1 ========================
2 ftrace - Function Tracer
3 ========================
5 Copyright 2008 Red Hat Inc.
7 :Author: Steven Rostedt <srostedt@redhat.com>
8 :License: The GNU Free Documentation License, Version 1.2
9 (dual licensed under the GPL v2)
10 :Original Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton,
11 John Kacur, and David Teigland.
13 - Written for: 2.6.28-rc2
15 - Updated for: 4.13 - Copyright 2017 VMware Inc. Steven Rostedt
16 - Converted to rst format - Changbin Du <changbin.du@intel.com>
21 Ftrace is an internal tracer designed to help out developers and
22 designers of systems to find what is going on inside the kernel.
23 It can be used for debugging or analyzing latencies and
24 performance issues that take place outside of user-space.
26 Although ftrace is typically considered the function tracer, it
27 is really a framework of several assorted tracing utilities.
28 There's latency tracing to examine what occurs between interrupts
29 disabled and enabled, as well as for preemption and from a time
30 a task is woken to the task is actually scheduled in.
32 One of the most common uses of ftrace is the event tracing.
33 Throughout the kernel is hundreds of static event points that
34 can be enabled via the tracefs file system to see what is
35 going on in certain parts of the kernel.
37 See events.rst for more information.
40 Implementation Details
41 ----------------------
43 See Documentation/trace/ftrace-design.rst for details for arch porters and such.
49 Ftrace uses the tracefs file system to hold the control files as
50 well as the files to display output.
52 When tracefs is configured into the kernel (which selecting any ftrace
53 option will do) the directory /sys/kernel/tracing will be created. To mount
54 this directory, you can add to your /etc/fstab file::
56 tracefs /sys/kernel/tracing tracefs defaults 0 0
58 Or you can mount it at run time with::
60 mount -t tracefs nodev /sys/kernel/tracing
62 For quicker access to that directory you may want to make a soft link to
65 ln -s /sys/kernel/tracing /tracing
69 Before 4.1, all ftrace tracing control files were within the debugfs
70 file system, which is typically located at /sys/kernel/debug/tracing.
71 For backward compatibility, when mounting the debugfs file system,
72 the tracefs file system will be automatically mounted at:
74 /sys/kernel/debug/tracing
76 All files located in the tracefs file system will be located in that
77 debugfs file system directory as well.
81 Any selected ftrace option will also create the tracefs file system.
82 The rest of the document will assume that you are in the ftrace directory
83 (cd /sys/kernel/tracing) and will only concentrate on the files within that
84 directory and not distract from the content with the extended
85 "/sys/kernel/tracing" path name.
87 That's it! (assuming that you have ftrace configured into your kernel)
89 After mounting tracefs you will have access to the control and output files
90 of ftrace. Here is a list of some of the key files:
93 Note: all time values are in microseconds.
97 This is used to set or display the current tracer
98 that is configured. Changing the current tracer clears
99 the ring buffer content as well as the "snapshot" buffer.
103 This holds the different types of tracers that
104 have been compiled into the kernel. The
105 tracers listed here can be configured by
106 echoing their name into current_tracer.
110 This sets or displays whether writing to the trace
111 ring buffer is enabled. Echo 0 into this file to disable
112 the tracer or 1 to enable it. Note, this only disables
113 writing to the ring buffer, the tracing overhead may
116 The kernel function tracing_off() can be used within the
117 kernel to disable writing to the ring buffer, which will
118 set this file to "0". User space can re-enable tracing by
119 echoing "1" into the file.
121 Note, the function and event trigger "traceoff" will also
122 set this file to zero and stop tracing. Which can also
123 be re-enabled by user space using this file.
127 This file holds the output of the trace in a human
128 readable format (described below). Opening this file for
129 writing with the O_TRUNC flag clears the ring buffer content.
130 Note, this file is not a consumer. If tracing is off
131 (no tracer running, or tracing_on is zero), it will produce
132 the same output each time it is read. When tracing is on,
133 it may produce inconsistent results as it tries to read
134 the entire buffer without consuming it.
138 The output is the same as the "trace" file but this
139 file is meant to be streamed with live tracing.
140 Reads from this file will block until new data is
141 retrieved. Unlike the "trace" file, this file is a
142 consumer. This means reading from this file causes
143 sequential reads to display more current data. Once
144 data is read from this file, it is consumed, and
145 will not be read again with a sequential read. The
146 "trace" file is static, and if the tracer is not
147 adding more data, it will display the same
148 information every time it is read.
152 This file lets the user control the amount of data
153 that is displayed in one of the above output
154 files. Options also exist to modify how a tracer
155 or events work (stack traces, timestamps, etc).
159 This is a directory that has a file for every available
160 trace option (also in trace_options). Options may also be set
161 or cleared by writing a "1" or "0" respectively into the
162 corresponding file with the option name.
166 Some of the tracers record the max latency.
167 For example, the maximum time that interrupts are disabled.
168 The maximum time is saved in this file. The max trace will also be
169 stored, and displayed by "trace". A new max trace will only be
170 recorded if the latency is greater than the value in this file
173 By echoing in a time into this file, no latency will be recorded
174 unless it is greater than the time in this file.
178 Some latency tracers will record a trace whenever the
179 latency is greater than the number in this file.
180 Only active when the file contains a number greater than 0.
185 This sets or displays the number of kilobytes each CPU
186 buffer holds. By default, the trace buffers are the same size
187 for each CPU. The displayed number is the size of the
188 CPU buffer and not total size of all buffers. The
189 trace buffers are allocated in pages (blocks of memory
190 that the kernel uses for allocation, usually 4 KB in size).
191 A few extra pages may be allocated to accommodate buffer management
192 meta-data. If the last page allocated has room for more bytes
193 than requested, the rest of the page will be used,
194 making the actual allocation bigger than requested or shown.
195 ( Note, the size may not be a multiple of the page size
196 due to buffer management meta-data. )
198 Buffer sizes for individual CPUs may vary
199 (see "per_cpu/cpu0/buffer_size_kb" below), and if they do
200 this file will show "X".
202 buffer_total_size_kb:
204 This displays the total combined size of all the trace buffers.
208 This sets or displays the sub buffer page size order. The ring buffer
209 is broken up into several same size "sub buffers". An event can not be
210 bigger than the size of the sub buffer. Normally, the sub buffer is
211 the size of the architecture's page (4K on x86). The sub buffer also
212 contains meta data at the start which also limits the size of an event.
213 That means when the sub buffer is a page size, no event can be larger
214 than the page size minus the sub buffer meta data.
216 The buffer_subbuf_order allows the user to change the size of the sub
217 buffer. As the sub buffer is a set of pages by the power of 2, thus
218 the sub buffer total size is defined by the order:
227 Changing the order will change the sub buffer size allowing for events
228 to be larger than the page size.
230 Note: When changing the order, tracing is stopped and any data in the
231 ring buffer and the snapshot buffer will be discarded.
235 If a process is performing tracing, and the ring buffer should be
236 shrunk "freed" when the process is finished, even if it were to be
237 killed by a signal, this file can be used for that purpose. On close
238 of this file, the ring buffer will be resized to its minimum size.
239 Having a process that is tracing also open this file, when the process
240 exits its file descriptor for this file will be closed, and in doing so,
241 the ring buffer will be "freed".
243 It may also stop tracing if disable_on_free option is set.
247 This is a mask that lets the user only trace on specified CPUs.
248 The format is a hex string representing the CPUs.
252 When dynamic ftrace is configured in (see the
253 section below "dynamic ftrace"), the code is dynamically
254 modified (code text rewrite) to disable calling of the
255 function profiler (mcount). This lets tracing be configured
256 in with practically no overhead in performance. This also
257 has a side effect of enabling or disabling specific functions
258 to be traced. Echoing names of functions into this file
259 will limit the trace to only those functions.
260 This influences the tracers "function" and "function_graph"
261 and thus also function profiling (see "function_profile_enabled").
263 The functions listed in "available_filter_functions" are what
264 can be written into this file.
266 This interface also allows for commands to be used. See the
267 "Filter commands" section for more details.
269 As a speed up, since processing strings can be quite expensive
270 and requires a check of all functions registered to tracing, instead
271 an index can be written into this file. A number (starting with "1")
272 written will instead select the same corresponding at the line position
273 of the "available_filter_functions" file.
277 This has an effect opposite to that of
278 set_ftrace_filter. Any function that is added here will not
279 be traced. If a function exists in both set_ftrace_filter
280 and set_ftrace_notrace, the function will _not_ be traced.
284 Have the function tracer only trace the threads whose PID are
287 If the "function-fork" option is set, then when a task whose
288 PID is listed in this file forks, the child's PID will
289 automatically be added to this file, and the child will be
290 traced by the function tracer as well. This option will also
291 cause PIDs of tasks that exit to be removed from the file.
293 set_ftrace_notrace_pid:
295 Have the function tracer ignore threads whose PID are listed in
298 If the "function-fork" option is set, then when a task whose
299 PID is listed in this file forks, the child's PID will
300 automatically be added to this file, and the child will not be
301 traced by the function tracer as well. This option will also
302 cause PIDs of tasks that exit to be removed from the file.
304 If a PID is in both this file and "set_ftrace_pid", then this
305 file takes precedence, and the thread will not be traced.
309 Have the events only trace a task with a PID listed in this file.
310 Note, sched_switch and sched_wake_up will also trace events
313 To have the PIDs of children of tasks with their PID in this file
314 added on fork, enable the "event-fork" option. That option will also
315 cause the PIDs of tasks to be removed from this file when the task
318 set_event_notrace_pid:
320 Have the events not trace a task with a PID listed in this file.
321 Note, sched_switch and sched_wakeup will trace threads not listed
322 in this file, even if a thread's PID is in the file if the
323 sched_switch or sched_wakeup events also trace a thread that should
326 To have the PIDs of children of tasks with their PID in this file
327 added on fork, enable the "event-fork" option. That option will also
328 cause the PIDs of tasks to be removed from this file when the task
333 Functions listed in this file will cause the function graph
334 tracer to only trace these functions and the functions that
335 they call. (See the section "dynamic ftrace" for more details).
336 Note, set_ftrace_filter and set_ftrace_notrace still affects
337 what functions are being traced.
341 Similar to set_graph_function, but will disable function graph
342 tracing when the function is hit until it exits the function.
343 This makes it possible to ignore tracing functions that are called
344 by a specific function.
346 available_filter_functions:
348 This lists the functions that ftrace has processed and can trace.
349 These are the function names that you can pass to
350 "set_ftrace_filter", "set_ftrace_notrace",
351 "set_graph_function", or "set_graph_notrace".
352 (See the section "dynamic ftrace" below for more details.)
354 available_filter_functions_addrs:
356 Similar to available_filter_functions, but with address displayed
357 for each function. The displayed address is the patch-site address
358 and can differ from /proc/kallsyms address.
360 dyn_ftrace_total_info:
362 This file is for debugging purposes. The number of functions that
363 have been converted to nops and are available to be traced.
367 This file is more for debugging ftrace, but can also be useful
368 in seeing if any function has a callback attached to it.
369 Not only does the trace infrastructure use ftrace function
370 trace utility, but other subsystems might too. This file
371 displays all functions that have a callback attached to them
372 as well as the number of callbacks that have been attached.
373 Note, a callback may also call multiple functions which will
374 not be listed in this count.
376 If the callback registered to be traced by a function with
377 the "save regs" attribute (thus even more overhead), a 'R'
378 will be displayed on the same line as the function that
379 is returning registers.
381 If the callback registered to be traced by a function with
382 the "ip modify" attribute (thus the regs->ip can be changed),
383 an 'I' will be displayed on the same line as the function that
386 If a non ftrace trampoline is attached (BPF) a 'D' will be displayed.
387 Note, normal ftrace trampolines can also be attached, but only one
388 "direct" trampoline can be attached to a given function at a time.
390 Some architectures can not call direct trampolines, but instead have
391 the ftrace ops function located above the function entry point. In
392 such cases an 'O' will be displayed.
394 If a function had either the "ip modify" or a "direct" call attached to
395 it in the past, a 'M' will be shown. This flag is never cleared. It is
396 used to know if a function was every modified by the ftrace infrastructure,
397 and can be used for debugging.
399 If the architecture supports it, it will also show what callback
400 is being directly called by the function. If the count is greater
401 than 1 it most likely will be ftrace_ops_list_func().
403 If the callback of a function jumps to a trampoline that is
404 specific to the callback and which is not the standard trampoline,
405 its address will be printed as well as the function that the
410 This file contains all the functions that ever had a function callback
411 to it via the ftrace infrastructure. It has the same format as
412 enabled_functions but shows all functions that have every been
415 To see any function that has every been modified by "ip modify" or a
416 direct trampoline, one can perform the following command:
418 grep ' M ' /sys/kernel/tracing/touched_functions
420 function_profile_enabled:
422 When set it will enable all functions with either the function
423 tracer, or if configured, the function graph tracer. It will
424 keep a histogram of the number of functions that were called
425 and if the function graph tracer was configured, it will also keep
426 track of the time spent in those functions. The histogram
427 content can be displayed in the files:
429 trace_stat/function<cpu> ( function0, function1, etc).
433 A directory that holds different tracing stats.
437 Enable dynamic trace points. See kprobetrace.rst.
441 Dynamic trace points stats. See kprobetrace.rst.
445 Used with the function graph tracer. This is the max depth
446 it will trace into a function. Setting this to a value of
447 one will show only the first kernel function that is called
452 This is for tools that read the raw format files. If an event in
453 the ring buffer references a string, only a pointer to the string
454 is recorded into the buffer and not the string itself. This prevents
455 tools from knowing what that string was. This file displays the string
456 and address for the string allowing tools to map the pointers to what
461 Only the pid of the task is recorded in a trace event unless
462 the event specifically saves the task comm as well. Ftrace
463 makes a cache of pid mappings to comms to try to display
464 comms for events. If a pid for a comm is not listed, then
465 "<...>" is displayed in the output.
467 If the option "record-cmd" is set to "0", then comms of tasks
468 will not be saved during recording. By default, it is enabled.
472 By default, 128 comms are saved (see "saved_cmdlines" above). To
473 increase or decrease the amount of comms that are cached, echo
474 the number of comms to cache into this file.
478 If the option "record-tgid" is set, on each scheduling context switch
479 the Task Group ID of a task is saved in a table mapping the PID of
480 the thread to its TGID. By default, the "record-tgid" option is
485 This displays the "snapshot" buffer and also lets the user
486 take a snapshot of the current running trace.
487 See the "Snapshot" section below for more details.
491 When the stack tracer is activated, this will display the
492 maximum stack size it has encountered.
493 See the "Stack Trace" section below.
497 This displays the stack back trace of the largest stack
498 that was encountered when the stack tracer is activated.
499 See the "Stack Trace" section below.
503 This is similar to "set_ftrace_filter" but it limits what
504 functions the stack tracer will check.
508 Whenever an event is recorded into the ring buffer, a
509 "timestamp" is added. This stamp comes from a specified
510 clock. By default, ftrace uses the "local" clock. This
511 clock is very fast and strictly per cpu, but on some
512 systems it may not be monotonic with respect to other
513 CPUs. In other words, the local clocks may not be in sync
514 with local clocks on other CPUs.
516 Usual clocks for tracing::
519 [local] global counter x86-tsc
521 The clock with the square brackets around it is the one in effect.
524 Default clock, but may not be in sync across CPUs
527 This clock is in sync with all CPUs but may
528 be a bit slower than the local clock.
531 This is not a clock at all, but literally an atomic
532 counter. It counts up one by one, but is in sync
533 with all CPUs. This is useful when you need to
534 know exactly the order events occurred with respect to
535 each other on different CPUs.
538 This uses the jiffies counter and the time stamp
539 is relative to the time since boot up.
542 This makes ftrace use the same clock that perf uses.
543 Eventually perf will be able to read ftrace buffers
544 and this will help out in interleaving the data.
547 Architectures may define their own clocks. For
548 example, x86 uses its own TSC cycle clock here.
551 This uses the powerpc timebase register value.
552 This is in sync across CPUs and can also be used
553 to correlate events across hypervisor/guest if
557 This uses the fast monotonic clock (CLOCK_MONOTONIC)
558 which is monotonic and is subject to NTP rate adjustments.
561 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
562 which is monotonic but is not subject to any rate adjustments
563 and ticks at the same rate as the hardware clocksource.
566 This is the boot clock (CLOCK_BOOTTIME) and is based on the
567 fast monotonic clock, but also accounts for time spent in
568 suspend. Since the clock access is designed for use in
569 tracing in the suspend path, some side effects are possible
570 if clock is accessed after the suspend time is accounted before
571 the fast mono clock is updated. In this case, the clock update
572 appears to happen slightly sooner than it normally would have.
573 Also on 32-bit systems, it's possible that the 64-bit boot offset
574 sees a partial update. These effects are rare and post
575 processing should be able to handle them. See comments in the
576 ktime_get_boot_fast_ns() function for more information.
579 This is the tai clock (CLOCK_TAI) and is derived from the wall-
580 clock time. However, this clock does not experience
581 discontinuities and backwards jumps caused by NTP inserting leap
582 seconds. Since the clock access is designed for use in tracing,
583 side effects are possible. The clock access may yield wrong
584 readouts in case the internal TAI offset is updated e.g., caused
585 by setting the system time or using adjtimex() with an offset.
586 These effects are rare and post processing should be able to
587 handle them. See comments in the ktime_get_tai_fast_ns()
588 function for more information.
590 To set a clock, simply echo the clock name into this file::
592 # echo global > trace_clock
594 Setting a clock clears the ring buffer content as well as the
599 This is a very useful file for synchronizing user space
600 with events happening in the kernel. Writing strings into
601 this file will be written into the ftrace buffer.
603 It is useful in applications to open this file at the start
604 of the application and just reference the file descriptor
607 void trace_write(const char *fmt, ...)
617 n = vsnprintf(buf, 256, fmt, ap);
620 write(trace_fd, buf, n);
625 trace_fd = open("trace_marker", O_WRONLY);
627 Note: Writing into the trace_marker file can also initiate triggers
628 that are written into /sys/kernel/tracing/events/ftrace/print/trigger
629 See "Event triggers" in Documentation/trace/events.rst and an
630 example in Documentation/trace/histogram.rst (Section 3.)
634 This is similar to trace_marker above, but is meant for binary data
635 to be written to it, where a tool can be used to parse the data
640 Add dynamic tracepoints in programs.
645 Uprobe statistics. See uprobetrace.txt
649 This is a way to make multiple trace buffers where different
650 events can be recorded in different buffers.
651 See "Instances" section below.
655 This is the trace event directory. It holds event tracepoints
656 (also known as static tracepoints) that have been compiled
657 into the kernel. It shows what event tracepoints exist
658 and how they are grouped by system. There are "enable"
659 files at various levels that can enable the tracepoints
660 when a "1" is written to them.
662 See events.rst for more information.
666 By echoing in the event into this file, will enable that event.
668 See events.rst for more information.
672 A list of events that can be enabled in tracing.
674 See events.rst for more information.
678 Certain tracers may change the timestamp mode used when
679 logging trace events into the event buffer. Events with
680 different modes can coexist within a buffer but the mode in
681 effect when an event is logged determines which timestamp mode
682 is used for that event. The default timestamp mode is
685 Usual timestamp modes for tracing:
690 The timestamp mode with the square brackets around it is the
693 delta: Default timestamp mode - timestamp is a delta against
694 a per-buffer timestamp.
696 absolute: The timestamp is a full timestamp, not a delta
697 against some other value. As such it takes up more
698 space and is less efficient.
702 Directory for the Hardware Latency Detector.
703 See "Hardware Latency Detector" section below.
707 This is a directory that contains the trace per_cpu information.
709 per_cpu/cpu0/buffer_size_kb:
711 The ftrace buffer is defined per_cpu. That is, there's a separate
712 buffer for each CPU to allow writes to be done atomically,
713 and free from cache bouncing. These buffers may have different
714 size buffers. This file is similar to the buffer_size_kb
715 file, but it only displays or sets the buffer size for the
716 specific CPU. (here cpu0).
720 This is similar to the "trace" file, but it will only display
721 the data specific for the CPU. If written to, it only clears
722 the specific CPU buffer.
724 per_cpu/cpu0/trace_pipe
726 This is similar to the "trace_pipe" file, and is a consuming
727 read, but it will only display (and consume) the data specific
730 per_cpu/cpu0/trace_pipe_raw
732 For tools that can parse the ftrace ring buffer binary format,
733 the trace_pipe_raw file can be used to extract the data
734 from the ring buffer directly. With the use of the splice()
735 system call, the buffer data can be quickly transferred to
736 a file or to the network where a server is collecting the
739 Like trace_pipe, this is a consuming reader, where multiple
740 reads will always produce different data.
742 per_cpu/cpu0/snapshot:
744 This is similar to the main "snapshot" file, but will only
745 snapshot the current CPU (if supported). It only displays
746 the content of the snapshot for a given CPU, and if
747 written to, only clears this CPU buffer.
749 per_cpu/cpu0/snapshot_raw:
751 Similar to the trace_pipe_raw, but will read the binary format
752 from the snapshot buffer for the given CPU.
756 This displays certain stats about the ring buffer:
759 The number of events that are still in the buffer.
762 The number of lost events due to overwriting when
766 Should always be zero.
767 This gets set if so many events happened within a nested
768 event (ring buffer is re-entrant), that it fills the
769 buffer and starts dropping events.
772 Bytes actually read (not overwritten).
775 The oldest timestamp in the buffer
778 The current timestamp
781 Events lost due to overwrite option being off.
784 The number of events read.
789 Here is the list of current tracers that may be configured.
793 Function call tracer to trace all kernel functions.
797 Similar to the function tracer except that the
798 function tracer probes the functions on their entry
799 whereas the function graph tracer traces on both entry
800 and exit of the functions. It then provides the ability
801 to draw a graph of function calls similar to C code
806 The block tracer. The tracer used by the blktrace user
811 The Hardware Latency tracer is used to detect if the hardware
812 produces any latency. See "Hardware Latency Detector" section
817 Traces the areas that disable interrupts and saves
818 the trace with the longest max latency.
819 See tracing_max_latency. When a new max is recorded,
820 it replaces the old trace. It is best to view this
821 trace with the latency-format option enabled, which
822 happens automatically when the tracer is selected.
826 Similar to irqsoff but traces and records the amount of
827 time for which preemption is disabled.
831 Similar to irqsoff and preemptoff, but traces and
832 records the largest time for which irqs and/or preemption
837 Traces and records the max latency that it takes for
838 the highest priority task to get scheduled after
839 it has been woken up.
840 Traces all tasks as an average developer would expect.
844 Traces and records the max latency that it takes for just
845 RT tasks (as the current "wakeup" does). This is useful
846 for those interested in wake up timings of RT tasks.
850 Traces and records the max latency that it takes for
851 a SCHED_DEADLINE task to be woken (as the "wakeup" and
856 A special tracer that is used to trace binary module.
857 It will trace all the calls that a module makes to the
858 hardware. Everything it writes and reads from the I/O
863 This tracer can be configured when tracing likely/unlikely
864 calls within the kernel. It will trace when a likely and
865 unlikely branch is hit and if it was correct in its prediction
870 This is the "trace nothing" tracer. To remove all
871 tracers from tracing simply echo "nop" into
877 For most ftrace commands, failure modes are obvious and communicated
878 using standard return codes.
880 For other more involved commands, extended error information may be
881 available via the tracing/error_log file. For the commands that
882 support it, reading the tracing/error_log file after an error will
883 display more detailed information about what went wrong, if
884 information is available. The tracing/error_log file is a circular
885 error log displaying a small number (currently, 8) of ftrace errors
886 for the last (8) failed commands.
888 The extended error information and usage takes the form shown in
891 # echo xxx > /sys/kernel/tracing/events/sched/sched_wakeup/trigger
892 echo: write error: Invalid argument
894 # cat /sys/kernel/tracing/error_log
895 [ 5348.887237] location: error: Couldn't yyy: zzz
898 [ 7517.023364] location: error: Bad rrr: sss
902 To clear the error log, echo the empty string into it::
904 # echo > /sys/kernel/tracing/error_log
906 Examples of using the tracer
907 ----------------------------
909 Here are typical examples of using the tracers when controlling
910 them only with the tracefs interface (without using any
911 user-land utilities).
916 Here is an example of the output format of the file "trace"::
920 # entries-in-buffer/entries-written: 140080/250280 #P:4
923 # / _----=> need-resched
924 # | / _---=> hardirq/softirq
925 # || / _--=> preempt-depth
927 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
929 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
930 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
931 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
932 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
933 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
934 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
935 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
936 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
937 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
938 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
941 A header is printed with the tracer name that is represented by
942 the trace. In this case the tracer is "function". Then it shows the
943 number of events in the buffer as well as the total number of entries
944 that were written. The difference is the number of entries that were
945 lost due to the buffer filling up (250280 - 140080 = 110200 events
948 The header explains the content of the events. Task name "bash", the task
949 PID "1977", the CPU that it was running on "000", the latency format
950 (explained below), the timestamp in <secs>.<usecs> format, the
951 function name that was traced "sys_close" and the parent function that
952 called this function "system_call_fastpath". The timestamp is the time
953 at which the function was entered.
958 When the latency-format option is enabled or when one of the latency
959 tracers is set, the trace file gives somewhat more information to see
960 why a latency happened. Here is a typical trace::
964 # irqsoff latency trace v1.1.5 on 3.8.0-test+
965 # --------------------------------------------------------------------
966 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
968 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
970 # => started at: __lock_task_sighand
971 # => ended at: _raw_spin_unlock_irqrestore
975 # / _-----=> irqs-off
976 # | / _----=> need-resched
977 # || / _---=> hardirq/softirq
978 # ||| / _--=> preempt-depth
980 # cmd pid ||||| time | caller
982 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
983 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
984 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
985 ps-6143 2d..1 306us : <stack trace>
986 => trace_hardirqs_on_caller
988 => _raw_spin_unlock_irqrestore
995 => system_call_fastpath
998 This shows that the current tracer is "irqsoff" tracing the time
999 for which interrupts were disabled. It gives the trace version (which
1000 never changes) and the version of the kernel upon which this was executed on
1001 (3.8). Then it displays the max latency in microseconds (259 us). The number
1002 of trace entries displayed and the total number (both are four: #4/4).
1003 VP, KP, SP, and HP are always zero and are reserved for later use.
1004 #P is the number of online CPUs (#P:4).
1006 The task is the process that was running when the latency
1007 occurred. (ps pid: 6143).
1009 The start and stop (the functions in which the interrupts were
1010 disabled and enabled respectively) that caused the latencies:
1012 - __lock_task_sighand is where the interrupts were disabled.
1013 - _raw_spin_unlock_irqrestore is where they were enabled again.
1015 The next lines after the header are the trace itself. The header
1016 explains which is which.
1018 cmd: The name of the process in the trace.
1020 pid: The PID of that process.
1022 CPU#: The CPU which the process was running on.
1024 irqs-off: 'd' interrupts are disabled. '.' otherwise.
1025 .. caution:: If the architecture does not support a way to
1026 read the irq flags variable, an 'X' will always
1030 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
1031 - 'n' only TIF_NEED_RESCHED is set,
1032 - 'p' only PREEMPT_NEED_RESCHED is set,
1036 - 'Z' - NMI occurred inside a hardirq
1037 - 'z' - NMI is running
1038 - 'H' - hard irq occurred inside a softirq.
1039 - 'h' - hard irq is running
1040 - 's' - soft irq is running
1041 - '.' - normal context.
1043 preempt-depth: The level of preempt_disabled
1045 The above is mostly meaningful for kernel developers.
1048 When the latency-format option is enabled, the trace file
1049 output includes a timestamp relative to the start of the
1050 trace. This differs from the output when latency-format
1051 is disabled, which includes an absolute timestamp.
1054 This is just to help catch your eye a bit better. And
1055 needs to be fixed to be only relative to the same CPU.
1056 The marks are determined by the difference between this
1057 current trace and the next trace.
1059 - '$' - greater than 1 second
1060 - '@' - greater than 100 millisecond
1061 - '*' - greater than 10 millisecond
1062 - '#' - greater than 1000 microsecond
1063 - '!' - greater than 100 microsecond
1064 - '+' - greater than 10 microsecond
1065 - ' ' - less than or equal to 10 microsecond.
1067 The rest is the same as the 'trace' file.
1069 Note, the latency tracers will usually end with a back trace
1070 to easily find where the latency occurred.
1075 The trace_options file (or the options directory) is used to control
1076 what gets printed in the trace output, or manipulate the tracers.
1077 To see what is available, simply cat the file::
1109 To disable one of the options, echo in the option prepended with
1112 echo noprint-parent > trace_options
1114 To enable an option, leave off the "no"::
1116 echo sym-offset > trace_options
1118 Here are the available options:
1121 On function traces, display the calling (parent)
1122 function as well as the function being traced.
1126 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
1129 bash-4000 [01] 1477.606694: simple_strtoul
1133 Display not only the function name, but also the
1134 offset in the function. For example, instead of
1135 seeing just "ktime_get", you will see
1136 "ktime_get+0xb/0x20".
1140 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
1143 This will also display the function address as well
1144 as the function name.
1148 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
1151 This deals with the trace file when the
1152 latency-format option is enabled.
1155 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
1156 (+0.000ms): simple_strtoul (kstrtoul)
1159 This will display raw numbers. This option is best for
1160 use with user applications that can translate the raw
1161 numbers better than having it done in the kernel.
1164 Similar to raw, but the numbers will be in a hexadecimal format.
1167 This will print out the formats in raw binary.
1170 When set, reading trace_pipe will not block when polled.
1173 Print the fields as described by their types. This is a better
1174 option than using hex, bin or raw, as it gives a better parsing
1175 of the content of the event.
1178 Can disable trace_printk() from writing into the buffer.
1181 It is sometimes confusing when the CPU buffers are full
1182 and one CPU buffer had a lot of events recently, thus
1183 a shorter time frame, were another CPU may have only had
1184 a few events, which lets it have older events. When
1185 the trace is reported, it shows the oldest events first,
1186 and it may look like only one CPU ran (the one with the
1187 oldest events). When the annotate option is set, it will
1188 display when a new CPU buffer started::
1190 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
1191 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
1192 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
1193 ##### CPU 2 buffer started ####
1194 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
1195 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
1196 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
1199 This option changes the trace. It records a
1200 stacktrace of the current user space thread after
1204 when user stacktrace are enabled, look up which
1205 object the address belongs to, and print a
1206 relative address. This is especially useful when
1207 ASLR is on, otherwise you don't get a chance to
1208 resolve the address to object/file/line after
1209 the app is no longer running
1211 The lookup is performed when you read
1212 trace,trace_pipe. Example::
1214 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
1215 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
1219 When set, trace_printk()s will only show the format
1220 and not their parameters (if trace_bprintk() or
1221 trace_bputs() was used to save the trace_printk()).
1224 Show only the event data. Hides the comm, PID,
1225 timestamp, CPU, and other useful data.
1228 This option changes the trace output. When it is enabled,
1229 the trace displays additional information about the
1230 latency, as described in "Latency trace format".
1233 When set, opening the trace file for read, will pause
1234 writing to the ring buffer (as if tracing_on was set to zero).
1235 This simulates the original behavior of the trace file.
1236 When the file is closed, tracing will be enabled again.
1239 When set, "%p" in the event printk format displays the
1240 hashed pointer value instead of real address.
1241 This will be useful if you want to find out which hashed
1242 value is corresponding to the real value in trace log.
1245 When any event or tracer is enabled, a hook is enabled
1246 in the sched_switch trace point to fill comm cache
1247 with mapped pids and comms. But this may cause some
1248 overhead, and if you only care about pids, and not the
1249 name of the task, disabling this option can lower the
1250 impact of tracing. See "saved_cmdlines".
1253 When any event or tracer is enabled, a hook is enabled
1254 in the sched_switch trace point to fill the cache of
1255 mapped Thread Group IDs (TGID) mapping to pids. See
1259 This controls what happens when the trace buffer is
1260 full. If "1" (default), the oldest events are
1261 discarded and overwritten. If "0", then the newest
1262 events are discarded.
1263 (see per_cpu/cpu0/stats for overrun and dropped)
1266 When the free_buffer is closed, tracing will
1267 stop (tracing_on set to 0).
1270 Shows the interrupt, preempt count, need resched data.
1271 When disabled, the trace looks like::
1275 # entries-in-buffer/entries-written: 144405/9452052 #P:4
1277 # TASK-PID CPU# TIMESTAMP FUNCTION
1279 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
1280 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
1281 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
1285 When set, the trace_marker is writable (only by root).
1286 When disabled, the trace_marker will error with EINVAL
1290 When set, tasks with PIDs listed in set_event_pid will have
1291 the PIDs of their children added to set_event_pid when those
1292 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
1293 their PIDs will be removed from the file.
1295 This affects PIDs listed in set_event_notrace_pid as well.
1298 The latency tracers will enable function tracing
1299 if this option is enabled (default it is). When
1300 it is disabled, the latency tracers do not trace
1301 functions. This keeps the overhead of the tracer down
1302 when performing latency tests.
1305 When set, tasks with PIDs listed in set_ftrace_pid will
1306 have the PIDs of their children added to set_ftrace_pid
1307 when those tasks fork. Also, when tasks with PIDs in
1308 set_ftrace_pid exit, their PIDs will be removed from the
1311 This affects PIDs in set_ftrace_notrace_pid as well.
1314 When set, the latency tracers (irqsoff, wakeup, etc) will
1315 use function graph tracing instead of function tracing.
1318 When set, a stack trace is recorded after any trace event
1322 Enable branch tracing with the tracer. This enables branch
1323 tracer along with the currently set tracer. Enabling this
1324 with the "nop" tracer is the same as just enabling the
1327 .. tip:: Some tracers have their own options. They only appear in this
1328 file when the tracer is active. They always appear in the
1332 Here are the per tracer options:
1334 Options for function tracer:
1337 When set, a stack trace is recorded after every
1338 function that is recorded. NOTE! Limit the functions
1339 that are recorded before enabling this, with
1340 "set_ftrace_filter" otherwise the system performance
1341 will be critically degraded. Remember to disable
1342 this option before clearing the function filter.
1344 Options for function_graph tracer:
1346 Since the function_graph tracer has a slightly different output
1347 it has its own options to control what is displayed.
1350 When set, the "overrun" of the graph stack is
1351 displayed after each function traced. The
1352 overrun, is when the stack depth of the calls
1353 is greater than what is reserved for each task.
1354 Each task has a fixed array of functions to
1355 trace in the call graph. If the depth of the
1356 calls exceeds that, the function is not traced.
1357 The overrun is the number of functions missed
1358 due to exceeding this array.
1361 When set, the CPU number of the CPU where the trace
1362 occurred is displayed.
1365 When set, if the function takes longer than
1366 A certain amount, then a delay marker is
1367 displayed. See "delay" above, under the
1371 Unlike other tracers, the process' command line
1372 is not displayed by default, but instead only
1373 when a task is traced in and out during a context
1374 switch. Enabling this options has the command
1375 of each process displayed at every line.
1378 At the end of each function (the return)
1379 the duration of the amount of time in the
1380 function is displayed in microseconds.
1383 When set, the timestamp is displayed at each line.
1386 When disabled, functions that happen inside an
1387 interrupt will not be traced.
1390 When set, the return event will include the function
1391 that it represents. By default this is off, and
1392 only a closing curly bracket "}" is displayed for
1393 the return of a function.
1396 When set, the return value of each traced function
1397 will be printed after an equal sign "=". By default
1400 funcgraph-retval-hex
1401 When set, the return value will always be printed
1402 in hexadecimal format. If the option is not set and
1403 the return value is an error code, it will be printed
1404 in signed decimal format; otherwise it will also be
1405 printed in hexadecimal format. By default, this option
1409 When running function graph tracer, to include
1410 the time a task schedules out in its function.
1411 When enabled, it will account time the task has been
1412 scheduled out as part of the function call.
1415 When running function profiler with function graph tracer,
1416 to include the time to call nested functions. When this is
1417 not set, the time reported for the function will only
1418 include the time the function itself executed for, not the
1419 time for functions that it called.
1421 Options for blk tracer:
1424 Shows a more minimalistic output.
1430 When interrupts are disabled, the CPU can not react to any other
1431 external event (besides NMIs and SMIs). This prevents the timer
1432 interrupt from triggering or the mouse interrupt from letting
1433 the kernel know of a new mouse event. The result is a latency
1434 with the reaction time.
1436 The irqsoff tracer tracks the time for which interrupts are
1437 disabled. When a new maximum latency is hit, the tracer saves
1438 the trace leading up to that latency point so that every time a
1439 new maximum is reached, the old saved trace is discarded and the
1442 To reset the maximum, echo 0 into tracing_max_latency. Here is
1445 # echo 0 > options/function-trace
1446 # echo irqsoff > current_tracer
1447 # echo 1 > tracing_on
1448 # echo 0 > tracing_max_latency
1451 # echo 0 > tracing_on
1455 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1456 # --------------------------------------------------------------------
1457 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1459 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
1461 # => started at: run_timer_softirq
1462 # => ended at: run_timer_softirq
1466 # / _-----=> irqs-off
1467 # | / _----=> need-resched
1468 # || / _---=> hardirq/softirq
1469 # ||| / _--=> preempt-depth
1471 # cmd pid ||||| time | caller
1473 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
1474 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
1475 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
1476 <idle>-0 0dNs3 25us : <stack trace>
1477 => _raw_spin_unlock_irq
1478 => run_timer_softirq
1483 => smp_apic_timer_interrupt
1484 => apic_timer_interrupt
1489 => x86_64_start_reservations
1490 => x86_64_start_kernel
1492 Here we see that we had a latency of 16 microseconds (which is
1493 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
1494 interrupts. The difference between the 16 and the displayed
1495 timestamp 25us occurred because the clock was incremented
1496 between the time of recording the max latency and the time of
1497 recording the function that had that latency.
1499 Note the above example had function-trace not set. If we set
1500 function-trace, we get a much larger output::
1502 with echo 1 > options/function-trace
1506 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1507 # --------------------------------------------------------------------
1508 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1510 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1512 # => started at: ata_scsi_queuecmd
1513 # => ended at: ata_scsi_queuecmd
1517 # / _-----=> irqs-off
1518 # | / _----=> need-resched
1519 # || / _---=> hardirq/softirq
1520 # ||| / _--=> preempt-depth
1522 # cmd pid ||||| time | caller
1524 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1525 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1526 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1527 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1528 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1529 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1530 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1531 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1532 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1534 bash-2042 3d..1 67us : delay_tsc <-__delay
1535 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1536 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1537 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1538 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1539 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1540 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1541 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1542 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1543 bash-2042 3d..1 120us : <stack trace>
1544 => _raw_spin_unlock_irqrestore
1545 => ata_scsi_queuecmd
1546 => scsi_dispatch_cmd
1548 => __blk_run_queue_uncond
1551 => submit_bio_noacct
1554 => __ext3_get_inode_loc
1563 => user_path_at_empty
1568 => system_call_fastpath
1571 Here we traced a 71 microsecond latency. But we also see all the
1572 functions that were called during that time. Note that by
1573 enabling function tracing, we incur an added overhead. This
1574 overhead may extend the latency times. But nevertheless, this
1575 trace has provided some very helpful debugging information.
1577 If we prefer function graph output instead of function, we can set
1578 display-graph option::
1580 with echo 1 > options/display-graph
1584 # irqsoff latency trace v1.1.5 on 4.20.0-rc6+
1585 # --------------------------------------------------------------------
1586 # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4)
1588 # | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0)
1590 # => started at: free_debug_processing
1591 # => ended at: return_to_handler
1595 # / _----=> need-resched
1596 # | / _---=> hardirq/softirq
1597 # || / _--=> preempt-depth
1599 # REL TIME CPU TASK/PID |||| DURATION FUNCTION CALLS
1600 # | | | | |||| | | | | | |
1601 0 us | 0) bash-1507 | d... | 0.000 us | _raw_spin_lock_irqsave();
1602 0 us | 0) bash-1507 | d..1 | 0.378 us | do_raw_spin_trylock();
1603 1 us | 0) bash-1507 | d..2 | | set_track() {
1604 2 us | 0) bash-1507 | d..2 | | save_stack_trace() {
1605 2 us | 0) bash-1507 | d..2 | | __save_stack_trace() {
1606 3 us | 0) bash-1507 | d..2 | | __unwind_start() {
1607 3 us | 0) bash-1507 | d..2 | | get_stack_info() {
1608 3 us | 0) bash-1507 | d..2 | 0.351 us | in_task_stack();
1609 4 us | 0) bash-1507 | d..2 | 1.107 us | }
1611 3750 us | 0) bash-1507 | d..1 | 0.516 us | do_raw_spin_unlock();
1612 3750 us | 0) bash-1507 | d..1 | 0.000 us | _raw_spin_unlock_irqrestore();
1613 3764 us | 0) bash-1507 | d..1 | 0.000 us | tracer_hardirqs_on();
1614 bash-1507 0d..1 3792us : <stack trace>
1615 => free_debug_processing
1624 => search_binary_handler
1625 => __do_execve_file.isra.32
1628 => entry_SYSCALL_64_after_hwframe
1633 When preemption is disabled, we may be able to receive
1634 interrupts but the task cannot be preempted and a higher
1635 priority task must wait for preemption to be enabled again
1636 before it can preempt a lower priority task.
1638 The preemptoff tracer traces the places that disable preemption.
1639 Like the irqsoff tracer, it records the maximum latency for
1640 which preemption was disabled. The control of preemptoff tracer
1641 is much like the irqsoff tracer.
1644 # echo 0 > options/function-trace
1645 # echo preemptoff > current_tracer
1646 # echo 1 > tracing_on
1647 # echo 0 > tracing_max_latency
1650 # echo 0 > tracing_on
1652 # tracer: preemptoff
1654 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1655 # --------------------------------------------------------------------
1656 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1658 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1660 # => started at: do_IRQ
1661 # => ended at: do_IRQ
1665 # / _-----=> irqs-off
1666 # | / _----=> need-resched
1667 # || / _---=> hardirq/softirq
1668 # ||| / _--=> preempt-depth
1670 # cmd pid ||||| time | caller
1672 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1673 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1674 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1675 sshd-1991 1d..1 52us : <stack trace>
1676 => sub_preempt_count
1682 This has some more changes. Preemption was disabled when an
1683 interrupt came in (notice the 'h'), and was enabled on exit.
1684 But we also see that interrupts have been disabled when entering
1685 the preempt off section and leaving it (the 'd'). We do not know if
1686 interrupts were enabled in the mean time or shortly after this
1690 # tracer: preemptoff
1692 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1693 # --------------------------------------------------------------------
1694 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1696 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1698 # => started at: wake_up_new_task
1699 # => ended at: task_rq_unlock
1703 # / _-----=> irqs-off
1704 # | / _----=> need-resched
1705 # || / _---=> hardirq/softirq
1706 # ||| / _--=> preempt-depth
1708 # cmd pid ||||| time | caller
1710 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1711 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1712 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1713 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1714 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1716 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1717 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1718 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1719 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1720 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1721 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1722 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1723 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1725 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1726 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1727 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1728 bash-1994 1d..2 36us : do_softirq <-irq_exit
1729 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1730 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1731 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1732 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1733 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1734 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1736 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1737 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1738 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1739 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1740 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1741 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1742 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1743 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1744 bash-1994 1.N.1 104us : <stack trace>
1745 => sub_preempt_count
1746 => _raw_spin_unlock_irqrestore
1754 The above is an example of the preemptoff trace with
1755 function-trace set. Here we see that interrupts were not disabled
1756 the entire time. The irq_enter code lets us know that we entered
1757 an interrupt 'h'. Before that, the functions being traced still
1758 show that it is not in an interrupt, but we can see from the
1759 functions themselves that this is not the case.
1764 Knowing the locations that have interrupts disabled or
1765 preemption disabled for the longest times is helpful. But
1766 sometimes we would like to know when either preemption and/or
1767 interrupts are disabled.
1769 Consider the following code::
1771 local_irq_disable();
1772 call_function_with_irqs_off();
1774 call_function_with_irqs_and_preemption_off();
1776 call_function_with_preemption_off();
1779 The irqsoff tracer will record the total length of
1780 call_function_with_irqs_off() and
1781 call_function_with_irqs_and_preemption_off().
1783 The preemptoff tracer will record the total length of
1784 call_function_with_irqs_and_preemption_off() and
1785 call_function_with_preemption_off().
1787 But neither will trace the time that interrupts and/or
1788 preemption is disabled. This total time is the time that we can
1789 not schedule. To record this time, use the preemptirqsoff
1792 Again, using this trace is much like the irqsoff and preemptoff
1796 # echo 0 > options/function-trace
1797 # echo preemptirqsoff > current_tracer
1798 # echo 1 > tracing_on
1799 # echo 0 > tracing_max_latency
1802 # echo 0 > tracing_on
1804 # tracer: preemptirqsoff
1806 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1807 # --------------------------------------------------------------------
1808 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1810 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1812 # => started at: ata_scsi_queuecmd
1813 # => ended at: ata_scsi_queuecmd
1817 # / _-----=> irqs-off
1818 # | / _----=> need-resched
1819 # || / _---=> hardirq/softirq
1820 # ||| / _--=> preempt-depth
1822 # cmd pid ||||| time | caller
1824 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1825 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1826 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1827 ls-2230 3...1 111us : <stack trace>
1828 => sub_preempt_count
1829 => _raw_spin_unlock_irqrestore
1830 => ata_scsi_queuecmd
1831 => scsi_dispatch_cmd
1833 => __blk_run_queue_uncond
1836 => submit_bio_noacct
1841 => htree_dirblock_to_tree
1842 => ext3_htree_fill_tree
1846 => system_call_fastpath
1849 The trace_hardirqs_off_thunk is called from assembly on x86 when
1850 interrupts are disabled in the assembly code. Without the
1851 function tracing, we do not know if interrupts were enabled
1852 within the preemption points. We do see that it started with
1855 Here is a trace with function-trace set::
1857 # tracer: preemptirqsoff
1859 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1860 # --------------------------------------------------------------------
1861 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1863 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1865 # => started at: schedule
1866 # => ended at: mutex_unlock
1870 # / _-----=> irqs-off
1871 # | / _----=> need-resched
1872 # || / _---=> hardirq/softirq
1873 # ||| / _--=> preempt-depth
1875 # cmd pid ||||| time | caller
1877 kworker/-59 3...1 0us : __schedule <-schedule
1878 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1879 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1880 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1881 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1882 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1883 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1884 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1885 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1886 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1887 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1888 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1889 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1890 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1891 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1892 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1893 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1894 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1895 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1896 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1897 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1898 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1899 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1900 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1901 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1902 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1903 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1904 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1905 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1906 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1907 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1908 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1910 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1911 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1912 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1913 ls-2269 3d..3 21us : do_softirq <-irq_exit
1914 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1915 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1916 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1917 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1918 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1919 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1920 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1922 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1923 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1924 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1925 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1926 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1927 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1929 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1930 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1931 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1932 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1933 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1934 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1935 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1936 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1937 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1938 ls-2269 3d... 186us : <stack trace>
1939 => __mutex_unlock_slowpath
1946 => system_call_fastpath
1948 This is an interesting trace. It started with kworker running and
1949 scheduling out and ls taking over. But as soon as ls released the
1950 rq lock and enabled interrupts (but not preemption) an interrupt
1951 triggered. When the interrupt finished, it started running softirqs.
1952 But while the softirq was running, another interrupt triggered.
1953 When an interrupt is running inside a softirq, the annotation is 'H'.
1959 One common case that people are interested in tracing is the
1960 time it takes for a task that is woken to actually wake up.
1961 Now for non Real-Time tasks, this can be arbitrary. But tracing
1962 it none the less can be interesting.
1964 Without function tracing::
1966 # echo 0 > options/function-trace
1967 # echo wakeup > current_tracer
1968 # echo 1 > tracing_on
1969 # echo 0 > tracing_max_latency
1971 # echo 0 > tracing_on
1975 # wakeup latency trace v1.1.5 on 3.8.0-test+
1976 # --------------------------------------------------------------------
1977 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1979 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1983 # / _-----=> irqs-off
1984 # | / _----=> need-resched
1985 # || / _---=> hardirq/softirq
1986 # ||| / _--=> preempt-depth
1988 # cmd pid ||||| time | caller
1990 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1991 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1992 <idle>-0 3d..3 15us : __schedule <-schedule
1993 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1995 The tracer only traces the highest priority task in the system
1996 to avoid tracing the normal circumstances. Here we see that
1997 the kworker with a nice priority of -20 (not very nice), took
1998 just 15 microseconds from the time it woke up, to the time it
2001 Non Real-Time tasks are not that interesting. A more interesting
2002 trace is to concentrate only on Real-Time tasks.
2007 In a Real-Time environment it is very important to know the
2008 wakeup time it takes for the highest priority task that is woken
2009 up to the time that it executes. This is also known as "schedule
2010 latency". I stress the point that this is about RT tasks. It is
2011 also important to know the scheduling latency of non-RT tasks,
2012 but the average schedule latency is better for non-RT tasks.
2013 Tools like LatencyTop are more appropriate for such
2016 Real-Time environments are interested in the worst case latency.
2017 That is the longest latency it takes for something to happen,
2018 and not the average. We can have a very fast scheduler that may
2019 only have a large latency once in a while, but that would not
2020 work well with Real-Time tasks. The wakeup_rt tracer was designed
2021 to record the worst case wakeups of RT tasks. Non-RT tasks are
2022 not recorded because the tracer only records one worst case and
2023 tracing non-RT tasks that are unpredictable will overwrite the
2024 worst case latency of RT tasks (just run the normal wakeup
2025 tracer for a while to see that effect).
2027 Since this tracer only deals with RT tasks, we will run this
2028 slightly differently than we did with the previous tracers.
2029 Instead of performing an 'ls', we will run 'sleep 1' under
2030 'chrt' which changes the priority of the task.
2033 # echo 0 > options/function-trace
2034 # echo wakeup_rt > current_tracer
2035 # echo 1 > tracing_on
2036 # echo 0 > tracing_max_latency
2038 # echo 0 > tracing_on
2044 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
2045 # --------------------------------------------------------------------
2046 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
2048 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
2052 # / _-----=> irqs-off
2053 # | / _----=> need-resched
2054 # || / _---=> hardirq/softirq
2055 # ||| / _--=> preempt-depth
2057 # cmd pid ||||| time | caller
2059 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
2060 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
2061 <idle>-0 3d..3 5us : __schedule <-schedule
2062 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
2065 Running this on an idle system, we see that it only took 5 microseconds
2066 to perform the task switch. Note, since the trace point in the schedule
2067 is before the actual "switch", we stop the tracing when the recorded task
2068 is about to schedule in. This may change if we add a new marker at the
2069 end of the scheduler.
2071 Notice that the recorded task is 'sleep' with the PID of 2389
2072 and it has an rt_prio of 5. This priority is user-space priority
2073 and not the internal kernel priority. The policy is 1 for
2074 SCHED_FIFO and 2 for SCHED_RR.
2076 Note, that the trace data shows the internal priority (99 - rtprio).
2079 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
2081 The 0:120:R means idle was running with a nice priority of 0 (120 - 120)
2082 and in the running state 'R'. The sleep task was scheduled in with
2083 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
2084 and it too is in the running state.
2086 Doing the same with chrt -r 5 and function-trace set.
2089 echo 1 > options/function-trace
2093 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
2094 # --------------------------------------------------------------------
2095 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
2097 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
2101 # / _-----=> irqs-off
2102 # | / _----=> need-resched
2103 # || / _---=> hardirq/softirq
2104 # ||| / _--=> preempt-depth
2106 # cmd pid ||||| time | caller
2108 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
2109 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
2110 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
2111 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
2112 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
2113 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
2114 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
2115 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
2116 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
2117 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2118 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
2119 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
2120 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
2121 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
2122 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
2123 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
2124 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
2125 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
2126 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
2127 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
2128 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
2129 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
2130 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
2131 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
2132 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
2133 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
2134 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
2135 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
2136 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
2137 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
2138 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
2139 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
2140 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
2141 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
2142 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
2143 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
2144 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
2145 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit
2146 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
2147 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
2148 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
2149 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
2150 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
2151 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
2152 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
2153 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
2154 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
2155 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
2156 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
2157 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
2158 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
2159 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2160 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
2161 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
2162 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
2163 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2164 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
2165 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
2166 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
2167 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
2168 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
2169 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
2170 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
2171 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
2172 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
2173 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
2174 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
2175 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2176 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
2177 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
2178 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
2179 <idle>-0 3.N.. 25us : schedule <-cpu_idle
2180 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
2181 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
2182 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
2183 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
2184 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
2185 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
2186 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
2187 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
2188 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
2189 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
2190 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
2191 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
2192 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
2194 This isn't that big of a trace, even with function tracing enabled,
2195 so I included the entire trace.
2197 The interrupt went off while when the system was idle. Somewhere
2198 before task_woken_rt() was called, the NEED_RESCHED flag was set,
2199 this is indicated by the first occurrence of the 'N' flag.
2201 Latency tracing and events
2202 --------------------------
2203 As function tracing can induce a much larger latency, but without
2204 seeing what happens within the latency it is hard to know what
2205 caused it. There is a middle ground, and that is with enabling
2209 # echo 0 > options/function-trace
2210 # echo wakeup_rt > current_tracer
2211 # echo 1 > events/enable
2212 # echo 1 > tracing_on
2213 # echo 0 > tracing_max_latency
2215 # echo 0 > tracing_on
2219 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
2220 # --------------------------------------------------------------------
2221 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
2223 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
2227 # / _-----=> irqs-off
2228 # | / _----=> need-resched
2229 # || / _---=> hardirq/softirq
2230 # ||| / _--=> preempt-depth
2232 # cmd pid ||||| time | caller
2234 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
2235 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
2236 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
2237 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
2238 <idle>-0 2.N.2 2us : power_end: cpu_id=2
2239 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
2240 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
2241 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
2242 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
2243 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
2244 <idle>-0 2d..3 6us : __schedule <-schedule
2245 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
2248 Hardware Latency Detector
2249 -------------------------
2251 The hardware latency detector is executed by enabling the "hwlat" tracer.
2253 NOTE, this tracer will affect the performance of the system as it will
2254 periodically make a CPU constantly busy with interrupts disabled.
2257 # echo hwlat > current_tracer
2262 # entries-in-buffer/entries-written: 13/13 #P:8
2265 # / _----=> need-resched
2266 # | / _---=> hardirq/softirq
2267 # || / _--=> preempt-depth
2269 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2271 <...>-1729 [001] d... 678.473449: #1 inner/outer(us): 11/12 ts:1581527483.343962693 count:6
2272 <...>-1729 [004] d... 689.556542: #2 inner/outer(us): 16/9 ts:1581527494.889008092 count:1
2273 <...>-1729 [005] d... 714.756290: #3 inner/outer(us): 16/16 ts:1581527519.678961629 count:5
2274 <...>-1729 [001] d... 718.788247: #4 inner/outer(us): 9/17 ts:1581527523.889012713 count:1
2275 <...>-1729 [002] d... 719.796341: #5 inner/outer(us): 13/9 ts:1581527524.912872606 count:1
2276 <...>-1729 [006] d... 844.787091: #6 inner/outer(us): 9/12 ts:1581527649.889048502 count:2
2277 <...>-1729 [003] d... 849.827033: #7 inner/outer(us): 18/9 ts:1581527654.889013793 count:1
2278 <...>-1729 [007] d... 853.859002: #8 inner/outer(us): 9/12 ts:1581527658.889065736 count:1
2279 <...>-1729 [001] d... 855.874978: #9 inner/outer(us): 9/11 ts:1581527660.861991877 count:1
2280 <...>-1729 [001] d... 863.938932: #10 inner/outer(us): 9/11 ts:1581527668.970010500 count:1 nmi-total:7 nmi-count:1
2281 <...>-1729 [007] d... 878.050780: #11 inner/outer(us): 9/12 ts:1581527683.385002600 count:1 nmi-total:5 nmi-count:1
2282 <...>-1729 [007] d... 886.114702: #12 inner/outer(us): 9/12 ts:1581527691.385001600 count:1
2285 The above output is somewhat the same in the header. All events will have
2286 interrupts disabled 'd'. Under the FUNCTION title there is:
2289 This is the count of events recorded that were greater than the
2290 tracing_threshold (See below).
2292 inner/outer(us): 11/11
2294 This shows two numbers as "inner latency" and "outer latency". The test
2295 runs in a loop checking a timestamp twice. The latency detected within
2296 the two timestamps is the "inner latency" and the latency detected
2297 after the previous timestamp and the next timestamp in the loop is
2298 the "outer latency".
2300 ts:1581527483.343962693
2302 The absolute timestamp that the first latency was recorded in the window.
2306 The number of times a latency was detected during the window.
2308 nmi-total:7 nmi-count:1
2310 On architectures that support it, if an NMI comes in during the
2311 test, the time spent in NMI is reported in "nmi-total" (in
2314 All architectures that have NMIs will show the "nmi-count" if an
2315 NMI comes in during the test.
2320 This gets automatically set to "10" to represent 10
2321 microseconds. This is the threshold of latency that
2322 needs to be detected before the trace will be recorded.
2324 Note, when hwlat tracer is finished (another tracer is
2325 written into "current_tracer"), the original value for
2326 tracing_threshold is placed back into this file.
2328 hwlat_detector/width
2329 The length of time the test runs with interrupts disabled.
2331 hwlat_detector/window
2332 The length of time of the window which the test
2333 runs. That is, the test will run for "width"
2334 microseconds per "window" microseconds
2337 When the test is started. A kernel thread is created that
2338 runs the test. This thread will alternate between CPUs
2339 listed in the tracing_cpumask between each period
2340 (one "window"). To limit the test to specific CPUs
2341 set the mask in this file to only the CPUs that the test
2347 This tracer is the function tracer. Enabling the function tracer
2348 can be done from the debug file system. Make sure the
2349 ftrace_enabled is set; otherwise this tracer is a nop.
2350 See the "ftrace_enabled" section below.
2353 # sysctl kernel.ftrace_enabled=1
2354 # echo function > current_tracer
2355 # echo 1 > tracing_on
2357 # echo 0 > tracing_on
2361 # entries-in-buffer/entries-written: 24799/24799 #P:4
2364 # / _----=> need-resched
2365 # | / _---=> hardirq/softirq
2366 # || / _--=> preempt-depth
2368 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2370 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
2371 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
2372 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
2373 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
2374 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
2375 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
2376 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
2377 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
2381 Note: function tracer uses ring buffers to store the above
2382 entries. The newest data may overwrite the oldest data.
2383 Sometimes using echo to stop the trace is not sufficient because
2384 the tracing could have overwritten the data that you wanted to
2385 record. For this reason, it is sometimes better to disable
2386 tracing directly from a program. This allows you to stop the
2387 tracing at the point that you hit the part that you are
2388 interested in. To disable the tracing directly from a C program,
2389 something like following code snippet can be used::
2393 int main(int argc, char *argv[]) {
2395 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
2397 if (condition_hit()) {
2398 write(trace_fd, "0", 1);
2404 Single thread tracing
2405 ---------------------
2407 By writing into set_ftrace_pid you can trace a
2408 single thread. For example::
2410 # cat set_ftrace_pid
2412 # echo 3111 > set_ftrace_pid
2413 # cat set_ftrace_pid
2415 # echo function > current_tracer
2419 # TASK-PID CPU# TIMESTAMP FUNCTION
2421 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
2422 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
2423 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
2424 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
2425 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
2426 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
2427 # echo > set_ftrace_pid
2431 # TASK-PID CPU# TIMESTAMP FUNCTION
2433 ##### CPU 3 buffer started ####
2434 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
2435 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
2436 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
2437 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
2438 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
2440 If you want to trace a function when executing, you could use
2441 something like this simple program.
2446 #include <sys/types.h>
2447 #include <sys/stat.h>
2453 #define STR(x) _STR(x)
2454 #define MAX_PATH 256
2456 const char *find_tracefs(void)
2458 static char tracefs[MAX_PATH+1];
2459 static int tracefs_found;
2466 if ((fp = fopen("/proc/mounts","r")) == NULL) {
2467 perror("/proc/mounts");
2471 while (fscanf(fp, "%*s %"
2473 "s %99s %*s %*d %*d\n",
2474 tracefs, type) == 2) {
2475 if (strcmp(type, "tracefs") == 0)
2480 if (strcmp(type, "tracefs") != 0) {
2481 fprintf(stderr, "tracefs not mounted");
2485 strcat(tracefs, "/tracing/");
2491 const char *tracing_file(const char *file_name)
2493 static char trace_file[MAX_PATH+1];
2494 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name);
2498 int main (int argc, char **argv)
2508 ffd = open(tracing_file("current_tracer"), O_WRONLY);
2511 write(ffd, "nop", 3);
2513 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
2514 s = sprintf(line, "%d\n", getpid());
2517 write(ffd, "function", 8);
2522 execvp(argv[1], argv+1);
2528 Or this simple script!
2533 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts`
2534 echo 0 > $tracefs/tracing_on
2535 echo $$ > $tracefs/set_ftrace_pid
2536 echo function > $tracefs/current_tracer
2537 echo 1 > $tracefs/tracing_on
2541 function graph tracer
2542 ---------------------------
2544 This tracer is similar to the function tracer except that it
2545 probes a function on its entry and its exit. This is done by
2546 using a dynamically allocated stack of return addresses in each
2547 task_struct. On function entry the tracer overwrites the return
2548 address of each function traced to set a custom probe. Thus the
2549 original return address is stored on the stack of return address
2552 Probing on both ends of a function leads to special features
2555 - measure of a function's time execution
2556 - having a reliable call stack to draw function calls graph
2558 This tracer is useful in several situations:
2560 - you want to find the reason of a strange kernel behavior and
2561 need to see what happens in detail on any areas (or specific
2564 - you are experiencing weird latencies but it's difficult to
2567 - you want to find quickly which path is taken by a specific
2570 - you just want to peek inside a working kernel and want to see
2575 # tracer: function_graph
2577 # CPU DURATION FUNCTION CALLS
2581 0) | do_sys_open() {
2583 0) | kmem_cache_alloc() {
2584 0) 1.382 us | __might_sleep();
2586 0) | strncpy_from_user() {
2587 0) | might_fault() {
2588 0) 1.389 us | __might_sleep();
2593 0) 0.668 us | _spin_lock();
2594 0) 0.570 us | expand_files();
2595 0) 0.586 us | _spin_unlock();
2598 There are several columns that can be dynamically
2599 enabled/disabled. You can use every combination of options you
2600 want, depending on your needs.
2602 - The cpu number on which the function executed is default
2603 enabled. It is sometimes better to only trace one cpu (see
2604 tracing_cpu_mask file) or you might sometimes see unordered
2605 function calls while cpu tracing switch.
2607 - hide: echo nofuncgraph-cpu > trace_options
2608 - show: echo funcgraph-cpu > trace_options
2610 - The duration (function's time of execution) is displayed on
2611 the closing bracket line of a function or on the same line
2612 than the current function in case of a leaf one. It is default
2615 - hide: echo nofuncgraph-duration > trace_options
2616 - show: echo funcgraph-duration > trace_options
2618 - The overhead field precedes the duration field in case of
2619 reached duration thresholds.
2621 - hide: echo nofuncgraph-overhead > trace_options
2622 - show: echo funcgraph-overhead > trace_options
2623 - depends on: funcgraph-duration
2627 3) # 1837.709 us | } /* __switch_to */
2628 3) | finish_task_switch() {
2629 3) 0.313 us | _raw_spin_unlock_irq();
2631 3) # 1889.063 us | } /* __schedule */
2632 3) ! 140.417 us | } /* __schedule */
2633 3) # 2034.948 us | } /* schedule */
2634 3) * 33998.59 us | } /* schedule_preempt_disabled */
2638 1) 0.260 us | msecs_to_jiffies();
2639 1) 0.313 us | __rcu_read_unlock();
2642 1) 0.313 us | rcu_bh_qs();
2643 1) 0.313 us | __local_bh_enable();
2645 1) 0.365 us | idle_cpu();
2646 1) | rcu_irq_exit() {
2647 1) 0.417 us | rcu_eqs_enter_common.isra.47();
2651 1) @ 119760.2 us | }
2657 2) 0.417 us | scheduler_ipi();
2667 + means that the function exceeded 10 usecs.
2668 ! means that the function exceeded 100 usecs.
2669 # means that the function exceeded 1000 usecs.
2670 * means that the function exceeded 10 msecs.
2671 @ means that the function exceeded 100 msecs.
2672 $ means that the function exceeded 1 sec.
2675 - The task/pid field displays the thread cmdline and pid which
2676 executed the function. It is default disabled.
2678 - hide: echo nofuncgraph-proc > trace_options
2679 - show: echo funcgraph-proc > trace_options
2683 # tracer: function_graph
2685 # CPU TASK/PID DURATION FUNCTION CALLS
2687 0) sh-4802 | | d_free() {
2688 0) sh-4802 | | call_rcu() {
2689 0) sh-4802 | | __call_rcu() {
2690 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2691 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2692 0) sh-4802 | 2.899 us | }
2693 0) sh-4802 | 4.040 us | }
2694 0) sh-4802 | 5.151 us | }
2695 0) sh-4802 | + 49.370 us | }
2698 - The absolute time field is an absolute timestamp given by the
2699 system clock since it started. A snapshot of this time is
2700 given on each entry/exit of functions
2702 - hide: echo nofuncgraph-abstime > trace_options
2703 - show: echo funcgraph-abstime > trace_options
2708 # TIME CPU DURATION FUNCTION CALLS
2710 360.774522 | 1) 0.541 us | }
2711 360.774522 | 1) 4.663 us | }
2712 360.774523 | 1) 0.541 us | __wake_up_bit();
2713 360.774524 | 1) 6.796 us | }
2714 360.774524 | 1) 7.952 us | }
2715 360.774525 | 1) 9.063 us | }
2716 360.774525 | 1) 0.615 us | journal_mark_dirty();
2717 360.774527 | 1) 0.578 us | __brelse();
2718 360.774528 | 1) | reiserfs_prepare_for_journal() {
2719 360.774528 | 1) | unlock_buffer() {
2720 360.774529 | 1) | wake_up_bit() {
2721 360.774529 | 1) | bit_waitqueue() {
2722 360.774530 | 1) 0.594 us | __phys_addr();
2725 The function name is always displayed after the closing bracket
2726 for a function if the start of that function is not in the
2729 Display of the function name after the closing bracket may be
2730 enabled for functions whose start is in the trace buffer,
2731 allowing easier searching with grep for function durations.
2732 It is default disabled.
2734 - hide: echo nofuncgraph-tail > trace_options
2735 - show: echo funcgraph-tail > trace_options
2737 Example with nofuncgraph-tail (default)::
2740 0) | kmem_cache_free() {
2741 0) 0.518 us | __phys_addr();
2745 Example with funcgraph-tail::
2748 0) | kmem_cache_free() {
2749 0) 0.518 us | __phys_addr();
2750 0) 1.757 us | } /* kmem_cache_free() */
2751 0) 2.861 us | } /* putname() */
2753 The return value of each traced function can be displayed after
2754 an equal sign "=". When encountering system call failures, it
2755 can be very helpful to quickly locate the function that first
2756 returns an error code.
2758 - hide: echo nofuncgraph-retval > trace_options
2759 - show: echo funcgraph-retval > trace_options
2761 Example with funcgraph-retval::
2763 1) | cgroup_migrate() {
2764 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */
2765 1) | cgroup_migrate_execute() {
2766 1) | cpu_cgroup_can_attach() {
2767 1) | cgroup_taskset_first() {
2768 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */
2769 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */
2770 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */
2771 1) 2.335 us | } /* cpu_cgroup_can_attach = -22 */
2772 1) 4.369 us | } /* cgroup_migrate_execute = -22 */
2773 1) 7.143 us | } /* cgroup_migrate = -22 */
2775 The above example shows that the function cpu_cgroup_can_attach
2776 returned the error code -22 firstly, then we can read the code
2777 of this function to get the root cause.
2779 When the option funcgraph-retval-hex is not set, the return value can
2780 be displayed in a smart way. Specifically, if it is an error code,
2781 it will be printed in signed decimal format, otherwise it will
2782 printed in hexadecimal format.
2784 - smart: echo nofuncgraph-retval-hex > trace_options
2785 - hexadecimal: echo funcgraph-retval-hex > trace_options
2787 Example with funcgraph-retval-hex::
2789 1) | cgroup_migrate() {
2790 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */
2791 1) | cgroup_migrate_execute() {
2792 1) | cpu_cgroup_can_attach() {
2793 1) | cgroup_taskset_first() {
2794 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */
2795 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */
2796 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */
2797 1) 2.335 us | } /* cpu_cgroup_can_attach = 0xffffffea */
2798 1) 4.369 us | } /* cgroup_migrate_execute = 0xffffffea */
2799 1) 7.143 us | } /* cgroup_migrate = 0xffffffea */
2801 At present, there are some limitations when using the funcgraph-retval
2802 option, and these limitations will be eliminated in the future:
2804 - Even if the function return type is void, a return value will still
2805 be printed, and you can just ignore it.
2807 - Even if return values are stored in multiple registers, only the
2808 value contained in the first register will be recorded and printed.
2809 To illustrate, in the x86 architecture, eax and edx are used to store
2810 a 64-bit return value, with the lower 32 bits saved in eax and the
2811 upper 32 bits saved in edx. However, only the value stored in eax
2812 will be recorded and printed.
2814 - In certain procedure call standards, such as arm64's AAPCS64, when a
2815 type is smaller than a GPR, it is the responsibility of the consumer
2816 to perform the narrowing, and the upper bits may contain UNKNOWN values.
2817 Therefore, it is advisable to check the code for such cases. For instance,
2818 when using a u8 in a 64-bit GPR, bits [63:8] may contain arbitrary values,
2819 especially when larger types are truncated, whether explicitly or implicitly.
2820 Here are some specific cases to illustrate this point:
2824 The function narrow_to_u8 is defined as follows::
2826 u8 narrow_to_u8(u64 val)
2828 // implicitly truncated
2832 It may be compiled to::
2835 < ... ftrace instrumentation ... >
2838 If you pass 0x123456789abcdef to this function and want to narrow it,
2839 it may be recorded as 0x123456789abcdef instead of 0xef.
2843 The function error_if_not_4g_aligned is defined as follows::
2845 int error_if_not_4g_aligned(u64 val)
2847 if (val & GENMASK(31, 0))
2853 It could be compiled to::
2855 error_if_not_4g_aligned:
2856 CBNZ w0, .Lnot_aligned
2857 RET // bits [31:0] are zero, bits
2858 // [63:32] are UNKNOWN
2863 When passing 0x2_0000_0000 to it, the return value may be recorded as
2864 0x2_0000_0000 instead of 0.
2866 You can put some comments on specific functions by using
2867 trace_printk() For example, if you want to put a comment inside
2868 the __might_sleep() function, you just have to include
2869 <linux/ftrace.h> and call trace_printk() inside __might_sleep()::
2871 trace_printk("I'm a comment!\n")
2875 1) | __might_sleep() {
2876 1) | /* I'm a comment! */
2880 You might find other useful features for this tracer in the
2881 following "dynamic ftrace" section such as tracing only specific
2887 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2888 virtually no overhead when function tracing is disabled. The way
2889 this works is the mcount function call (placed at the start of
2890 every kernel function, produced by the -pg switch in gcc),
2891 starts of pointing to a simple return. (Enabling FTRACE will
2892 include the -pg switch in the compiling of the kernel.)
2894 At compile time every C file object is run through the
2895 recordmcount program (located in the scripts directory). This
2896 program will parse the ELF headers in the C object to find all
2897 the locations in the .text section that call mcount. Starting
2898 with gcc version 4.6, the -mfentry has been added for x86, which
2899 calls "__fentry__" instead of "mcount". Which is called before
2900 the creation of the stack frame.
2902 Note, not all sections are traced. They may be prevented by either
2903 a notrace, or blocked another way and all inline functions are not
2904 traced. Check the "available_filter_functions" file to see what functions
2907 A section called "__mcount_loc" is created that holds
2908 references to all the mcount/fentry call sites in the .text section.
2909 The recordmcount program re-links this section back into the
2910 original object. The final linking stage of the kernel will add all these
2911 references into a single table.
2913 On boot up, before SMP is initialized, the dynamic ftrace code
2914 scans this table and updates all the locations into nops. It
2915 also records the locations, which are added to the
2916 available_filter_functions list. Modules are processed as they
2917 are loaded and before they are executed. When a module is
2918 unloaded, it also removes its functions from the ftrace function
2919 list. This is automatic in the module unload code, and the
2920 module author does not need to worry about it.
2922 When tracing is enabled, the process of modifying the function
2923 tracepoints is dependent on architecture. The old method is to use
2924 kstop_machine to prevent races with the CPUs executing code being
2925 modified (which can cause the CPU to do undesirable things, especially
2926 if the modified code crosses cache (or page) boundaries), and the nops are
2927 patched back to calls. But this time, they do not call mcount
2928 (which is just a function stub). They now call into the ftrace
2931 The new method of modifying the function tracepoints is to place
2932 a breakpoint at the location to be modified, sync all CPUs, modify
2933 the rest of the instruction not covered by the breakpoint. Sync
2934 all CPUs again, and then remove the breakpoint with the finished
2935 version to the ftrace call site.
2937 Some archs do not even need to monkey around with the synchronization,
2938 and can just slap the new code on top of the old without any
2939 problems with other CPUs executing it at the same time.
2941 One special side-effect to the recording of the functions being
2942 traced is that we can now selectively choose which functions we
2943 wish to trace and which ones we want the mcount calls to remain
2946 Two files are used, one for enabling and one for disabling the
2947 tracing of specified functions. They are:
2955 A list of available functions that you can add to these files is
2958 available_filter_functions
2962 # cat available_filter_functions
2971 If I am only interested in sys_nanosleep and hrtimer_interrupt::
2973 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2974 # echo function > current_tracer
2975 # echo 1 > tracing_on
2977 # echo 0 > tracing_on
2981 # entries-in-buffer/entries-written: 5/5 #P:4
2984 # / _----=> need-resched
2985 # | / _---=> hardirq/softirq
2986 # || / _--=> preempt-depth
2988 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2990 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2991 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2992 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2993 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2994 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2996 To see which functions are being traced, you can cat the file:
2999 # cat set_ftrace_filter
3004 Perhaps this is not enough. The filters also allow glob(7) matching.
3007 will match functions that begin with <match>
3009 will match functions that end with <match>
3011 will match functions that have <match> in it
3012 ``<match1>*<match2>``
3013 will match functions that begin with <match1> and end with <match2>
3016 It is better to use quotes to enclose the wild cards,
3017 otherwise the shell may expand the parameters into names
3018 of files in the local directory.
3022 # echo 'hrtimer_*' > set_ftrace_filter
3028 # entries-in-buffer/entries-written: 897/897 #P:4
3031 # / _----=> need-resched
3032 # | / _---=> hardirq/softirq
3033 # || / _--=> preempt-depth
3035 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3037 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
3038 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
3039 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
3040 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
3041 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
3042 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
3043 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
3044 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
3046 Notice that we lost the sys_nanosleep.
3049 # cat set_ftrace_filter
3054 hrtimer_try_to_cancel
3058 hrtimer_force_reprogram
3059 hrtimer_get_next_event
3063 hrtimer_get_remaining
3065 hrtimer_init_sleeper
3068 This is because the '>' and '>>' act just like they do in bash.
3069 To rewrite the filters, use '>'
3070 To append to the filters, use '>>'
3072 To clear out a filter so that all functions will be recorded
3075 # echo > set_ftrace_filter
3076 # cat set_ftrace_filter
3079 Again, now we want to append.
3083 # echo sys_nanosleep > set_ftrace_filter
3084 # cat set_ftrace_filter
3086 # echo 'hrtimer_*' >> set_ftrace_filter
3087 # cat set_ftrace_filter
3092 hrtimer_try_to_cancel
3096 hrtimer_force_reprogram
3097 hrtimer_get_next_event
3102 hrtimer_get_remaining
3104 hrtimer_init_sleeper
3107 The set_ftrace_notrace prevents those functions from being
3111 # echo '*preempt*' '*lock*' > set_ftrace_notrace
3117 # entries-in-buffer/entries-written: 39608/39608 #P:4
3120 # / _----=> need-resched
3121 # | / _---=> hardirq/softirq
3122 # || / _--=> preempt-depth
3124 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3126 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
3127 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
3128 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
3129 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
3130 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
3131 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
3132 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
3133 bash-1994 [000] .... 4342.324899: setattr_should_drop_suidgid <-do_truncate
3134 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
3135 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
3136 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
3137 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
3139 We can see that there's no more lock or preempt tracing.
3141 Selecting function filters via index
3142 ------------------------------------
3144 Because processing of strings is expensive (the address of the function
3145 needs to be looked up before comparing to the string being passed in),
3146 an index can be used as well to enable functions. This is useful in the
3147 case of setting thousands of specific functions at a time. By passing
3148 in a list of numbers, no string processing will occur. Instead, the function
3149 at the specific location in the internal array (which corresponds to the
3150 functions in the "available_filter_functions" file), is selected.
3154 # echo 1 > set_ftrace_filter
3156 Will select the first function listed in "available_filter_functions"
3160 # head -1 available_filter_functions
3161 trace_initcall_finish_cb
3163 # cat set_ftrace_filter
3164 trace_initcall_finish_cb
3166 # head -50 available_filter_functions | tail -1
3169 # echo 1 50 > set_ftrace_filter
3170 # cat set_ftrace_filter
3171 trace_initcall_finish_cb
3174 Dynamic ftrace with the function graph tracer
3175 ---------------------------------------------
3177 Although what has been explained above concerns both the
3178 function tracer and the function-graph-tracer, there are some
3179 special features only available in the function-graph tracer.
3181 If you want to trace only one function and all of its children,
3182 you just have to echo its name into set_graph_function::
3184 echo __do_fault > set_graph_function
3186 will produce the following "expanded" trace of the __do_fault()
3190 0) | filemap_fault() {
3191 0) | find_lock_page() {
3192 0) 0.804 us | find_get_page();
3193 0) | __might_sleep() {
3197 0) 0.653 us | _spin_lock();
3198 0) 0.578 us | page_add_file_rmap();
3199 0) 0.525 us | native_set_pte_at();
3200 0) 0.585 us | _spin_unlock();
3201 0) | unlock_page() {
3202 0) 0.541 us | page_waitqueue();
3203 0) 0.639 us | __wake_up_bit();
3207 0) | filemap_fault() {
3208 0) | find_lock_page() {
3209 0) 0.698 us | find_get_page();
3210 0) | __might_sleep() {
3214 0) 0.631 us | _spin_lock();
3215 0) 0.571 us | page_add_file_rmap();
3216 0) 0.526 us | native_set_pte_at();
3217 0) 0.586 us | _spin_unlock();
3218 0) | unlock_page() {
3219 0) 0.533 us | page_waitqueue();
3220 0) 0.638 us | __wake_up_bit();
3224 You can also expand several functions at once::
3226 echo sys_open > set_graph_function
3227 echo sys_close >> set_graph_function
3229 Now if you want to go back to trace all functions you can clear
3230 this special filter via::
3232 echo > set_graph_function
3238 Note, the proc sysctl ftrace_enable is a big on/off switch for the
3239 function tracer. By default it is enabled (when function tracing is
3240 enabled in the kernel). If it is disabled, all function tracing is
3241 disabled. This includes not only the function tracers for ftrace, but
3242 also for any other uses (perf, kprobes, stack tracing, profiling, etc). It
3243 cannot be disabled if there is a callback with FTRACE_OPS_FL_PERMANENT set
3246 Please disable this with care.
3248 This can be disable (and enabled) with::
3250 sysctl kernel.ftrace_enabled=0
3251 sysctl kernel.ftrace_enabled=1
3255 echo 0 > /proc/sys/kernel/ftrace_enabled
3256 echo 1 > /proc/sys/kernel/ftrace_enabled
3262 A few commands are supported by the set_ftrace_filter interface.
3263 Trace commands have the following format::
3265 <function>:<command>:<parameter>
3267 The following commands are supported:
3270 This command enables function filtering per module. The
3271 parameter defines the module. For example, if only the write*
3272 functions in the ext3 module are desired, run:
3274 echo 'write*:mod:ext3' > set_ftrace_filter
3276 This command interacts with the filter in the same way as
3277 filtering based on function names. Thus, adding more functions
3278 in a different module is accomplished by appending (>>) to the
3279 filter file. Remove specific module functions by prepending
3282 echo '!writeback*:mod:ext3' >> set_ftrace_filter
3284 Mod command supports module globbing. Disable tracing for all
3285 functions except a specific module::
3287 echo '!*:mod:!ext3' >> set_ftrace_filter
3289 Disable tracing for all modules, but still trace kernel::
3291 echo '!*:mod:*' >> set_ftrace_filter
3293 Enable filter only for kernel::
3295 echo '*write*:mod:!*' >> set_ftrace_filter
3297 Enable filter for module globbing::
3299 echo '*write*:mod:*snd*' >> set_ftrace_filter
3302 These commands turn tracing on and off when the specified
3303 functions are hit. The parameter determines how many times the
3304 tracing system is turned on and off. If unspecified, there is
3305 no limit. For example, to disable tracing when a schedule bug
3306 is hit the first 5 times, run::
3308 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
3310 To always disable tracing when __schedule_bug is hit::
3312 echo '__schedule_bug:traceoff' > set_ftrace_filter
3314 These commands are cumulative whether or not they are appended
3315 to set_ftrace_filter. To remove a command, prepend it by '!'
3316 and drop the parameter::
3318 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
3320 The above removes the traceoff command for __schedule_bug
3321 that have a counter. To remove commands without counters::
3323 echo '!__schedule_bug:traceoff' > set_ftrace_filter
3326 Will cause a snapshot to be triggered when the function is hit.
3329 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
3331 To only snapshot once:
3334 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
3336 To remove the above commands::
3338 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
3339 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
3341 - enable_event/disable_event:
3342 These commands can enable or disable a trace event. Note, because
3343 function tracing callbacks are very sensitive, when these commands
3344 are registered, the trace point is activated, but disabled in
3345 a "soft" mode. That is, the tracepoint will be called, but
3346 just will not be traced. The event tracepoint stays in this mode
3347 as long as there's a command that triggers it.
3350 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
3355 <function>:enable_event:<system>:<event>[:count]
3356 <function>:disable_event:<system>:<event>[:count]
3358 To remove the events commands::
3360 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
3362 echo '!schedule:disable_event:sched:sched_switch' > \
3366 When the function is hit, it will dump the contents of the ftrace
3367 ring buffer to the console. This is useful if you need to debug
3368 something, and want to dump the trace when a certain function
3369 is hit. Perhaps it's a function that is called before a triple
3370 fault happens and does not allow you to get a regular dump.
3373 When the function is hit, it will dump the contents of the ftrace
3374 ring buffer for the current CPU to the console. Unlike the "dump"
3375 command, it only prints out the contents of the ring buffer for the
3376 CPU that executed the function that triggered the dump.
3379 When the function is hit, a stack trace is recorded.
3384 The trace_pipe outputs the same content as the trace file, but
3385 the effect on the tracing is different. Every read from
3386 trace_pipe is consumed. This means that subsequent reads will be
3387 different. The trace is live.
3390 # echo function > current_tracer
3391 # cat trace_pipe > /tmp/trace.out &
3393 # echo 1 > tracing_on
3395 # echo 0 > tracing_on
3399 # entries-in-buffer/entries-written: 0/0 #P:4
3402 # / _----=> need-resched
3403 # | / _---=> hardirq/softirq
3404 # || / _--=> preempt-depth
3406 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3410 # cat /tmp/trace.out
3411 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
3412 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
3413 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
3414 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
3415 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
3416 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
3417 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
3418 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
3419 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
3422 Note, reading the trace_pipe file will block until more input is
3423 added. This is contrary to the trace file. If any process opened
3424 the trace file for reading, it will actually disable tracing and
3425 prevent new entries from being added. The trace_pipe file does
3426 not have this limitation.
3431 Having too much or not enough data can be troublesome in
3432 diagnosing an issue in the kernel. The file buffer_size_kb is
3433 used to modify the size of the internal trace buffers. The
3434 number listed is the number of entries that can be recorded per
3435 CPU. To know the full size, multiply the number of possible CPUs
3436 with the number of entries.
3439 # cat buffer_size_kb
3440 1408 (units kilobytes)
3442 Or simply read buffer_total_size_kb
3445 # cat buffer_total_size_kb
3448 To modify the buffer, simple echo in a number (in 1024 byte segments).
3451 # echo 10000 > buffer_size_kb
3452 # cat buffer_size_kb
3453 10000 (units kilobytes)
3455 It will try to allocate as much as possible. If you allocate too
3456 much, it can cause Out-Of-Memory to trigger.
3459 # echo 1000000000000 > buffer_size_kb
3460 -bash: echo: write error: Cannot allocate memory
3461 # cat buffer_size_kb
3464 The per_cpu buffers can be changed individually as well:
3467 # echo 10000 > per_cpu/cpu0/buffer_size_kb
3468 # echo 100 > per_cpu/cpu1/buffer_size_kb
3470 When the per_cpu buffers are not the same, the buffer_size_kb
3471 at the top level will just show an X
3474 # cat buffer_size_kb
3477 This is where the buffer_total_size_kb is useful:
3480 # cat buffer_total_size_kb
3483 Writing to the top level buffer_size_kb will reset all the buffers
3484 to be the same again.
3488 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
3489 available to all non latency tracers. (Latency tracers which
3490 record max latency, such as "irqsoff" or "wakeup", can't use
3491 this feature, since those are already using the snapshot
3492 mechanism internally.)
3494 Snapshot preserves a current trace buffer at a particular point
3495 in time without stopping tracing. Ftrace swaps the current
3496 buffer with a spare buffer, and tracing continues in the new
3497 current (=previous spare) buffer.
3499 The following tracefs files in "tracing" are related to this
3504 This is used to take a snapshot and to read the output
3505 of the snapshot. Echo 1 into this file to allocate a
3506 spare buffer and to take a snapshot (swap), then read
3507 the snapshot from this file in the same format as
3508 "trace" (described above in the section "The File
3509 System"). Both reads snapshot and tracing are executable
3510 in parallel. When the spare buffer is allocated, echoing
3511 0 frees it, and echoing else (positive) values clear the
3513 More details are shown in the table below.
3515 +--------------+------------+------------+------------+
3516 |status\\input | 0 | 1 | else |
3517 +==============+============+============+============+
3518 |not allocated |(do nothing)| alloc+swap |(do nothing)|
3519 +--------------+------------+------------+------------+
3520 |allocated | free | swap | clear |
3521 +--------------+------------+------------+------------+
3523 Here is an example of using the snapshot feature.
3526 # echo 1 > events/sched/enable
3531 # entries-in-buffer/entries-written: 71/71 #P:8
3534 # / _----=> need-resched
3535 # | / _---=> hardirq/softirq
3536 # || / _--=> preempt-depth
3538 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3540 <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
3541 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
3543 <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
3548 # entries-in-buffer/entries-written: 77/77 #P:8
3551 # / _----=> need-resched
3552 # | / _---=> hardirq/softirq
3553 # || / _--=> preempt-depth
3555 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3557 <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
3558 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
3562 If you try to use this snapshot feature when current tracer is
3563 one of the latency tracers, you will get the following results.
3566 # echo wakeup > current_tracer
3568 bash: echo: write error: Device or resource busy
3570 cat: snapshot: Device or resource busy
3575 In the tracefs tracing directory, there is a directory called "instances".
3576 This directory can have new directories created inside of it using
3577 mkdir, and removing directories with rmdir. The directory created
3578 with mkdir in this directory will already contain files and other
3579 directories after it is created.
3582 # mkdir instances/foo
3584 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
3585 set_event snapshot trace trace_clock trace_marker trace_options
3586 trace_pipe tracing_on
3588 As you can see, the new directory looks similar to the tracing directory
3589 itself. In fact, it is very similar, except that the buffer and
3590 events are agnostic from the main directory, or from any other
3591 instances that are created.
3593 The files in the new directory work just like the files with the
3594 same name in the tracing directory except the buffer that is used
3595 is a separate and new buffer. The files affect that buffer but do not
3596 affect the main buffer with the exception of trace_options. Currently,
3597 the trace_options affect all instances and the top level buffer
3598 the same, but this may change in future releases. That is, options
3599 may become specific to the instance they reside in.
3601 Notice that none of the function tracer files are there, nor is
3602 current_tracer and available_tracers. This is because the buffers
3603 can currently only have events enabled for them.
3606 # mkdir instances/foo
3607 # mkdir instances/bar
3608 # mkdir instances/zoot
3609 # echo 100000 > buffer_size_kb
3610 # echo 1000 > instances/foo/buffer_size_kb
3611 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
3612 # echo function > current_trace
3613 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
3614 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
3615 # echo 1 > instances/foo/events/sched/sched_switch/enable
3616 # echo 1 > instances/bar/events/irq/enable
3617 # echo 1 > instances/zoot/events/syscalls/enable
3619 CPU:2 [LOST 11745 EVENTS]
3620 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
3621 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
3622 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
3623 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
3624 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
3625 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
3626 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
3627 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
3628 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3629 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3630 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
3633 # cat instances/foo/trace_pipe
3634 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3635 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3636 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
3637 <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
3638 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
3639 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3640 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3641 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
3642 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
3643 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
3646 # cat instances/bar/trace_pipe
3647 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
3648 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
3649 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
3650 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
3651 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
3652 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
3653 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
3654 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
3655 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
3656 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
3657 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
3658 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
3661 # cat instances/zoot/trace
3664 # entries-in-buffer/entries-written: 18996/18996 #P:4
3667 # / _----=> need-resched
3668 # | / _---=> hardirq/softirq
3669 # || / _--=> preempt-depth
3671 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3673 bash-1998 [000] d... 140.733501: sys_write -> 0x2
3674 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
3675 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
3676 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
3677 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
3678 bash-1998 [000] d... 140.733510: sys_close(fd: a)
3679 bash-1998 [000] d... 140.733510: sys_close -> 0x0
3680 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
3681 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
3682 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
3683 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
3685 You can see that the trace of the top most trace buffer shows only
3686 the function tracing. The foo instance displays wakeups and task
3689 To remove the instances, simply delete their directories:
3692 # rmdir instances/foo
3693 # rmdir instances/bar
3694 # rmdir instances/zoot
3696 Note, if a process has a trace file open in one of the instance
3697 directories, the rmdir will fail with EBUSY.
3702 Since the kernel has a fixed sized stack, it is important not to
3703 waste it in functions. A kernel developer must be conscious of
3704 what they allocate on the stack. If they add too much, the system
3705 can be in danger of a stack overflow, and corruption will occur,
3706 usually leading to a system panic.
3708 There are some tools that check this, usually with interrupts
3709 periodically checking usage. But if you can perform a check
3710 at every function call that will become very useful. As ftrace provides
3711 a function tracer, it makes it convenient to check the stack size
3712 at every function call. This is enabled via the stack tracer.
3714 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
3715 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
3718 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
3720 You can also enable it from the kernel command line to trace
3721 the stack size of the kernel during boot up, by adding "stacktrace"
3722 to the kernel command line parameter.
3724 After running it for a few minutes, the output looks like:
3727 # cat stack_max_size
3731 Depth Size Location (18 entries)
3733 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
3734 1) 2704 160 find_busiest_group+0x31/0x1f1
3735 2) 2544 256 load_balance+0xd9/0x662
3736 3) 2288 80 idle_balance+0xbb/0x130
3737 4) 2208 128 __schedule+0x26e/0x5b9
3738 5) 2080 16 schedule+0x64/0x66
3739 6) 2064 128 schedule_timeout+0x34/0xe0
3740 7) 1936 112 wait_for_common+0x97/0xf1
3741 8) 1824 16 wait_for_completion+0x1d/0x1f
3742 9) 1808 128 flush_work+0xfe/0x119
3743 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
3744 11) 1664 48 input_available_p+0x1d/0x5c
3745 12) 1616 48 n_tty_poll+0x6d/0x134
3746 13) 1568 64 tty_poll+0x64/0x7f
3747 14) 1504 880 do_select+0x31e/0x511
3748 15) 624 400 core_sys_select+0x177/0x216
3749 16) 224 96 sys_select+0x91/0xb9
3750 17) 128 128 system_call_fastpath+0x16/0x1b
3752 Note, if -mfentry is being used by gcc, functions get traced before
3753 they set up the stack frame. This means that leaf level functions
3754 are not tested by the stack tracer when -mfentry is used.
3756 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
3760 More details can be found in the source code, in the `kernel/trace/*.c` files.