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 Directory for the Hardware Latency Detector.
549 See "Hardware Latency Detector" section below.
553 This is a directory that contains the trace per_cpu information.
555 per_cpu/cpu0/buffer_size_kb:
557 The ftrace buffer is defined per_cpu. That is, there's a separate
558 buffer for each CPU to allow writes to be done atomically,
559 and free from cache bouncing. These buffers may have different
560 size buffers. This file is similar to the buffer_size_kb
561 file, but it only displays or sets the buffer size for the
562 specific CPU. (here cpu0).
566 This is similar to the "trace" file, but it will only display
567 the data specific for the CPU. If written to, it only clears
568 the specific CPU buffer.
570 per_cpu/cpu0/trace_pipe
572 This is similar to the "trace_pipe" file, and is a consuming
573 read, but it will only display (and consume) the data specific
576 per_cpu/cpu0/trace_pipe_raw
578 For tools that can parse the ftrace ring buffer binary format,
579 the trace_pipe_raw file can be used to extract the data
580 from the ring buffer directly. With the use of the splice()
581 system call, the buffer data can be quickly transferred to
582 a file or to the network where a server is collecting the
585 Like trace_pipe, this is a consuming reader, where multiple
586 reads will always produce different data.
588 per_cpu/cpu0/snapshot:
590 This is similar to the main "snapshot" file, but will only
591 snapshot the current CPU (if supported). It only displays
592 the content of the snapshot for a given CPU, and if
593 written to, only clears this CPU buffer.
595 per_cpu/cpu0/snapshot_raw:
597 Similar to the trace_pipe_raw, but will read the binary format
598 from the snapshot buffer for the given CPU.
602 This displays certain stats about the ring buffer:
605 The number of events that are still in the buffer.
608 The number of lost events due to overwriting when
612 Should always be zero.
613 This gets set if so many events happened within a nested
614 event (ring buffer is re-entrant), that it fills the
615 buffer and starts dropping events.
618 Bytes actually read (not overwritten).
621 The oldest timestamp in the buffer
624 The current timestamp
627 Events lost due to overwrite option being off.
630 The number of events read.
635 Here is the list of current tracers that may be configured.
639 Function call tracer to trace all kernel functions.
643 Similar to the function tracer except that the
644 function tracer probes the functions on their entry
645 whereas the function graph tracer traces on both entry
646 and exit of the functions. It then provides the ability
647 to draw a graph of function calls similar to C code
652 The block tracer. The tracer used by the blktrace user
657 The Hardware Latency tracer is used to detect if the hardware
658 produces any latency. See "Hardware Latency Detector" section
663 Traces the areas that disable interrupts and saves
664 the trace with the longest max latency.
665 See tracing_max_latency. When a new max is recorded,
666 it replaces the old trace. It is best to view this
667 trace with the latency-format option enabled, which
668 happens automatically when the tracer is selected.
672 Similar to irqsoff but traces and records the amount of
673 time for which preemption is disabled.
677 Similar to irqsoff and preemptoff, but traces and
678 records the largest time for which irqs and/or preemption
683 Traces and records the max latency that it takes for
684 the highest priority task to get scheduled after
685 it has been woken up.
686 Traces all tasks as an average developer would expect.
690 Traces and records the max latency that it takes for just
691 RT tasks (as the current "wakeup" does). This is useful
692 for those interested in wake up timings of RT tasks.
696 Traces and records the max latency that it takes for
697 a SCHED_DEADLINE task to be woken (as the "wakeup" and
702 A special tracer that is used to trace binary module.
703 It will trace all the calls that a module makes to the
704 hardware. Everything it writes and reads from the I/O
709 This tracer can be configured when tracing likely/unlikely
710 calls within the kernel. It will trace when a likely and
711 unlikely branch is hit and if it was correct in its prediction
716 This is the "trace nothing" tracer. To remove all
717 tracers from tracing simply echo "nop" into
721 Examples of using the tracer
722 ----------------------------
724 Here are typical examples of using the tracers when controlling
725 them only with the tracefs interface (without using any
726 user-land utilities).
731 Here is an example of the output format of the file "trace"::
735 # entries-in-buffer/entries-written: 140080/250280 #P:4
738 # / _----=> need-resched
739 # | / _---=> hardirq/softirq
740 # || / _--=> preempt-depth
742 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
744 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
745 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
746 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
747 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
748 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
749 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
750 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
751 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
752 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
753 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
756 A header is printed with the tracer name that is represented by
757 the trace. In this case the tracer is "function". Then it shows the
758 number of events in the buffer as well as the total number of entries
759 that were written. The difference is the number of entries that were
760 lost due to the buffer filling up (250280 - 140080 = 110200 events
763 The header explains the content of the events. Task name "bash", the task
764 PID "1977", the CPU that it was running on "000", the latency format
765 (explained below), the timestamp in <secs>.<usecs> format, the
766 function name that was traced "sys_close" and the parent function that
767 called this function "system_call_fastpath". The timestamp is the time
768 at which the function was entered.
773 When the latency-format option is enabled or when one of the latency
774 tracers is set, the trace file gives somewhat more information to see
775 why a latency happened. Here is a typical trace::
779 # irqsoff latency trace v1.1.5 on 3.8.0-test+
780 # --------------------------------------------------------------------
781 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
783 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
785 # => started at: __lock_task_sighand
786 # => ended at: _raw_spin_unlock_irqrestore
790 # / _-----=> irqs-off
791 # | / _----=> need-resched
792 # || / _---=> hardirq/softirq
793 # ||| / _--=> preempt-depth
795 # cmd pid ||||| time | caller
797 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
798 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
799 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
800 ps-6143 2d..1 306us : <stack trace>
801 => trace_hardirqs_on_caller
803 => _raw_spin_unlock_irqrestore
810 => system_call_fastpath
813 This shows that the current tracer is "irqsoff" tracing the time
814 for which interrupts were disabled. It gives the trace version (which
815 never changes) and the version of the kernel upon which this was executed on
816 (3.8). Then it displays the max latency in microseconds (259 us). The number
817 of trace entries displayed and the total number (both are four: #4/4).
818 VP, KP, SP, and HP are always zero and are reserved for later use.
819 #P is the number of online CPUs (#P:4).
821 The task is the process that was running when the latency
822 occurred. (ps pid: 6143).
824 The start and stop (the functions in which the interrupts were
825 disabled and enabled respectively) that caused the latencies:
827 - __lock_task_sighand is where the interrupts were disabled.
828 - _raw_spin_unlock_irqrestore is where they were enabled again.
830 The next lines after the header are the trace itself. The header
831 explains which is which.
833 cmd: The name of the process in the trace.
835 pid: The PID of that process.
837 CPU#: The CPU which the process was running on.
839 irqs-off: 'd' interrupts are disabled. '.' otherwise.
840 .. caution:: If the architecture does not support a way to
841 read the irq flags variable, an 'X' will always
845 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
846 - 'n' only TIF_NEED_RESCHED is set,
847 - 'p' only PREEMPT_NEED_RESCHED is set,
851 - 'Z' - NMI occurred inside a hardirq
852 - 'z' - NMI is running
853 - 'H' - hard irq occurred inside a softirq.
854 - 'h' - hard irq is running
855 - 's' - soft irq is running
856 - '.' - normal context.
858 preempt-depth: The level of preempt_disabled
860 The above is mostly meaningful for kernel developers.
863 When the latency-format option is enabled, the trace file
864 output includes a timestamp relative to the start of the
865 trace. This differs from the output when latency-format
866 is disabled, which includes an absolute timestamp.
869 This is just to help catch your eye a bit better. And
870 needs to be fixed to be only relative to the same CPU.
871 The marks are determined by the difference between this
872 current trace and the next trace.
874 - '$' - greater than 1 second
875 - '@' - greater than 100 milisecond
876 - '*' - greater than 10 milisecond
877 - '#' - greater than 1000 microsecond
878 - '!' - greater than 100 microsecond
879 - '+' - greater than 10 microsecond
880 - ' ' - less than or equal to 10 microsecond.
882 The rest is the same as the 'trace' file.
884 Note, the latency tracers will usually end with a back trace
885 to easily find where the latency occurred.
890 The trace_options file (or the options directory) is used to control
891 what gets printed in the trace output, or manipulate the tracers.
892 To see what is available, simply cat the file::
923 To disable one of the options, echo in the option prepended with
926 echo noprint-parent > trace_options
928 To enable an option, leave off the "no"::
930 echo sym-offset > trace_options
932 Here are the available options:
935 On function traces, display the calling (parent)
936 function as well as the function being traced.
940 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
943 bash-4000 [01] 1477.606694: simple_strtoul
947 Display not only the function name, but also the
948 offset in the function. For example, instead of
949 seeing just "ktime_get", you will see
950 "ktime_get+0xb/0x20".
954 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
957 This will also display the function address as well
958 as the function name.
962 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
965 This deals with the trace file when the
966 latency-format option is enabled.
969 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
970 (+0.000ms): simple_strtoul (kstrtoul)
973 This will display raw numbers. This option is best for
974 use with user applications that can translate the raw
975 numbers better than having it done in the kernel.
978 Similar to raw, but the numbers will be in a hexadecimal format.
981 This will print out the formats in raw binary.
984 When set, reading trace_pipe will not block when polled.
987 Can disable trace_printk() from writing into the buffer.
990 It is sometimes confusing when the CPU buffers are full
991 and one CPU buffer had a lot of events recently, thus
992 a shorter time frame, were another CPU may have only had
993 a few events, which lets it have older events. When
994 the trace is reported, it shows the oldest events first,
995 and it may look like only one CPU ran (the one with the
996 oldest events). When the annotate option is set, it will
997 display when a new CPU buffer started::
999 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
1000 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
1001 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
1002 ##### CPU 2 buffer started ####
1003 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
1004 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
1005 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
1008 This option changes the trace. It records a
1009 stacktrace of the current user space thread after
1013 when user stacktrace are enabled, look up which
1014 object the address belongs to, and print a
1015 relative address. This is especially useful when
1016 ASLR is on, otherwise you don't get a chance to
1017 resolve the address to object/file/line after
1018 the app is no longer running
1020 The lookup is performed when you read
1021 trace,trace_pipe. Example::
1023 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
1024 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
1028 When set, trace_printk()s will only show the format
1029 and not their parameters (if trace_bprintk() or
1030 trace_bputs() was used to save the trace_printk()).
1033 Show only the event data. Hides the comm, PID,
1034 timestamp, CPU, and other useful data.
1037 This option changes the trace output. When it is enabled,
1038 the trace displays additional information about the
1039 latency, as described in "Latency trace format".
1042 When any event or tracer is enabled, a hook is enabled
1043 in the sched_switch trace point to fill comm cache
1044 with mapped pids and comms. But this may cause some
1045 overhead, and if you only care about pids, and not the
1046 name of the task, disabling this option can lower the
1047 impact of tracing. See "saved_cmdlines".
1050 When any event or tracer is enabled, a hook is enabled
1051 in the sched_switch trace point to fill the cache of
1052 mapped Thread Group IDs (TGID) mapping to pids. See
1056 This controls what happens when the trace buffer is
1057 full. If "1" (default), the oldest events are
1058 discarded and overwritten. If "0", then the newest
1059 events are discarded.
1060 (see per_cpu/cpu0/stats for overrun and dropped)
1063 When the free_buffer is closed, tracing will
1064 stop (tracing_on set to 0).
1067 Shows the interrupt, preempt count, need resched data.
1068 When disabled, the trace looks like::
1072 # entries-in-buffer/entries-written: 144405/9452052 #P:4
1074 # TASK-PID CPU# TIMESTAMP FUNCTION
1076 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
1077 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
1078 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
1082 When set, the trace_marker is writable (only by root).
1083 When disabled, the trace_marker will error with EINVAL
1087 When set, tasks with PIDs listed in set_event_pid will have
1088 the PIDs of their children added to set_event_pid when those
1089 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
1090 their PIDs will be removed from the file.
1093 The latency tracers will enable function tracing
1094 if this option is enabled (default it is). When
1095 it is disabled, the latency tracers do not trace
1096 functions. This keeps the overhead of the tracer down
1097 when performing latency tests.
1100 When set, tasks with PIDs listed in set_ftrace_pid will
1101 have the PIDs of their children added to set_ftrace_pid
1102 when those tasks fork. Also, when tasks with PIDs in
1103 set_ftrace_pid exit, their PIDs will be removed from the
1107 When set, the latency tracers (irqsoff, wakeup, etc) will
1108 use function graph tracing instead of function tracing.
1111 When set, a stack trace is recorded after any trace event
1115 Enable branch tracing with the tracer. This enables branch
1116 tracer along with the currently set tracer. Enabling this
1117 with the "nop" tracer is the same as just enabling the
1120 .. tip:: Some tracers have their own options. They only appear in this
1121 file when the tracer is active. They always appear in the
1125 Here are the per tracer options:
1127 Options for function tracer:
1130 When set, a stack trace is recorded after every
1131 function that is recorded. NOTE! Limit the functions
1132 that are recorded before enabling this, with
1133 "set_ftrace_filter" otherwise the system performance
1134 will be critically degraded. Remember to disable
1135 this option before clearing the function filter.
1137 Options for function_graph tracer:
1139 Since the function_graph tracer has a slightly different output
1140 it has its own options to control what is displayed.
1143 When set, the "overrun" of the graph stack is
1144 displayed after each function traced. The
1145 overrun, is when the stack depth of the calls
1146 is greater than what is reserved for each task.
1147 Each task has a fixed array of functions to
1148 trace in the call graph. If the depth of the
1149 calls exceeds that, the function is not traced.
1150 The overrun is the number of functions missed
1151 due to exceeding this array.
1154 When set, the CPU number of the CPU where the trace
1155 occurred is displayed.
1158 When set, if the function takes longer than
1159 A certain amount, then a delay marker is
1160 displayed. See "delay" above, under the
1164 Unlike other tracers, the process' command line
1165 is not displayed by default, but instead only
1166 when a task is traced in and out during a context
1167 switch. Enabling this options has the command
1168 of each process displayed at every line.
1171 At the end of each function (the return)
1172 the duration of the amount of time in the
1173 function is displayed in microseconds.
1176 When set, the timestamp is displayed at each line.
1179 When disabled, functions that happen inside an
1180 interrupt will not be traced.
1183 When set, the return event will include the function
1184 that it represents. By default this is off, and
1185 only a closing curly bracket "}" is displayed for
1186 the return of a function.
1189 When running function graph tracer, to include
1190 the time a task schedules out in its function.
1191 When enabled, it will account time the task has been
1192 scheduled out as part of the function call.
1195 When running function profiler with function graph tracer,
1196 to include the time to call nested functions. When this is
1197 not set, the time reported for the function will only
1198 include the time the function itself executed for, not the
1199 time for functions that it called.
1201 Options for blk tracer:
1204 Shows a more minimalistic output.
1210 When interrupts are disabled, the CPU can not react to any other
1211 external event (besides NMIs and SMIs). This prevents the timer
1212 interrupt from triggering or the mouse interrupt from letting
1213 the kernel know of a new mouse event. The result is a latency
1214 with the reaction time.
1216 The irqsoff tracer tracks the time for which interrupts are
1217 disabled. When a new maximum latency is hit, the tracer saves
1218 the trace leading up to that latency point so that every time a
1219 new maximum is reached, the old saved trace is discarded and the
1222 To reset the maximum, echo 0 into tracing_max_latency. Here is
1225 # echo 0 > options/function-trace
1226 # echo irqsoff > current_tracer
1227 # echo 1 > tracing_on
1228 # echo 0 > tracing_max_latency
1231 # echo 0 > tracing_on
1235 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1236 # --------------------------------------------------------------------
1237 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1239 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
1241 # => started at: run_timer_softirq
1242 # => ended at: run_timer_softirq
1246 # / _-----=> irqs-off
1247 # | / _----=> need-resched
1248 # || / _---=> hardirq/softirq
1249 # ||| / _--=> preempt-depth
1251 # cmd pid ||||| time | caller
1253 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
1254 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
1255 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
1256 <idle>-0 0dNs3 25us : <stack trace>
1257 => _raw_spin_unlock_irq
1258 => run_timer_softirq
1263 => smp_apic_timer_interrupt
1264 => apic_timer_interrupt
1269 => x86_64_start_reservations
1270 => x86_64_start_kernel
1272 Here we see that that we had a latency of 16 microseconds (which is
1273 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
1274 interrupts. The difference between the 16 and the displayed
1275 timestamp 25us occurred because the clock was incremented
1276 between the time of recording the max latency and the time of
1277 recording the function that had that latency.
1279 Note the above example had function-trace not set. If we set
1280 function-trace, we get a much larger output::
1282 with echo 1 > options/function-trace
1286 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1287 # --------------------------------------------------------------------
1288 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1290 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1292 # => started at: ata_scsi_queuecmd
1293 # => ended at: ata_scsi_queuecmd
1297 # / _-----=> irqs-off
1298 # | / _----=> need-resched
1299 # || / _---=> hardirq/softirq
1300 # ||| / _--=> preempt-depth
1302 # cmd pid ||||| time | caller
1304 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1305 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1306 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1307 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1308 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1309 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1310 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1311 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1312 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1314 bash-2042 3d..1 67us : delay_tsc <-__delay
1315 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1316 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1317 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1318 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1319 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1320 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1321 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1322 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1323 bash-2042 3d..1 120us : <stack trace>
1324 => _raw_spin_unlock_irqrestore
1325 => ata_scsi_queuecmd
1326 => scsi_dispatch_cmd
1328 => __blk_run_queue_uncond
1331 => generic_make_request
1334 => __ext3_get_inode_loc
1343 => user_path_at_empty
1348 => system_call_fastpath
1351 Here we traced a 71 microsecond latency. But we also see all the
1352 functions that were called during that time. Note that by
1353 enabling function tracing, we incur an added overhead. This
1354 overhead may extend the latency times. But nevertheless, this
1355 trace has provided some very helpful debugging information.
1361 When preemption is disabled, we may be able to receive
1362 interrupts but the task cannot be preempted and a higher
1363 priority task must wait for preemption to be enabled again
1364 before it can preempt a lower priority task.
1366 The preemptoff tracer traces the places that disable preemption.
1367 Like the irqsoff tracer, it records the maximum latency for
1368 which preemption was disabled. The control of preemptoff tracer
1369 is much like the irqsoff tracer.
1372 # echo 0 > options/function-trace
1373 # echo preemptoff > current_tracer
1374 # echo 1 > tracing_on
1375 # echo 0 > tracing_max_latency
1378 # echo 0 > tracing_on
1380 # tracer: preemptoff
1382 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1383 # --------------------------------------------------------------------
1384 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1386 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1388 # => started at: do_IRQ
1389 # => ended at: do_IRQ
1393 # / _-----=> irqs-off
1394 # | / _----=> need-resched
1395 # || / _---=> hardirq/softirq
1396 # ||| / _--=> preempt-depth
1398 # cmd pid ||||| time | caller
1400 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1401 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1402 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1403 sshd-1991 1d..1 52us : <stack trace>
1404 => sub_preempt_count
1410 This has some more changes. Preemption was disabled when an
1411 interrupt came in (notice the 'h'), and was enabled on exit.
1412 But we also see that interrupts have been disabled when entering
1413 the preempt off section and leaving it (the 'd'). We do not know if
1414 interrupts were enabled in the mean time or shortly after this
1418 # tracer: preemptoff
1420 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1421 # --------------------------------------------------------------------
1422 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1424 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1426 # => started at: wake_up_new_task
1427 # => ended at: task_rq_unlock
1431 # / _-----=> irqs-off
1432 # | / _----=> need-resched
1433 # || / _---=> hardirq/softirq
1434 # ||| / _--=> preempt-depth
1436 # cmd pid ||||| time | caller
1438 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1439 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1440 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1441 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1442 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1444 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1445 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1446 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1447 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1448 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1449 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1450 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1451 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1453 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1454 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1455 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1456 bash-1994 1d..2 36us : do_softirq <-irq_exit
1457 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1458 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1459 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1460 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1461 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1462 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1464 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1465 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1466 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1467 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1468 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1469 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1470 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1471 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1472 bash-1994 1.N.1 104us : <stack trace>
1473 => sub_preempt_count
1474 => _raw_spin_unlock_irqrestore
1482 The above is an example of the preemptoff trace with
1483 function-trace set. Here we see that interrupts were not disabled
1484 the entire time. The irq_enter code lets us know that we entered
1485 an interrupt 'h'. Before that, the functions being traced still
1486 show that it is not in an interrupt, but we can see from the
1487 functions themselves that this is not the case.
1492 Knowing the locations that have interrupts disabled or
1493 preemption disabled for the longest times is helpful. But
1494 sometimes we would like to know when either preemption and/or
1495 interrupts are disabled.
1497 Consider the following code::
1499 local_irq_disable();
1500 call_function_with_irqs_off();
1502 call_function_with_irqs_and_preemption_off();
1504 call_function_with_preemption_off();
1507 The irqsoff tracer will record the total length of
1508 call_function_with_irqs_off() and
1509 call_function_with_irqs_and_preemption_off().
1511 The preemptoff tracer will record the total length of
1512 call_function_with_irqs_and_preemption_off() and
1513 call_function_with_preemption_off().
1515 But neither will trace the time that interrupts and/or
1516 preemption is disabled. This total time is the time that we can
1517 not schedule. To record this time, use the preemptirqsoff
1520 Again, using this trace is much like the irqsoff and preemptoff
1524 # echo 0 > options/function-trace
1525 # echo preemptirqsoff > current_tracer
1526 # echo 1 > tracing_on
1527 # echo 0 > tracing_max_latency
1530 # echo 0 > tracing_on
1532 # tracer: preemptirqsoff
1534 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1535 # --------------------------------------------------------------------
1536 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1538 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1540 # => started at: ata_scsi_queuecmd
1541 # => ended at: ata_scsi_queuecmd
1545 # / _-----=> irqs-off
1546 # | / _----=> need-resched
1547 # || / _---=> hardirq/softirq
1548 # ||| / _--=> preempt-depth
1550 # cmd pid ||||| time | caller
1552 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1553 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1554 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1555 ls-2230 3...1 111us : <stack trace>
1556 => sub_preempt_count
1557 => _raw_spin_unlock_irqrestore
1558 => ata_scsi_queuecmd
1559 => scsi_dispatch_cmd
1561 => __blk_run_queue_uncond
1564 => generic_make_request
1569 => htree_dirblock_to_tree
1570 => ext3_htree_fill_tree
1574 => system_call_fastpath
1577 The trace_hardirqs_off_thunk is called from assembly on x86 when
1578 interrupts are disabled in the assembly code. Without the
1579 function tracing, we do not know if interrupts were enabled
1580 within the preemption points. We do see that it started with
1583 Here is a trace with function-trace set::
1585 # tracer: preemptirqsoff
1587 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1588 # --------------------------------------------------------------------
1589 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1591 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1593 # => started at: schedule
1594 # => ended at: mutex_unlock
1598 # / _-----=> irqs-off
1599 # | / _----=> need-resched
1600 # || / _---=> hardirq/softirq
1601 # ||| / _--=> preempt-depth
1603 # cmd pid ||||| time | caller
1605 kworker/-59 3...1 0us : __schedule <-schedule
1606 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1607 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1608 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1609 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1610 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1611 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1612 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1613 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1614 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1615 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1616 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1617 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1618 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1619 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1620 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1621 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1622 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1623 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1624 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1625 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1626 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1627 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1628 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1629 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1630 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1631 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1632 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1633 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1634 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1635 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1636 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1638 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1639 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1640 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1641 ls-2269 3d..3 21us : do_softirq <-irq_exit
1642 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1643 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1644 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1645 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1646 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1647 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1648 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1650 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1651 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1652 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1653 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1654 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1655 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1657 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1658 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1659 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1660 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1661 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1662 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1663 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1664 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1665 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1666 ls-2269 3d... 186us : <stack trace>
1667 => __mutex_unlock_slowpath
1674 => system_call_fastpath
1676 This is an interesting trace. It started with kworker running and
1677 scheduling out and ls taking over. But as soon as ls released the
1678 rq lock and enabled interrupts (but not preemption) an interrupt
1679 triggered. When the interrupt finished, it started running softirqs.
1680 But while the softirq was running, another interrupt triggered.
1681 When an interrupt is running inside a softirq, the annotation is 'H'.
1687 One common case that people are interested in tracing is the
1688 time it takes for a task that is woken to actually wake up.
1689 Now for non Real-Time tasks, this can be arbitrary. But tracing
1690 it none the less can be interesting.
1692 Without function tracing::
1694 # echo 0 > options/function-trace
1695 # echo wakeup > current_tracer
1696 # echo 1 > tracing_on
1697 # echo 0 > tracing_max_latency
1699 # echo 0 > tracing_on
1703 # wakeup latency trace v1.1.5 on 3.8.0-test+
1704 # --------------------------------------------------------------------
1705 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1707 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1711 # / _-----=> irqs-off
1712 # | / _----=> need-resched
1713 # || / _---=> hardirq/softirq
1714 # ||| / _--=> preempt-depth
1716 # cmd pid ||||| time | caller
1718 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1719 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1720 <idle>-0 3d..3 15us : __schedule <-schedule
1721 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1723 The tracer only traces the highest priority task in the system
1724 to avoid tracing the normal circumstances. Here we see that
1725 the kworker with a nice priority of -20 (not very nice), took
1726 just 15 microseconds from the time it woke up, to the time it
1729 Non Real-Time tasks are not that interesting. A more interesting
1730 trace is to concentrate only on Real-Time tasks.
1735 In a Real-Time environment it is very important to know the
1736 wakeup time it takes for the highest priority task that is woken
1737 up to the time that it executes. This is also known as "schedule
1738 latency". I stress the point that this is about RT tasks. It is
1739 also important to know the scheduling latency of non-RT tasks,
1740 but the average schedule latency is better for non-RT tasks.
1741 Tools like LatencyTop are more appropriate for such
1744 Real-Time environments are interested in the worst case latency.
1745 That is the longest latency it takes for something to happen,
1746 and not the average. We can have a very fast scheduler that may
1747 only have a large latency once in a while, but that would not
1748 work well with Real-Time tasks. The wakeup_rt tracer was designed
1749 to record the worst case wakeups of RT tasks. Non-RT tasks are
1750 not recorded because the tracer only records one worst case and
1751 tracing non-RT tasks that are unpredictable will overwrite the
1752 worst case latency of RT tasks (just run the normal wakeup
1753 tracer for a while to see that effect).
1755 Since this tracer only deals with RT tasks, we will run this
1756 slightly differently than we did with the previous tracers.
1757 Instead of performing an 'ls', we will run 'sleep 1' under
1758 'chrt' which changes the priority of the task.
1761 # echo 0 > options/function-trace
1762 # echo wakeup_rt > current_tracer
1763 # echo 1 > tracing_on
1764 # echo 0 > tracing_max_latency
1766 # echo 0 > tracing_on
1772 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1773 # --------------------------------------------------------------------
1774 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1776 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1780 # / _-----=> irqs-off
1781 # | / _----=> need-resched
1782 # || / _---=> hardirq/softirq
1783 # ||| / _--=> preempt-depth
1785 # cmd pid ||||| time | caller
1787 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1788 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1789 <idle>-0 3d..3 5us : __schedule <-schedule
1790 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1793 Running this on an idle system, we see that it only took 5 microseconds
1794 to perform the task switch. Note, since the trace point in the schedule
1795 is before the actual "switch", we stop the tracing when the recorded task
1796 is about to schedule in. This may change if we add a new marker at the
1797 end of the scheduler.
1799 Notice that the recorded task is 'sleep' with the PID of 2389
1800 and it has an rt_prio of 5. This priority is user-space priority
1801 and not the internal kernel priority. The policy is 1 for
1802 SCHED_FIFO and 2 for SCHED_RR.
1804 Note, that the trace data shows the internal priority (99 - rtprio).
1807 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1809 The 0:120:R means idle was running with a nice priority of 0 (120 - 120)
1810 and in the running state 'R'. The sleep task was scheduled in with
1811 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1812 and it too is in the running state.
1814 Doing the same with chrt -r 5 and function-trace set.
1817 echo 1 > options/function-trace
1821 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1822 # --------------------------------------------------------------------
1823 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1825 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
1829 # / _-----=> irqs-off
1830 # | / _----=> need-resched
1831 # || / _---=> hardirq/softirq
1832 # ||| / _--=> preempt-depth
1834 # cmd pid ||||| time | caller
1836 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
1837 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1838 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
1839 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
1840 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
1841 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
1842 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
1843 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
1844 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
1845 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1846 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
1847 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
1848 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
1849 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
1850 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
1851 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
1852 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
1853 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
1854 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
1855 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
1856 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
1857 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
1858 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
1859 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
1860 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
1861 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
1862 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
1863 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
1864 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
1865 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
1866 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
1867 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
1868 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
1869 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
1870 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
1871 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
1872 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
1873 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit
1874 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
1875 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
1876 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
1877 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
1878 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1879 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
1880 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
1881 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
1882 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
1883 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
1884 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
1885 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
1886 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
1887 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1888 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
1889 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1890 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1891 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
1892 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
1893 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
1894 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1895 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
1896 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
1897 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
1898 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
1899 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
1900 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
1901 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
1902 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
1903 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1904 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
1905 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
1906 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
1907 <idle>-0 3.N.. 25us : schedule <-cpu_idle
1908 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
1909 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
1910 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
1911 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
1912 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
1913 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
1914 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
1915 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
1916 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
1917 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
1918 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
1919 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
1920 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
1922 This isn't that big of a trace, even with function tracing enabled,
1923 so I included the entire trace.
1925 The interrupt went off while when the system was idle. Somewhere
1926 before task_woken_rt() was called, the NEED_RESCHED flag was set,
1927 this is indicated by the first occurrence of the 'N' flag.
1929 Latency tracing and events
1930 --------------------------
1931 As function tracing can induce a much larger latency, but without
1932 seeing what happens within the latency it is hard to know what
1933 caused it. There is a middle ground, and that is with enabling
1937 # echo 0 > options/function-trace
1938 # echo wakeup_rt > current_tracer
1939 # echo 1 > events/enable
1940 # echo 1 > tracing_on
1941 # echo 0 > tracing_max_latency
1943 # echo 0 > tracing_on
1947 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1948 # --------------------------------------------------------------------
1949 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1951 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
1955 # / _-----=> irqs-off
1956 # | / _----=> need-resched
1957 # || / _---=> hardirq/softirq
1958 # ||| / _--=> preempt-depth
1960 # cmd pid ||||| time | caller
1962 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
1963 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1964 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
1965 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
1966 <idle>-0 2.N.2 2us : power_end: cpu_id=2
1967 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
1968 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
1969 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
1970 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
1971 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
1972 <idle>-0 2d..3 6us : __schedule <-schedule
1973 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
1976 Hardware Latency Detector
1977 -------------------------
1979 The hardware latency detector is executed by enabling the "hwlat" tracer.
1981 NOTE, this tracer will affect the performance of the system as it will
1982 periodically make a CPU constantly busy with interrupts disabled.
1985 # echo hwlat > current_tracer
1991 # / _----=> need-resched
1992 # | / _---=> hardirq/softirq
1993 # || / _--=> preempt-depth
1995 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
1997 <...>-3638 [001] d... 19452.055471: #1 inner/outer(us): 12/14 ts:1499801089.066141940
1998 <...>-3638 [003] d... 19454.071354: #2 inner/outer(us): 11/9 ts:1499801091.082164365
1999 <...>-3638 [002] dn.. 19461.126852: #3 inner/outer(us): 12/9 ts:1499801098.138150062
2000 <...>-3638 [001] d... 19488.340960: #4 inner/outer(us): 8/12 ts:1499801125.354139633
2001 <...>-3638 [003] d... 19494.388553: #5 inner/outer(us): 8/12 ts:1499801131.402150961
2002 <...>-3638 [003] d... 19501.283419: #6 inner/outer(us): 0/12 ts:1499801138.297435289 nmi-total:4 nmi-count:1
2005 The above output is somewhat the same in the header. All events will have
2006 interrupts disabled 'd'. Under the FUNCTION title there is:
2009 This is the count of events recorded that were greater than the
2010 tracing_threshold (See below).
2012 inner/outer(us): 12/14
2014 This shows two numbers as "inner latency" and "outer latency". The test
2015 runs in a loop checking a timestamp twice. The latency detected within
2016 the two timestamps is the "inner latency" and the latency detected
2017 after the previous timestamp and the next timestamp in the loop is
2018 the "outer latency".
2020 ts:1499801089.066141940
2022 The absolute timestamp that the event happened.
2024 nmi-total:4 nmi-count:1
2026 On architectures that support it, if an NMI comes in during the
2027 test, the time spent in NMI is reported in "nmi-total" (in
2030 All architectures that have NMIs will show the "nmi-count" if an
2031 NMI comes in during the test.
2036 This gets automatically set to "10" to represent 10
2037 microseconds. This is the threshold of latency that
2038 needs to be detected before the trace will be recorded.
2040 Note, when hwlat tracer is finished (another tracer is
2041 written into "current_tracer"), the original value for
2042 tracing_threshold is placed back into this file.
2044 hwlat_detector/width
2045 The length of time the test runs with interrupts disabled.
2047 hwlat_detector/window
2048 The length of time of the window which the test
2049 runs. That is, the test will run for "width"
2050 microseconds per "window" microseconds
2053 When the test is started. A kernel thread is created that
2054 runs the test. This thread will alternate between CPUs
2055 listed in the tracing_cpumask between each period
2056 (one "window"). To limit the test to specific CPUs
2057 set the mask in this file to only the CPUs that the test
2063 This tracer is the function tracer. Enabling the function tracer
2064 can be done from the debug file system. Make sure the
2065 ftrace_enabled is set; otherwise this tracer is a nop.
2066 See the "ftrace_enabled" section below.
2069 # sysctl kernel.ftrace_enabled=1
2070 # echo function > current_tracer
2071 # echo 1 > tracing_on
2073 # echo 0 > tracing_on
2077 # entries-in-buffer/entries-written: 24799/24799 #P:4
2080 # / _----=> need-resched
2081 # | / _---=> hardirq/softirq
2082 # || / _--=> preempt-depth
2084 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2086 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
2087 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
2088 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
2089 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
2090 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
2091 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
2092 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
2093 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
2097 Note: function tracer uses ring buffers to store the above
2098 entries. The newest data may overwrite the oldest data.
2099 Sometimes using echo to stop the trace is not sufficient because
2100 the tracing could have overwritten the data that you wanted to
2101 record. For this reason, it is sometimes better to disable
2102 tracing directly from a program. This allows you to stop the
2103 tracing at the point that you hit the part that you are
2104 interested in. To disable the tracing directly from a C program,
2105 something like following code snippet can be used::
2109 int main(int argc, char *argv[]) {
2111 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
2113 if (condition_hit()) {
2114 write(trace_fd, "0", 1);
2120 Single thread tracing
2121 ---------------------
2123 By writing into set_ftrace_pid you can trace a
2124 single thread. For example::
2126 # cat set_ftrace_pid
2128 # echo 3111 > set_ftrace_pid
2129 # cat set_ftrace_pid
2131 # echo function > current_tracer
2135 # TASK-PID CPU# TIMESTAMP FUNCTION
2137 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
2138 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
2139 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
2140 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
2141 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
2142 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
2143 # echo > set_ftrace_pid
2147 # TASK-PID CPU# TIMESTAMP FUNCTION
2149 ##### CPU 3 buffer started ####
2150 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
2151 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
2152 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
2153 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
2154 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
2156 If you want to trace a function when executing, you could use
2157 something like this simple program.
2162 #include <sys/types.h>
2163 #include <sys/stat.h>
2169 #define STR(x) _STR(x)
2170 #define MAX_PATH 256
2172 const char *find_tracefs(void)
2174 static char tracefs[MAX_PATH+1];
2175 static int tracefs_found;
2182 if ((fp = fopen("/proc/mounts","r")) == NULL) {
2183 perror("/proc/mounts");
2187 while (fscanf(fp, "%*s %"
2189 "s %99s %*s %*d %*d\n",
2190 tracefs, type) == 2) {
2191 if (strcmp(type, "tracefs") == 0)
2196 if (strcmp(type, "tracefs") != 0) {
2197 fprintf(stderr, "tracefs not mounted");
2201 strcat(tracefs, "/tracing/");
2207 const char *tracing_file(const char *file_name)
2209 static char trace_file[MAX_PATH+1];
2210 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name);
2214 int main (int argc, char **argv)
2224 ffd = open(tracing_file("current_tracer"), O_WRONLY);
2227 write(ffd, "nop", 3);
2229 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
2230 s = sprintf(line, "%d\n", getpid());
2233 write(ffd, "function", 8);
2238 execvp(argv[1], argv+1);
2244 Or this simple script!
2249 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts`
2250 echo nop > $tracefs/tracing/current_tracer
2251 echo 0 > $tracefs/tracing/tracing_on
2252 echo $$ > $tracefs/tracing/set_ftrace_pid
2253 echo function > $tracefs/tracing/current_tracer
2254 echo 1 > $tracefs/tracing/tracing_on
2258 function graph tracer
2259 ---------------------------
2261 This tracer is similar to the function tracer except that it
2262 probes a function on its entry and its exit. This is done by
2263 using a dynamically allocated stack of return addresses in each
2264 task_struct. On function entry the tracer overwrites the return
2265 address of each function traced to set a custom probe. Thus the
2266 original return address is stored on the stack of return address
2269 Probing on both ends of a function leads to special features
2272 - measure of a function's time execution
2273 - having a reliable call stack to draw function calls graph
2275 This tracer is useful in several situations:
2277 - you want to find the reason of a strange kernel behavior and
2278 need to see what happens in detail on any areas (or specific
2281 - you are experiencing weird latencies but it's difficult to
2284 - you want to find quickly which path is taken by a specific
2287 - you just want to peek inside a working kernel and want to see
2292 # tracer: function_graph
2294 # CPU DURATION FUNCTION CALLS
2298 0) | do_sys_open() {
2300 0) | kmem_cache_alloc() {
2301 0) 1.382 us | __might_sleep();
2303 0) | strncpy_from_user() {
2304 0) | might_fault() {
2305 0) 1.389 us | __might_sleep();
2310 0) 0.668 us | _spin_lock();
2311 0) 0.570 us | expand_files();
2312 0) 0.586 us | _spin_unlock();
2315 There are several columns that can be dynamically
2316 enabled/disabled. You can use every combination of options you
2317 want, depending on your needs.
2319 - The cpu number on which the function executed is default
2320 enabled. It is sometimes better to only trace one cpu (see
2321 tracing_cpu_mask file) or you might sometimes see unordered
2322 function calls while cpu tracing switch.
2324 - hide: echo nofuncgraph-cpu > trace_options
2325 - show: echo funcgraph-cpu > trace_options
2327 - The duration (function's time of execution) is displayed on
2328 the closing bracket line of a function or on the same line
2329 than the current function in case of a leaf one. It is default
2332 - hide: echo nofuncgraph-duration > trace_options
2333 - show: echo funcgraph-duration > trace_options
2335 - The overhead field precedes the duration field in case of
2336 reached duration thresholds.
2338 - hide: echo nofuncgraph-overhead > trace_options
2339 - show: echo funcgraph-overhead > trace_options
2340 - depends on: funcgraph-duration
2344 3) # 1837.709 us | } /* __switch_to */
2345 3) | finish_task_switch() {
2346 3) 0.313 us | _raw_spin_unlock_irq();
2348 3) # 1889.063 us | } /* __schedule */
2349 3) ! 140.417 us | } /* __schedule */
2350 3) # 2034.948 us | } /* schedule */
2351 3) * 33998.59 us | } /* schedule_preempt_disabled */
2355 1) 0.260 us | msecs_to_jiffies();
2356 1) 0.313 us | __rcu_read_unlock();
2359 1) 0.313 us | rcu_bh_qs();
2360 1) 0.313 us | __local_bh_enable();
2362 1) 0.365 us | idle_cpu();
2363 1) | rcu_irq_exit() {
2364 1) 0.417 us | rcu_eqs_enter_common.isra.47();
2368 1) @ 119760.2 us | }
2374 2) 0.417 us | scheduler_ipi();
2384 + means that the function exceeded 10 usecs.
2385 ! means that the function exceeded 100 usecs.
2386 # means that the function exceeded 1000 usecs.
2387 * means that the function exceeded 10 msecs.
2388 @ means that the function exceeded 100 msecs.
2389 $ means that the function exceeded 1 sec.
2392 - The task/pid field displays the thread cmdline and pid which
2393 executed the function. It is default disabled.
2395 - hide: echo nofuncgraph-proc > trace_options
2396 - show: echo funcgraph-proc > trace_options
2400 # tracer: function_graph
2402 # CPU TASK/PID DURATION FUNCTION CALLS
2404 0) sh-4802 | | d_free() {
2405 0) sh-4802 | | call_rcu() {
2406 0) sh-4802 | | __call_rcu() {
2407 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2408 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2409 0) sh-4802 | 2.899 us | }
2410 0) sh-4802 | 4.040 us | }
2411 0) sh-4802 | 5.151 us | }
2412 0) sh-4802 | + 49.370 us | }
2415 - The absolute time field is an absolute timestamp given by the
2416 system clock since it started. A snapshot of this time is
2417 given on each entry/exit of functions
2419 - hide: echo nofuncgraph-abstime > trace_options
2420 - show: echo funcgraph-abstime > trace_options
2425 # TIME CPU DURATION FUNCTION CALLS
2427 360.774522 | 1) 0.541 us | }
2428 360.774522 | 1) 4.663 us | }
2429 360.774523 | 1) 0.541 us | __wake_up_bit();
2430 360.774524 | 1) 6.796 us | }
2431 360.774524 | 1) 7.952 us | }
2432 360.774525 | 1) 9.063 us | }
2433 360.774525 | 1) 0.615 us | journal_mark_dirty();
2434 360.774527 | 1) 0.578 us | __brelse();
2435 360.774528 | 1) | reiserfs_prepare_for_journal() {
2436 360.774528 | 1) | unlock_buffer() {
2437 360.774529 | 1) | wake_up_bit() {
2438 360.774529 | 1) | bit_waitqueue() {
2439 360.774530 | 1) 0.594 us | __phys_addr();
2442 The function name is always displayed after the closing bracket
2443 for a function if the start of that function is not in the
2446 Display of the function name after the closing bracket may be
2447 enabled for functions whose start is in the trace buffer,
2448 allowing easier searching with grep for function durations.
2449 It is default disabled.
2451 - hide: echo nofuncgraph-tail > trace_options
2452 - show: echo funcgraph-tail > trace_options
2454 Example with nofuncgraph-tail (default)::
2457 0) | kmem_cache_free() {
2458 0) 0.518 us | __phys_addr();
2462 Example with funcgraph-tail::
2465 0) | kmem_cache_free() {
2466 0) 0.518 us | __phys_addr();
2467 0) 1.757 us | } /* kmem_cache_free() */
2468 0) 2.861 us | } /* putname() */
2470 You can put some comments on specific functions by using
2471 trace_printk() For example, if you want to put a comment inside
2472 the __might_sleep() function, you just have to include
2473 <linux/ftrace.h> and call trace_printk() inside __might_sleep()::
2475 trace_printk("I'm a comment!\n")
2479 1) | __might_sleep() {
2480 1) | /* I'm a comment! */
2484 You might find other useful features for this tracer in the
2485 following "dynamic ftrace" section such as tracing only specific
2491 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2492 virtually no overhead when function tracing is disabled. The way
2493 this works is the mcount function call (placed at the start of
2494 every kernel function, produced by the -pg switch in gcc),
2495 starts of pointing to a simple return. (Enabling FTRACE will
2496 include the -pg switch in the compiling of the kernel.)
2498 At compile time every C file object is run through the
2499 recordmcount program (located in the scripts directory). This
2500 program will parse the ELF headers in the C object to find all
2501 the locations in the .text section that call mcount. Starting
2502 with gcc verson 4.6, the -mfentry has been added for x86, which
2503 calls "__fentry__" instead of "mcount". Which is called before
2504 the creation of the stack frame.
2506 Note, not all sections are traced. They may be prevented by either
2507 a notrace, or blocked another way and all inline functions are not
2508 traced. Check the "available_filter_functions" file to see what functions
2511 A section called "__mcount_loc" is created that holds
2512 references to all the mcount/fentry call sites in the .text section.
2513 The recordmcount program re-links this section back into the
2514 original object. The final linking stage of the kernel will add all these
2515 references into a single table.
2517 On boot up, before SMP is initialized, the dynamic ftrace code
2518 scans this table and updates all the locations into nops. It
2519 also records the locations, which are added to the
2520 available_filter_functions list. Modules are processed as they
2521 are loaded and before they are executed. When a module is
2522 unloaded, it also removes its functions from the ftrace function
2523 list. This is automatic in the module unload code, and the
2524 module author does not need to worry about it.
2526 When tracing is enabled, the process of modifying the function
2527 tracepoints is dependent on architecture. The old method is to use
2528 kstop_machine to prevent races with the CPUs executing code being
2529 modified (which can cause the CPU to do undesirable things, especially
2530 if the modified code crosses cache (or page) boundaries), and the nops are
2531 patched back to calls. But this time, they do not call mcount
2532 (which is just a function stub). They now call into the ftrace
2535 The new method of modifying the function tracepoints is to place
2536 a breakpoint at the location to be modified, sync all CPUs, modify
2537 the rest of the instruction not covered by the breakpoint. Sync
2538 all CPUs again, and then remove the breakpoint with the finished
2539 version to the ftrace call site.
2541 Some archs do not even need to monkey around with the synchronization,
2542 and can just slap the new code on top of the old without any
2543 problems with other CPUs executing it at the same time.
2545 One special side-effect to the recording of the functions being
2546 traced is that we can now selectively choose which functions we
2547 wish to trace and which ones we want the mcount calls to remain
2550 Two files are used, one for enabling and one for disabling the
2551 tracing of specified functions. They are:
2559 A list of available functions that you can add to these files is
2562 available_filter_functions
2566 # cat available_filter_functions
2575 If I am only interested in sys_nanosleep and hrtimer_interrupt::
2577 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2578 # echo function > current_tracer
2579 # echo 1 > tracing_on
2581 # echo 0 > tracing_on
2585 # entries-in-buffer/entries-written: 5/5 #P:4
2588 # / _----=> need-resched
2589 # | / _---=> hardirq/softirq
2590 # || / _--=> preempt-depth
2592 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2594 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2595 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2596 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2597 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2598 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2600 To see which functions are being traced, you can cat the file:
2603 # cat set_ftrace_filter
2608 Perhaps this is not enough. The filters also allow glob(7) matching.
2611 will match functions that begin with <match>
2613 will match functions that end with <match>
2615 will match functions that have <match> in it
2616 ``<match1>*<match2>``
2617 will match functions that begin with <match1> and end with <match2>
2620 It is better to use quotes to enclose the wild cards,
2621 otherwise the shell may expand the parameters into names
2622 of files in the local directory.
2626 # echo 'hrtimer_*' > set_ftrace_filter
2632 # entries-in-buffer/entries-written: 897/897 #P:4
2635 # / _----=> need-resched
2636 # | / _---=> hardirq/softirq
2637 # || / _--=> preempt-depth
2639 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2641 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2642 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2643 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2644 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2645 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2646 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2647 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2648 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2650 Notice that we lost the sys_nanosleep.
2653 # cat set_ftrace_filter
2658 hrtimer_try_to_cancel
2662 hrtimer_force_reprogram
2663 hrtimer_get_next_event
2667 hrtimer_get_remaining
2669 hrtimer_init_sleeper
2672 This is because the '>' and '>>' act just like they do in bash.
2673 To rewrite the filters, use '>'
2674 To append to the filters, use '>>'
2676 To clear out a filter so that all functions will be recorded
2679 # echo > set_ftrace_filter
2680 # cat set_ftrace_filter
2683 Again, now we want to append.
2687 # echo sys_nanosleep > set_ftrace_filter
2688 # cat set_ftrace_filter
2690 # echo 'hrtimer_*' >> set_ftrace_filter
2691 # cat set_ftrace_filter
2696 hrtimer_try_to_cancel
2700 hrtimer_force_reprogram
2701 hrtimer_get_next_event
2706 hrtimer_get_remaining
2708 hrtimer_init_sleeper
2711 The set_ftrace_notrace prevents those functions from being
2715 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2721 # entries-in-buffer/entries-written: 39608/39608 #P:4
2724 # / _----=> need-resched
2725 # | / _---=> hardirq/softirq
2726 # || / _--=> preempt-depth
2728 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2730 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2731 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2732 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2733 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2734 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2735 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2736 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2737 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2738 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2739 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2740 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2741 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2743 We can see that there's no more lock or preempt tracing.
2746 Dynamic ftrace with the function graph tracer
2747 ---------------------------------------------
2749 Although what has been explained above concerns both the
2750 function tracer and the function-graph-tracer, there are some
2751 special features only available in the function-graph tracer.
2753 If you want to trace only one function and all of its children,
2754 you just have to echo its name into set_graph_function::
2756 echo __do_fault > set_graph_function
2758 will produce the following "expanded" trace of the __do_fault()
2762 0) | filemap_fault() {
2763 0) | find_lock_page() {
2764 0) 0.804 us | find_get_page();
2765 0) | __might_sleep() {
2769 0) 0.653 us | _spin_lock();
2770 0) 0.578 us | page_add_file_rmap();
2771 0) 0.525 us | native_set_pte_at();
2772 0) 0.585 us | _spin_unlock();
2773 0) | unlock_page() {
2774 0) 0.541 us | page_waitqueue();
2775 0) 0.639 us | __wake_up_bit();
2779 0) | filemap_fault() {
2780 0) | find_lock_page() {
2781 0) 0.698 us | find_get_page();
2782 0) | __might_sleep() {
2786 0) 0.631 us | _spin_lock();
2787 0) 0.571 us | page_add_file_rmap();
2788 0) 0.526 us | native_set_pte_at();
2789 0) 0.586 us | _spin_unlock();
2790 0) | unlock_page() {
2791 0) 0.533 us | page_waitqueue();
2792 0) 0.638 us | __wake_up_bit();
2796 You can also expand several functions at once::
2798 echo sys_open > set_graph_function
2799 echo sys_close >> set_graph_function
2801 Now if you want to go back to trace all functions you can clear
2802 this special filter via::
2804 echo > set_graph_function
2810 Note, the proc sysctl ftrace_enable is a big on/off switch for the
2811 function tracer. By default it is enabled (when function tracing is
2812 enabled in the kernel). If it is disabled, all function tracing is
2813 disabled. This includes not only the function tracers for ftrace, but
2814 also for any other uses (perf, kprobes, stack tracing, profiling, etc).
2816 Please disable this with care.
2818 This can be disable (and enabled) with::
2820 sysctl kernel.ftrace_enabled=0
2821 sysctl kernel.ftrace_enabled=1
2825 echo 0 > /proc/sys/kernel/ftrace_enabled
2826 echo 1 > /proc/sys/kernel/ftrace_enabled
2832 A few commands are supported by the set_ftrace_filter interface.
2833 Trace commands have the following format::
2835 <function>:<command>:<parameter>
2837 The following commands are supported:
2840 This command enables function filtering per module. The
2841 parameter defines the module. For example, if only the write*
2842 functions in the ext3 module are desired, run:
2844 echo 'write*:mod:ext3' > set_ftrace_filter
2846 This command interacts with the filter in the same way as
2847 filtering based on function names. Thus, adding more functions
2848 in a different module is accomplished by appending (>>) to the
2849 filter file. Remove specific module functions by prepending
2852 echo '!writeback*:mod:ext3' >> set_ftrace_filter
2854 Mod command supports module globbing. Disable tracing for all
2855 functions except a specific module::
2857 echo '!*:mod:!ext3' >> set_ftrace_filter
2859 Disable tracing for all modules, but still trace kernel::
2861 echo '!*:mod:*' >> set_ftrace_filter
2863 Enable filter only for kernel::
2865 echo '*write*:mod:!*' >> set_ftrace_filter
2867 Enable filter for module globbing::
2869 echo '*write*:mod:*snd*' >> set_ftrace_filter
2872 These commands turn tracing on and off when the specified
2873 functions are hit. The parameter determines how many times the
2874 tracing system is turned on and off. If unspecified, there is
2875 no limit. For example, to disable tracing when a schedule bug
2876 is hit the first 5 times, run::
2878 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
2880 To always disable tracing when __schedule_bug is hit::
2882 echo '__schedule_bug:traceoff' > set_ftrace_filter
2884 These commands are cumulative whether or not they are appended
2885 to set_ftrace_filter. To remove a command, prepend it by '!'
2886 and drop the parameter::
2888 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
2890 The above removes the traceoff command for __schedule_bug
2891 that have a counter. To remove commands without counters::
2893 echo '!__schedule_bug:traceoff' > set_ftrace_filter
2896 Will cause a snapshot to be triggered when the function is hit.
2899 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
2901 To only snapshot once:
2904 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
2906 To remove the above commands::
2908 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
2909 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
2911 - enable_event/disable_event:
2912 These commands can enable or disable a trace event. Note, because
2913 function tracing callbacks are very sensitive, when these commands
2914 are registered, the trace point is activated, but disabled in
2915 a "soft" mode. That is, the tracepoint will be called, but
2916 just will not be traced. The event tracepoint stays in this mode
2917 as long as there's a command that triggers it.
2920 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
2925 <function>:enable_event:<system>:<event>[:count]
2926 <function>:disable_event:<system>:<event>[:count]
2928 To remove the events commands::
2930 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
2932 echo '!schedule:disable_event:sched:sched_switch' > \
2936 When the function is hit, it will dump the contents of the ftrace
2937 ring buffer to the console. This is useful if you need to debug
2938 something, and want to dump the trace when a certain function
2939 is hit. Perhaps its a function that is called before a tripple
2940 fault happens and does not allow you to get a regular dump.
2943 When the function is hit, it will dump the contents of the ftrace
2944 ring buffer for the current CPU to the console. Unlike the "dump"
2945 command, it only prints out the contents of the ring buffer for the
2946 CPU that executed the function that triggered the dump.
2951 The trace_pipe outputs the same content as the trace file, but
2952 the effect on the tracing is different. Every read from
2953 trace_pipe is consumed. This means that subsequent reads will be
2954 different. The trace is live.
2957 # echo function > current_tracer
2958 # cat trace_pipe > /tmp/trace.out &
2960 # echo 1 > tracing_on
2962 # echo 0 > tracing_on
2966 # entries-in-buffer/entries-written: 0/0 #P:4
2969 # / _----=> need-resched
2970 # | / _---=> hardirq/softirq
2971 # || / _--=> preempt-depth
2973 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2977 # cat /tmp/trace.out
2978 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
2979 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
2980 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
2981 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
2982 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
2983 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
2984 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
2985 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
2986 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
2989 Note, reading the trace_pipe file will block until more input is
2995 Having too much or not enough data can be troublesome in
2996 diagnosing an issue in the kernel. The file buffer_size_kb is
2997 used to modify the size of the internal trace buffers. The
2998 number listed is the number of entries that can be recorded per
2999 CPU. To know the full size, multiply the number of possible CPUs
3000 with the number of entries.
3003 # cat buffer_size_kb
3004 1408 (units kilobytes)
3006 Or simply read buffer_total_size_kb
3009 # cat buffer_total_size_kb
3012 To modify the buffer, simple echo in a number (in 1024 byte segments).
3015 # echo 10000 > buffer_size_kb
3016 # cat buffer_size_kb
3017 10000 (units kilobytes)
3019 It will try to allocate as much as possible. If you allocate too
3020 much, it can cause Out-Of-Memory to trigger.
3023 # echo 1000000000000 > buffer_size_kb
3024 -bash: echo: write error: Cannot allocate memory
3025 # cat buffer_size_kb
3028 The per_cpu buffers can be changed individually as well:
3031 # echo 10000 > per_cpu/cpu0/buffer_size_kb
3032 # echo 100 > per_cpu/cpu1/buffer_size_kb
3034 When the per_cpu buffers are not the same, the buffer_size_kb
3035 at the top level will just show an X
3038 # cat buffer_size_kb
3041 This is where the buffer_total_size_kb is useful:
3044 # cat buffer_total_size_kb
3047 Writing to the top level buffer_size_kb will reset all the buffers
3048 to be the same again.
3052 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
3053 available to all non latency tracers. (Latency tracers which
3054 record max latency, such as "irqsoff" or "wakeup", can't use
3055 this feature, since those are already using the snapshot
3056 mechanism internally.)
3058 Snapshot preserves a current trace buffer at a particular point
3059 in time without stopping tracing. Ftrace swaps the current
3060 buffer with a spare buffer, and tracing continues in the new
3061 current (=previous spare) buffer.
3063 The following tracefs files in "tracing" are related to this
3068 This is used to take a snapshot and to read the output
3069 of the snapshot. Echo 1 into this file to allocate a
3070 spare buffer and to take a snapshot (swap), then read
3071 the snapshot from this file in the same format as
3072 "trace" (described above in the section "The File
3073 System"). Both reads snapshot and tracing are executable
3074 in parallel. When the spare buffer is allocated, echoing
3075 0 frees it, and echoing else (positive) values clear the
3077 More details are shown in the table below.
3079 +--------------+------------+------------+------------+
3080 |status\\input | 0 | 1 | else |
3081 +==============+============+============+============+
3082 |not allocated |(do nothing)| alloc+swap |(do nothing)|
3083 +--------------+------------+------------+------------+
3084 |allocated | free | swap | clear |
3085 +--------------+------------+------------+------------+
3087 Here is an example of using the snapshot feature.
3090 # echo 1 > events/sched/enable
3095 # entries-in-buffer/entries-written: 71/71 #P:8
3098 # / _----=> need-resched
3099 # | / _---=> hardirq/softirq
3100 # || / _--=> preempt-depth
3102 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3104 <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
3105 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
3107 <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
3112 # entries-in-buffer/entries-written: 77/77 #P:8
3115 # / _----=> need-resched
3116 # | / _---=> hardirq/softirq
3117 # || / _--=> preempt-depth
3119 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3121 <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
3122 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
3126 If you try to use this snapshot feature when current tracer is
3127 one of the latency tracers, you will get the following results.
3130 # echo wakeup > current_tracer
3132 bash: echo: write error: Device or resource busy
3134 cat: snapshot: Device or resource busy
3139 In the tracefs tracing directory is a directory called "instances".
3140 This directory can have new directories created inside of it using
3141 mkdir, and removing directories with rmdir. The directory created
3142 with mkdir in this directory will already contain files and other
3143 directories after it is created.
3146 # mkdir instances/foo
3148 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
3149 set_event snapshot trace trace_clock trace_marker trace_options
3150 trace_pipe tracing_on
3152 As you can see, the new directory looks similar to the tracing directory
3153 itself. In fact, it is very similar, except that the buffer and
3154 events are agnostic from the main director, or from any other
3155 instances that are created.
3157 The files in the new directory work just like the files with the
3158 same name in the tracing directory except the buffer that is used
3159 is a separate and new buffer. The files affect that buffer but do not
3160 affect the main buffer with the exception of trace_options. Currently,
3161 the trace_options affect all instances and the top level buffer
3162 the same, but this may change in future releases. That is, options
3163 may become specific to the instance they reside in.
3165 Notice that none of the function tracer files are there, nor is
3166 current_tracer and available_tracers. This is because the buffers
3167 can currently only have events enabled for them.
3170 # mkdir instances/foo
3171 # mkdir instances/bar
3172 # mkdir instances/zoot
3173 # echo 100000 > buffer_size_kb
3174 # echo 1000 > instances/foo/buffer_size_kb
3175 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
3176 # echo function > current_trace
3177 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
3178 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
3179 # echo 1 > instances/foo/events/sched/sched_switch/enable
3180 # echo 1 > instances/bar/events/irq/enable
3181 # echo 1 > instances/zoot/events/syscalls/enable
3183 CPU:2 [LOST 11745 EVENTS]
3184 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
3185 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
3186 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
3187 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
3188 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
3189 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
3190 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
3191 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
3192 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3193 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3194 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
3197 # cat instances/foo/trace_pipe
3198 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3199 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3200 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
3201 <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
3202 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
3203 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3204 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3205 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
3206 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
3207 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
3210 # cat instances/bar/trace_pipe
3211 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
3212 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
3213 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
3214 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
3215 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
3216 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
3217 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
3218 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
3219 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
3220 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
3221 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
3222 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
3225 # cat instances/zoot/trace
3228 # entries-in-buffer/entries-written: 18996/18996 #P:4
3231 # / _----=> need-resched
3232 # | / _---=> hardirq/softirq
3233 # || / _--=> preempt-depth
3235 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3237 bash-1998 [000] d... 140.733501: sys_write -> 0x2
3238 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
3239 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
3240 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
3241 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
3242 bash-1998 [000] d... 140.733510: sys_close(fd: a)
3243 bash-1998 [000] d... 140.733510: sys_close -> 0x0
3244 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
3245 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
3246 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
3247 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
3249 You can see that the trace of the top most trace buffer shows only
3250 the function tracing. The foo instance displays wakeups and task
3253 To remove the instances, simply delete their directories:
3256 # rmdir instances/foo
3257 # rmdir instances/bar
3258 # rmdir instances/zoot
3260 Note, if a process has a trace file open in one of the instance
3261 directories, the rmdir will fail with EBUSY.
3266 Since the kernel has a fixed sized stack, it is important not to
3267 waste it in functions. A kernel developer must be conscience of
3268 what they allocate on the stack. If they add too much, the system
3269 can be in danger of a stack overflow, and corruption will occur,
3270 usually leading to a system panic.
3272 There are some tools that check this, usually with interrupts
3273 periodically checking usage. But if you can perform a check
3274 at every function call that will become very useful. As ftrace provides
3275 a function tracer, it makes it convenient to check the stack size
3276 at every function call. This is enabled via the stack tracer.
3278 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
3279 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
3282 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
3284 You can also enable it from the kernel command line to trace
3285 the stack size of the kernel during boot up, by adding "stacktrace"
3286 to the kernel command line parameter.
3288 After running it for a few minutes, the output looks like:
3291 # cat stack_max_size
3295 Depth Size Location (18 entries)
3297 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
3298 1) 2704 160 find_busiest_group+0x31/0x1f1
3299 2) 2544 256 load_balance+0xd9/0x662
3300 3) 2288 80 idle_balance+0xbb/0x130
3301 4) 2208 128 __schedule+0x26e/0x5b9
3302 5) 2080 16 schedule+0x64/0x66
3303 6) 2064 128 schedule_timeout+0x34/0xe0
3304 7) 1936 112 wait_for_common+0x97/0xf1
3305 8) 1824 16 wait_for_completion+0x1d/0x1f
3306 9) 1808 128 flush_work+0xfe/0x119
3307 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
3308 11) 1664 48 input_available_p+0x1d/0x5c
3309 12) 1616 48 n_tty_poll+0x6d/0x134
3310 13) 1568 64 tty_poll+0x64/0x7f
3311 14) 1504 880 do_select+0x31e/0x511
3312 15) 624 400 core_sys_select+0x177/0x216
3313 16) 224 96 sys_select+0x91/0xb9
3314 17) 128 128 system_call_fastpath+0x16/0x1b
3316 Note, if -mfentry is being used by gcc, functions get traced before
3317 they set up the stack frame. This means that leaf level functions
3318 are not tested by the stack tracer when -mfentry is used.
3320 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
3324 More details can be found in the source code, in the `kernel/trace/*.c` files.