1 ========================
2 ftrace - Function Tracer
3 ========================
5 Copyright 2008 Red Hat Inc.
7 :Author: Steven Rostedt <srostedt@redhat.com>
8 :License: The GNU Free Documentation License, Version 1.2
9 (dual licensed under the GPL v2)
10 :Original Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton,
11 John Kacur, and David Teigland.
13 - Written for: 2.6.28-rc2
15 - Updated for: 4.13 - Copyright 2017 VMware Inc. Steven Rostedt
16 - Converted to rst format - Changbin Du <changbin.du@intel.com>
21 Ftrace is an internal tracer designed to help out developers and
22 designers of systems to find what is going on inside the kernel.
23 It can be used for debugging or analyzing latencies and
24 performance issues that take place outside of user-space.
26 Although ftrace is typically considered the function tracer, it
27 is really a framework of several assorted tracing utilities.
28 There's latency tracing to examine what occurs between interrupts
29 disabled and enabled, as well as for preemption and from a time
30 a task is woken to the task is actually scheduled in.
32 One of the most common uses of ftrace is the event tracing.
33 Throughout the kernel is hundreds of static event points that
34 can be enabled via the tracefs file system to see what is
35 going on in certain parts of the kernel.
37 See events.rst for more information.
40 Implementation Details
41 ----------------------
43 See Documentation/trace/ftrace-design.rst for details for arch porters and such.
49 Ftrace uses the tracefs file system to hold the control files as
50 well as the files to display output.
52 When tracefs is configured into the kernel (which selecting any ftrace
53 option will do) the directory /sys/kernel/tracing will be created. To mount
54 this directory, you can add to your /etc/fstab file::
56 tracefs /sys/kernel/tracing tracefs defaults 0 0
58 Or you can mount it at run time with::
60 mount -t tracefs nodev /sys/kernel/tracing
62 For quicker access to that directory you may want to make a soft link to
65 ln -s /sys/kernel/tracing /tracing
69 Before 4.1, all ftrace tracing control files were within the debugfs
70 file system, which is typically located at /sys/kernel/debug/tracing.
71 For backward compatibility, when mounting the debugfs file system,
72 the tracefs file system will be automatically mounted at:
74 /sys/kernel/debug/tracing
76 All files located in the tracefs file system will be located in that
77 debugfs file system directory as well.
81 Any selected ftrace option will also create the tracefs file system.
82 The rest of the document will assume that you are in the ftrace directory
83 (cd /sys/kernel/tracing) and will only concentrate on the files within that
84 directory and not distract from the content with the extended
85 "/sys/kernel/tracing" path name.
87 That's it! (assuming that you have ftrace configured into your kernel)
89 After mounting tracefs you will have access to the control and output files
90 of ftrace. Here is a list of some of the key files:
93 Note: all time values are in microseconds.
97 This is used to set or display the current tracer
98 that is configured. Changing the current tracer clears
99 the ring buffer content as well as the "snapshot" buffer.
103 This holds the different types of tracers that
104 have been compiled into the kernel. The
105 tracers listed here can be configured by
106 echoing their name into current_tracer.
110 This sets or displays whether writing to the trace
111 ring buffer is enabled. Echo 0 into this file to disable
112 the tracer or 1 to enable it. Note, this only disables
113 writing to the ring buffer, the tracing overhead may
116 The kernel function tracing_off() can be used within the
117 kernel to disable writing to the ring buffer, which will
118 set this file to "0". User space can re-enable tracing by
119 echoing "1" into the file.
121 Note, the function and event trigger "traceoff" will also
122 set this file to zero and stop tracing. Which can also
123 be re-enabled by user space using this file.
127 This file holds the output of the trace in a human
128 readable format (described below). Opening this file for
129 writing with the O_TRUNC flag clears the ring buffer content.
130 Note, this file is not a consumer. If tracing is off
131 (no tracer running, or tracing_on is zero), it will produce
132 the same output each time it is read. When tracing is on,
133 it may produce inconsistent results as it tries to read
134 the entire buffer without consuming it.
138 The output is the same as the "trace" file but this
139 file is meant to be streamed with live tracing.
140 Reads from this file will block until new data is
141 retrieved. Unlike the "trace" file, this file is a
142 consumer. This means reading from this file causes
143 sequential reads to display more current data. Once
144 data is read from this file, it is consumed, and
145 will not be read again with a sequential read. The
146 "trace" file is static, and if the tracer is not
147 adding more data, it will display the same
148 information every time it is read.
152 This file lets the user control the amount of data
153 that is displayed in one of the above output
154 files. Options also exist to modify how a tracer
155 or events work (stack traces, timestamps, etc).
159 This is a directory that has a file for every available
160 trace option (also in trace_options). Options may also be set
161 or cleared by writing a "1" or "0" respectively into the
162 corresponding file with the option name.
166 Some of the tracers record the max latency.
167 For example, the maximum time that interrupts are disabled.
168 The maximum time is saved in this file. The max trace will also be
169 stored, and displayed by "trace". A new max trace will only be
170 recorded if the latency is greater than the value in this file
173 By echoing in a time into this file, no latency will be recorded
174 unless it is greater than the time in this file.
178 Some latency tracers will record a trace whenever the
179 latency is greater than the number in this file.
180 Only active when the file contains a number greater than 0.
185 This sets or displays the number of kilobytes each CPU
186 buffer holds. By default, the trace buffers are the same size
187 for each CPU. The displayed number is the size of the
188 CPU buffer and not total size of all buffers. The
189 trace buffers are allocated in pages (blocks of memory
190 that the kernel uses for allocation, usually 4 KB in size).
191 A few extra pages may be allocated to accommodate buffer management
192 meta-data. If the last page allocated has room for more bytes
193 than requested, the rest of the page will be used,
194 making the actual allocation bigger than requested or shown.
195 ( Note, the size may not be a multiple of the page size
196 due to buffer management meta-data. )
198 Buffer sizes for individual CPUs may vary
199 (see "per_cpu/cpu0/buffer_size_kb" below), and if they do
200 this file will show "X".
202 buffer_total_size_kb:
204 This displays the total combined size of all the trace buffers.
208 If a process is performing tracing, and the ring buffer should be
209 shrunk "freed" when the process is finished, even if it were to be
210 killed by a signal, this file can be used for that purpose. On close
211 of this file, the ring buffer will be resized to its minimum size.
212 Having a process that is tracing also open this file, when the process
213 exits its file descriptor for this file will be closed, and in doing so,
214 the ring buffer will be "freed".
216 It may also stop tracing if disable_on_free option is set.
220 This is a mask that lets the user only trace on specified CPUs.
221 The format is a hex string representing the CPUs.
225 When dynamic ftrace is configured in (see the
226 section below "dynamic ftrace"), the code is dynamically
227 modified (code text rewrite) to disable calling of the
228 function profiler (mcount). This lets tracing be configured
229 in with practically no overhead in performance. This also
230 has a side effect of enabling or disabling specific functions
231 to be traced. Echoing names of functions into this file
232 will limit the trace to only those functions.
233 This influences the tracers "function" and "function_graph"
234 and thus also function profiling (see "function_profile_enabled").
236 The functions listed in "available_filter_functions" are what
237 can be written into this file.
239 This interface also allows for commands to be used. See the
240 "Filter commands" section for more details.
242 As a speed up, since processing strings can be quite expensive
243 and requires a check of all functions registered to tracing, instead
244 an index can be written into this file. A number (starting with "1")
245 written will instead select the same corresponding at the line position
246 of the "available_filter_functions" file.
250 This has an effect opposite to that of
251 set_ftrace_filter. Any function that is added here will not
252 be traced. If a function exists in both set_ftrace_filter
253 and set_ftrace_notrace, the function will _not_ be traced.
257 Have the function tracer only trace the threads whose PID are
260 If the "function-fork" option is set, then when a task whose
261 PID is listed in this file forks, the child's PID will
262 automatically be added to this file, and the child will be
263 traced by the function tracer as well. This option will also
264 cause PIDs of tasks that exit to be removed from the file.
266 set_ftrace_notrace_pid:
268 Have the function tracer ignore threads whose PID are listed in
271 If the "function-fork" option is set, then when a task whose
272 PID is listed in this file forks, the child's PID will
273 automatically be added to this file, and the child will not be
274 traced by the function tracer as well. This option will also
275 cause PIDs of tasks that exit to be removed from the file.
277 If a PID is in both this file and "set_ftrace_pid", then this
278 file takes precedence, and the thread will not be traced.
282 Have the events only trace a task with a PID listed in this file.
283 Note, sched_switch and sched_wake_up will also trace events
286 To have the PIDs of children of tasks with their PID in this file
287 added on fork, enable the "event-fork" option. That option will also
288 cause the PIDs of tasks to be removed from this file when the task
291 set_event_notrace_pid:
293 Have the events not trace a task with a PID listed in this file.
294 Note, sched_switch and sched_wakeup will trace threads not listed
295 in this file, even if a thread's PID is in the file if the
296 sched_switch or sched_wakeup events also trace a thread that should
299 To have the PIDs of children of tasks with their PID in this file
300 added on fork, enable the "event-fork" option. That option will also
301 cause the PIDs of tasks to be removed from this file when the task
306 Functions listed in this file will cause the function graph
307 tracer to only trace these functions and the functions that
308 they call. (See the section "dynamic ftrace" for more details).
309 Note, set_ftrace_filter and set_ftrace_notrace still affects
310 what functions are being traced.
314 Similar to set_graph_function, but will disable function graph
315 tracing when the function is hit until it exits the function.
316 This makes it possible to ignore tracing functions that are called
317 by a specific function.
319 available_filter_functions:
321 This lists the functions that ftrace has processed and can trace.
322 These are the function names that you can pass to
323 "set_ftrace_filter", "set_ftrace_notrace",
324 "set_graph_function", or "set_graph_notrace".
325 (See the section "dynamic ftrace" below for more details.)
327 dyn_ftrace_total_info:
329 This file is for debugging purposes. The number of functions that
330 have been converted to nops and are available to be traced.
334 This file is more for debugging ftrace, but can also be useful
335 in seeing if any function has a callback attached to it.
336 Not only does the trace infrastructure use ftrace function
337 trace utility, but other subsystems might too. This file
338 displays all functions that have a callback attached to them
339 as well as the number of callbacks that have been attached.
340 Note, a callback may also call multiple functions which will
341 not be listed in this count.
343 If the callback registered to be traced by a function with
344 the "save regs" attribute (thus even more overhead), a 'R'
345 will be displayed on the same line as the function that
346 is returning registers.
348 If the callback registered to be traced by a function with
349 the "ip modify" attribute (thus the regs->ip can be changed),
350 an 'I' will be displayed on the same line as the function that
353 If the architecture supports it, it will also show what callback
354 is being directly called by the function. If the count is greater
355 than 1 it most likely will be ftrace_ops_list_func().
357 If the callback of a function jumps to a trampoline that is
358 specific to the callback and which is not the standard trampoline,
359 its address will be printed as well as the function that the
362 function_profile_enabled:
364 When set it will enable all functions with either the function
365 tracer, or if configured, the function graph tracer. It will
366 keep a histogram of the number of functions that were called
367 and if the function graph tracer was configured, it will also keep
368 track of the time spent in those functions. The histogram
369 content can be displayed in the files:
371 trace_stat/function<cpu> ( function0, function1, etc).
375 A directory that holds different tracing stats.
379 Enable dynamic trace points. See kprobetrace.rst.
383 Dynamic trace points stats. See kprobetrace.rst.
387 Used with the function graph tracer. This is the max depth
388 it will trace into a function. Setting this to a value of
389 one will show only the first kernel function that is called
394 This is for tools that read the raw format files. If an event in
395 the ring buffer references a string, only a pointer to the string
396 is recorded into the buffer and not the string itself. This prevents
397 tools from knowing what that string was. This file displays the string
398 and address for the string allowing tools to map the pointers to what
403 Only the pid of the task is recorded in a trace event unless
404 the event specifically saves the task comm as well. Ftrace
405 makes a cache of pid mappings to comms to try to display
406 comms for events. If a pid for a comm is not listed, then
407 "<...>" is displayed in the output.
409 If the option "record-cmd" is set to "0", then comms of tasks
410 will not be saved during recording. By default, it is enabled.
414 By default, 128 comms are saved (see "saved_cmdlines" above). To
415 increase or decrease the amount of comms that are cached, echo
416 the number of comms to cache into this file.
420 If the option "record-tgid" is set, on each scheduling context switch
421 the Task Group ID of a task is saved in a table mapping the PID of
422 the thread to its TGID. By default, the "record-tgid" option is
427 This displays the "snapshot" buffer and also lets the user
428 take a snapshot of the current running trace.
429 See the "Snapshot" section below for more details.
433 When the stack tracer is activated, this will display the
434 maximum stack size it has encountered.
435 See the "Stack Trace" section below.
439 This displays the stack back trace of the largest stack
440 that was encountered when the stack tracer is activated.
441 See the "Stack Trace" section below.
445 This is similar to "set_ftrace_filter" but it limits what
446 functions the stack tracer will check.
450 Whenever an event is recorded into the ring buffer, a
451 "timestamp" is added. This stamp comes from a specified
452 clock. By default, ftrace uses the "local" clock. This
453 clock is very fast and strictly per cpu, but on some
454 systems it may not be monotonic with respect to other
455 CPUs. In other words, the local clocks may not be in sync
456 with local clocks on other CPUs.
458 Usual clocks for tracing::
461 [local] global counter x86-tsc
463 The clock with the square brackets around it is the one in effect.
466 Default clock, but may not be in sync across CPUs
469 This clock is in sync with all CPUs but may
470 be a bit slower than the local clock.
473 This is not a clock at all, but literally an atomic
474 counter. It counts up one by one, but is in sync
475 with all CPUs. This is useful when you need to
476 know exactly the order events occurred with respect to
477 each other on different CPUs.
480 This uses the jiffies counter and the time stamp
481 is relative to the time since boot up.
484 This makes ftrace use the same clock that perf uses.
485 Eventually perf will be able to read ftrace buffers
486 and this will help out in interleaving the data.
489 Architectures may define their own clocks. For
490 example, x86 uses its own TSC cycle clock here.
493 This uses the powerpc timebase register value.
494 This is in sync across CPUs and can also be used
495 to correlate events across hypervisor/guest if
499 This uses the fast monotonic clock (CLOCK_MONOTONIC)
500 which is monotonic and is subject to NTP rate adjustments.
503 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
504 which is monotonic but is not subject to any rate adjustments
505 and ticks at the same rate as the hardware clocksource.
508 This is the boot clock (CLOCK_BOOTTIME) and is based on the
509 fast monotonic clock, but also accounts for time spent in
510 suspend. Since the clock access is designed for use in
511 tracing in the suspend path, some side effects are possible
512 if clock is accessed after the suspend time is accounted before
513 the fast mono clock is updated. In this case, the clock update
514 appears to happen slightly sooner than it normally would have.
515 Also on 32-bit systems, it's possible that the 64-bit boot offset
516 sees a partial update. These effects are rare and post
517 processing should be able to handle them. See comments in the
518 ktime_get_boot_fast_ns() function for more information.
521 This is the tai clock (CLOCK_TAI) and is derived from the wall-
522 clock time. However, this clock does not experience
523 discontinuities and backwards jumps caused by NTP inserting leap
524 seconds. Since the clock access is designed for use in tracing,
525 side effects are possible. The clock access may yield wrong
526 readouts in case the internal TAI offset is updated e.g., caused
527 by setting the system time or using adjtimex() with an offset.
528 These effects are rare and post processing should be able to
529 handle them. See comments in the ktime_get_tai_fast_ns()
530 function for more information.
532 To set a clock, simply echo the clock name into this file::
534 # echo global > trace_clock
536 Setting a clock clears the ring buffer content as well as the
541 This is a very useful file for synchronizing user space
542 with events happening in the kernel. Writing strings into
543 this file will be written into the ftrace buffer.
545 It is useful in applications to open this file at the start
546 of the application and just reference the file descriptor
549 void trace_write(const char *fmt, ...)
559 n = vsnprintf(buf, 256, fmt, ap);
562 write(trace_fd, buf, n);
567 trace_fd = open("trace_marker", O_WRONLY);
569 Note: Writing into the trace_marker file can also initiate triggers
570 that are written into /sys/kernel/tracing/events/ftrace/print/trigger
571 See "Event triggers" in Documentation/trace/events.rst and an
572 example in Documentation/trace/histogram.rst (Section 3.)
576 This is similar to trace_marker above, but is meant for binary data
577 to be written to it, where a tool can be used to parse the data
582 Add dynamic tracepoints in programs.
587 Uprobe statistics. See uprobetrace.txt
591 This is a way to make multiple trace buffers where different
592 events can be recorded in different buffers.
593 See "Instances" section below.
597 This is the trace event directory. It holds event tracepoints
598 (also known as static tracepoints) that have been compiled
599 into the kernel. It shows what event tracepoints exist
600 and how they are grouped by system. There are "enable"
601 files at various levels that can enable the tracepoints
602 when a "1" is written to them.
604 See events.rst for more information.
608 By echoing in the event into this file, will enable that event.
610 See events.rst for more information.
614 A list of events that can be enabled in tracing.
616 See events.rst for more information.
620 Certain tracers may change the timestamp mode used when
621 logging trace events into the event buffer. Events with
622 different modes can coexist within a buffer but the mode in
623 effect when an event is logged determines which timestamp mode
624 is used for that event. The default timestamp mode is
627 Usual timestamp modes for tracing:
632 The timestamp mode with the square brackets around it is the
635 delta: Default timestamp mode - timestamp is a delta against
636 a per-buffer timestamp.
638 absolute: The timestamp is a full timestamp, not a delta
639 against some other value. As such it takes up more
640 space and is less efficient.
644 Directory for the Hardware Latency Detector.
645 See "Hardware Latency Detector" section below.
649 This is a directory that contains the trace per_cpu information.
651 per_cpu/cpu0/buffer_size_kb:
653 The ftrace buffer is defined per_cpu. That is, there's a separate
654 buffer for each CPU to allow writes to be done atomically,
655 and free from cache bouncing. These buffers may have different
656 size buffers. This file is similar to the buffer_size_kb
657 file, but it only displays or sets the buffer size for the
658 specific CPU. (here cpu0).
662 This is similar to the "trace" file, but it will only display
663 the data specific for the CPU. If written to, it only clears
664 the specific CPU buffer.
666 per_cpu/cpu0/trace_pipe
668 This is similar to the "trace_pipe" file, and is a consuming
669 read, but it will only display (and consume) the data specific
672 per_cpu/cpu0/trace_pipe_raw
674 For tools that can parse the ftrace ring buffer binary format,
675 the trace_pipe_raw file can be used to extract the data
676 from the ring buffer directly. With the use of the splice()
677 system call, the buffer data can be quickly transferred to
678 a file or to the network where a server is collecting the
681 Like trace_pipe, this is a consuming reader, where multiple
682 reads will always produce different data.
684 per_cpu/cpu0/snapshot:
686 This is similar to the main "snapshot" file, but will only
687 snapshot the current CPU (if supported). It only displays
688 the content of the snapshot for a given CPU, and if
689 written to, only clears this CPU buffer.
691 per_cpu/cpu0/snapshot_raw:
693 Similar to the trace_pipe_raw, but will read the binary format
694 from the snapshot buffer for the given CPU.
698 This displays certain stats about the ring buffer:
701 The number of events that are still in the buffer.
704 The number of lost events due to overwriting when
708 Should always be zero.
709 This gets set if so many events happened within a nested
710 event (ring buffer is re-entrant), that it fills the
711 buffer and starts dropping events.
714 Bytes actually read (not overwritten).
717 The oldest timestamp in the buffer
720 The current timestamp
723 Events lost due to overwrite option being off.
726 The number of events read.
731 Here is the list of current tracers that may be configured.
735 Function call tracer to trace all kernel functions.
739 Similar to the function tracer except that the
740 function tracer probes the functions on their entry
741 whereas the function graph tracer traces on both entry
742 and exit of the functions. It then provides the ability
743 to draw a graph of function calls similar to C code
748 The block tracer. The tracer used by the blktrace user
753 The Hardware Latency tracer is used to detect if the hardware
754 produces any latency. See "Hardware Latency Detector" section
759 Traces the areas that disable interrupts and saves
760 the trace with the longest max latency.
761 See tracing_max_latency. When a new max is recorded,
762 it replaces the old trace. It is best to view this
763 trace with the latency-format option enabled, which
764 happens automatically when the tracer is selected.
768 Similar to irqsoff but traces and records the amount of
769 time for which preemption is disabled.
773 Similar to irqsoff and preemptoff, but traces and
774 records the largest time for which irqs and/or preemption
779 Traces and records the max latency that it takes for
780 the highest priority task to get scheduled after
781 it has been woken up.
782 Traces all tasks as an average developer would expect.
786 Traces and records the max latency that it takes for just
787 RT tasks (as the current "wakeup" does). This is useful
788 for those interested in wake up timings of RT tasks.
792 Traces and records the max latency that it takes for
793 a SCHED_DEADLINE task to be woken (as the "wakeup" and
798 A special tracer that is used to trace binary module.
799 It will trace all the calls that a module makes to the
800 hardware. Everything it writes and reads from the I/O
805 This tracer can be configured when tracing likely/unlikely
806 calls within the kernel. It will trace when a likely and
807 unlikely branch is hit and if it was correct in its prediction
812 This is the "trace nothing" tracer. To remove all
813 tracers from tracing simply echo "nop" into
819 For most ftrace commands, failure modes are obvious and communicated
820 using standard return codes.
822 For other more involved commands, extended error information may be
823 available via the tracing/error_log file. For the commands that
824 support it, reading the tracing/error_log file after an error will
825 display more detailed information about what went wrong, if
826 information is available. The tracing/error_log file is a circular
827 error log displaying a small number (currently, 8) of ftrace errors
828 for the last (8) failed commands.
830 The extended error information and usage takes the form shown in
833 # echo xxx > /sys/kernel/debug/tracing/events/sched/sched_wakeup/trigger
834 echo: write error: Invalid argument
836 # cat /sys/kernel/debug/tracing/error_log
837 [ 5348.887237] location: error: Couldn't yyy: zzz
840 [ 7517.023364] location: error: Bad rrr: sss
844 To clear the error log, echo the empty string into it::
846 # echo > /sys/kernel/debug/tracing/error_log
848 Examples of using the tracer
849 ----------------------------
851 Here are typical examples of using the tracers when controlling
852 them only with the tracefs interface (without using any
853 user-land utilities).
858 Here is an example of the output format of the file "trace"::
862 # entries-in-buffer/entries-written: 140080/250280 #P:4
865 # / _----=> need-resched
866 # | / _---=> hardirq/softirq
867 # || / _--=> preempt-depth
869 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
871 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
872 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
873 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
874 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
875 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
876 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
877 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
878 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
879 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
880 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
883 A header is printed with the tracer name that is represented by
884 the trace. In this case the tracer is "function". Then it shows the
885 number of events in the buffer as well as the total number of entries
886 that were written. The difference is the number of entries that were
887 lost due to the buffer filling up (250280 - 140080 = 110200 events
890 The header explains the content of the events. Task name "bash", the task
891 PID "1977", the CPU that it was running on "000", the latency format
892 (explained below), the timestamp in <secs>.<usecs> format, the
893 function name that was traced "sys_close" and the parent function that
894 called this function "system_call_fastpath". The timestamp is the time
895 at which the function was entered.
900 When the latency-format option is enabled or when one of the latency
901 tracers is set, the trace file gives somewhat more information to see
902 why a latency happened. Here is a typical trace::
906 # irqsoff latency trace v1.1.5 on 3.8.0-test+
907 # --------------------------------------------------------------------
908 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
910 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
912 # => started at: __lock_task_sighand
913 # => ended at: _raw_spin_unlock_irqrestore
917 # / _-----=> irqs-off
918 # | / _----=> need-resched
919 # || / _---=> hardirq/softirq
920 # ||| / _--=> preempt-depth
922 # cmd pid ||||| time | caller
924 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
925 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
926 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
927 ps-6143 2d..1 306us : <stack trace>
928 => trace_hardirqs_on_caller
930 => _raw_spin_unlock_irqrestore
937 => system_call_fastpath
940 This shows that the current tracer is "irqsoff" tracing the time
941 for which interrupts were disabled. It gives the trace version (which
942 never changes) and the version of the kernel upon which this was executed on
943 (3.8). Then it displays the max latency in microseconds (259 us). The number
944 of trace entries displayed and the total number (both are four: #4/4).
945 VP, KP, SP, and HP are always zero and are reserved for later use.
946 #P is the number of online CPUs (#P:4).
948 The task is the process that was running when the latency
949 occurred. (ps pid: 6143).
951 The start and stop (the functions in which the interrupts were
952 disabled and enabled respectively) that caused the latencies:
954 - __lock_task_sighand is where the interrupts were disabled.
955 - _raw_spin_unlock_irqrestore is where they were enabled again.
957 The next lines after the header are the trace itself. The header
958 explains which is which.
960 cmd: The name of the process in the trace.
962 pid: The PID of that process.
964 CPU#: The CPU which the process was running on.
966 irqs-off: 'd' interrupts are disabled. '.' otherwise.
967 .. caution:: If the architecture does not support a way to
968 read the irq flags variable, an 'X' will always
972 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
973 - 'n' only TIF_NEED_RESCHED is set,
974 - 'p' only PREEMPT_NEED_RESCHED is set,
978 - 'Z' - NMI occurred inside a hardirq
979 - 'z' - NMI is running
980 - 'H' - hard irq occurred inside a softirq.
981 - 'h' - hard irq is running
982 - 's' - soft irq is running
983 - '.' - normal context.
985 preempt-depth: The level of preempt_disabled
987 The above is mostly meaningful for kernel developers.
990 When the latency-format option is enabled, the trace file
991 output includes a timestamp relative to the start of the
992 trace. This differs from the output when latency-format
993 is disabled, which includes an absolute timestamp.
996 This is just to help catch your eye a bit better. And
997 needs to be fixed to be only relative to the same CPU.
998 The marks are determined by the difference between this
999 current trace and the next trace.
1001 - '$' - greater than 1 second
1002 - '@' - greater than 100 millisecond
1003 - '*' - greater than 10 millisecond
1004 - '#' - greater than 1000 microsecond
1005 - '!' - greater than 100 microsecond
1006 - '+' - greater than 10 microsecond
1007 - ' ' - less than or equal to 10 microsecond.
1009 The rest is the same as the 'trace' file.
1011 Note, the latency tracers will usually end with a back trace
1012 to easily find where the latency occurred.
1017 The trace_options file (or the options directory) is used to control
1018 what gets printed in the trace output, or manipulate the tracers.
1019 To see what is available, simply cat the file::
1050 To disable one of the options, echo in the option prepended with
1053 echo noprint-parent > trace_options
1055 To enable an option, leave off the "no"::
1057 echo sym-offset > trace_options
1059 Here are the available options:
1062 On function traces, display the calling (parent)
1063 function as well as the function being traced.
1067 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
1070 bash-4000 [01] 1477.606694: simple_strtoul
1074 Display not only the function name, but also the
1075 offset in the function. For example, instead of
1076 seeing just "ktime_get", you will see
1077 "ktime_get+0xb/0x20".
1081 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
1084 This will also display the function address as well
1085 as the function name.
1089 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
1092 This deals with the trace file when the
1093 latency-format option is enabled.
1096 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
1097 (+0.000ms): simple_strtoul (kstrtoul)
1100 This will display raw numbers. This option is best for
1101 use with user applications that can translate the raw
1102 numbers better than having it done in the kernel.
1105 Similar to raw, but the numbers will be in a hexadecimal format.
1108 This will print out the formats in raw binary.
1111 When set, reading trace_pipe will not block when polled.
1114 Can disable trace_printk() from writing into the buffer.
1117 It is sometimes confusing when the CPU buffers are full
1118 and one CPU buffer had a lot of events recently, thus
1119 a shorter time frame, were another CPU may have only had
1120 a few events, which lets it have older events. When
1121 the trace is reported, it shows the oldest events first,
1122 and it may look like only one CPU ran (the one with the
1123 oldest events). When the annotate option is set, it will
1124 display when a new CPU buffer started::
1126 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
1127 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
1128 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
1129 ##### CPU 2 buffer started ####
1130 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
1131 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
1132 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
1135 This option changes the trace. It records a
1136 stacktrace of the current user space thread after
1140 when user stacktrace are enabled, look up which
1141 object the address belongs to, and print a
1142 relative address. This is especially useful when
1143 ASLR is on, otherwise you don't get a chance to
1144 resolve the address to object/file/line after
1145 the app is no longer running
1147 The lookup is performed when you read
1148 trace,trace_pipe. Example::
1150 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
1151 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
1155 When set, trace_printk()s will only show the format
1156 and not their parameters (if trace_bprintk() or
1157 trace_bputs() was used to save the trace_printk()).
1160 Show only the event data. Hides the comm, PID,
1161 timestamp, CPU, and other useful data.
1164 This option changes the trace output. When it is enabled,
1165 the trace displays additional information about the
1166 latency, as described in "Latency trace format".
1169 When set, opening the trace file for read, will pause
1170 writing to the ring buffer (as if tracing_on was set to zero).
1171 This simulates the original behavior of the trace file.
1172 When the file is closed, tracing will be enabled again.
1175 When set, "%p" in the event printk format displays the
1176 hashed pointer value instead of real address.
1177 This will be useful if you want to find out which hashed
1178 value is corresponding to the real value in trace log.
1181 When any event or tracer is enabled, a hook is enabled
1182 in the sched_switch trace point to fill comm cache
1183 with mapped pids and comms. But this may cause some
1184 overhead, and if you only care about pids, and not the
1185 name of the task, disabling this option can lower the
1186 impact of tracing. See "saved_cmdlines".
1189 When any event or tracer is enabled, a hook is enabled
1190 in the sched_switch trace point to fill the cache of
1191 mapped Thread Group IDs (TGID) mapping to pids. See
1195 This controls what happens when the trace buffer is
1196 full. If "1" (default), the oldest events are
1197 discarded and overwritten. If "0", then the newest
1198 events are discarded.
1199 (see per_cpu/cpu0/stats for overrun and dropped)
1202 When the free_buffer is closed, tracing will
1203 stop (tracing_on set to 0).
1206 Shows the interrupt, preempt count, need resched data.
1207 When disabled, the trace looks like::
1211 # entries-in-buffer/entries-written: 144405/9452052 #P:4
1213 # TASK-PID CPU# TIMESTAMP FUNCTION
1215 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
1216 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
1217 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
1221 When set, the trace_marker is writable (only by root).
1222 When disabled, the trace_marker will error with EINVAL
1226 When set, tasks with PIDs listed in set_event_pid will have
1227 the PIDs of their children added to set_event_pid when those
1228 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
1229 their PIDs will be removed from the file.
1231 This affects PIDs listed in set_event_notrace_pid as well.
1234 The latency tracers will enable function tracing
1235 if this option is enabled (default it is). When
1236 it is disabled, the latency tracers do not trace
1237 functions. This keeps the overhead of the tracer down
1238 when performing latency tests.
1241 When set, tasks with PIDs listed in set_ftrace_pid will
1242 have the PIDs of their children added to set_ftrace_pid
1243 when those tasks fork. Also, when tasks with PIDs in
1244 set_ftrace_pid exit, their PIDs will be removed from the
1247 This affects PIDs in set_ftrace_notrace_pid as well.
1250 When set, the latency tracers (irqsoff, wakeup, etc) will
1251 use function graph tracing instead of function tracing.
1254 When set, a stack trace is recorded after any trace event
1258 Enable branch tracing with the tracer. This enables branch
1259 tracer along with the currently set tracer. Enabling this
1260 with the "nop" tracer is the same as just enabling the
1263 .. tip:: Some tracers have their own options. They only appear in this
1264 file when the tracer is active. They always appear in the
1268 Here are the per tracer options:
1270 Options for function tracer:
1273 When set, a stack trace is recorded after every
1274 function that is recorded. NOTE! Limit the functions
1275 that are recorded before enabling this, with
1276 "set_ftrace_filter" otherwise the system performance
1277 will be critically degraded. Remember to disable
1278 this option before clearing the function filter.
1280 Options for function_graph tracer:
1282 Since the function_graph tracer has a slightly different output
1283 it has its own options to control what is displayed.
1286 When set, the "overrun" of the graph stack is
1287 displayed after each function traced. The
1288 overrun, is when the stack depth of the calls
1289 is greater than what is reserved for each task.
1290 Each task has a fixed array of functions to
1291 trace in the call graph. If the depth of the
1292 calls exceeds that, the function is not traced.
1293 The overrun is the number of functions missed
1294 due to exceeding this array.
1297 When set, the CPU number of the CPU where the trace
1298 occurred is displayed.
1301 When set, if the function takes longer than
1302 A certain amount, then a delay marker is
1303 displayed. See "delay" above, under the
1307 Unlike other tracers, the process' command line
1308 is not displayed by default, but instead only
1309 when a task is traced in and out during a context
1310 switch. Enabling this options has the command
1311 of each process displayed at every line.
1314 At the end of each function (the return)
1315 the duration of the amount of time in the
1316 function is displayed in microseconds.
1319 When set, the timestamp is displayed at each line.
1322 When disabled, functions that happen inside an
1323 interrupt will not be traced.
1326 When set, the return event will include the function
1327 that it represents. By default this is off, and
1328 only a closing curly bracket "}" is displayed for
1329 the return of a function.
1332 When running function graph tracer, to include
1333 the time a task schedules out in its function.
1334 When enabled, it will account time the task has been
1335 scheduled out as part of the function call.
1338 When running function profiler with function graph tracer,
1339 to include the time to call nested functions. When this is
1340 not set, the time reported for the function will only
1341 include the time the function itself executed for, not the
1342 time for functions that it called.
1344 Options for blk tracer:
1347 Shows a more minimalistic output.
1353 When interrupts are disabled, the CPU can not react to any other
1354 external event (besides NMIs and SMIs). This prevents the timer
1355 interrupt from triggering or the mouse interrupt from letting
1356 the kernel know of a new mouse event. The result is a latency
1357 with the reaction time.
1359 The irqsoff tracer tracks the time for which interrupts are
1360 disabled. When a new maximum latency is hit, the tracer saves
1361 the trace leading up to that latency point so that every time a
1362 new maximum is reached, the old saved trace is discarded and the
1365 To reset the maximum, echo 0 into tracing_max_latency. Here is
1368 # echo 0 > options/function-trace
1369 # echo irqsoff > current_tracer
1370 # echo 1 > tracing_on
1371 # echo 0 > tracing_max_latency
1374 # echo 0 > tracing_on
1378 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1379 # --------------------------------------------------------------------
1380 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1382 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
1384 # => started at: run_timer_softirq
1385 # => ended at: run_timer_softirq
1389 # / _-----=> irqs-off
1390 # | / _----=> need-resched
1391 # || / _---=> hardirq/softirq
1392 # ||| / _--=> preempt-depth
1394 # cmd pid ||||| time | caller
1396 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
1397 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
1398 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
1399 <idle>-0 0dNs3 25us : <stack trace>
1400 => _raw_spin_unlock_irq
1401 => run_timer_softirq
1406 => smp_apic_timer_interrupt
1407 => apic_timer_interrupt
1412 => x86_64_start_reservations
1413 => x86_64_start_kernel
1415 Here we see that we had a latency of 16 microseconds (which is
1416 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
1417 interrupts. The difference between the 16 and the displayed
1418 timestamp 25us occurred because the clock was incremented
1419 between the time of recording the max latency and the time of
1420 recording the function that had that latency.
1422 Note the above example had function-trace not set. If we set
1423 function-trace, we get a much larger output::
1425 with echo 1 > options/function-trace
1429 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1430 # --------------------------------------------------------------------
1431 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1433 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1435 # => started at: ata_scsi_queuecmd
1436 # => ended at: ata_scsi_queuecmd
1440 # / _-----=> irqs-off
1441 # | / _----=> need-resched
1442 # || / _---=> hardirq/softirq
1443 # ||| / _--=> preempt-depth
1445 # cmd pid ||||| time | caller
1447 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1448 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1449 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1450 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1451 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1452 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1453 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1454 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1455 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1457 bash-2042 3d..1 67us : delay_tsc <-__delay
1458 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1459 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1460 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1461 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1462 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1463 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1464 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1465 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1466 bash-2042 3d..1 120us : <stack trace>
1467 => _raw_spin_unlock_irqrestore
1468 => ata_scsi_queuecmd
1469 => scsi_dispatch_cmd
1471 => __blk_run_queue_uncond
1474 => submit_bio_noacct
1477 => __ext3_get_inode_loc
1486 => user_path_at_empty
1491 => system_call_fastpath
1494 Here we traced a 71 microsecond latency. But we also see all the
1495 functions that were called during that time. Note that by
1496 enabling function tracing, we incur an added overhead. This
1497 overhead may extend the latency times. But nevertheless, this
1498 trace has provided some very helpful debugging information.
1500 If we prefer function graph output instead of function, we can set
1501 display-graph option::
1503 with echo 1 > options/display-graph
1507 # irqsoff latency trace v1.1.5 on 4.20.0-rc6+
1508 # --------------------------------------------------------------------
1509 # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4)
1511 # | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0)
1513 # => started at: free_debug_processing
1514 # => ended at: return_to_handler
1518 # / _----=> need-resched
1519 # | / _---=> hardirq/softirq
1520 # || / _--=> preempt-depth
1522 # REL TIME CPU TASK/PID |||| DURATION FUNCTION CALLS
1523 # | | | | |||| | | | | | |
1524 0 us | 0) bash-1507 | d... | 0.000 us | _raw_spin_lock_irqsave();
1525 0 us | 0) bash-1507 | d..1 | 0.378 us | do_raw_spin_trylock();
1526 1 us | 0) bash-1507 | d..2 | | set_track() {
1527 2 us | 0) bash-1507 | d..2 | | save_stack_trace() {
1528 2 us | 0) bash-1507 | d..2 | | __save_stack_trace() {
1529 3 us | 0) bash-1507 | d..2 | | __unwind_start() {
1530 3 us | 0) bash-1507 | d..2 | | get_stack_info() {
1531 3 us | 0) bash-1507 | d..2 | 0.351 us | in_task_stack();
1532 4 us | 0) bash-1507 | d..2 | 1.107 us | }
1534 3750 us | 0) bash-1507 | d..1 | 0.516 us | do_raw_spin_unlock();
1535 3750 us | 0) bash-1507 | d..1 | 0.000 us | _raw_spin_unlock_irqrestore();
1536 3764 us | 0) bash-1507 | d..1 | 0.000 us | tracer_hardirqs_on();
1537 bash-1507 0d..1 3792us : <stack trace>
1538 => free_debug_processing
1547 => search_binary_handler
1548 => __do_execve_file.isra.32
1551 => entry_SYSCALL_64_after_hwframe
1556 When preemption is disabled, we may be able to receive
1557 interrupts but the task cannot be preempted and a higher
1558 priority task must wait for preemption to be enabled again
1559 before it can preempt a lower priority task.
1561 The preemptoff tracer traces the places that disable preemption.
1562 Like the irqsoff tracer, it records the maximum latency for
1563 which preemption was disabled. The control of preemptoff tracer
1564 is much like the irqsoff tracer.
1567 # echo 0 > options/function-trace
1568 # echo preemptoff > current_tracer
1569 # echo 1 > tracing_on
1570 # echo 0 > tracing_max_latency
1573 # echo 0 > tracing_on
1575 # tracer: preemptoff
1577 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1578 # --------------------------------------------------------------------
1579 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1581 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1583 # => started at: do_IRQ
1584 # => ended at: do_IRQ
1588 # / _-----=> irqs-off
1589 # | / _----=> need-resched
1590 # || / _---=> hardirq/softirq
1591 # ||| / _--=> preempt-depth
1593 # cmd pid ||||| time | caller
1595 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1596 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1597 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1598 sshd-1991 1d..1 52us : <stack trace>
1599 => sub_preempt_count
1605 This has some more changes. Preemption was disabled when an
1606 interrupt came in (notice the 'h'), and was enabled on exit.
1607 But we also see that interrupts have been disabled when entering
1608 the preempt off section and leaving it (the 'd'). We do not know if
1609 interrupts were enabled in the mean time or shortly after this
1613 # tracer: preemptoff
1615 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1616 # --------------------------------------------------------------------
1617 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1619 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1621 # => started at: wake_up_new_task
1622 # => ended at: task_rq_unlock
1626 # / _-----=> irqs-off
1627 # | / _----=> need-resched
1628 # || / _---=> hardirq/softirq
1629 # ||| / _--=> preempt-depth
1631 # cmd pid ||||| time | caller
1633 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1634 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1635 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1636 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1637 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1639 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1640 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1641 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1642 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1643 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1644 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1645 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1646 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1648 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1649 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1650 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1651 bash-1994 1d..2 36us : do_softirq <-irq_exit
1652 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1653 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1654 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1655 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1656 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1657 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1659 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1660 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1661 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1662 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1663 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1664 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1665 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1666 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1667 bash-1994 1.N.1 104us : <stack trace>
1668 => sub_preempt_count
1669 => _raw_spin_unlock_irqrestore
1677 The above is an example of the preemptoff trace with
1678 function-trace set. Here we see that interrupts were not disabled
1679 the entire time. The irq_enter code lets us know that we entered
1680 an interrupt 'h'. Before that, the functions being traced still
1681 show that it is not in an interrupt, but we can see from the
1682 functions themselves that this is not the case.
1687 Knowing the locations that have interrupts disabled or
1688 preemption disabled for the longest times is helpful. But
1689 sometimes we would like to know when either preemption and/or
1690 interrupts are disabled.
1692 Consider the following code::
1694 local_irq_disable();
1695 call_function_with_irqs_off();
1697 call_function_with_irqs_and_preemption_off();
1699 call_function_with_preemption_off();
1702 The irqsoff tracer will record the total length of
1703 call_function_with_irqs_off() and
1704 call_function_with_irqs_and_preemption_off().
1706 The preemptoff tracer will record the total length of
1707 call_function_with_irqs_and_preemption_off() and
1708 call_function_with_preemption_off().
1710 But neither will trace the time that interrupts and/or
1711 preemption is disabled. This total time is the time that we can
1712 not schedule. To record this time, use the preemptirqsoff
1715 Again, using this trace is much like the irqsoff and preemptoff
1719 # echo 0 > options/function-trace
1720 # echo preemptirqsoff > current_tracer
1721 # echo 1 > tracing_on
1722 # echo 0 > tracing_max_latency
1725 # echo 0 > tracing_on
1727 # tracer: preemptirqsoff
1729 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1730 # --------------------------------------------------------------------
1731 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1733 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1735 # => started at: ata_scsi_queuecmd
1736 # => ended at: ata_scsi_queuecmd
1740 # / _-----=> irqs-off
1741 # | / _----=> need-resched
1742 # || / _---=> hardirq/softirq
1743 # ||| / _--=> preempt-depth
1745 # cmd pid ||||| time | caller
1747 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1748 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1749 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1750 ls-2230 3...1 111us : <stack trace>
1751 => sub_preempt_count
1752 => _raw_spin_unlock_irqrestore
1753 => ata_scsi_queuecmd
1754 => scsi_dispatch_cmd
1756 => __blk_run_queue_uncond
1759 => submit_bio_noacct
1764 => htree_dirblock_to_tree
1765 => ext3_htree_fill_tree
1769 => system_call_fastpath
1772 The trace_hardirqs_off_thunk is called from assembly on x86 when
1773 interrupts are disabled in the assembly code. Without the
1774 function tracing, we do not know if interrupts were enabled
1775 within the preemption points. We do see that it started with
1778 Here is a trace with function-trace set::
1780 # tracer: preemptirqsoff
1782 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1783 # --------------------------------------------------------------------
1784 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1786 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1788 # => started at: schedule
1789 # => ended at: mutex_unlock
1793 # / _-----=> irqs-off
1794 # | / _----=> need-resched
1795 # || / _---=> hardirq/softirq
1796 # ||| / _--=> preempt-depth
1798 # cmd pid ||||| time | caller
1800 kworker/-59 3...1 0us : __schedule <-schedule
1801 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1802 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1803 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1804 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1805 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1806 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1807 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1808 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1809 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1810 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1811 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1812 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1813 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1814 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1815 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1816 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1817 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1818 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1819 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1820 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1821 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1822 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1823 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1824 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1825 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1826 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1827 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1828 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1829 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1830 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1831 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1833 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1834 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1835 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1836 ls-2269 3d..3 21us : do_softirq <-irq_exit
1837 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1838 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1839 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1840 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1841 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1842 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1843 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1845 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1846 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1847 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1848 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1849 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1850 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1852 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1853 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1854 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1855 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1856 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1857 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1858 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1859 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1860 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1861 ls-2269 3d... 186us : <stack trace>
1862 => __mutex_unlock_slowpath
1869 => system_call_fastpath
1871 This is an interesting trace. It started with kworker running and
1872 scheduling out and ls taking over. But as soon as ls released the
1873 rq lock and enabled interrupts (but not preemption) an interrupt
1874 triggered. When the interrupt finished, it started running softirqs.
1875 But while the softirq was running, another interrupt triggered.
1876 When an interrupt is running inside a softirq, the annotation is 'H'.
1882 One common case that people are interested in tracing is the
1883 time it takes for a task that is woken to actually wake up.
1884 Now for non Real-Time tasks, this can be arbitrary. But tracing
1885 it none the less can be interesting.
1887 Without function tracing::
1889 # echo 0 > options/function-trace
1890 # echo wakeup > current_tracer
1891 # echo 1 > tracing_on
1892 # echo 0 > tracing_max_latency
1894 # echo 0 > tracing_on
1898 # wakeup latency trace v1.1.5 on 3.8.0-test+
1899 # --------------------------------------------------------------------
1900 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1902 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1906 # / _-----=> irqs-off
1907 # | / _----=> need-resched
1908 # || / _---=> hardirq/softirq
1909 # ||| / _--=> preempt-depth
1911 # cmd pid ||||| time | caller
1913 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1914 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1915 <idle>-0 3d..3 15us : __schedule <-schedule
1916 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1918 The tracer only traces the highest priority task in the system
1919 to avoid tracing the normal circumstances. Here we see that
1920 the kworker with a nice priority of -20 (not very nice), took
1921 just 15 microseconds from the time it woke up, to the time it
1924 Non Real-Time tasks are not that interesting. A more interesting
1925 trace is to concentrate only on Real-Time tasks.
1930 In a Real-Time environment it is very important to know the
1931 wakeup time it takes for the highest priority task that is woken
1932 up to the time that it executes. This is also known as "schedule
1933 latency". I stress the point that this is about RT tasks. It is
1934 also important to know the scheduling latency of non-RT tasks,
1935 but the average schedule latency is better for non-RT tasks.
1936 Tools like LatencyTop are more appropriate for such
1939 Real-Time environments are interested in the worst case latency.
1940 That is the longest latency it takes for something to happen,
1941 and not the average. We can have a very fast scheduler that may
1942 only have a large latency once in a while, but that would not
1943 work well with Real-Time tasks. The wakeup_rt tracer was designed
1944 to record the worst case wakeups of RT tasks. Non-RT tasks are
1945 not recorded because the tracer only records one worst case and
1946 tracing non-RT tasks that are unpredictable will overwrite the
1947 worst case latency of RT tasks (just run the normal wakeup
1948 tracer for a while to see that effect).
1950 Since this tracer only deals with RT tasks, we will run this
1951 slightly differently than we did with the previous tracers.
1952 Instead of performing an 'ls', we will run 'sleep 1' under
1953 'chrt' which changes the priority of the task.
1956 # echo 0 > options/function-trace
1957 # echo wakeup_rt > current_tracer
1958 # echo 1 > tracing_on
1959 # echo 0 > tracing_max_latency
1961 # echo 0 > tracing_on
1967 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1968 # --------------------------------------------------------------------
1969 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1971 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1975 # / _-----=> irqs-off
1976 # | / _----=> need-resched
1977 # || / _---=> hardirq/softirq
1978 # ||| / _--=> preempt-depth
1980 # cmd pid ||||| time | caller
1982 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1983 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1984 <idle>-0 3d..3 5us : __schedule <-schedule
1985 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1988 Running this on an idle system, we see that it only took 5 microseconds
1989 to perform the task switch. Note, since the trace point in the schedule
1990 is before the actual "switch", we stop the tracing when the recorded task
1991 is about to schedule in. This may change if we add a new marker at the
1992 end of the scheduler.
1994 Notice that the recorded task is 'sleep' with the PID of 2389
1995 and it has an rt_prio of 5. This priority is user-space priority
1996 and not the internal kernel priority. The policy is 1 for
1997 SCHED_FIFO and 2 for SCHED_RR.
1999 Note, that the trace data shows the internal priority (99 - rtprio).
2002 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
2004 The 0:120:R means idle was running with a nice priority of 0 (120 - 120)
2005 and in the running state 'R'. The sleep task was scheduled in with
2006 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
2007 and it too is in the running state.
2009 Doing the same with chrt -r 5 and function-trace set.
2012 echo 1 > options/function-trace
2016 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
2017 # --------------------------------------------------------------------
2018 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
2020 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
2024 # / _-----=> irqs-off
2025 # | / _----=> need-resched
2026 # || / _---=> hardirq/softirq
2027 # ||| / _--=> preempt-depth
2029 # cmd pid ||||| time | caller
2031 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
2032 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
2033 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
2034 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
2035 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
2036 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
2037 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
2038 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
2039 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
2040 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2041 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
2042 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
2043 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
2044 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
2045 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
2046 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
2047 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
2048 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
2049 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
2050 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
2051 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
2052 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
2053 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
2054 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
2055 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
2056 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
2057 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
2058 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
2059 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
2060 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
2061 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
2062 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
2063 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
2064 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
2065 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
2066 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
2067 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
2068 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit
2069 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
2070 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
2071 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
2072 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
2073 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
2074 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
2075 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
2076 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
2077 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
2078 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
2079 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
2080 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
2081 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
2082 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2083 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
2084 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
2085 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
2086 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2087 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
2088 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
2089 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
2090 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
2091 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
2092 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
2093 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
2094 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
2095 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
2096 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
2097 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
2098 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2099 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
2100 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
2101 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
2102 <idle>-0 3.N.. 25us : schedule <-cpu_idle
2103 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
2104 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
2105 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
2106 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
2107 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
2108 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
2109 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
2110 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
2111 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
2112 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
2113 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
2114 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
2115 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
2117 This isn't that big of a trace, even with function tracing enabled,
2118 so I included the entire trace.
2120 The interrupt went off while when the system was idle. Somewhere
2121 before task_woken_rt() was called, the NEED_RESCHED flag was set,
2122 this is indicated by the first occurrence of the 'N' flag.
2124 Latency tracing and events
2125 --------------------------
2126 As function tracing can induce a much larger latency, but without
2127 seeing what happens within the latency it is hard to know what
2128 caused it. There is a middle ground, and that is with enabling
2132 # echo 0 > options/function-trace
2133 # echo wakeup_rt > current_tracer
2134 # echo 1 > events/enable
2135 # echo 1 > tracing_on
2136 # echo 0 > tracing_max_latency
2138 # echo 0 > tracing_on
2142 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
2143 # --------------------------------------------------------------------
2144 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
2146 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
2150 # / _-----=> irqs-off
2151 # | / _----=> need-resched
2152 # || / _---=> hardirq/softirq
2153 # ||| / _--=> preempt-depth
2155 # cmd pid ||||| time | caller
2157 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
2158 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
2159 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
2160 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
2161 <idle>-0 2.N.2 2us : power_end: cpu_id=2
2162 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
2163 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
2164 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
2165 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
2166 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
2167 <idle>-0 2d..3 6us : __schedule <-schedule
2168 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
2171 Hardware Latency Detector
2172 -------------------------
2174 The hardware latency detector is executed by enabling the "hwlat" tracer.
2176 NOTE, this tracer will affect the performance of the system as it will
2177 periodically make a CPU constantly busy with interrupts disabled.
2180 # echo hwlat > current_tracer
2185 # entries-in-buffer/entries-written: 13/13 #P:8
2188 # / _----=> need-resched
2189 # | / _---=> hardirq/softirq
2190 # || / _--=> preempt-depth
2192 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2194 <...>-1729 [001] d... 678.473449: #1 inner/outer(us): 11/12 ts:1581527483.343962693 count:6
2195 <...>-1729 [004] d... 689.556542: #2 inner/outer(us): 16/9 ts:1581527494.889008092 count:1
2196 <...>-1729 [005] d... 714.756290: #3 inner/outer(us): 16/16 ts:1581527519.678961629 count:5
2197 <...>-1729 [001] d... 718.788247: #4 inner/outer(us): 9/17 ts:1581527523.889012713 count:1
2198 <...>-1729 [002] d... 719.796341: #5 inner/outer(us): 13/9 ts:1581527524.912872606 count:1
2199 <...>-1729 [006] d... 844.787091: #6 inner/outer(us): 9/12 ts:1581527649.889048502 count:2
2200 <...>-1729 [003] d... 849.827033: #7 inner/outer(us): 18/9 ts:1581527654.889013793 count:1
2201 <...>-1729 [007] d... 853.859002: #8 inner/outer(us): 9/12 ts:1581527658.889065736 count:1
2202 <...>-1729 [001] d... 855.874978: #9 inner/outer(us): 9/11 ts:1581527660.861991877 count:1
2203 <...>-1729 [001] d... 863.938932: #10 inner/outer(us): 9/11 ts:1581527668.970010500 count:1 nmi-total:7 nmi-count:1
2204 <...>-1729 [007] d... 878.050780: #11 inner/outer(us): 9/12 ts:1581527683.385002600 count:1 nmi-total:5 nmi-count:1
2205 <...>-1729 [007] d... 886.114702: #12 inner/outer(us): 9/12 ts:1581527691.385001600 count:1
2208 The above output is somewhat the same in the header. All events will have
2209 interrupts disabled 'd'. Under the FUNCTION title there is:
2212 This is the count of events recorded that were greater than the
2213 tracing_threshold (See below).
2215 inner/outer(us): 11/11
2217 This shows two numbers as "inner latency" and "outer latency". The test
2218 runs in a loop checking a timestamp twice. The latency detected within
2219 the two timestamps is the "inner latency" and the latency detected
2220 after the previous timestamp and the next timestamp in the loop is
2221 the "outer latency".
2223 ts:1581527483.343962693
2225 The absolute timestamp that the first latency was recorded in the window.
2229 The number of times a latency was detected during the window.
2231 nmi-total:7 nmi-count:1
2233 On architectures that support it, if an NMI comes in during the
2234 test, the time spent in NMI is reported in "nmi-total" (in
2237 All architectures that have NMIs will show the "nmi-count" if an
2238 NMI comes in during the test.
2243 This gets automatically set to "10" to represent 10
2244 microseconds. This is the threshold of latency that
2245 needs to be detected before the trace will be recorded.
2247 Note, when hwlat tracer is finished (another tracer is
2248 written into "current_tracer"), the original value for
2249 tracing_threshold is placed back into this file.
2251 hwlat_detector/width
2252 The length of time the test runs with interrupts disabled.
2254 hwlat_detector/window
2255 The length of time of the window which the test
2256 runs. That is, the test will run for "width"
2257 microseconds per "window" microseconds
2260 When the test is started. A kernel thread is created that
2261 runs the test. This thread will alternate between CPUs
2262 listed in the tracing_cpumask between each period
2263 (one "window"). To limit the test to specific CPUs
2264 set the mask in this file to only the CPUs that the test
2270 This tracer is the function tracer. Enabling the function tracer
2271 can be done from the debug file system. Make sure the
2272 ftrace_enabled is set; otherwise this tracer is a nop.
2273 See the "ftrace_enabled" section below.
2276 # sysctl kernel.ftrace_enabled=1
2277 # echo function > current_tracer
2278 # echo 1 > tracing_on
2280 # echo 0 > tracing_on
2284 # entries-in-buffer/entries-written: 24799/24799 #P:4
2287 # / _----=> need-resched
2288 # | / _---=> hardirq/softirq
2289 # || / _--=> preempt-depth
2291 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2293 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
2294 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
2295 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
2296 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
2297 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
2298 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
2299 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
2300 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
2304 Note: function tracer uses ring buffers to store the above
2305 entries. The newest data may overwrite the oldest data.
2306 Sometimes using echo to stop the trace is not sufficient because
2307 the tracing could have overwritten the data that you wanted to
2308 record. For this reason, it is sometimes better to disable
2309 tracing directly from a program. This allows you to stop the
2310 tracing at the point that you hit the part that you are
2311 interested in. To disable the tracing directly from a C program,
2312 something like following code snippet can be used::
2316 int main(int argc, char *argv[]) {
2318 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
2320 if (condition_hit()) {
2321 write(trace_fd, "0", 1);
2327 Single thread tracing
2328 ---------------------
2330 By writing into set_ftrace_pid you can trace a
2331 single thread. For example::
2333 # cat set_ftrace_pid
2335 # echo 3111 > set_ftrace_pid
2336 # cat set_ftrace_pid
2338 # echo function > current_tracer
2342 # TASK-PID CPU# TIMESTAMP FUNCTION
2344 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
2345 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
2346 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
2347 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
2348 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
2349 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
2350 # echo > set_ftrace_pid
2354 # TASK-PID CPU# TIMESTAMP FUNCTION
2356 ##### CPU 3 buffer started ####
2357 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
2358 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
2359 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
2360 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
2361 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
2363 If you want to trace a function when executing, you could use
2364 something like this simple program.
2369 #include <sys/types.h>
2370 #include <sys/stat.h>
2376 #define STR(x) _STR(x)
2377 #define MAX_PATH 256
2379 const char *find_tracefs(void)
2381 static char tracefs[MAX_PATH+1];
2382 static int tracefs_found;
2389 if ((fp = fopen("/proc/mounts","r")) == NULL) {
2390 perror("/proc/mounts");
2394 while (fscanf(fp, "%*s %"
2396 "s %99s %*s %*d %*d\n",
2397 tracefs, type) == 2) {
2398 if (strcmp(type, "tracefs") == 0)
2403 if (strcmp(type, "tracefs") != 0) {
2404 fprintf(stderr, "tracefs not mounted");
2408 strcat(tracefs, "/tracing/");
2414 const char *tracing_file(const char *file_name)
2416 static char trace_file[MAX_PATH+1];
2417 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name);
2421 int main (int argc, char **argv)
2431 ffd = open(tracing_file("current_tracer"), O_WRONLY);
2434 write(ffd, "nop", 3);
2436 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
2437 s = sprintf(line, "%d\n", getpid());
2440 write(ffd, "function", 8);
2445 execvp(argv[1], argv+1);
2451 Or this simple script!
2456 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts`
2457 echo 0 > $tracefs/tracing_on
2458 echo $$ > $tracefs/set_ftrace_pid
2459 echo function > $tracefs/current_tracer
2460 echo 1 > $tracefs/tracing_on
2464 function graph tracer
2465 ---------------------------
2467 This tracer is similar to the function tracer except that it
2468 probes a function on its entry and its exit. This is done by
2469 using a dynamically allocated stack of return addresses in each
2470 task_struct. On function entry the tracer overwrites the return
2471 address of each function traced to set a custom probe. Thus the
2472 original return address is stored on the stack of return address
2475 Probing on both ends of a function leads to special features
2478 - measure of a function's time execution
2479 - having a reliable call stack to draw function calls graph
2481 This tracer is useful in several situations:
2483 - you want to find the reason of a strange kernel behavior and
2484 need to see what happens in detail on any areas (or specific
2487 - you are experiencing weird latencies but it's difficult to
2490 - you want to find quickly which path is taken by a specific
2493 - you just want to peek inside a working kernel and want to see
2498 # tracer: function_graph
2500 # CPU DURATION FUNCTION CALLS
2504 0) | do_sys_open() {
2506 0) | kmem_cache_alloc() {
2507 0) 1.382 us | __might_sleep();
2509 0) | strncpy_from_user() {
2510 0) | might_fault() {
2511 0) 1.389 us | __might_sleep();
2516 0) 0.668 us | _spin_lock();
2517 0) 0.570 us | expand_files();
2518 0) 0.586 us | _spin_unlock();
2521 There are several columns that can be dynamically
2522 enabled/disabled. You can use every combination of options you
2523 want, depending on your needs.
2525 - The cpu number on which the function executed is default
2526 enabled. It is sometimes better to only trace one cpu (see
2527 tracing_cpu_mask file) or you might sometimes see unordered
2528 function calls while cpu tracing switch.
2530 - hide: echo nofuncgraph-cpu > trace_options
2531 - show: echo funcgraph-cpu > trace_options
2533 - The duration (function's time of execution) is displayed on
2534 the closing bracket line of a function or on the same line
2535 than the current function in case of a leaf one. It is default
2538 - hide: echo nofuncgraph-duration > trace_options
2539 - show: echo funcgraph-duration > trace_options
2541 - The overhead field precedes the duration field in case of
2542 reached duration thresholds.
2544 - hide: echo nofuncgraph-overhead > trace_options
2545 - show: echo funcgraph-overhead > trace_options
2546 - depends on: funcgraph-duration
2550 3) # 1837.709 us | } /* __switch_to */
2551 3) | finish_task_switch() {
2552 3) 0.313 us | _raw_spin_unlock_irq();
2554 3) # 1889.063 us | } /* __schedule */
2555 3) ! 140.417 us | } /* __schedule */
2556 3) # 2034.948 us | } /* schedule */
2557 3) * 33998.59 us | } /* schedule_preempt_disabled */
2561 1) 0.260 us | msecs_to_jiffies();
2562 1) 0.313 us | __rcu_read_unlock();
2565 1) 0.313 us | rcu_bh_qs();
2566 1) 0.313 us | __local_bh_enable();
2568 1) 0.365 us | idle_cpu();
2569 1) | rcu_irq_exit() {
2570 1) 0.417 us | rcu_eqs_enter_common.isra.47();
2574 1) @ 119760.2 us | }
2580 2) 0.417 us | scheduler_ipi();
2590 + means that the function exceeded 10 usecs.
2591 ! means that the function exceeded 100 usecs.
2592 # means that the function exceeded 1000 usecs.
2593 * means that the function exceeded 10 msecs.
2594 @ means that the function exceeded 100 msecs.
2595 $ means that the function exceeded 1 sec.
2598 - The task/pid field displays the thread cmdline and pid which
2599 executed the function. It is default disabled.
2601 - hide: echo nofuncgraph-proc > trace_options
2602 - show: echo funcgraph-proc > trace_options
2606 # tracer: function_graph
2608 # CPU TASK/PID DURATION FUNCTION CALLS
2610 0) sh-4802 | | d_free() {
2611 0) sh-4802 | | call_rcu() {
2612 0) sh-4802 | | __call_rcu() {
2613 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2614 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2615 0) sh-4802 | 2.899 us | }
2616 0) sh-4802 | 4.040 us | }
2617 0) sh-4802 | 5.151 us | }
2618 0) sh-4802 | + 49.370 us | }
2621 - The absolute time field is an absolute timestamp given by the
2622 system clock since it started. A snapshot of this time is
2623 given on each entry/exit of functions
2625 - hide: echo nofuncgraph-abstime > trace_options
2626 - show: echo funcgraph-abstime > trace_options
2631 # TIME CPU DURATION FUNCTION CALLS
2633 360.774522 | 1) 0.541 us | }
2634 360.774522 | 1) 4.663 us | }
2635 360.774523 | 1) 0.541 us | __wake_up_bit();
2636 360.774524 | 1) 6.796 us | }
2637 360.774524 | 1) 7.952 us | }
2638 360.774525 | 1) 9.063 us | }
2639 360.774525 | 1) 0.615 us | journal_mark_dirty();
2640 360.774527 | 1) 0.578 us | __brelse();
2641 360.774528 | 1) | reiserfs_prepare_for_journal() {
2642 360.774528 | 1) | unlock_buffer() {
2643 360.774529 | 1) | wake_up_bit() {
2644 360.774529 | 1) | bit_waitqueue() {
2645 360.774530 | 1) 0.594 us | __phys_addr();
2648 The function name is always displayed after the closing bracket
2649 for a function if the start of that function is not in the
2652 Display of the function name after the closing bracket may be
2653 enabled for functions whose start is in the trace buffer,
2654 allowing easier searching with grep for function durations.
2655 It is default disabled.
2657 - hide: echo nofuncgraph-tail > trace_options
2658 - show: echo funcgraph-tail > trace_options
2660 Example with nofuncgraph-tail (default)::
2663 0) | kmem_cache_free() {
2664 0) 0.518 us | __phys_addr();
2668 Example with funcgraph-tail::
2671 0) | kmem_cache_free() {
2672 0) 0.518 us | __phys_addr();
2673 0) 1.757 us | } /* kmem_cache_free() */
2674 0) 2.861 us | } /* putname() */
2676 You can put some comments on specific functions by using
2677 trace_printk() For example, if you want to put a comment inside
2678 the __might_sleep() function, you just have to include
2679 <linux/ftrace.h> and call trace_printk() inside __might_sleep()::
2681 trace_printk("I'm a comment!\n")
2685 1) | __might_sleep() {
2686 1) | /* I'm a comment! */
2690 You might find other useful features for this tracer in the
2691 following "dynamic ftrace" section such as tracing only specific
2697 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2698 virtually no overhead when function tracing is disabled. The way
2699 this works is the mcount function call (placed at the start of
2700 every kernel function, produced by the -pg switch in gcc),
2701 starts of pointing to a simple return. (Enabling FTRACE will
2702 include the -pg switch in the compiling of the kernel.)
2704 At compile time every C file object is run through the
2705 recordmcount program (located in the scripts directory). This
2706 program will parse the ELF headers in the C object to find all
2707 the locations in the .text section that call mcount. Starting
2708 with gcc version 4.6, the -mfentry has been added for x86, which
2709 calls "__fentry__" instead of "mcount". Which is called before
2710 the creation of the stack frame.
2712 Note, not all sections are traced. They may be prevented by either
2713 a notrace, or blocked another way and all inline functions are not
2714 traced. Check the "available_filter_functions" file to see what functions
2717 A section called "__mcount_loc" is created that holds
2718 references to all the mcount/fentry call sites in the .text section.
2719 The recordmcount program re-links this section back into the
2720 original object. The final linking stage of the kernel will add all these
2721 references into a single table.
2723 On boot up, before SMP is initialized, the dynamic ftrace code
2724 scans this table and updates all the locations into nops. It
2725 also records the locations, which are added to the
2726 available_filter_functions list. Modules are processed as they
2727 are loaded and before they are executed. When a module is
2728 unloaded, it also removes its functions from the ftrace function
2729 list. This is automatic in the module unload code, and the
2730 module author does not need to worry about it.
2732 When tracing is enabled, the process of modifying the function
2733 tracepoints is dependent on architecture. The old method is to use
2734 kstop_machine to prevent races with the CPUs executing code being
2735 modified (which can cause the CPU to do undesirable things, especially
2736 if the modified code crosses cache (or page) boundaries), and the nops are
2737 patched back to calls. But this time, they do not call mcount
2738 (which is just a function stub). They now call into the ftrace
2741 The new method of modifying the function tracepoints is to place
2742 a breakpoint at the location to be modified, sync all CPUs, modify
2743 the rest of the instruction not covered by the breakpoint. Sync
2744 all CPUs again, and then remove the breakpoint with the finished
2745 version to the ftrace call site.
2747 Some archs do not even need to monkey around with the synchronization,
2748 and can just slap the new code on top of the old without any
2749 problems with other CPUs executing it at the same time.
2751 One special side-effect to the recording of the functions being
2752 traced is that we can now selectively choose which functions we
2753 wish to trace and which ones we want the mcount calls to remain
2756 Two files are used, one for enabling and one for disabling the
2757 tracing of specified functions. They are:
2765 A list of available functions that you can add to these files is
2768 available_filter_functions
2772 # cat available_filter_functions
2781 If I am only interested in sys_nanosleep and hrtimer_interrupt::
2783 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2784 # echo function > current_tracer
2785 # echo 1 > tracing_on
2787 # echo 0 > tracing_on
2791 # entries-in-buffer/entries-written: 5/5 #P:4
2794 # / _----=> need-resched
2795 # | / _---=> hardirq/softirq
2796 # || / _--=> preempt-depth
2798 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2800 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2801 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2802 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2803 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2804 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2806 To see which functions are being traced, you can cat the file:
2809 # cat set_ftrace_filter
2814 Perhaps this is not enough. The filters also allow glob(7) matching.
2817 will match functions that begin with <match>
2819 will match functions that end with <match>
2821 will match functions that have <match> in it
2822 ``<match1>*<match2>``
2823 will match functions that begin with <match1> and end with <match2>
2826 It is better to use quotes to enclose the wild cards,
2827 otherwise the shell may expand the parameters into names
2828 of files in the local directory.
2832 # echo 'hrtimer_*' > set_ftrace_filter
2838 # entries-in-buffer/entries-written: 897/897 #P:4
2841 # / _----=> need-resched
2842 # | / _---=> hardirq/softirq
2843 # || / _--=> preempt-depth
2845 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2847 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2848 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2849 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2850 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2851 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2852 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2853 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2854 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2856 Notice that we lost the sys_nanosleep.
2859 # cat set_ftrace_filter
2864 hrtimer_try_to_cancel
2868 hrtimer_force_reprogram
2869 hrtimer_get_next_event
2873 hrtimer_get_remaining
2875 hrtimer_init_sleeper
2878 This is because the '>' and '>>' act just like they do in bash.
2879 To rewrite the filters, use '>'
2880 To append to the filters, use '>>'
2882 To clear out a filter so that all functions will be recorded
2885 # echo > set_ftrace_filter
2886 # cat set_ftrace_filter
2889 Again, now we want to append.
2893 # echo sys_nanosleep > set_ftrace_filter
2894 # cat set_ftrace_filter
2896 # echo 'hrtimer_*' >> set_ftrace_filter
2897 # cat set_ftrace_filter
2902 hrtimer_try_to_cancel
2906 hrtimer_force_reprogram
2907 hrtimer_get_next_event
2912 hrtimer_get_remaining
2914 hrtimer_init_sleeper
2917 The set_ftrace_notrace prevents those functions from being
2921 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2927 # entries-in-buffer/entries-written: 39608/39608 #P:4
2930 # / _----=> need-resched
2931 # | / _---=> hardirq/softirq
2932 # || / _--=> preempt-depth
2934 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2936 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2937 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2938 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2939 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2940 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2941 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2942 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2943 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2944 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2945 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2946 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2947 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2949 We can see that there's no more lock or preempt tracing.
2951 Selecting function filters via index
2952 ------------------------------------
2954 Because processing of strings is expensive (the address of the function
2955 needs to be looked up before comparing to the string being passed in),
2956 an index can be used as well to enable functions. This is useful in the
2957 case of setting thousands of specific functions at a time. By passing
2958 in a list of numbers, no string processing will occur. Instead, the function
2959 at the specific location in the internal array (which corresponds to the
2960 functions in the "available_filter_functions" file), is selected.
2964 # echo 1 > set_ftrace_filter
2966 Will select the first function listed in "available_filter_functions"
2970 # head -1 available_filter_functions
2971 trace_initcall_finish_cb
2973 # cat set_ftrace_filter
2974 trace_initcall_finish_cb
2976 # head -50 available_filter_functions | tail -1
2979 # echo 1 50 > set_ftrace_filter
2980 # cat set_ftrace_filter
2981 trace_initcall_finish_cb
2984 Dynamic ftrace with the function graph tracer
2985 ---------------------------------------------
2987 Although what has been explained above concerns both the
2988 function tracer and the function-graph-tracer, there are some
2989 special features only available in the function-graph tracer.
2991 If you want to trace only one function and all of its children,
2992 you just have to echo its name into set_graph_function::
2994 echo __do_fault > set_graph_function
2996 will produce the following "expanded" trace of the __do_fault()
3000 0) | filemap_fault() {
3001 0) | find_lock_page() {
3002 0) 0.804 us | find_get_page();
3003 0) | __might_sleep() {
3007 0) 0.653 us | _spin_lock();
3008 0) 0.578 us | page_add_file_rmap();
3009 0) 0.525 us | native_set_pte_at();
3010 0) 0.585 us | _spin_unlock();
3011 0) | unlock_page() {
3012 0) 0.541 us | page_waitqueue();
3013 0) 0.639 us | __wake_up_bit();
3017 0) | filemap_fault() {
3018 0) | find_lock_page() {
3019 0) 0.698 us | find_get_page();
3020 0) | __might_sleep() {
3024 0) 0.631 us | _spin_lock();
3025 0) 0.571 us | page_add_file_rmap();
3026 0) 0.526 us | native_set_pte_at();
3027 0) 0.586 us | _spin_unlock();
3028 0) | unlock_page() {
3029 0) 0.533 us | page_waitqueue();
3030 0) 0.638 us | __wake_up_bit();
3034 You can also expand several functions at once::
3036 echo sys_open > set_graph_function
3037 echo sys_close >> set_graph_function
3039 Now if you want to go back to trace all functions you can clear
3040 this special filter via::
3042 echo > set_graph_function
3048 Note, the proc sysctl ftrace_enable is a big on/off switch for the
3049 function tracer. By default it is enabled (when function tracing is
3050 enabled in the kernel). If it is disabled, all function tracing is
3051 disabled. This includes not only the function tracers for ftrace, but
3052 also for any other uses (perf, kprobes, stack tracing, profiling, etc). It
3053 cannot be disabled if there is a callback with FTRACE_OPS_FL_PERMANENT set
3056 Please disable this with care.
3058 This can be disable (and enabled) with::
3060 sysctl kernel.ftrace_enabled=0
3061 sysctl kernel.ftrace_enabled=1
3065 echo 0 > /proc/sys/kernel/ftrace_enabled
3066 echo 1 > /proc/sys/kernel/ftrace_enabled
3072 A few commands are supported by the set_ftrace_filter interface.
3073 Trace commands have the following format::
3075 <function>:<command>:<parameter>
3077 The following commands are supported:
3080 This command enables function filtering per module. The
3081 parameter defines the module. For example, if only the write*
3082 functions in the ext3 module are desired, run:
3084 echo 'write*:mod:ext3' > set_ftrace_filter
3086 This command interacts with the filter in the same way as
3087 filtering based on function names. Thus, adding more functions
3088 in a different module is accomplished by appending (>>) to the
3089 filter file. Remove specific module functions by prepending
3092 echo '!writeback*:mod:ext3' >> set_ftrace_filter
3094 Mod command supports module globbing. Disable tracing for all
3095 functions except a specific module::
3097 echo '!*:mod:!ext3' >> set_ftrace_filter
3099 Disable tracing for all modules, but still trace kernel::
3101 echo '!*:mod:*' >> set_ftrace_filter
3103 Enable filter only for kernel::
3105 echo '*write*:mod:!*' >> set_ftrace_filter
3107 Enable filter for module globbing::
3109 echo '*write*:mod:*snd*' >> set_ftrace_filter
3112 These commands turn tracing on and off when the specified
3113 functions are hit. The parameter determines how many times the
3114 tracing system is turned on and off. If unspecified, there is
3115 no limit. For example, to disable tracing when a schedule bug
3116 is hit the first 5 times, run::
3118 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
3120 To always disable tracing when __schedule_bug is hit::
3122 echo '__schedule_bug:traceoff' > set_ftrace_filter
3124 These commands are cumulative whether or not they are appended
3125 to set_ftrace_filter. To remove a command, prepend it by '!'
3126 and drop the parameter::
3128 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
3130 The above removes the traceoff command for __schedule_bug
3131 that have a counter. To remove commands without counters::
3133 echo '!__schedule_bug:traceoff' > set_ftrace_filter
3136 Will cause a snapshot to be triggered when the function is hit.
3139 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
3141 To only snapshot once:
3144 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
3146 To remove the above commands::
3148 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
3149 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
3151 - enable_event/disable_event:
3152 These commands can enable or disable a trace event. Note, because
3153 function tracing callbacks are very sensitive, when these commands
3154 are registered, the trace point is activated, but disabled in
3155 a "soft" mode. That is, the tracepoint will be called, but
3156 just will not be traced. The event tracepoint stays in this mode
3157 as long as there's a command that triggers it.
3160 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
3165 <function>:enable_event:<system>:<event>[:count]
3166 <function>:disable_event:<system>:<event>[:count]
3168 To remove the events commands::
3170 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
3172 echo '!schedule:disable_event:sched:sched_switch' > \
3176 When the function is hit, it will dump the contents of the ftrace
3177 ring buffer to the console. This is useful if you need to debug
3178 something, and want to dump the trace when a certain function
3179 is hit. Perhaps it's a function that is called before a triple
3180 fault happens and does not allow you to get a regular dump.
3183 When the function is hit, it will dump the contents of the ftrace
3184 ring buffer for the current CPU to the console. Unlike the "dump"
3185 command, it only prints out the contents of the ring buffer for the
3186 CPU that executed the function that triggered the dump.
3189 When the function is hit, a stack trace is recorded.
3194 The trace_pipe outputs the same content as the trace file, but
3195 the effect on the tracing is different. Every read from
3196 trace_pipe is consumed. This means that subsequent reads will be
3197 different. The trace is live.
3200 # echo function > current_tracer
3201 # cat trace_pipe > /tmp/trace.out &
3203 # echo 1 > tracing_on
3205 # echo 0 > tracing_on
3209 # entries-in-buffer/entries-written: 0/0 #P:4
3212 # / _----=> need-resched
3213 # | / _---=> hardirq/softirq
3214 # || / _--=> preempt-depth
3216 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3220 # cat /tmp/trace.out
3221 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
3222 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
3223 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
3224 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
3225 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
3226 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
3227 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
3228 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
3229 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
3232 Note, reading the trace_pipe file will block until more input is
3233 added. This is contrary to the trace file. If any process opened
3234 the trace file for reading, it will actually disable tracing and
3235 prevent new entries from being added. The trace_pipe file does
3236 not have this limitation.
3241 Having too much or not enough data can be troublesome in
3242 diagnosing an issue in the kernel. The file buffer_size_kb is
3243 used to modify the size of the internal trace buffers. The
3244 number listed is the number of entries that can be recorded per
3245 CPU. To know the full size, multiply the number of possible CPUs
3246 with the number of entries.
3249 # cat buffer_size_kb
3250 1408 (units kilobytes)
3252 Or simply read buffer_total_size_kb
3255 # cat buffer_total_size_kb
3258 To modify the buffer, simple echo in a number (in 1024 byte segments).
3261 # echo 10000 > buffer_size_kb
3262 # cat buffer_size_kb
3263 10000 (units kilobytes)
3265 It will try to allocate as much as possible. If you allocate too
3266 much, it can cause Out-Of-Memory to trigger.
3269 # echo 1000000000000 > buffer_size_kb
3270 -bash: echo: write error: Cannot allocate memory
3271 # cat buffer_size_kb
3274 The per_cpu buffers can be changed individually as well:
3277 # echo 10000 > per_cpu/cpu0/buffer_size_kb
3278 # echo 100 > per_cpu/cpu1/buffer_size_kb
3280 When the per_cpu buffers are not the same, the buffer_size_kb
3281 at the top level will just show an X
3284 # cat buffer_size_kb
3287 This is where the buffer_total_size_kb is useful:
3290 # cat buffer_total_size_kb
3293 Writing to the top level buffer_size_kb will reset all the buffers
3294 to be the same again.
3298 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
3299 available to all non latency tracers. (Latency tracers which
3300 record max latency, such as "irqsoff" or "wakeup", can't use
3301 this feature, since those are already using the snapshot
3302 mechanism internally.)
3304 Snapshot preserves a current trace buffer at a particular point
3305 in time without stopping tracing. Ftrace swaps the current
3306 buffer with a spare buffer, and tracing continues in the new
3307 current (=previous spare) buffer.
3309 The following tracefs files in "tracing" are related to this
3314 This is used to take a snapshot and to read the output
3315 of the snapshot. Echo 1 into this file to allocate a
3316 spare buffer and to take a snapshot (swap), then read
3317 the snapshot from this file in the same format as
3318 "trace" (described above in the section "The File
3319 System"). Both reads snapshot and tracing are executable
3320 in parallel. When the spare buffer is allocated, echoing
3321 0 frees it, and echoing else (positive) values clear the
3323 More details are shown in the table below.
3325 +--------------+------------+------------+------------+
3326 |status\\input | 0 | 1 | else |
3327 +==============+============+============+============+
3328 |not allocated |(do nothing)| alloc+swap |(do nothing)|
3329 +--------------+------------+------------+------------+
3330 |allocated | free | swap | clear |
3331 +--------------+------------+------------+------------+
3333 Here is an example of using the snapshot feature.
3336 # echo 1 > events/sched/enable
3341 # entries-in-buffer/entries-written: 71/71 #P:8
3344 # / _----=> need-resched
3345 # | / _---=> hardirq/softirq
3346 # || / _--=> preempt-depth
3348 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3350 <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
3351 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
3353 <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
3358 # entries-in-buffer/entries-written: 77/77 #P:8
3361 # / _----=> need-resched
3362 # | / _---=> hardirq/softirq
3363 # || / _--=> preempt-depth
3365 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3367 <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
3368 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
3372 If you try to use this snapshot feature when current tracer is
3373 one of the latency tracers, you will get the following results.
3376 # echo wakeup > current_tracer
3378 bash: echo: write error: Device or resource busy
3380 cat: snapshot: Device or resource busy
3385 In the tracefs tracing directory, there is a directory called "instances".
3386 This directory can have new directories created inside of it using
3387 mkdir, and removing directories with rmdir. The directory created
3388 with mkdir in this directory will already contain files and other
3389 directories after it is created.
3392 # mkdir instances/foo
3394 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
3395 set_event snapshot trace trace_clock trace_marker trace_options
3396 trace_pipe tracing_on
3398 As you can see, the new directory looks similar to the tracing directory
3399 itself. In fact, it is very similar, except that the buffer and
3400 events are agnostic from the main directory, or from any other
3401 instances that are created.
3403 The files in the new directory work just like the files with the
3404 same name in the tracing directory except the buffer that is used
3405 is a separate and new buffer. The files affect that buffer but do not
3406 affect the main buffer with the exception of trace_options. Currently,
3407 the trace_options affect all instances and the top level buffer
3408 the same, but this may change in future releases. That is, options
3409 may become specific to the instance they reside in.
3411 Notice that none of the function tracer files are there, nor is
3412 current_tracer and available_tracers. This is because the buffers
3413 can currently only have events enabled for them.
3416 # mkdir instances/foo
3417 # mkdir instances/bar
3418 # mkdir instances/zoot
3419 # echo 100000 > buffer_size_kb
3420 # echo 1000 > instances/foo/buffer_size_kb
3421 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
3422 # echo function > current_trace
3423 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
3424 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
3425 # echo 1 > instances/foo/events/sched/sched_switch/enable
3426 # echo 1 > instances/bar/events/irq/enable
3427 # echo 1 > instances/zoot/events/syscalls/enable
3429 CPU:2 [LOST 11745 EVENTS]
3430 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
3431 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
3432 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
3433 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
3434 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
3435 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
3436 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
3437 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
3438 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3439 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3440 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
3443 # cat instances/foo/trace_pipe
3444 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3445 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3446 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
3447 <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
3448 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
3449 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3450 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3451 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
3452 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
3453 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
3456 # cat instances/bar/trace_pipe
3457 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
3458 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
3459 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
3460 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
3461 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
3462 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
3463 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
3464 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
3465 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
3466 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
3467 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
3468 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
3471 # cat instances/zoot/trace
3474 # entries-in-buffer/entries-written: 18996/18996 #P:4
3477 # / _----=> need-resched
3478 # | / _---=> hardirq/softirq
3479 # || / _--=> preempt-depth
3481 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3483 bash-1998 [000] d... 140.733501: sys_write -> 0x2
3484 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
3485 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
3486 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
3487 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
3488 bash-1998 [000] d... 140.733510: sys_close(fd: a)
3489 bash-1998 [000] d... 140.733510: sys_close -> 0x0
3490 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
3491 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
3492 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
3493 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
3495 You can see that the trace of the top most trace buffer shows only
3496 the function tracing. The foo instance displays wakeups and task
3499 To remove the instances, simply delete their directories:
3502 # rmdir instances/foo
3503 # rmdir instances/bar
3504 # rmdir instances/zoot
3506 Note, if a process has a trace file open in one of the instance
3507 directories, the rmdir will fail with EBUSY.
3512 Since the kernel has a fixed sized stack, it is important not to
3513 waste it in functions. A kernel developer must be conscience of
3514 what they allocate on the stack. If they add too much, the system
3515 can be in danger of a stack overflow, and corruption will occur,
3516 usually leading to a system panic.
3518 There are some tools that check this, usually with interrupts
3519 periodically checking usage. But if you can perform a check
3520 at every function call that will become very useful. As ftrace provides
3521 a function tracer, it makes it convenient to check the stack size
3522 at every function call. This is enabled via the stack tracer.
3524 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
3525 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
3528 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
3530 You can also enable it from the kernel command line to trace
3531 the stack size of the kernel during boot up, by adding "stacktrace"
3532 to the kernel command line parameter.
3534 After running it for a few minutes, the output looks like:
3537 # cat stack_max_size
3541 Depth Size Location (18 entries)
3543 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
3544 1) 2704 160 find_busiest_group+0x31/0x1f1
3545 2) 2544 256 load_balance+0xd9/0x662
3546 3) 2288 80 idle_balance+0xbb/0x130
3547 4) 2208 128 __schedule+0x26e/0x5b9
3548 5) 2080 16 schedule+0x64/0x66
3549 6) 2064 128 schedule_timeout+0x34/0xe0
3550 7) 1936 112 wait_for_common+0x97/0xf1
3551 8) 1824 16 wait_for_completion+0x1d/0x1f
3552 9) 1808 128 flush_work+0xfe/0x119
3553 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
3554 11) 1664 48 input_available_p+0x1d/0x5c
3555 12) 1616 48 n_tty_poll+0x6d/0x134
3556 13) 1568 64 tty_poll+0x64/0x7f
3557 14) 1504 880 do_select+0x31e/0x511
3558 15) 624 400 core_sys_select+0x177/0x216
3559 16) 224 96 sys_select+0x91/0xb9
3560 17) 128 128 system_call_fastpath+0x16/0x1b
3562 Note, if -mfentry is being used by gcc, functions get traced before
3563 they set up the stack frame. This means that leaf level functions
3564 are not tested by the stack tracer when -mfentry is used.
3566 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
3570 More details can be found in the source code, in the `kernel/trace/*.c` files.
git.samba.org / sfrench / cifs-2.6.git / blob