1 ========================
2 ftrace - Function Tracer
3 ========================
5 Copyright 2008 Red Hat Inc.
7 :Author: Steven Rostedt <srostedt@redhat.com>
8 :License: The GNU Free Documentation License, Version 1.2
9 (dual licensed under the GPL v2)
10 :Original Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton,
11 John Kacur, and David Teigland.
13 - Written for: 2.6.28-rc2
15 - Updated for: 4.13 - Copyright 2017 VMware Inc. Steven Rostedt
16 - Converted to rst format - Changbin Du <changbin.du@intel.com>
21 Ftrace is an internal tracer designed to help out developers and
22 designers of systems to find what is going on inside the kernel.
23 It can be used for debugging or analyzing latencies and
24 performance issues that take place outside of user-space.
26 Although ftrace is typically considered the function tracer, it
27 is really a framework of several assorted tracing utilities.
28 There's latency tracing to examine what occurs between interrupts
29 disabled and enabled, as well as for preemption and from a time
30 a task is woken to the task is actually scheduled in.
32 One of the most common uses of ftrace is the event tracing.
33 Throughout the kernel is hundreds of static event points that
34 can be enabled via the tracefs file system to see what is
35 going on in certain parts of the kernel.
37 See events.txt for more information.
40 Implementation Details
41 ----------------------
43 See :doc:`ftrace-design` for details for arch porters and such.
49 Ftrace uses the tracefs file system to hold the control files as
50 well as the files to display output.
52 When tracefs is configured into the kernel (which selecting any ftrace
53 option will do) the directory /sys/kernel/tracing will be created. To mount
54 this directory, you can add to your /etc/fstab file::
56 tracefs /sys/kernel/tracing tracefs defaults 0 0
58 Or you can mount it at run time with::
60 mount -t tracefs nodev /sys/kernel/tracing
62 For quicker access to that directory you may want to make a soft link to
65 ln -s /sys/kernel/tracing /tracing
69 Before 4.1, all ftrace tracing control files were within the debugfs
70 file system, which is typically located at /sys/kernel/debug/tracing.
71 For backward compatibility, when mounting the debugfs file system,
72 the tracefs file system will be automatically mounted at:
74 /sys/kernel/debug/tracing
76 All files located in the tracefs file system will be located in that
77 debugfs file system directory as well.
81 Any selected ftrace option will also create the tracefs file system.
82 The rest of the document will assume that you are in the ftrace directory
83 (cd /sys/kernel/tracing) and will only concentrate on the files within that
84 directory and not distract from the content with the extended
85 "/sys/kernel/tracing" path name.
87 That's it! (assuming that you have ftrace configured into your kernel)
89 After mounting tracefs you will have access to the control and output files
90 of ftrace. Here is a list of some of the key files:
93 Note: all time values are in microseconds.
97 This is used to set or display the current tracer
102 This holds the different types of tracers that
103 have been compiled into the kernel. The
104 tracers listed here can be configured by
105 echoing their name into current_tracer.
109 This sets or displays whether writing to the trace
110 ring buffer is enabled. Echo 0 into this file to disable
111 the tracer or 1 to enable it. Note, this only disables
112 writing to the ring buffer, the tracing overhead may
115 The kernel function tracing_off() can be used within the
116 kernel to disable writing to the ring buffer, which will
117 set this file to "0". User space can re-enable tracing by
118 echoing "1" into the file.
120 Note, the function and event trigger "traceoff" will also
121 set this file to zero and stop tracing. Which can also
122 be re-enabled by user space using this file.
126 This file holds the output of the trace in a human
127 readable format (described below). Note, tracing is temporarily
128 disabled while this file is being read (opened).
132 The output is the same as the "trace" file but this
133 file is meant to be streamed with live tracing.
134 Reads from this file will block until new data is
135 retrieved. Unlike the "trace" file, this file is a
136 consumer. This means reading from this file causes
137 sequential reads to display more current data. Once
138 data is read from this file, it is consumed, and
139 will not be read again with a sequential read. The
140 "trace" file is static, and if the tracer is not
141 adding more data, it will display the same
142 information every time it is read. This file will not
143 disable tracing while being read.
147 This file lets the user control the amount of data
148 that is displayed in one of the above output
149 files. Options also exist to modify how a tracer
150 or events work (stack traces, timestamps, etc).
154 This is a directory that has a file for every available
155 trace option (also in trace_options). Options may also be set
156 or cleared by writing a "1" or "0" respectively into the
157 corresponding file with the option name.
161 Some of the tracers record the max latency.
162 For example, the maximum time that interrupts are disabled.
163 The maximum time is saved in this file. The max trace will also be
164 stored, and displayed by "trace". A new max trace will only be
165 recorded if the latency is greater than the value in this file
168 By echoing in a time into this file, no latency will be recorded
169 unless it is greater than the time in this file.
173 Some latency tracers will record a trace whenever the
174 latency is greater than the number in this file.
175 Only active when the file contains a number greater than 0.
180 This sets or displays the number of kilobytes each CPU
181 buffer holds. By default, the trace buffers are the same size
182 for each CPU. The displayed number is the size of the
183 CPU buffer and not total size of all buffers. The
184 trace buffers are allocated in pages (blocks of memory
185 that the kernel uses for allocation, usually 4 KB in size).
186 If the last page allocated has room for more bytes
187 than requested, the rest of the page will be used,
188 making the actual allocation bigger than requested or shown.
189 ( Note, the size may not be a multiple of the page size
190 due to buffer management meta-data. )
192 Buffer sizes for individual CPUs may vary
193 (see "per_cpu/cpu0/buffer_size_kb" below), and if they do
194 this file will show "X".
196 buffer_total_size_kb:
198 This displays the total combined size of all the trace buffers.
202 If a process is performing tracing, and the ring buffer should be
203 shrunk "freed" when the process is finished, even if it were to be
204 killed by a signal, this file can be used for that purpose. On close
205 of this file, the ring buffer will be resized to its minimum size.
206 Having a process that is tracing also open this file, when the process
207 exits its file descriptor for this file will be closed, and in doing so,
208 the ring buffer will be "freed".
210 It may also stop tracing if disable_on_free option is set.
214 This is a mask that lets the user only trace on specified CPUs.
215 The format is a hex string representing the CPUs.
219 When dynamic ftrace is configured in (see the
220 section below "dynamic ftrace"), the code is dynamically
221 modified (code text rewrite) to disable calling of the
222 function profiler (mcount). This lets tracing be configured
223 in with practically no overhead in performance. This also
224 has a side effect of enabling or disabling specific functions
225 to be traced. Echoing names of functions into this file
226 will limit the trace to only those functions.
227 This influences the tracers "function" and "function_graph"
228 and thus also function profiling (see "function_profile_enabled").
230 The functions listed in "available_filter_functions" are what
231 can be written into this file.
233 This interface also allows for commands to be used. See the
234 "Filter commands" section for more details.
236 As a speed up, since processing strings can't be quite expensive
237 and requires a check of all functions registered to tracing, instead
238 an index can be written into this file. A number (starting with "1")
239 written will instead select the same corresponding at the line position
240 of the "available_filter_functions" file.
244 This has an effect opposite to that of
245 set_ftrace_filter. Any function that is added here will not
246 be traced. If a function exists in both set_ftrace_filter
247 and set_ftrace_notrace, the function will _not_ be traced.
251 Have the function tracer only trace the threads whose PID are
254 If the "function-fork" option is set, then when a task whose
255 PID is listed in this file forks, the child's PID will
256 automatically be added to this file, and the child will be
257 traced by the function tracer as well. This option will also
258 cause PIDs of tasks that exit to be removed from the file.
262 Have the events only trace a task with a PID listed in this file.
263 Note, sched_switch and sched_wake_up will also trace events
266 To have the PIDs of children of tasks with their PID in this file
267 added on fork, enable the "event-fork" option. That option will also
268 cause the PIDs of tasks to be removed from this file when the task
273 Functions listed in this file will cause the function graph
274 tracer to only trace these functions and the functions that
275 they call. (See the section "dynamic ftrace" for more details).
276 Note, set_ftrace_filter and set_ftrace_notrace still affects
277 what functions are being traced.
281 Similar to set_graph_function, but will disable function graph
282 tracing when the function is hit until it exits the function.
283 This makes it possible to ignore tracing functions that are called
284 by a specific function.
286 available_filter_functions:
288 This lists the functions that ftrace has processed and can trace.
289 These are the function names that you can pass to
290 "set_ftrace_filter", "set_ftrace_notrace",
291 "set_graph_function", or "set_graph_notrace".
292 (See the section "dynamic ftrace" below for more details.)
294 dyn_ftrace_total_info:
296 This file is for debugging purposes. The number of functions that
297 have been converted to nops and are available to be traced.
301 This file is more for debugging ftrace, but can also be useful
302 in seeing if any function has a callback attached to it.
303 Not only does the trace infrastructure use ftrace function
304 trace utility, but other subsystems might too. This file
305 displays all functions that have a callback attached to them
306 as well as the number of callbacks that have been attached.
307 Note, a callback may also call multiple functions which will
308 not be listed in this count.
310 If the callback registered to be traced by a function with
311 the "save regs" attribute (thus even more overhead), a 'R'
312 will be displayed on the same line as the function that
313 is returning registers.
315 If the callback registered to be traced by a function with
316 the "ip modify" attribute (thus the regs->ip can be changed),
317 an 'I' will be displayed on the same line as the function that
320 If the architecture supports it, it will also show what callback
321 is being directly called by the function. If the count is greater
322 than 1 it most likely will be ftrace_ops_list_func().
324 If the callback of the function jumps to a trampoline that is
325 specific to a the callback and not the standard trampoline,
326 its address will be printed as well as the function that the
329 function_profile_enabled:
331 When set it will enable all functions with either the function
332 tracer, or if configured, the function graph tracer. It will
333 keep a histogram of the number of functions that were called
334 and if the function graph tracer was configured, it will also keep
335 track of the time spent in those functions. The histogram
336 content can be displayed in the files:
338 trace_stat/function<cpu> ( function0, function1, etc).
342 A directory that holds different tracing stats.
346 Enable dynamic trace points. See kprobetrace.txt.
350 Dynamic trace points stats. See kprobetrace.txt.
354 Used with the function graph tracer. This is the max depth
355 it will trace into a function. Setting this to a value of
356 one will show only the first kernel function that is called
361 This is for tools that read the raw format files. If an event in
362 the ring buffer references a string, only a pointer to the string
363 is recorded into the buffer and not the string itself. This prevents
364 tools from knowing what that string was. This file displays the string
365 and address for the string allowing tools to map the pointers to what
370 Only the pid of the task is recorded in a trace event unless
371 the event specifically saves the task comm as well. Ftrace
372 makes a cache of pid mappings to comms to try to display
373 comms for events. If a pid for a comm is not listed, then
374 "<...>" is displayed in the output.
376 If the option "record-cmd" is set to "0", then comms of tasks
377 will not be saved during recording. By default, it is enabled.
381 By default, 128 comms are saved (see "saved_cmdlines" above). To
382 increase or decrease the amount of comms that are cached, echo
383 in a the number of comms to cache, into this file.
387 If the option "record-tgid" is set, on each scheduling context switch
388 the Task Group ID of a task is saved in a table mapping the PID of
389 the thread to its TGID. By default, the "record-tgid" option is
394 This displays the "snapshot" buffer and also lets the user
395 take a snapshot of the current running trace.
396 See the "Snapshot" section below for more details.
400 When the stack tracer is activated, this will display the
401 maximum stack size it has encountered.
402 See the "Stack Trace" section below.
406 This displays the stack back trace of the largest stack
407 that was encountered when the stack tracer is activated.
408 See the "Stack Trace" section below.
412 This is similar to "set_ftrace_filter" but it limits what
413 functions the stack tracer will check.
417 Whenever an event is recorded into the ring buffer, a
418 "timestamp" is added. This stamp comes from a specified
419 clock. By default, ftrace uses the "local" clock. This
420 clock is very fast and strictly per cpu, but on some
421 systems it may not be monotonic with respect to other
422 CPUs. In other words, the local clocks may not be in sync
423 with local clocks on other CPUs.
425 Usual clocks for tracing::
428 [local] global counter x86-tsc
430 The clock with the square brackets around it is the one in effect.
433 Default clock, but may not be in sync across CPUs
436 This clock is in sync with all CPUs but may
437 be a bit slower than the local clock.
440 This is not a clock at all, but literally an atomic
441 counter. It counts up one by one, but is in sync
442 with all CPUs. This is useful when you need to
443 know exactly the order events occurred with respect to
444 each other on different CPUs.
447 This uses the jiffies counter and the time stamp
448 is relative to the time since boot up.
451 This makes ftrace use the same clock that perf uses.
452 Eventually perf will be able to read ftrace buffers
453 and this will help out in interleaving the data.
456 Architectures may define their own clocks. For
457 example, x86 uses its own TSC cycle clock here.
460 This uses the powerpc timebase register value.
461 This is in sync across CPUs and can also be used
462 to correlate events across hypervisor/guest if
466 This uses the fast monotonic clock (CLOCK_MONOTONIC)
467 which is monotonic and is subject to NTP rate adjustments.
470 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
471 which is monotonic but is not subject to any rate adjustments
472 and ticks at the same rate as the hardware clocksource.
475 This is the boot clock (CLOCK_BOOTTIME) and is based on the
476 fast monotonic clock, but also accounts for time spent in
477 suspend. Since the clock access is designed for use in
478 tracing in the suspend path, some side effects are possible
479 if clock is accessed after the suspend time is accounted before
480 the fast mono clock is updated. In this case, the clock update
481 appears to happen slightly sooner than it normally would have.
482 Also on 32-bit systems, it's possible that the 64-bit boot offset
483 sees a partial update. These effects are rare and post
484 processing should be able to handle them. See comments in the
485 ktime_get_boot_fast_ns() function for more information.
487 To set a clock, simply echo the clock name into this file::
489 # echo global > trace_clock
493 This is a very useful file for synchronizing user space
494 with events happening in the kernel. Writing strings into
495 this file will be written into the ftrace buffer.
497 It is useful in applications to open this file at the start
498 of the application and just reference the file descriptor
501 void trace_write(const char *fmt, ...)
511 n = vsnprintf(buf, 256, fmt, ap);
514 write(trace_fd, buf, n);
519 trace_fd = open("trace_marker", WR_ONLY);
521 Note: Writing into the trace_marker file can also initiate triggers
522 that are written into /sys/kernel/tracing/events/ftrace/print/trigger
523 See "Event triggers" in Documentation/trace/events.rst and an
524 example in Documentation/trace/histogram.rst (Section 3.)
528 This is similar to trace_marker above, but is meant for for binary data
529 to be written to it, where a tool can be used to parse the data
534 Add dynamic tracepoints in programs.
539 Uprobe statistics. See uprobetrace.txt
543 This is a way to make multiple trace buffers where different
544 events can be recorded in different buffers.
545 See "Instances" section below.
549 This is the trace event directory. It holds event tracepoints
550 (also known as static tracepoints) that have been compiled
551 into the kernel. It shows what event tracepoints exist
552 and how they are grouped by system. There are "enable"
553 files at various levels that can enable the tracepoints
554 when a "1" is written to them.
556 See events.txt for more information.
560 By echoing in the event into this file, will enable that event.
562 See events.txt for more information.
566 A list of events that can be enabled in tracing.
568 See events.txt for more information.
572 Certain tracers may change the timestamp mode used when
573 logging trace events into the event buffer. Events with
574 different modes can coexist within a buffer but the mode in
575 effect when an event is logged determines which timestamp mode
576 is used for that event. The default timestamp mode is
579 Usual timestamp modes for tracing:
584 The timestamp mode with the square brackets around it is the
587 delta: Default timestamp mode - timestamp is a delta against
588 a per-buffer timestamp.
590 absolute: The timestamp is a full timestamp, not a delta
591 against some other value. As such it takes up more
592 space and is less efficient.
596 Directory for the Hardware Latency Detector.
597 See "Hardware Latency Detector" section below.
601 This is a directory that contains the trace per_cpu information.
603 per_cpu/cpu0/buffer_size_kb:
605 The ftrace buffer is defined per_cpu. That is, there's a separate
606 buffer for each CPU to allow writes to be done atomically,
607 and free from cache bouncing. These buffers may have different
608 size buffers. This file is similar to the buffer_size_kb
609 file, but it only displays or sets the buffer size for the
610 specific CPU. (here cpu0).
614 This is similar to the "trace" file, but it will only display
615 the data specific for the CPU. If written to, it only clears
616 the specific CPU buffer.
618 per_cpu/cpu0/trace_pipe
620 This is similar to the "trace_pipe" file, and is a consuming
621 read, but it will only display (and consume) the data specific
624 per_cpu/cpu0/trace_pipe_raw
626 For tools that can parse the ftrace ring buffer binary format,
627 the trace_pipe_raw file can be used to extract the data
628 from the ring buffer directly. With the use of the splice()
629 system call, the buffer data can be quickly transferred to
630 a file or to the network where a server is collecting the
633 Like trace_pipe, this is a consuming reader, where multiple
634 reads will always produce different data.
636 per_cpu/cpu0/snapshot:
638 This is similar to the main "snapshot" file, but will only
639 snapshot the current CPU (if supported). It only displays
640 the content of the snapshot for a given CPU, and if
641 written to, only clears this CPU buffer.
643 per_cpu/cpu0/snapshot_raw:
645 Similar to the trace_pipe_raw, but will read the binary format
646 from the snapshot buffer for the given CPU.
650 This displays certain stats about the ring buffer:
653 The number of events that are still in the buffer.
656 The number of lost events due to overwriting when
660 Should always be zero.
661 This gets set if so many events happened within a nested
662 event (ring buffer is re-entrant), that it fills the
663 buffer and starts dropping events.
666 Bytes actually read (not overwritten).
669 The oldest timestamp in the buffer
672 The current timestamp
675 Events lost due to overwrite option being off.
678 The number of events read.
683 Here is the list of current tracers that may be configured.
687 Function call tracer to trace all kernel functions.
691 Similar to the function tracer except that the
692 function tracer probes the functions on their entry
693 whereas the function graph tracer traces on both entry
694 and exit of the functions. It then provides the ability
695 to draw a graph of function calls similar to C code
700 The block tracer. The tracer used by the blktrace user
705 The Hardware Latency tracer is used to detect if the hardware
706 produces any latency. See "Hardware Latency Detector" section
711 Traces the areas that disable interrupts and saves
712 the trace with the longest max latency.
713 See tracing_max_latency. When a new max is recorded,
714 it replaces the old trace. It is best to view this
715 trace with the latency-format option enabled, which
716 happens automatically when the tracer is selected.
720 Similar to irqsoff but traces and records the amount of
721 time for which preemption is disabled.
725 Similar to irqsoff and preemptoff, but traces and
726 records the largest time for which irqs and/or preemption
731 Traces and records the max latency that it takes for
732 the highest priority task to get scheduled after
733 it has been woken up.
734 Traces all tasks as an average developer would expect.
738 Traces and records the max latency that it takes for just
739 RT tasks (as the current "wakeup" does). This is useful
740 for those interested in wake up timings of RT tasks.
744 Traces and records the max latency that it takes for
745 a SCHED_DEADLINE task to be woken (as the "wakeup" and
750 A special tracer that is used to trace binary module.
751 It will trace all the calls that a module makes to the
752 hardware. Everything it writes and reads from the I/O
757 This tracer can be configured when tracing likely/unlikely
758 calls within the kernel. It will trace when a likely and
759 unlikely branch is hit and if it was correct in its prediction
764 This is the "trace nothing" tracer. To remove all
765 tracers from tracing simply echo "nop" into
769 Examples of using the tracer
770 ----------------------------
772 Here are typical examples of using the tracers when controlling
773 them only with the tracefs interface (without using any
774 user-land utilities).
779 Here is an example of the output format of the file "trace"::
783 # entries-in-buffer/entries-written: 140080/250280 #P:4
786 # / _----=> need-resched
787 # | / _---=> hardirq/softirq
788 # || / _--=> preempt-depth
790 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
792 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
793 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
794 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
795 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
796 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
797 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
798 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
799 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
800 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
801 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
804 A header is printed with the tracer name that is represented by
805 the trace. In this case the tracer is "function". Then it shows the
806 number of events in the buffer as well as the total number of entries
807 that were written. The difference is the number of entries that were
808 lost due to the buffer filling up (250280 - 140080 = 110200 events
811 The header explains the content of the events. Task name "bash", the task
812 PID "1977", the CPU that it was running on "000", the latency format
813 (explained below), the timestamp in <secs>.<usecs> format, the
814 function name that was traced "sys_close" and the parent function that
815 called this function "system_call_fastpath". The timestamp is the time
816 at which the function was entered.
821 When the latency-format option is enabled or when one of the latency
822 tracers is set, the trace file gives somewhat more information to see
823 why a latency happened. Here is a typical trace::
827 # irqsoff latency trace v1.1.5 on 3.8.0-test+
828 # --------------------------------------------------------------------
829 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
831 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
833 # => started at: __lock_task_sighand
834 # => ended at: _raw_spin_unlock_irqrestore
838 # / _-----=> irqs-off
839 # | / _----=> need-resched
840 # || / _---=> hardirq/softirq
841 # ||| / _--=> preempt-depth
843 # cmd pid ||||| time | caller
845 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
846 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
847 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
848 ps-6143 2d..1 306us : <stack trace>
849 => trace_hardirqs_on_caller
851 => _raw_spin_unlock_irqrestore
858 => system_call_fastpath
861 This shows that the current tracer is "irqsoff" tracing the time
862 for which interrupts were disabled. It gives the trace version (which
863 never changes) and the version of the kernel upon which this was executed on
864 (3.8). Then it displays the max latency in microseconds (259 us). The number
865 of trace entries displayed and the total number (both are four: #4/4).
866 VP, KP, SP, and HP are always zero and are reserved for later use.
867 #P is the number of online CPUs (#P:4).
869 The task is the process that was running when the latency
870 occurred. (ps pid: 6143).
872 The start and stop (the functions in which the interrupts were
873 disabled and enabled respectively) that caused the latencies:
875 - __lock_task_sighand is where the interrupts were disabled.
876 - _raw_spin_unlock_irqrestore is where they were enabled again.
878 The next lines after the header are the trace itself. The header
879 explains which is which.
881 cmd: The name of the process in the trace.
883 pid: The PID of that process.
885 CPU#: The CPU which the process was running on.
887 irqs-off: 'd' interrupts are disabled. '.' otherwise.
888 .. caution:: If the architecture does not support a way to
889 read the irq flags variable, an 'X' will always
893 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
894 - 'n' only TIF_NEED_RESCHED is set,
895 - 'p' only PREEMPT_NEED_RESCHED is set,
899 - 'Z' - NMI occurred inside a hardirq
900 - 'z' - NMI is running
901 - 'H' - hard irq occurred inside a softirq.
902 - 'h' - hard irq is running
903 - 's' - soft irq is running
904 - '.' - normal context.
906 preempt-depth: The level of preempt_disabled
908 The above is mostly meaningful for kernel developers.
911 When the latency-format option is enabled, the trace file
912 output includes a timestamp relative to the start of the
913 trace. This differs from the output when latency-format
914 is disabled, which includes an absolute timestamp.
917 This is just to help catch your eye a bit better. And
918 needs to be fixed to be only relative to the same CPU.
919 The marks are determined by the difference between this
920 current trace and the next trace.
922 - '$' - greater than 1 second
923 - '@' - greater than 100 millisecond
924 - '*' - greater than 10 millisecond
925 - '#' - greater than 1000 microsecond
926 - '!' - greater than 100 microsecond
927 - '+' - greater than 10 microsecond
928 - ' ' - less than or equal to 10 microsecond.
930 The rest is the same as the 'trace' file.
932 Note, the latency tracers will usually end with a back trace
933 to easily find where the latency occurred.
938 The trace_options file (or the options directory) is used to control
939 what gets printed in the trace output, or manipulate the tracers.
940 To see what is available, simply cat the file::
971 To disable one of the options, echo in the option prepended with
974 echo noprint-parent > trace_options
976 To enable an option, leave off the "no"::
978 echo sym-offset > trace_options
980 Here are the available options:
983 On function traces, display the calling (parent)
984 function as well as the function being traced.
988 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
991 bash-4000 [01] 1477.606694: simple_strtoul
995 Display not only the function name, but also the
996 offset in the function. For example, instead of
997 seeing just "ktime_get", you will see
998 "ktime_get+0xb/0x20".
1002 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
1005 This will also display the function address as well
1006 as the function name.
1010 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
1013 This deals with the trace file when the
1014 latency-format option is enabled.
1017 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
1018 (+0.000ms): simple_strtoul (kstrtoul)
1021 This will display raw numbers. This option is best for
1022 use with user applications that can translate the raw
1023 numbers better than having it done in the kernel.
1026 Similar to raw, but the numbers will be in a hexadecimal format.
1029 This will print out the formats in raw binary.
1032 When set, reading trace_pipe will not block when polled.
1035 Can disable trace_printk() from writing into the buffer.
1038 It is sometimes confusing when the CPU buffers are full
1039 and one CPU buffer had a lot of events recently, thus
1040 a shorter time frame, were another CPU may have only had
1041 a few events, which lets it have older events. When
1042 the trace is reported, it shows the oldest events first,
1043 and it may look like only one CPU ran (the one with the
1044 oldest events). When the annotate option is set, it will
1045 display when a new CPU buffer started::
1047 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
1048 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
1049 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
1050 ##### CPU 2 buffer started ####
1051 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
1052 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
1053 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
1056 This option changes the trace. It records a
1057 stacktrace of the current user space thread after
1061 when user stacktrace are enabled, look up which
1062 object the address belongs to, and print a
1063 relative address. This is especially useful when
1064 ASLR is on, otherwise you don't get a chance to
1065 resolve the address to object/file/line after
1066 the app is no longer running
1068 The lookup is performed when you read
1069 trace,trace_pipe. Example::
1071 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
1072 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
1076 When set, trace_printk()s will only show the format
1077 and not their parameters (if trace_bprintk() or
1078 trace_bputs() was used to save the trace_printk()).
1081 Show only the event data. Hides the comm, PID,
1082 timestamp, CPU, and other useful data.
1085 This option changes the trace output. When it is enabled,
1086 the trace displays additional information about the
1087 latency, as described in "Latency trace format".
1090 When any event or tracer is enabled, a hook is enabled
1091 in the sched_switch trace point to fill comm cache
1092 with mapped pids and comms. But this may cause some
1093 overhead, and if you only care about pids, and not the
1094 name of the task, disabling this option can lower the
1095 impact of tracing. See "saved_cmdlines".
1098 When any event or tracer is enabled, a hook is enabled
1099 in the sched_switch trace point to fill the cache of
1100 mapped Thread Group IDs (TGID) mapping to pids. See
1104 This controls what happens when the trace buffer is
1105 full. If "1" (default), the oldest events are
1106 discarded and overwritten. If "0", then the newest
1107 events are discarded.
1108 (see per_cpu/cpu0/stats for overrun and dropped)
1111 When the free_buffer is closed, tracing will
1112 stop (tracing_on set to 0).
1115 Shows the interrupt, preempt count, need resched data.
1116 When disabled, the trace looks like::
1120 # entries-in-buffer/entries-written: 144405/9452052 #P:4
1122 # TASK-PID CPU# TIMESTAMP FUNCTION
1124 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
1125 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
1126 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
1130 When set, the trace_marker is writable (only by root).
1131 When disabled, the trace_marker will error with EINVAL
1135 When set, tasks with PIDs listed in set_event_pid will have
1136 the PIDs of their children added to set_event_pid when those
1137 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
1138 their PIDs will be removed from the file.
1141 The latency tracers will enable function tracing
1142 if this option is enabled (default it is). When
1143 it is disabled, the latency tracers do not trace
1144 functions. This keeps the overhead of the tracer down
1145 when performing latency tests.
1148 When set, tasks with PIDs listed in set_ftrace_pid will
1149 have the PIDs of their children added to set_ftrace_pid
1150 when those tasks fork. Also, when tasks with PIDs in
1151 set_ftrace_pid exit, their PIDs will be removed from the
1155 When set, the latency tracers (irqsoff, wakeup, etc) will
1156 use function graph tracing instead of function tracing.
1159 When set, a stack trace is recorded after any trace event
1163 Enable branch tracing with the tracer. This enables branch
1164 tracer along with the currently set tracer. Enabling this
1165 with the "nop" tracer is the same as just enabling the
1168 .. tip:: Some tracers have their own options. They only appear in this
1169 file when the tracer is active. They always appear in the
1173 Here are the per tracer options:
1175 Options for function tracer:
1178 When set, a stack trace is recorded after every
1179 function that is recorded. NOTE! Limit the functions
1180 that are recorded before enabling this, with
1181 "set_ftrace_filter" otherwise the system performance
1182 will be critically degraded. Remember to disable
1183 this option before clearing the function filter.
1185 Options for function_graph tracer:
1187 Since the function_graph tracer has a slightly different output
1188 it has its own options to control what is displayed.
1191 When set, the "overrun" of the graph stack is
1192 displayed after each function traced. The
1193 overrun, is when the stack depth of the calls
1194 is greater than what is reserved for each task.
1195 Each task has a fixed array of functions to
1196 trace in the call graph. If the depth of the
1197 calls exceeds that, the function is not traced.
1198 The overrun is the number of functions missed
1199 due to exceeding this array.
1202 When set, the CPU number of the CPU where the trace
1203 occurred is displayed.
1206 When set, if the function takes longer than
1207 A certain amount, then a delay marker is
1208 displayed. See "delay" above, under the
1212 Unlike other tracers, the process' command line
1213 is not displayed by default, but instead only
1214 when a task is traced in and out during a context
1215 switch. Enabling this options has the command
1216 of each process displayed at every line.
1219 At the end of each function (the return)
1220 the duration of the amount of time in the
1221 function is displayed in microseconds.
1224 When set, the timestamp is displayed at each line.
1227 When disabled, functions that happen inside an
1228 interrupt will not be traced.
1231 When set, the return event will include the function
1232 that it represents. By default this is off, and
1233 only a closing curly bracket "}" is displayed for
1234 the return of a function.
1237 When running function graph tracer, to include
1238 the time a task schedules out in its function.
1239 When enabled, it will account time the task has been
1240 scheduled out as part of the function call.
1243 When running function profiler with function graph tracer,
1244 to include the time to call nested functions. When this is
1245 not set, the time reported for the function will only
1246 include the time the function itself executed for, not the
1247 time for functions that it called.
1249 Options for blk tracer:
1252 Shows a more minimalistic output.
1258 When interrupts are disabled, the CPU can not react to any other
1259 external event (besides NMIs and SMIs). This prevents the timer
1260 interrupt from triggering or the mouse interrupt from letting
1261 the kernel know of a new mouse event. The result is a latency
1262 with the reaction time.
1264 The irqsoff tracer tracks the time for which interrupts are
1265 disabled. When a new maximum latency is hit, the tracer saves
1266 the trace leading up to that latency point so that every time a
1267 new maximum is reached, the old saved trace is discarded and the
1270 To reset the maximum, echo 0 into tracing_max_latency. Here is
1273 # echo 0 > options/function-trace
1274 # echo irqsoff > current_tracer
1275 # echo 1 > tracing_on
1276 # echo 0 > tracing_max_latency
1279 # echo 0 > tracing_on
1283 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1284 # --------------------------------------------------------------------
1285 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1287 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
1289 # => started at: run_timer_softirq
1290 # => ended at: run_timer_softirq
1294 # / _-----=> irqs-off
1295 # | / _----=> need-resched
1296 # || / _---=> hardirq/softirq
1297 # ||| / _--=> preempt-depth
1299 # cmd pid ||||| time | caller
1301 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
1302 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
1303 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
1304 <idle>-0 0dNs3 25us : <stack trace>
1305 => _raw_spin_unlock_irq
1306 => run_timer_softirq
1311 => smp_apic_timer_interrupt
1312 => apic_timer_interrupt
1317 => x86_64_start_reservations
1318 => x86_64_start_kernel
1320 Here we see that that we had a latency of 16 microseconds (which is
1321 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
1322 interrupts. The difference between the 16 and the displayed
1323 timestamp 25us occurred because the clock was incremented
1324 between the time of recording the max latency and the time of
1325 recording the function that had that latency.
1327 Note the above example had function-trace not set. If we set
1328 function-trace, we get a much larger output::
1330 with echo 1 > options/function-trace
1334 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1335 # --------------------------------------------------------------------
1336 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1338 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1340 # => started at: ata_scsi_queuecmd
1341 # => ended at: ata_scsi_queuecmd
1345 # / _-----=> irqs-off
1346 # | / _----=> need-resched
1347 # || / _---=> hardirq/softirq
1348 # ||| / _--=> preempt-depth
1350 # cmd pid ||||| time | caller
1352 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1353 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1354 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1355 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1356 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1357 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1358 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1359 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1360 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1362 bash-2042 3d..1 67us : delay_tsc <-__delay
1363 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1364 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1365 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1366 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1367 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1368 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1369 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1370 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1371 bash-2042 3d..1 120us : <stack trace>
1372 => _raw_spin_unlock_irqrestore
1373 => ata_scsi_queuecmd
1374 => scsi_dispatch_cmd
1376 => __blk_run_queue_uncond
1379 => generic_make_request
1382 => __ext3_get_inode_loc
1391 => user_path_at_empty
1396 => system_call_fastpath
1399 Here we traced a 71 microsecond latency. But we also see all the
1400 functions that were called during that time. Note that by
1401 enabling function tracing, we incur an added overhead. This
1402 overhead may extend the latency times. But nevertheless, this
1403 trace has provided some very helpful debugging information.
1405 If we prefer function graph output instead of function, we can set
1406 display-graph option::
1408 with echo 1 > options/display-graph
1412 # irqsoff latency trace v1.1.5 on 4.20.0-rc6+
1413 # --------------------------------------------------------------------
1414 # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4)
1416 # | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0)
1418 # => started at: free_debug_processing
1419 # => ended at: return_to_handler
1423 # / _----=> need-resched
1424 # | / _---=> hardirq/softirq
1425 # || / _--=> preempt-depth
1427 # REL TIME CPU TASK/PID |||| DURATION FUNCTION CALLS
1428 # | | | | |||| | | | | | |
1429 0 us | 0) bash-1507 | d... | 0.000 us | _raw_spin_lock_irqsave();
1430 0 us | 0) bash-1507 | d..1 | 0.378 us | do_raw_spin_trylock();
1431 1 us | 0) bash-1507 | d..2 | | set_track() {
1432 2 us | 0) bash-1507 | d..2 | | save_stack_trace() {
1433 2 us | 0) bash-1507 | d..2 | | __save_stack_trace() {
1434 3 us | 0) bash-1507 | d..2 | | __unwind_start() {
1435 3 us | 0) bash-1507 | d..2 | | get_stack_info() {
1436 3 us | 0) bash-1507 | d..2 | 0.351 us | in_task_stack();
1437 4 us | 0) bash-1507 | d..2 | 1.107 us | }
1439 3750 us | 0) bash-1507 | d..1 | 0.516 us | do_raw_spin_unlock();
1440 3750 us | 0) bash-1507 | d..1 | 0.000 us | _raw_spin_unlock_irqrestore();
1441 3764 us | 0) bash-1507 | d..1 | 0.000 us | tracer_hardirqs_on();
1442 bash-1507 0d..1 3792us : <stack trace>
1443 => free_debug_processing
1452 => search_binary_handler
1453 => __do_execve_file.isra.32
1456 => entry_SYSCALL_64_after_hwframe
1461 When preemption is disabled, we may be able to receive
1462 interrupts but the task cannot be preempted and a higher
1463 priority task must wait for preemption to be enabled again
1464 before it can preempt a lower priority task.
1466 The preemptoff tracer traces the places that disable preemption.
1467 Like the irqsoff tracer, it records the maximum latency for
1468 which preemption was disabled. The control of preemptoff tracer
1469 is much like the irqsoff tracer.
1472 # echo 0 > options/function-trace
1473 # echo preemptoff > current_tracer
1474 # echo 1 > tracing_on
1475 # echo 0 > tracing_max_latency
1478 # echo 0 > tracing_on
1480 # tracer: preemptoff
1482 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1483 # --------------------------------------------------------------------
1484 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1486 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1488 # => started at: do_IRQ
1489 # => ended at: do_IRQ
1493 # / _-----=> irqs-off
1494 # | / _----=> need-resched
1495 # || / _---=> hardirq/softirq
1496 # ||| / _--=> preempt-depth
1498 # cmd pid ||||| time | caller
1500 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1501 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1502 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1503 sshd-1991 1d..1 52us : <stack trace>
1504 => sub_preempt_count
1510 This has some more changes. Preemption was disabled when an
1511 interrupt came in (notice the 'h'), and was enabled on exit.
1512 But we also see that interrupts have been disabled when entering
1513 the preempt off section and leaving it (the 'd'). We do not know if
1514 interrupts were enabled in the mean time or shortly after this
1518 # tracer: preemptoff
1520 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1521 # --------------------------------------------------------------------
1522 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1524 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1526 # => started at: wake_up_new_task
1527 # => ended at: task_rq_unlock
1531 # / _-----=> irqs-off
1532 # | / _----=> need-resched
1533 # || / _---=> hardirq/softirq
1534 # ||| / _--=> preempt-depth
1536 # cmd pid ||||| time | caller
1538 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1539 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1540 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1541 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1542 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1544 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1545 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1546 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1547 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1548 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1549 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1550 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1551 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1553 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1554 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1555 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1556 bash-1994 1d..2 36us : do_softirq <-irq_exit
1557 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1558 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1559 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1560 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1561 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1562 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1564 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1565 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1566 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1567 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1568 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1569 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1570 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1571 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1572 bash-1994 1.N.1 104us : <stack trace>
1573 => sub_preempt_count
1574 => _raw_spin_unlock_irqrestore
1582 The above is an example of the preemptoff trace with
1583 function-trace set. Here we see that interrupts were not disabled
1584 the entire time. The irq_enter code lets us know that we entered
1585 an interrupt 'h'. Before that, the functions being traced still
1586 show that it is not in an interrupt, but we can see from the
1587 functions themselves that this is not the case.
1592 Knowing the locations that have interrupts disabled or
1593 preemption disabled for the longest times is helpful. But
1594 sometimes we would like to know when either preemption and/or
1595 interrupts are disabled.
1597 Consider the following code::
1599 local_irq_disable();
1600 call_function_with_irqs_off();
1602 call_function_with_irqs_and_preemption_off();
1604 call_function_with_preemption_off();
1607 The irqsoff tracer will record the total length of
1608 call_function_with_irqs_off() and
1609 call_function_with_irqs_and_preemption_off().
1611 The preemptoff tracer will record the total length of
1612 call_function_with_irqs_and_preemption_off() and
1613 call_function_with_preemption_off().
1615 But neither will trace the time that interrupts and/or
1616 preemption is disabled. This total time is the time that we can
1617 not schedule. To record this time, use the preemptirqsoff
1620 Again, using this trace is much like the irqsoff and preemptoff
1624 # echo 0 > options/function-trace
1625 # echo preemptirqsoff > current_tracer
1626 # echo 1 > tracing_on
1627 # echo 0 > tracing_max_latency
1630 # echo 0 > tracing_on
1632 # tracer: preemptirqsoff
1634 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1635 # --------------------------------------------------------------------
1636 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1638 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1640 # => started at: ata_scsi_queuecmd
1641 # => ended at: ata_scsi_queuecmd
1645 # / _-----=> irqs-off
1646 # | / _----=> need-resched
1647 # || / _---=> hardirq/softirq
1648 # ||| / _--=> preempt-depth
1650 # cmd pid ||||| time | caller
1652 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1653 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1654 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1655 ls-2230 3...1 111us : <stack trace>
1656 => sub_preempt_count
1657 => _raw_spin_unlock_irqrestore
1658 => ata_scsi_queuecmd
1659 => scsi_dispatch_cmd
1661 => __blk_run_queue_uncond
1664 => generic_make_request
1669 => htree_dirblock_to_tree
1670 => ext3_htree_fill_tree
1674 => system_call_fastpath
1677 The trace_hardirqs_off_thunk is called from assembly on x86 when
1678 interrupts are disabled in the assembly code. Without the
1679 function tracing, we do not know if interrupts were enabled
1680 within the preemption points. We do see that it started with
1683 Here is a trace with function-trace set::
1685 # tracer: preemptirqsoff
1687 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1688 # --------------------------------------------------------------------
1689 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1691 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1693 # => started at: schedule
1694 # => ended at: mutex_unlock
1698 # / _-----=> irqs-off
1699 # | / _----=> need-resched
1700 # || / _---=> hardirq/softirq
1701 # ||| / _--=> preempt-depth
1703 # cmd pid ||||| time | caller
1705 kworker/-59 3...1 0us : __schedule <-schedule
1706 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1707 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1708 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1709 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1710 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1711 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1712 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1713 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1714 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1715 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1716 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1717 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1718 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1719 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1720 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1721 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1722 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1723 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1724 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1725 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1726 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1727 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1728 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1729 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1730 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1731 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1732 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1733 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1734 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1735 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1736 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1738 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1739 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1740 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1741 ls-2269 3d..3 21us : do_softirq <-irq_exit
1742 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1743 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1744 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1745 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1746 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1747 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1748 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1750 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1751 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1752 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1753 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1754 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1755 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1757 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1758 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1759 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1760 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1761 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1762 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1763 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1764 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1765 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1766 ls-2269 3d... 186us : <stack trace>
1767 => __mutex_unlock_slowpath
1774 => system_call_fastpath
1776 This is an interesting trace. It started with kworker running and
1777 scheduling out and ls taking over. But as soon as ls released the
1778 rq lock and enabled interrupts (but not preemption) an interrupt
1779 triggered. When the interrupt finished, it started running softirqs.
1780 But while the softirq was running, another interrupt triggered.
1781 When an interrupt is running inside a softirq, the annotation is 'H'.
1787 One common case that people are interested in tracing is the
1788 time it takes for a task that is woken to actually wake up.
1789 Now for non Real-Time tasks, this can be arbitrary. But tracing
1790 it none the less can be interesting.
1792 Without function tracing::
1794 # echo 0 > options/function-trace
1795 # echo wakeup > current_tracer
1796 # echo 1 > tracing_on
1797 # echo 0 > tracing_max_latency
1799 # echo 0 > tracing_on
1803 # wakeup latency trace v1.1.5 on 3.8.0-test+
1804 # --------------------------------------------------------------------
1805 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1807 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1811 # / _-----=> irqs-off
1812 # | / _----=> need-resched
1813 # || / _---=> hardirq/softirq
1814 # ||| / _--=> preempt-depth
1816 # cmd pid ||||| time | caller
1818 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1819 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1820 <idle>-0 3d..3 15us : __schedule <-schedule
1821 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1823 The tracer only traces the highest priority task in the system
1824 to avoid tracing the normal circumstances. Here we see that
1825 the kworker with a nice priority of -20 (not very nice), took
1826 just 15 microseconds from the time it woke up, to the time it
1829 Non Real-Time tasks are not that interesting. A more interesting
1830 trace is to concentrate only on Real-Time tasks.
1835 In a Real-Time environment it is very important to know the
1836 wakeup time it takes for the highest priority task that is woken
1837 up to the time that it executes. This is also known as "schedule
1838 latency". I stress the point that this is about RT tasks. It is
1839 also important to know the scheduling latency of non-RT tasks,
1840 but the average schedule latency is better for non-RT tasks.
1841 Tools like LatencyTop are more appropriate for such
1844 Real-Time environments are interested in the worst case latency.
1845 That is the longest latency it takes for something to happen,
1846 and not the average. We can have a very fast scheduler that may
1847 only have a large latency once in a while, but that would not
1848 work well with Real-Time tasks. The wakeup_rt tracer was designed
1849 to record the worst case wakeups of RT tasks. Non-RT tasks are
1850 not recorded because the tracer only records one worst case and
1851 tracing non-RT tasks that are unpredictable will overwrite the
1852 worst case latency of RT tasks (just run the normal wakeup
1853 tracer for a while to see that effect).
1855 Since this tracer only deals with RT tasks, we will run this
1856 slightly differently than we did with the previous tracers.
1857 Instead of performing an 'ls', we will run 'sleep 1' under
1858 'chrt' which changes the priority of the task.
1861 # echo 0 > options/function-trace
1862 # echo wakeup_rt > current_tracer
1863 # echo 1 > tracing_on
1864 # echo 0 > tracing_max_latency
1866 # echo 0 > tracing_on
1872 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1873 # --------------------------------------------------------------------
1874 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1876 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1880 # / _-----=> irqs-off
1881 # | / _----=> need-resched
1882 # || / _---=> hardirq/softirq
1883 # ||| / _--=> preempt-depth
1885 # cmd pid ||||| time | caller
1887 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1888 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1889 <idle>-0 3d..3 5us : __schedule <-schedule
1890 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1893 Running this on an idle system, we see that it only took 5 microseconds
1894 to perform the task switch. Note, since the trace point in the schedule
1895 is before the actual "switch", we stop the tracing when the recorded task
1896 is about to schedule in. This may change if we add a new marker at the
1897 end of the scheduler.
1899 Notice that the recorded task is 'sleep' with the PID of 2389
1900 and it has an rt_prio of 5. This priority is user-space priority
1901 and not the internal kernel priority. The policy is 1 for
1902 SCHED_FIFO and 2 for SCHED_RR.
1904 Note, that the trace data shows the internal priority (99 - rtprio).
1907 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1909 The 0:120:R means idle was running with a nice priority of 0 (120 - 120)
1910 and in the running state 'R'. The sleep task was scheduled in with
1911 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1912 and it too is in the running state.
1914 Doing the same with chrt -r 5 and function-trace set.
1917 echo 1 > options/function-trace
1921 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1922 # --------------------------------------------------------------------
1923 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1925 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
1929 # / _-----=> irqs-off
1930 # | / _----=> need-resched
1931 # || / _---=> hardirq/softirq
1932 # ||| / _--=> preempt-depth
1934 # cmd pid ||||| time | caller
1936 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
1937 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1938 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
1939 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
1940 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
1941 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
1942 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
1943 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
1944 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
1945 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1946 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
1947 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
1948 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
1949 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
1950 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
1951 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
1952 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
1953 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
1954 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
1955 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
1956 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
1957 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
1958 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
1959 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
1960 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
1961 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
1962 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
1963 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
1964 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
1965 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
1966 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
1967 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
1968 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
1969 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
1970 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
1971 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
1972 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
1973 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit
1974 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
1975 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
1976 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
1977 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
1978 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1979 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
1980 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
1981 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
1982 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
1983 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
1984 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
1985 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
1986 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
1987 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1988 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
1989 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1990 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1991 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
1992 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
1993 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
1994 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1995 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
1996 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
1997 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
1998 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
1999 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
2000 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
2001 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
2002 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
2003 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2004 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
2005 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
2006 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
2007 <idle>-0 3.N.. 25us : schedule <-cpu_idle
2008 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
2009 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
2010 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
2011 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
2012 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
2013 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
2014 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
2015 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
2016 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
2017 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
2018 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
2019 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
2020 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
2022 This isn't that big of a trace, even with function tracing enabled,
2023 so I included the entire trace.
2025 The interrupt went off while when the system was idle. Somewhere
2026 before task_woken_rt() was called, the NEED_RESCHED flag was set,
2027 this is indicated by the first occurrence of the 'N' flag.
2029 Latency tracing and events
2030 --------------------------
2031 As function tracing can induce a much larger latency, but without
2032 seeing what happens within the latency it is hard to know what
2033 caused it. There is a middle ground, and that is with enabling
2037 # echo 0 > options/function-trace
2038 # echo wakeup_rt > current_tracer
2039 # echo 1 > events/enable
2040 # echo 1 > tracing_on
2041 # echo 0 > tracing_max_latency
2043 # echo 0 > tracing_on
2047 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
2048 # --------------------------------------------------------------------
2049 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
2051 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
2055 # / _-----=> irqs-off
2056 # | / _----=> need-resched
2057 # || / _---=> hardirq/softirq
2058 # ||| / _--=> preempt-depth
2060 # cmd pid ||||| time | caller
2062 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
2063 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
2064 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
2065 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
2066 <idle>-0 2.N.2 2us : power_end: cpu_id=2
2067 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
2068 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
2069 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
2070 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
2071 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
2072 <idle>-0 2d..3 6us : __schedule <-schedule
2073 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
2076 Hardware Latency Detector
2077 -------------------------
2079 The hardware latency detector is executed by enabling the "hwlat" tracer.
2081 NOTE, this tracer will affect the performance of the system as it will
2082 periodically make a CPU constantly busy with interrupts disabled.
2085 # echo hwlat > current_tracer
2091 # / _----=> need-resched
2092 # | / _---=> hardirq/softirq
2093 # || / _--=> preempt-depth
2095 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2097 <...>-3638 [001] d... 19452.055471: #1 inner/outer(us): 12/14 ts:1499801089.066141940
2098 <...>-3638 [003] d... 19454.071354: #2 inner/outer(us): 11/9 ts:1499801091.082164365
2099 <...>-3638 [002] dn.. 19461.126852: #3 inner/outer(us): 12/9 ts:1499801098.138150062
2100 <...>-3638 [001] d... 19488.340960: #4 inner/outer(us): 8/12 ts:1499801125.354139633
2101 <...>-3638 [003] d... 19494.388553: #5 inner/outer(us): 8/12 ts:1499801131.402150961
2102 <...>-3638 [003] d... 19501.283419: #6 inner/outer(us): 0/12 ts:1499801138.297435289 nmi-total:4 nmi-count:1
2105 The above output is somewhat the same in the header. All events will have
2106 interrupts disabled 'd'. Under the FUNCTION title there is:
2109 This is the count of events recorded that were greater than the
2110 tracing_threshold (See below).
2112 inner/outer(us): 12/14
2114 This shows two numbers as "inner latency" and "outer latency". The test
2115 runs in a loop checking a timestamp twice. The latency detected within
2116 the two timestamps is the "inner latency" and the latency detected
2117 after the previous timestamp and the next timestamp in the loop is
2118 the "outer latency".
2120 ts:1499801089.066141940
2122 The absolute timestamp that the event happened.
2124 nmi-total:4 nmi-count:1
2126 On architectures that support it, if an NMI comes in during the
2127 test, the time spent in NMI is reported in "nmi-total" (in
2130 All architectures that have NMIs will show the "nmi-count" if an
2131 NMI comes in during the test.
2136 This gets automatically set to "10" to represent 10
2137 microseconds. This is the threshold of latency that
2138 needs to be detected before the trace will be recorded.
2140 Note, when hwlat tracer is finished (another tracer is
2141 written into "current_tracer"), the original value for
2142 tracing_threshold is placed back into this file.
2144 hwlat_detector/width
2145 The length of time the test runs with interrupts disabled.
2147 hwlat_detector/window
2148 The length of time of the window which the test
2149 runs. That is, the test will run for "width"
2150 microseconds per "window" microseconds
2153 When the test is started. A kernel thread is created that
2154 runs the test. This thread will alternate between CPUs
2155 listed in the tracing_cpumask between each period
2156 (one "window"). To limit the test to specific CPUs
2157 set the mask in this file to only the CPUs that the test
2163 This tracer is the function tracer. Enabling the function tracer
2164 can be done from the debug file system. Make sure the
2165 ftrace_enabled is set; otherwise this tracer is a nop.
2166 See the "ftrace_enabled" section below.
2169 # sysctl kernel.ftrace_enabled=1
2170 # echo function > current_tracer
2171 # echo 1 > tracing_on
2173 # echo 0 > tracing_on
2177 # entries-in-buffer/entries-written: 24799/24799 #P:4
2180 # / _----=> need-resched
2181 # | / _---=> hardirq/softirq
2182 # || / _--=> preempt-depth
2184 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2186 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
2187 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
2188 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
2189 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
2190 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
2191 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
2192 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
2193 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
2197 Note: function tracer uses ring buffers to store the above
2198 entries. The newest data may overwrite the oldest data.
2199 Sometimes using echo to stop the trace is not sufficient because
2200 the tracing could have overwritten the data that you wanted to
2201 record. For this reason, it is sometimes better to disable
2202 tracing directly from a program. This allows you to stop the
2203 tracing at the point that you hit the part that you are
2204 interested in. To disable the tracing directly from a C program,
2205 something like following code snippet can be used::
2209 int main(int argc, char *argv[]) {
2211 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
2213 if (condition_hit()) {
2214 write(trace_fd, "0", 1);
2220 Single thread tracing
2221 ---------------------
2223 By writing into set_ftrace_pid you can trace a
2224 single thread. For example::
2226 # cat set_ftrace_pid
2228 # echo 3111 > set_ftrace_pid
2229 # cat set_ftrace_pid
2231 # echo function > current_tracer
2235 # TASK-PID CPU# TIMESTAMP FUNCTION
2237 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
2238 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
2239 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
2240 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
2241 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
2242 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
2243 # echo > set_ftrace_pid
2247 # TASK-PID CPU# TIMESTAMP FUNCTION
2249 ##### CPU 3 buffer started ####
2250 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
2251 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
2252 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
2253 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
2254 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
2256 If you want to trace a function when executing, you could use
2257 something like this simple program.
2262 #include <sys/types.h>
2263 #include <sys/stat.h>
2269 #define STR(x) _STR(x)
2270 #define MAX_PATH 256
2272 const char *find_tracefs(void)
2274 static char tracefs[MAX_PATH+1];
2275 static int tracefs_found;
2282 if ((fp = fopen("/proc/mounts","r")) == NULL) {
2283 perror("/proc/mounts");
2287 while (fscanf(fp, "%*s %"
2289 "s %99s %*s %*d %*d\n",
2290 tracefs, type) == 2) {
2291 if (strcmp(type, "tracefs") == 0)
2296 if (strcmp(type, "tracefs") != 0) {
2297 fprintf(stderr, "tracefs not mounted");
2301 strcat(tracefs, "/tracing/");
2307 const char *tracing_file(const char *file_name)
2309 static char trace_file[MAX_PATH+1];
2310 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name);
2314 int main (int argc, char **argv)
2324 ffd = open(tracing_file("current_tracer"), O_WRONLY);
2327 write(ffd, "nop", 3);
2329 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
2330 s = sprintf(line, "%d\n", getpid());
2333 write(ffd, "function", 8);
2338 execvp(argv[1], argv+1);
2344 Or this simple script!
2349 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts`
2350 echo nop > $tracefs/tracing/current_tracer
2351 echo 0 > $tracefs/tracing/tracing_on
2352 echo $$ > $tracefs/tracing/set_ftrace_pid
2353 echo function > $tracefs/tracing/current_tracer
2354 echo 1 > $tracefs/tracing/tracing_on
2358 function graph tracer
2359 ---------------------------
2361 This tracer is similar to the function tracer except that it
2362 probes a function on its entry and its exit. This is done by
2363 using a dynamically allocated stack of return addresses in each
2364 task_struct. On function entry the tracer overwrites the return
2365 address of each function traced to set a custom probe. Thus the
2366 original return address is stored on the stack of return address
2369 Probing on both ends of a function leads to special features
2372 - measure of a function's time execution
2373 - having a reliable call stack to draw function calls graph
2375 This tracer is useful in several situations:
2377 - you want to find the reason of a strange kernel behavior and
2378 need to see what happens in detail on any areas (or specific
2381 - you are experiencing weird latencies but it's difficult to
2384 - you want to find quickly which path is taken by a specific
2387 - you just want to peek inside a working kernel and want to see
2392 # tracer: function_graph
2394 # CPU DURATION FUNCTION CALLS
2398 0) | do_sys_open() {
2400 0) | kmem_cache_alloc() {
2401 0) 1.382 us | __might_sleep();
2403 0) | strncpy_from_user() {
2404 0) | might_fault() {
2405 0) 1.389 us | __might_sleep();
2410 0) 0.668 us | _spin_lock();
2411 0) 0.570 us | expand_files();
2412 0) 0.586 us | _spin_unlock();
2415 There are several columns that can be dynamically
2416 enabled/disabled. You can use every combination of options you
2417 want, depending on your needs.
2419 - The cpu number on which the function executed is default
2420 enabled. It is sometimes better to only trace one cpu (see
2421 tracing_cpu_mask file) or you might sometimes see unordered
2422 function calls while cpu tracing switch.
2424 - hide: echo nofuncgraph-cpu > trace_options
2425 - show: echo funcgraph-cpu > trace_options
2427 - The duration (function's time of execution) is displayed on
2428 the closing bracket line of a function or on the same line
2429 than the current function in case of a leaf one. It is default
2432 - hide: echo nofuncgraph-duration > trace_options
2433 - show: echo funcgraph-duration > trace_options
2435 - The overhead field precedes the duration field in case of
2436 reached duration thresholds.
2438 - hide: echo nofuncgraph-overhead > trace_options
2439 - show: echo funcgraph-overhead > trace_options
2440 - depends on: funcgraph-duration
2444 3) # 1837.709 us | } /* __switch_to */
2445 3) | finish_task_switch() {
2446 3) 0.313 us | _raw_spin_unlock_irq();
2448 3) # 1889.063 us | } /* __schedule */
2449 3) ! 140.417 us | } /* __schedule */
2450 3) # 2034.948 us | } /* schedule */
2451 3) * 33998.59 us | } /* schedule_preempt_disabled */
2455 1) 0.260 us | msecs_to_jiffies();
2456 1) 0.313 us | __rcu_read_unlock();
2459 1) 0.313 us | rcu_bh_qs();
2460 1) 0.313 us | __local_bh_enable();
2462 1) 0.365 us | idle_cpu();
2463 1) | rcu_irq_exit() {
2464 1) 0.417 us | rcu_eqs_enter_common.isra.47();
2468 1) @ 119760.2 us | }
2474 2) 0.417 us | scheduler_ipi();
2484 + means that the function exceeded 10 usecs.
2485 ! means that the function exceeded 100 usecs.
2486 # means that the function exceeded 1000 usecs.
2487 * means that the function exceeded 10 msecs.
2488 @ means that the function exceeded 100 msecs.
2489 $ means that the function exceeded 1 sec.
2492 - The task/pid field displays the thread cmdline and pid which
2493 executed the function. It is default disabled.
2495 - hide: echo nofuncgraph-proc > trace_options
2496 - show: echo funcgraph-proc > trace_options
2500 # tracer: function_graph
2502 # CPU TASK/PID DURATION FUNCTION CALLS
2504 0) sh-4802 | | d_free() {
2505 0) sh-4802 | | call_rcu() {
2506 0) sh-4802 | | __call_rcu() {
2507 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2508 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2509 0) sh-4802 | 2.899 us | }
2510 0) sh-4802 | 4.040 us | }
2511 0) sh-4802 | 5.151 us | }
2512 0) sh-4802 | + 49.370 us | }
2515 - The absolute time field is an absolute timestamp given by the
2516 system clock since it started. A snapshot of this time is
2517 given on each entry/exit of functions
2519 - hide: echo nofuncgraph-abstime > trace_options
2520 - show: echo funcgraph-abstime > trace_options
2525 # TIME CPU DURATION FUNCTION CALLS
2527 360.774522 | 1) 0.541 us | }
2528 360.774522 | 1) 4.663 us | }
2529 360.774523 | 1) 0.541 us | __wake_up_bit();
2530 360.774524 | 1) 6.796 us | }
2531 360.774524 | 1) 7.952 us | }
2532 360.774525 | 1) 9.063 us | }
2533 360.774525 | 1) 0.615 us | journal_mark_dirty();
2534 360.774527 | 1) 0.578 us | __brelse();
2535 360.774528 | 1) | reiserfs_prepare_for_journal() {
2536 360.774528 | 1) | unlock_buffer() {
2537 360.774529 | 1) | wake_up_bit() {
2538 360.774529 | 1) | bit_waitqueue() {
2539 360.774530 | 1) 0.594 us | __phys_addr();
2542 The function name is always displayed after the closing bracket
2543 for a function if the start of that function is not in the
2546 Display of the function name after the closing bracket may be
2547 enabled for functions whose start is in the trace buffer,
2548 allowing easier searching with grep for function durations.
2549 It is default disabled.
2551 - hide: echo nofuncgraph-tail > trace_options
2552 - show: echo funcgraph-tail > trace_options
2554 Example with nofuncgraph-tail (default)::
2557 0) | kmem_cache_free() {
2558 0) 0.518 us | __phys_addr();
2562 Example with funcgraph-tail::
2565 0) | kmem_cache_free() {
2566 0) 0.518 us | __phys_addr();
2567 0) 1.757 us | } /* kmem_cache_free() */
2568 0) 2.861 us | } /* putname() */
2570 You can put some comments on specific functions by using
2571 trace_printk() For example, if you want to put a comment inside
2572 the __might_sleep() function, you just have to include
2573 <linux/ftrace.h> and call trace_printk() inside __might_sleep()::
2575 trace_printk("I'm a comment!\n")
2579 1) | __might_sleep() {
2580 1) | /* I'm a comment! */
2584 You might find other useful features for this tracer in the
2585 following "dynamic ftrace" section such as tracing only specific
2591 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2592 virtually no overhead when function tracing is disabled. The way
2593 this works is the mcount function call (placed at the start of
2594 every kernel function, produced by the -pg switch in gcc),
2595 starts of pointing to a simple return. (Enabling FTRACE will
2596 include the -pg switch in the compiling of the kernel.)
2598 At compile time every C file object is run through the
2599 recordmcount program (located in the scripts directory). This
2600 program will parse the ELF headers in the C object to find all
2601 the locations in the .text section that call mcount. Starting
2602 with gcc version 4.6, the -mfentry has been added for x86, which
2603 calls "__fentry__" instead of "mcount". Which is called before
2604 the creation of the stack frame.
2606 Note, not all sections are traced. They may be prevented by either
2607 a notrace, or blocked another way and all inline functions are not
2608 traced. Check the "available_filter_functions" file to see what functions
2611 A section called "__mcount_loc" is created that holds
2612 references to all the mcount/fentry call sites in the .text section.
2613 The recordmcount program re-links this section back into the
2614 original object. The final linking stage of the kernel will add all these
2615 references into a single table.
2617 On boot up, before SMP is initialized, the dynamic ftrace code
2618 scans this table and updates all the locations into nops. It
2619 also records the locations, which are added to the
2620 available_filter_functions list. Modules are processed as they
2621 are loaded and before they are executed. When a module is
2622 unloaded, it also removes its functions from the ftrace function
2623 list. This is automatic in the module unload code, and the
2624 module author does not need to worry about it.
2626 When tracing is enabled, the process of modifying the function
2627 tracepoints is dependent on architecture. The old method is to use
2628 kstop_machine to prevent races with the CPUs executing code being
2629 modified (which can cause the CPU to do undesirable things, especially
2630 if the modified code crosses cache (or page) boundaries), and the nops are
2631 patched back to calls. But this time, they do not call mcount
2632 (which is just a function stub). They now call into the ftrace
2635 The new method of modifying the function tracepoints is to place
2636 a breakpoint at the location to be modified, sync all CPUs, modify
2637 the rest of the instruction not covered by the breakpoint. Sync
2638 all CPUs again, and then remove the breakpoint with the finished
2639 version to the ftrace call site.
2641 Some archs do not even need to monkey around with the synchronization,
2642 and can just slap the new code on top of the old without any
2643 problems with other CPUs executing it at the same time.
2645 One special side-effect to the recording of the functions being
2646 traced is that we can now selectively choose which functions we
2647 wish to trace and which ones we want the mcount calls to remain
2650 Two files are used, one for enabling and one for disabling the
2651 tracing of specified functions. They are:
2659 A list of available functions that you can add to these files is
2662 available_filter_functions
2666 # cat available_filter_functions
2675 If I am only interested in sys_nanosleep and hrtimer_interrupt::
2677 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2678 # echo function > current_tracer
2679 # echo 1 > tracing_on
2681 # echo 0 > tracing_on
2685 # entries-in-buffer/entries-written: 5/5 #P:4
2688 # / _----=> need-resched
2689 # | / _---=> hardirq/softirq
2690 # || / _--=> preempt-depth
2692 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2694 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2695 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2696 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2697 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2698 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2700 To see which functions are being traced, you can cat the file:
2703 # cat set_ftrace_filter
2708 Perhaps this is not enough. The filters also allow glob(7) matching.
2711 will match functions that begin with <match>
2713 will match functions that end with <match>
2715 will match functions that have <match> in it
2716 ``<match1>*<match2>``
2717 will match functions that begin with <match1> and end with <match2>
2720 It is better to use quotes to enclose the wild cards,
2721 otherwise the shell may expand the parameters into names
2722 of files in the local directory.
2726 # echo 'hrtimer_*' > set_ftrace_filter
2732 # entries-in-buffer/entries-written: 897/897 #P:4
2735 # / _----=> need-resched
2736 # | / _---=> hardirq/softirq
2737 # || / _--=> preempt-depth
2739 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2741 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2742 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2743 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2744 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2745 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2746 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2747 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2748 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2750 Notice that we lost the sys_nanosleep.
2753 # cat set_ftrace_filter
2758 hrtimer_try_to_cancel
2762 hrtimer_force_reprogram
2763 hrtimer_get_next_event
2767 hrtimer_get_remaining
2769 hrtimer_init_sleeper
2772 This is because the '>' and '>>' act just like they do in bash.
2773 To rewrite the filters, use '>'
2774 To append to the filters, use '>>'
2776 To clear out a filter so that all functions will be recorded
2779 # echo > set_ftrace_filter
2780 # cat set_ftrace_filter
2783 Again, now we want to append.
2787 # echo sys_nanosleep > set_ftrace_filter
2788 # cat set_ftrace_filter
2790 # echo 'hrtimer_*' >> set_ftrace_filter
2791 # cat set_ftrace_filter
2796 hrtimer_try_to_cancel
2800 hrtimer_force_reprogram
2801 hrtimer_get_next_event
2806 hrtimer_get_remaining
2808 hrtimer_init_sleeper
2811 The set_ftrace_notrace prevents those functions from being
2815 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2821 # entries-in-buffer/entries-written: 39608/39608 #P:4
2824 # / _----=> need-resched
2825 # | / _---=> hardirq/softirq
2826 # || / _--=> preempt-depth
2828 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2830 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2831 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2832 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2833 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2834 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2835 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2836 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2837 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2838 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2839 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2840 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2841 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2843 We can see that there's no more lock or preempt tracing.
2845 Selecting function filters via index
2846 ------------------------------------
2848 Because processing of strings is expensive (the address of the function
2849 needs to be looked up before comparing to the string being passed in),
2850 an index can be used as well to enable functions. This is useful in the
2851 case of setting thousands of specific functions at a time. By passing
2852 in a list of numbers, no string processing will occur. Instead, the function
2853 at the specific location in the internal array (which corresponds to the
2854 functions in the "available_filter_functions" file), is selected.
2858 # echo 1 > set_ftrace_filter
2860 Will select the first function listed in "available_filter_functions"
2864 # head -1 available_filter_functions
2865 trace_initcall_finish_cb
2867 # cat set_ftrace_filter
2868 trace_initcall_finish_cb
2870 # head -50 available_filter_functions | tail -1
2873 # echo 1 50 > set_ftrace_filter
2874 # cat set_ftrace_filter
2875 trace_initcall_finish_cb
2878 Dynamic ftrace with the function graph tracer
2879 ---------------------------------------------
2881 Although what has been explained above concerns both the
2882 function tracer and the function-graph-tracer, there are some
2883 special features only available in the function-graph tracer.
2885 If you want to trace only one function and all of its children,
2886 you just have to echo its name into set_graph_function::
2888 echo __do_fault > set_graph_function
2890 will produce the following "expanded" trace of the __do_fault()
2894 0) | filemap_fault() {
2895 0) | find_lock_page() {
2896 0) 0.804 us | find_get_page();
2897 0) | __might_sleep() {
2901 0) 0.653 us | _spin_lock();
2902 0) 0.578 us | page_add_file_rmap();
2903 0) 0.525 us | native_set_pte_at();
2904 0) 0.585 us | _spin_unlock();
2905 0) | unlock_page() {
2906 0) 0.541 us | page_waitqueue();
2907 0) 0.639 us | __wake_up_bit();
2911 0) | filemap_fault() {
2912 0) | find_lock_page() {
2913 0) 0.698 us | find_get_page();
2914 0) | __might_sleep() {
2918 0) 0.631 us | _spin_lock();
2919 0) 0.571 us | page_add_file_rmap();
2920 0) 0.526 us | native_set_pte_at();
2921 0) 0.586 us | _spin_unlock();
2922 0) | unlock_page() {
2923 0) 0.533 us | page_waitqueue();
2924 0) 0.638 us | __wake_up_bit();
2928 You can also expand several functions at once::
2930 echo sys_open > set_graph_function
2931 echo sys_close >> set_graph_function
2933 Now if you want to go back to trace all functions you can clear
2934 this special filter via::
2936 echo > set_graph_function
2942 Note, the proc sysctl ftrace_enable is a big on/off switch for the
2943 function tracer. By default it is enabled (when function tracing is
2944 enabled in the kernel). If it is disabled, all function tracing is
2945 disabled. This includes not only the function tracers for ftrace, but
2946 also for any other uses (perf, kprobes, stack tracing, profiling, etc).
2948 Please disable this with care.
2950 This can be disable (and enabled) with::
2952 sysctl kernel.ftrace_enabled=0
2953 sysctl kernel.ftrace_enabled=1
2957 echo 0 > /proc/sys/kernel/ftrace_enabled
2958 echo 1 > /proc/sys/kernel/ftrace_enabled
2964 A few commands are supported by the set_ftrace_filter interface.
2965 Trace commands have the following format::
2967 <function>:<command>:<parameter>
2969 The following commands are supported:
2972 This command enables function filtering per module. The
2973 parameter defines the module. For example, if only the write*
2974 functions in the ext3 module are desired, run:
2976 echo 'write*:mod:ext3' > set_ftrace_filter
2978 This command interacts with the filter in the same way as
2979 filtering based on function names. Thus, adding more functions
2980 in a different module is accomplished by appending (>>) to the
2981 filter file. Remove specific module functions by prepending
2984 echo '!writeback*:mod:ext3' >> set_ftrace_filter
2986 Mod command supports module globbing. Disable tracing for all
2987 functions except a specific module::
2989 echo '!*:mod:!ext3' >> set_ftrace_filter
2991 Disable tracing for all modules, but still trace kernel::
2993 echo '!*:mod:*' >> set_ftrace_filter
2995 Enable filter only for kernel::
2997 echo '*write*:mod:!*' >> set_ftrace_filter
2999 Enable filter for module globbing::
3001 echo '*write*:mod:*snd*' >> set_ftrace_filter
3004 These commands turn tracing on and off when the specified
3005 functions are hit. The parameter determines how many times the
3006 tracing system is turned on and off. If unspecified, there is
3007 no limit. For example, to disable tracing when a schedule bug
3008 is hit the first 5 times, run::
3010 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
3012 To always disable tracing when __schedule_bug is hit::
3014 echo '__schedule_bug:traceoff' > set_ftrace_filter
3016 These commands are cumulative whether or not they are appended
3017 to set_ftrace_filter. To remove a command, prepend it by '!'
3018 and drop the parameter::
3020 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
3022 The above removes the traceoff command for __schedule_bug
3023 that have a counter. To remove commands without counters::
3025 echo '!__schedule_bug:traceoff' > set_ftrace_filter
3028 Will cause a snapshot to be triggered when the function is hit.
3031 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
3033 To only snapshot once:
3036 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
3038 To remove the above commands::
3040 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
3041 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
3043 - enable_event/disable_event:
3044 These commands can enable or disable a trace event. Note, because
3045 function tracing callbacks are very sensitive, when these commands
3046 are registered, the trace point is activated, but disabled in
3047 a "soft" mode. That is, the tracepoint will be called, but
3048 just will not be traced. The event tracepoint stays in this mode
3049 as long as there's a command that triggers it.
3052 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
3057 <function>:enable_event:<system>:<event>[:count]
3058 <function>:disable_event:<system>:<event>[:count]
3060 To remove the events commands::
3062 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
3064 echo '!schedule:disable_event:sched:sched_switch' > \
3068 When the function is hit, it will dump the contents of the ftrace
3069 ring buffer to the console. This is useful if you need to debug
3070 something, and want to dump the trace when a certain function
3071 is hit. Perhaps it's a function that is called before a triple
3072 fault happens and does not allow you to get a regular dump.
3075 When the function is hit, it will dump the contents of the ftrace
3076 ring buffer for the current CPU to the console. Unlike the "dump"
3077 command, it only prints out the contents of the ring buffer for the
3078 CPU that executed the function that triggered the dump.
3081 When the function is hit, a stack trace is recorded.
3086 The trace_pipe outputs the same content as the trace file, but
3087 the effect on the tracing is different. Every read from
3088 trace_pipe is consumed. This means that subsequent reads will be
3089 different. The trace is live.
3092 # echo function > current_tracer
3093 # cat trace_pipe > /tmp/trace.out &
3095 # echo 1 > tracing_on
3097 # echo 0 > tracing_on
3101 # entries-in-buffer/entries-written: 0/0 #P:4
3104 # / _----=> need-resched
3105 # | / _---=> hardirq/softirq
3106 # || / _--=> preempt-depth
3108 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3112 # cat /tmp/trace.out
3113 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
3114 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
3115 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
3116 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
3117 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
3118 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
3119 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
3120 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
3121 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
3124 Note, reading the trace_pipe file will block until more input is
3130 Having too much or not enough data can be troublesome in
3131 diagnosing an issue in the kernel. The file buffer_size_kb is
3132 used to modify the size of the internal trace buffers. The
3133 number listed is the number of entries that can be recorded per
3134 CPU. To know the full size, multiply the number of possible CPUs
3135 with the number of entries.
3138 # cat buffer_size_kb
3139 1408 (units kilobytes)
3141 Or simply read buffer_total_size_kb
3144 # cat buffer_total_size_kb
3147 To modify the buffer, simple echo in a number (in 1024 byte segments).
3150 # echo 10000 > buffer_size_kb
3151 # cat buffer_size_kb
3152 10000 (units kilobytes)
3154 It will try to allocate as much as possible. If you allocate too
3155 much, it can cause Out-Of-Memory to trigger.
3158 # echo 1000000000000 > buffer_size_kb
3159 -bash: echo: write error: Cannot allocate memory
3160 # cat buffer_size_kb
3163 The per_cpu buffers can be changed individually as well:
3166 # echo 10000 > per_cpu/cpu0/buffer_size_kb
3167 # echo 100 > per_cpu/cpu1/buffer_size_kb
3169 When the per_cpu buffers are not the same, the buffer_size_kb
3170 at the top level will just show an X
3173 # cat buffer_size_kb
3176 This is where the buffer_total_size_kb is useful:
3179 # cat buffer_total_size_kb
3182 Writing to the top level buffer_size_kb will reset all the buffers
3183 to be the same again.
3187 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
3188 available to all non latency tracers. (Latency tracers which
3189 record max latency, such as "irqsoff" or "wakeup", can't use
3190 this feature, since those are already using the snapshot
3191 mechanism internally.)
3193 Snapshot preserves a current trace buffer at a particular point
3194 in time without stopping tracing. Ftrace swaps the current
3195 buffer with a spare buffer, and tracing continues in the new
3196 current (=previous spare) buffer.
3198 The following tracefs files in "tracing" are related to this
3203 This is used to take a snapshot and to read the output
3204 of the snapshot. Echo 1 into this file to allocate a
3205 spare buffer and to take a snapshot (swap), then read
3206 the snapshot from this file in the same format as
3207 "trace" (described above in the section "The File
3208 System"). Both reads snapshot and tracing are executable
3209 in parallel. When the spare buffer is allocated, echoing
3210 0 frees it, and echoing else (positive) values clear the
3212 More details are shown in the table below.
3214 +--------------+------------+------------+------------+
3215 |status\\input | 0 | 1 | else |
3216 +==============+============+============+============+
3217 |not allocated |(do nothing)| alloc+swap |(do nothing)|
3218 +--------------+------------+------------+------------+
3219 |allocated | free | swap | clear |
3220 +--------------+------------+------------+------------+
3222 Here is an example of using the snapshot feature.
3225 # echo 1 > events/sched/enable
3230 # entries-in-buffer/entries-written: 71/71 #P:8
3233 # / _----=> need-resched
3234 # | / _---=> hardirq/softirq
3235 # || / _--=> preempt-depth
3237 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3239 <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
3240 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
3242 <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
3247 # entries-in-buffer/entries-written: 77/77 #P:8
3250 # / _----=> need-resched
3251 # | / _---=> hardirq/softirq
3252 # || / _--=> preempt-depth
3254 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3256 <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
3257 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
3261 If you try to use this snapshot feature when current tracer is
3262 one of the latency tracers, you will get the following results.
3265 # echo wakeup > current_tracer
3267 bash: echo: write error: Device or resource busy
3269 cat: snapshot: Device or resource busy
3274 In the tracefs tracing directory is a directory called "instances".
3275 This directory can have new directories created inside of it using
3276 mkdir, and removing directories with rmdir. The directory created
3277 with mkdir in this directory will already contain files and other
3278 directories after it is created.
3281 # mkdir instances/foo
3283 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
3284 set_event snapshot trace trace_clock trace_marker trace_options
3285 trace_pipe tracing_on
3287 As you can see, the new directory looks similar to the tracing directory
3288 itself. In fact, it is very similar, except that the buffer and
3289 events are agnostic from the main director, or from any other
3290 instances that are created.
3292 The files in the new directory work just like the files with the
3293 same name in the tracing directory except the buffer that is used
3294 is a separate and new buffer. The files affect that buffer but do not
3295 affect the main buffer with the exception of trace_options. Currently,
3296 the trace_options affect all instances and the top level buffer
3297 the same, but this may change in future releases. That is, options
3298 may become specific to the instance they reside in.
3300 Notice that none of the function tracer files are there, nor is
3301 current_tracer and available_tracers. This is because the buffers
3302 can currently only have events enabled for them.
3305 # mkdir instances/foo
3306 # mkdir instances/bar
3307 # mkdir instances/zoot
3308 # echo 100000 > buffer_size_kb
3309 # echo 1000 > instances/foo/buffer_size_kb
3310 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
3311 # echo function > current_trace
3312 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
3313 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
3314 # echo 1 > instances/foo/events/sched/sched_switch/enable
3315 # echo 1 > instances/bar/events/irq/enable
3316 # echo 1 > instances/zoot/events/syscalls/enable
3318 CPU:2 [LOST 11745 EVENTS]
3319 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
3320 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
3321 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
3322 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
3323 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
3324 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
3325 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
3326 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
3327 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3328 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3329 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
3332 # cat instances/foo/trace_pipe
3333 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3334 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3335 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
3336 <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
3337 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
3338 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3339 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3340 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
3341 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
3342 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
3345 # cat instances/bar/trace_pipe
3346 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
3347 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
3348 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
3349 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
3350 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
3351 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
3352 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
3353 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
3354 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
3355 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
3356 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
3357 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
3360 # cat instances/zoot/trace
3363 # entries-in-buffer/entries-written: 18996/18996 #P:4
3366 # / _----=> need-resched
3367 # | / _---=> hardirq/softirq
3368 # || / _--=> preempt-depth
3370 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3372 bash-1998 [000] d... 140.733501: sys_write -> 0x2
3373 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
3374 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
3375 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
3376 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
3377 bash-1998 [000] d... 140.733510: sys_close(fd: a)
3378 bash-1998 [000] d... 140.733510: sys_close -> 0x0
3379 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
3380 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
3381 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
3382 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
3384 You can see that the trace of the top most trace buffer shows only
3385 the function tracing. The foo instance displays wakeups and task
3388 To remove the instances, simply delete their directories:
3391 # rmdir instances/foo
3392 # rmdir instances/bar
3393 # rmdir instances/zoot
3395 Note, if a process has a trace file open in one of the instance
3396 directories, the rmdir will fail with EBUSY.
3401 Since the kernel has a fixed sized stack, it is important not to
3402 waste it in functions. A kernel developer must be conscience of
3403 what they allocate on the stack. If they add too much, the system
3404 can be in danger of a stack overflow, and corruption will occur,
3405 usually leading to a system panic.
3407 There are some tools that check this, usually with interrupts
3408 periodically checking usage. But if you can perform a check
3409 at every function call that will become very useful. As ftrace provides
3410 a function tracer, it makes it convenient to check the stack size
3411 at every function call. This is enabled via the stack tracer.
3413 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
3414 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
3417 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
3419 You can also enable it from the kernel command line to trace
3420 the stack size of the kernel during boot up, by adding "stacktrace"
3421 to the kernel command line parameter.
3423 After running it for a few minutes, the output looks like:
3426 # cat stack_max_size
3430 Depth Size Location (18 entries)
3432 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
3433 1) 2704 160 find_busiest_group+0x31/0x1f1
3434 2) 2544 256 load_balance+0xd9/0x662
3435 3) 2288 80 idle_balance+0xbb/0x130
3436 4) 2208 128 __schedule+0x26e/0x5b9
3437 5) 2080 16 schedule+0x64/0x66
3438 6) 2064 128 schedule_timeout+0x34/0xe0
3439 7) 1936 112 wait_for_common+0x97/0xf1
3440 8) 1824 16 wait_for_completion+0x1d/0x1f
3441 9) 1808 128 flush_work+0xfe/0x119
3442 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
3443 11) 1664 48 input_available_p+0x1d/0x5c
3444 12) 1616 48 n_tty_poll+0x6d/0x134
3445 13) 1568 64 tty_poll+0x64/0x7f
3446 14) 1504 880 do_select+0x31e/0x511
3447 15) 624 400 core_sys_select+0x177/0x216
3448 16) 224 96 sys_select+0x91/0xb9
3449 17) 128 128 system_call_fastpath+0x16/0x1b
3451 Note, if -mfentry is being used by gcc, functions get traced before
3452 they set up the stack frame. This means that leaf level functions
3453 are not tested by the stack tracer when -mfentry is used.
3455 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
3459 More details can be found in the source code, in the `kernel/trace/*.c` files.