1 ========================
2 ftrace - Function Tracer
3 ========================
5 Copyright 2008 Red Hat Inc.
7 :Author: Steven Rostedt <srostedt@redhat.com>
8 :License: The GNU Free Documentation License, Version 1.2
9 (dual licensed under the GPL v2)
10 :Original Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton,
11 John Kacur, and David Teigland.
13 - Written for: 2.6.28-rc2
15 - Updated for: 4.13 - Copyright 2017 VMware Inc. Steven Rostedt
16 - Converted to rst format - Changbin Du <changbin.du@intel.com>
21 Ftrace is an internal tracer designed to help out developers and
22 designers of systems to find what is going on inside the kernel.
23 It can be used for debugging or analyzing latencies and
24 performance issues that take place outside of user-space.
26 Although ftrace is typically considered the function tracer, it
27 is really a framework of several assorted tracing utilities.
28 There's latency tracing to examine what occurs between interrupts
29 disabled and enabled, as well as for preemption and from a time
30 a task is woken to the task is actually scheduled in.
32 One of the most common uses of ftrace is the event tracing.
33 Throughout the kernel is hundreds of static event points that
34 can be enabled via the tracefs file system to see what is
35 going on in certain parts of the kernel.
37 See events.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.
227 This influences the tracers "function" and "function_graph"
228 and thus also function profiling (see "function_profile_enabled").
230 The functions listed in "available_filter_functions" are what
231 can be written into this file.
233 This interface also allows for commands to be used. See the
234 "Filter commands" section for more details.
238 This has an effect opposite to that of
239 set_ftrace_filter. Any function that is added here will not
240 be traced. If a function exists in both set_ftrace_filter
241 and set_ftrace_notrace, the function will _not_ be traced.
245 Have the function tracer only trace the threads whose PID are
248 If the "function-fork" option is set, then when a task whose
249 PID is listed in this file forks, the child's PID will
250 automatically be added to this file, and the child will be
251 traced by the function tracer as well. This option will also
252 cause PIDs of tasks that exit to be removed from the file.
256 Have the events only trace a task with a PID listed in this file.
257 Note, sched_switch and sched_wake_up will also trace events
260 To have the PIDs of children of tasks with their PID in this file
261 added on fork, enable the "event-fork" option. That option will also
262 cause the PIDs of tasks to be removed from this file when the task
267 Functions listed in this file will cause the function graph
268 tracer to only trace these functions and the functions that
269 they call. (See the section "dynamic ftrace" for more details).
270 Note, set_ftrace_filter and set_ftrace_notrace still affects
271 what functions are being traced.
275 Similar to set_graph_function, but will disable function graph
276 tracing when the function is hit until it exits the function.
277 This makes it possible to ignore tracing functions that are called
278 by a specific function.
280 available_filter_functions:
282 This lists the functions that ftrace has processed and can trace.
283 These are the function names that you can pass to
284 "set_ftrace_filter", "set_ftrace_notrace",
285 "set_graph_function", or "set_graph_notrace".
286 (See the section "dynamic ftrace" below for more details.)
288 dyn_ftrace_total_info:
290 This file is for debugging purposes. The number of functions that
291 have been converted to nops and are available to be traced.
295 This file is more for debugging ftrace, but can also be useful
296 in seeing if any function has a callback attached to it.
297 Not only does the trace infrastructure use ftrace function
298 trace utility, but other subsystems might too. This file
299 displays all functions that have a callback attached to them
300 as well as the number of callbacks that have been attached.
301 Note, a callback may also call multiple functions which will
302 not be listed in this count.
304 If the callback registered to be traced by a function with
305 the "save regs" attribute (thus even more overhead), a 'R'
306 will be displayed on the same line as the function that
307 is returning registers.
309 If the callback registered to be traced by a function with
310 the "ip modify" attribute (thus the regs->ip can be changed),
311 an 'I' will be displayed on the same line as the function that
314 If the architecture supports it, it will also show what callback
315 is being directly called by the function. If the count is greater
316 than 1 it most likely will be ftrace_ops_list_func().
318 If the callback of the function jumps to a trampoline that is
319 specific to a the callback and not the standard trampoline,
320 its address will be printed as well as the function that the
323 function_profile_enabled:
325 When set it will enable all functions with either the function
326 tracer, or if configured, the function graph tracer. It will
327 keep a histogram of the number of functions that were called
328 and if the function graph tracer was configured, it will also keep
329 track of the time spent in those functions. The histogram
330 content can be displayed in the files:
332 trace_stat/function<cpu> ( function0, function1, etc).
336 A directory that holds different tracing stats.
340 Enable dynamic trace points. See kprobetrace.txt.
344 Dynamic trace points stats. See kprobetrace.txt.
348 Used with the function graph tracer. This is the max depth
349 it will trace into a function. Setting this to a value of
350 one will show only the first kernel function that is called
355 This is for tools that read the raw format files. If an event in
356 the ring buffer references a string, only a pointer to the string
357 is recorded into the buffer and not the string itself. This prevents
358 tools from knowing what that string was. This file displays the string
359 and address for the string allowing tools to map the pointers to what
364 Only the pid of the task is recorded in a trace event unless
365 the event specifically saves the task comm as well. Ftrace
366 makes a cache of pid mappings to comms to try to display
367 comms for events. If a pid for a comm is not listed, then
368 "<...>" is displayed in the output.
370 If the option "record-cmd" is set to "0", then comms of tasks
371 will not be saved during recording. By default, it is enabled.
375 By default, 128 comms are saved (see "saved_cmdlines" above). To
376 increase or decrease the amount of comms that are cached, echo
377 in a the number of comms to cache, into this file.
381 If the option "record-tgid" is set, on each scheduling context switch
382 the Task Group ID of a task is saved in a table mapping the PID of
383 the thread to its TGID. By default, the "record-tgid" option is
388 This displays the "snapshot" buffer and also lets the user
389 take a snapshot of the current running trace.
390 See the "Snapshot" section below for more details.
394 When the stack tracer is activated, this will display the
395 maximum stack size it has encountered.
396 See the "Stack Trace" section below.
400 This displays the stack back trace of the largest stack
401 that was encountered when the stack tracer is activated.
402 See the "Stack Trace" section below.
406 This is similar to "set_ftrace_filter" but it limits what
407 functions the stack tracer will check.
411 Whenever an event is recorded into the ring buffer, a
412 "timestamp" is added. This stamp comes from a specified
413 clock. By default, ftrace uses the "local" clock. This
414 clock is very fast and strictly per cpu, but on some
415 systems it may not be monotonic with respect to other
416 CPUs. In other words, the local clocks may not be in sync
417 with local clocks on other CPUs.
419 Usual clocks for tracing::
422 [local] global counter x86-tsc
424 The clock with the square brackets around it is the one in effect.
427 Default clock, but may not be in sync across CPUs
430 This clock is in sync with all CPUs but may
431 be a bit slower than the local clock.
434 This is not a clock at all, but literally an atomic
435 counter. It counts up one by one, but is in sync
436 with all CPUs. This is useful when you need to
437 know exactly the order events occurred with respect to
438 each other on different CPUs.
441 This uses the jiffies counter and the time stamp
442 is relative to the time since boot up.
445 This makes ftrace use the same clock that perf uses.
446 Eventually perf will be able to read ftrace buffers
447 and this will help out in interleaving the data.
450 Architectures may define their own clocks. For
451 example, x86 uses its own TSC cycle clock here.
454 This uses the powerpc timebase register value.
455 This is in sync across CPUs and can also be used
456 to correlate events across hypervisor/guest if
460 This uses the fast monotonic clock (CLOCK_MONOTONIC)
461 which is monotonic and is subject to NTP rate adjustments.
464 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
465 which is monotonic but is not subject to any rate adjustments
466 and ticks at the same rate as the hardware clocksource.
469 This is the boot clock (CLOCK_BOOTTIME) and is based on the
470 fast monotonic clock, but also accounts for time spent in
471 suspend. Since the clock access is designed for use in
472 tracing in the suspend path, some side effects are possible
473 if clock is accessed after the suspend time is accounted before
474 the fast mono clock is updated. In this case, the clock update
475 appears to happen slightly sooner than it normally would have.
476 Also on 32-bit systems, it's possible that the 64-bit boot offset
477 sees a partial update. These effects are rare and post
478 processing should be able to handle them. See comments in the
479 ktime_get_boot_fast_ns() function for more information.
481 To set a clock, simply echo the clock name into this file::
483 # echo global > trace_clock
487 This is a very useful file for synchronizing user space
488 with events happening in the kernel. Writing strings into
489 this file will be written into the ftrace buffer.
491 It is useful in applications to open this file at the start
492 of the application and just reference the file descriptor
495 void trace_write(const char *fmt, ...)
505 n = vsnprintf(buf, 256, fmt, ap);
508 write(trace_fd, buf, n);
513 trace_fd = open("trace_marker", WR_ONLY);
515 Note: Writing into the trace_marker file can also initiate triggers
516 that are written into /sys/kernel/tracing/events/ftrace/print/trigger
517 See "Event triggers" in Documentation/trace/events.rst and an
518 example in Documentation/trace/histogram.rst (Section 3.)
522 This is similar to trace_marker above, but is meant for for binary data
523 to be written to it, where a tool can be used to parse the data
528 Add dynamic tracepoints in programs.
533 Uprobe statistics. See uprobetrace.txt
537 This is a way to make multiple trace buffers where different
538 events can be recorded in different buffers.
539 See "Instances" section below.
543 This is the trace event directory. It holds event tracepoints
544 (also known as static tracepoints) that have been compiled
545 into the kernel. It shows what event tracepoints exist
546 and how they are grouped by system. There are "enable"
547 files at various levels that can enable the tracepoints
548 when a "1" is written to them.
550 See events.txt for more information.
554 By echoing in the event into this file, will enable that event.
556 See events.txt for more information.
560 A list of events that can be enabled in tracing.
562 See events.txt for more information.
566 Certain tracers may change the timestamp mode used when
567 logging trace events into the event buffer. Events with
568 different modes can coexist within a buffer but the mode in
569 effect when an event is logged determines which timestamp mode
570 is used for that event. The default timestamp mode is
573 Usual timestamp modes for tracing:
578 The timestamp mode with the square brackets around it is the
581 delta: Default timestamp mode - timestamp is a delta against
582 a per-buffer timestamp.
584 absolute: The timestamp is a full timestamp, not a delta
585 against some other value. As such it takes up more
586 space and is less efficient.
590 Directory for the Hardware Latency Detector.
591 See "Hardware Latency Detector" section below.
595 This is a directory that contains the trace per_cpu information.
597 per_cpu/cpu0/buffer_size_kb:
599 The ftrace buffer is defined per_cpu. That is, there's a separate
600 buffer for each CPU to allow writes to be done atomically,
601 and free from cache bouncing. These buffers may have different
602 size buffers. This file is similar to the buffer_size_kb
603 file, but it only displays or sets the buffer size for the
604 specific CPU. (here cpu0).
608 This is similar to the "trace" file, but it will only display
609 the data specific for the CPU. If written to, it only clears
610 the specific CPU buffer.
612 per_cpu/cpu0/trace_pipe
614 This is similar to the "trace_pipe" file, and is a consuming
615 read, but it will only display (and consume) the data specific
618 per_cpu/cpu0/trace_pipe_raw
620 For tools that can parse the ftrace ring buffer binary format,
621 the trace_pipe_raw file can be used to extract the data
622 from the ring buffer directly. With the use of the splice()
623 system call, the buffer data can be quickly transferred to
624 a file or to the network where a server is collecting the
627 Like trace_pipe, this is a consuming reader, where multiple
628 reads will always produce different data.
630 per_cpu/cpu0/snapshot:
632 This is similar to the main "snapshot" file, but will only
633 snapshot the current CPU (if supported). It only displays
634 the content of the snapshot for a given CPU, and if
635 written to, only clears this CPU buffer.
637 per_cpu/cpu0/snapshot_raw:
639 Similar to the trace_pipe_raw, but will read the binary format
640 from the snapshot buffer for the given CPU.
644 This displays certain stats about the ring buffer:
647 The number of events that are still in the buffer.
650 The number of lost events due to overwriting when
654 Should always be zero.
655 This gets set if so many events happened within a nested
656 event (ring buffer is re-entrant), that it fills the
657 buffer and starts dropping events.
660 Bytes actually read (not overwritten).
663 The oldest timestamp in the buffer
666 The current timestamp
669 Events lost due to overwrite option being off.
672 The number of events read.
677 Here is the list of current tracers that may be configured.
681 Function call tracer to trace all kernel functions.
685 Similar to the function tracer except that the
686 function tracer probes the functions on their entry
687 whereas the function graph tracer traces on both entry
688 and exit of the functions. It then provides the ability
689 to draw a graph of function calls similar to C code
694 The block tracer. The tracer used by the blktrace user
699 The Hardware Latency tracer is used to detect if the hardware
700 produces any latency. See "Hardware Latency Detector" section
705 Traces the areas that disable interrupts and saves
706 the trace with the longest max latency.
707 See tracing_max_latency. When a new max is recorded,
708 it replaces the old trace. It is best to view this
709 trace with the latency-format option enabled, which
710 happens automatically when the tracer is selected.
714 Similar to irqsoff but traces and records the amount of
715 time for which preemption is disabled.
719 Similar to irqsoff and preemptoff, but traces and
720 records the largest time for which irqs and/or preemption
725 Traces and records the max latency that it takes for
726 the highest priority task to get scheduled after
727 it has been woken up.
728 Traces all tasks as an average developer would expect.
732 Traces and records the max latency that it takes for just
733 RT tasks (as the current "wakeup" does). This is useful
734 for those interested in wake up timings of RT tasks.
738 Traces and records the max latency that it takes for
739 a SCHED_DEADLINE task to be woken (as the "wakeup" and
744 A special tracer that is used to trace binary module.
745 It will trace all the calls that a module makes to the
746 hardware. Everything it writes and reads from the I/O
751 This tracer can be configured when tracing likely/unlikely
752 calls within the kernel. It will trace when a likely and
753 unlikely branch is hit and if it was correct in its prediction
758 This is the "trace nothing" tracer. To remove all
759 tracers from tracing simply echo "nop" into
763 Examples of using the tracer
764 ----------------------------
766 Here are typical examples of using the tracers when controlling
767 them only with the tracefs interface (without using any
768 user-land utilities).
773 Here is an example of the output format of the file "trace"::
777 # entries-in-buffer/entries-written: 140080/250280 #P:4
780 # / _----=> need-resched
781 # | / _---=> hardirq/softirq
782 # || / _--=> preempt-depth
784 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
786 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
787 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
788 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
789 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
790 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
791 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
792 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
793 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
794 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
795 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
798 A header is printed with the tracer name that is represented by
799 the trace. In this case the tracer is "function". Then it shows the
800 number of events in the buffer as well as the total number of entries
801 that were written. The difference is the number of entries that were
802 lost due to the buffer filling up (250280 - 140080 = 110200 events
805 The header explains the content of the events. Task name "bash", the task
806 PID "1977", the CPU that it was running on "000", the latency format
807 (explained below), the timestamp in <secs>.<usecs> format, the
808 function name that was traced "sys_close" and the parent function that
809 called this function "system_call_fastpath". The timestamp is the time
810 at which the function was entered.
815 When the latency-format option is enabled or when one of the latency
816 tracers is set, the trace file gives somewhat more information to see
817 why a latency happened. Here is a typical trace::
821 # irqsoff latency trace v1.1.5 on 3.8.0-test+
822 # --------------------------------------------------------------------
823 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
825 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
827 # => started at: __lock_task_sighand
828 # => ended at: _raw_spin_unlock_irqrestore
832 # / _-----=> irqs-off
833 # | / _----=> need-resched
834 # || / _---=> hardirq/softirq
835 # ||| / _--=> preempt-depth
837 # cmd pid ||||| time | caller
839 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
840 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
841 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
842 ps-6143 2d..1 306us : <stack trace>
843 => trace_hardirqs_on_caller
845 => _raw_spin_unlock_irqrestore
852 => system_call_fastpath
855 This shows that the current tracer is "irqsoff" tracing the time
856 for which interrupts were disabled. It gives the trace version (which
857 never changes) and the version of the kernel upon which this was executed on
858 (3.8). Then it displays the max latency in microseconds (259 us). The number
859 of trace entries displayed and the total number (both are four: #4/4).
860 VP, KP, SP, and HP are always zero and are reserved for later use.
861 #P is the number of online CPUs (#P:4).
863 The task is the process that was running when the latency
864 occurred. (ps pid: 6143).
866 The start and stop (the functions in which the interrupts were
867 disabled and enabled respectively) that caused the latencies:
869 - __lock_task_sighand is where the interrupts were disabled.
870 - _raw_spin_unlock_irqrestore is where they were enabled again.
872 The next lines after the header are the trace itself. The header
873 explains which is which.
875 cmd: The name of the process in the trace.
877 pid: The PID of that process.
879 CPU#: The CPU which the process was running on.
881 irqs-off: 'd' interrupts are disabled. '.' otherwise.
882 .. caution:: If the architecture does not support a way to
883 read the irq flags variable, an 'X' will always
887 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
888 - 'n' only TIF_NEED_RESCHED is set,
889 - 'p' only PREEMPT_NEED_RESCHED is set,
893 - 'Z' - NMI occurred inside a hardirq
894 - 'z' - NMI is running
895 - 'H' - hard irq occurred inside a softirq.
896 - 'h' - hard irq is running
897 - 's' - soft irq is running
898 - '.' - normal context.
900 preempt-depth: The level of preempt_disabled
902 The above is mostly meaningful for kernel developers.
905 When the latency-format option is enabled, the trace file
906 output includes a timestamp relative to the start of the
907 trace. This differs from the output when latency-format
908 is disabled, which includes an absolute timestamp.
911 This is just to help catch your eye a bit better. And
912 needs to be fixed to be only relative to the same CPU.
913 The marks are determined by the difference between this
914 current trace and the next trace.
916 - '$' - greater than 1 second
917 - '@' - greater than 100 millisecond
918 - '*' - greater than 10 millisecond
919 - '#' - greater than 1000 microsecond
920 - '!' - greater than 100 microsecond
921 - '+' - greater than 10 microsecond
922 - ' ' - less than or equal to 10 microsecond.
924 The rest is the same as the 'trace' file.
926 Note, the latency tracers will usually end with a back trace
927 to easily find where the latency occurred.
932 The trace_options file (or the options directory) is used to control
933 what gets printed in the trace output, or manipulate the tracers.
934 To see what is available, simply cat the file::
965 To disable one of the options, echo in the option prepended with
968 echo noprint-parent > trace_options
970 To enable an option, leave off the "no"::
972 echo sym-offset > trace_options
974 Here are the available options:
977 On function traces, display the calling (parent)
978 function as well as the function being traced.
982 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
985 bash-4000 [01] 1477.606694: simple_strtoul
989 Display not only the function name, but also the
990 offset in the function. For example, instead of
991 seeing just "ktime_get", you will see
992 "ktime_get+0xb/0x20".
996 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
999 This will also display the function address as well
1000 as the function name.
1004 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
1007 This deals with the trace file when the
1008 latency-format option is enabled.
1011 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
1012 (+0.000ms): simple_strtoul (kstrtoul)
1015 This will display raw numbers. This option is best for
1016 use with user applications that can translate the raw
1017 numbers better than having it done in the kernel.
1020 Similar to raw, but the numbers will be in a hexadecimal format.
1023 This will print out the formats in raw binary.
1026 When set, reading trace_pipe will not block when polled.
1029 Can disable trace_printk() from writing into the buffer.
1032 It is sometimes confusing when the CPU buffers are full
1033 and one CPU buffer had a lot of events recently, thus
1034 a shorter time frame, were another CPU may have only had
1035 a few events, which lets it have older events. When
1036 the trace is reported, it shows the oldest events first,
1037 and it may look like only one CPU ran (the one with the
1038 oldest events). When the annotate option is set, it will
1039 display when a new CPU buffer started::
1041 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
1042 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
1043 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
1044 ##### CPU 2 buffer started ####
1045 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
1046 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
1047 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
1050 This option changes the trace. It records a
1051 stacktrace of the current user space thread after
1055 when user stacktrace are enabled, look up which
1056 object the address belongs to, and print a
1057 relative address. This is especially useful when
1058 ASLR is on, otherwise you don't get a chance to
1059 resolve the address to object/file/line after
1060 the app is no longer running
1062 The lookup is performed when you read
1063 trace,trace_pipe. Example::
1065 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
1066 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
1070 When set, trace_printk()s will only show the format
1071 and not their parameters (if trace_bprintk() or
1072 trace_bputs() was used to save the trace_printk()).
1075 Show only the event data. Hides the comm, PID,
1076 timestamp, CPU, and other useful data.
1079 This option changes the trace output. When it is enabled,
1080 the trace displays additional information about the
1081 latency, as described in "Latency trace format".
1084 When any event or tracer is enabled, a hook is enabled
1085 in the sched_switch trace point to fill comm cache
1086 with mapped pids and comms. But this may cause some
1087 overhead, and if you only care about pids, and not the
1088 name of the task, disabling this option can lower the
1089 impact of tracing. See "saved_cmdlines".
1092 When any event or tracer is enabled, a hook is enabled
1093 in the sched_switch trace point to fill the cache of
1094 mapped Thread Group IDs (TGID) mapping to pids. See
1098 This controls what happens when the trace buffer is
1099 full. If "1" (default), the oldest events are
1100 discarded and overwritten. If "0", then the newest
1101 events are discarded.
1102 (see per_cpu/cpu0/stats for overrun and dropped)
1105 When the free_buffer is closed, tracing will
1106 stop (tracing_on set to 0).
1109 Shows the interrupt, preempt count, need resched data.
1110 When disabled, the trace looks like::
1114 # entries-in-buffer/entries-written: 144405/9452052 #P:4
1116 # TASK-PID CPU# TIMESTAMP FUNCTION
1118 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
1119 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
1120 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
1124 When set, the trace_marker is writable (only by root).
1125 When disabled, the trace_marker will error with EINVAL
1129 When set, tasks with PIDs listed in set_event_pid will have
1130 the PIDs of their children added to set_event_pid when those
1131 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
1132 their PIDs will be removed from the file.
1135 The latency tracers will enable function tracing
1136 if this option is enabled (default it is). When
1137 it is disabled, the latency tracers do not trace
1138 functions. This keeps the overhead of the tracer down
1139 when performing latency tests.
1142 When set, tasks with PIDs listed in set_ftrace_pid will
1143 have the PIDs of their children added to set_ftrace_pid
1144 when those tasks fork. Also, when tasks with PIDs in
1145 set_ftrace_pid exit, their PIDs will be removed from the
1149 When set, the latency tracers (irqsoff, wakeup, etc) will
1150 use function graph tracing instead of function tracing.
1153 When set, a stack trace is recorded after any trace event
1157 Enable branch tracing with the tracer. This enables branch
1158 tracer along with the currently set tracer. Enabling this
1159 with the "nop" tracer is the same as just enabling the
1162 .. tip:: Some tracers have their own options. They only appear in this
1163 file when the tracer is active. They always appear in the
1167 Here are the per tracer options:
1169 Options for function tracer:
1172 When set, a stack trace is recorded after every
1173 function that is recorded. NOTE! Limit the functions
1174 that are recorded before enabling this, with
1175 "set_ftrace_filter" otherwise the system performance
1176 will be critically degraded. Remember to disable
1177 this option before clearing the function filter.
1179 Options for function_graph tracer:
1181 Since the function_graph tracer has a slightly different output
1182 it has its own options to control what is displayed.
1185 When set, the "overrun" of the graph stack is
1186 displayed after each function traced. The
1187 overrun, is when the stack depth of the calls
1188 is greater than what is reserved for each task.
1189 Each task has a fixed array of functions to
1190 trace in the call graph. If the depth of the
1191 calls exceeds that, the function is not traced.
1192 The overrun is the number of functions missed
1193 due to exceeding this array.
1196 When set, the CPU number of the CPU where the trace
1197 occurred is displayed.
1200 When set, if the function takes longer than
1201 A certain amount, then a delay marker is
1202 displayed. See "delay" above, under the
1206 Unlike other tracers, the process' command line
1207 is not displayed by default, but instead only
1208 when a task is traced in and out during a context
1209 switch. Enabling this options has the command
1210 of each process displayed at every line.
1213 At the end of each function (the return)
1214 the duration of the amount of time in the
1215 function is displayed in microseconds.
1218 When set, the timestamp is displayed at each line.
1221 When disabled, functions that happen inside an
1222 interrupt will not be traced.
1225 When set, the return event will include the function
1226 that it represents. By default this is off, and
1227 only a closing curly bracket "}" is displayed for
1228 the return of a function.
1231 When running function graph tracer, to include
1232 the time a task schedules out in its function.
1233 When enabled, it will account time the task has been
1234 scheduled out as part of the function call.
1237 When running function profiler with function graph tracer,
1238 to include the time to call nested functions. When this is
1239 not set, the time reported for the function will only
1240 include the time the function itself executed for, not the
1241 time for functions that it called.
1243 Options for blk tracer:
1246 Shows a more minimalistic output.
1252 When interrupts are disabled, the CPU can not react to any other
1253 external event (besides NMIs and SMIs). This prevents the timer
1254 interrupt from triggering or the mouse interrupt from letting
1255 the kernel know of a new mouse event. The result is a latency
1256 with the reaction time.
1258 The irqsoff tracer tracks the time for which interrupts are
1259 disabled. When a new maximum latency is hit, the tracer saves
1260 the trace leading up to that latency point so that every time a
1261 new maximum is reached, the old saved trace is discarded and the
1264 To reset the maximum, echo 0 into tracing_max_latency. Here is
1267 # echo 0 > options/function-trace
1268 # echo irqsoff > current_tracer
1269 # echo 1 > tracing_on
1270 # echo 0 > tracing_max_latency
1273 # echo 0 > tracing_on
1277 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1278 # --------------------------------------------------------------------
1279 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1281 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
1283 # => started at: run_timer_softirq
1284 # => ended at: run_timer_softirq
1288 # / _-----=> irqs-off
1289 # | / _----=> need-resched
1290 # || / _---=> hardirq/softirq
1291 # ||| / _--=> preempt-depth
1293 # cmd pid ||||| time | caller
1295 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
1296 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
1297 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
1298 <idle>-0 0dNs3 25us : <stack trace>
1299 => _raw_spin_unlock_irq
1300 => run_timer_softirq
1305 => smp_apic_timer_interrupt
1306 => apic_timer_interrupt
1311 => x86_64_start_reservations
1312 => x86_64_start_kernel
1314 Here we see that that we had a latency of 16 microseconds (which is
1315 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
1316 interrupts. The difference between the 16 and the displayed
1317 timestamp 25us occurred because the clock was incremented
1318 between the time of recording the max latency and the time of
1319 recording the function that had that latency.
1321 Note the above example had function-trace not set. If we set
1322 function-trace, we get a much larger output::
1324 with echo 1 > options/function-trace
1328 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1329 # --------------------------------------------------------------------
1330 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1332 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1334 # => started at: ata_scsi_queuecmd
1335 # => ended at: ata_scsi_queuecmd
1339 # / _-----=> irqs-off
1340 # | / _----=> need-resched
1341 # || / _---=> hardirq/softirq
1342 # ||| / _--=> preempt-depth
1344 # cmd pid ||||| time | caller
1346 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1347 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1348 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1349 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1350 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1351 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1352 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1353 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1354 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1356 bash-2042 3d..1 67us : delay_tsc <-__delay
1357 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1358 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1359 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1360 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1361 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1362 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1363 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1364 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1365 bash-2042 3d..1 120us : <stack trace>
1366 => _raw_spin_unlock_irqrestore
1367 => ata_scsi_queuecmd
1368 => scsi_dispatch_cmd
1370 => __blk_run_queue_uncond
1373 => generic_make_request
1376 => __ext3_get_inode_loc
1385 => user_path_at_empty
1390 => system_call_fastpath
1393 Here we traced a 71 microsecond latency. But we also see all the
1394 functions that were called during that time. Note that by
1395 enabling function tracing, we incur an added overhead. This
1396 overhead may extend the latency times. But nevertheless, this
1397 trace has provided some very helpful debugging information.
1399 If we prefer function graph output instead of function, we can set
1400 display-graph option::
1401 with echo 1 > options/display-graph
1405 # irqsoff latency trace v1.1.5 on 4.20.0-rc6+
1406 # --------------------------------------------------------------------
1407 # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4)
1409 # | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0)
1411 # => started at: free_debug_processing
1412 # => ended at: return_to_handler
1416 # / _----=> need-resched
1417 # | / _---=> hardirq/softirq
1418 # || / _--=> preempt-depth
1420 # REL TIME CPU TASK/PID |||| DURATION FUNCTION CALLS
1421 # | | | | |||| | | | | | |
1422 0 us | 0) bash-1507 | d... | 0.000 us | _raw_spin_lock_irqsave();
1423 0 us | 0) bash-1507 | d..1 | 0.378 us | do_raw_spin_trylock();
1424 1 us | 0) bash-1507 | d..2 | | set_track() {
1425 2 us | 0) bash-1507 | d..2 | | save_stack_trace() {
1426 2 us | 0) bash-1507 | d..2 | | __save_stack_trace() {
1427 3 us | 0) bash-1507 | d..2 | | __unwind_start() {
1428 3 us | 0) bash-1507 | d..2 | | get_stack_info() {
1429 3 us | 0) bash-1507 | d..2 | 0.351 us | in_task_stack();
1430 4 us | 0) bash-1507 | d..2 | 1.107 us | }
1432 3750 us | 0) bash-1507 | d..1 | 0.516 us | do_raw_spin_unlock();
1433 3750 us | 0) bash-1507 | d..1 | 0.000 us | _raw_spin_unlock_irqrestore();
1434 3764 us | 0) bash-1507 | d..1 | 0.000 us | tracer_hardirqs_on();
1435 bash-1507 0d..1 3792us : <stack trace>
1436 => free_debug_processing
1445 => search_binary_handler
1446 => __do_execve_file.isra.32
1449 => entry_SYSCALL_64_after_hwframe
1454 When preemption is disabled, we may be able to receive
1455 interrupts but the task cannot be preempted and a higher
1456 priority task must wait for preemption to be enabled again
1457 before it can preempt a lower priority task.
1459 The preemptoff tracer traces the places that disable preemption.
1460 Like the irqsoff tracer, it records the maximum latency for
1461 which preemption was disabled. The control of preemptoff tracer
1462 is much like the irqsoff tracer.
1465 # echo 0 > options/function-trace
1466 # echo preemptoff > current_tracer
1467 # echo 1 > tracing_on
1468 # echo 0 > tracing_max_latency
1471 # echo 0 > tracing_on
1473 # tracer: preemptoff
1475 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1476 # --------------------------------------------------------------------
1477 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1479 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1481 # => started at: do_IRQ
1482 # => ended at: do_IRQ
1486 # / _-----=> irqs-off
1487 # | / _----=> need-resched
1488 # || / _---=> hardirq/softirq
1489 # ||| / _--=> preempt-depth
1491 # cmd pid ||||| time | caller
1493 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1494 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1495 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1496 sshd-1991 1d..1 52us : <stack trace>
1497 => sub_preempt_count
1503 This has some more changes. Preemption was disabled when an
1504 interrupt came in (notice the 'h'), and was enabled on exit.
1505 But we also see that interrupts have been disabled when entering
1506 the preempt off section and leaving it (the 'd'). We do not know if
1507 interrupts were enabled in the mean time or shortly after this
1511 # tracer: preemptoff
1513 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1514 # --------------------------------------------------------------------
1515 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1517 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1519 # => started at: wake_up_new_task
1520 # => ended at: task_rq_unlock
1524 # / _-----=> irqs-off
1525 # | / _----=> need-resched
1526 # || / _---=> hardirq/softirq
1527 # ||| / _--=> preempt-depth
1529 # cmd pid ||||| time | caller
1531 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1532 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1533 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1534 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1535 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1537 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1538 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1539 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1540 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1541 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1542 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1543 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1544 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1546 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1547 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1548 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1549 bash-1994 1d..2 36us : do_softirq <-irq_exit
1550 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1551 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1552 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1553 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1554 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1555 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1557 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1558 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1559 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1560 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1561 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1562 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1563 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1564 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1565 bash-1994 1.N.1 104us : <stack trace>
1566 => sub_preempt_count
1567 => _raw_spin_unlock_irqrestore
1575 The above is an example of the preemptoff trace with
1576 function-trace set. Here we see that interrupts were not disabled
1577 the entire time. The irq_enter code lets us know that we entered
1578 an interrupt 'h'. Before that, the functions being traced still
1579 show that it is not in an interrupt, but we can see from the
1580 functions themselves that this is not the case.
1585 Knowing the locations that have interrupts disabled or
1586 preemption disabled for the longest times is helpful. But
1587 sometimes we would like to know when either preemption and/or
1588 interrupts are disabled.
1590 Consider the following code::
1592 local_irq_disable();
1593 call_function_with_irqs_off();
1595 call_function_with_irqs_and_preemption_off();
1597 call_function_with_preemption_off();
1600 The irqsoff tracer will record the total length of
1601 call_function_with_irqs_off() and
1602 call_function_with_irqs_and_preemption_off().
1604 The preemptoff tracer will record the total length of
1605 call_function_with_irqs_and_preemption_off() and
1606 call_function_with_preemption_off().
1608 But neither will trace the time that interrupts and/or
1609 preemption is disabled. This total time is the time that we can
1610 not schedule. To record this time, use the preemptirqsoff
1613 Again, using this trace is much like the irqsoff and preemptoff
1617 # echo 0 > options/function-trace
1618 # echo preemptirqsoff > current_tracer
1619 # echo 1 > tracing_on
1620 # echo 0 > tracing_max_latency
1623 # echo 0 > tracing_on
1625 # tracer: preemptirqsoff
1627 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1628 # --------------------------------------------------------------------
1629 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1631 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1633 # => started at: ata_scsi_queuecmd
1634 # => ended at: ata_scsi_queuecmd
1638 # / _-----=> irqs-off
1639 # | / _----=> need-resched
1640 # || / _---=> hardirq/softirq
1641 # ||| / _--=> preempt-depth
1643 # cmd pid ||||| time | caller
1645 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1646 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1647 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1648 ls-2230 3...1 111us : <stack trace>
1649 => sub_preempt_count
1650 => _raw_spin_unlock_irqrestore
1651 => ata_scsi_queuecmd
1652 => scsi_dispatch_cmd
1654 => __blk_run_queue_uncond
1657 => generic_make_request
1662 => htree_dirblock_to_tree
1663 => ext3_htree_fill_tree
1667 => system_call_fastpath
1670 The trace_hardirqs_off_thunk is called from assembly on x86 when
1671 interrupts are disabled in the assembly code. Without the
1672 function tracing, we do not know if interrupts were enabled
1673 within the preemption points. We do see that it started with
1676 Here is a trace with function-trace set::
1678 # tracer: preemptirqsoff
1680 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1681 # --------------------------------------------------------------------
1682 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1684 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1686 # => started at: schedule
1687 # => ended at: mutex_unlock
1691 # / _-----=> irqs-off
1692 # | / _----=> need-resched
1693 # || / _---=> hardirq/softirq
1694 # ||| / _--=> preempt-depth
1696 # cmd pid ||||| time | caller
1698 kworker/-59 3...1 0us : __schedule <-schedule
1699 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1700 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1701 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1702 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1703 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1704 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1705 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1706 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1707 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1708 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1709 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1710 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1711 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1712 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1713 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1714 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1715 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1716 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1717 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1718 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1719 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1720 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1721 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1722 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1723 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1724 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1725 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1726 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1727 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1728 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1729 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1731 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1732 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1733 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1734 ls-2269 3d..3 21us : do_softirq <-irq_exit
1735 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1736 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1737 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1738 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1739 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1740 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1741 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1743 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1744 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1745 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1746 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1747 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1748 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1750 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1751 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1752 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1753 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1754 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1755 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1756 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1757 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1758 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1759 ls-2269 3d... 186us : <stack trace>
1760 => __mutex_unlock_slowpath
1767 => system_call_fastpath
1769 This is an interesting trace. It started with kworker running and
1770 scheduling out and ls taking over. But as soon as ls released the
1771 rq lock and enabled interrupts (but not preemption) an interrupt
1772 triggered. When the interrupt finished, it started running softirqs.
1773 But while the softirq was running, another interrupt triggered.
1774 When an interrupt is running inside a softirq, the annotation is 'H'.
1780 One common case that people are interested in tracing is the
1781 time it takes for a task that is woken to actually wake up.
1782 Now for non Real-Time tasks, this can be arbitrary. But tracing
1783 it none the less can be interesting.
1785 Without function tracing::
1787 # echo 0 > options/function-trace
1788 # echo wakeup > current_tracer
1789 # echo 1 > tracing_on
1790 # echo 0 > tracing_max_latency
1792 # echo 0 > tracing_on
1796 # wakeup latency trace v1.1.5 on 3.8.0-test+
1797 # --------------------------------------------------------------------
1798 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1800 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1804 # / _-----=> irqs-off
1805 # | / _----=> need-resched
1806 # || / _---=> hardirq/softirq
1807 # ||| / _--=> preempt-depth
1809 # cmd pid ||||| time | caller
1811 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1812 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1813 <idle>-0 3d..3 15us : __schedule <-schedule
1814 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1816 The tracer only traces the highest priority task in the system
1817 to avoid tracing the normal circumstances. Here we see that
1818 the kworker with a nice priority of -20 (not very nice), took
1819 just 15 microseconds from the time it woke up, to the time it
1822 Non Real-Time tasks are not that interesting. A more interesting
1823 trace is to concentrate only on Real-Time tasks.
1828 In a Real-Time environment it is very important to know the
1829 wakeup time it takes for the highest priority task that is woken
1830 up to the time that it executes. This is also known as "schedule
1831 latency". I stress the point that this is about RT tasks. It is
1832 also important to know the scheduling latency of non-RT tasks,
1833 but the average schedule latency is better for non-RT tasks.
1834 Tools like LatencyTop are more appropriate for such
1837 Real-Time environments are interested in the worst case latency.
1838 That is the longest latency it takes for something to happen,
1839 and not the average. We can have a very fast scheduler that may
1840 only have a large latency once in a while, but that would not
1841 work well with Real-Time tasks. The wakeup_rt tracer was designed
1842 to record the worst case wakeups of RT tasks. Non-RT tasks are
1843 not recorded because the tracer only records one worst case and
1844 tracing non-RT tasks that are unpredictable will overwrite the
1845 worst case latency of RT tasks (just run the normal wakeup
1846 tracer for a while to see that effect).
1848 Since this tracer only deals with RT tasks, we will run this
1849 slightly differently than we did with the previous tracers.
1850 Instead of performing an 'ls', we will run 'sleep 1' under
1851 'chrt' which changes the priority of the task.
1854 # echo 0 > options/function-trace
1855 # echo wakeup_rt > current_tracer
1856 # echo 1 > tracing_on
1857 # echo 0 > tracing_max_latency
1859 # echo 0 > tracing_on
1865 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1866 # --------------------------------------------------------------------
1867 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1869 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1873 # / _-----=> irqs-off
1874 # | / _----=> need-resched
1875 # || / _---=> hardirq/softirq
1876 # ||| / _--=> preempt-depth
1878 # cmd pid ||||| time | caller
1880 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1881 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1882 <idle>-0 3d..3 5us : __schedule <-schedule
1883 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1886 Running this on an idle system, we see that it only took 5 microseconds
1887 to perform the task switch. Note, since the trace point in the schedule
1888 is before the actual "switch", we stop the tracing when the recorded task
1889 is about to schedule in. This may change if we add a new marker at the
1890 end of the scheduler.
1892 Notice that the recorded task is 'sleep' with the PID of 2389
1893 and it has an rt_prio of 5. This priority is user-space priority
1894 and not the internal kernel priority. The policy is 1 for
1895 SCHED_FIFO and 2 for SCHED_RR.
1897 Note, that the trace data shows the internal priority (99 - rtprio).
1900 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1902 The 0:120:R means idle was running with a nice priority of 0 (120 - 120)
1903 and in the running state 'R'. The sleep task was scheduled in with
1904 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1905 and it too is in the running state.
1907 Doing the same with chrt -r 5 and function-trace set.
1910 echo 1 > options/function-trace
1914 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1915 # --------------------------------------------------------------------
1916 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1918 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
1922 # / _-----=> irqs-off
1923 # | / _----=> need-resched
1924 # || / _---=> hardirq/softirq
1925 # ||| / _--=> preempt-depth
1927 # cmd pid ||||| time | caller
1929 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
1930 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1931 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
1932 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
1933 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
1934 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
1935 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
1936 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
1937 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
1938 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1939 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
1940 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
1941 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
1942 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
1943 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
1944 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
1945 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
1946 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
1947 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
1948 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
1949 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
1950 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
1951 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
1952 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
1953 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
1954 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
1955 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
1956 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
1957 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
1958 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
1959 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
1960 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
1961 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
1962 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
1963 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
1964 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
1965 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
1966 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit
1967 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
1968 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
1969 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
1970 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
1971 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1972 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
1973 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
1974 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
1975 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
1976 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
1977 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
1978 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
1979 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
1980 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1981 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
1982 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1983 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1984 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
1985 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
1986 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
1987 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1988 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
1989 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
1990 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
1991 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
1992 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
1993 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
1994 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
1995 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
1996 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1997 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
1998 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
1999 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
2000 <idle>-0 3.N.. 25us : schedule <-cpu_idle
2001 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
2002 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
2003 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
2004 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
2005 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
2006 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
2007 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
2008 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
2009 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
2010 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
2011 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
2012 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
2013 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
2015 This isn't that big of a trace, even with function tracing enabled,
2016 so I included the entire trace.
2018 The interrupt went off while when the system was idle. Somewhere
2019 before task_woken_rt() was called, the NEED_RESCHED flag was set,
2020 this is indicated by the first occurrence of the 'N' flag.
2022 Latency tracing and events
2023 --------------------------
2024 As function tracing can induce a much larger latency, but without
2025 seeing what happens within the latency it is hard to know what
2026 caused it. There is a middle ground, and that is with enabling
2030 # echo 0 > options/function-trace
2031 # echo wakeup_rt > current_tracer
2032 # echo 1 > events/enable
2033 # echo 1 > tracing_on
2034 # echo 0 > tracing_max_latency
2036 # echo 0 > tracing_on
2040 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
2041 # --------------------------------------------------------------------
2042 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
2044 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
2048 # / _-----=> irqs-off
2049 # | / _----=> need-resched
2050 # || / _---=> hardirq/softirq
2051 # ||| / _--=> preempt-depth
2053 # cmd pid ||||| time | caller
2055 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
2056 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
2057 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
2058 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
2059 <idle>-0 2.N.2 2us : power_end: cpu_id=2
2060 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
2061 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
2062 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
2063 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
2064 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
2065 <idle>-0 2d..3 6us : __schedule <-schedule
2066 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
2069 Hardware Latency Detector
2070 -------------------------
2072 The hardware latency detector is executed by enabling the "hwlat" tracer.
2074 NOTE, this tracer will affect the performance of the system as it will
2075 periodically make a CPU constantly busy with interrupts disabled.
2078 # echo hwlat > current_tracer
2084 # / _----=> need-resched
2085 # | / _---=> hardirq/softirq
2086 # || / _--=> preempt-depth
2088 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2090 <...>-3638 [001] d... 19452.055471: #1 inner/outer(us): 12/14 ts:1499801089.066141940
2091 <...>-3638 [003] d... 19454.071354: #2 inner/outer(us): 11/9 ts:1499801091.082164365
2092 <...>-3638 [002] dn.. 19461.126852: #3 inner/outer(us): 12/9 ts:1499801098.138150062
2093 <...>-3638 [001] d... 19488.340960: #4 inner/outer(us): 8/12 ts:1499801125.354139633
2094 <...>-3638 [003] d... 19494.388553: #5 inner/outer(us): 8/12 ts:1499801131.402150961
2095 <...>-3638 [003] d... 19501.283419: #6 inner/outer(us): 0/12 ts:1499801138.297435289 nmi-total:4 nmi-count:1
2098 The above output is somewhat the same in the header. All events will have
2099 interrupts disabled 'd'. Under the FUNCTION title there is:
2102 This is the count of events recorded that were greater than the
2103 tracing_threshold (See below).
2105 inner/outer(us): 12/14
2107 This shows two numbers as "inner latency" and "outer latency". The test
2108 runs in a loop checking a timestamp twice. The latency detected within
2109 the two timestamps is the "inner latency" and the latency detected
2110 after the previous timestamp and the next timestamp in the loop is
2111 the "outer latency".
2113 ts:1499801089.066141940
2115 The absolute timestamp that the event happened.
2117 nmi-total:4 nmi-count:1
2119 On architectures that support it, if an NMI comes in during the
2120 test, the time spent in NMI is reported in "nmi-total" (in
2123 All architectures that have NMIs will show the "nmi-count" if an
2124 NMI comes in during the test.
2129 This gets automatically set to "10" to represent 10
2130 microseconds. This is the threshold of latency that
2131 needs to be detected before the trace will be recorded.
2133 Note, when hwlat tracer is finished (another tracer is
2134 written into "current_tracer"), the original value for
2135 tracing_threshold is placed back into this file.
2137 hwlat_detector/width
2138 The length of time the test runs with interrupts disabled.
2140 hwlat_detector/window
2141 The length of time of the window which the test
2142 runs. That is, the test will run for "width"
2143 microseconds per "window" microseconds
2146 When the test is started. A kernel thread is created that
2147 runs the test. This thread will alternate between CPUs
2148 listed in the tracing_cpumask between each period
2149 (one "window"). To limit the test to specific CPUs
2150 set the mask in this file to only the CPUs that the test
2156 This tracer is the function tracer. Enabling the function tracer
2157 can be done from the debug file system. Make sure the
2158 ftrace_enabled is set; otherwise this tracer is a nop.
2159 See the "ftrace_enabled" section below.
2162 # sysctl kernel.ftrace_enabled=1
2163 # echo function > current_tracer
2164 # echo 1 > tracing_on
2166 # echo 0 > tracing_on
2170 # entries-in-buffer/entries-written: 24799/24799 #P:4
2173 # / _----=> need-resched
2174 # | / _---=> hardirq/softirq
2175 # || / _--=> preempt-depth
2177 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2179 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
2180 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
2181 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
2182 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
2183 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
2184 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
2185 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
2186 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
2190 Note: function tracer uses ring buffers to store the above
2191 entries. The newest data may overwrite the oldest data.
2192 Sometimes using echo to stop the trace is not sufficient because
2193 the tracing could have overwritten the data that you wanted to
2194 record. For this reason, it is sometimes better to disable
2195 tracing directly from a program. This allows you to stop the
2196 tracing at the point that you hit the part that you are
2197 interested in. To disable the tracing directly from a C program,
2198 something like following code snippet can be used::
2202 int main(int argc, char *argv[]) {
2204 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
2206 if (condition_hit()) {
2207 write(trace_fd, "0", 1);
2213 Single thread tracing
2214 ---------------------
2216 By writing into set_ftrace_pid you can trace a
2217 single thread. For example::
2219 # cat set_ftrace_pid
2221 # echo 3111 > set_ftrace_pid
2222 # cat set_ftrace_pid
2224 # echo function > current_tracer
2228 # TASK-PID CPU# TIMESTAMP FUNCTION
2230 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
2231 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
2232 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
2233 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
2234 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
2235 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
2236 # echo > set_ftrace_pid
2240 # TASK-PID CPU# TIMESTAMP FUNCTION
2242 ##### CPU 3 buffer started ####
2243 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
2244 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
2245 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
2246 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
2247 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
2249 If you want to trace a function when executing, you could use
2250 something like this simple program.
2255 #include <sys/types.h>
2256 #include <sys/stat.h>
2262 #define STR(x) _STR(x)
2263 #define MAX_PATH 256
2265 const char *find_tracefs(void)
2267 static char tracefs[MAX_PATH+1];
2268 static int tracefs_found;
2275 if ((fp = fopen("/proc/mounts","r")) == NULL) {
2276 perror("/proc/mounts");
2280 while (fscanf(fp, "%*s %"
2282 "s %99s %*s %*d %*d\n",
2283 tracefs, type) == 2) {
2284 if (strcmp(type, "tracefs") == 0)
2289 if (strcmp(type, "tracefs") != 0) {
2290 fprintf(stderr, "tracefs not mounted");
2294 strcat(tracefs, "/tracing/");
2300 const char *tracing_file(const char *file_name)
2302 static char trace_file[MAX_PATH+1];
2303 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name);
2307 int main (int argc, char **argv)
2317 ffd = open(tracing_file("current_tracer"), O_WRONLY);
2320 write(ffd, "nop", 3);
2322 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
2323 s = sprintf(line, "%d\n", getpid());
2326 write(ffd, "function", 8);
2331 execvp(argv[1], argv+1);
2337 Or this simple script!
2342 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts`
2343 echo nop > $tracefs/tracing/current_tracer
2344 echo 0 > $tracefs/tracing/tracing_on
2345 echo $$ > $tracefs/tracing/set_ftrace_pid
2346 echo function > $tracefs/tracing/current_tracer
2347 echo 1 > $tracefs/tracing/tracing_on
2351 function graph tracer
2352 ---------------------------
2354 This tracer is similar to the function tracer except that it
2355 probes a function on its entry and its exit. This is done by
2356 using a dynamically allocated stack of return addresses in each
2357 task_struct. On function entry the tracer overwrites the return
2358 address of each function traced to set a custom probe. Thus the
2359 original return address is stored on the stack of return address
2362 Probing on both ends of a function leads to special features
2365 - measure of a function's time execution
2366 - having a reliable call stack to draw function calls graph
2368 This tracer is useful in several situations:
2370 - you want to find the reason of a strange kernel behavior and
2371 need to see what happens in detail on any areas (or specific
2374 - you are experiencing weird latencies but it's difficult to
2377 - you want to find quickly which path is taken by a specific
2380 - you just want to peek inside a working kernel and want to see
2385 # tracer: function_graph
2387 # CPU DURATION FUNCTION CALLS
2391 0) | do_sys_open() {
2393 0) | kmem_cache_alloc() {
2394 0) 1.382 us | __might_sleep();
2396 0) | strncpy_from_user() {
2397 0) | might_fault() {
2398 0) 1.389 us | __might_sleep();
2403 0) 0.668 us | _spin_lock();
2404 0) 0.570 us | expand_files();
2405 0) 0.586 us | _spin_unlock();
2408 There are several columns that can be dynamically
2409 enabled/disabled. You can use every combination of options you
2410 want, depending on your needs.
2412 - The cpu number on which the function executed is default
2413 enabled. It is sometimes better to only trace one cpu (see
2414 tracing_cpu_mask file) or you might sometimes see unordered
2415 function calls while cpu tracing switch.
2417 - hide: echo nofuncgraph-cpu > trace_options
2418 - show: echo funcgraph-cpu > trace_options
2420 - The duration (function's time of execution) is displayed on
2421 the closing bracket line of a function or on the same line
2422 than the current function in case of a leaf one. It is default
2425 - hide: echo nofuncgraph-duration > trace_options
2426 - show: echo funcgraph-duration > trace_options
2428 - The overhead field precedes the duration field in case of
2429 reached duration thresholds.
2431 - hide: echo nofuncgraph-overhead > trace_options
2432 - show: echo funcgraph-overhead > trace_options
2433 - depends on: funcgraph-duration
2437 3) # 1837.709 us | } /* __switch_to */
2438 3) | finish_task_switch() {
2439 3) 0.313 us | _raw_spin_unlock_irq();
2441 3) # 1889.063 us | } /* __schedule */
2442 3) ! 140.417 us | } /* __schedule */
2443 3) # 2034.948 us | } /* schedule */
2444 3) * 33998.59 us | } /* schedule_preempt_disabled */
2448 1) 0.260 us | msecs_to_jiffies();
2449 1) 0.313 us | __rcu_read_unlock();
2452 1) 0.313 us | rcu_bh_qs();
2453 1) 0.313 us | __local_bh_enable();
2455 1) 0.365 us | idle_cpu();
2456 1) | rcu_irq_exit() {
2457 1) 0.417 us | rcu_eqs_enter_common.isra.47();
2461 1) @ 119760.2 us | }
2467 2) 0.417 us | scheduler_ipi();
2477 + means that the function exceeded 10 usecs.
2478 ! means that the function exceeded 100 usecs.
2479 # means that the function exceeded 1000 usecs.
2480 * means that the function exceeded 10 msecs.
2481 @ means that the function exceeded 100 msecs.
2482 $ means that the function exceeded 1 sec.
2485 - The task/pid field displays the thread cmdline and pid which
2486 executed the function. It is default disabled.
2488 - hide: echo nofuncgraph-proc > trace_options
2489 - show: echo funcgraph-proc > trace_options
2493 # tracer: function_graph
2495 # CPU TASK/PID DURATION FUNCTION CALLS
2497 0) sh-4802 | | d_free() {
2498 0) sh-4802 | | call_rcu() {
2499 0) sh-4802 | | __call_rcu() {
2500 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2501 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2502 0) sh-4802 | 2.899 us | }
2503 0) sh-4802 | 4.040 us | }
2504 0) sh-4802 | 5.151 us | }
2505 0) sh-4802 | + 49.370 us | }
2508 - The absolute time field is an absolute timestamp given by the
2509 system clock since it started. A snapshot of this time is
2510 given on each entry/exit of functions
2512 - hide: echo nofuncgraph-abstime > trace_options
2513 - show: echo funcgraph-abstime > trace_options
2518 # TIME CPU DURATION FUNCTION CALLS
2520 360.774522 | 1) 0.541 us | }
2521 360.774522 | 1) 4.663 us | }
2522 360.774523 | 1) 0.541 us | __wake_up_bit();
2523 360.774524 | 1) 6.796 us | }
2524 360.774524 | 1) 7.952 us | }
2525 360.774525 | 1) 9.063 us | }
2526 360.774525 | 1) 0.615 us | journal_mark_dirty();
2527 360.774527 | 1) 0.578 us | __brelse();
2528 360.774528 | 1) | reiserfs_prepare_for_journal() {
2529 360.774528 | 1) | unlock_buffer() {
2530 360.774529 | 1) | wake_up_bit() {
2531 360.774529 | 1) | bit_waitqueue() {
2532 360.774530 | 1) 0.594 us | __phys_addr();
2535 The function name is always displayed after the closing bracket
2536 for a function if the start of that function is not in the
2539 Display of the function name after the closing bracket may be
2540 enabled for functions whose start is in the trace buffer,
2541 allowing easier searching with grep for function durations.
2542 It is default disabled.
2544 - hide: echo nofuncgraph-tail > trace_options
2545 - show: echo funcgraph-tail > trace_options
2547 Example with nofuncgraph-tail (default)::
2550 0) | kmem_cache_free() {
2551 0) 0.518 us | __phys_addr();
2555 Example with funcgraph-tail::
2558 0) | kmem_cache_free() {
2559 0) 0.518 us | __phys_addr();
2560 0) 1.757 us | } /* kmem_cache_free() */
2561 0) 2.861 us | } /* putname() */
2563 You can put some comments on specific functions by using
2564 trace_printk() For example, if you want to put a comment inside
2565 the __might_sleep() function, you just have to include
2566 <linux/ftrace.h> and call trace_printk() inside __might_sleep()::
2568 trace_printk("I'm a comment!\n")
2572 1) | __might_sleep() {
2573 1) | /* I'm a comment! */
2577 You might find other useful features for this tracer in the
2578 following "dynamic ftrace" section such as tracing only specific
2584 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2585 virtually no overhead when function tracing is disabled. The way
2586 this works is the mcount function call (placed at the start of
2587 every kernel function, produced by the -pg switch in gcc),
2588 starts of pointing to a simple return. (Enabling FTRACE will
2589 include the -pg switch in the compiling of the kernel.)
2591 At compile time every C file object is run through the
2592 recordmcount program (located in the scripts directory). This
2593 program will parse the ELF headers in the C object to find all
2594 the locations in the .text section that call mcount. Starting
2595 with gcc version 4.6, the -mfentry has been added for x86, which
2596 calls "__fentry__" instead of "mcount". Which is called before
2597 the creation of the stack frame.
2599 Note, not all sections are traced. They may be prevented by either
2600 a notrace, or blocked another way and all inline functions are not
2601 traced. Check the "available_filter_functions" file to see what functions
2604 A section called "__mcount_loc" is created that holds
2605 references to all the mcount/fentry call sites in the .text section.
2606 The recordmcount program re-links this section back into the
2607 original object. The final linking stage of the kernel will add all these
2608 references into a single table.
2610 On boot up, before SMP is initialized, the dynamic ftrace code
2611 scans this table and updates all the locations into nops. It
2612 also records the locations, which are added to the
2613 available_filter_functions list. Modules are processed as they
2614 are loaded and before they are executed. When a module is
2615 unloaded, it also removes its functions from the ftrace function
2616 list. This is automatic in the module unload code, and the
2617 module author does not need to worry about it.
2619 When tracing is enabled, the process of modifying the function
2620 tracepoints is dependent on architecture. The old method is to use
2621 kstop_machine to prevent races with the CPUs executing code being
2622 modified (which can cause the CPU to do undesirable things, especially
2623 if the modified code crosses cache (or page) boundaries), and the nops are
2624 patched back to calls. But this time, they do not call mcount
2625 (which is just a function stub). They now call into the ftrace
2628 The new method of modifying the function tracepoints is to place
2629 a breakpoint at the location to be modified, sync all CPUs, modify
2630 the rest of the instruction not covered by the breakpoint. Sync
2631 all CPUs again, and then remove the breakpoint with the finished
2632 version to the ftrace call site.
2634 Some archs do not even need to monkey around with the synchronization,
2635 and can just slap the new code on top of the old without any
2636 problems with other CPUs executing it at the same time.
2638 One special side-effect to the recording of the functions being
2639 traced is that we can now selectively choose which functions we
2640 wish to trace and which ones we want the mcount calls to remain
2643 Two files are used, one for enabling and one for disabling the
2644 tracing of specified functions. They are:
2652 A list of available functions that you can add to these files is
2655 available_filter_functions
2659 # cat available_filter_functions
2668 If I am only interested in sys_nanosleep and hrtimer_interrupt::
2670 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2671 # echo function > current_tracer
2672 # echo 1 > tracing_on
2674 # echo 0 > tracing_on
2678 # entries-in-buffer/entries-written: 5/5 #P:4
2681 # / _----=> need-resched
2682 # | / _---=> hardirq/softirq
2683 # || / _--=> preempt-depth
2685 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2687 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2688 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2689 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2690 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2691 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2693 To see which functions are being traced, you can cat the file:
2696 # cat set_ftrace_filter
2701 Perhaps this is not enough. The filters also allow glob(7) matching.
2704 will match functions that begin with <match>
2706 will match functions that end with <match>
2708 will match functions that have <match> in it
2709 ``<match1>*<match2>``
2710 will match functions that begin with <match1> and end with <match2>
2713 It is better to use quotes to enclose the wild cards,
2714 otherwise the shell may expand the parameters into names
2715 of files in the local directory.
2719 # echo 'hrtimer_*' > set_ftrace_filter
2725 # entries-in-buffer/entries-written: 897/897 #P:4
2728 # / _----=> need-resched
2729 # | / _---=> hardirq/softirq
2730 # || / _--=> preempt-depth
2732 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2734 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2735 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2736 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2737 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2738 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2739 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2740 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2741 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2743 Notice that we lost the sys_nanosleep.
2746 # cat set_ftrace_filter
2751 hrtimer_try_to_cancel
2755 hrtimer_force_reprogram
2756 hrtimer_get_next_event
2760 hrtimer_get_remaining
2762 hrtimer_init_sleeper
2765 This is because the '>' and '>>' act just like they do in bash.
2766 To rewrite the filters, use '>'
2767 To append to the filters, use '>>'
2769 To clear out a filter so that all functions will be recorded
2772 # echo > set_ftrace_filter
2773 # cat set_ftrace_filter
2776 Again, now we want to append.
2780 # echo sys_nanosleep > set_ftrace_filter
2781 # cat set_ftrace_filter
2783 # echo 'hrtimer_*' >> set_ftrace_filter
2784 # cat set_ftrace_filter
2789 hrtimer_try_to_cancel
2793 hrtimer_force_reprogram
2794 hrtimer_get_next_event
2799 hrtimer_get_remaining
2801 hrtimer_init_sleeper
2804 The set_ftrace_notrace prevents those functions from being
2808 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2814 # entries-in-buffer/entries-written: 39608/39608 #P:4
2817 # / _----=> need-resched
2818 # | / _---=> hardirq/softirq
2819 # || / _--=> preempt-depth
2821 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2823 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2824 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2825 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2826 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2827 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2828 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2829 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2830 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2831 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2832 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2833 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2834 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2836 We can see that there's no more lock or preempt tracing.
2839 Dynamic ftrace with the function graph tracer
2840 ---------------------------------------------
2842 Although what has been explained above concerns both the
2843 function tracer and the function-graph-tracer, there are some
2844 special features only available in the function-graph tracer.
2846 If you want to trace only one function and all of its children,
2847 you just have to echo its name into set_graph_function::
2849 echo __do_fault > set_graph_function
2851 will produce the following "expanded" trace of the __do_fault()
2855 0) | filemap_fault() {
2856 0) | find_lock_page() {
2857 0) 0.804 us | find_get_page();
2858 0) | __might_sleep() {
2862 0) 0.653 us | _spin_lock();
2863 0) 0.578 us | page_add_file_rmap();
2864 0) 0.525 us | native_set_pte_at();
2865 0) 0.585 us | _spin_unlock();
2866 0) | unlock_page() {
2867 0) 0.541 us | page_waitqueue();
2868 0) 0.639 us | __wake_up_bit();
2872 0) | filemap_fault() {
2873 0) | find_lock_page() {
2874 0) 0.698 us | find_get_page();
2875 0) | __might_sleep() {
2879 0) 0.631 us | _spin_lock();
2880 0) 0.571 us | page_add_file_rmap();
2881 0) 0.526 us | native_set_pte_at();
2882 0) 0.586 us | _spin_unlock();
2883 0) | unlock_page() {
2884 0) 0.533 us | page_waitqueue();
2885 0) 0.638 us | __wake_up_bit();
2889 You can also expand several functions at once::
2891 echo sys_open > set_graph_function
2892 echo sys_close >> set_graph_function
2894 Now if you want to go back to trace all functions you can clear
2895 this special filter via::
2897 echo > set_graph_function
2903 Note, the proc sysctl ftrace_enable is a big on/off switch for the
2904 function tracer. By default it is enabled (when function tracing is
2905 enabled in the kernel). If it is disabled, all function tracing is
2906 disabled. This includes not only the function tracers for ftrace, but
2907 also for any other uses (perf, kprobes, stack tracing, profiling, etc).
2909 Please disable this with care.
2911 This can be disable (and enabled) with::
2913 sysctl kernel.ftrace_enabled=0
2914 sysctl kernel.ftrace_enabled=1
2918 echo 0 > /proc/sys/kernel/ftrace_enabled
2919 echo 1 > /proc/sys/kernel/ftrace_enabled
2925 A few commands are supported by the set_ftrace_filter interface.
2926 Trace commands have the following format::
2928 <function>:<command>:<parameter>
2930 The following commands are supported:
2933 This command enables function filtering per module. The
2934 parameter defines the module. For example, if only the write*
2935 functions in the ext3 module are desired, run:
2937 echo 'write*:mod:ext3' > set_ftrace_filter
2939 This command interacts with the filter in the same way as
2940 filtering based on function names. Thus, adding more functions
2941 in a different module is accomplished by appending (>>) to the
2942 filter file. Remove specific module functions by prepending
2945 echo '!writeback*:mod:ext3' >> set_ftrace_filter
2947 Mod command supports module globbing. Disable tracing for all
2948 functions except a specific module::
2950 echo '!*:mod:!ext3' >> set_ftrace_filter
2952 Disable tracing for all modules, but still trace kernel::
2954 echo '!*:mod:*' >> set_ftrace_filter
2956 Enable filter only for kernel::
2958 echo '*write*:mod:!*' >> set_ftrace_filter
2960 Enable filter for module globbing::
2962 echo '*write*:mod:*snd*' >> set_ftrace_filter
2965 These commands turn tracing on and off when the specified
2966 functions are hit. The parameter determines how many times the
2967 tracing system is turned on and off. If unspecified, there is
2968 no limit. For example, to disable tracing when a schedule bug
2969 is hit the first 5 times, run::
2971 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
2973 To always disable tracing when __schedule_bug is hit::
2975 echo '__schedule_bug:traceoff' > set_ftrace_filter
2977 These commands are cumulative whether or not they are appended
2978 to set_ftrace_filter. To remove a command, prepend it by '!'
2979 and drop the parameter::
2981 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
2983 The above removes the traceoff command for __schedule_bug
2984 that have a counter. To remove commands without counters::
2986 echo '!__schedule_bug:traceoff' > set_ftrace_filter
2989 Will cause a snapshot to be triggered when the function is hit.
2992 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
2994 To only snapshot once:
2997 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
2999 To remove the above commands::
3001 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
3002 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
3004 - enable_event/disable_event:
3005 These commands can enable or disable a trace event. Note, because
3006 function tracing callbacks are very sensitive, when these commands
3007 are registered, the trace point is activated, but disabled in
3008 a "soft" mode. That is, the tracepoint will be called, but
3009 just will not be traced. The event tracepoint stays in this mode
3010 as long as there's a command that triggers it.
3013 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
3018 <function>:enable_event:<system>:<event>[:count]
3019 <function>:disable_event:<system>:<event>[:count]
3021 To remove the events commands::
3023 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
3025 echo '!schedule:disable_event:sched:sched_switch' > \
3029 When the function is hit, it will dump the contents of the ftrace
3030 ring buffer to the console. This is useful if you need to debug
3031 something, and want to dump the trace when a certain function
3032 is hit. Perhaps it's a function that is called before a triple
3033 fault happens and does not allow you to get a regular dump.
3036 When the function is hit, it will dump the contents of the ftrace
3037 ring buffer for the current CPU to the console. Unlike the "dump"
3038 command, it only prints out the contents of the ring buffer for the
3039 CPU that executed the function that triggered the dump.
3042 When the function is hit, a stack trace is recorded.
3047 The trace_pipe outputs the same content as the trace file, but
3048 the effect on the tracing is different. Every read from
3049 trace_pipe is consumed. This means that subsequent reads will be
3050 different. The trace is live.
3053 # echo function > current_tracer
3054 # cat trace_pipe > /tmp/trace.out &
3056 # echo 1 > tracing_on
3058 # echo 0 > tracing_on
3062 # entries-in-buffer/entries-written: 0/0 #P:4
3065 # / _----=> need-resched
3066 # | / _---=> hardirq/softirq
3067 # || / _--=> preempt-depth
3069 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3073 # cat /tmp/trace.out
3074 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
3075 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
3076 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
3077 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
3078 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
3079 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
3080 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
3081 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
3082 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
3085 Note, reading the trace_pipe file will block until more input is
3091 Having too much or not enough data can be troublesome in
3092 diagnosing an issue in the kernel. The file buffer_size_kb is
3093 used to modify the size of the internal trace buffers. The
3094 number listed is the number of entries that can be recorded per
3095 CPU. To know the full size, multiply the number of possible CPUs
3096 with the number of entries.
3099 # cat buffer_size_kb
3100 1408 (units kilobytes)
3102 Or simply read buffer_total_size_kb
3105 # cat buffer_total_size_kb
3108 To modify the buffer, simple echo in a number (in 1024 byte segments).
3111 # echo 10000 > buffer_size_kb
3112 # cat buffer_size_kb
3113 10000 (units kilobytes)
3115 It will try to allocate as much as possible. If you allocate too
3116 much, it can cause Out-Of-Memory to trigger.
3119 # echo 1000000000000 > buffer_size_kb
3120 -bash: echo: write error: Cannot allocate memory
3121 # cat buffer_size_kb
3124 The per_cpu buffers can be changed individually as well:
3127 # echo 10000 > per_cpu/cpu0/buffer_size_kb
3128 # echo 100 > per_cpu/cpu1/buffer_size_kb
3130 When the per_cpu buffers are not the same, the buffer_size_kb
3131 at the top level will just show an X
3134 # cat buffer_size_kb
3137 This is where the buffer_total_size_kb is useful:
3140 # cat buffer_total_size_kb
3143 Writing to the top level buffer_size_kb will reset all the buffers
3144 to be the same again.
3148 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
3149 available to all non latency tracers. (Latency tracers which
3150 record max latency, such as "irqsoff" or "wakeup", can't use
3151 this feature, since those are already using the snapshot
3152 mechanism internally.)
3154 Snapshot preserves a current trace buffer at a particular point
3155 in time without stopping tracing. Ftrace swaps the current
3156 buffer with a spare buffer, and tracing continues in the new
3157 current (=previous spare) buffer.
3159 The following tracefs files in "tracing" are related to this
3164 This is used to take a snapshot and to read the output
3165 of the snapshot. Echo 1 into this file to allocate a
3166 spare buffer and to take a snapshot (swap), then read
3167 the snapshot from this file in the same format as
3168 "trace" (described above in the section "The File
3169 System"). Both reads snapshot and tracing are executable
3170 in parallel. When the spare buffer is allocated, echoing
3171 0 frees it, and echoing else (positive) values clear the
3173 More details are shown in the table below.
3175 +--------------+------------+------------+------------+
3176 |status\\input | 0 | 1 | else |
3177 +==============+============+============+============+
3178 |not allocated |(do nothing)| alloc+swap |(do nothing)|
3179 +--------------+------------+------------+------------+
3180 |allocated | free | swap | clear |
3181 +--------------+------------+------------+------------+
3183 Here is an example of using the snapshot feature.
3186 # echo 1 > events/sched/enable
3191 # entries-in-buffer/entries-written: 71/71 #P:8
3194 # / _----=> need-resched
3195 # | / _---=> hardirq/softirq
3196 # || / _--=> preempt-depth
3198 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3200 <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
3201 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
3203 <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
3208 # entries-in-buffer/entries-written: 77/77 #P:8
3211 # / _----=> need-resched
3212 # | / _---=> hardirq/softirq
3213 # || / _--=> preempt-depth
3215 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3217 <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
3218 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
3222 If you try to use this snapshot feature when current tracer is
3223 one of the latency tracers, you will get the following results.
3226 # echo wakeup > current_tracer
3228 bash: echo: write error: Device or resource busy
3230 cat: snapshot: Device or resource busy
3235 In the tracefs tracing directory is a directory called "instances".
3236 This directory can have new directories created inside of it using
3237 mkdir, and removing directories with rmdir. The directory created
3238 with mkdir in this directory will already contain files and other
3239 directories after it is created.
3242 # mkdir instances/foo
3244 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
3245 set_event snapshot trace trace_clock trace_marker trace_options
3246 trace_pipe tracing_on
3248 As you can see, the new directory looks similar to the tracing directory
3249 itself. In fact, it is very similar, except that the buffer and
3250 events are agnostic from the main director, or from any other
3251 instances that are created.
3253 The files in the new directory work just like the files with the
3254 same name in the tracing directory except the buffer that is used
3255 is a separate and new buffer. The files affect that buffer but do not
3256 affect the main buffer with the exception of trace_options. Currently,
3257 the trace_options affect all instances and the top level buffer
3258 the same, but this may change in future releases. That is, options
3259 may become specific to the instance they reside in.
3261 Notice that none of the function tracer files are there, nor is
3262 current_tracer and available_tracers. This is because the buffers
3263 can currently only have events enabled for them.
3266 # mkdir instances/foo
3267 # mkdir instances/bar
3268 # mkdir instances/zoot
3269 # echo 100000 > buffer_size_kb
3270 # echo 1000 > instances/foo/buffer_size_kb
3271 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
3272 # echo function > current_trace
3273 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
3274 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
3275 # echo 1 > instances/foo/events/sched/sched_switch/enable
3276 # echo 1 > instances/bar/events/irq/enable
3277 # echo 1 > instances/zoot/events/syscalls/enable
3279 CPU:2 [LOST 11745 EVENTS]
3280 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
3281 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
3282 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
3283 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
3284 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
3285 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
3286 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
3287 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
3288 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3289 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3290 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
3293 # cat instances/foo/trace_pipe
3294 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3295 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3296 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
3297 <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
3298 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
3299 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3300 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3301 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
3302 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
3303 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
3306 # cat instances/bar/trace_pipe
3307 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
3308 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
3309 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
3310 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
3311 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
3312 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
3313 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
3314 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
3315 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
3316 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
3317 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
3318 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
3321 # cat instances/zoot/trace
3324 # entries-in-buffer/entries-written: 18996/18996 #P:4
3327 # / _----=> need-resched
3328 # | / _---=> hardirq/softirq
3329 # || / _--=> preempt-depth
3331 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3333 bash-1998 [000] d... 140.733501: sys_write -> 0x2
3334 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
3335 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
3336 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
3337 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
3338 bash-1998 [000] d... 140.733510: sys_close(fd: a)
3339 bash-1998 [000] d... 140.733510: sys_close -> 0x0
3340 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
3341 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
3342 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
3343 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
3345 You can see that the trace of the top most trace buffer shows only
3346 the function tracing. The foo instance displays wakeups and task
3349 To remove the instances, simply delete their directories:
3352 # rmdir instances/foo
3353 # rmdir instances/bar
3354 # rmdir instances/zoot
3356 Note, if a process has a trace file open in one of the instance
3357 directories, the rmdir will fail with EBUSY.
3362 Since the kernel has a fixed sized stack, it is important not to
3363 waste it in functions. A kernel developer must be conscience of
3364 what they allocate on the stack. If they add too much, the system
3365 can be in danger of a stack overflow, and corruption will occur,
3366 usually leading to a system panic.
3368 There are some tools that check this, usually with interrupts
3369 periodically checking usage. But if you can perform a check
3370 at every function call that will become very useful. As ftrace provides
3371 a function tracer, it makes it convenient to check the stack size
3372 at every function call. This is enabled via the stack tracer.
3374 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
3375 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
3378 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
3380 You can also enable it from the kernel command line to trace
3381 the stack size of the kernel during boot up, by adding "stacktrace"
3382 to the kernel command line parameter.
3384 After running it for a few minutes, the output looks like:
3387 # cat stack_max_size
3391 Depth Size Location (18 entries)
3393 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
3394 1) 2704 160 find_busiest_group+0x31/0x1f1
3395 2) 2544 256 load_balance+0xd9/0x662
3396 3) 2288 80 idle_balance+0xbb/0x130
3397 4) 2208 128 __schedule+0x26e/0x5b9
3398 5) 2080 16 schedule+0x64/0x66
3399 6) 2064 128 schedule_timeout+0x34/0xe0
3400 7) 1936 112 wait_for_common+0x97/0xf1
3401 8) 1824 16 wait_for_completion+0x1d/0x1f
3402 9) 1808 128 flush_work+0xfe/0x119
3403 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
3404 11) 1664 48 input_available_p+0x1d/0x5c
3405 12) 1616 48 n_tty_poll+0x6d/0x134
3406 13) 1568 64 tty_poll+0x64/0x7f
3407 14) 1504 880 do_select+0x31e/0x511
3408 15) 624 400 core_sys_select+0x177/0x216
3409 16) 224 96 sys_select+0x91/0xb9
3410 17) 128 128 system_call_fastpath+0x16/0x1b
3412 Note, if -mfentry is being used by gcc, functions get traced before
3413 they set up the stack frame. This means that leaf level functions
3414 are not tested by the stack tracer when -mfentry is used.
3416 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
3420 More details can be found in the source code, in the `kernel/trace/*.c` files.
projects / platform / kernel / linux-starfive.git / blob