| blob | 006f47c7d9135530d473ac80df5f83008ee408b0 |
1 ftrace - Function Tracer
2 ========================
4 Copyright 2008 Red Hat Inc.
5 Author: Steven Rostedt <srostedt@redhat.com>
6 License: The GNU Free Documentation License, Version 1.2
7 (dual licensed under the GPL v2)
8 Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton,
9 John Kacur, and David Teigland.
10 Written for: 2.6.28-rc2
11 Updated for: 3.10
13 Introduction
14 ------------
16 Ftrace is an internal tracer designed to help out developers and
17 designers of systems to find what is going on inside the kernel.
18 It can be used for debugging or analyzing latencies and
19 performance issues that take place outside of user-space.
21 Although ftrace is typically considered the function tracer, it
22 is really a frame work of several assorted tracing utilities.
23 There's latency tracing to examine what occurs between interrupts
24 disabled and enabled, as well as for preemption and from a time
25 a task is woken to the task is actually scheduled in.
27 One of the most common uses of ftrace is the event tracing.
28 Through out the kernel is hundreds of static event points that
29 can be enabled via the debugfs file system to see what is
30 going on in certain parts of the kernel.
33 Implementation Details
34 ----------------------
36 See ftrace-design.txt for details for arch porters and such.
39 The File System
40 ---------------
42 Ftrace uses the debugfs file system to hold the control files as
43 well as the files to display output.
45 When debugfs is configured into the kernel (which selecting any ftrace
46 option will do) the directory /sys/kernel/debug will be created. To mount
47 this directory, you can add to your /etc/fstab file:
49 debugfs /sys/kernel/debug debugfs defaults 0 0
51 Or you can mount it at run time with:
53 mount -t debugfs nodev /sys/kernel/debug
55 For quicker access to that directory you may want to make a soft link to
56 it:
58 ln -s /sys/kernel/debug /debug
60 Any selected ftrace option will also create a directory called tracing
61 within the debugfs. The rest of the document will assume that you are in
62 the ftrace directory (cd /sys/kernel/debug/tracing) and will only concentrate
63 on the files within that directory and not distract from the content with
64 the extended "/sys/kernel/debug/tracing" path name.
66 That's it! (assuming that you have ftrace configured into your kernel)
68 After mounting debugfs, you can see a directory called
69 "tracing". This directory contains the control and output files
70 of ftrace. Here is a list of some of the key files:
73 Note: all time values are in microseconds.
75 current_tracer:
77 This is used to set or display the current tracer
78 that is configured.
80 available_tracers:
82 This holds the different types of tracers that
83 have been compiled into the kernel. The
84 tracers listed here can be configured by
85 echoing their name into current_tracer.
87 tracing_on:
89 This sets or displays whether writing to the trace
90 ring buffer is enabled. Echo 0 into this file to disable
91 the tracer or 1 to enable it. Note, this only disables
92 writing to the ring buffer, the tracing overhead may
93 still be occurring.
95 trace:
97 This file holds the output of the trace in a human
98 readable format (described below).
100 trace_pipe:
102 The output is the same as the "trace" file but this
103 file is meant to be streamed with live tracing.
104 Reads from this file will block until new data is
105 retrieved. Unlike the "trace" file, this file is a
106 consumer. This means reading from this file causes
107 sequential reads to display more current data. Once
108 data is read from this file, it is consumed, and
109 will not be read again with a sequential read. The
110 "trace" file is static, and if the tracer is not
111 adding more data, it will display the same
112 information every time it is read.
114 trace_options:
116 This file lets the user control the amount of data
117 that is displayed in one of the above output
118 files. Options also exist to modify how a tracer
119 or events work (stack traces, timestamps, etc).
121 options:
123 This is a directory that has a file for every available
124 trace option (also in trace_options). Options may also be set
125 or cleared by writing a "1" or "0" respectively into the
126 corresponding file with the option name.
128 tracing_max_latency:
130 Some of the tracers record the max latency.
131 For example, the time interrupts are disabled.
132 This time is saved in this file. The max trace
133 will also be stored, and displayed by "trace".
134 A new max trace will only be recorded if the
135 latency is greater than the value in this
136 file. (in microseconds)
138 tracing_thresh:
140 Some latency tracers will record a trace whenever the
141 latency is greater than the number in this file.
142 Only active when the file contains a number greater than 0.
143 (in microseconds)
145 buffer_size_kb:
147 This sets or displays the number of kilobytes each CPU
148 buffer holds. By default, the trace buffers are the same size
149 for each CPU. The displayed number is the size of the
150 CPU buffer and not total size of all buffers. The
151 trace buffers are allocated in pages (blocks of memory
152 that the kernel uses for allocation, usually 4 KB in size).
153 If the last page allocated has room for more bytes
154 than requested, the rest of the page will be used,
155 making the actual allocation bigger than requested.
156 ( Note, the size may not be a multiple of the page size
157 due to buffer management meta-data. )
159 buffer_total_size_kb:
161 This displays the total combined size of all the trace buffers.
163 free_buffer:
165 If a process is performing the tracing, and the ring buffer
166 should be shrunk "freed" when the process is finished, even
167 if it were to be killed by a signal, this file can be used
168 for that purpose. On close of this file, the ring buffer will
169 be resized to its minimum size. Having a process that is tracing
170 also open this file, when the process exits its file descriptor
171 for this file will be closed, and in doing so, the ring buffer
172 will be "freed".
174 It may also stop tracing if disable_on_free option is set.
176 tracing_cpumask:
178 This is a mask that lets the user only trace
179 on specified CPUs. The format is a hex string
180 representing the CPUs.
182 set_ftrace_filter:
184 When dynamic ftrace is configured in (see the
185 section below "dynamic ftrace"), the code is dynamically
186 modified (code text rewrite) to disable calling of the
187 function profiler (mcount). This lets tracing be configured
188 in with practically no overhead in performance. This also
189 has a side effect of enabling or disabling specific functions
190 to be traced. Echoing names of functions into this file
191 will limit the trace to only those functions.
193 This interface also allows for commands to be used. See the
194 "Filter commands" section for more details.
196 set_ftrace_notrace:
198 This has an effect opposite to that of
199 set_ftrace_filter. Any function that is added here will not
200 be traced. If a function exists in both set_ftrace_filter
201 and set_ftrace_notrace, the function will _not_ be traced.
203 set_ftrace_pid:
205 Have the function tracer only trace a single thread.
207 set_event_pid:
209 Have the events only trace a task with a PID listed in this file.
210 Note, sched_switch and sched_wake_up will also trace events
211 listed in this file.
213 To have the PIDs of children of tasks with their PID in this file
214 added on fork, enable the "event-fork" option. That option will also
215 cause the PIDs of tasks to be removed from this file when the task
216 exits.
218 set_graph_function:
220 Set a "trigger" function where tracing should start
221 with the function graph tracer (See the section
222 "dynamic ftrace" for more details).
224 available_filter_functions:
226 This lists the functions that ftrace
227 has processed and can trace. These are the function
228 names that you can pass to "set_ftrace_filter" or
229 "set_ftrace_notrace". (See the section "dynamic ftrace"
230 below for more details.)
232 enabled_functions:
234 This file is more for debugging ftrace, but can also be useful
235 in seeing if any function has a callback attached to it.
236 Not only does the trace infrastructure use ftrace function
237 trace utility, but other subsystems might too. This file
238 displays all functions that have a callback attached to them
239 as well as the number of callbacks that have been attached.
240 Note, a callback may also call multiple functions which will
241 not be listed in this count.
243 If the callback registered to be traced by a function with
244 the "save regs" attribute (thus even more overhead), a 'R'
245 will be displayed on the same line as the function that
246 is returning registers.
248 If the callback registered to be traced by a function with
249 the "ip modify" attribute (thus the regs->ip can be changed),
250 an 'I' will be displayed on the same line as the function that
251 can be overridden.
253 function_profile_enabled:
255 When set it will enable all functions with either the function
256 tracer, or if enabled, the function graph tracer. It will
257 keep a histogram of the number of functions that were called
258 and if run with the function graph tracer, it will also keep
259 track of the time spent in those functions. The histogram
260 content can be displayed in the files:
262 trace_stats/function<cpu> ( function0, function1, etc).
264 trace_stats:
266 A directory that holds different tracing stats.
268 kprobe_events:
270 Enable dynamic trace points. See kprobetrace.txt.
272 kprobe_profile:
274 Dynamic trace points stats. See kprobetrace.txt.
276 max_graph_depth:
278 Used with the function graph tracer. This is the max depth
279 it will trace into a function. Setting this to a value of
280 one will show only the first kernel function that is called
281 from user space.
283 printk_formats:
285 This is for tools that read the raw format files. If an event in
286 the ring buffer references a string (currently only trace_printk()
287 does this), only a pointer to the string is recorded into the buffer
288 and not the string itself. This prevents tools from knowing what
289 that string was. This file displays the string and address for
290 the string allowing tools to map the pointers to what the
291 strings were.
293 saved_cmdlines:
295 Only the pid of the task is recorded in a trace event unless
296 the event specifically saves the task comm as well. Ftrace
297 makes a cache of pid mappings to comms to try to display
298 comms for events. If a pid for a comm is not listed, then
299 "<...>" is displayed in the output.
301 snapshot:
303 This displays the "snapshot" buffer and also lets the user
304 take a snapshot of the current running trace.
305 See the "Snapshot" section below for more details.
307 stack_max_size:
309 When the stack tracer is activated, this will display the
310 maximum stack size it has encountered.
311 See the "Stack Trace" section below.
313 stack_trace:
315 This displays the stack back trace of the largest stack
316 that was encountered when the stack tracer is activated.
317 See the "Stack Trace" section below.
319 stack_trace_filter:
321 This is similar to "set_ftrace_filter" but it limits what
322 functions the stack tracer will check.
324 trace_clock:
326 Whenever an event is recorded into the ring buffer, a
327 "timestamp" is added. This stamp comes from a specified
328 clock. By default, ftrace uses the "local" clock. This
329 clock is very fast and strictly per cpu, but on some
330 systems it may not be monotonic with respect to other
331 CPUs. In other words, the local clocks may not be in sync
332 with local clocks on other CPUs.
334 Usual clocks for tracing:
336 # cat trace_clock
337 [local] global counter x86-tsc
339 local: Default clock, but may not be in sync across CPUs
341 global: This clock is in sync with all CPUs but may
342 be a bit slower than the local clock.
344 counter: This is not a clock at all, but literally an atomic
345 counter. It counts up one by one, but is in sync
346 with all CPUs. This is useful when you need to
347 know exactly the order events occurred with respect to
348 each other on different CPUs.
350 uptime: This uses the jiffies counter and the time stamp
351 is relative to the time since boot up.
353 perf: This makes ftrace use the same clock that perf uses.
354 Eventually perf will be able to read ftrace buffers
355 and this will help out in interleaving the data.
357 x86-tsc: Architectures may define their own clocks. For
358 example, x86 uses its own TSC cycle clock here.
360 ppc-tb: This uses the powerpc timebase register value.
361 This is in sync across CPUs and can also be used
362 to correlate events across hypervisor/guest if
363 tb_offset is known.
365 mono: This uses the fast monotonic clock (CLOCK_MONOTONIC)
366 which is monotonic and is subject to NTP rate adjustments.
368 mono_raw:
369 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
370 which is montonic but is not subject to any rate adjustments
371 and ticks at the same rate as the hardware clocksource.
373 boot: This is the boot clock (CLOCK_BOOTTIME) and is based on the
374 fast monotonic clock, but also accounts for time spent in
375 suspend. Since the clock access is designed for use in
376 tracing in the suspend path, some side effects are possible
377 if clock is accessed after the suspend time is accounted before
378 the fast mono clock is updated. In this case, the clock update
379 appears to happen slightly sooner than it normally would have.
380 Also on 32-bit systems, it's possible that the 64-bit boot offset
381 sees a partial update. These effects are rare and post
382 processing should be able to handle them. See comments in the
383 ktime_get_boot_fast_ns() function for more information.
385 To set a clock, simply echo the clock name into this file.
387 echo global > trace_clock
389 trace_marker:
391 This is a very useful file for synchronizing user space
392 with events happening in the kernel. Writing strings into
393 this file will be written into the ftrace buffer.
395 It is useful in applications to open this file at the start
396 of the application and just reference the file descriptor
397 for the file.
399 void trace_write(const char *fmt, ...)
400 {
401 va_list ap;
402 char buf[256];
403 int n;
405 if (trace_fd < 0)
406 return;
408 va_start(ap, fmt);
409 n = vsnprintf(buf, 256, fmt, ap);
410 va_end(ap);
412 write(trace_fd, buf, n);
413 }
415 start:
417 trace_fd = open("trace_marker", WR_ONLY);
419 trace_marker_raw:
421 This is similar to trace_marker above, but is meant for for binary data
422 to be written to it, where a tool can be used to parse the data
423 from trace_pipe_raw.
425 uprobe_events:
427 Add dynamic tracepoints in programs.
428 See uprobetracer.txt
430 uprobe_profile:
432 Uprobe statistics. See uprobetrace.txt
434 instances:
436 This is a way to make multiple trace buffers where different
437 events can be recorded in different buffers.
438 See "Instances" section below.
440 events:
442 This is the trace event directory. It holds event tracepoints
443 (also known as static tracepoints) that have been compiled
444 into the kernel. It shows what event tracepoints exist
445 and how they are grouped by system. There are "enable"
446 files at various levels that can enable the tracepoints
447 when a "1" is written to them.
449 See events.txt for more information.
451 per_cpu:
453 This is a directory that contains the trace per_cpu information.
455 per_cpu/cpu0/buffer_size_kb:
457 The ftrace buffer is defined per_cpu. That is, there's a separate
458 buffer for each CPU to allow writes to be done atomically,
459 and free from cache bouncing. These buffers may have different
460 size buffers. This file is similar to the buffer_size_kb
461 file, but it only displays or sets the buffer size for the
462 specific CPU. (here cpu0).
464 per_cpu/cpu0/trace:
466 This is similar to the "trace" file, but it will only display
467 the data specific for the CPU. If written to, it only clears
468 the specific CPU buffer.
470 per_cpu/cpu0/trace_pipe
472 This is similar to the "trace_pipe" file, and is a consuming
473 read, but it will only display (and consume) the data specific
474 for the CPU.
476 per_cpu/cpu0/trace_pipe_raw
478 For tools that can parse the ftrace ring buffer binary format,
479 the trace_pipe_raw file can be used to extract the data
480 from the ring buffer directly. With the use of the splice()
481 system call, the buffer data can be quickly transferred to
482 a file or to the network where a server is collecting the
483 data.
485 Like trace_pipe, this is a consuming reader, where multiple
486 reads will always produce different data.
488 per_cpu/cpu0/snapshot:
490 This is similar to the main "snapshot" file, but will only
491 snapshot the current CPU (if supported). It only displays
492 the content of the snapshot for a given CPU, and if
493 written to, only clears this CPU buffer.
495 per_cpu/cpu0/snapshot_raw:
497 Similar to the trace_pipe_raw, but will read the binary format
498 from the snapshot buffer for the given CPU.
500 per_cpu/cpu0/stats:
502 This displays certain stats about the ring buffer:
504 entries: The number of events that are still in the buffer.
506 overrun: The number of lost events due to overwriting when
507 the buffer was full.
509 commit overrun: Should always be zero.
510 This gets set if so many events happened within a nested
511 event (ring buffer is re-entrant), that it fills the
512 buffer and starts dropping events.
514 bytes: Bytes actually read (not overwritten).
516 oldest event ts: The oldest timestamp in the buffer
518 now ts: The current timestamp
520 dropped events: Events lost due to overwrite option being off.
522 read events: The number of events read.
524 The Tracers
525 -----------
527 Here is the list of current tracers that may be configured.
529 "function"
531 Function call tracer to trace all kernel functions.
533 "function_graph"
535 Similar to the function tracer except that the
536 function tracer probes the functions on their entry
537 whereas the function graph tracer traces on both entry
538 and exit of the functions. It then provides the ability
539 to draw a graph of function calls similar to C code
540 source.
542 "irqsoff"
544 Traces the areas that disable interrupts and saves
545 the trace with the longest max latency.
546 See tracing_max_latency. When a new max is recorded,
547 it replaces the old trace. It is best to view this
548 trace with the latency-format option enabled.
550 "preemptoff"
552 Similar to irqsoff but traces and records the amount of
553 time for which preemption is disabled.
555 "preemptirqsoff"
557 Similar to irqsoff and preemptoff, but traces and
558 records the largest time for which irqs and/or preemption
559 is disabled.
561 "wakeup"
563 Traces and records the max latency that it takes for
564 the highest priority task to get scheduled after
565 it has been woken up.
566 Traces all tasks as an average developer would expect.
568 "wakeup_rt"
570 Traces and records the max latency that it takes for just
571 RT tasks (as the current "wakeup" does). This is useful
572 for those interested in wake up timings of RT tasks.
574 "nop"
576 This is the "trace nothing" tracer. To remove all
577 tracers from tracing simply echo "nop" into
578 current_tracer.
581 Examples of using the tracer
582 ----------------------------
584 Here are typical examples of using the tracers when controlling
585 them only with the debugfs interface (without using any
586 user-land utilities).
588 Output format:
589 --------------
591 Here is an example of the output format of the file "trace"
593 --------
594 # tracer: function
595 #
596 # entries-in-buffer/entries-written: 140080/250280 #P:4
597 #
598 # _-----=> irqs-off
599 # / _----=> need-resched
600 # | / _---=> hardirq/softirq
601 # || / _--=> preempt-depth
602 # ||| / delay
603 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
604 # | | | |||| | |
605 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
606 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
607 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
608 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
609 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
610 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
611 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
612 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
613 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
614 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
615 --------
617 A header is printed with the tracer name that is represented by
618 the trace. In this case the tracer is "function". Then it shows the
619 number of events in the buffer as well as the total number of entries
620 that were written. The difference is the number of entries that were
621 lost due to the buffer filling up (250280 - 140080 = 110200 events
622 lost).
624 The header explains the content of the events. Task name "bash", the task
625 PID "1977", the CPU that it was running on "000", the latency format
626 (explained below), the timestamp in <secs>.<usecs> format, the
627 function name that was traced "sys_close" and the parent function that
628 called this function "system_call_fastpath". The timestamp is the time
629 at which the function was entered.
631 Latency trace format
632 --------------------
634 When the latency-format option is enabled or when one of the latency
635 tracers is set, the trace file gives somewhat more information to see
636 why a latency happened. Here is a typical trace.
638 # tracer: irqsoff
639 #
640 # irqsoff latency trace v1.1.5 on 3.8.0-test+
641 # --------------------------------------------------------------------
642 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
643 # -----------------
644 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
645 # -----------------
646 # => started at: __lock_task_sighand
647 # => ended at: _raw_spin_unlock_irqrestore
648 #
649 #
650 # _------=> CPU#
651 # / _-----=> irqs-off
652 # | / _----=> need-resched
653 # || / _---=> hardirq/softirq
654 # ||| / _--=> preempt-depth
655 # |||| / delay
656 # cmd pid ||||| time | caller
657 # \ / ||||| \ | /
658 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
659 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
660 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
661 ps-6143 2d..1 306us : <stack trace>
662 => trace_hardirqs_on_caller
663 => trace_hardirqs_on
664 => _raw_spin_unlock_irqrestore
665 => do_task_stat
666 => proc_tgid_stat
667 => proc_single_show
668 => seq_read
669 => vfs_read
670 => sys_read
671 => system_call_fastpath
674 This shows that the current tracer is "irqsoff" tracing the time
675 for which interrupts were disabled. It gives the trace version (which
676 never changes) and the version of the kernel upon which this was executed on
677 (3.10). Then it displays the max latency in microseconds (259 us). The number
678 of trace entries displayed and the total number (both are four: #4/4).
679 VP, KP, SP, and HP are always zero and are reserved for later use.
680 #P is the number of online CPUs (#P:4).
682 The task is the process that was running when the latency
683 occurred. (ps pid: 6143).
685 The start and stop (the functions in which the interrupts were
686 disabled and enabled respectively) that caused the latencies:
688 __lock_task_sighand is where the interrupts were disabled.
689 _raw_spin_unlock_irqrestore is where they were enabled again.
691 The next lines after the header are the trace itself. The header
692 explains which is which.
694 cmd: The name of the process in the trace.
696 pid: The PID of that process.
698 CPU#: The CPU which the process was running on.
700 irqs-off: 'd' interrupts are disabled. '.' otherwise.
701 Note: If the architecture does not support a way to
702 read the irq flags variable, an 'X' will always
703 be printed here.
705 need-resched:
706 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
707 'n' only TIF_NEED_RESCHED is set,
708 'p' only PREEMPT_NEED_RESCHED is set,
709 '.' otherwise.
711 hardirq/softirq:
712 'H' - hard irq occurred inside a softirq.
713 'h' - hard irq is running
714 's' - soft irq is running
715 '.' - normal context.
717 preempt-depth: The level of preempt_disabled
719 The above is mostly meaningful for kernel developers.
721 time: When the latency-format option is enabled, the trace file
722 output includes a timestamp relative to the start of the
723 trace. This differs from the output when latency-format
724 is disabled, which includes an absolute timestamp.
726 delay: This is just to help catch your eye a bit better. And
727 needs to be fixed to be only relative to the same CPU.
728 The marks are determined by the difference between this
729 current trace and the next trace.
730 '$' - greater than 1 second
731 '@' - greater than 100 milisecond
732 '*' - greater than 10 milisecond
733 '#' - greater than 1000 microsecond
734 '!' - greater than 100 microsecond
735 '+' - greater than 10 microsecond
736 ' ' - less than or equal to 10 microsecond.
738 The rest is the same as the 'trace' file.
740 Note, the latency tracers will usually end with a back trace
741 to easily find where the latency occurred.
743 trace_options
744 -------------
746 The trace_options file (or the options directory) is used to control
747 what gets printed in the trace output, or manipulate the tracers.
748 To see what is available, simply cat the file:
750 cat trace_options
751 print-parent
752 nosym-offset
753 nosym-addr
754 noverbose
755 noraw
756 nohex
757 nobin
758 noblock
759 trace_printk
760 nobranch
761 annotate
762 nouserstacktrace
763 nosym-userobj
764 noprintk-msg-only
765 context-info
766 nolatency-format
767 sleep-time
768 graph-time
769 record-cmd
770 overwrite
771 nodisable_on_free
772 irq-info
773 markers
774 noevent-fork
775 function-trace
776 nodisplay-graph
777 nostacktrace
779 To disable one of the options, echo in the option prepended with
780 "no".
782 echo noprint-parent > trace_options
784 To enable an option, leave off the "no".
786 echo sym-offset > trace_options
788 Here are the available options:
790 print-parent - On function traces, display the calling (parent)
791 function as well as the function being traced.
793 print-parent:
794 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
796 noprint-parent:
797 bash-4000 [01] 1477.606694: simple_strtoul
800 sym-offset - Display not only the function name, but also the
801 offset in the function. For example, instead of
802 seeing just "ktime_get", you will see
803 "ktime_get+0xb/0x20".
805 sym-offset:
806 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
808 sym-addr - this will also display the function address as well
809 as the function name.
811 sym-addr:
812 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
814 verbose - This deals with the trace file when the
815 latency-format option is enabled.
817 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
818 (+0.000ms): simple_strtoul (kstrtoul)
820 raw - This will display raw numbers. This option is best for
821 use with user applications that can translate the raw
822 numbers better than having it done in the kernel.
824 hex - Similar to raw, but the numbers will be in a hexadecimal
825 format.
827 bin - This will print out the formats in raw binary.
829 block - When set, reading trace_pipe will not block when polled.
831 trace_printk - Can disable trace_printk() from writing into the buffer.
833 branch - Enable branch tracing with the tracer.
835 annotate - It is sometimes confusing when the CPU buffers are full
836 and one CPU buffer had a lot of events recently, thus
837 a shorter time frame, were another CPU may have only had
838 a few events, which lets it have older events. When
839 the trace is reported, it shows the oldest events first,
840 and it may look like only one CPU ran (the one with the
841 oldest events). When the annotate option is set, it will
842 display when a new CPU buffer started:
844 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
845 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
846 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
847 ##### CPU 2 buffer started ####
848 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
849 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
850 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
852 userstacktrace - This option changes the trace. It records a
853 stacktrace of the current userspace thread.
855 sym-userobj - when user stacktrace are enabled, look up which
856 object the address belongs to, and print a
857 relative address. This is especially useful when
858 ASLR is on, otherwise you don't get a chance to
859 resolve the address to object/file/line after
860 the app is no longer running
862 The lookup is performed when you read
863 trace,trace_pipe. Example:
865 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
866 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
869 printk-msg-only - When set, trace_printk()s will only show the format
870 and not their parameters (if trace_bprintk() or
871 trace_bputs() was used to save the trace_printk()).
873 context-info - Show only the event data. Hides the comm, PID,
874 timestamp, CPU, and other useful data.
876 latency-format - This option changes the trace. When
877 it is enabled, the trace displays
878 additional information about the
879 latencies, as described in "Latency
880 trace format".
882 sleep-time - When running function graph tracer, to include
883 the time a task schedules out in its function.
884 When enabled, it will account time the task has been
885 scheduled out as part of the function call.
887 graph-time - When running function profiler with function graph tracer,
888 to include the time to call nested functions. When this is
889 not set, the time reported for the function will only
890 include the time the function itself executed for, not the
891 time for functions that it called.
893 record-cmd - When any event or tracer is enabled, a hook is enabled
894 in the sched_switch trace point to fill comm cache
895 with mapped pids and comms. But this may cause some
896 overhead, and if you only care about pids, and not the
897 name of the task, disabling this option can lower the
898 impact of tracing.
900 overwrite - This controls what happens when the trace buffer is
901 full. If "1" (default), the oldest events are
902 discarded and overwritten. If "0", then the newest
903 events are discarded.
904 (see per_cpu/cpu0/stats for overrun and dropped)
906 disable_on_free - When the free_buffer is closed, tracing will
907 stop (tracing_on set to 0).
909 irq-info - Shows the interrupt, preempt count, need resched data.
910 When disabled, the trace looks like:
912 # tracer: function
913 #
914 # entries-in-buffer/entries-written: 144405/9452052 #P:4
915 #
916 # TASK-PID CPU# TIMESTAMP FUNCTION
917 # | | | | |
918 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
919 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
920 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
923 markers - When set, the trace_marker is writable (only by root).
924 When disabled, the trace_marker will error with EINVAL
925 on write.
927 event-fork - When set, tasks with PIDs listed in set_event_pid will have
928 the PIDs of their children added to set_event_pid when those
929 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
930 their PIDs will be removed from the file.
932 function-trace - The latency tracers will enable function tracing
933 if this option is enabled (default it is). When
934 it is disabled, the latency tracers do not trace
935 functions. This keeps the overhead of the tracer down
936 when performing latency tests.
938 display-graph - When set, the latency tracers (irqsoff, wakeup, etc) will
939 use function graph tracing instead of function tracing.
941 stacktrace - This is one of the options that changes the trace
942 itself. When a trace is recorded, so is the stack
943 of functions. This allows for back traces of
944 trace sites.
946 Note: Some tracers have their own options. They only appear in this
947 file when the tracer is active. They always appear in the
948 options directory.
952 irqsoff
953 -------
955 When interrupts are disabled, the CPU can not react to any other
956 external event (besides NMIs and SMIs). This prevents the timer
957 interrupt from triggering or the mouse interrupt from letting
958 the kernel know of a new mouse event. The result is a latency
959 with the reaction time.
961 The irqsoff tracer tracks the time for which interrupts are
962 disabled. When a new maximum latency is hit, the tracer saves
963 the trace leading up to that latency point so that every time a
964 new maximum is reached, the old saved trace is discarded and the
965 new trace is saved.
967 To reset the maximum, echo 0 into tracing_max_latency. Here is
968 an example:
970 # echo 0 > options/function-trace
971 # echo irqsoff > current_tracer
972 # echo 1 > tracing_on
973 # echo 0 > tracing_max_latency
974 # ls -ltr
975 [...]
976 # echo 0 > tracing_on
977 # cat trace
978 # tracer: irqsoff
979 #
980 # irqsoff latency trace v1.1.5 on 3.8.0-test+
981 # --------------------------------------------------------------------
982 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
983 # -----------------
984 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
985 # -----------------
986 # => started at: run_timer_softirq
987 # => ended at: run_timer_softirq
988 #
989 #
990 # _------=> CPU#
991 # / _-----=> irqs-off
992 # | / _----=> need-resched
993 # || / _---=> hardirq/softirq
994 # ||| / _--=> preempt-depth
995 # |||| / delay
996 # cmd pid ||||| time | caller
997 # \ / ||||| \ | /
998 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
999 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
1000 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
1001 <idle>-0 0dNs3 25us : <stack trace>
1002 => _raw_spin_unlock_irq
1003 => run_timer_softirq
1004 => __do_softirq
1005 => call_softirq
1006 => do_softirq
1007 => irq_exit
1008 => smp_apic_timer_interrupt
1009 => apic_timer_interrupt
1010 => rcu_idle_exit
1011 => cpu_idle
1012 => rest_init
1013 => start_kernel
1014 => x86_64_start_reservations
1015 => x86_64_start_kernel
1017 Here we see that that we had a latency of 16 microseconds (which is
1018 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
1019 interrupts. The difference between the 16 and the displayed
1020 timestamp 25us occurred because the clock was incremented
1021 between the time of recording the max latency and the time of
1022 recording the function that had that latency.
1024 Note the above example had function-trace not set. If we set
1025 function-trace, we get a much larger output:
1027 with echo 1 > options/function-trace
1029 # tracer: irqsoff
1030 #
1031 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1032 # --------------------------------------------------------------------
1033 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1034 # -----------------
1035 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1036 # -----------------
1037 # => started at: ata_scsi_queuecmd
1038 # => ended at: ata_scsi_queuecmd
1039 #
1040 #
1041 # _------=> CPU#
1042 # / _-----=> irqs-off
1043 # | / _----=> need-resched
1044 # || / _---=> hardirq/softirq
1045 # ||| / _--=> preempt-depth
1046 # |||| / delay
1047 # cmd pid ||||| time | caller
1048 # \ / ||||| \ | /
1049 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1050 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1051 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1052 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1053 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1054 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1055 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1056 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1057 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1058 [...]
1059 bash-2042 3d..1 67us : delay_tsc <-__delay
1060 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1061 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1062 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1063 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1064 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1065 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1066 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1067 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1068 bash-2042 3d..1 120us : <stack trace>
1069 => _raw_spin_unlock_irqrestore
1070 => ata_scsi_queuecmd
1071 => scsi_dispatch_cmd
1072 => scsi_request_fn
1073 => __blk_run_queue_uncond
1074 => __blk_run_queue
1075 => blk_queue_bio
1076 => generic_make_request
1077 => submit_bio
1078 => submit_bh
1079 => __ext3_get_inode_loc
1080 => ext3_iget
1081 => ext3_lookup
1082 => lookup_real
1083 => __lookup_hash
1084 => walk_component
1085 => lookup_last
1086 => path_lookupat
1087 => filename_lookup
1088 => user_path_at_empty
1089 => user_path_at
1090 => vfs_fstatat
1091 => vfs_stat
1092 => sys_newstat
1093 => system_call_fastpath
1096 Here we traced a 71 microsecond latency. But we also see all the
1097 functions that were called during that time. Note that by
1098 enabling function tracing, we incur an added overhead. This
1099 overhead may extend the latency times. But nevertheless, this
1100 trace has provided some very helpful debugging information.
1103 preemptoff
1104 ----------
1106 When preemption is disabled, we may be able to receive
1107 interrupts but the task cannot be preempted and a higher
1108 priority task must wait for preemption to be enabled again
1109 before it can preempt a lower priority task.
1111 The preemptoff tracer traces the places that disable preemption.
1112 Like the irqsoff tracer, it records the maximum latency for
1113 which preemption was disabled. The control of preemptoff tracer
1114 is much like the irqsoff tracer.
1116 # echo 0 > options/function-trace
1117 # echo preemptoff > current_tracer
1118 # echo 1 > tracing_on
1119 # echo 0 > tracing_max_latency
1120 # ls -ltr
1121 [...]
1122 # echo 0 > tracing_on
1123 # cat trace
1124 # tracer: preemptoff
1125 #
1126 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1127 # --------------------------------------------------------------------
1128 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1129 # -----------------
1130 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1131 # -----------------
1132 # => started at: do_IRQ
1133 # => ended at: do_IRQ
1134 #
1135 #
1136 # _------=> CPU#
1137 # / _-----=> irqs-off
1138 # | / _----=> need-resched
1139 # || / _---=> hardirq/softirq
1140 # ||| / _--=> preempt-depth
1141 # |||| / delay
1142 # cmd pid ||||| time | caller
1143 # \ / ||||| \ | /
1144 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1145 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1146 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1147 sshd-1991 1d..1 52us : <stack trace>
1148 => sub_preempt_count
1149 => irq_exit
1150 => do_IRQ
1151 => ret_from_intr
1154 This has some more changes. Preemption was disabled when an
1155 interrupt came in (notice the 'h'), and was enabled on exit.
1156 But we also see that interrupts have been disabled when entering
1157 the preempt off section and leaving it (the 'd'). We do not know if
1158 interrupts were enabled in the mean time or shortly after this
1159 was over.
1161 # tracer: preemptoff
1162 #
1163 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1164 # --------------------------------------------------------------------
1165 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1166 # -----------------
1167 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1168 # -----------------
1169 # => started at: wake_up_new_task
1170 # => ended at: task_rq_unlock
1171 #
1172 #
1173 # _------=> CPU#
1174 # / _-----=> irqs-off
1175 # | / _----=> need-resched
1176 # || / _---=> hardirq/softirq
1177 # ||| / _--=> preempt-depth
1178 # |||| / delay
1179 # cmd pid ||||| time | caller
1180 # \ / ||||| \ | /
1181 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1182 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1183 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1184 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1185 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1186 [...]
1187 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1188 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1189 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1190 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1191 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1192 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1193 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1194 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1195 [...]
1196 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1197 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1198 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1199 bash-1994 1d..2 36us : do_softirq <-irq_exit
1200 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1201 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1202 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1203 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1204 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1205 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1206 [...]
1207 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1208 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1209 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1210 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1211 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1212 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1213 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1214 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1215 bash-1994 1.N.1 104us : <stack trace>
1216 => sub_preempt_count
1217 => _raw_spin_unlock_irqrestore
1218 => task_rq_unlock
1219 => wake_up_new_task
1220 => do_fork
1221 => sys_clone
1222 => stub_clone
1225 The above is an example of the preemptoff trace with
1226 function-trace set. Here we see that interrupts were not disabled
1227 the entire time. The irq_enter code lets us know that we entered
1228 an interrupt 'h'. Before that, the functions being traced still
1229 show that it is not in an interrupt, but we can see from the
1230 functions themselves that this is not the case.
1232 preemptirqsoff
1233 --------------
1235 Knowing the locations that have interrupts disabled or
1236 preemption disabled for the longest times is helpful. But
1237 sometimes we would like to know when either preemption and/or
1238 interrupts are disabled.
1240 Consider the following code:
1242 local_irq_disable();
1243 call_function_with_irqs_off();
1244 preempt_disable();
1245 call_function_with_irqs_and_preemption_off();
1246 local_irq_enable();
1247 call_function_with_preemption_off();
1248 preempt_enable();
1250 The irqsoff tracer will record the total length of
1251 call_function_with_irqs_off() and
1252 call_function_with_irqs_and_preemption_off().
1254 The preemptoff tracer will record the total length of
1255 call_function_with_irqs_and_preemption_off() and
1256 call_function_with_preemption_off().
1258 But neither will trace the time that interrupts and/or
1259 preemption is disabled. This total time is the time that we can
1260 not schedule. To record this time, use the preemptirqsoff
1261 tracer.
1263 Again, using this trace is much like the irqsoff and preemptoff
1264 tracers.
1266 # echo 0 > options/function-trace
1267 # echo preemptirqsoff > current_tracer
1268 # echo 1 > tracing_on
1269 # echo 0 > tracing_max_latency
1270 # ls -ltr
1271 [...]
1272 # echo 0 > tracing_on
1273 # cat trace
1274 # tracer: preemptirqsoff
1275 #
1276 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1277 # --------------------------------------------------------------------
1278 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1279 # -----------------
1280 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1281 # -----------------
1282 # => started at: ata_scsi_queuecmd
1283 # => ended at: ata_scsi_queuecmd
1284 #
1285 #
1286 # _------=> CPU#
1287 # / _-----=> irqs-off
1288 # | / _----=> need-resched
1289 # || / _---=> hardirq/softirq
1290 # ||| / _--=> preempt-depth
1291 # |||| / delay
1292 # cmd pid ||||| time | caller
1293 # \ / ||||| \ | /
1294 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1295 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1296 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1297 ls-2230 3...1 111us : <stack trace>
1298 => sub_preempt_count
1299 => _raw_spin_unlock_irqrestore
1300 => ata_scsi_queuecmd
1301 => scsi_dispatch_cmd
1302 => scsi_request_fn
1303 => __blk_run_queue_uncond
1304 => __blk_run_queue
1305 => blk_queue_bio
1306 => generic_make_request
1307 => submit_bio
1308 => submit_bh
1309 => ext3_bread
1310 => ext3_dir_bread
1311 => htree_dirblock_to_tree
1312 => ext3_htree_fill_tree
1313 => ext3_readdir
1314 => vfs_readdir
1315 => sys_getdents
1316 => system_call_fastpath
1319 The trace_hardirqs_off_thunk is called from assembly on x86 when
1320 interrupts are disabled in the assembly code. Without the
1321 function tracing, we do not know if interrupts were enabled
1322 within the preemption points. We do see that it started with
1323 preemption enabled.
1325 Here is a trace with function-trace set:
1327 # tracer: preemptirqsoff
1328 #
1329 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1330 # --------------------------------------------------------------------
1331 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1332 # -----------------
1333 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1334 # -----------------
1335 # => started at: schedule
1336 # => ended at: mutex_unlock
1337 #
1338 #
1339 # _------=> CPU#
1340 # / _-----=> irqs-off
1341 # | / _----=> need-resched
1342 # || / _---=> hardirq/softirq
1343 # ||| / _--=> preempt-depth
1344 # |||| / delay
1345 # cmd pid ||||| time | caller
1346 # \ / ||||| \ | /
1347 kworker/-59 3...1 0us : __schedule <-schedule
1348 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1349 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1350 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1351 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1352 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1353 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1354 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1355 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1356 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1357 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1358 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1359 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1360 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1361 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1362 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1363 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1364 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1365 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1366 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1367 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1368 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1369 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1370 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1371 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1372 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1373 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1374 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1375 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1376 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1377 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1378 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1379 [...]
1380 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1381 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1382 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1383 ls-2269 3d..3 21us : do_softirq <-irq_exit
1384 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1385 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1386 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1387 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1388 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1389 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1390 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1391 [...]
1392 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1393 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1394 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1395 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1396 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1397 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1398 [...]
1399 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1400 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1401 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1402 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1403 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1404 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1405 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1406 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1407 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1408 ls-2269 3d... 186us : <stack trace>
1409 => __mutex_unlock_slowpath
1410 => mutex_unlock
1411 => process_output
1412 => n_tty_write
1413 => tty_write
1414 => vfs_write
1415 => sys_write
1416 => system_call_fastpath
1418 This is an interesting trace. It started with kworker running and
1419 scheduling out and ls taking over. But as soon as ls released the
1420 rq lock and enabled interrupts (but not preemption) an interrupt
1421 triggered. When the interrupt finished, it started running softirqs.
1422 But while the softirq was running, another interrupt triggered.
1423 When an interrupt is running inside a softirq, the annotation is 'H'.
1426 wakeup
1427 ------
1429 One common case that people are interested in tracing is the
1430 time it takes for a task that is woken to actually wake up.
1431 Now for non Real-Time tasks, this can be arbitrary. But tracing
1432 it none the less can be interesting.
1434 Without function tracing:
1436 # echo 0 > options/function-trace
1437 # echo wakeup > current_tracer
1438 # echo 1 > tracing_on
1439 # echo 0 > tracing_max_latency
1440 # chrt -f 5 sleep 1
1441 # echo 0 > tracing_on
1442 # cat trace
1443 # tracer: wakeup
1444 #
1445 # wakeup latency trace v1.1.5 on 3.8.0-test+
1446 # --------------------------------------------------------------------
1447 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1448 # -----------------
1449 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1450 # -----------------
1451 #
1452 # _------=> CPU#
1453 # / _-----=> irqs-off
1454 # | / _----=> need-resched
1455 # || / _---=> hardirq/softirq
1456 # ||| / _--=> preempt-depth
1457 # |||| / delay
1458 # cmd pid ||||| time | caller
1459 # \ / ||||| \ | /
1460 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1461 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1462 <idle>-0 3d..3 15us : __schedule <-schedule
1463 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1465 The tracer only traces the highest priority task in the system
1466 to avoid tracing the normal circumstances. Here we see that
1467 the kworker with a nice priority of -20 (not very nice), took
1468 just 15 microseconds from the time it woke up, to the time it
1469 ran.
1471 Non Real-Time tasks are not that interesting. A more interesting
1472 trace is to concentrate only on Real-Time tasks.
1474 wakeup_rt
1475 ---------
1477 In a Real-Time environment it is very important to know the
1478 wakeup time it takes for the highest priority task that is woken
1479 up to the time that it executes. This is also known as "schedule
1480 latency". I stress the point that this is about RT tasks. It is
1481 also important to know the scheduling latency of non-RT tasks,
1482 but the average schedule latency is better for non-RT tasks.
1483 Tools like LatencyTop are more appropriate for such
1484 measurements.
1486 Real-Time environments are interested in the worst case latency.
1487 That is the longest latency it takes for something to happen,
1488 and not the average. We can have a very fast scheduler that may
1489 only have a large latency once in a while, but that would not
1490 work well with Real-Time tasks. The wakeup_rt tracer was designed
1491 to record the worst case wakeups of RT tasks. Non-RT tasks are
1492 not recorded because the tracer only records one worst case and
1493 tracing non-RT tasks that are unpredictable will overwrite the
1494 worst case latency of RT tasks (just run the normal wakeup
1495 tracer for a while to see that effect).
1497 Since this tracer only deals with RT tasks, we will run this
1498 slightly differently than we did with the previous tracers.
1499 Instead of performing an 'ls', we will run 'sleep 1' under
1500 'chrt' which changes the priority of the task.
1502 # echo 0 > options/function-trace
1503 # echo wakeup_rt > current_tracer
1504 # echo 1 > tracing_on
1505 # echo 0 > tracing_max_latency
1506 # chrt -f 5 sleep 1
1507 # echo 0 > tracing_on
1508 # cat trace
1509 # tracer: wakeup
1510 #
1511 # tracer: wakeup_rt
1512 #
1513 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1514 # --------------------------------------------------------------------
1515 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1516 # -----------------
1517 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1518 # -----------------
1519 #
1520 # _------=> CPU#
1521 # / _-----=> irqs-off
1522 # | / _----=> need-resched
1523 # || / _---=> hardirq/softirq
1524 # ||| / _--=> preempt-depth
1525 # |||| / delay
1526 # cmd pid ||||| time | caller
1527 # \ / ||||| \ | /
1528 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1529 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1530 <idle>-0 3d..3 5us : __schedule <-schedule
1531 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1534 Running this on an idle system, we see that it only took 5 microseconds
1535 to perform the task switch. Note, since the trace point in the schedule
1536 is before the actual "switch", we stop the tracing when the recorded task
1537 is about to schedule in. This may change if we add a new marker at the
1538 end of the scheduler.
1540 Notice that the recorded task is 'sleep' with the PID of 2389
1541 and it has an rt_prio of 5. This priority is user-space priority
1542 and not the internal kernel priority. The policy is 1 for
1543 SCHED_FIFO and 2 for SCHED_RR.
1545 Note, that the trace data shows the internal priority (99 - rtprio).
1547 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1549 The 0:120:R means idle was running with a nice priority of 0 (120 - 20)
1550 and in the running state 'R'. The sleep task was scheduled in with
1551 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1552 and it too is in the running state.
1554 Doing the same with chrt -r 5 and function-trace set.
1556 echo 1 > options/function-trace
1558 # tracer: wakeup_rt
1559 #
1560 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1561 # --------------------------------------------------------------------
1562 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1563 # -----------------
1564 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
1565 # -----------------
1566 #
1567 # _------=> CPU#
1568 # / _-----=> irqs-off
1569 # | / _----=> need-resched
1570 # || / _---=> hardirq/softirq
1571 # ||| / _--=> preempt-depth
1572 # |||| / delay
1573 # cmd pid ||||| time | caller
1574 # \ / ||||| \ | /
1575 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
1576 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1577 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
1578 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
1579 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
1580 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
1581 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
1582 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
1583 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
1584 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1585 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
1586 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
1587 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
1588 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
1589 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
1590 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
1591 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
1592 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
1593 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
1594 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
1595 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
1596 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
1597 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
1598 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
1599 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
1600 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
1601 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
1602 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
1603 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
1604 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
1605 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
1606 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
1607 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
1608 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
1609 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
1610 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
1611 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
1612 <idle>-0 3dN.1 15us : calc_load_exit_idle <-tick_nohz_idle_exit
1613 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
1614 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
1615 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
1616 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
1617 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1618 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
1619 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
1620 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
1621 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
1622 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
1623 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
1624 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
1625 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
1626 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1627 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
1628 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1629 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1630 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
1631 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
1632 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
1633 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1634 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
1635 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
1636 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
1637 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
1638 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
1639 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
1640 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
1641 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
1642 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1643 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
1644 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
1645 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
1646 <idle>-0 3.N.. 25us : schedule <-cpu_idle
1647 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
1648 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
1649 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
1650 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
1651 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
1652 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
1653 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
1654 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
1655 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
1656 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
1657 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
1658 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
1659 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
1661 This isn't that big of a trace, even with function tracing enabled,
1662 so I included the entire trace.
1664 The interrupt went off while when the system was idle. Somewhere
1665 before task_woken_rt() was called, the NEED_RESCHED flag was set,
1666 this is indicated by the first occurrence of the 'N' flag.
1668 Latency tracing and events
1669 --------------------------
1670 As function tracing can induce a much larger latency, but without
1671 seeing what happens within the latency it is hard to know what
1672 caused it. There is a middle ground, and that is with enabling
1673 events.
1675 # echo 0 > options/function-trace
1676 # echo wakeup_rt > current_tracer
1677 # echo 1 > events/enable
1678 # echo 1 > tracing_on
1679 # echo 0 > tracing_max_latency
1680 # chrt -f 5 sleep 1
1681 # echo 0 > tracing_on
1682 # cat trace
1683 # tracer: wakeup_rt
1684 #
1685 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1686 # --------------------------------------------------------------------
1687 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1688 # -----------------
1689 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
1690 # -----------------
1691 #
1692 # _------=> CPU#
1693 # / _-----=> irqs-off
1694 # | / _----=> need-resched
1695 # || / _---=> hardirq/softirq
1696 # ||| / _--=> preempt-depth
1697 # |||| / delay
1698 # cmd pid ||||| time | caller
1699 # \ / ||||| \ | /
1700 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
1701 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1702 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
1703 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
1704 <idle>-0 2.N.2 2us : power_end: cpu_id=2
1705 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
1706 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
1707 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
1708 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
1709 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
1710 <idle>-0 2d..3 6us : __schedule <-schedule
1711 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
1714 function
1715 --------
1717 This tracer is the function tracer. Enabling the function tracer
1718 can be done from the debug file system. Make sure the
1719 ftrace_enabled is set; otherwise this tracer is a nop.
1720 See the "ftrace_enabled" section below.
1722 # sysctl kernel.ftrace_enabled=1
1723 # echo function > current_tracer
1724 # echo 1 > tracing_on
1725 # usleep 1
1726 # echo 0 > tracing_on
1727 # cat trace
1728 # tracer: function
1729 #
1730 # entries-in-buffer/entries-written: 24799/24799 #P:4
1731 #
1732 # _-----=> irqs-off
1733 # / _----=> need-resched
1734 # | / _---=> hardirq/softirq
1735 # || / _--=> preempt-depth
1736 # ||| / delay
1737 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
1738 # | | | |||| | |
1739 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
1740 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
1741 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
1742 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
1743 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
1744 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
1745 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
1746 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
1747 [...]
1750 Note: function tracer uses ring buffers to store the above
1751 entries. The newest data may overwrite the oldest data.
1752 Sometimes using echo to stop the trace is not sufficient because
1753 the tracing could have overwritten the data that you wanted to
1754 record. For this reason, it is sometimes better to disable
1755 tracing directly from a program. This allows you to stop the
1756 tracing at the point that you hit the part that you are
1757 interested in. To disable the tracing directly from a C program,
1758 something like following code snippet can be used:
1760 int trace_fd;
1761 [...]
1762 int main(int argc, char *argv[]) {
1763 [...]
1764 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
1765 [...]
1766 if (condition_hit()) {
1767 write(trace_fd, "0", 1);
1768 }
1769 [...]
1770 }
1773 Single thread tracing
1774 ---------------------
1776 By writing into set_ftrace_pid you can trace a
1777 single thread. For example:
1779 # cat set_ftrace_pid
1780 no pid
1781 # echo 3111 > set_ftrace_pid
1782 # cat set_ftrace_pid
1783 3111
1784 # echo function > current_tracer
1785 # cat trace | head
1786 # tracer: function
1787 #
1788 # TASK-PID CPU# TIMESTAMP FUNCTION
1789 # | | | | |
1790 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
1791 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
1792 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
1793 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
1794 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
1795 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
1796 # echo > set_ftrace_pid
1797 # cat trace |head
1798 # tracer: function
1799 #
1800 # TASK-PID CPU# TIMESTAMP FUNCTION
1801 # | | | | |
1802 ##### CPU 3 buffer started ####
1803 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
1804 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
1805 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
1806 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
1807 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
1809 If you want to trace a function when executing, you could use
1810 something like this simple program:
1812 #include <stdio.h>
1813 #include <stdlib.h>
1814 #include <sys/types.h>
1815 #include <sys/stat.h>
1816 #include <fcntl.h>
1817 #include <unistd.h>
1818 #include <string.h>
1820 #define _STR(x) #x
1821 #define STR(x) _STR(x)
1822 #define MAX_PATH 256
1824 const char *find_debugfs(void)
1825 {
1826 static char debugfs[MAX_PATH+1];
1827 static int debugfs_found;
1828 char type[100];
1829 FILE *fp;
1831 if (debugfs_found)
1832 return debugfs;
1834 if ((fp = fopen("/proc/mounts","r")) == NULL) {
1835 perror("/proc/mounts");
1836 return NULL;
1837 }
1839 while (fscanf(fp, "%*s %"
1840 STR(MAX_PATH)
1841 "s %99s %*s %*d %*d\n",
1842 debugfs, type) == 2) {
1843 if (strcmp(type, "debugfs") == 0)
1844 break;
1845 }
1846 fclose(fp);
1848 if (strcmp(type, "debugfs") != 0) {
1849 fprintf(stderr, "debugfs not mounted");
1850 return NULL;
1851 }
1853 strcat(debugfs, "/tracing/");
1854 debugfs_found = 1;
1856 return debugfs;
1857 }
1859 const char *tracing_file(const char *file_name)
1860 {
1861 static char trace_file[MAX_PATH+1];
1862 snprintf(trace_file, MAX_PATH, "%s/%s", find_debugfs(), file_name);
1863 return trace_file;
1864 }
1866 int main (int argc, char **argv)
1867 {
1868 if (argc < 1)
1869 exit(-1);
1871 if (fork() > 0) {
1872 int fd, ffd;
1873 char line[64];
1874 int s;
1876 ffd = open(tracing_file("current_tracer"), O_WRONLY);
1877 if (ffd < 0)
1878 exit(-1);
1879 write(ffd, "nop", 3);
1881 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
1882 s = sprintf(line, "%d\n", getpid());
1883 write(fd, line, s);
1885 write(ffd, "function", 8);
1887 close(fd);
1888 close(ffd);
1890 execvp(argv[1], argv+1);
1891 }
1893 return 0;
1894 }
1896 Or this simple script!
1898 ------
1899 #!/bin/bash
1901 debugfs=`sed -ne 's/^debugfs \(.*\) debugfs.*/\1/p' /proc/mounts`
1902 echo nop > $debugfs/tracing/current_tracer
1903 echo 0 > $debugfs/tracing/tracing_on
1904 echo $$ > $debugfs/tracing/set_ftrace_pid
1905 echo function > $debugfs/tracing/current_tracer
1906 echo 1 > $debugfs/tracing/tracing_on
1907 exec "$@"
1908 ------
1911 function graph tracer
1912 ---------------------------
1914 This tracer is similar to the function tracer except that it
1915 probes a function on its entry and its exit. This is done by
1916 using a dynamically allocated stack of return addresses in each
1917 task_struct. On function entry the tracer overwrites the return
1918 address of each function traced to set a custom probe. Thus the
1919 original return address is stored on the stack of return address
1920 in the task_struct.
1922 Probing on both ends of a function leads to special features
1923 such as:
1925 - measure of a function's time execution
1926 - having a reliable call stack to draw function calls graph
1928 This tracer is useful in several situations:
1930 - you want to find the reason of a strange kernel behavior and
1931 need to see what happens in detail on any areas (or specific
1932 ones).
1934 - you are experiencing weird latencies but it's difficult to
1935 find its origin.
1937 - you want to find quickly which path is taken by a specific
1938 function
1940 - you just want to peek inside a working kernel and want to see
1941 what happens there.
1943 # tracer: function_graph
1944 #
1945 # CPU DURATION FUNCTION CALLS
1946 # | | | | | | |
1948 0) | sys_open() {
1949 0) | do_sys_open() {
1950 0) | getname() {
1951 0) | kmem_cache_alloc() {
1952 0) 1.382 us | __might_sleep();
1953 0) 2.478 us | }
1954 0) | strncpy_from_user() {
1955 0) | might_fault() {
1956 0) 1.389 us | __might_sleep();
1957 0) 2.553 us | }
1958 0) 3.807 us | }
1959 0) 7.876 us | }
1960 0) | alloc_fd() {
1961 0) 0.668 us | _spin_lock();
1962 0) 0.570 us | expand_files();
1963 0) 0.586 us | _spin_unlock();
1966 There are several columns that can be dynamically
1967 enabled/disabled. You can use every combination of options you
1968 want, depending on your needs.
1970 - The cpu number on which the function executed is default
1971 enabled. It is sometimes better to only trace one cpu (see
1972 tracing_cpu_mask file) or you might sometimes see unordered
1973 function calls while cpu tracing switch.
1975 hide: echo nofuncgraph-cpu > trace_options
1976 show: echo funcgraph-cpu > trace_options
1978 - The duration (function's time of execution) is displayed on
1979 the closing bracket line of a function or on the same line
1980 than the current function in case of a leaf one. It is default
1981 enabled.
1983 hide: echo nofuncgraph-duration > trace_options
1984 show: echo funcgraph-duration > trace_options
1986 - The overhead field precedes the duration field in case of
1987 reached duration thresholds.
1989 hide: echo nofuncgraph-overhead > trace_options
1990 show: echo funcgraph-overhead > trace_options
1991 depends on: funcgraph-duration
1993 ie:
1995 3) # 1837.709 us | } /* __switch_to */
1996 3) | finish_task_switch() {
1997 3) 0.313 us | _raw_spin_unlock_irq();
1998 3) 3.177 us | }
1999 3) # 1889.063 us | } /* __schedule */
2000 3) ! 140.417 us | } /* __schedule */
2001 3) # 2034.948 us | } /* schedule */
2002 3) * 33998.59 us | } /* schedule_preempt_disabled */
2004 [...]
2006 1) 0.260 us | msecs_to_jiffies();
2007 1) 0.313 us | __rcu_read_unlock();
2008 1) + 61.770 us | }
2009 1) + 64.479 us | }
2010 1) 0.313 us | rcu_bh_qs();
2011 1) 0.313 us | __local_bh_enable();
2012 1) ! 217.240 us | }
2013 1) 0.365 us | idle_cpu();
2014 1) | rcu_irq_exit() {
2015 1) 0.417 us | rcu_eqs_enter_common.isra.47();
2016 1) 3.125 us | }
2017 1) ! 227.812 us | }
2018 1) ! 457.395 us | }
2019 1) @ 119760.2 us | }
2021 [...]
2023 2) | handle_IPI() {
2024 1) 6.979 us | }
2025 2) 0.417 us | scheduler_ipi();
2026 1) 9.791 us | }
2027 1) + 12.917 us | }
2028 2) 3.490 us | }
2029 1) + 15.729 us | }
2030 1) + 18.542 us | }
2031 2) $ 3594274 us | }
2033 + means that the function exceeded 10 usecs.
2034 ! means that the function exceeded 100 usecs.
2035 # means that the function exceeded 1000 usecs.
2036 * means that the function exceeded 10 msecs.
2037 @ means that the function exceeded 100 msecs.
2038 $ means that the function exceeded 1 sec.
2041 - The task/pid field displays the thread cmdline and pid which
2042 executed the function. It is default disabled.
2044 hide: echo nofuncgraph-proc > trace_options
2045 show: echo funcgraph-proc > trace_options
2047 ie:
2049 # tracer: function_graph
2050 #
2051 # CPU TASK/PID DURATION FUNCTION CALLS
2052 # | | | | | | | | |
2053 0) sh-4802 | | d_free() {
2054 0) sh-4802 | | call_rcu() {
2055 0) sh-4802 | | __call_rcu() {
2056 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2057 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2058 0) sh-4802 | 2.899 us | }
2059 0) sh-4802 | 4.040 us | }
2060 0) sh-4802 | 5.151 us | }
2061 0) sh-4802 | + 49.370 us | }
2064 - The absolute time field is an absolute timestamp given by the
2065 system clock since it started. A snapshot of this time is
2066 given on each entry/exit of functions
2068 hide: echo nofuncgraph-abstime > trace_options
2069 show: echo funcgraph-abstime > trace_options
2071 ie:
2073 #
2074 # TIME CPU DURATION FUNCTION CALLS
2075 # | | | | | | | |
2076 360.774522 | 1) 0.541 us | }
2077 360.774522 | 1) 4.663 us | }
2078 360.774523 | 1) 0.541 us | __wake_up_bit();
2079 360.774524 | 1) 6.796 us | }
2080 360.774524 | 1) 7.952 us | }
2081 360.774525 | 1) 9.063 us | }
2082 360.774525 | 1) 0.615 us | journal_mark_dirty();
2083 360.774527 | 1) 0.578 us | __brelse();
2084 360.774528 | 1) | reiserfs_prepare_for_journal() {
2085 360.774528 | 1) | unlock_buffer() {
2086 360.774529 | 1) | wake_up_bit() {
2087 360.774529 | 1) | bit_waitqueue() {
2088 360.774530 | 1) 0.594 us | __phys_addr();
2091 The function name is always displayed after the closing bracket
2092 for a function if the start of that function is not in the
2093 trace buffer.
2095 Display of the function name after the closing bracket may be
2096 enabled for functions whose start is in the trace buffer,
2097 allowing easier searching with grep for function durations.
2098 It is default disabled.
2100 hide: echo nofuncgraph-tail > trace_options
2101 show: echo funcgraph-tail > trace_options
2103 Example with nofuncgraph-tail (default):
2104 0) | putname() {
2105 0) | kmem_cache_free() {
2106 0) 0.518 us | __phys_addr();
2107 0) 1.757 us | }
2108 0) 2.861 us | }
2110 Example with funcgraph-tail:
2111 0) | putname() {
2112 0) | kmem_cache_free() {
2113 0) 0.518 us | __phys_addr();
2114 0) 1.757 us | } /* kmem_cache_free() */
2115 0) 2.861 us | } /* putname() */
2117 You can put some comments on specific functions by using
2118 trace_printk() For example, if you want to put a comment inside
2119 the __might_sleep() function, you just have to include
2120 <linux/ftrace.h> and call trace_printk() inside __might_sleep()
2122 trace_printk("I'm a comment!\n")
2124 will produce:
2126 1) | __might_sleep() {
2127 1) | /* I'm a comment! */
2128 1) 1.449 us | }
2131 You might find other useful features for this tracer in the
2132 following "dynamic ftrace" section such as tracing only specific
2133 functions or tasks.
2135 dynamic ftrace
2136 --------------
2138 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2139 virtually no overhead when function tracing is disabled. The way
2140 this works is the mcount function call (placed at the start of
2141 every kernel function, produced by the -pg switch in gcc),
2142 starts of pointing to a simple return. (Enabling FTRACE will
2143 include the -pg switch in the compiling of the kernel.)
2145 At compile time every C file object is run through the
2146 recordmcount program (located in the scripts directory). This
2147 program will parse the ELF headers in the C object to find all
2148 the locations in the .text section that call mcount. (Note, only
2149 white listed .text sections are processed, since processing other
2150 sections like .init.text may cause races due to those sections
2151 being freed unexpectedly).
2153 A new section called "__mcount_loc" is created that holds
2154 references to all the mcount call sites in the .text section.
2155 The recordmcount program re-links this section back into the
2156 original object. The final linking stage of the kernel will add all these
2157 references into a single table.
2159 On boot up, before SMP is initialized, the dynamic ftrace code
2160 scans this table and updates all the locations into nops. It
2161 also records the locations, which are added to the
2162 available_filter_functions list. Modules are processed as they
2163 are loaded and before they are executed. When a module is
2164 unloaded, it also removes its functions from the ftrace function
2165 list. This is automatic in the module unload code, and the
2166 module author does not need to worry about it.
2168 When tracing is enabled, the process of modifying the function
2169 tracepoints is dependent on architecture. The old method is to use
2170 kstop_machine to prevent races with the CPUs executing code being
2171 modified (which can cause the CPU to do undesirable things, especially
2172 if the modified code crosses cache (or page) boundaries), and the nops are
2173 patched back to calls. But this time, they do not call mcount
2174 (which is just a function stub). They now call into the ftrace
2175 infrastructure.
2177 The new method of modifying the function tracepoints is to place
2178 a breakpoint at the location to be modified, sync all CPUs, modify
2179 the rest of the instruction not covered by the breakpoint. Sync
2180 all CPUs again, and then remove the breakpoint with the finished
2181 version to the ftrace call site.
2183 Some archs do not even need to monkey around with the synchronization,
2184 and can just slap the new code on top of the old without any
2185 problems with other CPUs executing it at the same time.
2187 One special side-effect to the recording of the functions being
2188 traced is that we can now selectively choose which functions we
2189 wish to trace and which ones we want the mcount calls to remain
2190 as nops.
2192 Two files are used, one for enabling and one for disabling the
2193 tracing of specified functions. They are:
2195 set_ftrace_filter
2197 and
2199 set_ftrace_notrace
2201 A list of available functions that you can add to these files is
2202 listed in:
2204 available_filter_functions
2206 # cat available_filter_functions
2207 put_prev_task_idle
2208 kmem_cache_create
2209 pick_next_task_rt
2210 get_online_cpus
2211 pick_next_task_fair
2212 mutex_lock
2213 [...]
2215 If I am only interested in sys_nanosleep and hrtimer_interrupt:
2217 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2218 # echo function > current_tracer
2219 # echo 1 > tracing_on
2220 # usleep 1
2221 # echo 0 > tracing_on
2222 # cat trace
2223 # tracer: function
2224 #
2225 # entries-in-buffer/entries-written: 5/5 #P:4
2226 #
2227 # _-----=> irqs-off
2228 # / _----=> need-resched
2229 # | / _---=> hardirq/softirq
2230 # || / _--=> preempt-depth
2231 # ||| / delay
2232 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2233 # | | | |||| | |
2234 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2235 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2236 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2237 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2238 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2240 To see which functions are being traced, you can cat the file:
2242 # cat set_ftrace_filter
2243 hrtimer_interrupt
2244 sys_nanosleep
2247 Perhaps this is not enough. The filters also allow glob(7) matching.
2249 <match>* - will match functions that begin with <match>
2250 *<match> - will match functions that end with <match>
2251 *<match>* - will match functions that have <match> in it
2252 <match1>*<match2> - will match functions that begin with
2253 <match1> and end with <match2>
2255 Note: It is better to use quotes to enclose the wild cards,
2256 otherwise the shell may expand the parameters into names
2257 of files in the local directory.
2259 # echo 'hrtimer_*' > set_ftrace_filter
2261 Produces:
2263 # tracer: function
2264 #
2265 # entries-in-buffer/entries-written: 897/897 #P:4
2266 #
2267 # _-----=> irqs-off
2268 # / _----=> need-resched
2269 # | / _---=> hardirq/softirq
2270 # || / _--=> preempt-depth
2271 # ||| / delay
2272 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2273 # | | | |||| | |
2274 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2275 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2276 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2277 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2278 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2279 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2280 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2281 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2283 Notice that we lost the sys_nanosleep.
2285 # cat set_ftrace_filter
2286 hrtimer_run_queues
2287 hrtimer_run_pending
2288 hrtimer_init
2289 hrtimer_cancel
2290 hrtimer_try_to_cancel
2291 hrtimer_forward
2292 hrtimer_start
2293 hrtimer_reprogram
2294 hrtimer_force_reprogram
2295 hrtimer_get_next_event
2296 hrtimer_interrupt
2297 hrtimer_nanosleep
2298 hrtimer_wakeup
2299 hrtimer_get_remaining
2300 hrtimer_get_res
2301 hrtimer_init_sleeper
2304 This is because the '>' and '>>' act just like they do in bash.
2305 To rewrite the filters, use '>'
2306 To append to the filters, use '>>'
2308 To clear out a filter so that all functions will be recorded
2309 again:
2311 # echo > set_ftrace_filter
2312 # cat set_ftrace_filter
2313 #
2315 Again, now we want to append.
2317 # echo sys_nanosleep > set_ftrace_filter
2318 # cat set_ftrace_filter
2319 sys_nanosleep
2320 # echo 'hrtimer_*' >> set_ftrace_filter
2321 # cat set_ftrace_filter
2322 hrtimer_run_queues
2323 hrtimer_run_pending
2324 hrtimer_init
2325 hrtimer_cancel
2326 hrtimer_try_to_cancel
2327 hrtimer_forward
2328 hrtimer_start
2329 hrtimer_reprogram
2330 hrtimer_force_reprogram
2331 hrtimer_get_next_event
2332 hrtimer_interrupt
2333 sys_nanosleep
2334 hrtimer_nanosleep
2335 hrtimer_wakeup
2336 hrtimer_get_remaining
2337 hrtimer_get_res
2338 hrtimer_init_sleeper
2341 The set_ftrace_notrace prevents those functions from being
2342 traced.
2344 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2346 Produces:
2348 # tracer: function
2349 #
2350 # entries-in-buffer/entries-written: 39608/39608 #P:4
2351 #
2352 # _-----=> irqs-off
2353 # / _----=> need-resched
2354 # | / _---=> hardirq/softirq
2355 # || / _--=> preempt-depth
2356 # ||| / delay
2357 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2358 # | | | |||| | |
2359 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2360 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2361 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2362 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2363 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2364 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2365 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2366 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2367 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2368 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2369 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2370 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2372 We can see that there's no more lock or preempt tracing.
2375 Dynamic ftrace with the function graph tracer
2376 ---------------------------------------------
2378 Although what has been explained above concerns both the
2379 function tracer and the function-graph-tracer, there are some
2380 special features only available in the function-graph tracer.
2382 If you want to trace only one function and all of its children,
2383 you just have to echo its name into set_graph_function:
2385 echo __do_fault > set_graph_function
2387 will produce the following "expanded" trace of the __do_fault()
2388 function:
2390 0) | __do_fault() {
2391 0) | filemap_fault() {
2392 0) | find_lock_page() {
2393 0) 0.804 us | find_get_page();
2394 0) | __might_sleep() {
2395 0) 1.329 us | }
2396 0) 3.904 us | }
2397 0) 4.979 us | }
2398 0) 0.653 us | _spin_lock();
2399 0) 0.578 us | page_add_file_rmap();
2400 0) 0.525 us | native_set_pte_at();
2401 0) 0.585 us | _spin_unlock();
2402 0) | unlock_page() {
2403 0) 0.541 us | page_waitqueue();
2404 0) 0.639 us | __wake_up_bit();
2405 0) 2.786 us | }
2406 0) + 14.237 us | }
2407 0) | __do_fault() {
2408 0) | filemap_fault() {
2409 0) | find_lock_page() {
2410 0) 0.698 us | find_get_page();
2411 0) | __might_sleep() {
2412 0) 1.412 us | }
2413 0) 3.950 us | }
2414 0) 5.098 us | }
2415 0) 0.631 us | _spin_lock();
2416 0) 0.571 us | page_add_file_rmap();
2417 0) 0.526 us | native_set_pte_at();
2418 0) 0.586 us | _spin_unlock();
2419 0) | unlock_page() {
2420 0) 0.533 us | page_waitqueue();
2421 0) 0.638 us | __wake_up_bit();
2422 0) 2.793 us | }
2423 0) + 14.012 us | }
2425 You can also expand several functions at once:
2427 echo sys_open > set_graph_function
2428 echo sys_close >> set_graph_function
2430 Now if you want to go back to trace all functions you can clear
2431 this special filter via:
2433 echo > set_graph_function
2436 ftrace_enabled
2437 --------------
2439 Note, the proc sysctl ftrace_enable is a big on/off switch for the
2440 function tracer. By default it is enabled (when function tracing is
2441 enabled in the kernel). If it is disabled, all function tracing is
2442 disabled. This includes not only the function tracers for ftrace, but
2443 also for any other uses (perf, kprobes, stack tracing, profiling, etc).
2445 Please disable this with care.
2447 This can be disable (and enabled) with:
2449 sysctl kernel.ftrace_enabled=0
2450 sysctl kernel.ftrace_enabled=1
2452 or
2454 echo 0 > /proc/sys/kernel/ftrace_enabled
2455 echo 1 > /proc/sys/kernel/ftrace_enabled
2458 Filter commands
2459 ---------------
2461 A few commands are supported by the set_ftrace_filter interface.
2462 Trace commands have the following format:
2464 <function>:<command>:<parameter>
2466 The following commands are supported:
2468 - mod
2469 This command enables function filtering per module. The
2470 parameter defines the module. For example, if only the write*
2471 functions in the ext3 module are desired, run:
2473 echo 'write*:mod:ext3' > set_ftrace_filter
2475 This command interacts with the filter in the same way as
2476 filtering based on function names. Thus, adding more functions
2477 in a different module is accomplished by appending (>>) to the
2478 filter file. Remove specific module functions by prepending
2479 '!':
2481 echo '!writeback*:mod:ext3' >> set_ftrace_filter
2483 Mod command supports module globbing. Disable tracing for all
2484 functions except a specific module:
2486 echo '!*:mod:!ext3' >> set_ftrace_filter
2488 Disable tracing for all modules, but still trace kernel:
2490 echo '!*:mod:*' >> set_ftrace_filter
2492 Enable filter only for kernel:
2494 echo '*write*:mod:!*' >> set_ftrace_filter
2496 Enable filter for module globbing:
2498 echo '*write*:mod:*snd*' >> set_ftrace_filter
2500 - traceon/traceoff
2501 These commands turn tracing on and off when the specified
2502 functions are hit. The parameter determines how many times the
2503 tracing system is turned on and off. If unspecified, there is
2504 no limit. For example, to disable tracing when a schedule bug
2505 is hit the first 5 times, run:
2507 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
2509 To always disable tracing when __schedule_bug is hit:
2511 echo '__schedule_bug:traceoff' > set_ftrace_filter
2513 These commands are cumulative whether or not they are appended
2514 to set_ftrace_filter. To remove a command, prepend it by '!'
2515 and drop the parameter:
2517 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
2519 The above removes the traceoff command for __schedule_bug
2520 that have a counter. To remove commands without counters:
2522 echo '!__schedule_bug:traceoff' > set_ftrace_filter
2524 - snapshot
2525 Will cause a snapshot to be triggered when the function is hit.
2527 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
2529 To only snapshot once:
2531 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
2533 To remove the above commands:
2535 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
2536 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
2538 - enable_event/disable_event
2539 These commands can enable or disable a trace event. Note, because
2540 function tracing callbacks are very sensitive, when these commands
2541 are registered, the trace point is activated, but disabled in
2542 a "soft" mode. That is, the tracepoint will be called, but
2543 just will not be traced. The event tracepoint stays in this mode
2544 as long as there's a command that triggers it.
2546 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
2547 set_ftrace_filter
2549 The format is:
2551 <function>:enable_event:<system>:<event>[:count]
2552 <function>:disable_event:<system>:<event>[:count]
2554 To remove the events commands:
2557 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
2558 set_ftrace_filter
2559 echo '!schedule:disable_event:sched:sched_switch' > \
2560 set_ftrace_filter
2562 - dump
2563 When the function is hit, it will dump the contents of the ftrace
2564 ring buffer to the console. This is useful if you need to debug
2565 something, and want to dump the trace when a certain function
2566 is hit. Perhaps its a function that is called before a tripple
2567 fault happens and does not allow you to get a regular dump.
2569 - cpudump
2570 When the function is hit, it will dump the contents of the ftrace
2571 ring buffer for the current CPU to the console. Unlike the "dump"
2572 command, it only prints out the contents of the ring buffer for the
2573 CPU that executed the function that triggered the dump.
2575 trace_pipe
2576 ----------
2578 The trace_pipe outputs the same content as the trace file, but
2579 the effect on the tracing is different. Every read from
2580 trace_pipe is consumed. This means that subsequent reads will be
2581 different. The trace is live.
2583 # echo function > current_tracer
2584 # cat trace_pipe > /tmp/trace.out &
2585 [1] 4153
2586 # echo 1 > tracing_on
2587 # usleep 1
2588 # echo 0 > tracing_on
2589 # cat trace
2590 # tracer: function
2591 #
2592 # entries-in-buffer/entries-written: 0/0 #P:4
2593 #
2594 # _-----=> irqs-off
2595 # / _----=> need-resched
2596 # | / _---=> hardirq/softirq
2597 # || / _--=> preempt-depth
2598 # ||| / delay
2599 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2600 # | | | |||| | |
2602 #
2603 # cat /tmp/trace.out
2604 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
2605 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
2606 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
2607 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
2608 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
2609 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
2610 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
2611 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
2612 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
2615 Note, reading the trace_pipe file will block until more input is
2616 added.
2618 trace entries
2619 -------------
2621 Having too much or not enough data can be troublesome in
2622 diagnosing an issue in the kernel. The file buffer_size_kb is
2623 used to modify the size of the internal trace buffers. The
2624 number listed is the number of entries that can be recorded per
2625 CPU. To know the full size, multiply the number of possible CPUs
2626 with the number of entries.
2628 # cat buffer_size_kb
2629 1408 (units kilobytes)
2631 Or simply read buffer_total_size_kb
2633 # cat buffer_total_size_kb
2634 5632
2636 To modify the buffer, simple echo in a number (in 1024 byte segments).
2638 # echo 10000 > buffer_size_kb
2639 # cat buffer_size_kb
2640 10000 (units kilobytes)
2642 It will try to allocate as much as possible. If you allocate too
2643 much, it can cause Out-Of-Memory to trigger.
2645 # echo 1000000000000 > buffer_size_kb
2646 -bash: echo: write error: Cannot allocate memory
2647 # cat buffer_size_kb
2648 85
2650 The per_cpu buffers can be changed individually as well:
2652 # echo 10000 > per_cpu/cpu0/buffer_size_kb
2653 # echo 100 > per_cpu/cpu1/buffer_size_kb
2655 When the per_cpu buffers are not the same, the buffer_size_kb
2656 at the top level will just show an X
2658 # cat buffer_size_kb
2659 X
2661 This is where the buffer_total_size_kb is useful:
2663 # cat buffer_total_size_kb
2664 12916
2666 Writing to the top level buffer_size_kb will reset all the buffers
2667 to be the same again.
2669 Snapshot
2670 --------
2671 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
2672 available to all non latency tracers. (Latency tracers which
2673 record max latency, such as "irqsoff" or "wakeup", can't use
2674 this feature, since those are already using the snapshot
2675 mechanism internally.)
2677 Snapshot preserves a current trace buffer at a particular point
2678 in time without stopping tracing. Ftrace swaps the current
2679 buffer with a spare buffer, and tracing continues in the new
2680 current (=previous spare) buffer.
2682 The following debugfs files in "tracing" are related to this
2683 feature:
2685 snapshot:
2687 This is used to take a snapshot and to read the output
2688 of the snapshot. Echo 1 into this file to allocate a
2689 spare buffer and to take a snapshot (swap), then read
2690 the snapshot from this file in the same format as
2691 "trace" (described above in the section "The File
2692 System"). Both reads snapshot and tracing are executable
2693 in parallel. When the spare buffer is allocated, echoing
2694 0 frees it, and echoing else (positive) values clear the
2695 snapshot contents.
2696 More details are shown in the table below.
2698 status\input | 0 | 1 | else |
2699 --------------+------------+------------+------------+
2700 not allocated |(do nothing)| alloc+swap |(do nothing)|
2701 --------------+------------+------------+------------+
2702 allocated | free | swap | clear |
2703 --------------+------------+------------+------------+
2705 Here is an example of using the snapshot feature.
2707 # echo 1 > events/sched/enable
2708 # echo 1 > snapshot
2709 # cat snapshot
2710 # tracer: nop
2711 #
2712 # entries-in-buffer/entries-written: 71/71 #P:8
2713 #
2714 # _-----=> irqs-off
2715 # / _----=> need-resched
2716 # | / _---=> hardirq/softirq
2717 # || / _--=> preempt-depth
2718 # ||| / delay
2719 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2720 # | | | |||| | |
2721 <idle>-0 [005] d... 2440.603828: sched_switch: prev_comm=swapper/5 prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=snapshot-test-2 next_pid=2242 next_prio=120
2722 sleep-2242 [005] d... 2440.603846: sched_switch: prev_comm=snapshot-test-2 prev_pid=2242 prev_prio=120 prev_state=R ==> next_comm=kworker/5:1 next_pid=60 next_prio=120
2723 [...]
2724 <idle>-0 [002] d... 2440.707230: sched_switch: prev_comm=swapper/2 prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=snapshot-test-2 next_pid=2229 next_prio=120
2726 # cat trace
2727 # tracer: nop
2728 #
2729 # entries-in-buffer/entries-written: 77/77 #P:8
2730 #
2731 # _-----=> irqs-off
2732 # / _----=> need-resched
2733 # | / _---=> hardirq/softirq
2734 # || / _--=> preempt-depth
2735 # ||| / delay
2736 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2737 # | | | |||| | |
2738 <idle>-0 [007] d... 2440.707395: sched_switch: prev_comm=swapper/7 prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=snapshot-test-2 next_pid=2243 next_prio=120
2739 snapshot-test-2-2229 [002] d... 2440.707438: sched_switch: prev_comm=snapshot-test-2 prev_pid=2229 prev_prio=120 prev_state=S ==> next_comm=swapper/2 next_pid=0 next_prio=120
2740 [...]
2743 If you try to use this snapshot feature when current tracer is
2744 one of the latency tracers, you will get the following results.
2746 # echo wakeup > current_tracer
2747 # echo 1 > snapshot
2748 bash: echo: write error: Device or resource busy
2749 # cat snapshot
2750 cat: snapshot: Device or resource busy
2753 Instances
2754 ---------
2755 In the debugfs tracing directory is a directory called "instances".
2756 This directory can have new directories created inside of it using
2757 mkdir, and removing directories with rmdir. The directory created
2758 with mkdir in this directory will already contain files and other
2759 directories after it is created.
2761 # mkdir instances/foo
2762 # ls instances/foo
2763 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
2764 set_event snapshot trace trace_clock trace_marker trace_options
2765 trace_pipe tracing_on
2767 As you can see, the new directory looks similar to the tracing directory
2768 itself. In fact, it is very similar, except that the buffer and
2769 events are agnostic from the main director, or from any other
2770 instances that are created.
2772 The files in the new directory work just like the files with the
2773 same name in the tracing directory except the buffer that is used
2774 is a separate and new buffer. The files affect that buffer but do not
2775 affect the main buffer with the exception of trace_options. Currently,
2776 the trace_options affect all instances and the top level buffer
2777 the same, but this may change in future releases. That is, options
2778 may become specific to the instance they reside in.
2780 Notice that none of the function tracer files are there, nor is
2781 current_tracer and available_tracers. This is because the buffers
2782 can currently only have events enabled for them.
2784 # mkdir instances/foo
2785 # mkdir instances/bar
2786 # mkdir instances/zoot
2787 # echo 100000 > buffer_size_kb
2788 # echo 1000 > instances/foo/buffer_size_kb
2789 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
2790 # echo function > current_trace
2791 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
2792 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
2793 # echo 1 > instances/foo/events/sched/sched_switch/enable
2794 # echo 1 > instances/bar/events/irq/enable
2795 # echo 1 > instances/zoot/events/syscalls/enable
2796 # cat trace_pipe
2797 CPU:2 [LOST 11745 EVENTS]
2798 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
2799 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
2800 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
2801 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
2802 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
2803 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
2804 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
2805 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
2806 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
2807 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
2808 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
2809 [...]
2811 # cat instances/foo/trace_pipe
2812 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
2813 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
2814 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
2815 <idle>-0 [003] d..3 136.676909: sched_switch: prev_comm=swapper/3 prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=rcu_preempt next_pid=9 next_prio=120
2816 rcu_preempt-9 [003] d..3 136.676916: sched_switch: prev_comm=rcu_preempt prev_pid=9 prev_prio=120 prev_state=S ==> next_comm=swapper/3 next_pid=0 next_prio=120
2817 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
2818 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
2819 bash-1998 [000] d..3 136.677018: sched_switch: prev_comm=bash prev_pid=1998 prev_prio=120 prev_state=R+ ==> next_comm=kworker/0:1 next_pid=59 next_prio=120
2820 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
2821 kworker/0:1-59 [000] d..3 136.677025: sched_switch: prev_comm=kworker/0:1 prev_pid=59 prev_prio=120 prev_state=S ==> next_comm=bash next_pid=1998 next_prio=120
2822 [...]
2824 # cat instances/bar/trace_pipe
2825 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
2826 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
2827 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
2828 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
2829 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
2830 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
2831 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
2832 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
2833 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
2834 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
2835 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
2836 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
2837 [...]
2839 # cat instances/zoot/trace
2840 # tracer: nop
2841 #
2842 # entries-in-buffer/entries-written: 18996/18996 #P:4
2843 #
2844 # _-----=> irqs-off
2845 # / _----=> need-resched
2846 # | / _---=> hardirq/softirq
2847 # || / _--=> preempt-depth
2848 # ||| / delay
2849 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2850 # | | | |||| | |
2851 bash-1998 [000] d... 140.733501: sys_write -> 0x2
2852 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
2853 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
2854 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
2855 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
2856 bash-1998 [000] d... 140.733510: sys_close(fd: a)
2857 bash-1998 [000] d... 140.733510: sys_close -> 0x0
2858 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
2859 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
2860 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
2861 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
2863 You can see that the trace of the top most trace buffer shows only
2864 the function tracing. The foo instance displays wakeups and task
2865 switches.
2867 To remove the instances, simply delete their directories:
2869 # rmdir instances/foo
2870 # rmdir instances/bar
2871 # rmdir instances/zoot
2873 Note, if a process has a trace file open in one of the instance
2874 directories, the rmdir will fail with EBUSY.
2877 Stack trace
2878 -----------
2879 Since the kernel has a fixed sized stack, it is important not to
2880 waste it in functions. A kernel developer must be conscience of
2881 what they allocate on the stack. If they add too much, the system
2882 can be in danger of a stack overflow, and corruption will occur,
2883 usually leading to a system panic.
2885 There are some tools that check this, usually with interrupts
2886 periodically checking usage. But if you can perform a check
2887 at every function call that will become very useful. As ftrace provides
2888 a function tracer, it makes it convenient to check the stack size
2889 at every function call. This is enabled via the stack tracer.
2891 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
2892 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
2894 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
2896 You can also enable it from the kernel command line to trace
2897 the stack size of the kernel during boot up, by adding "stacktrace"
2898 to the kernel command line parameter.
2900 After running it for a few minutes, the output looks like:
2902 # cat stack_max_size
2903 2928
2905 # cat stack_trace
2906 Depth Size Location (18 entries)
2907 ----- ---- --------
2908 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
2909 1) 2704 160 find_busiest_group+0x31/0x1f1
2910 2) 2544 256 load_balance+0xd9/0x662
2911 3) 2288 80 idle_balance+0xbb/0x130
2912 4) 2208 128 __schedule+0x26e/0x5b9
2913 5) 2080 16 schedule+0x64/0x66
2914 6) 2064 128 schedule_timeout+0x34/0xe0
2915 7) 1936 112 wait_for_common+0x97/0xf1
2916 8) 1824 16 wait_for_completion+0x1d/0x1f
2917 9) 1808 128 flush_work+0xfe/0x119
2918 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
2919 11) 1664 48 input_available_p+0x1d/0x5c
2920 12) 1616 48 n_tty_poll+0x6d/0x134
2921 13) 1568 64 tty_poll+0x64/0x7f
2922 14) 1504 880 do_select+0x31e/0x511
2923 15) 624 400 core_sys_select+0x177/0x216
2924 16) 224 96 sys_select+0x91/0xb9
2925 17) 128 128 system_call_fastpath+0x16/0x1b
2927 Note, if -mfentry is being used by gcc, functions get traced before
2928 they set up the stack frame. This means that leaf level functions
2929 are not tested by the stack tracer when -mfentry is used.
2931 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
2933 ---------
2935 More details can be found in the source code, in the
2936 kernel/trace/*.c files.