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 frame work 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 Through out 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.txt for more information.
40 Implementation Details
41 ----------------------
43 See :doc:`ftrace-design` 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
102 This holds the different types of tracers that
103 have been compiled into the kernel. The
104 tracers listed here can be configured by
105 echoing their name into current_tracer.
109 This sets or displays whether writing to the trace
110 ring buffer is enabled. Echo 0 into this file to disable
111 the tracer or 1 to enable it. Note, this only disables
112 writing to the ring buffer, the tracing overhead may
115 The kernel function tracing_off() can be used within the
116 kernel to disable writing to the ring buffer, which will
117 set this file to "0". User space can re-enable tracing by
118 echoing "1" into the file.
120 Note, the function and event trigger "traceoff" will also
121 set this file to zero and stop tracing. Which can also
122 be re-enabled by user space using this file.
126 This file holds the output of the trace in a human
127 readable format (described below). Note, tracing is temporarily
128 disabled while this file is being read (opened).
132 The output is the same as the "trace" file but this
133 file is meant to be streamed with live tracing.
134 Reads from this file will block until new data is
135 retrieved. Unlike the "trace" file, this file is a
136 consumer. This means reading from this file causes
137 sequential reads to display more current data. Once
138 data is read from this file, it is consumed, and
139 will not be read again with a sequential read. The
140 "trace" file is static, and if the tracer is not
141 adding more data, it will display the same
142 information every time it is read. This file will not
143 disable tracing while being read.
147 This file lets the user control the amount of data
148 that is displayed in one of the above output
149 files. Options also exist to modify how a tracer
150 or events work (stack traces, timestamps, etc).
154 This is a directory that has a file for every available
155 trace option (also in trace_options). Options may also be set
156 or cleared by writing a "1" or "0" respectively into the
157 corresponding file with the option name.
161 Some of the tracers record the max latency.
162 For example, the maximum time that interrupts are disabled.
163 The maximum time is saved in this file. The max trace will also be
164 stored, and displayed by "trace". A new max trace will only be
165 recorded if the latency is greater than the value in this file
168 By echoing in a time into this file, no latency will be recorded
169 unless it is greater than the time in this file.
173 Some latency tracers will record a trace whenever the
174 latency is greater than the number in this file.
175 Only active when the file contains a number greater than 0.
180 This sets or displays the number of kilobytes each CPU
181 buffer holds. By default, the trace buffers are the same size
182 for each CPU. The displayed number is the size of the
183 CPU buffer and not total size of all buffers. The
184 trace buffers are allocated in pages (blocks of memory
185 that the kernel uses for allocation, usually 4 KB in size).
186 If the last page allocated has room for more bytes
187 than requested, the rest of the page will be used,
188 making the actual allocation bigger than requested or shown.
189 ( Note, the size may not be a multiple of the page size
190 due to buffer management meta-data. )
192 Buffer sizes for individual CPUs may vary
193 (see "per_cpu/cpu0/buffer_size_kb" below), and if they do
194 this file will show "X".
196 buffer_total_size_kb:
198 This displays the total combined size of all the trace buffers.
202 If a process is performing tracing, and the ring buffer should be
203 shrunk "freed" when the process is finished, even if it were to be
204 killed by a signal, this file can be used for that purpose. On close
205 of this file, the ring buffer will be resized to its minimum size.
206 Having a process that is tracing also open this file, when the process
207 exits its file descriptor for this file will be closed, and in doing so,
208 the ring buffer will be "freed".
210 It may also stop tracing if disable_on_free option is set.
214 This is a mask that lets the user only trace on specified CPUs.
215 The format is a hex string representing the CPUs.
219 When dynamic ftrace is configured in (see the
220 section below "dynamic ftrace"), the code is dynamically
221 modified (code text rewrite) to disable calling of the
222 function profiler (mcount). This lets tracing be configured
223 in with practically no overhead in performance. This also
224 has a side effect of enabling or disabling specific functions
225 to be traced. Echoing names of functions into this file
226 will limit the trace to only those functions.
228 The functions listed in "available_filter_functions" are what
229 can be written into this file.
231 This interface also allows for commands to be used. See the
232 "Filter commands" section for more details.
236 This has an effect opposite to that of
237 set_ftrace_filter. Any function that is added here will not
238 be traced. If a function exists in both set_ftrace_filter
239 and set_ftrace_notrace, the function will _not_ be traced.
243 Have the function tracer only trace the threads whose PID are
246 If the "function-fork" option is set, then when a task whose
247 PID is listed in this file forks, the child's PID will
248 automatically be added to this file, and the child will be
249 traced by the function tracer as well. This option will also
250 cause PIDs of tasks that exit to be removed from the file.
254 Have the events only trace a task with a PID listed in this file.
255 Note, sched_switch and sched_wake_up will also trace events
258 To have the PIDs of children of tasks with their PID in this file
259 added on fork, enable the "event-fork" option. That option will also
260 cause the PIDs of tasks to be removed from this file when the task
265 Functions listed in this file will cause the function graph
266 tracer to only trace these functions and the functions that
267 they call. (See the section "dynamic ftrace" for more details).
271 Similar to set_graph_function, but will disable function graph
272 tracing when the function is hit until it exits the function.
273 This makes it possible to ignore tracing functions that are called
274 by a specific function.
276 available_filter_functions:
278 This lists the functions that ftrace has processed and can trace.
279 These are the function names that you can pass to
280 "set_ftrace_filter" or "set_ftrace_notrace".
281 (See the section "dynamic ftrace" below for more details.)
283 dyn_ftrace_total_info:
285 This file is for debugging purposes. The number of functions that
286 have been converted to nops and are available to be traced.
290 This file is more for debugging ftrace, but can also be useful
291 in seeing if any function has a callback attached to it.
292 Not only does the trace infrastructure use ftrace function
293 trace utility, but other subsystems might too. This file
294 displays all functions that have a callback attached to them
295 as well as the number of callbacks that have been attached.
296 Note, a callback may also call multiple functions which will
297 not be listed in this count.
299 If the callback registered to be traced by a function with
300 the "save regs" attribute (thus even more overhead), a 'R'
301 will be displayed on the same line as the function that
302 is returning registers.
304 If the callback registered to be traced by a function with
305 the "ip modify" attribute (thus the regs->ip can be changed),
306 an 'I' will be displayed on the same line as the function that
309 If the architecture supports it, it will also show what callback
310 is being directly called by the function. If the count is greater
311 than 1 it most likely will be ftrace_ops_list_func().
313 If the callback of the function jumps to a trampoline that is
314 specific to a the callback and not the standard trampoline,
315 its address will be printed as well as the function that the
318 function_profile_enabled:
320 When set it will enable all functions with either the function
321 tracer, or if configured, the function graph tracer. It will
322 keep a histogram of the number of functions that were called
323 and if the function graph tracer was configured, it will also keep
324 track of the time spent in those functions. The histogram
325 content can be displayed in the files:
327 trace_stats/function<cpu> ( function0, function1, etc).
331 A directory that holds different tracing stats.
335 Enable dynamic trace points. See kprobetrace.txt.
339 Dynamic trace points stats. See kprobetrace.txt.
343 Used with the function graph tracer. This is the max depth
344 it will trace into a function. Setting this to a value of
345 one will show only the first kernel function that is called
350 This is for tools that read the raw format files. If an event in
351 the ring buffer references a string, only a pointer to the string
352 is recorded into the buffer and not the string itself. This prevents
353 tools from knowing what that string was. This file displays the string
354 and address for the string allowing tools to map the pointers to what
359 Only the pid of the task is recorded in a trace event unless
360 the event specifically saves the task comm as well. Ftrace
361 makes a cache of pid mappings to comms to try to display
362 comms for events. If a pid for a comm is not listed, then
363 "<...>" is displayed in the output.
365 If the option "record-cmd" is set to "0", then comms of tasks
366 will not be saved during recording. By default, it is enabled.
370 By default, 128 comms are saved (see "saved_cmdlines" above). To
371 increase or decrease the amount of comms that are cached, echo
372 in a the number of comms to cache, into this file.
376 If the option "record-tgid" is set, on each scheduling context switch
377 the Task Group ID of a task is saved in a table mapping the PID of
378 the thread to its TGID. By default, the "record-tgid" option is
383 This displays the "snapshot" buffer and also lets the user
384 take a snapshot of the current running trace.
385 See the "Snapshot" section below for more details.
389 When the stack tracer is activated, this will display the
390 maximum stack size it has encountered.
391 See the "Stack Trace" section below.
395 This displays the stack back trace of the largest stack
396 that was encountered when the stack tracer is activated.
397 See the "Stack Trace" section below.
401 This is similar to "set_ftrace_filter" but it limits what
402 functions the stack tracer will check.
406 Whenever an event is recorded into the ring buffer, a
407 "timestamp" is added. This stamp comes from a specified
408 clock. By default, ftrace uses the "local" clock. This
409 clock is very fast and strictly per cpu, but on some
410 systems it may not be monotonic with respect to other
411 CPUs. In other words, the local clocks may not be in sync
412 with local clocks on other CPUs.
414 Usual clocks for tracing::
417 [local] global counter x86-tsc
419 The clock with the square brackets around it is the one in effect.
422 Default clock, but may not be in sync across CPUs
425 This clock is in sync with all CPUs but may
426 be a bit slower than the local clock.
429 This is not a clock at all, but literally an atomic
430 counter. It counts up one by one, but is in sync
431 with all CPUs. This is useful when you need to
432 know exactly the order events occurred with respect to
433 each other on different CPUs.
436 This uses the jiffies counter and the time stamp
437 is relative to the time since boot up.
440 This makes ftrace use the same clock that perf uses.
441 Eventually perf will be able to read ftrace buffers
442 and this will help out in interleaving the data.
445 Architectures may define their own clocks. For
446 example, x86 uses its own TSC cycle clock here.
449 This uses the powerpc timebase register value.
450 This is in sync across CPUs and can also be used
451 to correlate events across hypervisor/guest if
455 This uses the fast monotonic clock (CLOCK_MONOTONIC)
456 which is monotonic and is subject to NTP rate adjustments.
459 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
460 which is montonic but is not subject to any rate adjustments
461 and ticks at the same rate as the hardware clocksource.
464 Same as mono. Used to be a separate clock which accounted
465 for the time spent in suspend while CLOCK_MONOTONIC did
468 To set a clock, simply echo the clock name into this file::
470 # echo global > trace_clock
474 This is a very useful file for synchronizing user space
475 with events happening in the kernel. Writing strings into
476 this file will be written into the ftrace buffer.
478 It is useful in applications to open this file at the start
479 of the application and just reference the file descriptor
482 void trace_write(const char *fmt, ...)
492 n = vsnprintf(buf, 256, fmt, ap);
495 write(trace_fd, buf, n);
500 trace_fd = open("trace_marker", WR_ONLY);
504 This is similar to trace_marker above, but is meant for for binary data
505 to be written to it, where a tool can be used to parse the data
510 Add dynamic tracepoints in programs.
515 Uprobe statistics. See uprobetrace.txt
519 This is a way to make multiple trace buffers where different
520 events can be recorded in different buffers.
521 See "Instances" section below.
525 This is the trace event directory. It holds event tracepoints
526 (also known as static tracepoints) that have been compiled
527 into the kernel. It shows what event tracepoints exist
528 and how they are grouped by system. There are "enable"
529 files at various levels that can enable the tracepoints
530 when a "1" is written to them.
532 See events.txt for more information.
536 By echoing in the event into this file, will enable that event.
538 See events.txt for more information.
542 A list of events that can be enabled in tracing.
544 See events.txt for more information.
548 Certain tracers may change the timestamp mode used when
549 logging trace events into the event buffer. Events with
550 different modes can coexist within a buffer but the mode in
551 effect when an event is logged determines which timestamp mode
552 is used for that event. The default timestamp mode is
555 Usual timestamp modes for tracing:
560 The timestamp mode with the square brackets around it is the
563 delta: Default timestamp mode - timestamp is a delta against
564 a per-buffer timestamp.
566 absolute: The timestamp is a full timestamp, not a delta
567 against some other value. As such it takes up more
568 space and is less efficient.
572 Directory for the Hardware Latency Detector.
573 See "Hardware Latency Detector" section below.
577 This is a directory that contains the trace per_cpu information.
579 per_cpu/cpu0/buffer_size_kb:
581 The ftrace buffer is defined per_cpu. That is, there's a separate
582 buffer for each CPU to allow writes to be done atomically,
583 and free from cache bouncing. These buffers may have different
584 size buffers. This file is similar to the buffer_size_kb
585 file, but it only displays or sets the buffer size for the
586 specific CPU. (here cpu0).
590 This is similar to the "trace" file, but it will only display
591 the data specific for the CPU. If written to, it only clears
592 the specific CPU buffer.
594 per_cpu/cpu0/trace_pipe
596 This is similar to the "trace_pipe" file, and is a consuming
597 read, but it will only display (and consume) the data specific
600 per_cpu/cpu0/trace_pipe_raw
602 For tools that can parse the ftrace ring buffer binary format,
603 the trace_pipe_raw file can be used to extract the data
604 from the ring buffer directly. With the use of the splice()
605 system call, the buffer data can be quickly transferred to
606 a file or to the network where a server is collecting the
609 Like trace_pipe, this is a consuming reader, where multiple
610 reads will always produce different data.
612 per_cpu/cpu0/snapshot:
614 This is similar to the main "snapshot" file, but will only
615 snapshot the current CPU (if supported). It only displays
616 the content of the snapshot for a given CPU, and if
617 written to, only clears this CPU buffer.
619 per_cpu/cpu0/snapshot_raw:
621 Similar to the trace_pipe_raw, but will read the binary format
622 from the snapshot buffer for the given CPU.
626 This displays certain stats about the ring buffer:
629 The number of events that are still in the buffer.
632 The number of lost events due to overwriting when
636 Should always be zero.
637 This gets set if so many events happened within a nested
638 event (ring buffer is re-entrant), that it fills the
639 buffer and starts dropping events.
642 Bytes actually read (not overwritten).
645 The oldest timestamp in the buffer
648 The current timestamp
651 Events lost due to overwrite option being off.
654 The number of events read.
659 Here is the list of current tracers that may be configured.
663 Function call tracer to trace all kernel functions.
667 Similar to the function tracer except that the
668 function tracer probes the functions on their entry
669 whereas the function graph tracer traces on both entry
670 and exit of the functions. It then provides the ability
671 to draw a graph of function calls similar to C code
676 The block tracer. The tracer used by the blktrace user
681 The Hardware Latency tracer is used to detect if the hardware
682 produces any latency. See "Hardware Latency Detector" section
687 Traces the areas that disable interrupts and saves
688 the trace with the longest max latency.
689 See tracing_max_latency. When a new max is recorded,
690 it replaces the old trace. It is best to view this
691 trace with the latency-format option enabled, which
692 happens automatically when the tracer is selected.
696 Similar to irqsoff but traces and records the amount of
697 time for which preemption is disabled.
701 Similar to irqsoff and preemptoff, but traces and
702 records the largest time for which irqs and/or preemption
707 Traces and records the max latency that it takes for
708 the highest priority task to get scheduled after
709 it has been woken up.
710 Traces all tasks as an average developer would expect.
714 Traces and records the max latency that it takes for just
715 RT tasks (as the current "wakeup" does). This is useful
716 for those interested in wake up timings of RT tasks.
720 Traces and records the max latency that it takes for
721 a SCHED_DEADLINE task to be woken (as the "wakeup" and
726 A special tracer that is used to trace binary module.
727 It will trace all the calls that a module makes to the
728 hardware. Everything it writes and reads from the I/O
733 This tracer can be configured when tracing likely/unlikely
734 calls within the kernel. It will trace when a likely and
735 unlikely branch is hit and if it was correct in its prediction
740 This is the "trace nothing" tracer. To remove all
741 tracers from tracing simply echo "nop" into
745 Examples of using the tracer
746 ----------------------------
748 Here are typical examples of using the tracers when controlling
749 them only with the tracefs interface (without using any
750 user-land utilities).
755 Here is an example of the output format of the file "trace"::
759 # entries-in-buffer/entries-written: 140080/250280 #P:4
762 # / _----=> need-resched
763 # | / _---=> hardirq/softirq
764 # || / _--=> preempt-depth
766 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
768 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
769 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
770 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
771 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
772 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
773 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
774 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
775 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
776 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
777 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
780 A header is printed with the tracer name that is represented by
781 the trace. In this case the tracer is "function". Then it shows the
782 number of events in the buffer as well as the total number of entries
783 that were written. The difference is the number of entries that were
784 lost due to the buffer filling up (250280 - 140080 = 110200 events
787 The header explains the content of the events. Task name "bash", the task
788 PID "1977", the CPU that it was running on "000", the latency format
789 (explained below), the timestamp in <secs>.<usecs> format, the
790 function name that was traced "sys_close" and the parent function that
791 called this function "system_call_fastpath". The timestamp is the time
792 at which the function was entered.
797 When the latency-format option is enabled or when one of the latency
798 tracers is set, the trace file gives somewhat more information to see
799 why a latency happened. Here is a typical trace::
803 # irqsoff latency trace v1.1.5 on 3.8.0-test+
804 # --------------------------------------------------------------------
805 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
807 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
809 # => started at: __lock_task_sighand
810 # => ended at: _raw_spin_unlock_irqrestore
814 # / _-----=> irqs-off
815 # | / _----=> need-resched
816 # || / _---=> hardirq/softirq
817 # ||| / _--=> preempt-depth
819 # cmd pid ||||| time | caller
821 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
822 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
823 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
824 ps-6143 2d..1 306us : <stack trace>
825 => trace_hardirqs_on_caller
827 => _raw_spin_unlock_irqrestore
834 => system_call_fastpath
837 This shows that the current tracer is "irqsoff" tracing the time
838 for which interrupts were disabled. It gives the trace version (which
839 never changes) and the version of the kernel upon which this was executed on
840 (3.8). Then it displays the max latency in microseconds (259 us). The number
841 of trace entries displayed and the total number (both are four: #4/4).
842 VP, KP, SP, and HP are always zero and are reserved for later use.
843 #P is the number of online CPUs (#P:4).
845 The task is the process that was running when the latency
846 occurred. (ps pid: 6143).
848 The start and stop (the functions in which the interrupts were
849 disabled and enabled respectively) that caused the latencies:
851 - __lock_task_sighand is where the interrupts were disabled.
852 - _raw_spin_unlock_irqrestore is where they were enabled again.
854 The next lines after the header are the trace itself. The header
855 explains which is which.
857 cmd: The name of the process in the trace.
859 pid: The PID of that process.
861 CPU#: The CPU which the process was running on.
863 irqs-off: 'd' interrupts are disabled. '.' otherwise.
864 .. caution:: If the architecture does not support a way to
865 read the irq flags variable, an 'X' will always
869 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
870 - 'n' only TIF_NEED_RESCHED is set,
871 - 'p' only PREEMPT_NEED_RESCHED is set,
875 - 'Z' - NMI occurred inside a hardirq
876 - 'z' - NMI is running
877 - 'H' - hard irq occurred inside a softirq.
878 - 'h' - hard irq is running
879 - 's' - soft irq is running
880 - '.' - normal context.
882 preempt-depth: The level of preempt_disabled
884 The above is mostly meaningful for kernel developers.
887 When the latency-format option is enabled, the trace file
888 output includes a timestamp relative to the start of the
889 trace. This differs from the output when latency-format
890 is disabled, which includes an absolute timestamp.
893 This is just to help catch your eye a bit better. And
894 needs to be fixed to be only relative to the same CPU.
895 The marks are determined by the difference between this
896 current trace and the next trace.
898 - '$' - greater than 1 second
899 - '@' - greater than 100 milisecond
900 - '*' - greater than 10 milisecond
901 - '#' - greater than 1000 microsecond
902 - '!' - greater than 100 microsecond
903 - '+' - greater than 10 microsecond
904 - ' ' - less than or equal to 10 microsecond.
906 The rest is the same as the 'trace' file.
908 Note, the latency tracers will usually end with a back trace
909 to easily find where the latency occurred.
914 The trace_options file (or the options directory) is used to control
915 what gets printed in the trace output, or manipulate the tracers.
916 To see what is available, simply cat the file::
947 To disable one of the options, echo in the option prepended with
950 echo noprint-parent > trace_options
952 To enable an option, leave off the "no"::
954 echo sym-offset > trace_options
956 Here are the available options:
959 On function traces, display the calling (parent)
960 function as well as the function being traced.
964 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
967 bash-4000 [01] 1477.606694: simple_strtoul
971 Display not only the function name, but also the
972 offset in the function. For example, instead of
973 seeing just "ktime_get", you will see
974 "ktime_get+0xb/0x20".
978 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
981 This will also display the function address as well
982 as the function name.
986 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
989 This deals with the trace file when the
990 latency-format option is enabled.
993 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
994 (+0.000ms): simple_strtoul (kstrtoul)
997 This will display raw numbers. This option is best for
998 use with user applications that can translate the raw
999 numbers better than having it done in the kernel.
1002 Similar to raw, but the numbers will be in a hexadecimal format.
1005 This will print out the formats in raw binary.
1008 When set, reading trace_pipe will not block when polled.
1011 Can disable trace_printk() from writing into the buffer.
1014 It is sometimes confusing when the CPU buffers are full
1015 and one CPU buffer had a lot of events recently, thus
1016 a shorter time frame, were another CPU may have only had
1017 a few events, which lets it have older events. When
1018 the trace is reported, it shows the oldest events first,
1019 and it may look like only one CPU ran (the one with the
1020 oldest events). When the annotate option is set, it will
1021 display when a new CPU buffer started::
1023 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
1024 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
1025 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
1026 ##### CPU 2 buffer started ####
1027 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
1028 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
1029 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
1032 This option changes the trace. It records a
1033 stacktrace of the current user space thread after
1037 when user stacktrace are enabled, look up which
1038 object the address belongs to, and print a
1039 relative address. This is especially useful when
1040 ASLR is on, otherwise you don't get a chance to
1041 resolve the address to object/file/line after
1042 the app is no longer running
1044 The lookup is performed when you read
1045 trace,trace_pipe. Example::
1047 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
1048 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
1052 When set, trace_printk()s will only show the format
1053 and not their parameters (if trace_bprintk() or
1054 trace_bputs() was used to save the trace_printk()).
1057 Show only the event data. Hides the comm, PID,
1058 timestamp, CPU, and other useful data.
1061 This option changes the trace output. When it is enabled,
1062 the trace displays additional information about the
1063 latency, as described in "Latency trace format".
1066 When any event or tracer is enabled, a hook is enabled
1067 in the sched_switch trace point to fill comm cache
1068 with mapped pids and comms. But this may cause some
1069 overhead, and if you only care about pids, and not the
1070 name of the task, disabling this option can lower the
1071 impact of tracing. See "saved_cmdlines".
1074 When any event or tracer is enabled, a hook is enabled
1075 in the sched_switch trace point to fill the cache of
1076 mapped Thread Group IDs (TGID) mapping to pids. See
1080 This controls what happens when the trace buffer is
1081 full. If "1" (default), the oldest events are
1082 discarded and overwritten. If "0", then the newest
1083 events are discarded.
1084 (see per_cpu/cpu0/stats for overrun and dropped)
1087 When the free_buffer is closed, tracing will
1088 stop (tracing_on set to 0).
1091 Shows the interrupt, preempt count, need resched data.
1092 When disabled, the trace looks like::
1096 # entries-in-buffer/entries-written: 144405/9452052 #P:4
1098 # TASK-PID CPU# TIMESTAMP FUNCTION
1100 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
1101 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
1102 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
1106 When set, the trace_marker is writable (only by root).
1107 When disabled, the trace_marker will error with EINVAL
1111 When set, tasks with PIDs listed in set_event_pid will have
1112 the PIDs of their children added to set_event_pid when those
1113 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
1114 their PIDs will be removed from the file.
1117 The latency tracers will enable function tracing
1118 if this option is enabled (default it is). When
1119 it is disabled, the latency tracers do not trace
1120 functions. This keeps the overhead of the tracer down
1121 when performing latency tests.
1124 When set, tasks with PIDs listed in set_ftrace_pid will
1125 have the PIDs of their children added to set_ftrace_pid
1126 when those tasks fork. Also, when tasks with PIDs in
1127 set_ftrace_pid exit, their PIDs will be removed from the
1131 When set, the latency tracers (irqsoff, wakeup, etc) will
1132 use function graph tracing instead of function tracing.
1135 When set, a stack trace is recorded after any trace event
1139 Enable branch tracing with the tracer. This enables branch
1140 tracer along with the currently set tracer. Enabling this
1141 with the "nop" tracer is the same as just enabling the
1144 .. tip:: Some tracers have their own options. They only appear in this
1145 file when the tracer is active. They always appear in the
1149 Here are the per tracer options:
1151 Options for function tracer:
1154 When set, a stack trace is recorded after every
1155 function that is recorded. NOTE! Limit the functions
1156 that are recorded before enabling this, with
1157 "set_ftrace_filter" otherwise the system performance
1158 will be critically degraded. Remember to disable
1159 this option before clearing the function filter.
1161 Options for function_graph tracer:
1163 Since the function_graph tracer has a slightly different output
1164 it has its own options to control what is displayed.
1167 When set, the "overrun" of the graph stack is
1168 displayed after each function traced. The
1169 overrun, is when the stack depth of the calls
1170 is greater than what is reserved for each task.
1171 Each task has a fixed array of functions to
1172 trace in the call graph. If the depth of the
1173 calls exceeds that, the function is not traced.
1174 The overrun is the number of functions missed
1175 due to exceeding this array.
1178 When set, the CPU number of the CPU where the trace
1179 occurred is displayed.
1182 When set, if the function takes longer than
1183 A certain amount, then a delay marker is
1184 displayed. See "delay" above, under the
1188 Unlike other tracers, the process' command line
1189 is not displayed by default, but instead only
1190 when a task is traced in and out during a context
1191 switch. Enabling this options has the command
1192 of each process displayed at every line.
1195 At the end of each function (the return)
1196 the duration of the amount of time in the
1197 function is displayed in microseconds.
1200 When set, the timestamp is displayed at each line.
1203 When disabled, functions that happen inside an
1204 interrupt will not be traced.
1207 When set, the return event will include the function
1208 that it represents. By default this is off, and
1209 only a closing curly bracket "}" is displayed for
1210 the return of a function.
1213 When running function graph tracer, to include
1214 the time a task schedules out in its function.
1215 When enabled, it will account time the task has been
1216 scheduled out as part of the function call.
1219 When running function profiler with function graph tracer,
1220 to include the time to call nested functions. When this is
1221 not set, the time reported for the function will only
1222 include the time the function itself executed for, not the
1223 time for functions that it called.
1225 Options for blk tracer:
1228 Shows a more minimalistic output.
1234 When interrupts are disabled, the CPU can not react to any other
1235 external event (besides NMIs and SMIs). This prevents the timer
1236 interrupt from triggering or the mouse interrupt from letting
1237 the kernel know of a new mouse event. The result is a latency
1238 with the reaction time.
1240 The irqsoff tracer tracks the time for which interrupts are
1241 disabled. When a new maximum latency is hit, the tracer saves
1242 the trace leading up to that latency point so that every time a
1243 new maximum is reached, the old saved trace is discarded and the
1246 To reset the maximum, echo 0 into tracing_max_latency. Here is
1249 # echo 0 > options/function-trace
1250 # echo irqsoff > current_tracer
1251 # echo 1 > tracing_on
1252 # echo 0 > tracing_max_latency
1255 # echo 0 > tracing_on
1259 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1260 # --------------------------------------------------------------------
1261 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1263 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
1265 # => started at: run_timer_softirq
1266 # => ended at: run_timer_softirq
1270 # / _-----=> irqs-off
1271 # | / _----=> need-resched
1272 # || / _---=> hardirq/softirq
1273 # ||| / _--=> preempt-depth
1275 # cmd pid ||||| time | caller
1277 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
1278 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
1279 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
1280 <idle>-0 0dNs3 25us : <stack trace>
1281 => _raw_spin_unlock_irq
1282 => run_timer_softirq
1287 => smp_apic_timer_interrupt
1288 => apic_timer_interrupt
1293 => x86_64_start_reservations
1294 => x86_64_start_kernel
1296 Here we see that that we had a latency of 16 microseconds (which is
1297 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
1298 interrupts. The difference between the 16 and the displayed
1299 timestamp 25us occurred because the clock was incremented
1300 between the time of recording the max latency and the time of
1301 recording the function that had that latency.
1303 Note the above example had function-trace not set. If we set
1304 function-trace, we get a much larger output::
1306 with echo 1 > options/function-trace
1310 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1311 # --------------------------------------------------------------------
1312 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1314 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1316 # => started at: ata_scsi_queuecmd
1317 # => ended at: ata_scsi_queuecmd
1321 # / _-----=> irqs-off
1322 # | / _----=> need-resched
1323 # || / _---=> hardirq/softirq
1324 # ||| / _--=> preempt-depth
1326 # cmd pid ||||| time | caller
1328 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1329 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1330 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1331 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1332 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1333 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1334 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1335 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1336 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1338 bash-2042 3d..1 67us : delay_tsc <-__delay
1339 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1340 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1341 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1342 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1343 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1344 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1345 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1346 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1347 bash-2042 3d..1 120us : <stack trace>
1348 => _raw_spin_unlock_irqrestore
1349 => ata_scsi_queuecmd
1350 => scsi_dispatch_cmd
1352 => __blk_run_queue_uncond
1355 => generic_make_request
1358 => __ext3_get_inode_loc
1367 => user_path_at_empty
1372 => system_call_fastpath
1375 Here we traced a 71 microsecond latency. But we also see all the
1376 functions that were called during that time. Note that by
1377 enabling function tracing, we incur an added overhead. This
1378 overhead may extend the latency times. But nevertheless, this
1379 trace has provided some very helpful debugging information.
1385 When preemption is disabled, we may be able to receive
1386 interrupts but the task cannot be preempted and a higher
1387 priority task must wait for preemption to be enabled again
1388 before it can preempt a lower priority task.
1390 The preemptoff tracer traces the places that disable preemption.
1391 Like the irqsoff tracer, it records the maximum latency for
1392 which preemption was disabled. The control of preemptoff tracer
1393 is much like the irqsoff tracer.
1396 # echo 0 > options/function-trace
1397 # echo preemptoff > current_tracer
1398 # echo 1 > tracing_on
1399 # echo 0 > tracing_max_latency
1402 # echo 0 > tracing_on
1404 # tracer: preemptoff
1406 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1407 # --------------------------------------------------------------------
1408 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1410 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1412 # => started at: do_IRQ
1413 # => ended at: do_IRQ
1417 # / _-----=> irqs-off
1418 # | / _----=> need-resched
1419 # || / _---=> hardirq/softirq
1420 # ||| / _--=> preempt-depth
1422 # cmd pid ||||| time | caller
1424 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1425 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1426 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1427 sshd-1991 1d..1 52us : <stack trace>
1428 => sub_preempt_count
1434 This has some more changes. Preemption was disabled when an
1435 interrupt came in (notice the 'h'), and was enabled on exit.
1436 But we also see that interrupts have been disabled when entering
1437 the preempt off section and leaving it (the 'd'). We do not know if
1438 interrupts were enabled in the mean time or shortly after this
1442 # tracer: preemptoff
1444 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1445 # --------------------------------------------------------------------
1446 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1448 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1450 # => started at: wake_up_new_task
1451 # => ended at: task_rq_unlock
1455 # / _-----=> irqs-off
1456 # | / _----=> need-resched
1457 # || / _---=> hardirq/softirq
1458 # ||| / _--=> preempt-depth
1460 # cmd pid ||||| time | caller
1462 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1463 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1464 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1465 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1466 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1468 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1469 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1470 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1471 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1472 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1473 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1474 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1475 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1477 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1478 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1479 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1480 bash-1994 1d..2 36us : do_softirq <-irq_exit
1481 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1482 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1483 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1484 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1485 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1486 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1488 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1489 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1490 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1491 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1492 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1493 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1494 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1495 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1496 bash-1994 1.N.1 104us : <stack trace>
1497 => sub_preempt_count
1498 => _raw_spin_unlock_irqrestore
1506 The above is an example of the preemptoff trace with
1507 function-trace set. Here we see that interrupts were not disabled
1508 the entire time. The irq_enter code lets us know that we entered
1509 an interrupt 'h'. Before that, the functions being traced still
1510 show that it is not in an interrupt, but we can see from the
1511 functions themselves that this is not the case.
1516 Knowing the locations that have interrupts disabled or
1517 preemption disabled for the longest times is helpful. But
1518 sometimes we would like to know when either preemption and/or
1519 interrupts are disabled.
1521 Consider the following code::
1523 local_irq_disable();
1524 call_function_with_irqs_off();
1526 call_function_with_irqs_and_preemption_off();
1528 call_function_with_preemption_off();
1531 The irqsoff tracer will record the total length of
1532 call_function_with_irqs_off() and
1533 call_function_with_irqs_and_preemption_off().
1535 The preemptoff tracer will record the total length of
1536 call_function_with_irqs_and_preemption_off() and
1537 call_function_with_preemption_off().
1539 But neither will trace the time that interrupts and/or
1540 preemption is disabled. This total time is the time that we can
1541 not schedule. To record this time, use the preemptirqsoff
1544 Again, using this trace is much like the irqsoff and preemptoff
1548 # echo 0 > options/function-trace
1549 # echo preemptirqsoff > current_tracer
1550 # echo 1 > tracing_on
1551 # echo 0 > tracing_max_latency
1554 # echo 0 > tracing_on
1556 # tracer: preemptirqsoff
1558 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1559 # --------------------------------------------------------------------
1560 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1562 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1564 # => started at: ata_scsi_queuecmd
1565 # => ended at: ata_scsi_queuecmd
1569 # / _-----=> irqs-off
1570 # | / _----=> need-resched
1571 # || / _---=> hardirq/softirq
1572 # ||| / _--=> preempt-depth
1574 # cmd pid ||||| time | caller
1576 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1577 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1578 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1579 ls-2230 3...1 111us : <stack trace>
1580 => sub_preempt_count
1581 => _raw_spin_unlock_irqrestore
1582 => ata_scsi_queuecmd
1583 => scsi_dispatch_cmd
1585 => __blk_run_queue_uncond
1588 => generic_make_request
1593 => htree_dirblock_to_tree
1594 => ext3_htree_fill_tree
1598 => system_call_fastpath
1601 The trace_hardirqs_off_thunk is called from assembly on x86 when
1602 interrupts are disabled in the assembly code. Without the
1603 function tracing, we do not know if interrupts were enabled
1604 within the preemption points. We do see that it started with
1607 Here is a trace with function-trace set::
1609 # tracer: preemptirqsoff
1611 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1612 # --------------------------------------------------------------------
1613 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1615 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1617 # => started at: schedule
1618 # => ended at: mutex_unlock
1622 # / _-----=> irqs-off
1623 # | / _----=> need-resched
1624 # || / _---=> hardirq/softirq
1625 # ||| / _--=> preempt-depth
1627 # cmd pid ||||| time | caller
1629 kworker/-59 3...1 0us : __schedule <-schedule
1630 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1631 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1632 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1633 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1634 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1635 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1636 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1637 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1638 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1639 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1640 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1641 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1642 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1643 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1644 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1645 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1646 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1647 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1648 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1649 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1650 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1651 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1652 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1653 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1654 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1655 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1656 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1657 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1658 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1659 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1660 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1662 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1663 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1664 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1665 ls-2269 3d..3 21us : do_softirq <-irq_exit
1666 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1667 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1668 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1669 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1670 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1671 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1672 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1674 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1675 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1676 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1677 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1678 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1679 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1681 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1682 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1683 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1684 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1685 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1686 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1687 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1688 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1689 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1690 ls-2269 3d... 186us : <stack trace>
1691 => __mutex_unlock_slowpath
1698 => system_call_fastpath
1700 This is an interesting trace. It started with kworker running and
1701 scheduling out and ls taking over. But as soon as ls released the
1702 rq lock and enabled interrupts (but not preemption) an interrupt
1703 triggered. When the interrupt finished, it started running softirqs.
1704 But while the softirq was running, another interrupt triggered.
1705 When an interrupt is running inside a softirq, the annotation is 'H'.
1711 One common case that people are interested in tracing is the
1712 time it takes for a task that is woken to actually wake up.
1713 Now for non Real-Time tasks, this can be arbitrary. But tracing
1714 it none the less can be interesting.
1716 Without function tracing::
1718 # echo 0 > options/function-trace
1719 # echo wakeup > current_tracer
1720 # echo 1 > tracing_on
1721 # echo 0 > tracing_max_latency
1723 # echo 0 > tracing_on
1727 # wakeup latency trace v1.1.5 on 3.8.0-test+
1728 # --------------------------------------------------------------------
1729 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1731 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1735 # / _-----=> irqs-off
1736 # | / _----=> need-resched
1737 # || / _---=> hardirq/softirq
1738 # ||| / _--=> preempt-depth
1740 # cmd pid ||||| time | caller
1742 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1743 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1744 <idle>-0 3d..3 15us : __schedule <-schedule
1745 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1747 The tracer only traces the highest priority task in the system
1748 to avoid tracing the normal circumstances. Here we see that
1749 the kworker with a nice priority of -20 (not very nice), took
1750 just 15 microseconds from the time it woke up, to the time it
1753 Non Real-Time tasks are not that interesting. A more interesting
1754 trace is to concentrate only on Real-Time tasks.
1759 In a Real-Time environment it is very important to know the
1760 wakeup time it takes for the highest priority task that is woken
1761 up to the time that it executes. This is also known as "schedule
1762 latency". I stress the point that this is about RT tasks. It is
1763 also important to know the scheduling latency of non-RT tasks,
1764 but the average schedule latency is better for non-RT tasks.
1765 Tools like LatencyTop are more appropriate for such
1768 Real-Time environments are interested in the worst case latency.
1769 That is the longest latency it takes for something to happen,
1770 and not the average. We can have a very fast scheduler that may
1771 only have a large latency once in a while, but that would not
1772 work well with Real-Time tasks. The wakeup_rt tracer was designed
1773 to record the worst case wakeups of RT tasks. Non-RT tasks are
1774 not recorded because the tracer only records one worst case and
1775 tracing non-RT tasks that are unpredictable will overwrite the
1776 worst case latency of RT tasks (just run the normal wakeup
1777 tracer for a while to see that effect).
1779 Since this tracer only deals with RT tasks, we will run this
1780 slightly differently than we did with the previous tracers.
1781 Instead of performing an 'ls', we will run 'sleep 1' under
1782 'chrt' which changes the priority of the task.
1785 # echo 0 > options/function-trace
1786 # echo wakeup_rt > current_tracer
1787 # echo 1 > tracing_on
1788 # echo 0 > tracing_max_latency
1790 # echo 0 > tracing_on
1796 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1797 # --------------------------------------------------------------------
1798 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1800 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1804 # / _-----=> irqs-off
1805 # | / _----=> need-resched
1806 # || / _---=> hardirq/softirq
1807 # ||| / _--=> preempt-depth
1809 # cmd pid ||||| time | caller
1811 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1812 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1813 <idle>-0 3d..3 5us : __schedule <-schedule
1814 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1817 Running this on an idle system, we see that it only took 5 microseconds
1818 to perform the task switch. Note, since the trace point in the schedule
1819 is before the actual "switch", we stop the tracing when the recorded task
1820 is about to schedule in. This may change if we add a new marker at the
1821 end of the scheduler.
1823 Notice that the recorded task is 'sleep' with the PID of 2389
1824 and it has an rt_prio of 5. This priority is user-space priority
1825 and not the internal kernel priority. The policy is 1 for
1826 SCHED_FIFO and 2 for SCHED_RR.
1828 Note, that the trace data shows the internal priority (99 - rtprio).
1831 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1833 The 0:120:R means idle was running with a nice priority of 0 (120 - 120)
1834 and in the running state 'R'. The sleep task was scheduled in with
1835 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1836 and it too is in the running state.
1838 Doing the same with chrt -r 5 and function-trace set.
1841 echo 1 > options/function-trace
1845 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1846 # --------------------------------------------------------------------
1847 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1849 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
1853 # / _-----=> irqs-off
1854 # | / _----=> need-resched
1855 # || / _---=> hardirq/softirq
1856 # ||| / _--=> preempt-depth
1858 # cmd pid ||||| time | caller
1860 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
1861 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1862 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
1863 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
1864 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
1865 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
1866 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
1867 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
1868 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
1869 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1870 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
1871 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
1872 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
1873 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
1874 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
1875 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
1876 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
1877 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
1878 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
1879 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
1880 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
1881 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
1882 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
1883 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
1884 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
1885 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
1886 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
1887 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
1888 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
1889 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
1890 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
1891 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
1892 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
1893 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
1894 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
1895 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
1896 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
1897 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit
1898 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
1899 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
1900 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
1901 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
1902 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1903 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
1904 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
1905 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
1906 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
1907 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
1908 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
1909 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
1910 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
1911 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1912 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
1913 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1914 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1915 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
1916 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
1917 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
1918 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1919 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
1920 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
1921 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
1922 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
1923 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
1924 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
1925 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
1926 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
1927 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1928 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
1929 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
1930 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
1931 <idle>-0 3.N.. 25us : schedule <-cpu_idle
1932 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
1933 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
1934 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
1935 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
1936 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
1937 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
1938 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
1939 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
1940 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
1941 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
1942 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
1943 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
1944 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
1946 This isn't that big of a trace, even with function tracing enabled,
1947 so I included the entire trace.
1949 The interrupt went off while when the system was idle. Somewhere
1950 before task_woken_rt() was called, the NEED_RESCHED flag was set,
1951 this is indicated by the first occurrence of the 'N' flag.
1953 Latency tracing and events
1954 --------------------------
1955 As function tracing can induce a much larger latency, but without
1956 seeing what happens within the latency it is hard to know what
1957 caused it. There is a middle ground, and that is with enabling
1961 # echo 0 > options/function-trace
1962 # echo wakeup_rt > current_tracer
1963 # echo 1 > events/enable
1964 # echo 1 > tracing_on
1965 # echo 0 > tracing_max_latency
1967 # echo 0 > tracing_on
1971 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1972 # --------------------------------------------------------------------
1973 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1975 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
1979 # / _-----=> irqs-off
1980 # | / _----=> need-resched
1981 # || / _---=> hardirq/softirq
1982 # ||| / _--=> preempt-depth
1984 # cmd pid ||||| time | caller
1986 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
1987 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1988 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
1989 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
1990 <idle>-0 2.N.2 2us : power_end: cpu_id=2
1991 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
1992 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
1993 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
1994 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
1995 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
1996 <idle>-0 2d..3 6us : __schedule <-schedule
1997 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
2000 Hardware Latency Detector
2001 -------------------------
2003 The hardware latency detector is executed by enabling the "hwlat" tracer.
2005 NOTE, this tracer will affect the performance of the system as it will
2006 periodically make a CPU constantly busy with interrupts disabled.
2009 # echo hwlat > current_tracer
2015 # / _----=> need-resched
2016 # | / _---=> hardirq/softirq
2017 # || / _--=> preempt-depth
2019 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2021 <...>-3638 [001] d... 19452.055471: #1 inner/outer(us): 12/14 ts:1499801089.066141940
2022 <...>-3638 [003] d... 19454.071354: #2 inner/outer(us): 11/9 ts:1499801091.082164365
2023 <...>-3638 [002] dn.. 19461.126852: #3 inner/outer(us): 12/9 ts:1499801098.138150062
2024 <...>-3638 [001] d... 19488.340960: #4 inner/outer(us): 8/12 ts:1499801125.354139633
2025 <...>-3638 [003] d... 19494.388553: #5 inner/outer(us): 8/12 ts:1499801131.402150961
2026 <...>-3638 [003] d... 19501.283419: #6 inner/outer(us): 0/12 ts:1499801138.297435289 nmi-total:4 nmi-count:1
2029 The above output is somewhat the same in the header. All events will have
2030 interrupts disabled 'd'. Under the FUNCTION title there is:
2033 This is the count of events recorded that were greater than the
2034 tracing_threshold (See below).
2036 inner/outer(us): 12/14
2038 This shows two numbers as "inner latency" and "outer latency". The test
2039 runs in a loop checking a timestamp twice. The latency detected within
2040 the two timestamps is the "inner latency" and the latency detected
2041 after the previous timestamp and the next timestamp in the loop is
2042 the "outer latency".
2044 ts:1499801089.066141940
2046 The absolute timestamp that the event happened.
2048 nmi-total:4 nmi-count:1
2050 On architectures that support it, if an NMI comes in during the
2051 test, the time spent in NMI is reported in "nmi-total" (in
2054 All architectures that have NMIs will show the "nmi-count" if an
2055 NMI comes in during the test.
2060 This gets automatically set to "10" to represent 10
2061 microseconds. This is the threshold of latency that
2062 needs to be detected before the trace will be recorded.
2064 Note, when hwlat tracer is finished (another tracer is
2065 written into "current_tracer"), the original value for
2066 tracing_threshold is placed back into this file.
2068 hwlat_detector/width
2069 The length of time the test runs with interrupts disabled.
2071 hwlat_detector/window
2072 The length of time of the window which the test
2073 runs. That is, the test will run for "width"
2074 microseconds per "window" microseconds
2077 When the test is started. A kernel thread is created that
2078 runs the test. This thread will alternate between CPUs
2079 listed in the tracing_cpumask between each period
2080 (one "window"). To limit the test to specific CPUs
2081 set the mask in this file to only the CPUs that the test
2087 This tracer is the function tracer. Enabling the function tracer
2088 can be done from the debug file system. Make sure the
2089 ftrace_enabled is set; otherwise this tracer is a nop.
2090 See the "ftrace_enabled" section below.
2093 # sysctl kernel.ftrace_enabled=1
2094 # echo function > current_tracer
2095 # echo 1 > tracing_on
2097 # echo 0 > tracing_on
2101 # entries-in-buffer/entries-written: 24799/24799 #P:4
2104 # / _----=> need-resched
2105 # | / _---=> hardirq/softirq
2106 # || / _--=> preempt-depth
2108 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2110 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
2111 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
2112 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
2113 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
2114 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
2115 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
2116 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
2117 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
2121 Note: function tracer uses ring buffers to store the above
2122 entries. The newest data may overwrite the oldest data.
2123 Sometimes using echo to stop the trace is not sufficient because
2124 the tracing could have overwritten the data that you wanted to
2125 record. For this reason, it is sometimes better to disable
2126 tracing directly from a program. This allows you to stop the
2127 tracing at the point that you hit the part that you are
2128 interested in. To disable the tracing directly from a C program,
2129 something like following code snippet can be used::
2133 int main(int argc, char *argv[]) {
2135 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
2137 if (condition_hit()) {
2138 write(trace_fd, "0", 1);
2144 Single thread tracing
2145 ---------------------
2147 By writing into set_ftrace_pid you can trace a
2148 single thread. For example::
2150 # cat set_ftrace_pid
2152 # echo 3111 > set_ftrace_pid
2153 # cat set_ftrace_pid
2155 # echo function > current_tracer
2159 # TASK-PID CPU# TIMESTAMP FUNCTION
2161 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
2162 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
2163 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
2164 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
2165 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
2166 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
2167 # echo > set_ftrace_pid
2171 # TASK-PID CPU# TIMESTAMP FUNCTION
2173 ##### CPU 3 buffer started ####
2174 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
2175 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
2176 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
2177 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
2178 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
2180 If you want to trace a function when executing, you could use
2181 something like this simple program.
2186 #include <sys/types.h>
2187 #include <sys/stat.h>
2193 #define STR(x) _STR(x)
2194 #define MAX_PATH 256
2196 const char *find_tracefs(void)
2198 static char tracefs[MAX_PATH+1];
2199 static int tracefs_found;
2206 if ((fp = fopen("/proc/mounts","r")) == NULL) {
2207 perror("/proc/mounts");
2211 while (fscanf(fp, "%*s %"
2213 "s %99s %*s %*d %*d\n",
2214 tracefs, type) == 2) {
2215 if (strcmp(type, "tracefs") == 0)
2220 if (strcmp(type, "tracefs") != 0) {
2221 fprintf(stderr, "tracefs not mounted");
2225 strcat(tracefs, "/tracing/");
2231 const char *tracing_file(const char *file_name)
2233 static char trace_file[MAX_PATH+1];
2234 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name);
2238 int main (int argc, char **argv)
2248 ffd = open(tracing_file("current_tracer"), O_WRONLY);
2251 write(ffd, "nop", 3);
2253 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
2254 s = sprintf(line, "%d\n", getpid());
2257 write(ffd, "function", 8);
2262 execvp(argv[1], argv+1);
2268 Or this simple script!
2273 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts`
2274 echo nop > $tracefs/tracing/current_tracer
2275 echo 0 > $tracefs/tracing/tracing_on
2276 echo $$ > $tracefs/tracing/set_ftrace_pid
2277 echo function > $tracefs/tracing/current_tracer
2278 echo 1 > $tracefs/tracing/tracing_on
2282 function graph tracer
2283 ---------------------------
2285 This tracer is similar to the function tracer except that it
2286 probes a function on its entry and its exit. This is done by
2287 using a dynamically allocated stack of return addresses in each
2288 task_struct. On function entry the tracer overwrites the return
2289 address of each function traced to set a custom probe. Thus the
2290 original return address is stored on the stack of return address
2293 Probing on both ends of a function leads to special features
2296 - measure of a function's time execution
2297 - having a reliable call stack to draw function calls graph
2299 This tracer is useful in several situations:
2301 - you want to find the reason of a strange kernel behavior and
2302 need to see what happens in detail on any areas (or specific
2305 - you are experiencing weird latencies but it's difficult to
2308 - you want to find quickly which path is taken by a specific
2311 - you just want to peek inside a working kernel and want to see
2316 # tracer: function_graph
2318 # CPU DURATION FUNCTION CALLS
2322 0) | do_sys_open() {
2324 0) | kmem_cache_alloc() {
2325 0) 1.382 us | __might_sleep();
2327 0) | strncpy_from_user() {
2328 0) | might_fault() {
2329 0) 1.389 us | __might_sleep();
2334 0) 0.668 us | _spin_lock();
2335 0) 0.570 us | expand_files();
2336 0) 0.586 us | _spin_unlock();
2339 There are several columns that can be dynamically
2340 enabled/disabled. You can use every combination of options you
2341 want, depending on your needs.
2343 - The cpu number on which the function executed is default
2344 enabled. It is sometimes better to only trace one cpu (see
2345 tracing_cpu_mask file) or you might sometimes see unordered
2346 function calls while cpu tracing switch.
2348 - hide: echo nofuncgraph-cpu > trace_options
2349 - show: echo funcgraph-cpu > trace_options
2351 - The duration (function's time of execution) is displayed on
2352 the closing bracket line of a function or on the same line
2353 than the current function in case of a leaf one. It is default
2356 - hide: echo nofuncgraph-duration > trace_options
2357 - show: echo funcgraph-duration > trace_options
2359 - The overhead field precedes the duration field in case of
2360 reached duration thresholds.
2362 - hide: echo nofuncgraph-overhead > trace_options
2363 - show: echo funcgraph-overhead > trace_options
2364 - depends on: funcgraph-duration
2368 3) # 1837.709 us | } /* __switch_to */
2369 3) | finish_task_switch() {
2370 3) 0.313 us | _raw_spin_unlock_irq();
2372 3) # 1889.063 us | } /* __schedule */
2373 3) ! 140.417 us | } /* __schedule */
2374 3) # 2034.948 us | } /* schedule */
2375 3) * 33998.59 us | } /* schedule_preempt_disabled */
2379 1) 0.260 us | msecs_to_jiffies();
2380 1) 0.313 us | __rcu_read_unlock();
2383 1) 0.313 us | rcu_bh_qs();
2384 1) 0.313 us | __local_bh_enable();
2386 1) 0.365 us | idle_cpu();
2387 1) | rcu_irq_exit() {
2388 1) 0.417 us | rcu_eqs_enter_common.isra.47();
2392 1) @ 119760.2 us | }
2398 2) 0.417 us | scheduler_ipi();
2408 + means that the function exceeded 10 usecs.
2409 ! means that the function exceeded 100 usecs.
2410 # means that the function exceeded 1000 usecs.
2411 * means that the function exceeded 10 msecs.
2412 @ means that the function exceeded 100 msecs.
2413 $ means that the function exceeded 1 sec.
2416 - The task/pid field displays the thread cmdline and pid which
2417 executed the function. It is default disabled.
2419 - hide: echo nofuncgraph-proc > trace_options
2420 - show: echo funcgraph-proc > trace_options
2424 # tracer: function_graph
2426 # CPU TASK/PID DURATION FUNCTION CALLS
2428 0) sh-4802 | | d_free() {
2429 0) sh-4802 | | call_rcu() {
2430 0) sh-4802 | | __call_rcu() {
2431 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2432 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2433 0) sh-4802 | 2.899 us | }
2434 0) sh-4802 | 4.040 us | }
2435 0) sh-4802 | 5.151 us | }
2436 0) sh-4802 | + 49.370 us | }
2439 - The absolute time field is an absolute timestamp given by the
2440 system clock since it started. A snapshot of this time is
2441 given on each entry/exit of functions
2443 - hide: echo nofuncgraph-abstime > trace_options
2444 - show: echo funcgraph-abstime > trace_options
2449 # TIME CPU DURATION FUNCTION CALLS
2451 360.774522 | 1) 0.541 us | }
2452 360.774522 | 1) 4.663 us | }
2453 360.774523 | 1) 0.541 us | __wake_up_bit();
2454 360.774524 | 1) 6.796 us | }
2455 360.774524 | 1) 7.952 us | }
2456 360.774525 | 1) 9.063 us | }
2457 360.774525 | 1) 0.615 us | journal_mark_dirty();
2458 360.774527 | 1) 0.578 us | __brelse();
2459 360.774528 | 1) | reiserfs_prepare_for_journal() {
2460 360.774528 | 1) | unlock_buffer() {
2461 360.774529 | 1) | wake_up_bit() {
2462 360.774529 | 1) | bit_waitqueue() {
2463 360.774530 | 1) 0.594 us | __phys_addr();
2466 The function name is always displayed after the closing bracket
2467 for a function if the start of that function is not in the
2470 Display of the function name after the closing bracket may be
2471 enabled for functions whose start is in the trace buffer,
2472 allowing easier searching with grep for function durations.
2473 It is default disabled.
2475 - hide: echo nofuncgraph-tail > trace_options
2476 - show: echo funcgraph-tail > trace_options
2478 Example with nofuncgraph-tail (default)::
2481 0) | kmem_cache_free() {
2482 0) 0.518 us | __phys_addr();
2486 Example with funcgraph-tail::
2489 0) | kmem_cache_free() {
2490 0) 0.518 us | __phys_addr();
2491 0) 1.757 us | } /* kmem_cache_free() */
2492 0) 2.861 us | } /* putname() */
2494 You can put some comments on specific functions by using
2495 trace_printk() For example, if you want to put a comment inside
2496 the __might_sleep() function, you just have to include
2497 <linux/ftrace.h> and call trace_printk() inside __might_sleep()::
2499 trace_printk("I'm a comment!\n")
2503 1) | __might_sleep() {
2504 1) | /* I'm a comment! */
2508 You might find other useful features for this tracer in the
2509 following "dynamic ftrace" section such as tracing only specific
2515 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2516 virtually no overhead when function tracing is disabled. The way
2517 this works is the mcount function call (placed at the start of
2518 every kernel function, produced by the -pg switch in gcc),
2519 starts of pointing to a simple return. (Enabling FTRACE will
2520 include the -pg switch in the compiling of the kernel.)
2522 At compile time every C file object is run through the
2523 recordmcount program (located in the scripts directory). This
2524 program will parse the ELF headers in the C object to find all
2525 the locations in the .text section that call mcount. Starting
2526 with gcc verson 4.6, the -mfentry has been added for x86, which
2527 calls "__fentry__" instead of "mcount". Which is called before
2528 the creation of the stack frame.
2530 Note, not all sections are traced. They may be prevented by either
2531 a notrace, or blocked another way and all inline functions are not
2532 traced. Check the "available_filter_functions" file to see what functions
2535 A section called "__mcount_loc" is created that holds
2536 references to all the mcount/fentry call sites in the .text section.
2537 The recordmcount program re-links this section back into the
2538 original object. The final linking stage of the kernel will add all these
2539 references into a single table.
2541 On boot up, before SMP is initialized, the dynamic ftrace code
2542 scans this table and updates all the locations into nops. It
2543 also records the locations, which are added to the
2544 available_filter_functions list. Modules are processed as they
2545 are loaded and before they are executed. When a module is
2546 unloaded, it also removes its functions from the ftrace function
2547 list. This is automatic in the module unload code, and the
2548 module author does not need to worry about it.
2550 When tracing is enabled, the process of modifying the function
2551 tracepoints is dependent on architecture. The old method is to use
2552 kstop_machine to prevent races with the CPUs executing code being
2553 modified (which can cause the CPU to do undesirable things, especially
2554 if the modified code crosses cache (or page) boundaries), and the nops are
2555 patched back to calls. But this time, they do not call mcount
2556 (which is just a function stub). They now call into the ftrace
2559 The new method of modifying the function tracepoints is to place
2560 a breakpoint at the location to be modified, sync all CPUs, modify
2561 the rest of the instruction not covered by the breakpoint. Sync
2562 all CPUs again, and then remove the breakpoint with the finished
2563 version to the ftrace call site.
2565 Some archs do not even need to monkey around with the synchronization,
2566 and can just slap the new code on top of the old without any
2567 problems with other CPUs executing it at the same time.
2569 One special side-effect to the recording of the functions being
2570 traced is that we can now selectively choose which functions we
2571 wish to trace and which ones we want the mcount calls to remain
2574 Two files are used, one for enabling and one for disabling the
2575 tracing of specified functions. They are:
2583 A list of available functions that you can add to these files is
2586 available_filter_functions
2590 # cat available_filter_functions
2599 If I am only interested in sys_nanosleep and hrtimer_interrupt::
2601 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2602 # echo function > current_tracer
2603 # echo 1 > tracing_on
2605 # echo 0 > tracing_on
2609 # entries-in-buffer/entries-written: 5/5 #P:4
2612 # / _----=> need-resched
2613 # | / _---=> hardirq/softirq
2614 # || / _--=> preempt-depth
2616 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2618 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2619 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2620 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2621 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2622 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2624 To see which functions are being traced, you can cat the file:
2627 # cat set_ftrace_filter
2632 Perhaps this is not enough. The filters also allow glob(7) matching.
2635 will match functions that begin with <match>
2637 will match functions that end with <match>
2639 will match functions that have <match> in it
2640 ``<match1>*<match2>``
2641 will match functions that begin with <match1> and end with <match2>
2644 It is better to use quotes to enclose the wild cards,
2645 otherwise the shell may expand the parameters into names
2646 of files in the local directory.
2650 # echo 'hrtimer_*' > set_ftrace_filter
2656 # entries-in-buffer/entries-written: 897/897 #P:4
2659 # / _----=> need-resched
2660 # | / _---=> hardirq/softirq
2661 # || / _--=> preempt-depth
2663 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2665 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2666 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2667 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2668 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2669 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2670 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2671 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2672 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2674 Notice that we lost the sys_nanosleep.
2677 # cat set_ftrace_filter
2682 hrtimer_try_to_cancel
2686 hrtimer_force_reprogram
2687 hrtimer_get_next_event
2691 hrtimer_get_remaining
2693 hrtimer_init_sleeper
2696 This is because the '>' and '>>' act just like they do in bash.
2697 To rewrite the filters, use '>'
2698 To append to the filters, use '>>'
2700 To clear out a filter so that all functions will be recorded
2703 # echo > set_ftrace_filter
2704 # cat set_ftrace_filter
2707 Again, now we want to append.
2711 # echo sys_nanosleep > set_ftrace_filter
2712 # cat set_ftrace_filter
2714 # echo 'hrtimer_*' >> set_ftrace_filter
2715 # cat set_ftrace_filter
2720 hrtimer_try_to_cancel
2724 hrtimer_force_reprogram
2725 hrtimer_get_next_event
2730 hrtimer_get_remaining
2732 hrtimer_init_sleeper
2735 The set_ftrace_notrace prevents those functions from being
2739 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2745 # entries-in-buffer/entries-written: 39608/39608 #P:4
2748 # / _----=> need-resched
2749 # | / _---=> hardirq/softirq
2750 # || / _--=> preempt-depth
2752 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2754 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2755 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2756 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2757 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2758 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2759 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2760 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2761 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2762 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2763 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2764 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2765 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2767 We can see that there's no more lock or preempt tracing.
2770 Dynamic ftrace with the function graph tracer
2771 ---------------------------------------------
2773 Although what has been explained above concerns both the
2774 function tracer and the function-graph-tracer, there are some
2775 special features only available in the function-graph tracer.
2777 If you want to trace only one function and all of its children,
2778 you just have to echo its name into set_graph_function::
2780 echo __do_fault > set_graph_function
2782 will produce the following "expanded" trace of the __do_fault()
2786 0) | filemap_fault() {
2787 0) | find_lock_page() {
2788 0) 0.804 us | find_get_page();
2789 0) | __might_sleep() {
2793 0) 0.653 us | _spin_lock();
2794 0) 0.578 us | page_add_file_rmap();
2795 0) 0.525 us | native_set_pte_at();
2796 0) 0.585 us | _spin_unlock();
2797 0) | unlock_page() {
2798 0) 0.541 us | page_waitqueue();
2799 0) 0.639 us | __wake_up_bit();
2803 0) | filemap_fault() {
2804 0) | find_lock_page() {
2805 0) 0.698 us | find_get_page();
2806 0) | __might_sleep() {
2810 0) 0.631 us | _spin_lock();
2811 0) 0.571 us | page_add_file_rmap();
2812 0) 0.526 us | native_set_pte_at();
2813 0) 0.586 us | _spin_unlock();
2814 0) | unlock_page() {
2815 0) 0.533 us | page_waitqueue();
2816 0) 0.638 us | __wake_up_bit();
2820 You can also expand several functions at once::
2822 echo sys_open > set_graph_function
2823 echo sys_close >> set_graph_function
2825 Now if you want to go back to trace all functions you can clear
2826 this special filter via::
2828 echo > set_graph_function
2834 Note, the proc sysctl ftrace_enable is a big on/off switch for the
2835 function tracer. By default it is enabled (when function tracing is
2836 enabled in the kernel). If it is disabled, all function tracing is
2837 disabled. This includes not only the function tracers for ftrace, but
2838 also for any other uses (perf, kprobes, stack tracing, profiling, etc).
2840 Please disable this with care.
2842 This can be disable (and enabled) with::
2844 sysctl kernel.ftrace_enabled=0
2845 sysctl kernel.ftrace_enabled=1
2849 echo 0 > /proc/sys/kernel/ftrace_enabled
2850 echo 1 > /proc/sys/kernel/ftrace_enabled
2856 A few commands are supported by the set_ftrace_filter interface.
2857 Trace commands have the following format::
2859 <function>:<command>:<parameter>
2861 The following commands are supported:
2864 This command enables function filtering per module. The
2865 parameter defines the module. For example, if only the write*
2866 functions in the ext3 module are desired, run:
2868 echo 'write*:mod:ext3' > set_ftrace_filter
2870 This command interacts with the filter in the same way as
2871 filtering based on function names. Thus, adding more functions
2872 in a different module is accomplished by appending (>>) to the
2873 filter file. Remove specific module functions by prepending
2876 echo '!writeback*:mod:ext3' >> set_ftrace_filter
2878 Mod command supports module globbing. Disable tracing for all
2879 functions except a specific module::
2881 echo '!*:mod:!ext3' >> set_ftrace_filter
2883 Disable tracing for all modules, but still trace kernel::
2885 echo '!*:mod:*' >> set_ftrace_filter
2887 Enable filter only for kernel::
2889 echo '*write*:mod:!*' >> set_ftrace_filter
2891 Enable filter for module globbing::
2893 echo '*write*:mod:*snd*' >> set_ftrace_filter
2896 These commands turn tracing on and off when the specified
2897 functions are hit. The parameter determines how many times the
2898 tracing system is turned on and off. If unspecified, there is
2899 no limit. For example, to disable tracing when a schedule bug
2900 is hit the first 5 times, run::
2902 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
2904 To always disable tracing when __schedule_bug is hit::
2906 echo '__schedule_bug:traceoff' > set_ftrace_filter
2908 These commands are cumulative whether or not they are appended
2909 to set_ftrace_filter. To remove a command, prepend it by '!'
2910 and drop the parameter::
2912 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
2914 The above removes the traceoff command for __schedule_bug
2915 that have a counter. To remove commands without counters::
2917 echo '!__schedule_bug:traceoff' > set_ftrace_filter
2920 Will cause a snapshot to be triggered when the function is hit.
2923 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
2925 To only snapshot once:
2928 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
2930 To remove the above commands::
2932 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
2933 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
2935 - enable_event/disable_event:
2936 These commands can enable or disable a trace event. Note, because
2937 function tracing callbacks are very sensitive, when these commands
2938 are registered, the trace point is activated, but disabled in
2939 a "soft" mode. That is, the tracepoint will be called, but
2940 just will not be traced. The event tracepoint stays in this mode
2941 as long as there's a command that triggers it.
2944 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
2949 <function>:enable_event:<system>:<event>[:count]
2950 <function>:disable_event:<system>:<event>[:count]
2952 To remove the events commands::
2954 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
2956 echo '!schedule:disable_event:sched:sched_switch' > \
2960 When the function is hit, it will dump the contents of the ftrace
2961 ring buffer to the console. This is useful if you need to debug
2962 something, and want to dump the trace when a certain function
2963 is hit. Perhaps its a function that is called before a tripple
2964 fault happens and does not allow you to get a regular dump.
2967 When the function is hit, it will dump the contents of the ftrace
2968 ring buffer for the current CPU to the console. Unlike the "dump"
2969 command, it only prints out the contents of the ring buffer for the
2970 CPU that executed the function that triggered the dump.
2975 The trace_pipe outputs the same content as the trace file, but
2976 the effect on the tracing is different. Every read from
2977 trace_pipe is consumed. This means that subsequent reads will be
2978 different. The trace is live.
2981 # echo function > current_tracer
2982 # cat trace_pipe > /tmp/trace.out &
2984 # echo 1 > tracing_on
2986 # echo 0 > tracing_on
2990 # entries-in-buffer/entries-written: 0/0 #P:4
2993 # / _----=> need-resched
2994 # | / _---=> hardirq/softirq
2995 # || / _--=> preempt-depth
2997 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3001 # cat /tmp/trace.out
3002 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
3003 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
3004 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
3005 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
3006 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
3007 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
3008 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
3009 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
3010 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
3013 Note, reading the trace_pipe file will block until more input is
3019 Having too much or not enough data can be troublesome in
3020 diagnosing an issue in the kernel. The file buffer_size_kb is
3021 used to modify the size of the internal trace buffers. The
3022 number listed is the number of entries that can be recorded per
3023 CPU. To know the full size, multiply the number of possible CPUs
3024 with the number of entries.
3027 # cat buffer_size_kb
3028 1408 (units kilobytes)
3030 Or simply read buffer_total_size_kb
3033 # cat buffer_total_size_kb
3036 To modify the buffer, simple echo in a number (in 1024 byte segments).
3039 # echo 10000 > buffer_size_kb
3040 # cat buffer_size_kb
3041 10000 (units kilobytes)
3043 It will try to allocate as much as possible. If you allocate too
3044 much, it can cause Out-Of-Memory to trigger.
3047 # echo 1000000000000 > buffer_size_kb
3048 -bash: echo: write error: Cannot allocate memory
3049 # cat buffer_size_kb
3052 The per_cpu buffers can be changed individually as well:
3055 # echo 10000 > per_cpu/cpu0/buffer_size_kb
3056 # echo 100 > per_cpu/cpu1/buffer_size_kb
3058 When the per_cpu buffers are not the same, the buffer_size_kb
3059 at the top level will just show an X
3062 # cat buffer_size_kb
3065 This is where the buffer_total_size_kb is useful:
3068 # cat buffer_total_size_kb
3071 Writing to the top level buffer_size_kb will reset all the buffers
3072 to be the same again.
3076 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
3077 available to all non latency tracers. (Latency tracers which
3078 record max latency, such as "irqsoff" or "wakeup", can't use
3079 this feature, since those are already using the snapshot
3080 mechanism internally.)
3082 Snapshot preserves a current trace buffer at a particular point
3083 in time without stopping tracing. Ftrace swaps the current
3084 buffer with a spare buffer, and tracing continues in the new
3085 current (=previous spare) buffer.
3087 The following tracefs files in "tracing" are related to this
3092 This is used to take a snapshot and to read the output
3093 of the snapshot. Echo 1 into this file to allocate a
3094 spare buffer and to take a snapshot (swap), then read
3095 the snapshot from this file in the same format as
3096 "trace" (described above in the section "The File
3097 System"). Both reads snapshot and tracing are executable
3098 in parallel. When the spare buffer is allocated, echoing
3099 0 frees it, and echoing else (positive) values clear the
3101 More details are shown in the table below.
3103 +--------------+------------+------------+------------+
3104 |status\\input | 0 | 1 | else |
3105 +==============+============+============+============+
3106 |not allocated |(do nothing)| alloc+swap |(do nothing)|
3107 +--------------+------------+------------+------------+
3108 |allocated | free | swap | clear |
3109 +--------------+------------+------------+------------+
3111 Here is an example of using the snapshot feature.
3114 # echo 1 > events/sched/enable
3119 # entries-in-buffer/entries-written: 71/71 #P:8
3122 # / _----=> need-resched
3123 # | / _---=> hardirq/softirq
3124 # || / _--=> preempt-depth
3126 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3128 <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
3129 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
3131 <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
3136 # entries-in-buffer/entries-written: 77/77 #P:8
3139 # / _----=> need-resched
3140 # | / _---=> hardirq/softirq
3141 # || / _--=> preempt-depth
3143 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3145 <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
3146 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
3150 If you try to use this snapshot feature when current tracer is
3151 one of the latency tracers, you will get the following results.
3154 # echo wakeup > current_tracer
3156 bash: echo: write error: Device or resource busy
3158 cat: snapshot: Device or resource busy
3163 In the tracefs tracing directory is a directory called "instances".
3164 This directory can have new directories created inside of it using
3165 mkdir, and removing directories with rmdir. The directory created
3166 with mkdir in this directory will already contain files and other
3167 directories after it is created.
3170 # mkdir instances/foo
3172 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
3173 set_event snapshot trace trace_clock trace_marker trace_options
3174 trace_pipe tracing_on
3176 As you can see, the new directory looks similar to the tracing directory
3177 itself. In fact, it is very similar, except that the buffer and
3178 events are agnostic from the main director, or from any other
3179 instances that are created.
3181 The files in the new directory work just like the files with the
3182 same name in the tracing directory except the buffer that is used
3183 is a separate and new buffer. The files affect that buffer but do not
3184 affect the main buffer with the exception of trace_options. Currently,
3185 the trace_options affect all instances and the top level buffer
3186 the same, but this may change in future releases. That is, options
3187 may become specific to the instance they reside in.
3189 Notice that none of the function tracer files are there, nor is
3190 current_tracer and available_tracers. This is because the buffers
3191 can currently only have events enabled for them.
3194 # mkdir instances/foo
3195 # mkdir instances/bar
3196 # mkdir instances/zoot
3197 # echo 100000 > buffer_size_kb
3198 # echo 1000 > instances/foo/buffer_size_kb
3199 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
3200 # echo function > current_trace
3201 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
3202 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
3203 # echo 1 > instances/foo/events/sched/sched_switch/enable
3204 # echo 1 > instances/bar/events/irq/enable
3205 # echo 1 > instances/zoot/events/syscalls/enable
3207 CPU:2 [LOST 11745 EVENTS]
3208 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
3209 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
3210 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
3211 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
3212 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
3213 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
3214 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
3215 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
3216 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3217 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3218 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
3221 # cat instances/foo/trace_pipe
3222 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3223 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3224 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
3225 <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
3226 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
3227 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3228 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3229 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
3230 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
3231 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
3234 # cat instances/bar/trace_pipe
3235 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
3236 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
3237 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
3238 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
3239 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
3240 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
3241 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
3242 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
3243 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
3244 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
3245 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
3246 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
3249 # cat instances/zoot/trace
3252 # entries-in-buffer/entries-written: 18996/18996 #P:4
3255 # / _----=> need-resched
3256 # | / _---=> hardirq/softirq
3257 # || / _--=> preempt-depth
3259 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3261 bash-1998 [000] d... 140.733501: sys_write -> 0x2
3262 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
3263 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
3264 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
3265 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
3266 bash-1998 [000] d... 140.733510: sys_close(fd: a)
3267 bash-1998 [000] d... 140.733510: sys_close -> 0x0
3268 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
3269 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
3270 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
3271 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
3273 You can see that the trace of the top most trace buffer shows only
3274 the function tracing. The foo instance displays wakeups and task
3277 To remove the instances, simply delete their directories:
3280 # rmdir instances/foo
3281 # rmdir instances/bar
3282 # rmdir instances/zoot
3284 Note, if a process has a trace file open in one of the instance
3285 directories, the rmdir will fail with EBUSY.
3290 Since the kernel has a fixed sized stack, it is important not to
3291 waste it in functions. A kernel developer must be conscience of
3292 what they allocate on the stack. If they add too much, the system
3293 can be in danger of a stack overflow, and corruption will occur,
3294 usually leading to a system panic.
3296 There are some tools that check this, usually with interrupts
3297 periodically checking usage. But if you can perform a check
3298 at every function call that will become very useful. As ftrace provides
3299 a function tracer, it makes it convenient to check the stack size
3300 at every function call. This is enabled via the stack tracer.
3302 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
3303 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
3306 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
3308 You can also enable it from the kernel command line to trace
3309 the stack size of the kernel during boot up, by adding "stacktrace"
3310 to the kernel command line parameter.
3312 After running it for a few minutes, the output looks like:
3315 # cat stack_max_size
3319 Depth Size Location (18 entries)
3321 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
3322 1) 2704 160 find_busiest_group+0x31/0x1f1
3323 2) 2544 256 load_balance+0xd9/0x662
3324 3) 2288 80 idle_balance+0xbb/0x130
3325 4) 2208 128 __schedule+0x26e/0x5b9
3326 5) 2080 16 schedule+0x64/0x66
3327 6) 2064 128 schedule_timeout+0x34/0xe0
3328 7) 1936 112 wait_for_common+0x97/0xf1
3329 8) 1824 16 wait_for_completion+0x1d/0x1f
3330 9) 1808 128 flush_work+0xfe/0x119
3331 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
3332 11) 1664 48 input_available_p+0x1d/0x5c
3333 12) 1616 48 n_tty_poll+0x6d/0x134
3334 13) 1568 64 tty_poll+0x64/0x7f
3335 14) 1504 880 do_select+0x31e/0x511
3336 15) 624 400 core_sys_select+0x177/0x216
3337 16) 224 96 sys_select+0x91/0xb9
3338 17) 128 128 system_call_fastpath+0x16/0x1b
3340 Note, if -mfentry is being used by gcc, functions get traced before
3341 they set up the stack frame. This means that leaf level functions
3342 are not tested by the stack tracer when -mfentry is used.
3344 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
3348 More details can be found in the source code, in the `kernel/trace/*.c` files.