1 ========================
2 ftrace - Function Tracer
3 ========================
5 Copyright 2008 Red Hat Inc.
7 :Author: Steven Rostedt <srostedt@redhat.com>
8 :License: The GNU Free Documentation License, Version 1.2
9 (dual licensed under the GPL v2)
10 :Original Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton,
11 John Kacur, and David Teigland.
13 - Written for: 2.6.28-rc2
15 - Updated for: 4.13 - Copyright 2017 VMware Inc. Steven Rostedt
16 - Converted to rst format - Changbin Du <changbin.du@intel.com>
21 Ftrace is an internal tracer designed to help out developers and
22 designers of systems to find what is going on inside the kernel.
23 It can be used for debugging or analyzing latencies and
24 performance issues that take place outside of user-space.
26 Although ftrace is typically considered the function tracer, it
27 is really a framework of several assorted tracing utilities.
28 There's latency tracing to examine what occurs between interrupts
29 disabled and enabled, as well as for preemption and from a time
30 a task is woken to the task is actually scheduled in.
32 One of the most common uses of ftrace is the event tracing.
33 Throughout the kernel is hundreds of static event points that
34 can be enabled via the tracefs file system to see what is
35 going on in certain parts of the kernel.
37 See events.rst for more information.
40 Implementation Details
41 ----------------------
43 See Documentation/trace/ftrace-design.rst for details for arch porters and such.
49 Ftrace uses the tracefs file system to hold the control files as
50 well as the files to display output.
52 When tracefs is configured into the kernel (which selecting any ftrace
53 option will do) the directory /sys/kernel/tracing will be created. To mount
54 this directory, you can add to your /etc/fstab file::
56 tracefs /sys/kernel/tracing tracefs defaults 0 0
58 Or you can mount it at run time with::
60 mount -t tracefs nodev /sys/kernel/tracing
62 For quicker access to that directory you may want to make a soft link to
65 ln -s /sys/kernel/tracing /tracing
69 Before 4.1, all ftrace tracing control files were within the debugfs
70 file system, which is typically located at /sys/kernel/debug/tracing.
71 For backward compatibility, when mounting the debugfs file system,
72 the tracefs file system will be automatically mounted at:
74 /sys/kernel/debug/tracing
76 All files located in the tracefs file system will be located in that
77 debugfs file system directory as well.
81 Any selected ftrace option will also create the tracefs file system.
82 The rest of the document will assume that you are in the ftrace directory
83 (cd /sys/kernel/tracing) and will only concentrate on the files within that
84 directory and not distract from the content with the extended
85 "/sys/kernel/tracing" path name.
87 That's it! (assuming that you have ftrace configured into your kernel)
89 After mounting tracefs you will have access to the control and output files
90 of ftrace. Here is a list of some of the key files:
93 Note: all time values are in microseconds.
97 This is used to set or display the current tracer
98 that is configured. Changing the current tracer clears
99 the ring buffer content as well as the "snapshot" buffer.
103 This holds the different types of tracers that
104 have been compiled into the kernel. The
105 tracers listed here can be configured by
106 echoing their name into current_tracer.
110 This sets or displays whether writing to the trace
111 ring buffer is enabled. Echo 0 into this file to disable
112 the tracer or 1 to enable it. Note, this only disables
113 writing to the ring buffer, the tracing overhead may
116 The kernel function tracing_off() can be used within the
117 kernel to disable writing to the ring buffer, which will
118 set this file to "0". User space can re-enable tracing by
119 echoing "1" into the file.
121 Note, the function and event trigger "traceoff" will also
122 set this file to zero and stop tracing. Which can also
123 be re-enabled by user space using this file.
127 This file holds the output of the trace in a human
128 readable format (described below). Opening this file for
129 writing with the O_TRUNC flag clears the ring buffer content.
130 Note, this file is not a consumer. If tracing is off
131 (no tracer running, or tracing_on is zero), it will produce
132 the same output each time it is read. When tracing is on,
133 it may produce inconsistent results as it tries to read
134 the entire buffer without consuming it.
138 The output is the same as the "trace" file but this
139 file is meant to be streamed with live tracing.
140 Reads from this file will block until new data is
141 retrieved. Unlike the "trace" file, this file is a
142 consumer. This means reading from this file causes
143 sequential reads to display more current data. Once
144 data is read from this file, it is consumed, and
145 will not be read again with a sequential read. The
146 "trace" file is static, and if the tracer is not
147 adding more data, it will display the same
148 information every time it is read.
152 This file lets the user control the amount of data
153 that is displayed in one of the above output
154 files. Options also exist to modify how a tracer
155 or events work (stack traces, timestamps, etc).
159 This is a directory that has a file for every available
160 trace option (also in trace_options). Options may also be set
161 or cleared by writing a "1" or "0" respectively into the
162 corresponding file with the option name.
166 Some of the tracers record the max latency.
167 For example, the maximum time that interrupts are disabled.
168 The maximum time is saved in this file. The max trace will also be
169 stored, and displayed by "trace". A new max trace will only be
170 recorded if the latency is greater than the value in this file
173 By echoing in a time into this file, no latency will be recorded
174 unless it is greater than the time in this file.
178 Some latency tracers will record a trace whenever the
179 latency is greater than the number in this file.
180 Only active when the file contains a number greater than 0.
185 This sets or displays the number of kilobytes each CPU
186 buffer holds. By default, the trace buffers are the same size
187 for each CPU. The displayed number is the size of the
188 CPU buffer and not total size of all buffers. The
189 trace buffers are allocated in pages (blocks of memory
190 that the kernel uses for allocation, usually 4 KB in size).
191 A few extra pages may be allocated to accommodate buffer management
192 meta-data. If the last page allocated has room for more bytes
193 than requested, the rest of the page will be used,
194 making the actual allocation bigger than requested or shown.
195 ( Note, the size may not be a multiple of the page size
196 due to buffer management meta-data. )
198 Buffer sizes for individual CPUs may vary
199 (see "per_cpu/cpu0/buffer_size_kb" below), and if they do
200 this file will show "X".
202 buffer_total_size_kb:
204 This displays the total combined size of all the trace buffers.
208 If a process is performing tracing, and the ring buffer should be
209 shrunk "freed" when the process is finished, even if it were to be
210 killed by a signal, this file can be used for that purpose. On close
211 of this file, the ring buffer will be resized to its minimum size.
212 Having a process that is tracing also open this file, when the process
213 exits its file descriptor for this file will be closed, and in doing so,
214 the ring buffer will be "freed".
216 It may also stop tracing if disable_on_free option is set.
220 This is a mask that lets the user only trace on specified CPUs.
221 The format is a hex string representing the CPUs.
225 When dynamic ftrace is configured in (see the
226 section below "dynamic ftrace"), the code is dynamically
227 modified (code text rewrite) to disable calling of the
228 function profiler (mcount). This lets tracing be configured
229 in with practically no overhead in performance. This also
230 has a side effect of enabling or disabling specific functions
231 to be traced. Echoing names of functions into this file
232 will limit the trace to only those functions.
233 This influences the tracers "function" and "function_graph"
234 and thus also function profiling (see "function_profile_enabled").
236 The functions listed in "available_filter_functions" are what
237 can be written into this file.
239 This interface also allows for commands to be used. See the
240 "Filter commands" section for more details.
242 As a speed up, since processing strings can be quite expensive
243 and requires a check of all functions registered to tracing, instead
244 an index can be written into this file. A number (starting with "1")
245 written will instead select the same corresponding at the line position
246 of the "available_filter_functions" file.
250 This has an effect opposite to that of
251 set_ftrace_filter. Any function that is added here will not
252 be traced. If a function exists in both set_ftrace_filter
253 and set_ftrace_notrace, the function will _not_ be traced.
257 Have the function tracer only trace the threads whose PID are
260 If the "function-fork" option is set, then when a task whose
261 PID is listed in this file forks, the child's PID will
262 automatically be added to this file, and the child will be
263 traced by the function tracer as well. This option will also
264 cause PIDs of tasks that exit to be removed from the file.
266 set_ftrace_notrace_pid:
268 Have the function tracer ignore threads whose PID are listed in
271 If the "function-fork" option is set, then when a task whose
272 PID is listed in this file forks, the child's PID will
273 automatically be added to this file, and the child will not be
274 traced by the function tracer as well. This option will also
275 cause PIDs of tasks that exit to be removed from the file.
277 If a PID is in both this file and "set_ftrace_pid", then this
278 file takes precedence, and the thread will not be traced.
282 Have the events only trace a task with a PID listed in this file.
283 Note, sched_switch and sched_wake_up will also trace events
286 To have the PIDs of children of tasks with their PID in this file
287 added on fork, enable the "event-fork" option. That option will also
288 cause the PIDs of tasks to be removed from this file when the task
291 set_event_notrace_pid:
293 Have the events not trace a task with a PID listed in this file.
294 Note, sched_switch and sched_wakeup will trace threads not listed
295 in this file, even if a thread's PID is in the file if the
296 sched_switch or sched_wakeup events also trace a thread that should
299 To have the PIDs of children of tasks with their PID in this file
300 added on fork, enable the "event-fork" option. That option will also
301 cause the PIDs of tasks to be removed from this file when the task
306 Functions listed in this file will cause the function graph
307 tracer to only trace these functions and the functions that
308 they call. (See the section "dynamic ftrace" for more details).
309 Note, set_ftrace_filter and set_ftrace_notrace still affects
310 what functions are being traced.
314 Similar to set_graph_function, but will disable function graph
315 tracing when the function is hit until it exits the function.
316 This makes it possible to ignore tracing functions that are called
317 by a specific function.
319 available_filter_functions:
321 This lists the functions that ftrace has processed and can trace.
322 These are the function names that you can pass to
323 "set_ftrace_filter", "set_ftrace_notrace",
324 "set_graph_function", or "set_graph_notrace".
325 (See the section "dynamic ftrace" below for more details.)
327 dyn_ftrace_total_info:
329 This file is for debugging purposes. The number of functions that
330 have been converted to nops and are available to be traced.
334 This file is more for debugging ftrace, but can also be useful
335 in seeing if any function has a callback attached to it.
336 Not only does the trace infrastructure use ftrace function
337 trace utility, but other subsystems might too. This file
338 displays all functions that have a callback attached to them
339 as well as the number of callbacks that have been attached.
340 Note, a callback may also call multiple functions which will
341 not be listed in this count.
343 If the callback registered to be traced by a function with
344 the "save regs" attribute (thus even more overhead), a 'R'
345 will be displayed on the same line as the function that
346 is returning registers.
348 If the callback registered to be traced by a function with
349 the "ip modify" attribute (thus the regs->ip can be changed),
350 an 'I' will be displayed on the same line as the function that
353 If the architecture supports it, it will also show what callback
354 is being directly called by the function. If the count is greater
355 than 1 it most likely will be ftrace_ops_list_func().
357 If the callback of a function jumps to a trampoline that is
358 specific to the callback and which is not the standard trampoline,
359 its address will be printed as well as the function that the
362 function_profile_enabled:
364 When set it will enable all functions with either the function
365 tracer, or if configured, the function graph tracer. It will
366 keep a histogram of the number of functions that were called
367 and if the function graph tracer was configured, it will also keep
368 track of the time spent in those functions. The histogram
369 content can be displayed in the files:
371 trace_stat/function<cpu> ( function0, function1, etc).
375 A directory that holds different tracing stats.
379 Enable dynamic trace points. See kprobetrace.rst.
383 Dynamic trace points stats. See kprobetrace.rst.
387 Used with the function graph tracer. This is the max depth
388 it will trace into a function. Setting this to a value of
389 one will show only the first kernel function that is called
394 This is for tools that read the raw format files. If an event in
395 the ring buffer references a string, only a pointer to the string
396 is recorded into the buffer and not the string itself. This prevents
397 tools from knowing what that string was. This file displays the string
398 and address for the string allowing tools to map the pointers to what
403 Only the pid of the task is recorded in a trace event unless
404 the event specifically saves the task comm as well. Ftrace
405 makes a cache of pid mappings to comms to try to display
406 comms for events. If a pid for a comm is not listed, then
407 "<...>" is displayed in the output.
409 If the option "record-cmd" is set to "0", then comms of tasks
410 will not be saved during recording. By default, it is enabled.
414 By default, 128 comms are saved (see "saved_cmdlines" above). To
415 increase or decrease the amount of comms that are cached, echo
416 the number of comms to cache into this file.
420 If the option "record-tgid" is set, on each scheduling context switch
421 the Task Group ID of a task is saved in a table mapping the PID of
422 the thread to its TGID. By default, the "record-tgid" option is
427 This displays the "snapshot" buffer and also lets the user
428 take a snapshot of the current running trace.
429 See the "Snapshot" section below for more details.
433 When the stack tracer is activated, this will display the
434 maximum stack size it has encountered.
435 See the "Stack Trace" section below.
439 This displays the stack back trace of the largest stack
440 that was encountered when the stack tracer is activated.
441 See the "Stack Trace" section below.
445 This is similar to "set_ftrace_filter" but it limits what
446 functions the stack tracer will check.
450 Whenever an event is recorded into the ring buffer, a
451 "timestamp" is added. This stamp comes from a specified
452 clock. By default, ftrace uses the "local" clock. This
453 clock is very fast and strictly per cpu, but on some
454 systems it may not be monotonic with respect to other
455 CPUs. In other words, the local clocks may not be in sync
456 with local clocks on other CPUs.
458 Usual clocks for tracing::
461 [local] global counter x86-tsc
463 The clock with the square brackets around it is the one in effect.
466 Default clock, but may not be in sync across CPUs
469 This clock is in sync with all CPUs but may
470 be a bit slower than the local clock.
473 This is not a clock at all, but literally an atomic
474 counter. It counts up one by one, but is in sync
475 with all CPUs. This is useful when you need to
476 know exactly the order events occurred with respect to
477 each other on different CPUs.
480 This uses the jiffies counter and the time stamp
481 is relative to the time since boot up.
484 This makes ftrace use the same clock that perf uses.
485 Eventually perf will be able to read ftrace buffers
486 and this will help out in interleaving the data.
489 Architectures may define their own clocks. For
490 example, x86 uses its own TSC cycle clock here.
493 This uses the powerpc timebase register value.
494 This is in sync across CPUs and can also be used
495 to correlate events across hypervisor/guest if
499 This uses the fast monotonic clock (CLOCK_MONOTONIC)
500 which is monotonic and is subject to NTP rate adjustments.
503 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
504 which is monotonic but is not subject to any rate adjustments
505 and ticks at the same rate as the hardware clocksource.
508 This is the boot clock (CLOCK_BOOTTIME) and is based on the
509 fast monotonic clock, but also accounts for time spent in
510 suspend. Since the clock access is designed for use in
511 tracing in the suspend path, some side effects are possible
512 if clock is accessed after the suspend time is accounted before
513 the fast mono clock is updated. In this case, the clock update
514 appears to happen slightly sooner than it normally would have.
515 Also on 32-bit systems, it's possible that the 64-bit boot offset
516 sees a partial update. These effects are rare and post
517 processing should be able to handle them. See comments in the
518 ktime_get_boot_fast_ns() function for more information.
520 To set a clock, simply echo the clock name into this file::
522 # echo global > trace_clock
524 Setting a clock clears the ring buffer content as well as the
529 This is a very useful file for synchronizing user space
530 with events happening in the kernel. Writing strings into
531 this file will be written into the ftrace buffer.
533 It is useful in applications to open this file at the start
534 of the application and just reference the file descriptor
537 void trace_write(const char *fmt, ...)
547 n = vsnprintf(buf, 256, fmt, ap);
550 write(trace_fd, buf, n);
555 trace_fd = open("trace_marker", WR_ONLY);
557 Note: Writing into the trace_marker file can also initiate triggers
558 that are written into /sys/kernel/tracing/events/ftrace/print/trigger
559 See "Event triggers" in Documentation/trace/events.rst and an
560 example in Documentation/trace/histogram.rst (Section 3.)
564 This is similar to trace_marker above, but is meant for binary data
565 to be written to it, where a tool can be used to parse the data
570 Add dynamic tracepoints in programs.
575 Uprobe statistics. See uprobetrace.txt
579 This is a way to make multiple trace buffers where different
580 events can be recorded in different buffers.
581 See "Instances" section below.
585 This is the trace event directory. It holds event tracepoints
586 (also known as static tracepoints) that have been compiled
587 into the kernel. It shows what event tracepoints exist
588 and how they are grouped by system. There are "enable"
589 files at various levels that can enable the tracepoints
590 when a "1" is written to them.
592 See events.rst for more information.
596 By echoing in the event into this file, will enable that event.
598 See events.rst for more information.
602 A list of events that can be enabled in tracing.
604 See events.rst for more information.
608 Certain tracers may change the timestamp mode used when
609 logging trace events into the event buffer. Events with
610 different modes can coexist within a buffer but the mode in
611 effect when an event is logged determines which timestamp mode
612 is used for that event. The default timestamp mode is
615 Usual timestamp modes for tracing:
620 The timestamp mode with the square brackets around it is the
623 delta: Default timestamp mode - timestamp is a delta against
624 a per-buffer timestamp.
626 absolute: The timestamp is a full timestamp, not a delta
627 against some other value. As such it takes up more
628 space and is less efficient.
632 Directory for the Hardware Latency Detector.
633 See "Hardware Latency Detector" section below.
637 This is a directory that contains the trace per_cpu information.
639 per_cpu/cpu0/buffer_size_kb:
641 The ftrace buffer is defined per_cpu. That is, there's a separate
642 buffer for each CPU to allow writes to be done atomically,
643 and free from cache bouncing. These buffers may have different
644 size buffers. This file is similar to the buffer_size_kb
645 file, but it only displays or sets the buffer size for the
646 specific CPU. (here cpu0).
650 This is similar to the "trace" file, but it will only display
651 the data specific for the CPU. If written to, it only clears
652 the specific CPU buffer.
654 per_cpu/cpu0/trace_pipe
656 This is similar to the "trace_pipe" file, and is a consuming
657 read, but it will only display (and consume) the data specific
660 per_cpu/cpu0/trace_pipe_raw
662 For tools that can parse the ftrace ring buffer binary format,
663 the trace_pipe_raw file can be used to extract the data
664 from the ring buffer directly. With the use of the splice()
665 system call, the buffer data can be quickly transferred to
666 a file or to the network where a server is collecting the
669 Like trace_pipe, this is a consuming reader, where multiple
670 reads will always produce different data.
672 per_cpu/cpu0/snapshot:
674 This is similar to the main "snapshot" file, but will only
675 snapshot the current CPU (if supported). It only displays
676 the content of the snapshot for a given CPU, and if
677 written to, only clears this CPU buffer.
679 per_cpu/cpu0/snapshot_raw:
681 Similar to the trace_pipe_raw, but will read the binary format
682 from the snapshot buffer for the given CPU.
686 This displays certain stats about the ring buffer:
689 The number of events that are still in the buffer.
692 The number of lost events due to overwriting when
696 Should always be zero.
697 This gets set if so many events happened within a nested
698 event (ring buffer is re-entrant), that it fills the
699 buffer and starts dropping events.
702 Bytes actually read (not overwritten).
705 The oldest timestamp in the buffer
708 The current timestamp
711 Events lost due to overwrite option being off.
714 The number of events read.
719 Here is the list of current tracers that may be configured.
723 Function call tracer to trace all kernel functions.
727 Similar to the function tracer except that the
728 function tracer probes the functions on their entry
729 whereas the function graph tracer traces on both entry
730 and exit of the functions. It then provides the ability
731 to draw a graph of function calls similar to C code
736 The block tracer. The tracer used by the blktrace user
741 The Hardware Latency tracer is used to detect if the hardware
742 produces any latency. See "Hardware Latency Detector" section
747 Traces the areas that disable interrupts and saves
748 the trace with the longest max latency.
749 See tracing_max_latency. When a new max is recorded,
750 it replaces the old trace. It is best to view this
751 trace with the latency-format option enabled, which
752 happens automatically when the tracer is selected.
756 Similar to irqsoff but traces and records the amount of
757 time for which preemption is disabled.
761 Similar to irqsoff and preemptoff, but traces and
762 records the largest time for which irqs and/or preemption
767 Traces and records the max latency that it takes for
768 the highest priority task to get scheduled after
769 it has been woken up.
770 Traces all tasks as an average developer would expect.
774 Traces and records the max latency that it takes for just
775 RT tasks (as the current "wakeup" does). This is useful
776 for those interested in wake up timings of RT tasks.
780 Traces and records the max latency that it takes for
781 a SCHED_DEADLINE task to be woken (as the "wakeup" and
786 A special tracer that is used to trace binary module.
787 It will trace all the calls that a module makes to the
788 hardware. Everything it writes and reads from the I/O
793 This tracer can be configured when tracing likely/unlikely
794 calls within the kernel. It will trace when a likely and
795 unlikely branch is hit and if it was correct in its prediction
800 This is the "trace nothing" tracer. To remove all
801 tracers from tracing simply echo "nop" into
807 For most ftrace commands, failure modes are obvious and communicated
808 using standard return codes.
810 For other more involved commands, extended error information may be
811 available via the tracing/error_log file. For the commands that
812 support it, reading the tracing/error_log file after an error will
813 display more detailed information about what went wrong, if
814 information is available. The tracing/error_log file is a circular
815 error log displaying a small number (currently, 8) of ftrace errors
816 for the last (8) failed commands.
818 The extended error information and usage takes the form shown in
821 # echo xxx > /sys/kernel/debug/tracing/events/sched/sched_wakeup/trigger
822 echo: write error: Invalid argument
824 # cat /sys/kernel/debug/tracing/error_log
825 [ 5348.887237] location: error: Couldn't yyy: zzz
828 [ 7517.023364] location: error: Bad rrr: sss
832 To clear the error log, echo the empty string into it::
834 # echo > /sys/kernel/debug/tracing/error_log
836 Examples of using the tracer
837 ----------------------------
839 Here are typical examples of using the tracers when controlling
840 them only with the tracefs interface (without using any
841 user-land utilities).
846 Here is an example of the output format of the file "trace"::
850 # entries-in-buffer/entries-written: 140080/250280 #P:4
853 # / _----=> need-resched
854 # | / _---=> hardirq/softirq
855 # || / _--=> preempt-depth
857 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
859 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
860 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
861 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
862 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
863 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
864 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
865 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
866 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
867 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
868 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
871 A header is printed with the tracer name that is represented by
872 the trace. In this case the tracer is "function". Then it shows the
873 number of events in the buffer as well as the total number of entries
874 that were written. The difference is the number of entries that were
875 lost due to the buffer filling up (250280 - 140080 = 110200 events
878 The header explains the content of the events. Task name "bash", the task
879 PID "1977", the CPU that it was running on "000", the latency format
880 (explained below), the timestamp in <secs>.<usecs> format, the
881 function name that was traced "sys_close" and the parent function that
882 called this function "system_call_fastpath". The timestamp is the time
883 at which the function was entered.
888 When the latency-format option is enabled or when one of the latency
889 tracers is set, the trace file gives somewhat more information to see
890 why a latency happened. Here is a typical trace::
894 # irqsoff latency trace v1.1.5 on 3.8.0-test+
895 # --------------------------------------------------------------------
896 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
898 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
900 # => started at: __lock_task_sighand
901 # => ended at: _raw_spin_unlock_irqrestore
905 # / _-----=> irqs-off
906 # | / _----=> need-resched
907 # || / _---=> hardirq/softirq
908 # ||| / _--=> preempt-depth
910 # cmd pid ||||| time | caller
912 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
913 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
914 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
915 ps-6143 2d..1 306us : <stack trace>
916 => trace_hardirqs_on_caller
918 => _raw_spin_unlock_irqrestore
925 => system_call_fastpath
928 This shows that the current tracer is "irqsoff" tracing the time
929 for which interrupts were disabled. It gives the trace version (which
930 never changes) and the version of the kernel upon which this was executed on
931 (3.8). Then it displays the max latency in microseconds (259 us). The number
932 of trace entries displayed and the total number (both are four: #4/4).
933 VP, KP, SP, and HP are always zero and are reserved for later use.
934 #P is the number of online CPUs (#P:4).
936 The task is the process that was running when the latency
937 occurred. (ps pid: 6143).
939 The start and stop (the functions in which the interrupts were
940 disabled and enabled respectively) that caused the latencies:
942 - __lock_task_sighand is where the interrupts were disabled.
943 - _raw_spin_unlock_irqrestore is where they were enabled again.
945 The next lines after the header are the trace itself. The header
946 explains which is which.
948 cmd: The name of the process in the trace.
950 pid: The PID of that process.
952 CPU#: The CPU which the process was running on.
954 irqs-off: 'd' interrupts are disabled. '.' otherwise.
955 .. caution:: If the architecture does not support a way to
956 read the irq flags variable, an 'X' will always
960 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
961 - 'n' only TIF_NEED_RESCHED is set,
962 - 'p' only PREEMPT_NEED_RESCHED is set,
966 - 'Z' - NMI occurred inside a hardirq
967 - 'z' - NMI is running
968 - 'H' - hard irq occurred inside a softirq.
969 - 'h' - hard irq is running
970 - 's' - soft irq is running
971 - '.' - normal context.
973 preempt-depth: The level of preempt_disabled
975 The above is mostly meaningful for kernel developers.
978 When the latency-format option is enabled, the trace file
979 output includes a timestamp relative to the start of the
980 trace. This differs from the output when latency-format
981 is disabled, which includes an absolute timestamp.
984 This is just to help catch your eye a bit better. And
985 needs to be fixed to be only relative to the same CPU.
986 The marks are determined by the difference between this
987 current trace and the next trace.
989 - '$' - greater than 1 second
990 - '@' - greater than 100 millisecond
991 - '*' - greater than 10 millisecond
992 - '#' - greater than 1000 microsecond
993 - '!' - greater than 100 microsecond
994 - '+' - greater than 10 microsecond
995 - ' ' - less than or equal to 10 microsecond.
997 The rest is the same as the 'trace' file.
999 Note, the latency tracers will usually end with a back trace
1000 to easily find where the latency occurred.
1005 The trace_options file (or the options directory) is used to control
1006 what gets printed in the trace output, or manipulate the tracers.
1007 To see what is available, simply cat the file::
1038 To disable one of the options, echo in the option prepended with
1041 echo noprint-parent > trace_options
1043 To enable an option, leave off the "no"::
1045 echo sym-offset > trace_options
1047 Here are the available options:
1050 On function traces, display the calling (parent)
1051 function as well as the function being traced.
1055 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
1058 bash-4000 [01] 1477.606694: simple_strtoul
1062 Display not only the function name, but also the
1063 offset in the function. For example, instead of
1064 seeing just "ktime_get", you will see
1065 "ktime_get+0xb/0x20".
1069 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
1072 This will also display the function address as well
1073 as the function name.
1077 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
1080 This deals with the trace file when the
1081 latency-format option is enabled.
1084 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
1085 (+0.000ms): simple_strtoul (kstrtoul)
1088 This will display raw numbers. This option is best for
1089 use with user applications that can translate the raw
1090 numbers better than having it done in the kernel.
1093 Similar to raw, but the numbers will be in a hexadecimal format.
1096 This will print out the formats in raw binary.
1099 When set, reading trace_pipe will not block when polled.
1102 Can disable trace_printk() from writing into the buffer.
1105 It is sometimes confusing when the CPU buffers are full
1106 and one CPU buffer had a lot of events recently, thus
1107 a shorter time frame, were another CPU may have only had
1108 a few events, which lets it have older events. When
1109 the trace is reported, it shows the oldest events first,
1110 and it may look like only one CPU ran (the one with the
1111 oldest events). When the annotate option is set, it will
1112 display when a new CPU buffer started::
1114 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
1115 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
1116 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
1117 ##### CPU 2 buffer started ####
1118 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
1119 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
1120 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
1123 This option changes the trace. It records a
1124 stacktrace of the current user space thread after
1128 when user stacktrace are enabled, look up which
1129 object the address belongs to, and print a
1130 relative address. This is especially useful when
1131 ASLR is on, otherwise you don't get a chance to
1132 resolve the address to object/file/line after
1133 the app is no longer running
1135 The lookup is performed when you read
1136 trace,trace_pipe. Example::
1138 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
1139 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
1143 When set, trace_printk()s will only show the format
1144 and not their parameters (if trace_bprintk() or
1145 trace_bputs() was used to save the trace_printk()).
1148 Show only the event data. Hides the comm, PID,
1149 timestamp, CPU, and other useful data.
1152 This option changes the trace output. When it is enabled,
1153 the trace displays additional information about the
1154 latency, as described in "Latency trace format".
1157 When set, opening the trace file for read, will pause
1158 writing to the ring buffer (as if tracing_on was set to zero).
1159 This simulates the original behavior of the trace file.
1160 When the file is closed, tracing will be enabled again.
1163 When set, "%p" in the event printk format displays the
1164 hashed pointer value instead of real address.
1165 This will be useful if you want to find out which hashed
1166 value is corresponding to the real value in trace log.
1169 When any event or tracer is enabled, a hook is enabled
1170 in the sched_switch trace point to fill comm cache
1171 with mapped pids and comms. But this may cause some
1172 overhead, and if you only care about pids, and not the
1173 name of the task, disabling this option can lower the
1174 impact of tracing. See "saved_cmdlines".
1177 When any event or tracer is enabled, a hook is enabled
1178 in the sched_switch trace point to fill the cache of
1179 mapped Thread Group IDs (TGID) mapping to pids. See
1183 This controls what happens when the trace buffer is
1184 full. If "1" (default), the oldest events are
1185 discarded and overwritten. If "0", then the newest
1186 events are discarded.
1187 (see per_cpu/cpu0/stats for overrun and dropped)
1190 When the free_buffer is closed, tracing will
1191 stop (tracing_on set to 0).
1194 Shows the interrupt, preempt count, need resched data.
1195 When disabled, the trace looks like::
1199 # entries-in-buffer/entries-written: 144405/9452052 #P:4
1201 # TASK-PID CPU# TIMESTAMP FUNCTION
1203 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
1204 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
1205 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
1209 When set, the trace_marker is writable (only by root).
1210 When disabled, the trace_marker will error with EINVAL
1214 When set, tasks with PIDs listed in set_event_pid will have
1215 the PIDs of their children added to set_event_pid when those
1216 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
1217 their PIDs will be removed from the file.
1219 This affects PIDs listed in set_event_notrace_pid as well.
1222 The latency tracers will enable function tracing
1223 if this option is enabled (default it is). When
1224 it is disabled, the latency tracers do not trace
1225 functions. This keeps the overhead of the tracer down
1226 when performing latency tests.
1229 When set, tasks with PIDs listed in set_ftrace_pid will
1230 have the PIDs of their children added to set_ftrace_pid
1231 when those tasks fork. Also, when tasks with PIDs in
1232 set_ftrace_pid exit, their PIDs will be removed from the
1235 This affects PIDs in set_ftrace_notrace_pid as well.
1238 When set, the latency tracers (irqsoff, wakeup, etc) will
1239 use function graph tracing instead of function tracing.
1242 When set, a stack trace is recorded after any trace event
1246 Enable branch tracing with the tracer. This enables branch
1247 tracer along with the currently set tracer. Enabling this
1248 with the "nop" tracer is the same as just enabling the
1251 .. tip:: Some tracers have their own options. They only appear in this
1252 file when the tracer is active. They always appear in the
1256 Here are the per tracer options:
1258 Options for function tracer:
1261 When set, a stack trace is recorded after every
1262 function that is recorded. NOTE! Limit the functions
1263 that are recorded before enabling this, with
1264 "set_ftrace_filter" otherwise the system performance
1265 will be critically degraded. Remember to disable
1266 this option before clearing the function filter.
1268 Options for function_graph tracer:
1270 Since the function_graph tracer has a slightly different output
1271 it has its own options to control what is displayed.
1274 When set, the "overrun" of the graph stack is
1275 displayed after each function traced. The
1276 overrun, is when the stack depth of the calls
1277 is greater than what is reserved for each task.
1278 Each task has a fixed array of functions to
1279 trace in the call graph. If the depth of the
1280 calls exceeds that, the function is not traced.
1281 The overrun is the number of functions missed
1282 due to exceeding this array.
1285 When set, the CPU number of the CPU where the trace
1286 occurred is displayed.
1289 When set, if the function takes longer than
1290 A certain amount, then a delay marker is
1291 displayed. See "delay" above, under the
1295 Unlike other tracers, the process' command line
1296 is not displayed by default, but instead only
1297 when a task is traced in and out during a context
1298 switch. Enabling this options has the command
1299 of each process displayed at every line.
1302 At the end of each function (the return)
1303 the duration of the amount of time in the
1304 function is displayed in microseconds.
1307 When set, the timestamp is displayed at each line.
1310 When disabled, functions that happen inside an
1311 interrupt will not be traced.
1314 When set, the return event will include the function
1315 that it represents. By default this is off, and
1316 only a closing curly bracket "}" is displayed for
1317 the return of a function.
1320 When running function graph tracer, to include
1321 the time a task schedules out in its function.
1322 When enabled, it will account time the task has been
1323 scheduled out as part of the function call.
1326 When running function profiler with function graph tracer,
1327 to include the time to call nested functions. When this is
1328 not set, the time reported for the function will only
1329 include the time the function itself executed for, not the
1330 time for functions that it called.
1332 Options for blk tracer:
1335 Shows a more minimalistic output.
1341 When interrupts are disabled, the CPU can not react to any other
1342 external event (besides NMIs and SMIs). This prevents the timer
1343 interrupt from triggering or the mouse interrupt from letting
1344 the kernel know of a new mouse event. The result is a latency
1345 with the reaction time.
1347 The irqsoff tracer tracks the time for which interrupts are
1348 disabled. When a new maximum latency is hit, the tracer saves
1349 the trace leading up to that latency point so that every time a
1350 new maximum is reached, the old saved trace is discarded and the
1353 To reset the maximum, echo 0 into tracing_max_latency. Here is
1356 # echo 0 > options/function-trace
1357 # echo irqsoff > current_tracer
1358 # echo 1 > tracing_on
1359 # echo 0 > tracing_max_latency
1362 # echo 0 > tracing_on
1366 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1367 # --------------------------------------------------------------------
1368 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1370 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
1372 # => started at: run_timer_softirq
1373 # => ended at: run_timer_softirq
1377 # / _-----=> irqs-off
1378 # | / _----=> need-resched
1379 # || / _---=> hardirq/softirq
1380 # ||| / _--=> preempt-depth
1382 # cmd pid ||||| time | caller
1384 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
1385 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
1386 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
1387 <idle>-0 0dNs3 25us : <stack trace>
1388 => _raw_spin_unlock_irq
1389 => run_timer_softirq
1394 => smp_apic_timer_interrupt
1395 => apic_timer_interrupt
1400 => x86_64_start_reservations
1401 => x86_64_start_kernel
1403 Here we see that we had a latency of 16 microseconds (which is
1404 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
1405 interrupts. The difference between the 16 and the displayed
1406 timestamp 25us occurred because the clock was incremented
1407 between the time of recording the max latency and the time of
1408 recording the function that had that latency.
1410 Note the above example had function-trace not set. If we set
1411 function-trace, we get a much larger output::
1413 with echo 1 > options/function-trace
1417 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1418 # --------------------------------------------------------------------
1419 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1421 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1423 # => started at: ata_scsi_queuecmd
1424 # => ended at: ata_scsi_queuecmd
1428 # / _-----=> irqs-off
1429 # | / _----=> need-resched
1430 # || / _---=> hardirq/softirq
1431 # ||| / _--=> preempt-depth
1433 # cmd pid ||||| time | caller
1435 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1436 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1437 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1438 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1439 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1440 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1441 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1442 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1443 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1445 bash-2042 3d..1 67us : delay_tsc <-__delay
1446 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1447 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1448 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1449 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1450 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1451 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1452 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1453 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1454 bash-2042 3d..1 120us : <stack trace>
1455 => _raw_spin_unlock_irqrestore
1456 => ata_scsi_queuecmd
1457 => scsi_dispatch_cmd
1459 => __blk_run_queue_uncond
1462 => submit_bio_noacct
1465 => __ext3_get_inode_loc
1474 => user_path_at_empty
1479 => system_call_fastpath
1482 Here we traced a 71 microsecond latency. But we also see all the
1483 functions that were called during that time. Note that by
1484 enabling function tracing, we incur an added overhead. This
1485 overhead may extend the latency times. But nevertheless, this
1486 trace has provided some very helpful debugging information.
1488 If we prefer function graph output instead of function, we can set
1489 display-graph option::
1491 with echo 1 > options/display-graph
1495 # irqsoff latency trace v1.1.5 on 4.20.0-rc6+
1496 # --------------------------------------------------------------------
1497 # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4)
1499 # | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0)
1501 # => started at: free_debug_processing
1502 # => ended at: return_to_handler
1506 # / _----=> need-resched
1507 # | / _---=> hardirq/softirq
1508 # || / _--=> preempt-depth
1510 # REL TIME CPU TASK/PID |||| DURATION FUNCTION CALLS
1511 # | | | | |||| | | | | | |
1512 0 us | 0) bash-1507 | d... | 0.000 us | _raw_spin_lock_irqsave();
1513 0 us | 0) bash-1507 | d..1 | 0.378 us | do_raw_spin_trylock();
1514 1 us | 0) bash-1507 | d..2 | | set_track() {
1515 2 us | 0) bash-1507 | d..2 | | save_stack_trace() {
1516 2 us | 0) bash-1507 | d..2 | | __save_stack_trace() {
1517 3 us | 0) bash-1507 | d..2 | | __unwind_start() {
1518 3 us | 0) bash-1507 | d..2 | | get_stack_info() {
1519 3 us | 0) bash-1507 | d..2 | 0.351 us | in_task_stack();
1520 4 us | 0) bash-1507 | d..2 | 1.107 us | }
1522 3750 us | 0) bash-1507 | d..1 | 0.516 us | do_raw_spin_unlock();
1523 3750 us | 0) bash-1507 | d..1 | 0.000 us | _raw_spin_unlock_irqrestore();
1524 3764 us | 0) bash-1507 | d..1 | 0.000 us | tracer_hardirqs_on();
1525 bash-1507 0d..1 3792us : <stack trace>
1526 => free_debug_processing
1535 => search_binary_handler
1536 => __do_execve_file.isra.32
1539 => entry_SYSCALL_64_after_hwframe
1544 When preemption is disabled, we may be able to receive
1545 interrupts but the task cannot be preempted and a higher
1546 priority task must wait for preemption to be enabled again
1547 before it can preempt a lower priority task.
1549 The preemptoff tracer traces the places that disable preemption.
1550 Like the irqsoff tracer, it records the maximum latency for
1551 which preemption was disabled. The control of preemptoff tracer
1552 is much like the irqsoff tracer.
1555 # echo 0 > options/function-trace
1556 # echo preemptoff > current_tracer
1557 # echo 1 > tracing_on
1558 # echo 0 > tracing_max_latency
1561 # echo 0 > tracing_on
1563 # tracer: preemptoff
1565 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1566 # --------------------------------------------------------------------
1567 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1569 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1571 # => started at: do_IRQ
1572 # => ended at: do_IRQ
1576 # / _-----=> irqs-off
1577 # | / _----=> need-resched
1578 # || / _---=> hardirq/softirq
1579 # ||| / _--=> preempt-depth
1581 # cmd pid ||||| time | caller
1583 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1584 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1585 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1586 sshd-1991 1d..1 52us : <stack trace>
1587 => sub_preempt_count
1593 This has some more changes. Preemption was disabled when an
1594 interrupt came in (notice the 'h'), and was enabled on exit.
1595 But we also see that interrupts have been disabled when entering
1596 the preempt off section and leaving it (the 'd'). We do not know if
1597 interrupts were enabled in the mean time or shortly after this
1601 # tracer: preemptoff
1603 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1604 # --------------------------------------------------------------------
1605 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1607 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1609 # => started at: wake_up_new_task
1610 # => ended at: task_rq_unlock
1614 # / _-----=> irqs-off
1615 # | / _----=> need-resched
1616 # || / _---=> hardirq/softirq
1617 # ||| / _--=> preempt-depth
1619 # cmd pid ||||| time | caller
1621 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1622 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1623 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1624 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1625 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1627 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1628 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1629 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1630 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1631 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1632 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1633 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1634 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1636 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1637 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1638 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1639 bash-1994 1d..2 36us : do_softirq <-irq_exit
1640 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1641 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1642 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1643 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1644 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1645 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1647 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1648 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1649 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1650 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1651 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1652 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1653 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1654 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1655 bash-1994 1.N.1 104us : <stack trace>
1656 => sub_preempt_count
1657 => _raw_spin_unlock_irqrestore
1665 The above is an example of the preemptoff trace with
1666 function-trace set. Here we see that interrupts were not disabled
1667 the entire time. The irq_enter code lets us know that we entered
1668 an interrupt 'h'. Before that, the functions being traced still
1669 show that it is not in an interrupt, but we can see from the
1670 functions themselves that this is not the case.
1675 Knowing the locations that have interrupts disabled or
1676 preemption disabled for the longest times is helpful. But
1677 sometimes we would like to know when either preemption and/or
1678 interrupts are disabled.
1680 Consider the following code::
1682 local_irq_disable();
1683 call_function_with_irqs_off();
1685 call_function_with_irqs_and_preemption_off();
1687 call_function_with_preemption_off();
1690 The irqsoff tracer will record the total length of
1691 call_function_with_irqs_off() and
1692 call_function_with_irqs_and_preemption_off().
1694 The preemptoff tracer will record the total length of
1695 call_function_with_irqs_and_preemption_off() and
1696 call_function_with_preemption_off().
1698 But neither will trace the time that interrupts and/or
1699 preemption is disabled. This total time is the time that we can
1700 not schedule. To record this time, use the preemptirqsoff
1703 Again, using this trace is much like the irqsoff and preemptoff
1707 # echo 0 > options/function-trace
1708 # echo preemptirqsoff > current_tracer
1709 # echo 1 > tracing_on
1710 # echo 0 > tracing_max_latency
1713 # echo 0 > tracing_on
1715 # tracer: preemptirqsoff
1717 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1718 # --------------------------------------------------------------------
1719 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1721 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1723 # => started at: ata_scsi_queuecmd
1724 # => ended at: ata_scsi_queuecmd
1728 # / _-----=> irqs-off
1729 # | / _----=> need-resched
1730 # || / _---=> hardirq/softirq
1731 # ||| / _--=> preempt-depth
1733 # cmd pid ||||| time | caller
1735 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1736 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1737 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1738 ls-2230 3...1 111us : <stack trace>
1739 => sub_preempt_count
1740 => _raw_spin_unlock_irqrestore
1741 => ata_scsi_queuecmd
1742 => scsi_dispatch_cmd
1744 => __blk_run_queue_uncond
1747 => submit_bio_noacct
1752 => htree_dirblock_to_tree
1753 => ext3_htree_fill_tree
1757 => system_call_fastpath
1760 The trace_hardirqs_off_thunk is called from assembly on x86 when
1761 interrupts are disabled in the assembly code. Without the
1762 function tracing, we do not know if interrupts were enabled
1763 within the preemption points. We do see that it started with
1766 Here is a trace with function-trace set::
1768 # tracer: preemptirqsoff
1770 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1771 # --------------------------------------------------------------------
1772 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1774 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1776 # => started at: schedule
1777 # => ended at: mutex_unlock
1781 # / _-----=> irqs-off
1782 # | / _----=> need-resched
1783 # || / _---=> hardirq/softirq
1784 # ||| / _--=> preempt-depth
1786 # cmd pid ||||| time | caller
1788 kworker/-59 3...1 0us : __schedule <-schedule
1789 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1790 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1791 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1792 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1793 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1794 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1795 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1796 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1797 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1798 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1799 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1800 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1801 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1802 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1803 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1804 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1805 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1806 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1807 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1808 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1809 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1810 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1811 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1812 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1813 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1814 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1815 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1816 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1817 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1818 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1819 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1821 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1822 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1823 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1824 ls-2269 3d..3 21us : do_softirq <-irq_exit
1825 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1826 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1827 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1828 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1829 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1830 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1831 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1833 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1834 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1835 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1836 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1837 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1838 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1840 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1841 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1842 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1843 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1844 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1845 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1846 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1847 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1848 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1849 ls-2269 3d... 186us : <stack trace>
1850 => __mutex_unlock_slowpath
1857 => system_call_fastpath
1859 This is an interesting trace. It started with kworker running and
1860 scheduling out and ls taking over. But as soon as ls released the
1861 rq lock and enabled interrupts (but not preemption) an interrupt
1862 triggered. When the interrupt finished, it started running softirqs.
1863 But while the softirq was running, another interrupt triggered.
1864 When an interrupt is running inside a softirq, the annotation is 'H'.
1870 One common case that people are interested in tracing is the
1871 time it takes for a task that is woken to actually wake up.
1872 Now for non Real-Time tasks, this can be arbitrary. But tracing
1873 it none the less can be interesting.
1875 Without function tracing::
1877 # echo 0 > options/function-trace
1878 # echo wakeup > current_tracer
1879 # echo 1 > tracing_on
1880 # echo 0 > tracing_max_latency
1882 # echo 0 > tracing_on
1886 # wakeup latency trace v1.1.5 on 3.8.0-test+
1887 # --------------------------------------------------------------------
1888 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1890 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1894 # / _-----=> irqs-off
1895 # | / _----=> need-resched
1896 # || / _---=> hardirq/softirq
1897 # ||| / _--=> preempt-depth
1899 # cmd pid ||||| time | caller
1901 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1902 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1903 <idle>-0 3d..3 15us : __schedule <-schedule
1904 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1906 The tracer only traces the highest priority task in the system
1907 to avoid tracing the normal circumstances. Here we see that
1908 the kworker with a nice priority of -20 (not very nice), took
1909 just 15 microseconds from the time it woke up, to the time it
1912 Non Real-Time tasks are not that interesting. A more interesting
1913 trace is to concentrate only on Real-Time tasks.
1918 In a Real-Time environment it is very important to know the
1919 wakeup time it takes for the highest priority task that is woken
1920 up to the time that it executes. This is also known as "schedule
1921 latency". I stress the point that this is about RT tasks. It is
1922 also important to know the scheduling latency of non-RT tasks,
1923 but the average schedule latency is better for non-RT tasks.
1924 Tools like LatencyTop are more appropriate for such
1927 Real-Time environments are interested in the worst case latency.
1928 That is the longest latency it takes for something to happen,
1929 and not the average. We can have a very fast scheduler that may
1930 only have a large latency once in a while, but that would not
1931 work well with Real-Time tasks. The wakeup_rt tracer was designed
1932 to record the worst case wakeups of RT tasks. Non-RT tasks are
1933 not recorded because the tracer only records one worst case and
1934 tracing non-RT tasks that are unpredictable will overwrite the
1935 worst case latency of RT tasks (just run the normal wakeup
1936 tracer for a while to see that effect).
1938 Since this tracer only deals with RT tasks, we will run this
1939 slightly differently than we did with the previous tracers.
1940 Instead of performing an 'ls', we will run 'sleep 1' under
1941 'chrt' which changes the priority of the task.
1944 # echo 0 > options/function-trace
1945 # echo wakeup_rt > current_tracer
1946 # echo 1 > tracing_on
1947 # echo 0 > tracing_max_latency
1949 # echo 0 > tracing_on
1955 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1956 # --------------------------------------------------------------------
1957 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1959 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1963 # / _-----=> irqs-off
1964 # | / _----=> need-resched
1965 # || / _---=> hardirq/softirq
1966 # ||| / _--=> preempt-depth
1968 # cmd pid ||||| time | caller
1970 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1971 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1972 <idle>-0 3d..3 5us : __schedule <-schedule
1973 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1976 Running this on an idle system, we see that it only took 5 microseconds
1977 to perform the task switch. Note, since the trace point in the schedule
1978 is before the actual "switch", we stop the tracing when the recorded task
1979 is about to schedule in. This may change if we add a new marker at the
1980 end of the scheduler.
1982 Notice that the recorded task is 'sleep' with the PID of 2389
1983 and it has an rt_prio of 5. This priority is user-space priority
1984 and not the internal kernel priority. The policy is 1 for
1985 SCHED_FIFO and 2 for SCHED_RR.
1987 Note, that the trace data shows the internal priority (99 - rtprio).
1990 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1992 The 0:120:R means idle was running with a nice priority of 0 (120 - 120)
1993 and in the running state 'R'. The sleep task was scheduled in with
1994 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1995 and it too is in the running state.
1997 Doing the same with chrt -r 5 and function-trace set.
2000 echo 1 > options/function-trace
2004 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
2005 # --------------------------------------------------------------------
2006 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
2008 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
2012 # / _-----=> irqs-off
2013 # | / _----=> need-resched
2014 # || / _---=> hardirq/softirq
2015 # ||| / _--=> preempt-depth
2017 # cmd pid ||||| time | caller
2019 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
2020 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
2021 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
2022 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
2023 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
2024 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
2025 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
2026 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
2027 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
2028 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2029 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
2030 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
2031 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
2032 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
2033 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
2034 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
2035 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
2036 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
2037 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
2038 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
2039 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
2040 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
2041 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
2042 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
2043 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
2044 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
2045 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
2046 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
2047 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
2048 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
2049 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
2050 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
2051 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
2052 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
2053 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
2054 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
2055 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
2056 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit
2057 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
2058 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
2059 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
2060 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
2061 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
2062 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
2063 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
2064 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
2065 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
2066 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
2067 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
2068 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
2069 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
2070 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2071 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
2072 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
2073 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
2074 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2075 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
2076 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
2077 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
2078 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
2079 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
2080 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
2081 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
2082 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
2083 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
2084 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
2085 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
2086 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2087 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
2088 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
2089 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
2090 <idle>-0 3.N.. 25us : schedule <-cpu_idle
2091 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
2092 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
2093 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
2094 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
2095 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
2096 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
2097 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
2098 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
2099 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
2100 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
2101 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
2102 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
2103 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
2105 This isn't that big of a trace, even with function tracing enabled,
2106 so I included the entire trace.
2108 The interrupt went off while when the system was idle. Somewhere
2109 before task_woken_rt() was called, the NEED_RESCHED flag was set,
2110 this is indicated by the first occurrence of the 'N' flag.
2112 Latency tracing and events
2113 --------------------------
2114 As function tracing can induce a much larger latency, but without
2115 seeing what happens within the latency it is hard to know what
2116 caused it. There is a middle ground, and that is with enabling
2120 # echo 0 > options/function-trace
2121 # echo wakeup_rt > current_tracer
2122 # echo 1 > events/enable
2123 # echo 1 > tracing_on
2124 # echo 0 > tracing_max_latency
2126 # echo 0 > tracing_on
2130 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
2131 # --------------------------------------------------------------------
2132 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
2134 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
2138 # / _-----=> irqs-off
2139 # | / _----=> need-resched
2140 # || / _---=> hardirq/softirq
2141 # ||| / _--=> preempt-depth
2143 # cmd pid ||||| time | caller
2145 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
2146 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
2147 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
2148 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
2149 <idle>-0 2.N.2 2us : power_end: cpu_id=2
2150 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
2151 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
2152 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
2153 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
2154 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
2155 <idle>-0 2d..3 6us : __schedule <-schedule
2156 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
2159 Hardware Latency Detector
2160 -------------------------
2162 The hardware latency detector is executed by enabling the "hwlat" tracer.
2164 NOTE, this tracer will affect the performance of the system as it will
2165 periodically make a CPU constantly busy with interrupts disabled.
2168 # echo hwlat > current_tracer
2173 # entries-in-buffer/entries-written: 13/13 #P:8
2176 # / _----=> need-resched
2177 # | / _---=> hardirq/softirq
2178 # || / _--=> preempt-depth
2180 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2182 <...>-1729 [001] d... 678.473449: #1 inner/outer(us): 11/12 ts:1581527483.343962693 count:6
2183 <...>-1729 [004] d... 689.556542: #2 inner/outer(us): 16/9 ts:1581527494.889008092 count:1
2184 <...>-1729 [005] d... 714.756290: #3 inner/outer(us): 16/16 ts:1581527519.678961629 count:5
2185 <...>-1729 [001] d... 718.788247: #4 inner/outer(us): 9/17 ts:1581527523.889012713 count:1
2186 <...>-1729 [002] d... 719.796341: #5 inner/outer(us): 13/9 ts:1581527524.912872606 count:1
2187 <...>-1729 [006] d... 844.787091: #6 inner/outer(us): 9/12 ts:1581527649.889048502 count:2
2188 <...>-1729 [003] d... 849.827033: #7 inner/outer(us): 18/9 ts:1581527654.889013793 count:1
2189 <...>-1729 [007] d... 853.859002: #8 inner/outer(us): 9/12 ts:1581527658.889065736 count:1
2190 <...>-1729 [001] d... 855.874978: #9 inner/outer(us): 9/11 ts:1581527660.861991877 count:1
2191 <...>-1729 [001] d... 863.938932: #10 inner/outer(us): 9/11 ts:1581527668.970010500 count:1 nmi-total:7 nmi-count:1
2192 <...>-1729 [007] d... 878.050780: #11 inner/outer(us): 9/12 ts:1581527683.385002600 count:1 nmi-total:5 nmi-count:1
2193 <...>-1729 [007] d... 886.114702: #12 inner/outer(us): 9/12 ts:1581527691.385001600 count:1
2196 The above output is somewhat the same in the header. All events will have
2197 interrupts disabled 'd'. Under the FUNCTION title there is:
2200 This is the count of events recorded that were greater than the
2201 tracing_threshold (See below).
2203 inner/outer(us): 11/11
2205 This shows two numbers as "inner latency" and "outer latency". The test
2206 runs in a loop checking a timestamp twice. The latency detected within
2207 the two timestamps is the "inner latency" and the latency detected
2208 after the previous timestamp and the next timestamp in the loop is
2209 the "outer latency".
2211 ts:1581527483.343962693
2213 The absolute timestamp that the first latency was recorded in the window.
2217 The number of times a latency was detected during the window.
2219 nmi-total:7 nmi-count:1
2221 On architectures that support it, if an NMI comes in during the
2222 test, the time spent in NMI is reported in "nmi-total" (in
2225 All architectures that have NMIs will show the "nmi-count" if an
2226 NMI comes in during the test.
2231 This gets automatically set to "10" to represent 10
2232 microseconds. This is the threshold of latency that
2233 needs to be detected before the trace will be recorded.
2235 Note, when hwlat tracer is finished (another tracer is
2236 written into "current_tracer"), the original value for
2237 tracing_threshold is placed back into this file.
2239 hwlat_detector/width
2240 The length of time the test runs with interrupts disabled.
2242 hwlat_detector/window
2243 The length of time of the window which the test
2244 runs. That is, the test will run for "width"
2245 microseconds per "window" microseconds
2248 When the test is started. A kernel thread is created that
2249 runs the test. This thread will alternate between CPUs
2250 listed in the tracing_cpumask between each period
2251 (one "window"). To limit the test to specific CPUs
2252 set the mask in this file to only the CPUs that the test
2258 This tracer is the function tracer. Enabling the function tracer
2259 can be done from the debug file system. Make sure the
2260 ftrace_enabled is set; otherwise this tracer is a nop.
2261 See the "ftrace_enabled" section below.
2264 # sysctl kernel.ftrace_enabled=1
2265 # echo function > current_tracer
2266 # echo 1 > tracing_on
2268 # echo 0 > tracing_on
2272 # entries-in-buffer/entries-written: 24799/24799 #P:4
2275 # / _----=> need-resched
2276 # | / _---=> hardirq/softirq
2277 # || / _--=> preempt-depth
2279 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2281 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
2282 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
2283 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
2284 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
2285 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
2286 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
2287 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
2288 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
2292 Note: function tracer uses ring buffers to store the above
2293 entries. The newest data may overwrite the oldest data.
2294 Sometimes using echo to stop the trace is not sufficient because
2295 the tracing could have overwritten the data that you wanted to
2296 record. For this reason, it is sometimes better to disable
2297 tracing directly from a program. This allows you to stop the
2298 tracing at the point that you hit the part that you are
2299 interested in. To disable the tracing directly from a C program,
2300 something like following code snippet can be used::
2304 int main(int argc, char *argv[]) {
2306 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
2308 if (condition_hit()) {
2309 write(trace_fd, "0", 1);
2315 Single thread tracing
2316 ---------------------
2318 By writing into set_ftrace_pid you can trace a
2319 single thread. For example::
2321 # cat set_ftrace_pid
2323 # echo 3111 > set_ftrace_pid
2324 # cat set_ftrace_pid
2326 # echo function > current_tracer
2330 # TASK-PID CPU# TIMESTAMP FUNCTION
2332 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
2333 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
2334 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
2335 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
2336 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
2337 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
2338 # echo > set_ftrace_pid
2342 # TASK-PID CPU# TIMESTAMP FUNCTION
2344 ##### CPU 3 buffer started ####
2345 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
2346 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
2347 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
2348 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
2349 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
2351 If you want to trace a function when executing, you could use
2352 something like this simple program.
2357 #include <sys/types.h>
2358 #include <sys/stat.h>
2364 #define STR(x) _STR(x)
2365 #define MAX_PATH 256
2367 const char *find_tracefs(void)
2369 static char tracefs[MAX_PATH+1];
2370 static int tracefs_found;
2377 if ((fp = fopen("/proc/mounts","r")) == NULL) {
2378 perror("/proc/mounts");
2382 while (fscanf(fp, "%*s %"
2384 "s %99s %*s %*d %*d\n",
2385 tracefs, type) == 2) {
2386 if (strcmp(type, "tracefs") == 0)
2391 if (strcmp(type, "tracefs") != 0) {
2392 fprintf(stderr, "tracefs not mounted");
2396 strcat(tracefs, "/tracing/");
2402 const char *tracing_file(const char *file_name)
2404 static char trace_file[MAX_PATH+1];
2405 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name);
2409 int main (int argc, char **argv)
2419 ffd = open(tracing_file("current_tracer"), O_WRONLY);
2422 write(ffd, "nop", 3);
2424 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
2425 s = sprintf(line, "%d\n", getpid());
2428 write(ffd, "function", 8);
2433 execvp(argv[1], argv+1);
2439 Or this simple script!
2444 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts`
2445 echo 0 > $tracefs/tracing_on
2446 echo $$ > $tracefs/set_ftrace_pid
2447 echo function > $tracefs/current_tracer
2448 echo 1 > $tracefs/tracing_on
2452 function graph tracer
2453 ---------------------------
2455 This tracer is similar to the function tracer except that it
2456 probes a function on its entry and its exit. This is done by
2457 using a dynamically allocated stack of return addresses in each
2458 task_struct. On function entry the tracer overwrites the return
2459 address of each function traced to set a custom probe. Thus the
2460 original return address is stored on the stack of return address
2463 Probing on both ends of a function leads to special features
2466 - measure of a function's time execution
2467 - having a reliable call stack to draw function calls graph
2469 This tracer is useful in several situations:
2471 - you want to find the reason of a strange kernel behavior and
2472 need to see what happens in detail on any areas (or specific
2475 - you are experiencing weird latencies but it's difficult to
2478 - you want to find quickly which path is taken by a specific
2481 - you just want to peek inside a working kernel and want to see
2486 # tracer: function_graph
2488 # CPU DURATION FUNCTION CALLS
2492 0) | do_sys_open() {
2494 0) | kmem_cache_alloc() {
2495 0) 1.382 us | __might_sleep();
2497 0) | strncpy_from_user() {
2498 0) | might_fault() {
2499 0) 1.389 us | __might_sleep();
2504 0) 0.668 us | _spin_lock();
2505 0) 0.570 us | expand_files();
2506 0) 0.586 us | _spin_unlock();
2509 There are several columns that can be dynamically
2510 enabled/disabled. You can use every combination of options you
2511 want, depending on your needs.
2513 - The cpu number on which the function executed is default
2514 enabled. It is sometimes better to only trace one cpu (see
2515 tracing_cpu_mask file) or you might sometimes see unordered
2516 function calls while cpu tracing switch.
2518 - hide: echo nofuncgraph-cpu > trace_options
2519 - show: echo funcgraph-cpu > trace_options
2521 - The duration (function's time of execution) is displayed on
2522 the closing bracket line of a function or on the same line
2523 than the current function in case of a leaf one. It is default
2526 - hide: echo nofuncgraph-duration > trace_options
2527 - show: echo funcgraph-duration > trace_options
2529 - The overhead field precedes the duration field in case of
2530 reached duration thresholds.
2532 - hide: echo nofuncgraph-overhead > trace_options
2533 - show: echo funcgraph-overhead > trace_options
2534 - depends on: funcgraph-duration
2538 3) # 1837.709 us | } /* __switch_to */
2539 3) | finish_task_switch() {
2540 3) 0.313 us | _raw_spin_unlock_irq();
2542 3) # 1889.063 us | } /* __schedule */
2543 3) ! 140.417 us | } /* __schedule */
2544 3) # 2034.948 us | } /* schedule */
2545 3) * 33998.59 us | } /* schedule_preempt_disabled */
2549 1) 0.260 us | msecs_to_jiffies();
2550 1) 0.313 us | __rcu_read_unlock();
2553 1) 0.313 us | rcu_bh_qs();
2554 1) 0.313 us | __local_bh_enable();
2556 1) 0.365 us | idle_cpu();
2557 1) | rcu_irq_exit() {
2558 1) 0.417 us | rcu_eqs_enter_common.isra.47();
2562 1) @ 119760.2 us | }
2568 2) 0.417 us | scheduler_ipi();
2578 + means that the function exceeded 10 usecs.
2579 ! means that the function exceeded 100 usecs.
2580 # means that the function exceeded 1000 usecs.
2581 * means that the function exceeded 10 msecs.
2582 @ means that the function exceeded 100 msecs.
2583 $ means that the function exceeded 1 sec.
2586 - The task/pid field displays the thread cmdline and pid which
2587 executed the function. It is default disabled.
2589 - hide: echo nofuncgraph-proc > trace_options
2590 - show: echo funcgraph-proc > trace_options
2594 # tracer: function_graph
2596 # CPU TASK/PID DURATION FUNCTION CALLS
2598 0) sh-4802 | | d_free() {
2599 0) sh-4802 | | call_rcu() {
2600 0) sh-4802 | | __call_rcu() {
2601 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2602 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2603 0) sh-4802 | 2.899 us | }
2604 0) sh-4802 | 4.040 us | }
2605 0) sh-4802 | 5.151 us | }
2606 0) sh-4802 | + 49.370 us | }
2609 - The absolute time field is an absolute timestamp given by the
2610 system clock since it started. A snapshot of this time is
2611 given on each entry/exit of functions
2613 - hide: echo nofuncgraph-abstime > trace_options
2614 - show: echo funcgraph-abstime > trace_options
2619 # TIME CPU DURATION FUNCTION CALLS
2621 360.774522 | 1) 0.541 us | }
2622 360.774522 | 1) 4.663 us | }
2623 360.774523 | 1) 0.541 us | __wake_up_bit();
2624 360.774524 | 1) 6.796 us | }
2625 360.774524 | 1) 7.952 us | }
2626 360.774525 | 1) 9.063 us | }
2627 360.774525 | 1) 0.615 us | journal_mark_dirty();
2628 360.774527 | 1) 0.578 us | __brelse();
2629 360.774528 | 1) | reiserfs_prepare_for_journal() {
2630 360.774528 | 1) | unlock_buffer() {
2631 360.774529 | 1) | wake_up_bit() {
2632 360.774529 | 1) | bit_waitqueue() {
2633 360.774530 | 1) 0.594 us | __phys_addr();
2636 The function name is always displayed after the closing bracket
2637 for a function if the start of that function is not in the
2640 Display of the function name after the closing bracket may be
2641 enabled for functions whose start is in the trace buffer,
2642 allowing easier searching with grep for function durations.
2643 It is default disabled.
2645 - hide: echo nofuncgraph-tail > trace_options
2646 - show: echo funcgraph-tail > trace_options
2648 Example with nofuncgraph-tail (default)::
2651 0) | kmem_cache_free() {
2652 0) 0.518 us | __phys_addr();
2656 Example with funcgraph-tail::
2659 0) | kmem_cache_free() {
2660 0) 0.518 us | __phys_addr();
2661 0) 1.757 us | } /* kmem_cache_free() */
2662 0) 2.861 us | } /* putname() */
2664 You can put some comments on specific functions by using
2665 trace_printk() For example, if you want to put a comment inside
2666 the __might_sleep() function, you just have to include
2667 <linux/ftrace.h> and call trace_printk() inside __might_sleep()::
2669 trace_printk("I'm a comment!\n")
2673 1) | __might_sleep() {
2674 1) | /* I'm a comment! */
2678 You might find other useful features for this tracer in the
2679 following "dynamic ftrace" section such as tracing only specific
2685 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2686 virtually no overhead when function tracing is disabled. The way
2687 this works is the mcount function call (placed at the start of
2688 every kernel function, produced by the -pg switch in gcc),
2689 starts of pointing to a simple return. (Enabling FTRACE will
2690 include the -pg switch in the compiling of the kernel.)
2692 At compile time every C file object is run through the
2693 recordmcount program (located in the scripts directory). This
2694 program will parse the ELF headers in the C object to find all
2695 the locations in the .text section that call mcount. Starting
2696 with gcc version 4.6, the -mfentry has been added for x86, which
2697 calls "__fentry__" instead of "mcount". Which is called before
2698 the creation of the stack frame.
2700 Note, not all sections are traced. They may be prevented by either
2701 a notrace, or blocked another way and all inline functions are not
2702 traced. Check the "available_filter_functions" file to see what functions
2705 A section called "__mcount_loc" is created that holds
2706 references to all the mcount/fentry call sites in the .text section.
2707 The recordmcount program re-links this section back into the
2708 original object. The final linking stage of the kernel will add all these
2709 references into a single table.
2711 On boot up, before SMP is initialized, the dynamic ftrace code
2712 scans this table and updates all the locations into nops. It
2713 also records the locations, which are added to the
2714 available_filter_functions list. Modules are processed as they
2715 are loaded and before they are executed. When a module is
2716 unloaded, it also removes its functions from the ftrace function
2717 list. This is automatic in the module unload code, and the
2718 module author does not need to worry about it.
2720 When tracing is enabled, the process of modifying the function
2721 tracepoints is dependent on architecture. The old method is to use
2722 kstop_machine to prevent races with the CPUs executing code being
2723 modified (which can cause the CPU to do undesirable things, especially
2724 if the modified code crosses cache (or page) boundaries), and the nops are
2725 patched back to calls. But this time, they do not call mcount
2726 (which is just a function stub). They now call into the ftrace
2729 The new method of modifying the function tracepoints is to place
2730 a breakpoint at the location to be modified, sync all CPUs, modify
2731 the rest of the instruction not covered by the breakpoint. Sync
2732 all CPUs again, and then remove the breakpoint with the finished
2733 version to the ftrace call site.
2735 Some archs do not even need to monkey around with the synchronization,
2736 and can just slap the new code on top of the old without any
2737 problems with other CPUs executing it at the same time.
2739 One special side-effect to the recording of the functions being
2740 traced is that we can now selectively choose which functions we
2741 wish to trace and which ones we want the mcount calls to remain
2744 Two files are used, one for enabling and one for disabling the
2745 tracing of specified functions. They are:
2753 A list of available functions that you can add to these files is
2756 available_filter_functions
2760 # cat available_filter_functions
2769 If I am only interested in sys_nanosleep and hrtimer_interrupt::
2771 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2772 # echo function > current_tracer
2773 # echo 1 > tracing_on
2775 # echo 0 > tracing_on
2779 # entries-in-buffer/entries-written: 5/5 #P:4
2782 # / _----=> need-resched
2783 # | / _---=> hardirq/softirq
2784 # || / _--=> preempt-depth
2786 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2788 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2789 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2790 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2791 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2792 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2794 To see which functions are being traced, you can cat the file:
2797 # cat set_ftrace_filter
2802 Perhaps this is not enough. The filters also allow glob(7) matching.
2805 will match functions that begin with <match>
2807 will match functions that end with <match>
2809 will match functions that have <match> in it
2810 ``<match1>*<match2>``
2811 will match functions that begin with <match1> and end with <match2>
2814 It is better to use quotes to enclose the wild cards,
2815 otherwise the shell may expand the parameters into names
2816 of files in the local directory.
2820 # echo 'hrtimer_*' > set_ftrace_filter
2826 # entries-in-buffer/entries-written: 897/897 #P:4
2829 # / _----=> need-resched
2830 # | / _---=> hardirq/softirq
2831 # || / _--=> preempt-depth
2833 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2835 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2836 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2837 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2838 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2839 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2840 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2841 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2842 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2844 Notice that we lost the sys_nanosleep.
2847 # cat set_ftrace_filter
2852 hrtimer_try_to_cancel
2856 hrtimer_force_reprogram
2857 hrtimer_get_next_event
2861 hrtimer_get_remaining
2863 hrtimer_init_sleeper
2866 This is because the '>' and '>>' act just like they do in bash.
2867 To rewrite the filters, use '>'
2868 To append to the filters, use '>>'
2870 To clear out a filter so that all functions will be recorded
2873 # echo > set_ftrace_filter
2874 # cat set_ftrace_filter
2877 Again, now we want to append.
2881 # echo sys_nanosleep > set_ftrace_filter
2882 # cat set_ftrace_filter
2884 # echo 'hrtimer_*' >> set_ftrace_filter
2885 # cat set_ftrace_filter
2890 hrtimer_try_to_cancel
2894 hrtimer_force_reprogram
2895 hrtimer_get_next_event
2900 hrtimer_get_remaining
2902 hrtimer_init_sleeper
2905 The set_ftrace_notrace prevents those functions from being
2909 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2915 # entries-in-buffer/entries-written: 39608/39608 #P:4
2918 # / _----=> need-resched
2919 # | / _---=> hardirq/softirq
2920 # || / _--=> preempt-depth
2922 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2924 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2925 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2926 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2927 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2928 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2929 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2930 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2931 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2932 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2933 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2934 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2935 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2937 We can see that there's no more lock or preempt tracing.
2939 Selecting function filters via index
2940 ------------------------------------
2942 Because processing of strings is expensive (the address of the function
2943 needs to be looked up before comparing to the string being passed in),
2944 an index can be used as well to enable functions. This is useful in the
2945 case of setting thousands of specific functions at a time. By passing
2946 in a list of numbers, no string processing will occur. Instead, the function
2947 at the specific location in the internal array (which corresponds to the
2948 functions in the "available_filter_functions" file), is selected.
2952 # echo 1 > set_ftrace_filter
2954 Will select the first function listed in "available_filter_functions"
2958 # head -1 available_filter_functions
2959 trace_initcall_finish_cb
2961 # cat set_ftrace_filter
2962 trace_initcall_finish_cb
2964 # head -50 available_filter_functions | tail -1
2967 # echo 1 50 > set_ftrace_filter
2968 # cat set_ftrace_filter
2969 trace_initcall_finish_cb
2972 Dynamic ftrace with the function graph tracer
2973 ---------------------------------------------
2975 Although what has been explained above concerns both the
2976 function tracer and the function-graph-tracer, there are some
2977 special features only available in the function-graph tracer.
2979 If you want to trace only one function and all of its children,
2980 you just have to echo its name into set_graph_function::
2982 echo __do_fault > set_graph_function
2984 will produce the following "expanded" trace of the __do_fault()
2988 0) | filemap_fault() {
2989 0) | find_lock_page() {
2990 0) 0.804 us | find_get_page();
2991 0) | __might_sleep() {
2995 0) 0.653 us | _spin_lock();
2996 0) 0.578 us | page_add_file_rmap();
2997 0) 0.525 us | native_set_pte_at();
2998 0) 0.585 us | _spin_unlock();
2999 0) | unlock_page() {
3000 0) 0.541 us | page_waitqueue();
3001 0) 0.639 us | __wake_up_bit();
3005 0) | filemap_fault() {
3006 0) | find_lock_page() {
3007 0) 0.698 us | find_get_page();
3008 0) | __might_sleep() {
3012 0) 0.631 us | _spin_lock();
3013 0) 0.571 us | page_add_file_rmap();
3014 0) 0.526 us | native_set_pte_at();
3015 0) 0.586 us | _spin_unlock();
3016 0) | unlock_page() {
3017 0) 0.533 us | page_waitqueue();
3018 0) 0.638 us | __wake_up_bit();
3022 You can also expand several functions at once::
3024 echo sys_open > set_graph_function
3025 echo sys_close >> set_graph_function
3027 Now if you want to go back to trace all functions you can clear
3028 this special filter via::
3030 echo > set_graph_function
3036 Note, the proc sysctl ftrace_enable is a big on/off switch for the
3037 function tracer. By default it is enabled (when function tracing is
3038 enabled in the kernel). If it is disabled, all function tracing is
3039 disabled. This includes not only the function tracers for ftrace, but
3040 also for any other uses (perf, kprobes, stack tracing, profiling, etc). It
3041 cannot be disabled if there is a callback with FTRACE_OPS_FL_PERMANENT set
3044 Please disable this with care.
3046 This can be disable (and enabled) with::
3048 sysctl kernel.ftrace_enabled=0
3049 sysctl kernel.ftrace_enabled=1
3053 echo 0 > /proc/sys/kernel/ftrace_enabled
3054 echo 1 > /proc/sys/kernel/ftrace_enabled
3060 A few commands are supported by the set_ftrace_filter interface.
3061 Trace commands have the following format::
3063 <function>:<command>:<parameter>
3065 The following commands are supported:
3068 This command enables function filtering per module. The
3069 parameter defines the module. For example, if only the write*
3070 functions in the ext3 module are desired, run:
3072 echo 'write*:mod:ext3' > set_ftrace_filter
3074 This command interacts with the filter in the same way as
3075 filtering based on function names. Thus, adding more functions
3076 in a different module is accomplished by appending (>>) to the
3077 filter file. Remove specific module functions by prepending
3080 echo '!writeback*:mod:ext3' >> set_ftrace_filter
3082 Mod command supports module globbing. Disable tracing for all
3083 functions except a specific module::
3085 echo '!*:mod:!ext3' >> set_ftrace_filter
3087 Disable tracing for all modules, but still trace kernel::
3089 echo '!*:mod:*' >> set_ftrace_filter
3091 Enable filter only for kernel::
3093 echo '*write*:mod:!*' >> set_ftrace_filter
3095 Enable filter for module globbing::
3097 echo '*write*:mod:*snd*' >> set_ftrace_filter
3100 These commands turn tracing on and off when the specified
3101 functions are hit. The parameter determines how many times the
3102 tracing system is turned on and off. If unspecified, there is
3103 no limit. For example, to disable tracing when a schedule bug
3104 is hit the first 5 times, run::
3106 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
3108 To always disable tracing when __schedule_bug is hit::
3110 echo '__schedule_bug:traceoff' > set_ftrace_filter
3112 These commands are cumulative whether or not they are appended
3113 to set_ftrace_filter. To remove a command, prepend it by '!'
3114 and drop the parameter::
3116 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
3118 The above removes the traceoff command for __schedule_bug
3119 that have a counter. To remove commands without counters::
3121 echo '!__schedule_bug:traceoff' > set_ftrace_filter
3124 Will cause a snapshot to be triggered when the function is hit.
3127 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
3129 To only snapshot once:
3132 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
3134 To remove the above commands::
3136 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
3137 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
3139 - enable_event/disable_event:
3140 These commands can enable or disable a trace event. Note, because
3141 function tracing callbacks are very sensitive, when these commands
3142 are registered, the trace point is activated, but disabled in
3143 a "soft" mode. That is, the tracepoint will be called, but
3144 just will not be traced. The event tracepoint stays in this mode
3145 as long as there's a command that triggers it.
3148 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
3153 <function>:enable_event:<system>:<event>[:count]
3154 <function>:disable_event:<system>:<event>[:count]
3156 To remove the events commands::
3158 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
3160 echo '!schedule:disable_event:sched:sched_switch' > \
3164 When the function is hit, it will dump the contents of the ftrace
3165 ring buffer to the console. This is useful if you need to debug
3166 something, and want to dump the trace when a certain function
3167 is hit. Perhaps it's a function that is called before a triple
3168 fault happens and does not allow you to get a regular dump.
3171 When the function is hit, it will dump the contents of the ftrace
3172 ring buffer for the current CPU to the console. Unlike the "dump"
3173 command, it only prints out the contents of the ring buffer for the
3174 CPU that executed the function that triggered the dump.
3177 When the function is hit, a stack trace is recorded.
3182 The trace_pipe outputs the same content as the trace file, but
3183 the effect on the tracing is different. Every read from
3184 trace_pipe is consumed. This means that subsequent reads will be
3185 different. The trace is live.
3188 # echo function > current_tracer
3189 # cat trace_pipe > /tmp/trace.out &
3191 # echo 1 > tracing_on
3193 # echo 0 > tracing_on
3197 # entries-in-buffer/entries-written: 0/0 #P:4
3200 # / _----=> need-resched
3201 # | / _---=> hardirq/softirq
3202 # || / _--=> preempt-depth
3204 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3208 # cat /tmp/trace.out
3209 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
3210 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
3211 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
3212 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
3213 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
3214 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
3215 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
3216 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
3217 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
3220 Note, reading the trace_pipe file will block until more input is
3221 added. This is contrary to the trace file. If any process opened
3222 the trace file for reading, it will actually disable tracing and
3223 prevent new entries from being added. The trace_pipe file does
3224 not have this limitation.
3229 Having too much or not enough data can be troublesome in
3230 diagnosing an issue in the kernel. The file buffer_size_kb is
3231 used to modify the size of the internal trace buffers. The
3232 number listed is the number of entries that can be recorded per
3233 CPU. To know the full size, multiply the number of possible CPUs
3234 with the number of entries.
3237 # cat buffer_size_kb
3238 1408 (units kilobytes)
3240 Or simply read buffer_total_size_kb
3243 # cat buffer_total_size_kb
3246 To modify the buffer, simple echo in a number (in 1024 byte segments).
3249 # echo 10000 > buffer_size_kb
3250 # cat buffer_size_kb
3251 10000 (units kilobytes)
3253 It will try to allocate as much as possible. If you allocate too
3254 much, it can cause Out-Of-Memory to trigger.
3257 # echo 1000000000000 > buffer_size_kb
3258 -bash: echo: write error: Cannot allocate memory
3259 # cat buffer_size_kb
3262 The per_cpu buffers can be changed individually as well:
3265 # echo 10000 > per_cpu/cpu0/buffer_size_kb
3266 # echo 100 > per_cpu/cpu1/buffer_size_kb
3268 When the per_cpu buffers are not the same, the buffer_size_kb
3269 at the top level will just show an X
3272 # cat buffer_size_kb
3275 This is where the buffer_total_size_kb is useful:
3278 # cat buffer_total_size_kb
3281 Writing to the top level buffer_size_kb will reset all the buffers
3282 to be the same again.
3286 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
3287 available to all non latency tracers. (Latency tracers which
3288 record max latency, such as "irqsoff" or "wakeup", can't use
3289 this feature, since those are already using the snapshot
3290 mechanism internally.)
3292 Snapshot preserves a current trace buffer at a particular point
3293 in time without stopping tracing. Ftrace swaps the current
3294 buffer with a spare buffer, and tracing continues in the new
3295 current (=previous spare) buffer.
3297 The following tracefs files in "tracing" are related to this
3302 This is used to take a snapshot and to read the output
3303 of the snapshot. Echo 1 into this file to allocate a
3304 spare buffer and to take a snapshot (swap), then read
3305 the snapshot from this file in the same format as
3306 "trace" (described above in the section "The File
3307 System"). Both reads snapshot and tracing are executable
3308 in parallel. When the spare buffer is allocated, echoing
3309 0 frees it, and echoing else (positive) values clear the
3311 More details are shown in the table below.
3313 +--------------+------------+------------+------------+
3314 |status\\input | 0 | 1 | else |
3315 +==============+============+============+============+
3316 |not allocated |(do nothing)| alloc+swap |(do nothing)|
3317 +--------------+------------+------------+------------+
3318 |allocated | free | swap | clear |
3319 +--------------+------------+------------+------------+
3321 Here is an example of using the snapshot feature.
3324 # echo 1 > events/sched/enable
3329 # entries-in-buffer/entries-written: 71/71 #P:8
3332 # / _----=> need-resched
3333 # | / _---=> hardirq/softirq
3334 # || / _--=> preempt-depth
3336 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3338 <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
3339 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
3341 <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
3346 # entries-in-buffer/entries-written: 77/77 #P:8
3349 # / _----=> need-resched
3350 # | / _---=> hardirq/softirq
3351 # || / _--=> preempt-depth
3353 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3355 <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
3356 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
3360 If you try to use this snapshot feature when current tracer is
3361 one of the latency tracers, you will get the following results.
3364 # echo wakeup > current_tracer
3366 bash: echo: write error: Device or resource busy
3368 cat: snapshot: Device or resource busy
3373 In the tracefs tracing directory, there is a directory called "instances".
3374 This directory can have new directories created inside of it using
3375 mkdir, and removing directories with rmdir. The directory created
3376 with mkdir in this directory will already contain files and other
3377 directories after it is created.
3380 # mkdir instances/foo
3382 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
3383 set_event snapshot trace trace_clock trace_marker trace_options
3384 trace_pipe tracing_on
3386 As you can see, the new directory looks similar to the tracing directory
3387 itself. In fact, it is very similar, except that the buffer and
3388 events are agnostic from the main directory, or from any other
3389 instances that are created.
3391 The files in the new directory work just like the files with the
3392 same name in the tracing directory except the buffer that is used
3393 is a separate and new buffer. The files affect that buffer but do not
3394 affect the main buffer with the exception of trace_options. Currently,
3395 the trace_options affect all instances and the top level buffer
3396 the same, but this may change in future releases. That is, options
3397 may become specific to the instance they reside in.
3399 Notice that none of the function tracer files are there, nor is
3400 current_tracer and available_tracers. This is because the buffers
3401 can currently only have events enabled for them.
3404 # mkdir instances/foo
3405 # mkdir instances/bar
3406 # mkdir instances/zoot
3407 # echo 100000 > buffer_size_kb
3408 # echo 1000 > instances/foo/buffer_size_kb
3409 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
3410 # echo function > current_trace
3411 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
3412 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
3413 # echo 1 > instances/foo/events/sched/sched_switch/enable
3414 # echo 1 > instances/bar/events/irq/enable
3415 # echo 1 > instances/zoot/events/syscalls/enable
3417 CPU:2 [LOST 11745 EVENTS]
3418 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
3419 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
3420 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
3421 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
3422 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
3423 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
3424 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
3425 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
3426 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3427 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3428 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
3431 # cat instances/foo/trace_pipe
3432 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3433 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3434 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
3435 <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
3436 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
3437 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3438 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3439 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
3440 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
3441 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
3444 # cat instances/bar/trace_pipe
3445 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
3446 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
3447 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
3448 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
3449 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
3450 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
3451 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
3452 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
3453 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
3454 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
3455 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
3456 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
3459 # cat instances/zoot/trace
3462 # entries-in-buffer/entries-written: 18996/18996 #P:4
3465 # / _----=> need-resched
3466 # | / _---=> hardirq/softirq
3467 # || / _--=> preempt-depth
3469 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3471 bash-1998 [000] d... 140.733501: sys_write -> 0x2
3472 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
3473 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
3474 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
3475 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
3476 bash-1998 [000] d... 140.733510: sys_close(fd: a)
3477 bash-1998 [000] d... 140.733510: sys_close -> 0x0
3478 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
3479 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
3480 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
3481 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
3483 You can see that the trace of the top most trace buffer shows only
3484 the function tracing. The foo instance displays wakeups and task
3487 To remove the instances, simply delete their directories:
3490 # rmdir instances/foo
3491 # rmdir instances/bar
3492 # rmdir instances/zoot
3494 Note, if a process has a trace file open in one of the instance
3495 directories, the rmdir will fail with EBUSY.
3500 Since the kernel has a fixed sized stack, it is important not to
3501 waste it in functions. A kernel developer must be conscience of
3502 what they allocate on the stack. If they add too much, the system
3503 can be in danger of a stack overflow, and corruption will occur,
3504 usually leading to a system panic.
3506 There are some tools that check this, usually with interrupts
3507 periodically checking usage. But if you can perform a check
3508 at every function call that will become very useful. As ftrace provides
3509 a function tracer, it makes it convenient to check the stack size
3510 at every function call. This is enabled via the stack tracer.
3512 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
3513 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
3516 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
3518 You can also enable it from the kernel command line to trace
3519 the stack size of the kernel during boot up, by adding "stacktrace"
3520 to the kernel command line parameter.
3522 After running it for a few minutes, the output looks like:
3525 # cat stack_max_size
3529 Depth Size Location (18 entries)
3531 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
3532 1) 2704 160 find_busiest_group+0x31/0x1f1
3533 2) 2544 256 load_balance+0xd9/0x662
3534 3) 2288 80 idle_balance+0xbb/0x130
3535 4) 2208 128 __schedule+0x26e/0x5b9
3536 5) 2080 16 schedule+0x64/0x66
3537 6) 2064 128 schedule_timeout+0x34/0xe0
3538 7) 1936 112 wait_for_common+0x97/0xf1
3539 8) 1824 16 wait_for_completion+0x1d/0x1f
3540 9) 1808 128 flush_work+0xfe/0x119
3541 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
3542 11) 1664 48 input_available_p+0x1d/0x5c
3543 12) 1616 48 n_tty_poll+0x6d/0x134
3544 13) 1568 64 tty_poll+0x64/0x7f
3545 14) 1504 880 do_select+0x31e/0x511
3546 15) 624 400 core_sys_select+0x177/0x216
3547 16) 224 96 sys_select+0x91/0xb9
3548 17) 128 128 system_call_fastpath+0x16/0x1b
3550 Note, if -mfentry is being used by gcc, functions get traced before
3551 they set up the stack frame. This means that leaf level functions
3552 are not tested by the stack tracer when -mfentry is used.
3554 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
3558 More details can be found in the source code, in the `kernel/trace/*.c` files.
projects / linux-2.6-microblaze.git / blob