1 ========================
2 ftrace - Function Tracer
3 ========================
5 Copyright 2008 Red Hat Inc.
7 :Author: Steven Rostedt <srostedt@redhat.com>
8 :License: The GNU Free Documentation License, Version 1.2
9 (dual licensed under the GPL v2)
10 :Original Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton,
11 John Kacur, and David Teigland.
13 - Written for: 2.6.28-rc2
15 - Updated for: 4.13 - Copyright 2017 VMware Inc. Steven Rostedt
16 - Converted to rst format - Changbin Du <changbin.du@intel.com>
21 Ftrace is an internal tracer designed to help out developers and
22 designers of systems to find what is going on inside the kernel.
23 It can be used for debugging or analyzing latencies and
24 performance issues that take place outside of user-space.
26 Although ftrace is typically considered the function tracer, it
27 is really a frame work of several assorted tracing utilities.
28 There's latency tracing to examine what occurs between interrupts
29 disabled and enabled, as well as for preemption and from a time
30 a task is woken to the task is actually scheduled in.
32 One of the most common uses of ftrace is the event tracing.
33 Through out the kernel is hundreds of static event points that
34 can be enabled via the tracefs file system to see what is
35 going on in certain parts of the kernel.
37 See events.txt for more information.
40 Implementation Details
41 ----------------------
43 See :doc:`ftrace-design` for details for arch porters and such.
49 Ftrace uses the tracefs file system to hold the control files as
50 well as the files to display output.
52 When tracefs is configured into the kernel (which selecting any ftrace
53 option will do) the directory /sys/kernel/tracing will be created. To mount
54 this directory, you can add to your /etc/fstab file::
56 tracefs /sys/kernel/tracing tracefs defaults 0 0
58 Or you can mount it at run time with::
60 mount -t tracefs nodev /sys/kernel/tracing
62 For quicker access to that directory you may want to make a soft link to
65 ln -s /sys/kernel/tracing /tracing
69 Before 4.1, all ftrace tracing control files were within the debugfs
70 file system, which is typically located at /sys/kernel/debug/tracing.
71 For backward compatibility, when mounting the debugfs file system,
72 the tracefs file system will be automatically mounted at:
74 /sys/kernel/debug/tracing
76 All files located in the tracefs file system will be located in that
77 debugfs file system directory as well.
81 Any selected ftrace option will also create the tracefs file system.
82 The rest of the document will assume that you are in the ftrace directory
83 (cd /sys/kernel/tracing) and will only concentrate on the files within that
84 directory and not distract from the content with the extended
85 "/sys/kernel/tracing" path name.
87 That's it! (assuming that you have ftrace configured into your kernel)
89 After mounting tracefs you will have access to the control and output files
90 of ftrace. Here is a list of some of the key files:
93 Note: all time values are in microseconds.
97 This is used to set or display the current tracer
102 This holds the different types of tracers that
103 have been compiled into the kernel. The
104 tracers listed here can be configured by
105 echoing their name into current_tracer.
109 This sets or displays whether writing to the trace
110 ring buffer is enabled. Echo 0 into this file to disable
111 the tracer or 1 to enable it. Note, this only disables
112 writing to the ring buffer, the tracing overhead may
115 The kernel function tracing_off() can be used within the
116 kernel to disable writing to the ring buffer, which will
117 set this file to "0". User space can re-enable tracing by
118 echoing "1" into the file.
120 Note, the function and event trigger "traceoff" will also
121 set this file to zero and stop tracing. Which can also
122 be re-enabled by user space using this file.
126 This file holds the output of the trace in a human
127 readable format (described below). Note, tracing is temporarily
128 disabled while this file is being read (opened).
132 The output is the same as the "trace" file but this
133 file is meant to be streamed with live tracing.
134 Reads from this file will block until new data is
135 retrieved. Unlike the "trace" file, this file is a
136 consumer. This means reading from this file causes
137 sequential reads to display more current data. Once
138 data is read from this file, it is consumed, and
139 will not be read again with a sequential read. The
140 "trace" file is static, and if the tracer is not
141 adding more data, it will display the same
142 information every time it is read. This file will not
143 disable tracing while being read.
147 This file lets the user control the amount of data
148 that is displayed in one of the above output
149 files. Options also exist to modify how a tracer
150 or events work (stack traces, timestamps, etc).
154 This is a directory that has a file for every available
155 trace option (also in trace_options). Options may also be set
156 or cleared by writing a "1" or "0" respectively into the
157 corresponding file with the option name.
161 Some of the tracers record the max latency.
162 For example, the maximum time that interrupts are disabled.
163 The maximum time is saved in this file. The max trace will also be
164 stored, and displayed by "trace". A new max trace will only be
165 recorded if the latency is greater than the value in this file
168 By echoing in a time into this file, no latency will be recorded
169 unless it is greater than the time in this file.
173 Some latency tracers will record a trace whenever the
174 latency is greater than the number in this file.
175 Only active when the file contains a number greater than 0.
180 This sets or displays the number of kilobytes each CPU
181 buffer holds. By default, the trace buffers are the same size
182 for each CPU. The displayed number is the size of the
183 CPU buffer and not total size of all buffers. The
184 trace buffers are allocated in pages (blocks of memory
185 that the kernel uses for allocation, usually 4 KB in size).
186 If the last page allocated has room for more bytes
187 than requested, the rest of the page will be used,
188 making the actual allocation bigger than requested or shown.
189 ( Note, the size may not be a multiple of the page size
190 due to buffer management meta-data. )
192 Buffer sizes for individual CPUs may vary
193 (see "per_cpu/cpu0/buffer_size_kb" below), and if they do
194 this file will show "X".
196 buffer_total_size_kb:
198 This displays the total combined size of all the trace buffers.
202 If a process is performing tracing, and the ring buffer should be
203 shrunk "freed" when the process is finished, even if it were to be
204 killed by a signal, this file can be used for that purpose. On close
205 of this file, the ring buffer will be resized to its minimum size.
206 Having a process that is tracing also open this file, when the process
207 exits its file descriptor for this file will be closed, and in doing so,
208 the ring buffer will be "freed".
210 It may also stop tracing if disable_on_free option is set.
214 This is a mask that lets the user only trace on specified CPUs.
215 The format is a hex string representing the CPUs.
219 When dynamic ftrace is configured in (see the
220 section below "dynamic ftrace"), the code is dynamically
221 modified (code text rewrite) to disable calling of the
222 function profiler (mcount). This lets tracing be configured
223 in with practically no overhead in performance. This also
224 has a side effect of enabling or disabling specific functions
225 to be traced. Echoing names of functions into this file
226 will limit the trace to only those functions.
227 This influences the tracers "function" and "function_graph"
228 and thus also function profiling (see "function_profile_enabled").
230 The functions listed in "available_filter_functions" are what
231 can be written into this file.
233 This interface also allows for commands to be used. See the
234 "Filter commands" section for more details.
238 This has an effect opposite to that of
239 set_ftrace_filter. Any function that is added here will not
240 be traced. If a function exists in both set_ftrace_filter
241 and set_ftrace_notrace, the function will _not_ be traced.
245 Have the function tracer only trace the threads whose PID are
248 If the "function-fork" option is set, then when a task whose
249 PID is listed in this file forks, the child's PID will
250 automatically be added to this file, and the child will be
251 traced by the function tracer as well. This option will also
252 cause PIDs of tasks that exit to be removed from the file.
256 Have the events only trace a task with a PID listed in this file.
257 Note, sched_switch and sched_wake_up will also trace events
260 To have the PIDs of children of tasks with their PID in this file
261 added on fork, enable the "event-fork" option. That option will also
262 cause the PIDs of tasks to be removed from this file when the task
267 Functions listed in this file will cause the function graph
268 tracer to only trace these functions and the functions that
269 they call. (See the section "dynamic ftrace" for more details).
270 Note, set_ftrace_filter and set_ftrace_notrace still affects
271 what functions are being traced.
275 Similar to set_graph_function, but will disable function graph
276 tracing when the function is hit until it exits the function.
277 This makes it possible to ignore tracing functions that are called
278 by a specific function.
280 available_filter_functions:
282 This lists the functions that ftrace has processed and can trace.
283 These are the function names that you can pass to
284 "set_ftrace_filter", "set_ftrace_notrace",
285 "set_graph_function", or "set_graph_notrace".
286 (See the section "dynamic ftrace" below for more details.)
288 dyn_ftrace_total_info:
290 This file is for debugging purposes. The number of functions that
291 have been converted to nops and are available to be traced.
295 This file is more for debugging ftrace, but can also be useful
296 in seeing if any function has a callback attached to it.
297 Not only does the trace infrastructure use ftrace function
298 trace utility, but other subsystems might too. This file
299 displays all functions that have a callback attached to them
300 as well as the number of callbacks that have been attached.
301 Note, a callback may also call multiple functions which will
302 not be listed in this count.
304 If the callback registered to be traced by a function with
305 the "save regs" attribute (thus even more overhead), a 'R'
306 will be displayed on the same line as the function that
307 is returning registers.
309 If the callback registered to be traced by a function with
310 the "ip modify" attribute (thus the regs->ip can be changed),
311 an 'I' will be displayed on the same line as the function that
314 If the architecture supports it, it will also show what callback
315 is being directly called by the function. If the count is greater
316 than 1 it most likely will be ftrace_ops_list_func().
318 If the callback of the function jumps to a trampoline that is
319 specific to a the callback and not the standard trampoline,
320 its address will be printed as well as the function that the
323 function_profile_enabled:
325 When set it will enable all functions with either the function
326 tracer, or if configured, the function graph tracer. It will
327 keep a histogram of the number of functions that were called
328 and if the function graph tracer was configured, it will also keep
329 track of the time spent in those functions. The histogram
330 content can be displayed in the files:
332 trace_stats/function<cpu> ( function0, function1, etc).
336 A directory that holds different tracing stats.
340 Enable dynamic trace points. See kprobetrace.txt.
344 Dynamic trace points stats. See kprobetrace.txt.
348 Used with the function graph tracer. This is the max depth
349 it will trace into a function. Setting this to a value of
350 one will show only the first kernel function that is called
355 This is for tools that read the raw format files. If an event in
356 the ring buffer references a string, only a pointer to the string
357 is recorded into the buffer and not the string itself. This prevents
358 tools from knowing what that string was. This file displays the string
359 and address for the string allowing tools to map the pointers to what
364 Only the pid of the task is recorded in a trace event unless
365 the event specifically saves the task comm as well. Ftrace
366 makes a cache of pid mappings to comms to try to display
367 comms for events. If a pid for a comm is not listed, then
368 "<...>" is displayed in the output.
370 If the option "record-cmd" is set to "0", then comms of tasks
371 will not be saved during recording. By default, it is enabled.
375 By default, 128 comms are saved (see "saved_cmdlines" above). To
376 increase or decrease the amount of comms that are cached, echo
377 in a the number of comms to cache, into this file.
381 If the option "record-tgid" is set, on each scheduling context switch
382 the Task Group ID of a task is saved in a table mapping the PID of
383 the thread to its TGID. By default, the "record-tgid" option is
388 This displays the "snapshot" buffer and also lets the user
389 take a snapshot of the current running trace.
390 See the "Snapshot" section below for more details.
394 When the stack tracer is activated, this will display the
395 maximum stack size it has encountered.
396 See the "Stack Trace" section below.
400 This displays the stack back trace of the largest stack
401 that was encountered when the stack tracer is activated.
402 See the "Stack Trace" section below.
406 This is similar to "set_ftrace_filter" but it limits what
407 functions the stack tracer will check.
411 Whenever an event is recorded into the ring buffer, a
412 "timestamp" is added. This stamp comes from a specified
413 clock. By default, ftrace uses the "local" clock. This
414 clock is very fast and strictly per cpu, but on some
415 systems it may not be monotonic with respect to other
416 CPUs. In other words, the local clocks may not be in sync
417 with local clocks on other CPUs.
419 Usual clocks for tracing::
422 [local] global counter x86-tsc
424 The clock with the square brackets around it is the one in effect.
427 Default clock, but may not be in sync across CPUs
430 This clock is in sync with all CPUs but may
431 be a bit slower than the local clock.
434 This is not a clock at all, but literally an atomic
435 counter. It counts up one by one, but is in sync
436 with all CPUs. This is useful when you need to
437 know exactly the order events occurred with respect to
438 each other on different CPUs.
441 This uses the jiffies counter and the time stamp
442 is relative to the time since boot up.
445 This makes ftrace use the same clock that perf uses.
446 Eventually perf will be able to read ftrace buffers
447 and this will help out in interleaving the data.
450 Architectures may define their own clocks. For
451 example, x86 uses its own TSC cycle clock here.
454 This uses the powerpc timebase register value.
455 This is in sync across CPUs and can also be used
456 to correlate events across hypervisor/guest if
460 This uses the fast monotonic clock (CLOCK_MONOTONIC)
461 which is monotonic and is subject to NTP rate adjustments.
464 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
465 which is montonic but is not subject to any rate adjustments
466 and ticks at the same rate as the hardware clocksource.
469 This is the boot clock (CLOCK_BOOTTIME) and is based on the
470 fast monotonic clock, but also accounts for time spent in
471 suspend. Since the clock access is designed for use in
472 tracing in the suspend path, some side effects are possible
473 if clock is accessed after the suspend time is accounted before
474 the fast mono clock is updated. In this case, the clock update
475 appears to happen slightly sooner than it normally would have.
476 Also on 32-bit systems, it's possible that the 64-bit boot offset
477 sees a partial update. These effects are rare and post
478 processing should be able to handle them. See comments in the
479 ktime_get_boot_fast_ns() function for more information.
481 To set a clock, simply echo the clock name into this file::
483 # echo global > trace_clock
487 This is a very useful file for synchronizing user space
488 with events happening in the kernel. Writing strings into
489 this file will be written into the ftrace buffer.
491 It is useful in applications to open this file at the start
492 of the application and just reference the file descriptor
495 void trace_write(const char *fmt, ...)
505 n = vsnprintf(buf, 256, fmt, ap);
508 write(trace_fd, buf, n);
513 trace_fd = open("trace_marker", WR_ONLY);
517 This is similar to trace_marker above, but is meant for for binary data
518 to be written to it, where a tool can be used to parse the data
523 Add dynamic tracepoints in programs.
528 Uprobe statistics. See uprobetrace.txt
532 This is a way to make multiple trace buffers where different
533 events can be recorded in different buffers.
534 See "Instances" section below.
538 This is the trace event directory. It holds event tracepoints
539 (also known as static tracepoints) that have been compiled
540 into the kernel. It shows what event tracepoints exist
541 and how they are grouped by system. There are "enable"
542 files at various levels that can enable the tracepoints
543 when a "1" is written to them.
545 See events.txt for more information.
549 By echoing in the event into this file, will enable that event.
551 See events.txt for more information.
555 A list of events that can be enabled in tracing.
557 See events.txt for more information.
561 Certain tracers may change the timestamp mode used when
562 logging trace events into the event buffer. Events with
563 different modes can coexist within a buffer but the mode in
564 effect when an event is logged determines which timestamp mode
565 is used for that event. The default timestamp mode is
568 Usual timestamp modes for tracing:
573 The timestamp mode with the square brackets around it is the
576 delta: Default timestamp mode - timestamp is a delta against
577 a per-buffer timestamp.
579 absolute: The timestamp is a full timestamp, not a delta
580 against some other value. As such it takes up more
581 space and is less efficient.
585 Directory for the Hardware Latency Detector.
586 See "Hardware Latency Detector" section below.
590 This is a directory that contains the trace per_cpu information.
592 per_cpu/cpu0/buffer_size_kb:
594 The ftrace buffer is defined per_cpu. That is, there's a separate
595 buffer for each CPU to allow writes to be done atomically,
596 and free from cache bouncing. These buffers may have different
597 size buffers. This file is similar to the buffer_size_kb
598 file, but it only displays or sets the buffer size for the
599 specific CPU. (here cpu0).
603 This is similar to the "trace" file, but it will only display
604 the data specific for the CPU. If written to, it only clears
605 the specific CPU buffer.
607 per_cpu/cpu0/trace_pipe
609 This is similar to the "trace_pipe" file, and is a consuming
610 read, but it will only display (and consume) the data specific
613 per_cpu/cpu0/trace_pipe_raw
615 For tools that can parse the ftrace ring buffer binary format,
616 the trace_pipe_raw file can be used to extract the data
617 from the ring buffer directly. With the use of the splice()
618 system call, the buffer data can be quickly transferred to
619 a file or to the network where a server is collecting the
622 Like trace_pipe, this is a consuming reader, where multiple
623 reads will always produce different data.
625 per_cpu/cpu0/snapshot:
627 This is similar to the main "snapshot" file, but will only
628 snapshot the current CPU (if supported). It only displays
629 the content of the snapshot for a given CPU, and if
630 written to, only clears this CPU buffer.
632 per_cpu/cpu0/snapshot_raw:
634 Similar to the trace_pipe_raw, but will read the binary format
635 from the snapshot buffer for the given CPU.
639 This displays certain stats about the ring buffer:
642 The number of events that are still in the buffer.
645 The number of lost events due to overwriting when
649 Should always be zero.
650 This gets set if so many events happened within a nested
651 event (ring buffer is re-entrant), that it fills the
652 buffer and starts dropping events.
655 Bytes actually read (not overwritten).
658 The oldest timestamp in the buffer
661 The current timestamp
664 Events lost due to overwrite option being off.
667 The number of events read.
672 Here is the list of current tracers that may be configured.
676 Function call tracer to trace all kernel functions.
680 Similar to the function tracer except that the
681 function tracer probes the functions on their entry
682 whereas the function graph tracer traces on both entry
683 and exit of the functions. It then provides the ability
684 to draw a graph of function calls similar to C code
689 The block tracer. The tracer used by the blktrace user
694 The Hardware Latency tracer is used to detect if the hardware
695 produces any latency. See "Hardware Latency Detector" section
700 Traces the areas that disable interrupts and saves
701 the trace with the longest max latency.
702 See tracing_max_latency. When a new max is recorded,
703 it replaces the old trace. It is best to view this
704 trace with the latency-format option enabled, which
705 happens automatically when the tracer is selected.
709 Similar to irqsoff but traces and records the amount of
710 time for which preemption is disabled.
714 Similar to irqsoff and preemptoff, but traces and
715 records the largest time for which irqs and/or preemption
720 Traces and records the max latency that it takes for
721 the highest priority task to get scheduled after
722 it has been woken up.
723 Traces all tasks as an average developer would expect.
727 Traces and records the max latency that it takes for just
728 RT tasks (as the current "wakeup" does). This is useful
729 for those interested in wake up timings of RT tasks.
733 Traces and records the max latency that it takes for
734 a SCHED_DEADLINE task to be woken (as the "wakeup" and
739 A special tracer that is used to trace binary module.
740 It will trace all the calls that a module makes to the
741 hardware. Everything it writes and reads from the I/O
746 This tracer can be configured when tracing likely/unlikely
747 calls within the kernel. It will trace when a likely and
748 unlikely branch is hit and if it was correct in its prediction
753 This is the "trace nothing" tracer. To remove all
754 tracers from tracing simply echo "nop" into
758 Examples of using the tracer
759 ----------------------------
761 Here are typical examples of using the tracers when controlling
762 them only with the tracefs interface (without using any
763 user-land utilities).
768 Here is an example of the output format of the file "trace"::
772 # entries-in-buffer/entries-written: 140080/250280 #P:4
775 # / _----=> need-resched
776 # | / _---=> hardirq/softirq
777 # || / _--=> preempt-depth
779 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
781 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
782 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
783 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
784 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
785 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
786 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
787 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
788 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
789 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
790 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
793 A header is printed with the tracer name that is represented by
794 the trace. In this case the tracer is "function". Then it shows the
795 number of events in the buffer as well as the total number of entries
796 that were written. The difference is the number of entries that were
797 lost due to the buffer filling up (250280 - 140080 = 110200 events
800 The header explains the content of the events. Task name "bash", the task
801 PID "1977", the CPU that it was running on "000", the latency format
802 (explained below), the timestamp in <secs>.<usecs> format, the
803 function name that was traced "sys_close" and the parent function that
804 called this function "system_call_fastpath". The timestamp is the time
805 at which the function was entered.
810 When the latency-format option is enabled or when one of the latency
811 tracers is set, the trace file gives somewhat more information to see
812 why a latency happened. Here is a typical trace::
816 # irqsoff latency trace v1.1.5 on 3.8.0-test+
817 # --------------------------------------------------------------------
818 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
820 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
822 # => started at: __lock_task_sighand
823 # => ended at: _raw_spin_unlock_irqrestore
827 # / _-----=> irqs-off
828 # | / _----=> need-resched
829 # || / _---=> hardirq/softirq
830 # ||| / _--=> preempt-depth
832 # cmd pid ||||| time | caller
834 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
835 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
836 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
837 ps-6143 2d..1 306us : <stack trace>
838 => trace_hardirqs_on_caller
840 => _raw_spin_unlock_irqrestore
847 => system_call_fastpath
850 This shows that the current tracer is "irqsoff" tracing the time
851 for which interrupts were disabled. It gives the trace version (which
852 never changes) and the version of the kernel upon which this was executed on
853 (3.8). Then it displays the max latency in microseconds (259 us). The number
854 of trace entries displayed and the total number (both are four: #4/4).
855 VP, KP, SP, and HP are always zero and are reserved for later use.
856 #P is the number of online CPUs (#P:4).
858 The task is the process that was running when the latency
859 occurred. (ps pid: 6143).
861 The start and stop (the functions in which the interrupts were
862 disabled and enabled respectively) that caused the latencies:
864 - __lock_task_sighand is where the interrupts were disabled.
865 - _raw_spin_unlock_irqrestore is where they were enabled again.
867 The next lines after the header are the trace itself. The header
868 explains which is which.
870 cmd: The name of the process in the trace.
872 pid: The PID of that process.
874 CPU#: The CPU which the process was running on.
876 irqs-off: 'd' interrupts are disabled. '.' otherwise.
877 .. caution:: If the architecture does not support a way to
878 read the irq flags variable, an 'X' will always
882 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
883 - 'n' only TIF_NEED_RESCHED is set,
884 - 'p' only PREEMPT_NEED_RESCHED is set,
888 - 'Z' - NMI occurred inside a hardirq
889 - 'z' - NMI is running
890 - 'H' - hard irq occurred inside a softirq.
891 - 'h' - hard irq is running
892 - 's' - soft irq is running
893 - '.' - normal context.
895 preempt-depth: The level of preempt_disabled
897 The above is mostly meaningful for kernel developers.
900 When the latency-format option is enabled, the trace file
901 output includes a timestamp relative to the start of the
902 trace. This differs from the output when latency-format
903 is disabled, which includes an absolute timestamp.
906 This is just to help catch your eye a bit better. And
907 needs to be fixed to be only relative to the same CPU.
908 The marks are determined by the difference between this
909 current trace and the next trace.
911 - '$' - greater than 1 second
912 - '@' - greater than 100 milisecond
913 - '*' - greater than 10 milisecond
914 - '#' - greater than 1000 microsecond
915 - '!' - greater than 100 microsecond
916 - '+' - greater than 10 microsecond
917 - ' ' - less than or equal to 10 microsecond.
919 The rest is the same as the 'trace' file.
921 Note, the latency tracers will usually end with a back trace
922 to easily find where the latency occurred.
927 The trace_options file (or the options directory) is used to control
928 what gets printed in the trace output, or manipulate the tracers.
929 To see what is available, simply cat the file::
960 To disable one of the options, echo in the option prepended with
963 echo noprint-parent > trace_options
965 To enable an option, leave off the "no"::
967 echo sym-offset > trace_options
969 Here are the available options:
972 On function traces, display the calling (parent)
973 function as well as the function being traced.
977 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
980 bash-4000 [01] 1477.606694: simple_strtoul
984 Display not only the function name, but also the
985 offset in the function. For example, instead of
986 seeing just "ktime_get", you will see
987 "ktime_get+0xb/0x20".
991 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
994 This will also display the function address as well
995 as the function name.
999 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
1002 This deals with the trace file when the
1003 latency-format option is enabled.
1006 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
1007 (+0.000ms): simple_strtoul (kstrtoul)
1010 This will display raw numbers. This option is best for
1011 use with user applications that can translate the raw
1012 numbers better than having it done in the kernel.
1015 Similar to raw, but the numbers will be in a hexadecimal format.
1018 This will print out the formats in raw binary.
1021 When set, reading trace_pipe will not block when polled.
1024 Can disable trace_printk() from writing into the buffer.
1027 It is sometimes confusing when the CPU buffers are full
1028 and one CPU buffer had a lot of events recently, thus
1029 a shorter time frame, were another CPU may have only had
1030 a few events, which lets it have older events. When
1031 the trace is reported, it shows the oldest events first,
1032 and it may look like only one CPU ran (the one with the
1033 oldest events). When the annotate option is set, it will
1034 display when a new CPU buffer started::
1036 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
1037 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
1038 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
1039 ##### CPU 2 buffer started ####
1040 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
1041 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
1042 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
1045 This option changes the trace. It records a
1046 stacktrace of the current user space thread after
1050 when user stacktrace are enabled, look up which
1051 object the address belongs to, and print a
1052 relative address. This is especially useful when
1053 ASLR is on, otherwise you don't get a chance to
1054 resolve the address to object/file/line after
1055 the app is no longer running
1057 The lookup is performed when you read
1058 trace,trace_pipe. Example::
1060 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
1061 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
1065 When set, trace_printk()s will only show the format
1066 and not their parameters (if trace_bprintk() or
1067 trace_bputs() was used to save the trace_printk()).
1070 Show only the event data. Hides the comm, PID,
1071 timestamp, CPU, and other useful data.
1074 This option changes the trace output. When it is enabled,
1075 the trace displays additional information about the
1076 latency, as described in "Latency trace format".
1079 When any event or tracer is enabled, a hook is enabled
1080 in the sched_switch trace point to fill comm cache
1081 with mapped pids and comms. But this may cause some
1082 overhead, and if you only care about pids, and not the
1083 name of the task, disabling this option can lower the
1084 impact of tracing. See "saved_cmdlines".
1087 When any event or tracer is enabled, a hook is enabled
1088 in the sched_switch trace point to fill the cache of
1089 mapped Thread Group IDs (TGID) mapping to pids. See
1093 This controls what happens when the trace buffer is
1094 full. If "1" (default), the oldest events are
1095 discarded and overwritten. If "0", then the newest
1096 events are discarded.
1097 (see per_cpu/cpu0/stats for overrun and dropped)
1100 When the free_buffer is closed, tracing will
1101 stop (tracing_on set to 0).
1104 Shows the interrupt, preempt count, need resched data.
1105 When disabled, the trace looks like::
1109 # entries-in-buffer/entries-written: 144405/9452052 #P:4
1111 # TASK-PID CPU# TIMESTAMP FUNCTION
1113 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
1114 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
1115 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
1119 When set, the trace_marker is writable (only by root).
1120 When disabled, the trace_marker will error with EINVAL
1124 When set, tasks with PIDs listed in set_event_pid will have
1125 the PIDs of their children added to set_event_pid when those
1126 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
1127 their PIDs will be removed from the file.
1130 The latency tracers will enable function tracing
1131 if this option is enabled (default it is). When
1132 it is disabled, the latency tracers do not trace
1133 functions. This keeps the overhead of the tracer down
1134 when performing latency tests.
1137 When set, tasks with PIDs listed in set_ftrace_pid will
1138 have the PIDs of their children added to set_ftrace_pid
1139 when those tasks fork. Also, when tasks with PIDs in
1140 set_ftrace_pid exit, their PIDs will be removed from the
1144 When set, the latency tracers (irqsoff, wakeup, etc) will
1145 use function graph tracing instead of function tracing.
1148 When set, a stack trace is recorded after any trace event
1152 Enable branch tracing with the tracer. This enables branch
1153 tracer along with the currently set tracer. Enabling this
1154 with the "nop" tracer is the same as just enabling the
1157 .. tip:: Some tracers have their own options. They only appear in this
1158 file when the tracer is active. They always appear in the
1162 Here are the per tracer options:
1164 Options for function tracer:
1167 When set, a stack trace is recorded after every
1168 function that is recorded. NOTE! Limit the functions
1169 that are recorded before enabling this, with
1170 "set_ftrace_filter" otherwise the system performance
1171 will be critically degraded. Remember to disable
1172 this option before clearing the function filter.
1174 Options for function_graph tracer:
1176 Since the function_graph tracer has a slightly different output
1177 it has its own options to control what is displayed.
1180 When set, the "overrun" of the graph stack is
1181 displayed after each function traced. The
1182 overrun, is when the stack depth of the calls
1183 is greater than what is reserved for each task.
1184 Each task has a fixed array of functions to
1185 trace in the call graph. If the depth of the
1186 calls exceeds that, the function is not traced.
1187 The overrun is the number of functions missed
1188 due to exceeding this array.
1191 When set, the CPU number of the CPU where the trace
1192 occurred is displayed.
1195 When set, if the function takes longer than
1196 A certain amount, then a delay marker is
1197 displayed. See "delay" above, under the
1201 Unlike other tracers, the process' command line
1202 is not displayed by default, but instead only
1203 when a task is traced in and out during a context
1204 switch. Enabling this options has the command
1205 of each process displayed at every line.
1208 At the end of each function (the return)
1209 the duration of the amount of time in the
1210 function is displayed in microseconds.
1213 When set, the timestamp is displayed at each line.
1216 When disabled, functions that happen inside an
1217 interrupt will not be traced.
1220 When set, the return event will include the function
1221 that it represents. By default this is off, and
1222 only a closing curly bracket "}" is displayed for
1223 the return of a function.
1226 When running function graph tracer, to include
1227 the time a task schedules out in its function.
1228 When enabled, it will account time the task has been
1229 scheduled out as part of the function call.
1232 When running function profiler with function graph tracer,
1233 to include the time to call nested functions. When this is
1234 not set, the time reported for the function will only
1235 include the time the function itself executed for, not the
1236 time for functions that it called.
1238 Options for blk tracer:
1241 Shows a more minimalistic output.
1247 When interrupts are disabled, the CPU can not react to any other
1248 external event (besides NMIs and SMIs). This prevents the timer
1249 interrupt from triggering or the mouse interrupt from letting
1250 the kernel know of a new mouse event. The result is a latency
1251 with the reaction time.
1253 The irqsoff tracer tracks the time for which interrupts are
1254 disabled. When a new maximum latency is hit, the tracer saves
1255 the trace leading up to that latency point so that every time a
1256 new maximum is reached, the old saved trace is discarded and the
1259 To reset the maximum, echo 0 into tracing_max_latency. Here is
1262 # echo 0 > options/function-trace
1263 # echo irqsoff > current_tracer
1264 # echo 1 > tracing_on
1265 # echo 0 > tracing_max_latency
1268 # echo 0 > tracing_on
1272 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1273 # --------------------------------------------------------------------
1274 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1276 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
1278 # => started at: run_timer_softirq
1279 # => ended at: run_timer_softirq
1283 # / _-----=> irqs-off
1284 # | / _----=> need-resched
1285 # || / _---=> hardirq/softirq
1286 # ||| / _--=> preempt-depth
1288 # cmd pid ||||| time | caller
1290 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
1291 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
1292 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
1293 <idle>-0 0dNs3 25us : <stack trace>
1294 => _raw_spin_unlock_irq
1295 => run_timer_softirq
1300 => smp_apic_timer_interrupt
1301 => apic_timer_interrupt
1306 => x86_64_start_reservations
1307 => x86_64_start_kernel
1309 Here we see that that we had a latency of 16 microseconds (which is
1310 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
1311 interrupts. The difference between the 16 and the displayed
1312 timestamp 25us occurred because the clock was incremented
1313 between the time of recording the max latency and the time of
1314 recording the function that had that latency.
1316 Note the above example had function-trace not set. If we set
1317 function-trace, we get a much larger output::
1319 with echo 1 > options/function-trace
1323 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1324 # --------------------------------------------------------------------
1325 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1327 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1329 # => started at: ata_scsi_queuecmd
1330 # => ended at: ata_scsi_queuecmd
1334 # / _-----=> irqs-off
1335 # | / _----=> need-resched
1336 # || / _---=> hardirq/softirq
1337 # ||| / _--=> preempt-depth
1339 # cmd pid ||||| time | caller
1341 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1342 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1343 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1344 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1345 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1346 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1347 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1348 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1349 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1351 bash-2042 3d..1 67us : delay_tsc <-__delay
1352 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1353 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1354 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1355 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1356 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1357 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1358 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1359 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1360 bash-2042 3d..1 120us : <stack trace>
1361 => _raw_spin_unlock_irqrestore
1362 => ata_scsi_queuecmd
1363 => scsi_dispatch_cmd
1365 => __blk_run_queue_uncond
1368 => generic_make_request
1371 => __ext3_get_inode_loc
1380 => user_path_at_empty
1385 => system_call_fastpath
1388 Here we traced a 71 microsecond latency. But we also see all the
1389 functions that were called during that time. Note that by
1390 enabling function tracing, we incur an added overhead. This
1391 overhead may extend the latency times. But nevertheless, this
1392 trace has provided some very helpful debugging information.
1398 When preemption is disabled, we may be able to receive
1399 interrupts but the task cannot be preempted and a higher
1400 priority task must wait for preemption to be enabled again
1401 before it can preempt a lower priority task.
1403 The preemptoff tracer traces the places that disable preemption.
1404 Like the irqsoff tracer, it records the maximum latency for
1405 which preemption was disabled. The control of preemptoff tracer
1406 is much like the irqsoff tracer.
1409 # echo 0 > options/function-trace
1410 # echo preemptoff > current_tracer
1411 # echo 1 > tracing_on
1412 # echo 0 > tracing_max_latency
1415 # echo 0 > tracing_on
1417 # tracer: preemptoff
1419 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1420 # --------------------------------------------------------------------
1421 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1423 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1425 # => started at: do_IRQ
1426 # => ended at: do_IRQ
1430 # / _-----=> irqs-off
1431 # | / _----=> need-resched
1432 # || / _---=> hardirq/softirq
1433 # ||| / _--=> preempt-depth
1435 # cmd pid ||||| time | caller
1437 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1438 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1439 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1440 sshd-1991 1d..1 52us : <stack trace>
1441 => sub_preempt_count
1447 This has some more changes. Preemption was disabled when an
1448 interrupt came in (notice the 'h'), and was enabled on exit.
1449 But we also see that interrupts have been disabled when entering
1450 the preempt off section and leaving it (the 'd'). We do not know if
1451 interrupts were enabled in the mean time or shortly after this
1455 # tracer: preemptoff
1457 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1458 # --------------------------------------------------------------------
1459 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1461 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1463 # => started at: wake_up_new_task
1464 # => ended at: task_rq_unlock
1468 # / _-----=> irqs-off
1469 # | / _----=> need-resched
1470 # || / _---=> hardirq/softirq
1471 # ||| / _--=> preempt-depth
1473 # cmd pid ||||| time | caller
1475 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1476 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1477 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1478 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1479 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1481 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1482 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1483 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1484 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1485 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1486 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1487 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1488 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1490 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1491 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1492 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1493 bash-1994 1d..2 36us : do_softirq <-irq_exit
1494 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1495 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1496 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1497 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1498 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1499 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1501 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1502 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1503 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1504 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1505 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1506 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1507 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1508 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1509 bash-1994 1.N.1 104us : <stack trace>
1510 => sub_preempt_count
1511 => _raw_spin_unlock_irqrestore
1519 The above is an example of the preemptoff trace with
1520 function-trace set. Here we see that interrupts were not disabled
1521 the entire time. The irq_enter code lets us know that we entered
1522 an interrupt 'h'. Before that, the functions being traced still
1523 show that it is not in an interrupt, but we can see from the
1524 functions themselves that this is not the case.
1529 Knowing the locations that have interrupts disabled or
1530 preemption disabled for the longest times is helpful. But
1531 sometimes we would like to know when either preemption and/or
1532 interrupts are disabled.
1534 Consider the following code::
1536 local_irq_disable();
1537 call_function_with_irqs_off();
1539 call_function_with_irqs_and_preemption_off();
1541 call_function_with_preemption_off();
1544 The irqsoff tracer will record the total length of
1545 call_function_with_irqs_off() and
1546 call_function_with_irqs_and_preemption_off().
1548 The preemptoff tracer will record the total length of
1549 call_function_with_irqs_and_preemption_off() and
1550 call_function_with_preemption_off().
1552 But neither will trace the time that interrupts and/or
1553 preemption is disabled. This total time is the time that we can
1554 not schedule. To record this time, use the preemptirqsoff
1557 Again, using this trace is much like the irqsoff and preemptoff
1561 # echo 0 > options/function-trace
1562 # echo preemptirqsoff > current_tracer
1563 # echo 1 > tracing_on
1564 # echo 0 > tracing_max_latency
1567 # echo 0 > tracing_on
1569 # tracer: preemptirqsoff
1571 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1572 # --------------------------------------------------------------------
1573 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1575 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1577 # => started at: ata_scsi_queuecmd
1578 # => ended at: ata_scsi_queuecmd
1582 # / _-----=> irqs-off
1583 # | / _----=> need-resched
1584 # || / _---=> hardirq/softirq
1585 # ||| / _--=> preempt-depth
1587 # cmd pid ||||| time | caller
1589 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1590 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1591 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1592 ls-2230 3...1 111us : <stack trace>
1593 => sub_preempt_count
1594 => _raw_spin_unlock_irqrestore
1595 => ata_scsi_queuecmd
1596 => scsi_dispatch_cmd
1598 => __blk_run_queue_uncond
1601 => generic_make_request
1606 => htree_dirblock_to_tree
1607 => ext3_htree_fill_tree
1611 => system_call_fastpath
1614 The trace_hardirqs_off_thunk is called from assembly on x86 when
1615 interrupts are disabled in the assembly code. Without the
1616 function tracing, we do not know if interrupts were enabled
1617 within the preemption points. We do see that it started with
1620 Here is a trace with function-trace set::
1622 # tracer: preemptirqsoff
1624 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1625 # --------------------------------------------------------------------
1626 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1628 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1630 # => started at: schedule
1631 # => ended at: mutex_unlock
1635 # / _-----=> irqs-off
1636 # | / _----=> need-resched
1637 # || / _---=> hardirq/softirq
1638 # ||| / _--=> preempt-depth
1640 # cmd pid ||||| time | caller
1642 kworker/-59 3...1 0us : __schedule <-schedule
1643 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1644 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1645 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1646 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1647 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1648 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1649 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1650 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1651 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1652 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1653 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1654 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1655 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1656 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1657 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1658 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1659 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1660 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1661 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1662 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1663 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1664 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1665 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1666 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1667 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1668 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1669 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1670 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1671 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1672 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1673 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1675 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1676 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1677 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1678 ls-2269 3d..3 21us : do_softirq <-irq_exit
1679 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1680 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1681 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1682 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1683 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1684 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1685 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1687 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1688 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1689 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1690 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1691 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1692 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1694 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1695 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1696 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1697 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1698 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1699 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1700 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1701 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1702 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1703 ls-2269 3d... 186us : <stack trace>
1704 => __mutex_unlock_slowpath
1711 => system_call_fastpath
1713 This is an interesting trace. It started with kworker running and
1714 scheduling out and ls taking over. But as soon as ls released the
1715 rq lock and enabled interrupts (but not preemption) an interrupt
1716 triggered. When the interrupt finished, it started running softirqs.
1717 But while the softirq was running, another interrupt triggered.
1718 When an interrupt is running inside a softirq, the annotation is 'H'.
1724 One common case that people are interested in tracing is the
1725 time it takes for a task that is woken to actually wake up.
1726 Now for non Real-Time tasks, this can be arbitrary. But tracing
1727 it none the less can be interesting.
1729 Without function tracing::
1731 # echo 0 > options/function-trace
1732 # echo wakeup > current_tracer
1733 # echo 1 > tracing_on
1734 # echo 0 > tracing_max_latency
1736 # echo 0 > tracing_on
1740 # wakeup latency trace v1.1.5 on 3.8.0-test+
1741 # --------------------------------------------------------------------
1742 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1744 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1748 # / _-----=> irqs-off
1749 # | / _----=> need-resched
1750 # || / _---=> hardirq/softirq
1751 # ||| / _--=> preempt-depth
1753 # cmd pid ||||| time | caller
1755 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1756 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1757 <idle>-0 3d..3 15us : __schedule <-schedule
1758 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1760 The tracer only traces the highest priority task in the system
1761 to avoid tracing the normal circumstances. Here we see that
1762 the kworker with a nice priority of -20 (not very nice), took
1763 just 15 microseconds from the time it woke up, to the time it
1766 Non Real-Time tasks are not that interesting. A more interesting
1767 trace is to concentrate only on Real-Time tasks.
1772 In a Real-Time environment it is very important to know the
1773 wakeup time it takes for the highest priority task that is woken
1774 up to the time that it executes. This is also known as "schedule
1775 latency". I stress the point that this is about RT tasks. It is
1776 also important to know the scheduling latency of non-RT tasks,
1777 but the average schedule latency is better for non-RT tasks.
1778 Tools like LatencyTop are more appropriate for such
1781 Real-Time environments are interested in the worst case latency.
1782 That is the longest latency it takes for something to happen,
1783 and not the average. We can have a very fast scheduler that may
1784 only have a large latency once in a while, but that would not
1785 work well with Real-Time tasks. The wakeup_rt tracer was designed
1786 to record the worst case wakeups of RT tasks. Non-RT tasks are
1787 not recorded because the tracer only records one worst case and
1788 tracing non-RT tasks that are unpredictable will overwrite the
1789 worst case latency of RT tasks (just run the normal wakeup
1790 tracer for a while to see that effect).
1792 Since this tracer only deals with RT tasks, we will run this
1793 slightly differently than we did with the previous tracers.
1794 Instead of performing an 'ls', we will run 'sleep 1' under
1795 'chrt' which changes the priority of the task.
1798 # echo 0 > options/function-trace
1799 # echo wakeup_rt > current_tracer
1800 # echo 1 > tracing_on
1801 # echo 0 > tracing_max_latency
1803 # echo 0 > tracing_on
1809 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1810 # --------------------------------------------------------------------
1811 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1813 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1817 # / _-----=> irqs-off
1818 # | / _----=> need-resched
1819 # || / _---=> hardirq/softirq
1820 # ||| / _--=> preempt-depth
1822 # cmd pid ||||| time | caller
1824 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1825 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1826 <idle>-0 3d..3 5us : __schedule <-schedule
1827 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1830 Running this on an idle system, we see that it only took 5 microseconds
1831 to perform the task switch. Note, since the trace point in the schedule
1832 is before the actual "switch", we stop the tracing when the recorded task
1833 is about to schedule in. This may change if we add a new marker at the
1834 end of the scheduler.
1836 Notice that the recorded task is 'sleep' with the PID of 2389
1837 and it has an rt_prio of 5. This priority is user-space priority
1838 and not the internal kernel priority. The policy is 1 for
1839 SCHED_FIFO and 2 for SCHED_RR.
1841 Note, that the trace data shows the internal priority (99 - rtprio).
1844 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1846 The 0:120:R means idle was running with a nice priority of 0 (120 - 120)
1847 and in the running state 'R'. The sleep task was scheduled in with
1848 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1849 and it too is in the running state.
1851 Doing the same with chrt -r 5 and function-trace set.
1854 echo 1 > options/function-trace
1858 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1859 # --------------------------------------------------------------------
1860 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1862 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
1866 # / _-----=> irqs-off
1867 # | / _----=> need-resched
1868 # || / _---=> hardirq/softirq
1869 # ||| / _--=> preempt-depth
1871 # cmd pid ||||| time | caller
1873 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
1874 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1875 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
1876 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
1877 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
1878 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
1879 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
1880 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
1881 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
1882 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1883 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
1884 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
1885 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
1886 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
1887 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
1888 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
1889 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
1890 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
1891 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
1892 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
1893 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
1894 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
1895 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
1896 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
1897 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
1898 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
1899 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
1900 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
1901 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
1902 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
1903 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
1904 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
1905 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
1906 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
1907 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
1908 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
1909 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
1910 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit
1911 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
1912 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
1913 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
1914 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
1915 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1916 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
1917 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
1918 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
1919 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
1920 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
1921 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
1922 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
1923 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
1924 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1925 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
1926 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1927 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1928 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
1929 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
1930 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
1931 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1932 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
1933 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
1934 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
1935 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
1936 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
1937 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
1938 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
1939 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
1940 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1941 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
1942 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
1943 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
1944 <idle>-0 3.N.. 25us : schedule <-cpu_idle
1945 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
1946 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
1947 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
1948 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
1949 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
1950 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
1951 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
1952 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
1953 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
1954 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
1955 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
1956 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
1957 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
1959 This isn't that big of a trace, even with function tracing enabled,
1960 so I included the entire trace.
1962 The interrupt went off while when the system was idle. Somewhere
1963 before task_woken_rt() was called, the NEED_RESCHED flag was set,
1964 this is indicated by the first occurrence of the 'N' flag.
1966 Latency tracing and events
1967 --------------------------
1968 As function tracing can induce a much larger latency, but without
1969 seeing what happens within the latency it is hard to know what
1970 caused it. There is a middle ground, and that is with enabling
1974 # echo 0 > options/function-trace
1975 # echo wakeup_rt > current_tracer
1976 # echo 1 > events/enable
1977 # echo 1 > tracing_on
1978 # echo 0 > tracing_max_latency
1980 # echo 0 > tracing_on
1984 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1985 # --------------------------------------------------------------------
1986 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1988 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
1992 # / _-----=> irqs-off
1993 # | / _----=> need-resched
1994 # || / _---=> hardirq/softirq
1995 # ||| / _--=> preempt-depth
1997 # cmd pid ||||| time | caller
1999 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
2000 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
2001 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
2002 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
2003 <idle>-0 2.N.2 2us : power_end: cpu_id=2
2004 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
2005 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
2006 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
2007 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
2008 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
2009 <idle>-0 2d..3 6us : __schedule <-schedule
2010 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
2013 Hardware Latency Detector
2014 -------------------------
2016 The hardware latency detector is executed by enabling the "hwlat" tracer.
2018 NOTE, this tracer will affect the performance of the system as it will
2019 periodically make a CPU constantly busy with interrupts disabled.
2022 # echo hwlat > current_tracer
2028 # / _----=> need-resched
2029 # | / _---=> hardirq/softirq
2030 # || / _--=> preempt-depth
2032 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2034 <...>-3638 [001] d... 19452.055471: #1 inner/outer(us): 12/14 ts:1499801089.066141940
2035 <...>-3638 [003] d... 19454.071354: #2 inner/outer(us): 11/9 ts:1499801091.082164365
2036 <...>-3638 [002] dn.. 19461.126852: #3 inner/outer(us): 12/9 ts:1499801098.138150062
2037 <...>-3638 [001] d... 19488.340960: #4 inner/outer(us): 8/12 ts:1499801125.354139633
2038 <...>-3638 [003] d... 19494.388553: #5 inner/outer(us): 8/12 ts:1499801131.402150961
2039 <...>-3638 [003] d... 19501.283419: #6 inner/outer(us): 0/12 ts:1499801138.297435289 nmi-total:4 nmi-count:1
2042 The above output is somewhat the same in the header. All events will have
2043 interrupts disabled 'd'. Under the FUNCTION title there is:
2046 This is the count of events recorded that were greater than the
2047 tracing_threshold (See below).
2049 inner/outer(us): 12/14
2051 This shows two numbers as "inner latency" and "outer latency". The test
2052 runs in a loop checking a timestamp twice. The latency detected within
2053 the two timestamps is the "inner latency" and the latency detected
2054 after the previous timestamp and the next timestamp in the loop is
2055 the "outer latency".
2057 ts:1499801089.066141940
2059 The absolute timestamp that the event happened.
2061 nmi-total:4 nmi-count:1
2063 On architectures that support it, if an NMI comes in during the
2064 test, the time spent in NMI is reported in "nmi-total" (in
2067 All architectures that have NMIs will show the "nmi-count" if an
2068 NMI comes in during the test.
2073 This gets automatically set to "10" to represent 10
2074 microseconds. This is the threshold of latency that
2075 needs to be detected before the trace will be recorded.
2077 Note, when hwlat tracer is finished (another tracer is
2078 written into "current_tracer"), the original value for
2079 tracing_threshold is placed back into this file.
2081 hwlat_detector/width
2082 The length of time the test runs with interrupts disabled.
2084 hwlat_detector/window
2085 The length of time of the window which the test
2086 runs. That is, the test will run for "width"
2087 microseconds per "window" microseconds
2090 When the test is started. A kernel thread is created that
2091 runs the test. This thread will alternate between CPUs
2092 listed in the tracing_cpumask between each period
2093 (one "window"). To limit the test to specific CPUs
2094 set the mask in this file to only the CPUs that the test
2100 This tracer is the function tracer. Enabling the function tracer
2101 can be done from the debug file system. Make sure the
2102 ftrace_enabled is set; otherwise this tracer is a nop.
2103 See the "ftrace_enabled" section below.
2106 # sysctl kernel.ftrace_enabled=1
2107 # echo function > current_tracer
2108 # echo 1 > tracing_on
2110 # echo 0 > tracing_on
2114 # entries-in-buffer/entries-written: 24799/24799 #P:4
2117 # / _----=> need-resched
2118 # | / _---=> hardirq/softirq
2119 # || / _--=> preempt-depth
2121 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2123 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
2124 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
2125 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
2126 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
2127 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
2128 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
2129 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
2130 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
2134 Note: function tracer uses ring buffers to store the above
2135 entries. The newest data may overwrite the oldest data.
2136 Sometimes using echo to stop the trace is not sufficient because
2137 the tracing could have overwritten the data that you wanted to
2138 record. For this reason, it is sometimes better to disable
2139 tracing directly from a program. This allows you to stop the
2140 tracing at the point that you hit the part that you are
2141 interested in. To disable the tracing directly from a C program,
2142 something like following code snippet can be used::
2146 int main(int argc, char *argv[]) {
2148 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
2150 if (condition_hit()) {
2151 write(trace_fd, "0", 1);
2157 Single thread tracing
2158 ---------------------
2160 By writing into set_ftrace_pid you can trace a
2161 single thread. For example::
2163 # cat set_ftrace_pid
2165 # echo 3111 > set_ftrace_pid
2166 # cat set_ftrace_pid
2168 # echo function > current_tracer
2172 # TASK-PID CPU# TIMESTAMP FUNCTION
2174 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
2175 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
2176 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
2177 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
2178 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
2179 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
2180 # echo > set_ftrace_pid
2184 # TASK-PID CPU# TIMESTAMP FUNCTION
2186 ##### CPU 3 buffer started ####
2187 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
2188 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
2189 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
2190 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
2191 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
2193 If you want to trace a function when executing, you could use
2194 something like this simple program.
2199 #include <sys/types.h>
2200 #include <sys/stat.h>
2206 #define STR(x) _STR(x)
2207 #define MAX_PATH 256
2209 const char *find_tracefs(void)
2211 static char tracefs[MAX_PATH+1];
2212 static int tracefs_found;
2219 if ((fp = fopen("/proc/mounts","r")) == NULL) {
2220 perror("/proc/mounts");
2224 while (fscanf(fp, "%*s %"
2226 "s %99s %*s %*d %*d\n",
2227 tracefs, type) == 2) {
2228 if (strcmp(type, "tracefs") == 0)
2233 if (strcmp(type, "tracefs") != 0) {
2234 fprintf(stderr, "tracefs not mounted");
2238 strcat(tracefs, "/tracing/");
2244 const char *tracing_file(const char *file_name)
2246 static char trace_file[MAX_PATH+1];
2247 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name);
2251 int main (int argc, char **argv)
2261 ffd = open(tracing_file("current_tracer"), O_WRONLY);
2264 write(ffd, "nop", 3);
2266 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
2267 s = sprintf(line, "%d\n", getpid());
2270 write(ffd, "function", 8);
2275 execvp(argv[1], argv+1);
2281 Or this simple script!
2286 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts`
2287 echo nop > $tracefs/tracing/current_tracer
2288 echo 0 > $tracefs/tracing/tracing_on
2289 echo $$ > $tracefs/tracing/set_ftrace_pid
2290 echo function > $tracefs/tracing/current_tracer
2291 echo 1 > $tracefs/tracing/tracing_on
2295 function graph tracer
2296 ---------------------------
2298 This tracer is similar to the function tracer except that it
2299 probes a function on its entry and its exit. This is done by
2300 using a dynamically allocated stack of return addresses in each
2301 task_struct. On function entry the tracer overwrites the return
2302 address of each function traced to set a custom probe. Thus the
2303 original return address is stored on the stack of return address
2306 Probing on both ends of a function leads to special features
2309 - measure of a function's time execution
2310 - having a reliable call stack to draw function calls graph
2312 This tracer is useful in several situations:
2314 - you want to find the reason of a strange kernel behavior and
2315 need to see what happens in detail on any areas (or specific
2318 - you are experiencing weird latencies but it's difficult to
2321 - you want to find quickly which path is taken by a specific
2324 - you just want to peek inside a working kernel and want to see
2329 # tracer: function_graph
2331 # CPU DURATION FUNCTION CALLS
2335 0) | do_sys_open() {
2337 0) | kmem_cache_alloc() {
2338 0) 1.382 us | __might_sleep();
2340 0) | strncpy_from_user() {
2341 0) | might_fault() {
2342 0) 1.389 us | __might_sleep();
2347 0) 0.668 us | _spin_lock();
2348 0) 0.570 us | expand_files();
2349 0) 0.586 us | _spin_unlock();
2352 There are several columns that can be dynamically
2353 enabled/disabled. You can use every combination of options you
2354 want, depending on your needs.
2356 - The cpu number on which the function executed is default
2357 enabled. It is sometimes better to only trace one cpu (see
2358 tracing_cpu_mask file) or you might sometimes see unordered
2359 function calls while cpu tracing switch.
2361 - hide: echo nofuncgraph-cpu > trace_options
2362 - show: echo funcgraph-cpu > trace_options
2364 - The duration (function's time of execution) is displayed on
2365 the closing bracket line of a function or on the same line
2366 than the current function in case of a leaf one. It is default
2369 - hide: echo nofuncgraph-duration > trace_options
2370 - show: echo funcgraph-duration > trace_options
2372 - The overhead field precedes the duration field in case of
2373 reached duration thresholds.
2375 - hide: echo nofuncgraph-overhead > trace_options
2376 - show: echo funcgraph-overhead > trace_options
2377 - depends on: funcgraph-duration
2381 3) # 1837.709 us | } /* __switch_to */
2382 3) | finish_task_switch() {
2383 3) 0.313 us | _raw_spin_unlock_irq();
2385 3) # 1889.063 us | } /* __schedule */
2386 3) ! 140.417 us | } /* __schedule */
2387 3) # 2034.948 us | } /* schedule */
2388 3) * 33998.59 us | } /* schedule_preempt_disabled */
2392 1) 0.260 us | msecs_to_jiffies();
2393 1) 0.313 us | __rcu_read_unlock();
2396 1) 0.313 us | rcu_bh_qs();
2397 1) 0.313 us | __local_bh_enable();
2399 1) 0.365 us | idle_cpu();
2400 1) | rcu_irq_exit() {
2401 1) 0.417 us | rcu_eqs_enter_common.isra.47();
2405 1) @ 119760.2 us | }
2411 2) 0.417 us | scheduler_ipi();
2421 + means that the function exceeded 10 usecs.
2422 ! means that the function exceeded 100 usecs.
2423 # means that the function exceeded 1000 usecs.
2424 * means that the function exceeded 10 msecs.
2425 @ means that the function exceeded 100 msecs.
2426 $ means that the function exceeded 1 sec.
2429 - The task/pid field displays the thread cmdline and pid which
2430 executed the function. It is default disabled.
2432 - hide: echo nofuncgraph-proc > trace_options
2433 - show: echo funcgraph-proc > trace_options
2437 # tracer: function_graph
2439 # CPU TASK/PID DURATION FUNCTION CALLS
2441 0) sh-4802 | | d_free() {
2442 0) sh-4802 | | call_rcu() {
2443 0) sh-4802 | | __call_rcu() {
2444 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2445 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2446 0) sh-4802 | 2.899 us | }
2447 0) sh-4802 | 4.040 us | }
2448 0) sh-4802 | 5.151 us | }
2449 0) sh-4802 | + 49.370 us | }
2452 - The absolute time field is an absolute timestamp given by the
2453 system clock since it started. A snapshot of this time is
2454 given on each entry/exit of functions
2456 - hide: echo nofuncgraph-abstime > trace_options
2457 - show: echo funcgraph-abstime > trace_options
2462 # TIME CPU DURATION FUNCTION CALLS
2464 360.774522 | 1) 0.541 us | }
2465 360.774522 | 1) 4.663 us | }
2466 360.774523 | 1) 0.541 us | __wake_up_bit();
2467 360.774524 | 1) 6.796 us | }
2468 360.774524 | 1) 7.952 us | }
2469 360.774525 | 1) 9.063 us | }
2470 360.774525 | 1) 0.615 us | journal_mark_dirty();
2471 360.774527 | 1) 0.578 us | __brelse();
2472 360.774528 | 1) | reiserfs_prepare_for_journal() {
2473 360.774528 | 1) | unlock_buffer() {
2474 360.774529 | 1) | wake_up_bit() {
2475 360.774529 | 1) | bit_waitqueue() {
2476 360.774530 | 1) 0.594 us | __phys_addr();
2479 The function name is always displayed after the closing bracket
2480 for a function if the start of that function is not in the
2483 Display of the function name after the closing bracket may be
2484 enabled for functions whose start is in the trace buffer,
2485 allowing easier searching with grep for function durations.
2486 It is default disabled.
2488 - hide: echo nofuncgraph-tail > trace_options
2489 - show: echo funcgraph-tail > trace_options
2491 Example with nofuncgraph-tail (default)::
2494 0) | kmem_cache_free() {
2495 0) 0.518 us | __phys_addr();
2499 Example with funcgraph-tail::
2502 0) | kmem_cache_free() {
2503 0) 0.518 us | __phys_addr();
2504 0) 1.757 us | } /* kmem_cache_free() */
2505 0) 2.861 us | } /* putname() */
2507 You can put some comments on specific functions by using
2508 trace_printk() For example, if you want to put a comment inside
2509 the __might_sleep() function, you just have to include
2510 <linux/ftrace.h> and call trace_printk() inside __might_sleep()::
2512 trace_printk("I'm a comment!\n")
2516 1) | __might_sleep() {
2517 1) | /* I'm a comment! */
2521 You might find other useful features for this tracer in the
2522 following "dynamic ftrace" section such as tracing only specific
2528 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2529 virtually no overhead when function tracing is disabled. The way
2530 this works is the mcount function call (placed at the start of
2531 every kernel function, produced by the -pg switch in gcc),
2532 starts of pointing to a simple return. (Enabling FTRACE will
2533 include the -pg switch in the compiling of the kernel.)
2535 At compile time every C file object is run through the
2536 recordmcount program (located in the scripts directory). This
2537 program will parse the ELF headers in the C object to find all
2538 the locations in the .text section that call mcount. Starting
2539 with gcc verson 4.6, the -mfentry has been added for x86, which
2540 calls "__fentry__" instead of "mcount". Which is called before
2541 the creation of the stack frame.
2543 Note, not all sections are traced. They may be prevented by either
2544 a notrace, or blocked another way and all inline functions are not
2545 traced. Check the "available_filter_functions" file to see what functions
2548 A section called "__mcount_loc" is created that holds
2549 references to all the mcount/fentry call sites in the .text section.
2550 The recordmcount program re-links this section back into the
2551 original object. The final linking stage of the kernel will add all these
2552 references into a single table.
2554 On boot up, before SMP is initialized, the dynamic ftrace code
2555 scans this table and updates all the locations into nops. It
2556 also records the locations, which are added to the
2557 available_filter_functions list. Modules are processed as they
2558 are loaded and before they are executed. When a module is
2559 unloaded, it also removes its functions from the ftrace function
2560 list. This is automatic in the module unload code, and the
2561 module author does not need to worry about it.
2563 When tracing is enabled, the process of modifying the function
2564 tracepoints is dependent on architecture. The old method is to use
2565 kstop_machine to prevent races with the CPUs executing code being
2566 modified (which can cause the CPU to do undesirable things, especially
2567 if the modified code crosses cache (or page) boundaries), and the nops are
2568 patched back to calls. But this time, they do not call mcount
2569 (which is just a function stub). They now call into the ftrace
2572 The new method of modifying the function tracepoints is to place
2573 a breakpoint at the location to be modified, sync all CPUs, modify
2574 the rest of the instruction not covered by the breakpoint. Sync
2575 all CPUs again, and then remove the breakpoint with the finished
2576 version to the ftrace call site.
2578 Some archs do not even need to monkey around with the synchronization,
2579 and can just slap the new code on top of the old without any
2580 problems with other CPUs executing it at the same time.
2582 One special side-effect to the recording of the functions being
2583 traced is that we can now selectively choose which functions we
2584 wish to trace and which ones we want the mcount calls to remain
2587 Two files are used, one for enabling and one for disabling the
2588 tracing of specified functions. They are:
2596 A list of available functions that you can add to these files is
2599 available_filter_functions
2603 # cat available_filter_functions
2612 If I am only interested in sys_nanosleep and hrtimer_interrupt::
2614 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2615 # echo function > current_tracer
2616 # echo 1 > tracing_on
2618 # echo 0 > tracing_on
2622 # entries-in-buffer/entries-written: 5/5 #P:4
2625 # / _----=> need-resched
2626 # | / _---=> hardirq/softirq
2627 # || / _--=> preempt-depth
2629 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2631 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2632 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2633 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2634 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2635 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2637 To see which functions are being traced, you can cat the file:
2640 # cat set_ftrace_filter
2645 Perhaps this is not enough. The filters also allow glob(7) matching.
2648 will match functions that begin with <match>
2650 will match functions that end with <match>
2652 will match functions that have <match> in it
2653 ``<match1>*<match2>``
2654 will match functions that begin with <match1> and end with <match2>
2657 It is better to use quotes to enclose the wild cards,
2658 otherwise the shell may expand the parameters into names
2659 of files in the local directory.
2663 # echo 'hrtimer_*' > set_ftrace_filter
2669 # entries-in-buffer/entries-written: 897/897 #P:4
2672 # / _----=> need-resched
2673 # | / _---=> hardirq/softirq
2674 # || / _--=> preempt-depth
2676 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2678 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2679 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2680 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2681 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2682 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2683 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2684 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2685 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2687 Notice that we lost the sys_nanosleep.
2690 # cat set_ftrace_filter
2695 hrtimer_try_to_cancel
2699 hrtimer_force_reprogram
2700 hrtimer_get_next_event
2704 hrtimer_get_remaining
2706 hrtimer_init_sleeper
2709 This is because the '>' and '>>' act just like they do in bash.
2710 To rewrite the filters, use '>'
2711 To append to the filters, use '>>'
2713 To clear out a filter so that all functions will be recorded
2716 # echo > set_ftrace_filter
2717 # cat set_ftrace_filter
2720 Again, now we want to append.
2724 # echo sys_nanosleep > set_ftrace_filter
2725 # cat set_ftrace_filter
2727 # echo 'hrtimer_*' >> set_ftrace_filter
2728 # cat set_ftrace_filter
2733 hrtimer_try_to_cancel
2737 hrtimer_force_reprogram
2738 hrtimer_get_next_event
2743 hrtimer_get_remaining
2745 hrtimer_init_sleeper
2748 The set_ftrace_notrace prevents those functions from being
2752 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2758 # entries-in-buffer/entries-written: 39608/39608 #P:4
2761 # / _----=> need-resched
2762 # | / _---=> hardirq/softirq
2763 # || / _--=> preempt-depth
2765 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2767 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2768 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2769 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2770 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2771 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2772 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2773 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2774 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2775 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2776 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2777 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2778 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2780 We can see that there's no more lock or preempt tracing.
2783 Dynamic ftrace with the function graph tracer
2784 ---------------------------------------------
2786 Although what has been explained above concerns both the
2787 function tracer and the function-graph-tracer, there are some
2788 special features only available in the function-graph tracer.
2790 If you want to trace only one function and all of its children,
2791 you just have to echo its name into set_graph_function::
2793 echo __do_fault > set_graph_function
2795 will produce the following "expanded" trace of the __do_fault()
2799 0) | filemap_fault() {
2800 0) | find_lock_page() {
2801 0) 0.804 us | find_get_page();
2802 0) | __might_sleep() {
2806 0) 0.653 us | _spin_lock();
2807 0) 0.578 us | page_add_file_rmap();
2808 0) 0.525 us | native_set_pte_at();
2809 0) 0.585 us | _spin_unlock();
2810 0) | unlock_page() {
2811 0) 0.541 us | page_waitqueue();
2812 0) 0.639 us | __wake_up_bit();
2816 0) | filemap_fault() {
2817 0) | find_lock_page() {
2818 0) 0.698 us | find_get_page();
2819 0) | __might_sleep() {
2823 0) 0.631 us | _spin_lock();
2824 0) 0.571 us | page_add_file_rmap();
2825 0) 0.526 us | native_set_pte_at();
2826 0) 0.586 us | _spin_unlock();
2827 0) | unlock_page() {
2828 0) 0.533 us | page_waitqueue();
2829 0) 0.638 us | __wake_up_bit();
2833 You can also expand several functions at once::
2835 echo sys_open > set_graph_function
2836 echo sys_close >> set_graph_function
2838 Now if you want to go back to trace all functions you can clear
2839 this special filter via::
2841 echo > set_graph_function
2847 Note, the proc sysctl ftrace_enable is a big on/off switch for the
2848 function tracer. By default it is enabled (when function tracing is
2849 enabled in the kernel). If it is disabled, all function tracing is
2850 disabled. This includes not only the function tracers for ftrace, but
2851 also for any other uses (perf, kprobes, stack tracing, profiling, etc).
2853 Please disable this with care.
2855 This can be disable (and enabled) with::
2857 sysctl kernel.ftrace_enabled=0
2858 sysctl kernel.ftrace_enabled=1
2862 echo 0 > /proc/sys/kernel/ftrace_enabled
2863 echo 1 > /proc/sys/kernel/ftrace_enabled
2869 A few commands are supported by the set_ftrace_filter interface.
2870 Trace commands have the following format::
2872 <function>:<command>:<parameter>
2874 The following commands are supported:
2877 This command enables function filtering per module. The
2878 parameter defines the module. For example, if only the write*
2879 functions in the ext3 module are desired, run:
2881 echo 'write*:mod:ext3' > set_ftrace_filter
2883 This command interacts with the filter in the same way as
2884 filtering based on function names. Thus, adding more functions
2885 in a different module is accomplished by appending (>>) to the
2886 filter file. Remove specific module functions by prepending
2889 echo '!writeback*:mod:ext3' >> set_ftrace_filter
2891 Mod command supports module globbing. Disable tracing for all
2892 functions except a specific module::
2894 echo '!*:mod:!ext3' >> set_ftrace_filter
2896 Disable tracing for all modules, but still trace kernel::
2898 echo '!*:mod:*' >> set_ftrace_filter
2900 Enable filter only for kernel::
2902 echo '*write*:mod:!*' >> set_ftrace_filter
2904 Enable filter for module globbing::
2906 echo '*write*:mod:*snd*' >> set_ftrace_filter
2909 These commands turn tracing on and off when the specified
2910 functions are hit. The parameter determines how many times the
2911 tracing system is turned on and off. If unspecified, there is
2912 no limit. For example, to disable tracing when a schedule bug
2913 is hit the first 5 times, run::
2915 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
2917 To always disable tracing when __schedule_bug is hit::
2919 echo '__schedule_bug:traceoff' > set_ftrace_filter
2921 These commands are cumulative whether or not they are appended
2922 to set_ftrace_filter. To remove a command, prepend it by '!'
2923 and drop the parameter::
2925 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
2927 The above removes the traceoff command for __schedule_bug
2928 that have a counter. To remove commands without counters::
2930 echo '!__schedule_bug:traceoff' > set_ftrace_filter
2933 Will cause a snapshot to be triggered when the function is hit.
2936 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
2938 To only snapshot once:
2941 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
2943 To remove the above commands::
2945 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
2946 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
2948 - enable_event/disable_event:
2949 These commands can enable or disable a trace event. Note, because
2950 function tracing callbacks are very sensitive, when these commands
2951 are registered, the trace point is activated, but disabled in
2952 a "soft" mode. That is, the tracepoint will be called, but
2953 just will not be traced. The event tracepoint stays in this mode
2954 as long as there's a command that triggers it.
2957 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
2962 <function>:enable_event:<system>:<event>[:count]
2963 <function>:disable_event:<system>:<event>[:count]
2965 To remove the events commands::
2967 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
2969 echo '!schedule:disable_event:sched:sched_switch' > \
2973 When the function is hit, it will dump the contents of the ftrace
2974 ring buffer to the console. This is useful if you need to debug
2975 something, and want to dump the trace when a certain function
2976 is hit. Perhaps its a function that is called before a tripple
2977 fault happens and does not allow you to get a regular dump.
2980 When the function is hit, it will dump the contents of the ftrace
2981 ring buffer for the current CPU to the console. Unlike the "dump"
2982 command, it only prints out the contents of the ring buffer for the
2983 CPU that executed the function that triggered the dump.
2988 The trace_pipe outputs the same content as the trace file, but
2989 the effect on the tracing is different. Every read from
2990 trace_pipe is consumed. This means that subsequent reads will be
2991 different. The trace is live.
2994 # echo function > current_tracer
2995 # cat trace_pipe > /tmp/trace.out &
2997 # echo 1 > tracing_on
2999 # echo 0 > tracing_on
3003 # entries-in-buffer/entries-written: 0/0 #P:4
3006 # / _----=> need-resched
3007 # | / _---=> hardirq/softirq
3008 # || / _--=> preempt-depth
3010 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3014 # cat /tmp/trace.out
3015 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
3016 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
3017 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
3018 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
3019 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
3020 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
3021 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
3022 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
3023 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
3026 Note, reading the trace_pipe file will block until more input is
3032 Having too much or not enough data can be troublesome in
3033 diagnosing an issue in the kernel. The file buffer_size_kb is
3034 used to modify the size of the internal trace buffers. The
3035 number listed is the number of entries that can be recorded per
3036 CPU. To know the full size, multiply the number of possible CPUs
3037 with the number of entries.
3040 # cat buffer_size_kb
3041 1408 (units kilobytes)
3043 Or simply read buffer_total_size_kb
3046 # cat buffer_total_size_kb
3049 To modify the buffer, simple echo in a number (in 1024 byte segments).
3052 # echo 10000 > buffer_size_kb
3053 # cat buffer_size_kb
3054 10000 (units kilobytes)
3056 It will try to allocate as much as possible. If you allocate too
3057 much, it can cause Out-Of-Memory to trigger.
3060 # echo 1000000000000 > buffer_size_kb
3061 -bash: echo: write error: Cannot allocate memory
3062 # cat buffer_size_kb
3065 The per_cpu buffers can be changed individually as well:
3068 # echo 10000 > per_cpu/cpu0/buffer_size_kb
3069 # echo 100 > per_cpu/cpu1/buffer_size_kb
3071 When the per_cpu buffers are not the same, the buffer_size_kb
3072 at the top level will just show an X
3075 # cat buffer_size_kb
3078 This is where the buffer_total_size_kb is useful:
3081 # cat buffer_total_size_kb
3084 Writing to the top level buffer_size_kb will reset all the buffers
3085 to be the same again.
3089 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
3090 available to all non latency tracers. (Latency tracers which
3091 record max latency, such as "irqsoff" or "wakeup", can't use
3092 this feature, since those are already using the snapshot
3093 mechanism internally.)
3095 Snapshot preserves a current trace buffer at a particular point
3096 in time without stopping tracing. Ftrace swaps the current
3097 buffer with a spare buffer, and tracing continues in the new
3098 current (=previous spare) buffer.
3100 The following tracefs files in "tracing" are related to this
3105 This is used to take a snapshot and to read the output
3106 of the snapshot. Echo 1 into this file to allocate a
3107 spare buffer and to take a snapshot (swap), then read
3108 the snapshot from this file in the same format as
3109 "trace" (described above in the section "The File
3110 System"). Both reads snapshot and tracing are executable
3111 in parallel. When the spare buffer is allocated, echoing
3112 0 frees it, and echoing else (positive) values clear the
3114 More details are shown in the table below.
3116 +--------------+------------+------------+------------+
3117 |status\\input | 0 | 1 | else |
3118 +==============+============+============+============+
3119 |not allocated |(do nothing)| alloc+swap |(do nothing)|
3120 +--------------+------------+------------+------------+
3121 |allocated | free | swap | clear |
3122 +--------------+------------+------------+------------+
3124 Here is an example of using the snapshot feature.
3127 # echo 1 > events/sched/enable
3132 # entries-in-buffer/entries-written: 71/71 #P:8
3135 # / _----=> need-resched
3136 # | / _---=> hardirq/softirq
3137 # || / _--=> preempt-depth
3139 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3141 <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
3142 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
3144 <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
3149 # entries-in-buffer/entries-written: 77/77 #P:8
3152 # / _----=> need-resched
3153 # | / _---=> hardirq/softirq
3154 # || / _--=> preempt-depth
3156 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3158 <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
3159 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
3163 If you try to use this snapshot feature when current tracer is
3164 one of the latency tracers, you will get the following results.
3167 # echo wakeup > current_tracer
3169 bash: echo: write error: Device or resource busy
3171 cat: snapshot: Device or resource busy
3176 In the tracefs tracing directory is a directory called "instances".
3177 This directory can have new directories created inside of it using
3178 mkdir, and removing directories with rmdir. The directory created
3179 with mkdir in this directory will already contain files and other
3180 directories after it is created.
3183 # mkdir instances/foo
3185 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
3186 set_event snapshot trace trace_clock trace_marker trace_options
3187 trace_pipe tracing_on
3189 As you can see, the new directory looks similar to the tracing directory
3190 itself. In fact, it is very similar, except that the buffer and
3191 events are agnostic from the main director, or from any other
3192 instances that are created.
3194 The files in the new directory work just like the files with the
3195 same name in the tracing directory except the buffer that is used
3196 is a separate and new buffer. The files affect that buffer but do not
3197 affect the main buffer with the exception of trace_options. Currently,
3198 the trace_options affect all instances and the top level buffer
3199 the same, but this may change in future releases. That is, options
3200 may become specific to the instance they reside in.
3202 Notice that none of the function tracer files are there, nor is
3203 current_tracer and available_tracers. This is because the buffers
3204 can currently only have events enabled for them.
3207 # mkdir instances/foo
3208 # mkdir instances/bar
3209 # mkdir instances/zoot
3210 # echo 100000 > buffer_size_kb
3211 # echo 1000 > instances/foo/buffer_size_kb
3212 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
3213 # echo function > current_trace
3214 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
3215 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
3216 # echo 1 > instances/foo/events/sched/sched_switch/enable
3217 # echo 1 > instances/bar/events/irq/enable
3218 # echo 1 > instances/zoot/events/syscalls/enable
3220 CPU:2 [LOST 11745 EVENTS]
3221 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
3222 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
3223 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
3224 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
3225 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
3226 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
3227 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
3228 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
3229 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3230 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3231 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
3234 # cat instances/foo/trace_pipe
3235 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3236 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3237 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
3238 <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
3239 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
3240 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3241 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3242 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
3243 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
3244 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
3247 # cat instances/bar/trace_pipe
3248 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
3249 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
3250 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
3251 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
3252 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
3253 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
3254 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
3255 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
3256 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
3257 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
3258 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
3259 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
3262 # cat instances/zoot/trace
3265 # entries-in-buffer/entries-written: 18996/18996 #P:4
3268 # / _----=> need-resched
3269 # | / _---=> hardirq/softirq
3270 # || / _--=> preempt-depth
3272 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3274 bash-1998 [000] d... 140.733501: sys_write -> 0x2
3275 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
3276 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
3277 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
3278 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
3279 bash-1998 [000] d... 140.733510: sys_close(fd: a)
3280 bash-1998 [000] d... 140.733510: sys_close -> 0x0
3281 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
3282 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
3283 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
3284 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
3286 You can see that the trace of the top most trace buffer shows only
3287 the function tracing. The foo instance displays wakeups and task
3290 To remove the instances, simply delete their directories:
3293 # rmdir instances/foo
3294 # rmdir instances/bar
3295 # rmdir instances/zoot
3297 Note, if a process has a trace file open in one of the instance
3298 directories, the rmdir will fail with EBUSY.
3303 Since the kernel has a fixed sized stack, it is important not to
3304 waste it in functions. A kernel developer must be conscience of
3305 what they allocate on the stack. If they add too much, the system
3306 can be in danger of a stack overflow, and corruption will occur,
3307 usually leading to a system panic.
3309 There are some tools that check this, usually with interrupts
3310 periodically checking usage. But if you can perform a check
3311 at every function call that will become very useful. As ftrace provides
3312 a function tracer, it makes it convenient to check the stack size
3313 at every function call. This is enabled via the stack tracer.
3315 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
3316 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
3319 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
3321 You can also enable it from the kernel command line to trace
3322 the stack size of the kernel during boot up, by adding "stacktrace"
3323 to the kernel command line parameter.
3325 After running it for a few minutes, the output looks like:
3328 # cat stack_max_size
3332 Depth Size Location (18 entries)
3334 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
3335 1) 2704 160 find_busiest_group+0x31/0x1f1
3336 2) 2544 256 load_balance+0xd9/0x662
3337 3) 2288 80 idle_balance+0xbb/0x130
3338 4) 2208 128 __schedule+0x26e/0x5b9
3339 5) 2080 16 schedule+0x64/0x66
3340 6) 2064 128 schedule_timeout+0x34/0xe0
3341 7) 1936 112 wait_for_common+0x97/0xf1
3342 8) 1824 16 wait_for_completion+0x1d/0x1f
3343 9) 1808 128 flush_work+0xfe/0x119
3344 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
3345 11) 1664 48 input_available_p+0x1d/0x5c
3346 12) 1616 48 n_tty_poll+0x6d/0x134
3347 13) 1568 64 tty_poll+0x64/0x7f
3348 14) 1504 880 do_select+0x31e/0x511
3349 15) 624 400 core_sys_select+0x177/0x216
3350 16) 224 96 sys_select+0x91/0xb9
3351 17) 128 128 system_call_fastpath+0x16/0x1b
3353 Note, if -mfentry is being used by gcc, functions get traced before
3354 they set up the stack frame. This means that leaf level functions
3355 are not tested by the stack tracer when -mfentry is used.
3357 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
3361 More details can be found in the source code, in the `kernel/trace/*.c` files.
projects / linux-2.6-microblaze.git / blob