
496 See the "Snapshot" section below for m
497
498 stack_max_size:
499
500 When the stack tracer is activated, th
501 maximum stack size it has encountered.
502 See the "Stack Trace" section below.
503
504 stack_trace:
505
506 This displays the stack back trace of
507 that was encountered when the stack tr
508 See the "Stack Trace" section below.
509
510 stack_trace_filter:
511
512 This is similar to "set_ftrace_filter"
513 functions the stack tracer will check.
514
515 trace_clock:
516
517 Whenever an event is recorded into the
518 "timestamp" is added. This stamp comes
519 clock. By default, ftrace uses the "lo
520 clock is very fast and strictly per cp
521 systems it may not be monotonic with r
522 CPUs. In other words, the local clocks
523 with local clocks on other CPUs.
524
525 Usual clocks for tracing::
526
527 # cat trace_clock
528 [local] global counter x86-tsc
529
530 The clock with the square brackets aro
531
532 local:
533 Default clock, but may not be
534
535 global:
536 This clock is in sync with all
537 be a bit slower than the local
538
539 counter:
540 This is not a clock at all, bu
541 counter. It counts up one by o
542 with all CPUs. This is useful
543 know exactly the order events
544 each other on different CPUs.
545
546 uptime:
547 This uses the jiffies counter
548 is relative to the time since
549
550 perf:
551 This makes ftrace use the same
552 Eventually perf will be able t
553 and this will help out in inte
554
555 x86-tsc:
556 Architectures may define their
557 example, x86 uses its own TSC
558
559 ppc-tb:
560 This uses the powerpc timebase
561 This is in sync across CPUs an
562 to correlate events across hyp
563 tb_offset is known.
564
565 mono:
566 This uses the fast monotonic c
567 which is monotonic and is subj
568
569 mono_raw:
570 This is the raw monotonic cloc
571 which is monotonic but is not
572 and ticks at the same rate as
573
574 boot:
575 This is the boot clock (CLOCK_
576 fast monotonic clock, but also
577 suspend. Since the clock acces
578 tracing in the suspend path, s
579 if clock is accessed after the
580 the fast mono clock is updated
581 appears to happen slightly soo
582 Also on 32-bit systems, it's p
583 sees a partial update. These e
584 processing should be able to h
585 ktime_get_boot_fast_ns() funct
586
587 tai:
588 This is the tai clock (CLOCK_T
589 clock time. However, this cloc
590 discontinuities and backwards
591 seconds. Since the clock acces
592 side effects are possible. The
593 readouts in case the internal
594 by setting the system time or
595 These effects are rare and pos
596 handle them. See comments in t
597 function for more information.
598
599 To set a clock, simply echo the clock
600
601 # echo global > trace_clock
602
603 Setting a clock clears the ring buffer
604 "snapshot" buffer.
605
606 trace_marker:
607
608 This is a very useful file for synchro
609 with events happening in the kernel. W
610 this file will be written into the ftr
611
612 It is useful in applications to open t
613 of the application and just reference
614 for the file::
615
616 void trace_write(const char *f
617 {
618 va_list ap;
619 char buf[256];
620 int n;
621
622 if (trace_fd < 0)
623 return;
624
625 va_start(ap, fmt);
626 n = vsnprintf(buf, 256
627 va_end(ap);
628
629 write(trace_fd, buf, n
630 }
631
632 start::
633
634 trace_fd = open("trace_marker"
635
636 Note: Writing into the trace_marker fi
637 that are written into /sys/kerne
638 See "Event triggers" in Document
639 example in Documentation/trace/h
640
641 trace_marker_raw:
642
643 This is similar to trace_marker above,
644 to be written to it, where a tool can
645 from trace_pipe_raw.
646
647 uprobe_events:
648
649 Add dynamic tracepoints in programs.
650 See uprobetracer.rst
651
652 uprobe_profile:
653
654 Uprobe statistics. See uprobetrace.txt
655
656 instances:
657
658 This is a way to make multiple trace b
659 events can be recorded in different bu
660 See "Instances" section below.
661
662 events:
663
664 This is the trace event directory. It
665 (also known as static tracepoints) tha
666 into the kernel. It shows what event t
667 and how they are grouped by system. Th
668 files at various levels that can enabl
669 when a "1" is written to them.
670
671 See events.rst for more information.
672
673 set_event:
674
675 By echoing in the event into this file
676
677 See events.rst for more information.
678
679 available_events:
680
681 A list of events that can be enabled i
682
683 See events.rst for more information.
684
685 timestamp_mode:
686
687 Certain tracers may change the timesta
688 logging trace events into the event bu
689 different modes can coexist within a b
690 effect when an event is logged determi
691 is used for that event. The default t
692 'delta'.
693
694 Usual timestamp modes for tracing:
695
696 # cat timestamp_mode
697 [delta] absolute
698
699 The timestamp mode with the square b
700 one in effect.
701
702 delta: Default timestamp mode - time
703 a per-buffer timestamp.
704
705 absolute: The timestamp is a full ti
706 against some other value. As
707 space and is less efficient.
708
709 hwlat_detector:
710
711 Directory for the Hardware Latency Det
712 See "Hardware Latency Detector" sectio
713
714 per_cpu:
715
716 This is a directory that contains the
717
718 per_cpu/cpu0/buffer_size_kb:
719
720 The ftrace buffer is defined per_cpu.
721 buffer for each CPU to allow writes to
722 and free from cache bouncing. These bu
723 size buffers. This file is similar to
724 file, but it only displays or sets the
725 specific CPU. (here cpu0).
726
727 per_cpu/cpu0/trace:
728
729 This is similar to the "trace" file, b
730 the data specific for the CPU. If writ
731 the specific CPU buffer.
732
733 per_cpu/cpu0/trace_pipe
734
735 This is similar to the "trace_pipe" fi
736 read, but it will only display (and co
737 for the CPU.
738
739 per_cpu/cpu0/trace_pipe_raw
740
741 For tools that can parse the ftrace ri
742 the trace_pipe_raw file can be used to
743 from the ring buffer directly. With th
744 system call, the buffer data can be qu
745 a file or to the network where a serve
746 data.
747
748 Like trace_pipe, this is a consuming r
749 reads will always produce different da
750
751 per_cpu/cpu0/snapshot:
752
753 This is similar to the main "snapshot"
754 snapshot the current CPU (if supported
755 the content of the snapshot for a give
756 written to, only clears this CPU buffe
757
758 per_cpu/cpu0/snapshot_raw:
759
760 Similar to the trace_pipe_raw, but wil
761 from the snapshot buffer for the given
762
763 per_cpu/cpu0/stats:
764
765 This displays certain stats about the
766
767 entries:
768 The number of events that are
769
770 overrun:
771 The number of lost events due
772 the buffer was full.
773
774 commit overrun:
775 Should always be zero.
776 This gets set if so many event
777 event (ring buffer is re-entra
778 buffer and starts dropping eve
779
780 bytes:
781 Bytes actually read (not overw
782
783 oldest event ts:
784 The oldest timestamp in the bu
785
786 now ts:
787 The current timestamp
788
789 dropped events:
790 Events lost due to overwrite o
791
792 read events:
793 The number of events read.
794
795 The Tracers
796 -----------
797
798 Here is the list of current tracers that may b
799
800 "function"
801
802 Function call tracer to trace all kern
803
804 "function_graph"
805
806 Similar to the function tracer except
807 function tracer probes the functions o
808 whereas the function graph tracer trac
809 and exit of the functions. It then pro
810 to draw a graph of function calls simi
811 source.
812
813 "blk"
814
815 The block tracer. The tracer used by t
816 application.
817
818 "hwlat"
819
820 The Hardware Latency tracer is used to
821 produces any latency. See "Hardware La
822 below.
823
824 "irqsoff"
825
826 Traces the areas that disable interrup
827 the trace with the longest max latency
828 See tracing_max_latency. When a new ma
829 it replaces the old trace. It is best
830 trace with the latency-format option e
831 happens automatically when the tracer
832
833 "preemptoff"
834
835 Similar to irqsoff but traces and reco
836 time for which preemption is disabled.
837
838 "preemptirqsoff"
839
840 Similar to irqsoff and preemptoff, but
841 records the largest time for which irq
842 is disabled.
843
844 "wakeup"
845
846 Traces and records the max latency tha
847 the highest priority task to get sched
848 it has been woken up.
849 Traces all tasks as an average develop
850
851 "wakeup_rt"
852
853 Traces and records the max latency tha
854 RT tasks (as the current "wakeup" does
855 for those interested in wake up timing
856
857 "wakeup_dl"
858
859 Traces and records the max latency tha
860 a SCHED_DEADLINE task to be woken (as
861 "wakeup_rt" does).
862
863 "mmiotrace"
864
865 A special tracer that is used to trace
866 It will trace all the calls that a mod
867 hardware. Everything it writes and rea
868 as well.
869
870 "branch"
871
872 This tracer can be configured when tra
873 calls within the kernel. It will trace
874 unlikely branch is hit and if it was c
875 of being correct.
876
877 "nop"
878
879 This is the "trace nothing" tracer. To
880 tracers from tracing simply echo "nop"
881 current_tracer.
882
883 Error conditions
884 ----------------
885
886 For most ftrace commands, failure modes are
887 using standard return codes.
888
889 For other more involved commands, extended e
890 available via the tracing/error_log file. F
891 support it, reading the tracing/error_log fi
892 display more detailed information about what
893 information is available. The tracing/error
894 error log displaying a small number (current
895 for the last (8) failed commands.
896
897 The extended error information and usage tak
898 this example::
899
900 # echo xxx > /sys/kernel/tracing/events/sc
901 echo: write error: Invalid argument
902
903 # cat /sys/kernel/tracing/error_log
904 [ 5348.887237] location: error: Couldn't y
905 Command: xxx
906 ^
907 [ 7517.023364] location: error: Bad rrr: s
908 Command: ppp qqq
909 ^
910
911 To clear the error log, echo the empty strin
912
913 # echo > /sys/kernel/tracing/error_log
914
915 Examples of using the tracer
916 ----------------------------
917
918 Here are typical examples of using the tracers
919 them only with the tracefs interface (without
920 user-land utilities).
921
922 Output format:
923 --------------
924
925 Here is an example of the output format of the
926
927 # tracer: function
928 #
929 # entries-in-buffer/entries-written: 140080/
930 #
931 # _-----=> irqs
932 # / _----=> need
933 # | / _---=> hard
934 # || / _--=> pree
935 # ||| / delay
936 # TASK-PID CPU# |||| TIMESTA
937 # | | | |||| |
938 bash-1977 [000] .... 17284.9936
939 bash-1977 [000] .... 17284.9936
940 bash-1977 [000] .... 17284.9936
941 sshd-1974 [003] .... 17284.9936
942 bash-1977 [000] .... 17284.9936
943 bash-1977 [000] ...1 17284.9936
944 bash-1977 [000] ...1 17284.9936
945 bash-1977 [000] .... 17284.9936
946 bash-1977 [000] .... 17284.9936
947 sshd-1974 [003] .... 17284.9936
948 ....
949
950 A header is printed with the tracer name that
951 the trace. In this case the tracer is "functio
952 number of events in the buffer as well as the
953 that were written. The difference is the numbe
954 lost due to the buffer filling up (250280 - 14
955 lost).
956
957 The header explains the content of the events.
958 PID "1977", the CPU that it was running on "00
959 (explained below), the timestamp in <secs>.<us
960 function name that was traced "sys_close" and
961 called this function "system_call_fastpath". T
962 at which the function was entered.
963
964 Latency trace format
965 --------------------
966
967 When the latency-format option is enabled or w
968 tracers is set, the trace file gives somewhat
969 why a latency happened. Here is a typical trac
970
971 # tracer: irqsoff
972 #
973 # irqsoff latency trace v1.1.5 on 3.8.0-test
974 # ------------------------------------------
975 # latency: 259 us, #4/4, CPU#2 | (M:preempt
976 # -----------------
977 # | task: ps-6143 (uid:0 nice:0 policy:0
978 # -----------------
979 # => started at: __lock_task_sighand
980 # => ended at: _raw_spin_unlock_irqrestor
981 #
982 #
983 # _------=> CPU#
984 # / _-----=> irqs-off
985 # | / _----=> need-resched
986 # || / _---=> hardirq/softirq
987 # ||| / _--=> preempt-depth
988 # |||| / delay
989 # cmd pid ||||| time | caller
990 # \ / ||||| \ | /
991 ps-6143 2d... 0us!: trace_hardir
992 ps-6143 2d..1 259us+: trace_hardir
993 ps-6143 2d..1 263us+: time_hardirq
994 ps-6143 2d..1 306us : <stack trace
995 => trace_hardirqs_on_caller
996 => trace_hardirqs_on
997 => _raw_spin_unlock_irqrestore
998 => do_task_stat
999 => proc_tgid_stat
1000 => proc_single_show
1001 => seq_read
1002 => vfs_read
1003 => sys_read
1004 => system_call_fastpath
1005
1006
1007 This shows that the current tracer is "irqsof
1008 for which interrupts were disabled. It gives
1009 never changes) and the version of the kernel
1010 (3.8). Then it displays the max latency in mi
1011 of trace entries displayed and the total numb
1012 VP, KP, SP, and HP are always zero and are re
1013 #P is the number of online CPUs (#P:4).
1014
1015 The task is the process that was running when
1016 occurred. (ps pid: 6143).
1017
1018 The start and stop (the functions in which th
1019 disabled and enabled respectively) that cause
1020
1021 - __lock_task_sighand is where the interrup
1022 - _raw_spin_unlock_irqrestore is where they
1023
1024 The next lines after the header are the trace
1025 explains which is which.
1026
1027 cmd: The name of the process in the trace.
1028
1029 pid: The PID of that process.
1030
1031 CPU#: The CPU which the process was running
1032
1033 irqs-off: 'd' interrupts are disabled. '.'
1034 .. caution:: If the architecture does
1035 read the irq flags variable,
1036 be printed here.
1037
1038 need-resched:
1039 - 'N' both TIF_NEED_RESCHED and PREEM
1040 - 'n' only TIF_NEED_RESCHED is set,
1041 - 'p' only PREEMPT_NEED_RESCHED is se
1042 - '.' otherwise.
1043
1044 hardirq/softirq:
1045 - 'Z' - NMI occurred inside a hardirq
1046 - 'z' - NMI is running
1047 - 'H' - hard irq occurred inside a so
1048 - 'h' - hard irq is running
1049 - 's' - soft irq is running
1050 - '.' - normal context.
1051
1052 preempt-depth: The level of preempt_disable
1053
1054 The above is mostly meaningful for kernel dev
1055
1056 time:
1057 When the latency-format option is ena
1058 output includes a timestamp relative
1059 trace. This differs from the output w
1060 is disabled, which includes an absolu
1061
1062 delay:
1063 This is just to help catch your eye a
1064 needs to be fixed to be only relative
1065 The marks are determined by the diffe
1066 current trace and the next trace.
1067
1068 - '$' - greater than 1 second
1069 - '@' - greater than 100 millisecon
1070 - '*' - greater than 10 millisecond
1071 - '#' - greater than 1000 microseco
1072 - '!' - greater than 100 microsecon
1073 - '+' - greater than 10 microsecond
1074 - ' ' - less than or equal to 10 mi
1075
1076 The rest is the same as the 'trace' file.
1077
1078 Note, the latency tracers will usually end
1079 to easily find where the latency occurred.
1080
1081 trace_options
1082 -------------
1083
1084 The trace_options file (or the options direct
1085 what gets printed in the trace output, or man
1086 To see what is available, simply cat the file
1087
1088 cat trace_options
1089 print-parent
1090 nosym-offset
1091 nosym-addr
1092 noverbose
1093 noraw
1094 nohex
1095 nobin
1096 noblock
1097 nofields
1098 trace_printk
1099 annotate
1100 nouserstacktrace
1101 nosym-userobj
1102 noprintk-msg-only
1103 context-info
1104 nolatency-format
1105 record-cmd
1106 norecord-tgid
1107 overwrite
1108 nodisable_on_free
1109 irq-info
1110 markers
1111 noevent-fork
1112 function-trace
1113 nofunction-fork
1114 nodisplay-graph
1115 nostacktrace
1116 nobranch
1117
1118 To disable one of the options, echo in the op
1119 "no"::
1120
1121 echo noprint-parent > trace_options
1122
1123 To enable an option, leave off the "no"::
1124
1125 echo sym-offset > trace_options
1126
1127 Here are the available options:
1128
1129 print-parent
1130 On function traces, display the calli
1131 function as well as the function bein
1132 ::
1133
1134 print-parent:
1135 bash-4000 [01] 1477.606694: simp
1136
1137 noprint-parent:
1138 bash-4000 [01] 1477.606694: simp
1139
1140
1141 sym-offset
1142 Display not only the function name, b
1143 offset in the function. For example,
1144 seeing just "ktime_get", you will see
1145 "ktime_get+0xb/0x20".
1146 ::
1147
1148 sym-offset:
1149 bash-4000 [01] 1477.606694: simp
1150
1151 sym-addr
1152 This will also display the function a
1153 as the function name.
1154 ::
1155
1156 sym-addr:
1157 bash-4000 [01] 1477.606694: simp
1158
1159 verbose
1160 This deals with the trace file when t
1161 latency-format option is enabled.
1162 ::
1163
1164 bash 4000 1 0 00000000 00010a95
1165 (+0.000ms): simple_strtoul (kstrt
1166
1167 raw
1168 This will display raw numbers. This o
1169 use with user applications that can t
1170 numbers better than having it done in
1171
1172 hex
1173 Similar to raw, but the numbers will
1174
1175 bin
1176 This will print out the formats in ra
1177
1178 block
1179 When set, reading trace_pipe will not
1180
1181 fields
1182 Print the fields as described by thei
1183 option than using hex, bin or raw, as
1184 of the content of the event.
1185
1186 trace_printk
1187 Can disable trace_printk() from writi
1188
1189 trace_printk_dest
1190 Set to have trace_printk() and simila
1191 write into this instance. Note, only
1192 this set. By setting this flag, it cl
1193 of the instance that had it set previ
1194 level trace has this set, and will ge
1195 instance has it set then clears it.
1196
1197 This flag cannot be cleared by the to
1198 default instance. The only way the to
1199 cleared, is by it being set in anothe
1200
1201 annotate
1202 It is sometimes confusing when the CP
1203 and one CPU buffer had a lot of event
1204 a shorter time frame, were another CP
1205 a few events, which lets it have olde
1206 the trace is reported, it shows the o
1207 and it may look like only one CPU ran
1208 oldest events). When the annotate opt
1209 display when a new CPU buffer started
1210
1211 <idle>-0 [001]
1212 <idle>-0 [001]
1213 <idle>-0 [001]
1214 ##### CPU 2 buffer started ##
1215 <idle>-0 [002]
1216 <idle>-0 [001]
1217 <idle>-0 [001]
1218
1219 userstacktrace
1220 This option changes the trace. It rec
1221 stacktrace of the current user space
1222 each trace event.
1223
1224 sym-userobj
1225 when user stacktrace are enabled, loo
1226 object the address belongs to, and pr
1227 relative address. This is especially
1228 ASLR is on, otherwise you don't get a
1229 resolve the address to object/file/li
1230 the app is no longer running
1231
1232 The lookup is performed when you read
1233 trace,trace_pipe. Example::
1234
1235 a.out-1623 [000] 40874.465
1236 x494] <- /root/a.out[+0x4a8
1237
1238
1239 printk-msg-only
1240 When set, trace_printk()s will only s
1241 and not their parameters (if trace_bp
1242 trace_bputs() was used to save the tr
1243
1244 context-info
1245 Show only the event data. Hides the c
1246 timestamp, CPU, and other useful data
1247
1248 latency-format
1249 This option changes the trace output.
1250 the trace displays additional informa
1251 latency, as described in "Latency tra
1252
1253 pause-on-trace
1254 When set, opening the trace file for
1255 writing to the ring buffer (as if tra
1256 This simulates the original behavior
1257 When the file is closed, tracing will
1258
1259 hash-ptr
1260 When set, "%p" in the event printk fo
1261 hashed pointer value instead of real
1262 This will be useful if you want to fi
1263 value is corresponding to the real va
1264
1265 record-cmd
1266 When any event or tracer is enabled,
1267 in the sched_switch trace point to fi
1268 with mapped pids and comms. But this
1269 overhead, and if you only care about
1270 name of the task, disabling this opti
1271 impact of tracing. See "saved_cmdline
1272
1273 record-tgid
1274 When any event or tracer is enabled,
1275 in the sched_switch trace point to fi
1276 mapped Thread Group IDs (TGID) mappin
1277 "saved_tgids".
1278
1279 overwrite
1280 This controls what happens when the t
1281 full. If "1" (default), the oldest ev
1282 discarded and overwritten. If "0", th
1283 events are discarded.
1284 (see per_cpu/cpu0/stats for overrun a
1285
1286 disable_on_free
1287 When the free_buffer is closed, traci
1288 stop (tracing_on set to 0).
1289
1290 irq-info
1291 Shows the interrupt, preempt count, n
1292 When disabled, the trace looks like::
1293
1294 # tracer: function
1295 #
1296 # entries-in-buffer/entries-w
1297 #
1298 # TASK-PID CPU#
1299 # | | |
1300 <idle>-0 [002]
1301 <idle>-0 [002]
1302 <idle>-0 [002]
1303
1304
1305 markers
1306 When set, the trace_marker is writabl
1307 When disabled, the trace_marker will
1308 on write.
1309
1310 event-fork
1311 When set, tasks with PIDs listed in s
1312 the PIDs of their children added to s
1313 tasks fork. Also, when tasks with PID
1314 their PIDs will be removed from the f
1315
1316 This affects PIDs listed in set_event
1317
1318 function-trace
1319 The latency tracers will enable funct
1320 if this option is enabled (default it
1321 it is disabled, the latency tracers d
1322 functions. This keeps the overhead of
1323 when performing latency tests.
1324
1325 function-fork
1326 When set, tasks with PIDs listed in s
1327 have the PIDs of their children added
1328 when those tasks fork. Also, when tas
1329 set_ftrace_pid exit, their PIDs will
1330 file.
1331
1332 This affects PIDs in set_ftrace_notra
1333
1334 display-graph
1335 When set, the latency tracers (irqsof
1336 use function graph tracing instead of
1337
1338 stacktrace
1339 When set, a stack trace is recorded a
1340 is recorded.
1341
1342 branch
1343 Enable branch tracing with the tracer
1344 tracer along with the currently set t
1345 with the "nop" tracer is the same as
1346 "branch" tracer.
1347
1348 .. tip:: Some tracers have their own options.
1349 file when the tracer is active. They a
1350 options directory.
1351
1352
1353 Here are the per tracer options:
1354
1355 Options for function tracer:
1356
1357 func_stack_trace
1358 When set, a stack trace is recorded a
1359 function that is recorded. NOTE! Limi
1360 that are recorded before enabling thi
1361 "set_ftrace_filter" otherwise the sys
1362 will be critically degraded. Remember
1363 this option before clearing the funct
1364
1365 Options for function_graph tracer:
1366
1367 Since the function_graph tracer has a slight
1368 it has its own options to control what is di
1369
1370 funcgraph-overrun
1371 When set, the "overrun" of the graph
1372 displayed after each function traced.
1373 overrun, is when the stack depth of t
1374 is greater than what is reserved for
1375 Each task has a fixed array of functi
1376 trace in the call graph. If the depth
1377 calls exceeds that, the function is n
1378 The overrun is the number of function
1379 due to exceeding this array.
1380
1381 funcgraph-cpu
1382 When set, the CPU number of the CPU w
1383 occurred is displayed.
1384
1385 funcgraph-overhead
1386 When set, if the function takes longe
1387 A certain amount, then a delay marker
1388 displayed. See "delay" above, under t
1389 header description.
1390
1391 funcgraph-proc
1392 Unlike other tracers, the process' co
1393 is not displayed by default, but inst
1394 when a task is traced in and out duri
1395 switch. Enabling this options has the
1396 of each process displayed at every li
1397
1398 funcgraph-duration
1399 At the end of each function (the retu
1400 the duration of the amount of time in
1401 function is displayed in microseconds
1402
1403 funcgraph-abstime
1404 When set, the timestamp is displayed
1405
1406 funcgraph-irqs
1407 When disabled, functions that happen
1408 interrupt will not be traced.
1409
1410 funcgraph-tail
1411 When set, the return event will inclu
1412 that it represents. By default this i
1413 only a closing curly bracket "}" is d
1414 the return of a function.
1415
1416 funcgraph-retval
1417 When set, the return value of each tr
1418 will be printed after an equal sign "
1419 this is off.
1420
1421 funcgraph-retval-hex
1422 When set, the return value will alway
1423 in hexadecimal format. If the option
1424 the return value is an error code, it
1425 in signed decimal format; otherwise i
1426 printed in hexadecimal format. By def
1427 is off.
1428
1429 sleep-time
1430 When running function graph tracer, t
1431 the time a task schedules out in its
1432 When enabled, it will account time th
1433 scheduled out as part of the function
1434
1435 graph-time
1436 When running function profiler with f
1437 to include the time to call nested fu
1438 not set, the time reported for the fu
1439 include the time the function itself
1440 time for functions that it called.
1441
1442 Options for blk tracer:
1443
1444 blk_classic
1445 Shows a more minimalistic output.
1446
1447
1448 irqsoff
1449 -------
1450
1451 When interrupts are disabled, the CPU can not
1452 external event (besides NMIs and SMIs). This
1453 interrupt from triggering or the mouse interr
1454 the kernel know of a new mouse event. The res
1455 with the reaction time.
1456
1457 The irqsoff tracer tracks the time for which
1458 disabled. When a new maximum latency is hit,
1459 the trace leading up to that latency point so
1460 new maximum is reached, the old saved trace i
1461 new trace is saved.
1462
1463 To reset the maximum, echo 0 into tracing_max
1464 an example::
1465
1466 # echo 0 > options/function-trace
1467 # echo irqsoff > current_tracer
1468 # echo 1 > tracing_on
1469 # echo 0 > tracing_max_latency
1470 # ls -ltr
1471 [...]
1472 # echo 0 > tracing_on
1473 # cat trace
1474 # tracer: irqsoff
1475 #
1476 # irqsoff latency trace v1.1.5 on 3.8.0-tes
1477 # -----------------------------------------
1478 # latency: 16 us, #4/4, CPU#0 | (M:preempt
1479 # -----------------
1480 # | task: swapper/0-0 (uid:0 nice:0 poli
1481 # -----------------
1482 # => started at: run_timer_softirq
1483 # => ended at: run_timer_softirq
1484 #
1485 #
1486 # _------=> CPU#
1487 # / _-----=> irqs-off
1488 # | / _----=> need-resched
1489 # || / _---=> hardirq/softir
1490 # ||| / _--=> preempt-depth
1491 # |||| / delay
1492 # cmd pid ||||| time | caller
1493 # \ / ||||| \ | /
1494 <idle>-0 0d.s2 0us+: _raw_spin_l
1495 <idle>-0 0dNs3 17us : _raw_spin_u
1496 <idle>-0 0dNs3 17us+: trace_hardi
1497 <idle>-0 0dNs3 25us : <stack trac
1498 => _raw_spin_unlock_irq
1499 => run_timer_softirq
1500 => __do_softirq
1501 => call_softirq
1502 => do_softirq
1503 => irq_exit
1504 => smp_apic_timer_interrupt
1505 => apic_timer_interrupt
1506 => rcu_idle_exit
1507 => cpu_idle
1508 => rest_init
1509 => start_kernel
1510 => x86_64_start_reservations
1511 => x86_64_start_kernel
1512
1513 Here we see that we had a latency of 16 micro
1514 very good). The _raw_spin_lock_irq in run_tim
1515 interrupts. The difference between the 16 and
1516 timestamp 25us occurred because the clock was
1517 between the time of recording the max latency
1518 recording the function that had that latency.
1519
1520 Note the above example had function-trace not
1521 function-trace, we get a much larger output::
1522
1523 with echo 1 > options/function-trace
1524
1525 # tracer: irqsoff
1526 #
1527 # irqsoff latency trace v1.1.5 on 3.8.0-tes
1528 # -----------------------------------------
1529 # latency: 71 us, #168/168, CPU#3 | (M:pree
1530 # -----------------
1531 # | task: bash-2042 (uid:0 nice:0 policy
1532 # -----------------
1533 # => started at: ata_scsi_queuecmd
1534 # => ended at: ata_scsi_queuecmd
1535 #
1536 #
1537 # _------=> CPU#
1538 # / _-----=> irqs-off
1539 # | / _----=> need-resched
1540 # || / _---=> hardirq/softir
1541 # ||| / _--=> preempt-depth
1542 # |||| / delay
1543 # cmd pid ||||| time | caller
1544 # \ / ||||| \ | /
1545 bash-2042 3d... 0us : _raw_spin_l
1546 bash-2042 3d... 0us : add_preempt
1547 bash-2042 3d..1 1us : ata_scsi_fi
1548 bash-2042 3d..1 1us : __ata_scsi_
1549 bash-2042 3d..1 2us : ata_find_de
1550 bash-2042 3d..1 2us : ata_qc_new_
1551 bash-2042 3d..1 3us : ata_sg_init
1552 bash-2042 3d..1 4us : ata_scsi_rw
1553 bash-2042 3d..1 4us : ata_build_r
1554 [...]
1555 bash-2042 3d..1 67us : delay_tsc <
1556 bash-2042 3d..1 67us : add_preempt
1557 bash-2042 3d..2 67us : sub_preempt
1558 bash-2042 3d..1 67us : add_preempt
1559 bash-2042 3d..2 68us : sub_preempt
1560 bash-2042 3d..1 68us+: ata_bmdma_s
1561 bash-2042 3d..1 71us : _raw_spin_u
1562 bash-2042 3d..1 71us : _raw_spin_u
1563 bash-2042 3d..1 72us+: trace_hardi
1564 bash-2042 3d..1 120us : <stack trac
1565 => _raw_spin_unlock_irqrestore
1566 => ata_scsi_queuecmd
1567 => scsi_dispatch_cmd
1568 => scsi_request_fn
1569 => __blk_run_queue_uncond
1570 => __blk_run_queue
1571 => blk_queue_bio
1572 => submit_bio_noacct
1573 => submit_bio
1574 => submit_bh
1575 => __ext3_get_inode_loc
1576 => ext3_iget
1577 => ext3_lookup
1578 => lookup_real
1579 => __lookup_hash
1580 => walk_component
1581 => lookup_last
1582 => path_lookupat
1583 => filename_lookup
1584 => user_path_at_empty
1585 => user_path_at
1586 => vfs_fstatat
1587 => vfs_stat
1588 => sys_newstat
1589 => system_call_fastpath
1590
1591
1592 Here we traced a 71 microsecond latency. But
1593 functions that were called during that time.
1594 enabling function tracing, we incur an added
1595 overhead may extend the latency times. But ne
1596 trace has provided some very helpful debuggin
1597
1598 If we prefer function graph output instead of
1599 display-graph option::
1600
1601 with echo 1 > options/display-graph
1602
1603 # tracer: irqsoff
1604 #
1605 # irqsoff latency trace v1.1.5 on 4.20.0-rc
1606 # -----------------------------------------
1607 # latency: 3751 us, #274/274, CPU#0 | (M:de
1608 # -----------------
1609 # | task: bash-1507 (uid:0 nice:0 policy
1610 # -----------------
1611 # => started at: free_debug_processing
1612 # => ended at: return_to_handler
1613 #
1614 #
1615 # _--
1616 # / _-
1617 # | / _
1618 # || /
1619 # ||| /
1620 # REL TIME CPU TASK/PID ||||
1621 # | | | | ||||
1622 0 us | 0) bash-1507 | d...
1623 0 us | 0) bash-1507 | d..1
1624 1 us | 0) bash-1507 | d..2
1625 2 us | 0) bash-1507 | d..2
1626 2 us | 0) bash-1507 | d..2
1627 3 us | 0) bash-1507 | d..2
1628 3 us | 0) bash-1507 | d..2
1629 3 us | 0) bash-1507 | d..2
1630 4 us | 0) bash-1507 | d..2
1631 [...]
1632 3750 us | 0) bash-1507 | d..1
1633 3750 us | 0) bash-1507 | d..1
1634 3764 us | 0) bash-1507 | d..1
1635 bash-1507 0d..1 3792us : <stack trac
1636 => free_debug_processing
1637 => __slab_free
1638 => kmem_cache_free
1639 => vm_area_free
1640 => remove_vma
1641 => exit_mmap
1642 => mmput
1643 => begin_new_exec
1644 => load_elf_binary
1645 => search_binary_handler
1646 => __do_execve_file.isra.32
1647 => __x64_sys_execve
1648 => do_syscall_64
1649 => entry_SYSCALL_64_after_hwframe
1650
1651 preemptoff
1652 ----------
1653
1654 When preemption is disabled, we may be able t
1655 interrupts but the task cannot be preempted a
1656 priority task must wait for preemption to be
1657 before it can preempt a lower priority task.
1658
1659 The preemptoff tracer traces the places that
1660 Like the irqsoff tracer, it records the maxim
1661 which preemption was disabled. The control of
1662 is much like the irqsoff tracer.
1663 ::
1664
1665 # echo 0 > options/function-trace
1666 # echo preemptoff > current_tracer
1667 # echo 1 > tracing_on
1668 # echo 0 > tracing_max_latency
1669 # ls -ltr
1670 [...]
1671 # echo 0 > tracing_on
1672 # cat trace
1673 # tracer: preemptoff
1674 #
1675 # preemptoff latency trace v1.1.5 on 3.8.0-
1676 # -----------------------------------------
1677 # latency: 46 us, #4/4, CPU#1 | (M:preempt
1678 # -----------------
1679 # | task: sshd-1991 (uid:0 nice:0 policy
1680 # -----------------
1681 # => started at: do_IRQ
1682 # => ended at: do_IRQ
1683 #
1684 #
1685 # _------=> CPU#
1686 # / _-----=> irqs-off
1687 # | / _----=> need-resched
1688 # || / _---=> hardirq/softir
1689 # ||| / _--=> preempt-depth
1690 # |||| / delay
1691 # cmd pid ||||| time | caller
1692 # \ / ||||| \ | /
1693 sshd-1991 1d.h. 0us+: irq_enter <
1694 sshd-1991 1d..1 46us : irq_exit <-
1695 sshd-1991 1d..1 47us+: trace_preem
1696 sshd-1991 1d..1 52us : <stack trac
1697 => sub_preempt_count
1698 => irq_exit
1699 => do_IRQ
1700 => ret_from_intr
1701
1702
1703 This has some more changes. Preemption was di
1704 interrupt came in (notice the 'h'), and was e
1705 But we also see that interrupts have been dis
1706 the preempt off section and leaving it (the '
1707 interrupts were enabled in the mean time or s
1708 was over.
1709 ::
1710
1711 # tracer: preemptoff
1712 #
1713 # preemptoff latency trace v1.1.5 on 3.8.0-
1714 # -----------------------------------------
1715 # latency: 83 us, #241/241, CPU#1 | (M:pree
1716 # -----------------
1717 # | task: bash-1994 (uid:0 nice:0 policy
1718 # -----------------
1719 # => started at: wake_up_new_task
1720 # => ended at: task_rq_unlock
1721 #
1722 #
1723 # _------=> CPU#
1724 # / _-----=> irqs-off
1725 # | / _----=> need-resched
1726 # || / _---=> hardirq/softir
1727 # ||| / _--=> preempt-depth
1728 # |||| / delay
1729 # cmd pid ||||| time | caller
1730 # \ / ||||| \ | /
1731 bash-1994 1d..1 0us : _raw_spin_l
1732 bash-1994 1d..1 0us : select_task
1733 bash-1994 1d..1 1us : __rcu_read_
1734 bash-1994 1d..1 1us : source_load
1735 bash-1994 1d..1 1us : source_load
1736 [...]
1737 bash-1994 1d..1 12us : irq_enter <
1738 bash-1994 1d..1 12us : rcu_irq_ent
1739 bash-1994 1d..1 13us : add_preempt
1740 bash-1994 1d.h1 13us : exit_idle <
1741 bash-1994 1d.h1 13us : hrtimer_int
1742 bash-1994 1d.h1 13us : _raw_spin_l
1743 bash-1994 1d.h1 14us : add_preempt
1744 bash-1994 1d.h2 14us : ktime_get_u
1745 [...]
1746 bash-1994 1d.h1 35us : lapic_next_
1747 bash-1994 1d.h1 35us : irq_exit <-
1748 bash-1994 1d.h1 36us : sub_preempt
1749 bash-1994 1d..2 36us : do_softirq
1750 bash-1994 1d..2 36us : __do_softir
1751 bash-1994 1d..2 36us : __local_bh_
1752 bash-1994 1d.s2 37us : add_preempt
1753 bash-1994 1d.s3 38us : _raw_spin_u
1754 bash-1994 1d.s3 39us : sub_preempt
1755 bash-1994 1d.s2 39us : call_timer_
1756 [...]
1757 bash-1994 1dNs2 81us : cpu_needs_a
1758 bash-1994 1dNs2 82us : __local_bh_
1759 bash-1994 1dNs2 82us : sub_preempt
1760 bash-1994 1dN.2 82us : idle_cpu <-
1761 bash-1994 1dN.2 83us : rcu_irq_exi
1762 bash-1994 1dN.2 83us : sub_preempt
1763 bash-1994 1.N.1 84us : _raw_spin_u
1764 bash-1994 1.N.1 84us+: trace_preem
1765 bash-1994 1.N.1 104us : <stack trac
1766 => sub_preempt_count
1767 => _raw_spin_unlock_irqrestore
1768 => task_rq_unlock
1769 => wake_up_new_task
1770 => do_fork
1771 => sys_clone
1772 => stub_clone
1773
1774
1775 The above is an example of the preemptoff tra
1776 function-trace set. Here we see that interrup
1777 the entire time. The irq_enter code lets us k
1778 an interrupt 'h'. Before that, the functions
1779 show that it is not in an interrupt, but we c
1780 functions themselves that this is not the cas
1781
1782 preemptirqsoff
1783 --------------
1784
1785 Knowing the locations that have interrupts di
1786 preemption disabled for the longest times is
1787 sometimes we would like to know when either p
1788 interrupts are disabled.
1789
1790 Consider the following code::
1791
1792 local_irq_disable();
1793 call_function_with_irqs_off();
1794 preempt_disable();
1795 call_function_with_irqs_and_preemption_of
1796 local_irq_enable();
1797 call_function_with_preemption_off();
1798 preempt_enable();
1799
1800 The irqsoff tracer will record the total leng
1801 call_function_with_irqs_off() and
1802 call_function_with_irqs_and_preemption_off().
1803
1804 The preemptoff tracer will record the total l
1805 call_function_with_irqs_and_preemption_off()
1806 call_function_with_preemption_off().
1807
1808 But neither will trace the time that interrup
1809 preemption is disabled. This total time is th
1810 not schedule. To record this time, use the pr
1811 tracer.
1812
1813 Again, using this trace is much like the irqs
1814 tracers.
1815 ::
1816
1817 # echo 0 > options/function-trace
1818 # echo preemptirqsoff > current_tracer
1819 # echo 1 > tracing_on
1820 # echo 0 > tracing_max_latency
1821 # ls -ltr
1822 [...]
1823 # echo 0 > tracing_on
1824 # cat trace
1825 # tracer: preemptirqsoff
1826 #
1827 # preemptirqsoff latency trace v1.1.5 on 3.
1828 # -----------------------------------------
1829 # latency: 100 us, #4/4, CPU#3 | (M:preempt
1830 # -----------------
1831 # | task: ls-2230 (uid:0 nice:0 policy:0
1832 # -----------------
1833 # => started at: ata_scsi_queuecmd
1834 # => ended at: ata_scsi_queuecmd
1835 #
1836 #
1837 # _------=> CPU#
1838 # / _-----=> irqs-off
1839 # | / _----=> need-resched
1840 # || / _---=> hardirq/softir
1841 # ||| / _--=> preempt-depth
1842 # |||| / delay
1843 # cmd pid ||||| time | caller
1844 # \ / ||||| \ | /
1845 ls-2230 3d... 0us+: _raw_spin_l
1846 ls-2230 3...1 100us : _raw_spin_u
1847 ls-2230 3...1 101us+: trace_preem
1848 ls-2230 3...1 111us : <stack trac
1849 => sub_preempt_count
1850 => _raw_spin_unlock_irqrestore
1851 => ata_scsi_queuecmd
1852 => scsi_dispatch_cmd
1853 => scsi_request_fn
1854 => __blk_run_queue_uncond
1855 => __blk_run_queue
1856 => blk_queue_bio
1857 => submit_bio_noacct
1858 => submit_bio
1859 => submit_bh
1860 => ext3_bread
1861 => ext3_dir_bread
1862 => htree_dirblock_to_tree
1863 => ext3_htree_fill_tree
1864 => ext3_readdir
1865 => vfs_readdir
1866 => sys_getdents
1867 => system_call_fastpath
1868
1869
1870 The trace_hardirqs_off_thunk is called from a
1871 interrupts are disabled in the assembly code.
1872 function tracing, we do not know if interrupt
1873 within the preemption points. We do see that
1874 preemption enabled.
1875
1876 Here is a trace with function-trace set::
1877
1878 # tracer: preemptirqsoff
1879 #
1880 # preemptirqsoff latency trace v1.1.5 on 3.
1881 # -----------------------------------------
1882 # latency: 161 us, #339/339, CPU#3 | (M:pre
1883 # -----------------
1884 # | task: ls-2269 (uid:0 nice:0 policy:0
1885 # -----------------
1886 # => started at: schedule
1887 # => ended at: mutex_unlock
1888 #
1889 #
1890 # _------=> CPU#
1891 # / _-----=> irqs-off
1892 # | / _----=> need-resched
1893 # || / _---=> hardirq/softir
1894 # ||| / _--=> preempt-depth
1895 # |||| / delay
1896 # cmd pid ||||| time | caller
1897 # \ / ||||| \ | /
1898 kworker/-59 3...1 0us : __schedule
1899 kworker/-59 3d..1 0us : rcu_preempt
1900 kworker/-59 3d..1 1us : add_preempt
1901 kworker/-59 3d..2 1us : deactivate_
1902 kworker/-59 3d..2 1us : dequeue_tas
1903 kworker/-59 3d..2 2us : update_rq_c
1904 kworker/-59 3d..2 2us : dequeue_tas
1905 kworker/-59 3d..2 2us : update_curr
1906 kworker/-59 3d..2 2us : update_min_
1907 kworker/-59 3d..2 3us : cpuacct_cha
1908 kworker/-59 3d..2 3us : __rcu_read_
1909 kworker/-59 3d..2 3us : __rcu_read_
1910 kworker/-59 3d..2 3us : update_cfs_
1911 kworker/-59 3d..2 4us : clear_buddi
1912 kworker/-59 3d..2 4us : account_ent
1913 kworker/-59 3d..2 4us : update_min_
1914 kworker/-59 3d..2 4us : update_cfs_
1915 kworker/-59 3d..2 5us : hrtick_upda
1916 kworker/-59 3d..2 5us : wq_worker_s
1917 kworker/-59 3d..2 5us : kthread_dat
1918 kworker/-59 3d..2 5us : put_prev_ta
1919 kworker/-59 3d..2 6us : pick_next_t
1920 kworker/-59 3d..2 6us : clear_buddi
1921 kworker/-59 3d..2 6us : set_next_en
1922 kworker/-59 3d..2 6us : update_stat
1923 ls-2269 3d..2 7us : finish_task
1924 ls-2269 3d..2 7us : _raw_spin_u
1925 ls-2269 3d..2 8us : do_IRQ <-re
1926 ls-2269 3d..2 8us : irq_enter <
1927 ls-2269 3d..2 8us : rcu_irq_ent
1928 ls-2269 3d..2 9us : add_preempt
1929 ls-2269 3d.h2 9us : exit_idle <
1930 [...]
1931 ls-2269 3d.h3 20us : sub_preempt
1932 ls-2269 3d.h2 20us : irq_exit <-
1933 ls-2269 3d.h2 21us : sub_preempt
1934 ls-2269 3d..3 21us : do_softirq
1935 ls-2269 3d..3 21us : __do_softir
1936 ls-2269 3d..3 21us+: __local_bh_
1937 ls-2269 3d.s4 29us : sub_preempt
1938 ls-2269 3d.s5 29us : sub_preempt
1939 ls-2269 3d.s5 31us : do_IRQ <-re
1940 ls-2269 3d.s5 31us : irq_enter <
1941 ls-2269 3d.s5 31us : rcu_irq_ent
1942 [...]
1943 ls-2269 3d.s5 31us : rcu_irq_ent
1944 ls-2269 3d.s5 32us : add_preempt
1945 ls-2269 3d.H5 32us : exit_idle <
1946 ls-2269 3d.H5 32us : handle_irq
1947 ls-2269 3d.H5 32us : irq_to_desc
1948 ls-2269 3d.H5 33us : handle_fast
1949 [...]
1950 ls-2269 3d.s5 158us : _raw_spin_u
1951 ls-2269 3d.s3 158us : net_rps_act
1952 ls-2269 3d.s3 159us : __local_bh_
1953 ls-2269 3d.s3 159us : sub_preempt
1954 ls-2269 3d..3 159us : idle_cpu <-
1955 ls-2269 3d..3 159us : rcu_irq_exi
1956 ls-2269 3d..3 160us : sub_preempt
1957 ls-2269 3d... 161us : __mutex_unl
1958 ls-2269 3d... 162us+: trace_hardi
1959 ls-2269 3d... 186us : <stack trac
1960 => __mutex_unlock_slowpath
1961 => mutex_unlock
1962 => process_output
1963 => n_tty_write
1964 => tty_write
1965 => vfs_write
1966 => sys_write
1967 => system_call_fastpath
1968
1969 This is an interesting trace. It started with
1970 scheduling out and ls taking over. But as soo
1971 rq lock and enabled interrupts (but not preem
1972 triggered. When the interrupt finished, it st
1973 But while the softirq was running, another in
1974 When an interrupt is running inside a softirq
1975
1976
1977 wakeup
1978 ------
1979
1980 One common case that people are interested in
1981 time it takes for a task that is woken to act
1982 Now for non Real-Time tasks, this can be arbi
1983 it nonetheless can be interesting.
1984
1985 Without function tracing::
1986
1987 # echo 0 > options/function-trace
1988 # echo wakeup > current_tracer
1989 # echo 1 > tracing_on
1990 # echo 0 > tracing_max_latency
1991 # chrt -f 5 sleep 1
1992 # echo 0 > tracing_on
1993 # cat trace
1994 # tracer: wakeup
1995 #
1996 # wakeup latency trace v1.1.5 on 3.8.0-test
1997 # -----------------------------------------
1998 # latency: 15 us, #4/4, CPU#3 | (M:preempt
1999 # -----------------
2000 # | task: kworker/3:1H-312 (uid:0 nice:-
2001 # -----------------
2002 #
2003 # _------=> CPU#
2004 # / _-----=> irqs-off
2005 # | / _----=> need-resched
2006 # || / _---=> hardirq/softir
2007 # ||| / _--=> preempt-depth
2008 # |||| / delay
2009 # cmd pid ||||| time | caller
2010 # \ / ||||| \ | /
2011 <idle>-0 3dNs7 0us : 0:120:
2012 <idle>-0 3dNs7 1us+: ttwu_do_act
2013 <idle>-0 3d..3 15us : __schedule
2014 <idle>-0 3d..3 15us : 0:120:
2015
2016 The tracer only traces the highest priority t
2017 to avoid tracing the normal circumstances. He
2018 the kworker with a nice priority of -20 (not
2019 just 15 microseconds from the time it woke up
2020 ran.
2021
2022 Non Real-Time tasks are not that interesting.
2023 trace is to concentrate only on Real-Time tas
2024
2025 wakeup_rt
2026 ---------
2027
2028 In a Real-Time environment it is very importa
2029 wakeup time it takes for the highest priority
2030 up to the time that it executes. This is also
2031 latency". I stress the point that this is abo
2032 also important to know the scheduling latency
2033 but the average schedule latency is better fo
2034 Tools like LatencyTop are more appropriate fo
2035 measurements.
2036
2037 Real-Time environments are interested in the
2038 That is the longest latency it takes for some
2039 and not the average. We can have a very fast
2040 only have a large latency once in a while, bu
2041 work well with Real-Time tasks. The wakeup_r
2042 to record the worst case wakeups of RT tasks.
2043 not recorded because the tracer only records
2044 tracing non-RT tasks that are unpredictable w
2045 worst case latency of RT tasks (just run the
2046 tracer for a while to see that effect).
2047
2048 Since this tracer only deals with RT tasks, w
2049 slightly differently than we did with the pre
2050 Instead of performing an 'ls', we will run 's
2051 'chrt' which changes the priority of the task
2052 ::
2053
2054 # echo 0 > options/function-trace
2055 # echo wakeup_rt > current_tracer
2056 # echo 1 > tracing_on
2057 # echo 0 > tracing_max_latency
2058 # chrt -f 5 sleep 1
2059 # echo 0 > tracing_on
2060 # cat trace
2061 # tracer: wakeup
2062 #
2063 # tracer: wakeup_rt
2064 #
2065 # wakeup_rt latency trace v1.1.5 on 3.8.0-t
2066 # -----------------------------------------
2067 # latency: 5 us, #4/4, CPU#3 | (M:preempt V
2068 # -----------------
2069 # | task: sleep-2389 (uid:0 nice:0 polic
2070 # -----------------
2071 #
2072 # _------=> CPU#
2073 # / _-----=> irqs-off
2074 # | / _----=> need-resched
2075 # || / _---=> hardirq/softir
2076 # ||| / _--=> preempt-depth
2077 # |||| / delay
2078 # cmd pid ||||| time | caller
2079 # \ / ||||| \ | /
2080 <idle>-0 3d.h4 0us : 0:120:
2081 <idle>-0 3d.h4 1us+: ttwu_do_act
2082 <idle>-0 3d..3 5us : __schedule
2083 <idle>-0 3d..3 5us : 0:120:
2084
2085
2086 Running this on an idle system, we see that i
2087 to perform the task switch. Note, since the
2088 is before the actual "switch", we stop the tr
2089 is about to schedule in. This may change if w
2090 end of the scheduler.
2091
2092 Notice that the recorded task is 'sleep' with
2093 and it has an rt_prio of 5. This priority is
2094 and not the internal kernel priority. The pol
2095 SCHED_FIFO and 2 for SCHED_RR.
2096
2097 Note, that the trace data shows the internal
2098 ::
2099
2100 <idle>-0 3d..3 5us : 0:120:R
2101
2102 The 0:120:R means idle was running with a nic
2103 and in the running state 'R'. The sleep task
2104 2389: 94:R. That is the priority is the kerne
2105 and it too is in the running state.
2106
2107 Doing the same with chrt -r 5 and function-tr
2108 ::
2109
2110 echo 1 > options/function-trace
2111
2112 # tracer: wakeup_rt
2113 #
2114 # wakeup_rt latency trace v1.1.5 on 3.8.0-t
2115 # -----------------------------------------
2116 # latency: 29 us, #85/85, CPU#3 | (M:preemp
2117 # -----------------
2118 # | task: sleep-2448 (uid:0 nice:0 polic
2119 # -----------------
2120 #
2121 # _------=> CPU#
2122 # / _-----=> irqs-off
2123 # | / _----=> need-resched
2124 # || / _---=> hardirq/softir
2125 # ||| / _--=> preempt-depth
2126 # |||| / delay
2127 # cmd pid ||||| time | caller
2128 # \ / ||||| \ | /
2129 <idle>-0 3d.h4 1us+: 0:120:
2130 <idle>-0 3d.h4 2us : ttwu_do_act
2131 <idle>-0 3d.h3 3us : check_preem
2132 <idle>-0 3d.h3 3us : resched_cur
2133 <idle>-0 3dNh3 4us : task_woken_
2134 <idle>-0 3dNh3 4us : _raw_spin_u
2135 <idle>-0 3dNh3 4us : sub_preempt
2136 <idle>-0 3dNh2 5us : ttwu_stat <
2137 <idle>-0 3dNh2 5us : _raw_spin_u
2138 <idle>-0 3dNh2 6us : sub_preempt
2139 <idle>-0 3dNh1 6us : _raw_spin_l
2140 <idle>-0 3dNh1 6us : add_preempt
2141 <idle>-0 3dNh2 7us : _raw_spin_u
2142 <idle>-0 3dNh2 7us : sub_preempt
2143 <idle>-0 3dNh1 7us : tick_progra
2144 <idle>-0 3dNh1 7us : clockevents
2145 <idle>-0 3dNh1 8us : ktime_get <
2146 <idle>-0 3dNh1 8us : lapic_next_
2147 <idle>-0 3dNh1 8us : irq_exit <-
2148 <idle>-0 3dNh1 9us : sub_preempt
2149 <idle>-0 3dN.2 9us : idle_cpu <-
2150 <idle>-0 3dN.2 9us : rcu_irq_exi
2151 <idle>-0 3dN.2 10us : rcu_eqs_ent
2152 <idle>-0 3dN.2 10us : sub_preempt
2153 <idle>-0 3.N.1 11us : rcu_idle_ex
2154 <idle>-0 3dN.1 11us : rcu_eqs_exi
2155 <idle>-0 3.N.1 11us : tick_nohz_i
2156 <idle>-0 3dN.1 12us : menu_hrtime
2157 <idle>-0 3dN.1 12us : ktime_get <
2158 <idle>-0 3dN.1 12us : tick_do_upd
2159 <idle>-0 3dN.1 13us : cpu_load_up
2160 <idle>-0 3dN.1 13us : _raw_spin_l
2161 <idle>-0 3dN.1 13us : add_preempt
2162 <idle>-0 3dN.2 13us : __cpu_load_
2163 <idle>-0 3dN.2 14us : sched_avg_u
2164 <idle>-0 3dN.2 14us : _raw_spin_u
2165 <idle>-0 3dN.2 14us : sub_preempt
2166 <idle>-0 3dN.1 15us : calc_load_n
2167 <idle>-0 3dN.1 15us : touch_softl
2168 <idle>-0 3dN.1 15us : hrtimer_can
2169 <idle>-0 3dN.1 15us : hrtimer_try
2170 <idle>-0 3dN.1 16us : lock_hrtime
2171 <idle>-0 3dN.1 16us : _raw_spin_l
2172 <idle>-0 3dN.1 16us : add_preempt
2173 <idle>-0 3dN.2 17us : __remove_hr
2174 <idle>-0 3dN.2 17us : hrtimer_for
2175 <idle>-0 3dN.2 17us : tick_progra
2176 <idle>-0 3dN.2 18us : clockevents
2177 <idle>-0 3dN.2 18us : ktime_get <
2178 <idle>-0 3dN.2 18us : lapic_next_
2179 <idle>-0 3dN.2 19us : _raw_spin_u
2180 <idle>-0 3dN.2 19us : sub_preempt
2181 <idle>-0 3dN.1 19us : hrtimer_for
2182 <idle>-0 3dN.1 20us : ktime_add_s
2183 <idle>-0 3dN.1 20us : ktime_add_s
2184 <idle>-0 3dN.1 20us : hrtimer_sta
2185 <idle>-0 3dN.1 20us : __hrtimer_s
2186 <idle>-0 3dN.1 21us : lock_hrtime
2187 <idle>-0 3dN.1 21us : _raw_spin_l
2188 <idle>-0 3dN.1 21us : add_preempt
2189 <idle>-0 3dN.2 22us : ktime_add_s
2190 <idle>-0 3dN.2 22us : enqueue_hrt
2191 <idle>-0 3dN.2 22us : tick_progra
2192 <idle>-0 3dN.2 23us : clockevents
2193 <idle>-0 3dN.2 23us : ktime_get <
2194 <idle>-0 3dN.2 23us : lapic_next_
2195 <idle>-0 3dN.2 24us : _raw_spin_u
2196 <idle>-0 3dN.2 24us : sub_preempt
2197 <idle>-0 3dN.1 24us : account_idl
2198 <idle>-0 3dN.1 24us : account_idl
2199 <idle>-0 3.N.1 25us : sub_preempt
2200 <idle>-0 3.N.. 25us : schedule <-
2201 <idle>-0 3.N.. 25us : __schedule
2202 <idle>-0 3.N.. 26us : add_preempt
2203 <idle>-0 3.N.1 26us : rcu_note_co
2204 <idle>-0 3.N.1 26us : rcu_sched_q
2205 <idle>-0 3dN.1 27us : rcu_preempt
2206 <idle>-0 3.N.1 27us : _raw_spin_l
2207 <idle>-0 3dN.1 27us : add_preempt
2208 <idle>-0 3dN.2 28us : put_prev_ta
2209 <idle>-0 3dN.2 28us : pick_next_t
2210 <idle>-0 3dN.2 28us : pick_next_t
2211 <idle>-0 3dN.2 29us : dequeue_pus
2212 <idle>-0 3d..3 29us : __schedule
2213 <idle>-0 3d..3 30us : 0:120:
2214
2215 This isn't that big of a trace, even with fun
2216 so I included the entire trace.
2217
2218 The interrupt went off while when the system
2219 before task_woken_rt() was called, the NEED_R
2220 this is indicated by the first occurrence of
2221
2222 Latency tracing and events
2223 --------------------------
2224 As function tracing can induce a much larger
2225 seeing what happens within the latency it is
2226 caused it. There is a middle ground, and that
2227 events.
2228 ::
2229
2230 # echo 0 > options/function-trace
2231 # echo wakeup_rt > current_tracer
2232 # echo 1 > events/enable
2233 # echo 1 > tracing_on
2234 # echo 0 > tracing_max_latency
2235 # chrt -f 5 sleep 1
2236 # echo 0 > tracing_on
2237 # cat trace
2238 # tracer: wakeup_rt
2239 #
2240 # wakeup_rt latency trace v1.1.5 on 3.8.0-t
2241 # -----------------------------------------
2242 # latency: 6 us, #12/12, CPU#2 | (M:preempt
2243 # -----------------
2244 # | task: sleep-5882 (uid:0 nice:0 polic
2245 # -----------------
2246 #
2247 # _------=> CPU#
2248 # / _-----=> irqs-off
2249 # | / _----=> need-resched
2250 # || / _---=> hardirq/softir
2251 # ||| / _--=> preempt-depth
2252 # |||| / delay
2253 # cmd pid ||||| time | caller
2254 # \ / ||||| \ | /
2255 <idle>-0 2d.h4 0us : 0:120:
2256 <idle>-0 2d.h4 0us : ttwu_do_act
2257 <idle>-0 2d.h4 1us : sched_wakeu
2258 <idle>-0 2dNh2 1us : hrtimer_exp
2259 <idle>-0 2.N.2 2us : power_end:
2260 <idle>-0 2.N.2 3us : cpu_idle: s
2261 <idle>-0 2dN.3 4us : hrtimer_can
2262 <idle>-0 2dN.3 4us : hrtimer_sta
2263 <idle>-0 2.N.2 5us : rcu_utiliza
2264 <idle>-0 2.N.2 5us : rcu_utiliza
2265 <idle>-0 2d..3 6us : __schedule
2266 <idle>-0 2d..3 6us : 0:120:
2267
2268
2269 Hardware Latency Detector
2270 -------------------------
2271
2272 The hardware latency detector is executed by
2273
2274 NOTE, this tracer will affect the performance
2275 periodically make a CPU constantly busy with
2276 ::
2277
2278 # echo hwlat > current_tracer
2279 # sleep 100
2280 # cat trace
2281 # tracer: hwlat
2282 #
2283 # entries-in-buffer/entries-written: 13/13
2284 #
2285 # _-----=> irq
2286 # / _----=> nee
2287 # | / _---=> har
2288 # || / _--=> pre
2289 # ||| / dela
2290 # TASK-PID CPU# |||| TIMEST
2291 # | | | |||| |
2292 <...>-1729 [001] d... 678.473
2293 <...>-1729 [004] d... 689.556
2294 <...>-1729 [005] d... 714.756
2295 <...>-1729 [001] d... 718.788
2296 <...>-1729 [002] d... 719.796
2297 <...>-1729 [006] d... 844.787
2298 <...>-1729 [003] d... 849.827
2299 <...>-1729 [007] d... 853.859
2300 <...>-1729 [001] d... 855.874
2301 <...>-1729 [001] d... 863.938
2302 <...>-1729 [007] d... 878.050
2303 <...>-1729 [007] d... 886.114
2304
2305
2306 The above output is somewhat the same in the
2307 interrupts disabled 'd'. Under the FUNCTION t
2308
2309 #1
2310 This is the count of events recorded
2311 tracing_threshold (See below).
2312
2313 inner/outer(us): 11/11
2314
2315 This shows two numbers as "inner latenc
2316 runs in a loop checking a timestamp twi
2317 the two timestamps is the "inner latenc
2318 after the previous timestamp and the ne
2319 the "outer latency".
2320
2321 ts:1581527483.343962693
2322
2323 The absolute timestamp that the first l
2324
2325 count:6
2326
2327 The number of times a latency was detec
2328
2329 nmi-total:7 nmi-count:1
2330
2331 On architectures that support it, if an
2332 test, the time spent in NMI is reported
2333 microseconds).
2334
2335 All architectures that have NMIs will s
2336 NMI comes in during the test.
2337
2338 hwlat files:
2339
2340 tracing_threshold
2341 This gets automatically set to "10" t
2342 microseconds. This is the threshold o
2343 needs to be detected before the trace
2344
2345 Note, when hwlat tracer is finished (
2346 written into "current_tracer"), the o
2347 tracing_threshold is placed back into
2348
2349 hwlat_detector/width
2350 The length of time the test runs with
2351
2352 hwlat_detector/window
2353 The length of time of the window whic
2354 runs. That is, the test will run for
2355 microseconds per "window" microsecond
2356
2357 tracing_cpumask
2358 When the test is started. A kernel th
2359 runs the test. This thread will alter
2360 listed in the tracing_cpumask between
2361 (one "window"). To limit the test to
2362 set the mask in this file to only the
2363 should run on.
2364
2365 function
2366 --------
2367
2368 This tracer is the function tracer. Enabling
2369 can be done from the debug file system. Make
2370 ftrace_enabled is set; otherwise this tracer
2371 See the "ftrace_enabled" section below.
2372 ::
2373
2374 # sysctl kernel.ftrace_enabled=1
2375 # echo function > current_tracer
2376 # echo 1 > tracing_on
2377 # usleep 1
2378 # echo 0 > tracing_on
2379 # cat trace
2380 # tracer: function
2381 #
2382 # entries-in-buffer/entries-written: 24799/
2383 #
2384 # _-----=> irq
2385 # / _----=> nee
2386 # | / _---=> har
2387 # || / _--=> pre
2388 # ||| / dela
2389 # TASK-PID CPU# |||| TIMEST
2390 # | | | |||| |
2391 bash-1994 [002] .... 3082.063
2392 bash-1994 [002] .... 3082.063
2393 bash-1994 [002] .... 3082.063
2394 bash-1994 [002] .... 3082.063
2395 bash-1994 [002] .... 3082.063
2396 bash-1994 [002] .... 3082.063
2397 bash-1994 [002] ...1 3082.063
2398 bash-1994 [002] .... 3082.063
2399 [...]
2400
2401
2402 Note: function tracer uses ring buffers to st
2403 entries. The newest data may overwrite the ol
2404 Sometimes using echo to stop the trace is not
2405 the tracing could have overwritten the data t
2406 record. For this reason, it is sometimes bett
2407 tracing directly from a program. This allows
2408 tracing at the point that you hit the part th
2409 interested in. To disable the tracing directl
2410 something like following code snippet can be
2411
2412 int trace_fd;
2413 [...]
2414 int main(int argc, char *argv[]) {
2415 [...]
2416 trace_fd = open(tracing_file(
2417 [...]
2418 if (condition_hit()) {
2419 write(trace_fd, "0",
2420 }
2421 [...]
2422 }
2423
2424
2425 Single thread tracing
2426 ---------------------
2427
2428 By writing into set_ftrace_pid you can trace
2429 single thread. For example::
2430
2431 # cat set_ftrace_pid
2432 no pid
2433 # echo 3111 > set_ftrace_pid
2434 # cat set_ftrace_pid
2435 3111
2436 # echo function > current_tracer
2437 # cat trace | head
2438 # tracer: function
2439 #
2440 # TASK-PID CPU# TIMESTAMP
2441 # | | | |
2442 yum-updatesd-3111 [003] 1637.254676:
2443 yum-updatesd-3111 [003] 1637.254681:
2444 yum-updatesd-3111 [003] 1637.254682:
2445 yum-updatesd-3111 [003] 1637.254683:
2446 yum-updatesd-3111 [003] 1637.254685:
2447 yum-updatesd-3111 [003] 1637.254686:
2448 # echo > set_ftrace_pid
2449 # cat trace |head
2450 # tracer: function
2451 #
2452 # TASK-PID CPU# TIMESTAMP
2453 # | | | |
2454 ##### CPU 3 buffer started ####
2455 yum-updatesd-3111 [003] 1701.957688:
2456 yum-updatesd-3111 [003] 1701.957689:
2457 yum-updatesd-3111 [003] 1701.957691:
2458 yum-updatesd-3111 [003] 1701.957692:
2459 yum-updatesd-3111 [003] 1701.957693:
2460
2461 If you want to trace a function when executin
2462 something like this simple program.
2463 ::
2464
2465 #include <stdio.h>
2466 #include <stdlib.h>
2467 #include <sys/types.h>
2468 #include <sys/stat.h>
2469 #include <fcntl.h>
2470 #include <unistd.h>
2471 #include <string.h>
2472
2473 #define _STR(x) #x
2474 #define STR(x) _STR(x)
2475 #define MAX_PATH 256
2476
2477 const char *find_tracefs(void)
2478 {
2479 static char tracefs[MAX_PATH+1
2480 static int tracefs_found;
2481 char type[100];
2482 FILE *fp;
2483
2484 if (tracefs_found)
2485 return tracefs;
2486
2487 if ((fp = fopen("/proc/mounts"
2488 perror("/proc/mounts")
2489 return NULL;
2490 }
2491
2492 while (fscanf(fp, "%*s %"
2493 STR(MAX_PATH)
2494 "s %99s %*s %*d
2495 tracefs, type) =
2496 if (strcmp(type, "trac
2497 break;
2498 }
2499 fclose(fp);
2500
2501 if (strcmp(type, "tracefs") !=
2502 fprintf(stderr, "trace
2503 return NULL;
2504 }
2505
2506 strcat(tracefs, "/tracing/");
2507 tracefs_found = 1;
2508
2509 return tracefs;
2510 }
2511
2512 const char *tracing_file(const char *
2513 {
2514 static char trace_file[MAX_PAT
2515 snprintf(trace_file, MAX_PATH,
2516 return trace_file;
2517 }
2518
2519 int main (int argc, char **argv)
2520 {
2521 if (argc < 1)
2522 exit(-1);
2523
2524 if (fork() > 0) {
2525 int fd, ffd;
2526 char line[64];
2527 int s;
2528
2529 ffd = open(tracing_fi
2530 if (ffd < 0)
2531 exit(-1);
2532 write(ffd, "nop", 3);
2533
2534 fd = open(tracing_fil
2535 s = sprintf(line, "%d
2536 write(fd, line, s);
2537
2538 write(ffd, "function"
2539
2540 close(fd);
2541 close(ffd);
2542
2543 execvp(argv[1], argv+
2544 }
2545
2546 return 0;
2547 }
2548
2549 Or this simple script!
2550 ::
2551
2552 #!/bin/bash
2553
2554 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs
2555 echo 0 > $tracefs/tracing_on
2556 echo $$ > $tracefs/set_ftrace_pid
2557 echo function > $tracefs/current_tracer
2558 echo 1 > $tracefs/tracing_on
2559 exec "$@"
2560
2561
2562 function graph tracer
2563 ---------------------------
2564
2565 This tracer is similar to the function tracer
2566 probes a function on its entry and its exit.
2567 using a dynamically allocated stack of return
2568 task_struct. On function entry the tracer ove
2569 address of each function traced to set a cust
2570 original return address is stored on the stac
2571 in the task_struct.
2572
2573 Probing on both ends of a function leads to s
2574 such as:
2575
2576 - measure of a function's time execution
2577 - having a reliable call stack to draw functi
2578
2579 This tracer is useful in several situations:
2580
2581 - you want to find the reason of a strange ke
2582 need to see what happens in detail on any a
2583 ones).
2584
2585 - you are experiencing weird latencies but it
2586 find its origin.
2587
2588 - you want to find quickly which path is take
2589 function
2590
2591 - you just want to peek inside a working kern
2592 what happens there.
2593
2594 ::
2595
2596 # tracer: function_graph
2597 #
2598 # CPU DURATION FUNCTION C
2599 # | | | | | |
2600
2601 0) | sys_open() {
2602 0) | do_sys_open() {
2603 0) | getname() {
2604 0) | kmem_cache_alloc
2605 0) 1.382 us | __might_sleep(
2606 0) 2.478 us | }
2607 0) | strncpy_from_use
2608 0) | might_fault()
2609 0) 1.389 us | __might_slee
2610 0) 2.553 us | }
2611 0) 3.807 us | }
2612 0) 7.876 us | }
2613 0) | alloc_fd() {
2614 0) 0.668 us | _spin_lock();
2615 0) 0.570 us | expand_files();
2616 0) 0.586 us | _spin_unlock();
2617
2618
2619 There are several columns that can be dynamic
2620 enabled/disabled. You can use every combinati
2621 want, depending on your needs.
2622
2623 - The cpu number on which the function execut
2624 enabled. It is sometimes better to only tr
2625 tracing_cpumask file) or you might sometime
2626 function calls while cpu tracing switch.
2627
2628 - hide: echo nofuncgraph-cpu > trace_
2629 - show: echo funcgraph-cpu > trace_op
2630
2631 - The duration (function's time of execution)
2632 the closing bracket line of a function or o
2633 than the current function in case of a leaf
2634 enabled.
2635
2636 - hide: echo nofuncgraph-duration > t
2637 - show: echo funcgraph-duration > tra
2638
2639 - The overhead field precedes the duration fi
2640 reached duration thresholds.
2641
2642 - hide: echo nofuncgraph-overhead > t
2643 - show: echo funcgraph-overhead > tra
2644 - depends on: funcgraph-duration
2645
2646 ie::
2647
2648 3) # 1837.709 us | } /* __switch
2649 3) | finish_task_s
2650 3) 0.313 us | _raw_spin_u
2651 3) 3.177 us | }
2652 3) # 1889.063 us | } /* __schedule
2653 3) ! 140.417 us | } /* __schedule *
2654 3) # 2034.948 us | } /* schedule */
2655 3) * 33998.59 us | } /* schedule_preempt
2656
2657 [...]
2658
2659 1) 0.260 us | msecs_to_
2660 1) 0.313 us | __rcu_rea
2661 1) + 61.770 us | }
2662 1) + 64.479 us | }
2663 1) 0.313 us | rcu_bh_qs();
2664 1) 0.313 us | __local_bh_en
2665 1) ! 217.240 us | }
2666 1) 0.365 us | idle_cpu();
2667 1) | rcu_irq_exit()
2668 1) 0.417 us | rcu_eqs_enter
2669 1) 3.125 us | }
2670 1) ! 227.812 us | }
2671 1) ! 457.395 us | }
2672 1) @ 119760.2 us | }
2673
2674 [...]
2675
2676 2) | handle_IPI() {
2677 1) 6.979 us | }
2678 2) 0.417 us | scheduler_ipi();
2679 1) 9.791 us | }
2680 1) + 12.917 us | }
2681 2) 3.490 us | }
2682 1) + 15.729 us | }
2683 1) + 18.542 us | }
2684 2) $ 3594274 us | }
2685
2686 Flags::
2687
2688 + means that the function exceeded 10 usecs
2689 ! means that the function exceeded 100 usec
2690 # means that the function exceeded 1000 use
2691 * means that the function exceeded 10 msecs
2692 @ means that the function exceeded 100 msec
2693 $ means that the function exceeded 1 sec.
2694
2695
2696 - The task/pid field displays the thread cmdl
2697 executed the function. It is default disabl
2698
2699 - hide: echo nofuncgraph-proc > trace
2700 - show: echo funcgraph-proc > trace_o
2701
2702 ie::
2703
2704 # tracer: function_graph
2705 #
2706 # CPU TASK/PID DURATION
2707 # | | | | |
2708 0) sh-4802 | |
2709 0) sh-4802 | |
2710 0) sh-4802 | |
2711 0) sh-4802 | 0.616 us |
2712 0) sh-4802 | 0.586 us |
2713 0) sh-4802 | 2.899 us |
2714 0) sh-4802 | 4.040 us |
2715 0) sh-4802 | 5.151 us |
2716 0) sh-4802 | + 49.370 us |
2717
2718
2719 - The absolute time field is an absolute time
2720 system clock since it started. A snapshot o
2721 given on each entry/exit of functions
2722
2723 - hide: echo nofuncgraph-abstime > tr
2724 - show: echo funcgraph-abstime > trac
2725
2726 ie::
2727
2728 #
2729 # TIME CPU DURATION
2730 # | | | |
2731 360.774522 | 1) 0.541 us |
2732 360.774522 | 1) 4.663 us |
2733 360.774523 | 1) 0.541 us |
2734 360.774524 | 1) 6.796 us |
2735 360.774524 | 1) 7.952 us |
2736 360.774525 | 1) 9.063 us |
2737 360.774525 | 1) 0.615 us |
2738 360.774527 | 1) 0.578 us |
2739 360.774528 | 1) |
2740 360.774528 | 1) |
2741 360.774529 | 1) |
2742 360.774529 | 1) |
2743 360.774530 | 1) 0.594 us |
2744
2745
2746 The function name is always displayed after t
2747 for a function if the start of that function
2748 trace buffer.
2749
2750 Display of the function name after the closin
2751 enabled for functions whose start is in the t
2752 allowing easier searching with grep for funct
2753 It is default disabled.
2754
2755 - hide: echo nofuncgraph-tail > trace
2756 - show: echo funcgraph-tail > trace_o
2757
2758 Example with nofuncgraph-tail (default)::
2759
2760 0) | putname() {
2761 0) | kmem_cache_free
2762 0) 0.518 us | __phys_addr()
2763 0) 1.757 us | }
2764 0) 2.861 us | }
2765
2766 Example with funcgraph-tail::
2767
2768 0) | putname() {
2769 0) | kmem_cache_free
2770 0) 0.518 us | __phys_addr()
2771 0) 1.757 us | } /* kmem_cache
2772 0) 2.861 us | } /* putname() */
2773
2774 The return value of each traced function can
2775 an equal sign "=". When encountering system c
2776 can be very helpful to quickly locate the fun
2777 returns an error code.
2778
2779 - hide: echo nofuncgraph-retval > tra
2780 - show: echo funcgraph-retval > trace
2781
2782 Example with funcgraph-retval::
2783
2784 1) | cgroup_migrate() {
2785 1) 0.651 us | cgroup_migrate_ad
2786 1) | cgroup_migrate_ex
2787 1) | cpu_cgroup_can_
2788 1) | cgroup_taskse
2789 1) 0.732 us | cgroup_task
2790 1) 1.232 us | } /* cgroup_t
2791 1) 0.380 us | sched_rt_can_
2792 1) 2.335 us | } /* cpu_cgroup
2793 1) 4.369 us | } /* cgroup_migra
2794 1) 7.143 us | } /* cgroup_migrate
2795
2796 The above example shows that the function cpu
2797 returned the error code -22 firstly, then we
2798 of this function to get the root cause.
2799
2800 When the option funcgraph-retval-hex is not s
2801 be displayed in a smart way. Specifically, if
2802 it will be printed in signed decimal format,
2803 printed in hexadecimal format.
2804
2805 - smart: echo nofuncgraph-retval-hex
2806 - hexadecimal: echo funcgraph-retval-
2807
2808 Example with funcgraph-retval-hex::
2809
2810 1) | cgroup_migrate()
2811 1) 0.651 us | cgroup_migrate_
2812 1) | cgroup_migrate_
2813 1) | cpu_cgroup_ca
2814 1) | cgroup_task
2815 1) 0.732 us | cgroup_ta
2816 1) 1.232 us | } /* cgroup
2817 1) 0.380 us | sched_rt_ca
2818 1) 2.335 us | } /* cpu_cgro
2819 1) 4.369 us | } /* cgroup_mig
2820 1) 7.143 us | } /* cgroup_migra
2821
2822 At present, there are some limitations when u
2823 option, and these limitations will be elimina
2824
2825 - Even if the function return type is void, a
2826 be printed, and you can just ignore it.
2827
2828 - Even if return values are stored in multipl
2829 value contained in the first register will
2830 To illustrate, in the x86 architecture, eax
2831 a 64-bit return value, with the lower 32 bi
2832 upper 32 bits saved in edx. However, only t
2833 will be recorded and printed.
2834
2835 - In certain procedure call standards, such a
2836 type is smaller than a GPR, it is the respo
2837 to perform the narrowing, and the upper bit
2838 Therefore, it is advisable to check the cod
2839 when using a u8 in a 64-bit GPR, bits [63:8
2840 especially when larger types are truncated,
2841 Here are some specific cases to illustrate
2842
2843 **Case One**:
2844
2845 The function narrow_to_u8 is defined as fol
2846
2847 u8 narrow_to_u8(u64 val)
2848 {
2849 // implicitly truncated
2850 return val;
2851 }
2852
2853 It may be compiled to::
2854
2855 narrow_to_u8:
2856 < ... ftrace instrumentation
2857 RET
2858
2859 If you pass 0x123456789abcdef to this funct
2860 it may be recorded as 0x123456789abcdef ins
2861
2862 **Case Two**:
2863
2864 The function error_if_not_4g_aligned is def
2865
2866 int error_if_not_4g_aligned(u64 val)
2867 {
2868 if (val & GENMASK(31, 0))
2869 return -EINVAL;
2870
2871 return 0;
2872 }
2873
2874 It could be compiled to::
2875
2876 error_if_not_4g_aligned:
2877 CBNZ w0, .Lnot_aligned
2878 RET // bi
2879 // [6
2880 .Lnot_aligned:
2881 MOV x0, #-EINVAL
2882 RET
2883
2884 When passing 0x2_0000_0000 to it, the retur
2885 0x2_0000_0000 instead of 0.
2886
2887 You can put some comments on specific functio
2888 trace_printk() For example, if you want to pu
2889 the __might_sleep() function, you just have t
2890 <linux/ftrace.h> and call trace_printk() insi
2891
2892 trace_printk("I'm a comment!\n")
2893
2894 will produce::
2895
2896 1) | __might_sle
2897 1) | /* I'm a
2898 1) 1.449 us | }
2899
2900
2901 You might find other useful features for this
2902 following "dynamic ftrace" section such as tr
2903 functions or tasks.
2904
2905 dynamic ftrace
2906 --------------
2907
2908 If CONFIG_DYNAMIC_FTRACE is set, the system w
2909 virtually no overhead when function tracing i
2910 this works is the mcount function call (place
2911 every kernel function, produced by the -pg sw
2912 starts of pointing to a simple return. (Enabl
2913 include the -pg switch in the compiling of th
2914
2915 At compile time every C file object is run th
2916 recordmcount program (located in the scripts
2917 program will parse the ELF headers in the C o
2918 the locations in the .text section that call
2919 with gcc version 4.6, the -mfentry has been a
2920 calls "__fentry__" instead of "mcount". Which
2921 the creation of the stack frame.
2922
2923 Note, not all sections are traced. They may b
2924 a notrace, or blocked another way and all inl
2925 traced. Check the "available_filter_functions
2926 can be traced.
2927
2928 A section called "__mcount_loc" is created th
2929 references to all the mcount/fentry call site
2930 The recordmcount program re-links this sectio
2931 original object. The final linking stage of t
2932 references into a single table.
2933
2934 On boot up, before SMP is initialized, the dy
2935 scans this table and updates all the location
2936 also records the locations, which are added t
2937 available_filter_functions list. Modules are
2938 are loaded and before they are executed. Whe
2939 unloaded, it also removes its functions from
2940 list. This is automatic in the module unload
2941 module author does not need to worry about it
2942
2943 When tracing is enabled, the process of modif
2944 tracepoints is dependent on architecture. The
2945 kstop_machine to prevent races with the CPUs
2946 modified (which can cause the CPU to do undes
2947 if the modified code crosses cache (or page)
2948 patched back to calls. But this time, they do
2949 (which is just a function stub). They now cal
2950 infrastructure.
2951
2952 The new method of modifying the function trac
2953 a breakpoint at the location to be modified,
2954 the rest of the instruction not covered by th
2955 all CPUs again, and then remove the breakpoin
2956 version to the ftrace call site.
2957
2958 Some archs do not even need to monkey around
2959 and can just slap the new code on top of the
2960 problems with other CPUs executing it at the
2961
2962 One special side-effect to the recording of t
2963 traced is that we can now selectively choose
2964 wish to trace and which ones we want the mcou
2965 as nops.
2966
2967 Two files are used, one for enabling and one
2968 tracing of specified functions. They are:
2969
2970 set_ftrace_filter
2971
2972 and
2973
2974 set_ftrace_notrace
2975
2976 A list of available functions that you can ad
2977 listed in:
2978
2979 available_filter_functions
2980
2981 ::
2982
2983 # cat available_filter_functions
2984 put_prev_task_idle
2985 kmem_cache_create
2986 pick_next_task_rt
2987 cpus_read_lock
2988 pick_next_task_fair
2989 mutex_lock
2990 [...]
2991
2992 If I am only interested in sys_nanosleep and
2993
2994 # echo sys_nanosleep hrtimer_interrupt > se
2995 # echo function > current_tracer
2996 # echo 1 > tracing_on
2997 # usleep 1
2998 # echo 0 > tracing_on
2999 # cat trace
3000 # tracer: function
3001 #
3002 # entries-in-buffer/entries-written: 5/5
3003 #
3004 # _-----=> irq
3005 # / _----=> nee
3006 # | / _---=> har
3007 # || / _--=> pre
3008 # ||| / dela
3009 # TASK-PID CPU# |||| TIMEST
3010 # | | | |||| |
3011 usleep-2665 [001] .... 4186.475
3012 <idle>-0 [001] d.h1 4186.475
3013 usleep-2665 [001] d.h1 4186.475
3014 <idle>-0 [003] d.h1 4186.475
3015 <idle>-0 [002] d.h1 4186.475
3016
3017 To see which functions are being traced, you
3018 ::
3019
3020 # cat set_ftrace_filter
3021 hrtimer_interrupt
3022 sys_nanosleep
3023
3024
3025 Perhaps this is not enough. The filters also
3026
3027 ``<match>*``
3028 will match functions that begin with
3029 ``*<match>``
3030 will match functions that end with <m
3031 ``*<match>*``
3032 will match functions that have <match
3033 ``<match1>*<match2>``
3034 will match functions that begin with
3035
3036 .. note::
3037 It is better to use quotes to enclose t
3038 otherwise the shell may expand the para
3039 of files in the local directory.
3040
3041 ::
3042
3043 # echo 'hrtimer_*' > set_ftrace_filter
3044
3045 Produces::
3046
3047 # tracer: function
3048 #
3049 # entries-in-buffer/entries-written: 897/89
3050 #
3051 # _-----=> irq
3052 # / _----=> nee
3053 # | / _---=> har
3054 # || / _--=> pre
3055 # ||| / dela
3056 # TASK-PID CPU# |||| TIMEST
3057 # | | | |||| |
3058 <idle>-0 [003] dN.1 4228.547
3059 <idle>-0 [003] dN.1 4228.547
3060 <idle>-0 [003] dN.2 4228.547
3061 <idle>-0 [003] dN.1 4228.547
3062 <idle>-0 [003] dN.1 4228.547
3063 <idle>-0 [003] d..1 4228.547
3064 <idle>-0 [003] d..1 4228.547
3065 <idle>-0 [003] d..2 4228.547
3066
3067 Notice that we lost the sys_nanosleep.
3068 ::
3069
3070 # cat set_ftrace_filter
3071 hrtimer_run_queues
3072 hrtimer_run_pending
3073 hrtimer_init
3074 hrtimer_cancel
3075 hrtimer_try_to_cancel
3076 hrtimer_forward
3077 hrtimer_start
3078 hrtimer_reprogram
3079 hrtimer_force_reprogram
3080 hrtimer_get_next_event
3081 hrtimer_interrupt
3082 hrtimer_nanosleep
3083 hrtimer_wakeup
3084 hrtimer_get_remaining
3085 hrtimer_get_res
3086 hrtimer_init_sleeper
3087
3088
3089 This is because the '>' and '>>' act just lik
3090 To rewrite the filters, use '>'
3091 To append to the filters, use '>>'
3092
3093 To clear out a filter so that all functions w
3094 again::
3095
3096 # echo > set_ftrace_filter
3097 # cat set_ftrace_filter
3098 #
3099
3100 Again, now we want to append.
3101
3102 ::
3103
3104 # echo sys_nanosleep > set_ftrace_filter
3105 # cat set_ftrace_filter
3106 sys_nanosleep
3107 # echo 'hrtimer_*' >> set_ftrace_filter
3108 # cat set_ftrace_filter
3109 hrtimer_run_queues
3110 hrtimer_run_pending
3111 hrtimer_init
3112 hrtimer_cancel
3113 hrtimer_try_to_cancel
3114 hrtimer_forward
3115 hrtimer_start
3116 hrtimer_reprogram
3117 hrtimer_force_reprogram
3118 hrtimer_get_next_event
3119 hrtimer_interrupt
3120 sys_nanosleep
3121 hrtimer_nanosleep
3122 hrtimer_wakeup
3123 hrtimer_get_remaining
3124 hrtimer_get_res
3125 hrtimer_init_sleeper
3126
3127
3128 The set_ftrace_notrace prevents those functio
3129 traced.
3130 ::
3131
3132 # echo '*preempt*' '*lock*' > set_ftrace_no
3133
3134 Produces::
3135
3136 # tracer: function
3137 #
3138 # entries-in-buffer/entries-written: 39608/
3139 #
3140 # _-----=> irq
3141 # / _----=> nee
3142 # | / _---=> har
3143 # || / _--=> pre
3144 # ||| / dela
3145 # TASK-PID CPU# |||| TIMEST
3146 # | | | |||| |
3147 bash-1994 [000] .... 4342.324
3148 bash-1994 [000] .... 4342.324
3149 bash-1994 [000] .... 4342.324
3150 bash-1994 [000] .... 4342.324
3151 bash-1994 [000] .... 4342.324
3152 bash-1994 [000] .... 4342.324
3153 bash-1994 [000] .... 4342.324
3154 bash-1994 [000] .... 4342.324
3155 bash-1994 [000] .... 4342.324
3156 bash-1994 [000] .... 4342.324
3157 bash-1994 [000] .... 4342.324
3158 bash-1994 [000] .... 4342.324
3159
3160 We can see that there's no more lock or preem
3161
3162 Selecting function filters via index
3163 ------------------------------------
3164
3165 Because processing of strings is expensive (t
3166 needs to be looked up before comparing to the
3167 an index can be used as well to enable functi
3168 case of setting thousands of specific functio
3169 in a list of numbers, no string processing wi
3170 at the specific location in the internal arra
3171 functions in the "available_filter_functions"
3172
3173 ::
3174
3175 # echo 1 > set_ftrace_filter
3176
3177 Will select the first function listed in "ava
3178
3179 ::
3180
3181 # head -1 available_filter_functions
3182 trace_initcall_finish_cb
3183
3184 # cat set_ftrace_filter
3185 trace_initcall_finish_cb
3186
3187 # head -50 available_filter_functions | tai
3188 x86_pmu_commit_txn
3189
3190 # echo 1 50 > set_ftrace_filter
3191 # cat set_ftrace_filter
3192 trace_initcall_finish_cb
3193 x86_pmu_commit_txn
3194
3195 Dynamic ftrace with the function graph tracer
3196 ---------------------------------------------
3197
3198 Although what has been explained above concer
3199 function tracer and the function-graph-tracer
3200 special features only available in the functi
3201
3202 If you want to trace only one function and al
3203 you just have to echo its name into set_graph
3204
3205 echo __do_fault > set_graph_function
3206
3207 will produce the following "expanded" trace o
3208 function::
3209
3210 0) | __do_fault() {
3211 0) | filemap_fault() {
3212 0) | find_lock_page() {
3213 0) 0.804 us | find_get_page();
3214 0) | __might_sleep()
3215 0) 1.329 us | }
3216 0) 3.904 us | }
3217 0) 4.979 us | }
3218 0) 0.653 us | _spin_lock();
3219 0) 0.578 us | page_add_file_rmap()
3220 0) 0.525 us | native_set_pte_at();
3221 0) 0.585 us | _spin_unlock();
3222 0) | unlock_page() {
3223 0) 0.541 us | page_waitqueue();
3224 0) 0.639 us | __wake_up_bit();
3225 0) 2.786 us | }
3226 0) + 14.237 us | }
3227 0) | __do_fault() {
3228 0) | filemap_fault() {
3229 0) | find_lock_page() {
3230 0) 0.698 us | find_get_page();
3231 0) | __might_sleep()
3232 0) 1.412 us | }
3233 0) 3.950 us | }
3234 0) 5.098 us | }
3235 0) 0.631 us | _spin_lock();
3236 0) 0.571 us | page_add_file_rmap()
3237 0) 0.526 us | native_set_pte_at();
3238 0) 0.586 us | _spin_unlock();
3239 0) | unlock_page() {
3240 0) 0.533 us | page_waitqueue();
3241 0) 0.638 us | __wake_up_bit();
3242 0) 2.793 us | }
3243 0) + 14.012 us | }
3244
3245 You can also expand several functions at once
3246
3247 echo sys_open > set_graph_function
3248 echo sys_close >> set_graph_function
3249
3250 Now if you want to go back to trace all funct
3251 this special filter via::
3252
3253 echo > set_graph_function
3254
3255
3256 ftrace_enabled
3257 --------------
3258
3259 Note, the proc sysctl ftrace_enable is a big
3260 function tracer. By default it is enabled (wh
3261 enabled in the kernel). If it is disabled, al
3262 disabled. This includes not only the function
3263 also for any other uses (perf, kprobes, stack
3264 cannot be disabled if there is a callback wit
3265 registered.
3266
3267 Please disable this with care.
3268
3269 This can be disable (and enabled) with::
3270
3271 sysctl kernel.ftrace_enabled=0
3272 sysctl kernel.ftrace_enabled=1
3273
3274 or
3275
3276 echo 0 > /proc/sys/kernel/ftrace_enabled
3277 echo 1 > /proc/sys/kernel/ftrace_enabled
3278
3279
3280 Filter commands
3281 ---------------
3282
3283 A few commands are supported by the set_ftrac
3284 Trace commands have the following format::
3285
3286 <function>:<command>:<parameter>
3287
3288 The following commands are supported:
3289
3290 - mod:
3291 This command enables function filtering per
3292 parameter defines the module. For example,
3293 functions in the ext3 module are desired, r
3294
3295 echo 'write*:mod:ext3' > set_ftrace_filter
3296
3297 This command interacts with the filter in t
3298 filtering based on function names. Thus, ad
3299 in a different module is accomplished by ap
3300 filter file. Remove specific module functio
3301 '!'::
3302
3303 echo '!writeback*:mod:ext3' >> set_ftrace_
3304
3305 Mod command supports module globbing. Disab
3306 functions except a specific module::
3307
3308 echo '!*:mod:!ext3' >> set_ftrace_filter
3309
3310 Disable tracing for all modules, but still
3311
3312 echo '!*:mod:*' >> set_ftrace_filter
3313
3314 Enable filter only for kernel::
3315
3316 echo '*write*:mod:!*' >> set_ftrace_filter
3317
3318 Enable filter for module globbing::
3319
3320 echo '*write*:mod:*snd*' >> set_ftrace_fil
3321
3322 - traceon/traceoff:
3323 These commands turn tracing on and off when
3324 functions are hit. The parameter determines
3325 tracing system is turned on and off. If uns
3326 no limit. For example, to disable tracing w
3327 is hit the first 5 times, run::
3328
3329 echo '__schedule_bug:traceoff:5' > set_ftr
3330
3331 To always disable tracing when __schedule_b
3332
3333 echo '__schedule_bug:traceoff' > set_ftrac
3334
3335 These commands are cumulative whether or no
3336 to set_ftrace_filter. To remove a command,
3337 and drop the parameter::
3338
3339 echo '!__schedule_bug:traceoff:0' > set_ft
3340
3341 The above removes the traceoff command for
3342 that have a counter. To remove commands wit
3343
3344 echo '!__schedule_bug:traceoff' > set_ftra
3345
3346 - snapshot:
3347 Will cause a snapshot to be triggered when
3348 ::
3349
3350 echo 'native_flush_tlb_others:snapshot' >
3351
3352 To only snapshot once:
3353 ::
3354
3355 echo 'native_flush_tlb_others:snapshot:1'
3356
3357 To remove the above commands::
3358
3359 echo '!native_flush_tlb_others:snapshot' >
3360 echo '!native_flush_tlb_others:snapshot:0'
3361
3362 - enable_event/disable_event:
3363 These commands can enable or disable a trac
3364 function tracing callbacks are very sensiti
3365 are registered, the trace point is activate
3366 a "soft" mode. That is, the tracepoint will
3367 just will not be traced. The event tracepoi
3368 as long as there's a command that triggers
3369 ::
3370
3371 echo 'try_to_wake_up:enable_event:sched:sc
3372 set_ftrace_filter
3373
3374 The format is::
3375
3376 <function>:enable_event:<system>:<event>[
3377 <function>:disable_event:<system>:<event>
3378
3379 To remove the events commands::
3380
3381 echo '!try_to_wake_up:enable_event:sched:s
3382 set_ftrace_filter
3383 echo '!schedule:disable_event:sched:sched_
3384 set_ftrace_filter
3385
3386 - dump:
3387 When the function is hit, it will dump the
3388 ring buffer to the console. This is useful
3389 something, and want to dump the trace when
3390 is hit. Perhaps it's a function that is cal
3391 fault happens and does not allow you to get
3392
3393 - cpudump:
3394 When the function is hit, it will dump the
3395 ring buffer for the current CPU to the cons
3396 command, it only prints out the contents of
3397 CPU that executed the function that trigger
3398
3399 - stacktrace:
3400 When the function is hit, a stack trace is
3401
3402 trace_pipe
3403 ----------
3404
3405 The trace_pipe outputs the same content as th
3406 the effect on the tracing is different. Every
3407 trace_pipe is consumed. This means that subse
3408 different. The trace is live.
3409 ::
3410
3411 # echo function > current_tracer
3412 # cat trace_pipe > /tmp/trace.out &
3413 [1] 4153
3414 # echo 1 > tracing_on
3415 # usleep 1
3416 # echo 0 > tracing_on
3417 # cat trace
3418 # tracer: function
3419 #
3420 # entries-in-buffer/entries-written: 0/0
3421 #
3422 # _-----=> irq
3423 # / _----=> nee
3424 # | / _---=> har
3425 # || / _--=> pre
3426 # ||| / dela
3427 # TASK-PID CPU# |||| TIMEST
3428 # | | | |||| |
3429
3430 #
3431 # cat /tmp/trace.out
3432 bash-1994 [000] .... 5281.5689
3433 bash-1994 [000] .... 5281.5689
3434 bash-1994 [000] .... 5281.5689
3435 bash-1994 [000] .... 5281.5689
3436 bash-1994 [000] .... 5281.5689
3437 bash-1994 [000] .... 5281.5689
3438 bash-1994 [000] ...1 5281.5689
3439 bash-1994 [000] .... 5281.5689
3440 bash-1994 [000] .... 5281.5689
3441
3442
3443 Note, reading the trace_pipe file will block
3444 added. This is contrary to the trace file. If
3445 the trace file for reading, it will actually
3446 prevent new entries from being added. The tra
3447 not have this limitation.
3448
3449 trace entries
3450 -------------
3451
3452 Having too much or not enough data can be tro
3453 diagnosing an issue in the kernel. The file b
3454 used to modify the size of the internal trace
3455 number listed is the number of entries that c
3456 CPU. To know the full size, multiply the numb
3457 with the number of entries.
3458 ::
3459
3460 # cat buffer_size_kb
3461 1408 (units kilobytes)
3462
3463 Or simply read buffer_total_size_kb
3464 ::
3465
3466 # cat buffer_total_size_kb
3467 5632
3468
3469 To modify the buffer, simple echo in a number
3470 ::
3471
3472 # echo 10000 > buffer_size_kb
3473 # cat buffer_size_kb
3474 10000 (units kilobytes)
3475
3476 It will try to allocate as much as possible.
3477 much, it can cause Out-Of-Memory to trigger.
3478 ::
3479
3480 # echo 1000000000000 > buffer_size_kb
3481 -bash: echo: write error: Cannot allocate m
3482 # cat buffer_size_kb
3483 85
3484
3485 The per_cpu buffers can be changed individual
3486 ::
3487
3488 # echo 10000 > per_cpu/cpu0/buffer_size_kb
3489 # echo 100 > per_cpu/cpu1/buffer_size_kb
3490
3491 When the per_cpu buffers are not the same, th
3492 at the top level will just show an X
3493 ::
3494
3495 # cat buffer_size_kb
3496 X
3497
3498 This is where the buffer_total_size_kb is use
3499 ::
3500
3501 # cat buffer_total_size_kb
3502 12916
3503
3504 Writing to the top level buffer_size_kb will
3505 to be the same again.
3506
3507 Snapshot
3508 --------
3509 CONFIG_TRACER_SNAPSHOT makes a generic snapsh
3510 available to all non latency tracers. (Latenc
3511 record max latency, such as "irqsoff" or "wak
3512 this feature, since those are already using t
3513 mechanism internally.)
3514
3515 Snapshot preserves a current trace buffer at
3516 in time without stopping tracing. Ftrace swap
3517 buffer with a spare buffer, and tracing conti
3518 current (=previous spare) buffer.
3519
3520 The following tracefs files in "tracing" are
3521 feature:
3522
3523 snapshot:
3524
3525 This is used to take a snapshot and t
3526 of the snapshot. Echo 1 into this fil
3527 spare buffer and to take a snapshot (
3528 the snapshot from this file in the sa
3529 "trace" (described above in the secti
3530 System"). Both reads snapshot and tra
3531 in parallel. When the spare buffer is
3532 0 frees it, and echoing else (positiv
3533 snapshot contents.
3534 More details are shown in the table b
3535
3536 +--------------+------------+--------
3537 |status\\input | 0 | 1
3538 +==============+============+========
3539 |not allocated |(do nothing)| alloc+s
3540 +--------------+------------+--------
3541 |allocated | free | swap
3542 +--------------+------------+--------
3543
3544 Here is an example of using the snapshot feat
3545 ::
3546
3547 # echo 1 > events/sched/enable
3548 # echo 1 > snapshot
3549 # cat snapshot
3550 # tracer: nop
3551 #
3552 # entries-in-buffer/entries-written: 71/71
3553 #
3554 # _-----=> irq
3555 # / _----=> nee
3556 # | / _---=> har
3557 # || / _--=> pre
3558 # ||| / dela
3559 # TASK-PID CPU# |||| TIMEST
3560 # | | | |||| |
3561 <idle>-0 [005] d... 2440.603
3562 sleep-2242 [005] d... 2440.603
3563 [...]
3564 <idle>-0 [002] d... 2440.70723
3565
3566 # cat trace
3567 # tracer: nop
3568 #
3569 # entries-in-buffer/entries-written: 77/77
3570 #
3571 # _-----=> irq
3572 # / _----=> nee
3573 # | / _---=> har
3574 # || / _--=> pre
3575 # ||| / dela
3576 # TASK-PID CPU# |||| TIMEST
3577 # | | | |||| |
3578 <idle>-0 [007] d... 2440.707
3579 snapshot-test-2-2229 [002] d... 2440.707
3580 [...]
3581
3582
3583 If you try to use this snapshot feature when
3584 one of the latency tracers, you will get the
3585 ::
3586
3587 # echo wakeup > current_tracer
3588 # echo 1 > snapshot
3589 bash: echo: write error: Device or resource
3590 # cat snapshot
3591 cat: snapshot: Device or resource busy
3592
3593
3594 Instances
3595 ---------
3596 In the tracefs tracing directory, there is a
3597 This directory can have new directories creat
3598 mkdir, and removing directories with rmdir. T
3599 with mkdir in this directory will already con
3600 directories after it is created.
3601 ::
3602
3603 # mkdir instances/foo
3604 # ls instances/foo
3605 buffer_size_kb buffer_total_size_kb event
3606 set_event snapshot trace trace_clock tr
3607 trace_pipe tracing_on
3608
3609 As you can see, the new directory looks simil
3610 itself. In fact, it is very similar, except t
3611 events are agnostic from the main directory,
3612 instances that are created.
3613
3614 The files in the new directory work just like
3615 same name in the tracing directory except the
3616 is a separate and new buffer. The files affec
3617 affect the main buffer with the exception of
3618 the trace_options affect all instances and th
3619 the same, but this may change in future relea
3620 may become specific to the instance they resi
3621
3622 Notice that none of the function tracer files
3623 current_tracer and available_tracers. This is
3624 can currently only have events enabled for th
3625 ::
3626
3627 # mkdir instances/foo
3628 # mkdir instances/bar
3629 # mkdir instances/zoot
3630 # echo 100000 > buffer_size_kb
3631 # echo 1000 > instances/foo/buffer_size_kb
3632 # echo 5000 > instances/bar/per_cpu/cpu1/bu
3633 # echo function > current_trace
3634 # echo 1 > instances/foo/events/sched/sched
3635 # echo 1 > instances/foo/events/sched/sched
3636 # echo 1 > instances/foo/events/sched/sched
3637 # echo 1 > instances/bar/events/irq/enable
3638 # echo 1 > instances/zoot/events/syscalls/e
3639 # cat trace_pipe
3640 CPU:2 [LOST 11745 EVENTS]
3641 bash-2044 [002] .... 10594.481
3642 bash-2044 [002] d... 10594.481
3643 bash-2044 [002] d..1 10594.481
3644 bash-2044 [002] d..1 10594.481
3645 bash-2044 [002] d..1 10594.481
3646 bash-2044 [002] d... 10594.481
3647 bash-2044 [002] d... 10594.481
3648 bash-2044 [002] d... 10594.481
3649 bash-2044 [002] d... 10594.481
3650 bash-2044 [002] d... 10594.481
3651 bash-2044 [002] .... 10594.481
3652 [...]
3653
3654 # cat instances/foo/trace_pipe
3655 bash-1998 [000] d..4 136.676
3656 bash-1998 [000] dN.4 136.676
3657 <idle>-0 [003] d.h3 136.676
3658 <idle>-0 [003] d..3 136.676
3659 rcu_preempt-9 [003] d..3 136.676
3660 bash-1998 [000] d..4 136.677
3661 bash-1998 [000] dN.4 136.677
3662 bash-1998 [000] d..3 136.677
3663 kworker/0:1-59 [000] d..4 136.677
3664 kworker/0:1-59 [000] d..3 136.677
3665 [...]
3666
3667 # cat instances/bar/trace_pipe
3668 migration/1-14 [001] d.h3 138.732
3669 <idle>-0 [001] dNh3 138.732
3670 bash-1998 [000] d.h1 138.733
3671 bash-1998 [000] d.h1 138.733
3672 bash-1998 [000] ..s2 138.733
3673 bash-1998 [000] ..s2 138.733
3674 bash-1998 [000] ..s2 138.733
3675 bash-1998 [000] ..s2 138.733
3676 sshd-1995 [001] d.h1 138.733
3677 sshd-1995 [001] d.h1 138.733
3678 sshd-1995 [001] d.h1 138.733
3679 sshd-1995 [001] d.h1 138.733
3680 [...]
3681
3682 # cat instances/zoot/trace
3683 # tracer: nop
3684 #
3685 # entries-in-buffer/entries-written: 18996/
3686 #
3687 # _-----=> irq
3688 # / _----=> nee
3689 # | / _---=> har
3690 # || / _--=> pre
3691 # ||| / dela
3692 # TASK-PID CPU# |||| TIMEST
3693 # | | | |||| |
3694 bash-1998 [000] d... 140.733
3695 bash-1998 [000] d... 140.733
3696 bash-1998 [000] d... 140.733
3697 bash-1998 [000] d... 140.733
3698 bash-1998 [000] d... 140.733
3699 bash-1998 [000] d... 140.733
3700 bash-1998 [000] d... 140.733
3701 bash-1998 [000] d... 140.733
3702 bash-1998 [000] d... 140.733
3703 bash-1998 [000] d... 140.733
3704 bash-1998 [000] d... 140.733
3705
3706 You can see that the trace of the top most tr
3707 the function tracing. The foo instance displa
3708 switches.
3709
3710 To remove the instances, simply delete their
3711 ::
3712
3713 # rmdir instances/foo
3714 # rmdir instances/bar
3715 # rmdir instances/zoot
3716
3717 Note, if a process has a trace file open in o
3718 directories, the rmdir will fail with EBUSY.
3719
3720
3721 Stack trace
3722 -----------
3723 Since the kernel has a fixed sized stack, it
3724 waste it in functions. A kernel developer mus
3725 what they allocate on the stack. If they add
3726 can be in danger of a stack overflow, and cor
3727 usually leading to a system panic.
3728
3729 There are some tools that check this, usually
3730 periodically checking usage. But if you can p
3731 at every function call that will become very
3732 a function tracer, it makes it convenient to
3733 at every function call. This is enabled via t
3734
3735 CONFIG_STACK_TRACER enables the ftrace stack
3736 To enable it, write a '1' into /proc/sys/kern
3737 ::
3738
3739 # echo 1 > /proc/sys/kernel/stack_tracer_ena
3740
3741 You can also enable it from the kernel comman
3742 the stack size of the kernel during boot up,
3743 to the kernel command line parameter.
3744
3745 After running it for a few minutes, the outpu
3746 ::
3747
3748 # cat stack_max_size
3749 2928
3750
3751 # cat stack_trace
3752 Depth Size Location (18 ent
3753 ----- ---- --------
3754 0) 2928 224 update_sd_lb_stats+
3755 1) 2704 160 find_busiest_group+
3756 2) 2544 256 load_balance+0xd9/0
3757 3) 2288 80 idle_balance+0xbb/0
3758 4) 2208 128 __schedule+0x26e/0x
3759 5) 2080 16 schedule+0x64/0x66
3760 6) 2064 128 schedule_timeout+0x
3761 7) 1936 112 wait_for_common+0x9
3762 8) 1824 16 wait_for_completion
3763 9) 1808 128 flush_work+0xfe/0x1
3764 10) 1680 16 tty_flush_to_ldisc+
3765 11) 1664 48 input_available_p+0
3766 12) 1616 48 n_tty_poll+0x6d/0x1
3767 13) 1568 64 tty_poll+0x64/0x7f
3768 14) 1504 880 do_select+0x31e/0x5
3769 15) 624 400 core_sys_select+0x1
3770 16) 224 96 sys_select+0x91/0xb
3771 17) 128 128 system_call_fastpat
3772
3773 Note, if -mfentry is being used by gcc, funct
3774 they set up the stack frame. This means that
3775 are not tested by the stack tracer when -mfen
3776
3777 Currently, -mfentry is used by gcc 4.6.0 and
3778
3779 More
3780 ----
3781 More details can be found in the source code,
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.