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.
228 The functions listed in "available_filter_functions" are what
229 can be written into this file.
231 This interface also allows for commands to be used. See the
232 "Filter commands" section for more details.
236 This has an effect opposite to that of
237 set_ftrace_filter. Any function that is added here will not
238 be traced. If a function exists in both set_ftrace_filter
239 and set_ftrace_notrace, the function will _not_ be traced.
243 Have the function tracer only trace the threads whose PID are
246 If the "function-fork" option is set, then when a task whose
247 PID is listed in this file forks, the child's PID will
248 automatically be added to this file, and the child will be
249 traced by the function tracer as well. This option will also
250 cause PIDs of tasks that exit to be removed from the file.
254 Have the events only trace a task with a PID listed in this file.
255 Note, sched_switch and sched_wake_up will also trace events
258 To have the PIDs of children of tasks with their PID in this file
259 added on fork, enable the "event-fork" option. That option will also
260 cause the PIDs of tasks to be removed from this file when the task
265 Functions listed in this file will cause the function graph
266 tracer to only trace these functions and the functions that
267 they call. (See the section "dynamic ftrace" for more details).
271 Similar to set_graph_function, but will disable function graph
272 tracing when the function is hit until it exits the function.
273 This makes it possible to ignore tracing functions that are called
274 by a specific function.
276 available_filter_functions:
278 This lists the functions that ftrace has processed and can trace.
279 These are the function names that you can pass to
280 "set_ftrace_filter" or "set_ftrace_notrace".
281 (See the section "dynamic ftrace" below for more details.)
283 dyn_ftrace_total_info:
285 This file is for debugging purposes. The number of functions that
286 have been converted to nops and are available to be traced.
290 This file is more for debugging ftrace, but can also be useful
291 in seeing if any function has a callback attached to it.
292 Not only does the trace infrastructure use ftrace function
293 trace utility, but other subsystems might too. This file
294 displays all functions that have a callback attached to them
295 as well as the number of callbacks that have been attached.
296 Note, a callback may also call multiple functions which will
297 not be listed in this count.
299 If the callback registered to be traced by a function with
300 the "save regs" attribute (thus even more overhead), a 'R'
301 will be displayed on the same line as the function that
302 is returning registers.
304 If the callback registered to be traced by a function with
305 the "ip modify" attribute (thus the regs->ip can be changed),
306 an 'I' will be displayed on the same line as the function that
309 If the architecture supports it, it will also show what callback
310 is being directly called by the function. If the count is greater
311 than 1 it most likely will be ftrace_ops_list_func().
313 If the callback of the function jumps to a trampoline that is
314 specific to a the callback and not the standard trampoline,
315 its address will be printed as well as the function that the
318 function_profile_enabled:
320 When set it will enable all functions with either the function
321 tracer, or if configured, the function graph tracer. It will
322 keep a histogram of the number of functions that were called
323 and if the function graph tracer was configured, it will also keep
324 track of the time spent in those functions. The histogram
325 content can be displayed in the files:
327 trace_stats/function<cpu> ( function0, function1, etc).
331 A directory that holds different tracing stats.
335 Enable dynamic trace points. See kprobetrace.txt.
339 Dynamic trace points stats. See kprobetrace.txt.
343 Used with the function graph tracer. This is the max depth
344 it will trace into a function. Setting this to a value of
345 one will show only the first kernel function that is called
350 This is for tools that read the raw format files. If an event in
351 the ring buffer references a string, only a pointer to the string
352 is recorded into the buffer and not the string itself. This prevents
353 tools from knowing what that string was. This file displays the string
354 and address for the string allowing tools to map the pointers to what
359 Only the pid of the task is recorded in a trace event unless
360 the event specifically saves the task comm as well. Ftrace
361 makes a cache of pid mappings to comms to try to display
362 comms for events. If a pid for a comm is not listed, then
363 "<...>" is displayed in the output.
365 If the option "record-cmd" is set to "0", then comms of tasks
366 will not be saved during recording. By default, it is enabled.
370 By default, 128 comms are saved (see "saved_cmdlines" above). To
371 increase or decrease the amount of comms that are cached, echo
372 in a the number of comms to cache, into this file.
376 If the option "record-tgid" is set, on each scheduling context switch
377 the Task Group ID of a task is saved in a table mapping the PID of
378 the thread to its TGID. By default, the "record-tgid" option is
383 This displays the "snapshot" buffer and also lets the user
384 take a snapshot of the current running trace.
385 See the "Snapshot" section below for more details.
389 When the stack tracer is activated, this will display the
390 maximum stack size it has encountered.
391 See the "Stack Trace" section below.
395 This displays the stack back trace of the largest stack
396 that was encountered when the stack tracer is activated.
397 See the "Stack Trace" section below.
401 This is similar to "set_ftrace_filter" but it limits what
402 functions the stack tracer will check.
406 Whenever an event is recorded into the ring buffer, a
407 "timestamp" is added. This stamp comes from a specified
408 clock. By default, ftrace uses the "local" clock. This
409 clock is very fast and strictly per cpu, but on some
410 systems it may not be monotonic with respect to other
411 CPUs. In other words, the local clocks may not be in sync
412 with local clocks on other CPUs.
414 Usual clocks for tracing::
417 [local] global counter x86-tsc
419 The clock with the square brackets around it is the one in effect.
422 Default clock, but may not be in sync across CPUs
425 This clock is in sync with all CPUs but may
426 be a bit slower than the local clock.
429 This is not a clock at all, but literally an atomic
430 counter. It counts up one by one, but is in sync
431 with all CPUs. This is useful when you need to
432 know exactly the order events occurred with respect to
433 each other on different CPUs.
436 This uses the jiffies counter and the time stamp
437 is relative to the time since boot up.
440 This makes ftrace use the same clock that perf uses.
441 Eventually perf will be able to read ftrace buffers
442 and this will help out in interleaving the data.
445 Architectures may define their own clocks. For
446 example, x86 uses its own TSC cycle clock here.
449 This uses the powerpc timebase register value.
450 This is in sync across CPUs and can also be used
451 to correlate events across hypervisor/guest if
455 This uses the fast monotonic clock (CLOCK_MONOTONIC)
456 which is monotonic and is subject to NTP rate adjustments.
459 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
460 which is montonic but is not subject to any rate adjustments
461 and ticks at the same rate as the hardware clocksource.
464 This is the boot clock (CLOCK_BOOTTIME) and is based on the
465 fast monotonic clock, but also accounts for time spent in
466 suspend. Since the clock access is designed for use in
467 tracing in the suspend path, some side effects are possible
468 if clock is accessed after the suspend time is accounted before
469 the fast mono clock is updated. In this case, the clock update
470 appears to happen slightly sooner than it normally would have.
471 Also on 32-bit systems, it's possible that the 64-bit boot offset
472 sees a partial update. These effects are rare and post
473 processing should be able to handle them. See comments in the
474 ktime_get_boot_fast_ns() function for more information.
476 To set a clock, simply echo the clock name into this file::
478 # echo global > trace_clock
482 This is a very useful file for synchronizing user space
483 with events happening in the kernel. Writing strings into
484 this file will be written into the ftrace buffer.
486 It is useful in applications to open this file at the start
487 of the application and just reference the file descriptor
490 void trace_write(const char *fmt, ...)
500 n = vsnprintf(buf, 256, fmt, ap);
503 write(trace_fd, buf, n);
508 trace_fd = open("trace_marker", WR_ONLY);
512 This is similar to trace_marker above, but is meant for for binary data
513 to be written to it, where a tool can be used to parse the data
518 Add dynamic tracepoints in programs.
523 Uprobe statistics. See uprobetrace.txt
527 This is a way to make multiple trace buffers where different
528 events can be recorded in different buffers.
529 See "Instances" section below.
533 This is the trace event directory. It holds event tracepoints
534 (also known as static tracepoints) that have been compiled
535 into the kernel. It shows what event tracepoints exist
536 and how they are grouped by system. There are "enable"
537 files at various levels that can enable the tracepoints
538 when a "1" is written to them.
540 See events.txt for more information.
544 By echoing in the event into this file, will enable that event.
546 See events.txt for more information.
550 A list of events that can be enabled in tracing.
552 See events.txt for more information.
556 Certain tracers may change the timestamp mode used when
557 logging trace events into the event buffer. Events with
558 different modes can coexist within a buffer but the mode in
559 effect when an event is logged determines which timestamp mode
560 is used for that event. The default timestamp mode is
563 Usual timestamp modes for tracing:
568 The timestamp mode with the square brackets around it is the
571 delta: Default timestamp mode - timestamp is a delta against
572 a per-buffer timestamp.
574 absolute: The timestamp is a full timestamp, not a delta
575 against some other value. As such it takes up more
576 space and is less efficient.
580 Directory for the Hardware Latency Detector.
581 See "Hardware Latency Detector" section below.
585 This is a directory that contains the trace per_cpu information.
587 per_cpu/cpu0/buffer_size_kb:
589 The ftrace buffer is defined per_cpu. That is, there's a separate
590 buffer for each CPU to allow writes to be done atomically,
591 and free from cache bouncing. These buffers may have different
592 size buffers. This file is similar to the buffer_size_kb
593 file, but it only displays or sets the buffer size for the
594 specific CPU. (here cpu0).
598 This is similar to the "trace" file, but it will only display
599 the data specific for the CPU. If written to, it only clears
600 the specific CPU buffer.
602 per_cpu/cpu0/trace_pipe
604 This is similar to the "trace_pipe" file, and is a consuming
605 read, but it will only display (and consume) the data specific
608 per_cpu/cpu0/trace_pipe_raw
610 For tools that can parse the ftrace ring buffer binary format,
611 the trace_pipe_raw file can be used to extract the data
612 from the ring buffer directly. With the use of the splice()
613 system call, the buffer data can be quickly transferred to
614 a file or to the network where a server is collecting the
617 Like trace_pipe, this is a consuming reader, where multiple
618 reads will always produce different data.
620 per_cpu/cpu0/snapshot:
622 This is similar to the main "snapshot" file, but will only
623 snapshot the current CPU (if supported). It only displays
624 the content of the snapshot for a given CPU, and if
625 written to, only clears this CPU buffer.
627 per_cpu/cpu0/snapshot_raw:
629 Similar to the trace_pipe_raw, but will read the binary format
630 from the snapshot buffer for the given CPU.
634 This displays certain stats about the ring buffer:
637 The number of events that are still in the buffer.
640 The number of lost events due to overwriting when
644 Should always be zero.
645 This gets set if so many events happened within a nested
646 event (ring buffer is re-entrant), that it fills the
647 buffer and starts dropping events.
650 Bytes actually read (not overwritten).
653 The oldest timestamp in the buffer
656 The current timestamp
659 Events lost due to overwrite option being off.
662 The number of events read.
667 Here is the list of current tracers that may be configured.
671 Function call tracer to trace all kernel functions.
675 Similar to the function tracer except that the
676 function tracer probes the functions on their entry
677 whereas the function graph tracer traces on both entry
678 and exit of the functions. It then provides the ability
679 to draw a graph of function calls similar to C code
684 The block tracer. The tracer used by the blktrace user
689 The Hardware Latency tracer is used to detect if the hardware
690 produces any latency. See "Hardware Latency Detector" section
695 Traces the areas that disable interrupts and saves
696 the trace with the longest max latency.
697 See tracing_max_latency. When a new max is recorded,
698 it replaces the old trace. It is best to view this
699 trace with the latency-format option enabled, which
700 happens automatically when the tracer is selected.
704 Similar to irqsoff but traces and records the amount of
705 time for which preemption is disabled.
709 Similar to irqsoff and preemptoff, but traces and
710 records the largest time for which irqs and/or preemption
715 Traces and records the max latency that it takes for
716 the highest priority task to get scheduled after
717 it has been woken up.
718 Traces all tasks as an average developer would expect.
722 Traces and records the max latency that it takes for just
723 RT tasks (as the current "wakeup" does). This is useful
724 for those interested in wake up timings of RT tasks.
728 Traces and records the max latency that it takes for
729 a SCHED_DEADLINE task to be woken (as the "wakeup" and
734 A special tracer that is used to trace binary module.
735 It will trace all the calls that a module makes to the
736 hardware. Everything it writes and reads from the I/O
741 This tracer can be configured when tracing likely/unlikely
742 calls within the kernel. It will trace when a likely and
743 unlikely branch is hit and if it was correct in its prediction
748 This is the "trace nothing" tracer. To remove all
749 tracers from tracing simply echo "nop" into
753 Examples of using the tracer
754 ----------------------------
756 Here are typical examples of using the tracers when controlling
757 them only with the tracefs interface (without using any
758 user-land utilities).
763 Here is an example of the output format of the file "trace"::
767 # entries-in-buffer/entries-written: 140080/250280 #P:4
770 # / _----=> need-resched
771 # | / _---=> hardirq/softirq
772 # || / _--=> preempt-depth
774 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
776 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
777 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
778 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
779 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
780 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
781 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
782 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
783 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
784 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
785 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
788 A header is printed with the tracer name that is represented by
789 the trace. In this case the tracer is "function". Then it shows the
790 number of events in the buffer as well as the total number of entries
791 that were written. The difference is the number of entries that were
792 lost due to the buffer filling up (250280 - 140080 = 110200 events
795 The header explains the content of the events. Task name "bash", the task
796 PID "1977", the CPU that it was running on "000", the latency format
797 (explained below), the timestamp in <secs>.<usecs> format, the
798 function name that was traced "sys_close" and the parent function that
799 called this function "system_call_fastpath". The timestamp is the time
800 at which the function was entered.
805 When the latency-format option is enabled or when one of the latency
806 tracers is set, the trace file gives somewhat more information to see
807 why a latency happened. Here is a typical trace::
811 # irqsoff latency trace v1.1.5 on 3.8.0-test+
812 # --------------------------------------------------------------------
813 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
815 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
817 # => started at: __lock_task_sighand
818 # => ended at: _raw_spin_unlock_irqrestore
822 # / _-----=> irqs-off
823 # | / _----=> need-resched
824 # || / _---=> hardirq/softirq
825 # ||| / _--=> preempt-depth
827 # cmd pid ||||| time | caller
829 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
830 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
831 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
832 ps-6143 2d..1 306us : <stack trace>
833 => trace_hardirqs_on_caller
835 => _raw_spin_unlock_irqrestore
842 => system_call_fastpath
845 This shows that the current tracer is "irqsoff" tracing the time
846 for which interrupts were disabled. It gives the trace version (which
847 never changes) and the version of the kernel upon which this was executed on
848 (3.8). Then it displays the max latency in microseconds (259 us). The number
849 of trace entries displayed and the total number (both are four: #4/4).
850 VP, KP, SP, and HP are always zero and are reserved for later use.
851 #P is the number of online CPUs (#P:4).
853 The task is the process that was running when the latency
854 occurred. (ps pid: 6143).
856 The start and stop (the functions in which the interrupts were
857 disabled and enabled respectively) that caused the latencies:
859 - __lock_task_sighand is where the interrupts were disabled.
860 - _raw_spin_unlock_irqrestore is where they were enabled again.
862 The next lines after the header are the trace itself. The header
863 explains which is which.
865 cmd: The name of the process in the trace.
867 pid: The PID of that process.
869 CPU#: The CPU which the process was running on.
871 irqs-off: 'd' interrupts are disabled. '.' otherwise.
872 .. caution:: If the architecture does not support a way to
873 read the irq flags variable, an 'X' will always
877 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
878 - 'n' only TIF_NEED_RESCHED is set,
879 - 'p' only PREEMPT_NEED_RESCHED is set,
883 - 'Z' - NMI occurred inside a hardirq
884 - 'z' - NMI is running
885 - 'H' - hard irq occurred inside a softirq.
886 - 'h' - hard irq is running
887 - 's' - soft irq is running
888 - '.' - normal context.
890 preempt-depth: The level of preempt_disabled
892 The above is mostly meaningful for kernel developers.
895 When the latency-format option is enabled, the trace file
896 output includes a timestamp relative to the start of the
897 trace. This differs from the output when latency-format
898 is disabled, which includes an absolute timestamp.
901 This is just to help catch your eye a bit better. And
902 needs to be fixed to be only relative to the same CPU.
903 The marks are determined by the difference between this
904 current trace and the next trace.
906 - '$' - greater than 1 second
907 - '@' - greater than 100 milisecond
908 - '*' - greater than 10 milisecond
909 - '#' - greater than 1000 microsecond
910 - '!' - greater than 100 microsecond
911 - '+' - greater than 10 microsecond
912 - ' ' - less than or equal to 10 microsecond.
914 The rest is the same as the 'trace' file.
916 Note, the latency tracers will usually end with a back trace
917 to easily find where the latency occurred.
922 The trace_options file (or the options directory) is used to control
923 what gets printed in the trace output, or manipulate the tracers.
924 To see what is available, simply cat the file::
955 To disable one of the options, echo in the option prepended with
958 echo noprint-parent > trace_options
960 To enable an option, leave off the "no"::
962 echo sym-offset > trace_options
964 Here are the available options:
967 On function traces, display the calling (parent)
968 function as well as the function being traced.
972 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
975 bash-4000 [01] 1477.606694: simple_strtoul
979 Display not only the function name, but also the
980 offset in the function. For example, instead of
981 seeing just "ktime_get", you will see
982 "ktime_get+0xb/0x20".
986 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
989 This will also display the function address as well
990 as the function name.
994 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
997 This deals with the trace file when the
998 latency-format option is enabled.
1001 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
1002 (+0.000ms): simple_strtoul (kstrtoul)
1005 This will display raw numbers. This option is best for
1006 use with user applications that can translate the raw
1007 numbers better than having it done in the kernel.
1010 Similar to raw, but the numbers will be in a hexadecimal format.
1013 This will print out the formats in raw binary.
1016 When set, reading trace_pipe will not block when polled.
1019 Can disable trace_printk() from writing into the buffer.
1022 It is sometimes confusing when the CPU buffers are full
1023 and one CPU buffer had a lot of events recently, thus
1024 a shorter time frame, were another CPU may have only had
1025 a few events, which lets it have older events. When
1026 the trace is reported, it shows the oldest events first,
1027 and it may look like only one CPU ran (the one with the
1028 oldest events). When the annotate option is set, it will
1029 display when a new CPU buffer started::
1031 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
1032 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
1033 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
1034 ##### CPU 2 buffer started ####
1035 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
1036 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
1037 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
1040 This option changes the trace. It records a
1041 stacktrace of the current user space thread after
1045 when user stacktrace are enabled, look up which
1046 object the address belongs to, and print a
1047 relative address. This is especially useful when
1048 ASLR is on, otherwise you don't get a chance to
1049 resolve the address to object/file/line after
1050 the app is no longer running
1052 The lookup is performed when you read
1053 trace,trace_pipe. Example::
1055 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
1056 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
1060 When set, trace_printk()s will only show the format
1061 and not their parameters (if trace_bprintk() or
1062 trace_bputs() was used to save the trace_printk()).
1065 Show only the event data. Hides the comm, PID,
1066 timestamp, CPU, and other useful data.
1069 This option changes the trace output. When it is enabled,
1070 the trace displays additional information about the
1071 latency, as described in "Latency trace format".
1074 When any event or tracer is enabled, a hook is enabled
1075 in the sched_switch trace point to fill comm cache
1076 with mapped pids and comms. But this may cause some
1077 overhead, and if you only care about pids, and not the
1078 name of the task, disabling this option can lower the
1079 impact of tracing. See "saved_cmdlines".
1082 When any event or tracer is enabled, a hook is enabled
1083 in the sched_switch trace point to fill the cache of
1084 mapped Thread Group IDs (TGID) mapping to pids. See
1088 This controls what happens when the trace buffer is
1089 full. If "1" (default), the oldest events are
1090 discarded and overwritten. If "0", then the newest
1091 events are discarded.
1092 (see per_cpu/cpu0/stats for overrun and dropped)
1095 When the free_buffer is closed, tracing will
1096 stop (tracing_on set to 0).
1099 Shows the interrupt, preempt count, need resched data.
1100 When disabled, the trace looks like::
1104 # entries-in-buffer/entries-written: 144405/9452052 #P:4
1106 # TASK-PID CPU# TIMESTAMP FUNCTION
1108 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
1109 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
1110 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
1114 When set, the trace_marker is writable (only by root).
1115 When disabled, the trace_marker will error with EINVAL
1119 When set, tasks with PIDs listed in set_event_pid will have
1120 the PIDs of their children added to set_event_pid when those
1121 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
1122 their PIDs will be removed from the file.
1125 The latency tracers will enable function tracing
1126 if this option is enabled (default it is). When
1127 it is disabled, the latency tracers do not trace
1128 functions. This keeps the overhead of the tracer down
1129 when performing latency tests.
1132 When set, tasks with PIDs listed in set_ftrace_pid will
1133 have the PIDs of their children added to set_ftrace_pid
1134 when those tasks fork. Also, when tasks with PIDs in
1135 set_ftrace_pid exit, their PIDs will be removed from the
1139 When set, the latency tracers (irqsoff, wakeup, etc) will
1140 use function graph tracing instead of function tracing.
1143 When set, a stack trace is recorded after any trace event
1147 Enable branch tracing with the tracer. This enables branch
1148 tracer along with the currently set tracer. Enabling this
1149 with the "nop" tracer is the same as just enabling the
1152 .. tip:: Some tracers have their own options. They only appear in this
1153 file when the tracer is active. They always appear in the
1157 Here are the per tracer options:
1159 Options for function tracer:
1162 When set, a stack trace is recorded after every
1163 function that is recorded. NOTE! Limit the functions
1164 that are recorded before enabling this, with
1165 "set_ftrace_filter" otherwise the system performance
1166 will be critically degraded. Remember to disable
1167 this option before clearing the function filter.
1169 Options for function_graph tracer:
1171 Since the function_graph tracer has a slightly different output
1172 it has its own options to control what is displayed.
1175 When set, the "overrun" of the graph stack is
1176 displayed after each function traced. The
1177 overrun, is when the stack depth of the calls
1178 is greater than what is reserved for each task.
1179 Each task has a fixed array of functions to
1180 trace in the call graph. If the depth of the
1181 calls exceeds that, the function is not traced.
1182 The overrun is the number of functions missed
1183 due to exceeding this array.
1186 When set, the CPU number of the CPU where the trace
1187 occurred is displayed.
1190 When set, if the function takes longer than
1191 A certain amount, then a delay marker is
1192 displayed. See "delay" above, under the
1196 Unlike other tracers, the process' command line
1197 is not displayed by default, but instead only
1198 when a task is traced in and out during a context
1199 switch. Enabling this options has the command
1200 of each process displayed at every line.
1203 At the end of each function (the return)
1204 the duration of the amount of time in the
1205 function is displayed in microseconds.
1208 When set, the timestamp is displayed at each line.
1211 When disabled, functions that happen inside an
1212 interrupt will not be traced.
1215 When set, the return event will include the function
1216 that it represents. By default this is off, and
1217 only a closing curly bracket "}" is displayed for
1218 the return of a function.
1221 When running function graph tracer, to include
1222 the time a task schedules out in its function.
1223 When enabled, it will account time the task has been
1224 scheduled out as part of the function call.
1227 When running function profiler with function graph tracer,
1228 to include the time to call nested functions. When this is
1229 not set, the time reported for the function will only
1230 include the time the function itself executed for, not the
1231 time for functions that it called.
1233 Options for blk tracer:
1236 Shows a more minimalistic output.
1242 When interrupts are disabled, the CPU can not react to any other
1243 external event (besides NMIs and SMIs). This prevents the timer
1244 interrupt from triggering or the mouse interrupt from letting
1245 the kernel know of a new mouse event. The result is a latency
1246 with the reaction time.
1248 The irqsoff tracer tracks the time for which interrupts are
1249 disabled. When a new maximum latency is hit, the tracer saves
1250 the trace leading up to that latency point so that every time a
1251 new maximum is reached, the old saved trace is discarded and the
1254 To reset the maximum, echo 0 into tracing_max_latency. Here is
1257 # echo 0 > options/function-trace
1258 # echo irqsoff > current_tracer
1259 # echo 1 > tracing_on
1260 # echo 0 > tracing_max_latency
1263 # echo 0 > tracing_on
1267 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1268 # --------------------------------------------------------------------
1269 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1271 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
1273 # => started at: run_timer_softirq
1274 # => ended at: run_timer_softirq
1278 # / _-----=> irqs-off
1279 # | / _----=> need-resched
1280 # || / _---=> hardirq/softirq
1281 # ||| / _--=> preempt-depth
1283 # cmd pid ||||| time | caller
1285 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
1286 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
1287 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
1288 <idle>-0 0dNs3 25us : <stack trace>
1289 => _raw_spin_unlock_irq
1290 => run_timer_softirq
1295 => smp_apic_timer_interrupt
1296 => apic_timer_interrupt
1301 => x86_64_start_reservations
1302 => x86_64_start_kernel
1304 Here we see that that we had a latency of 16 microseconds (which is
1305 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
1306 interrupts. The difference between the 16 and the displayed
1307 timestamp 25us occurred because the clock was incremented
1308 between the time of recording the max latency and the time of
1309 recording the function that had that latency.
1311 Note the above example had function-trace not set. If we set
1312 function-trace, we get a much larger output::
1314 with echo 1 > options/function-trace
1318 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1319 # --------------------------------------------------------------------
1320 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1322 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1324 # => started at: ata_scsi_queuecmd
1325 # => ended at: ata_scsi_queuecmd
1329 # / _-----=> irqs-off
1330 # | / _----=> need-resched
1331 # || / _---=> hardirq/softirq
1332 # ||| / _--=> preempt-depth
1334 # cmd pid ||||| time | caller
1336 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1337 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1338 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1339 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1340 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1341 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1342 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1343 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1344 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1346 bash-2042 3d..1 67us : delay_tsc <-__delay
1347 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1348 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1349 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1350 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1351 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1352 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1353 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1354 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1355 bash-2042 3d..1 120us : <stack trace>
1356 => _raw_spin_unlock_irqrestore
1357 => ata_scsi_queuecmd
1358 => scsi_dispatch_cmd
1360 => __blk_run_queue_uncond
1363 => generic_make_request
1366 => __ext3_get_inode_loc
1375 => user_path_at_empty
1380 => system_call_fastpath
1383 Here we traced a 71 microsecond latency. But we also see all the
1384 functions that were called during that time. Note that by
1385 enabling function tracing, we incur an added overhead. This
1386 overhead may extend the latency times. But nevertheless, this
1387 trace has provided some very helpful debugging information.
1393 When preemption is disabled, we may be able to receive
1394 interrupts but the task cannot be preempted and a higher
1395 priority task must wait for preemption to be enabled again
1396 before it can preempt a lower priority task.
1398 The preemptoff tracer traces the places that disable preemption.
1399 Like the irqsoff tracer, it records the maximum latency for
1400 which preemption was disabled. The control of preemptoff tracer
1401 is much like the irqsoff tracer.
1404 # echo 0 > options/function-trace
1405 # echo preemptoff > current_tracer
1406 # echo 1 > tracing_on
1407 # echo 0 > tracing_max_latency
1410 # echo 0 > tracing_on
1412 # tracer: preemptoff
1414 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1415 # --------------------------------------------------------------------
1416 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1418 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1420 # => started at: do_IRQ
1421 # => ended at: do_IRQ
1425 # / _-----=> irqs-off
1426 # | / _----=> need-resched
1427 # || / _---=> hardirq/softirq
1428 # ||| / _--=> preempt-depth
1430 # cmd pid ||||| time | caller
1432 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1433 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1434 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1435 sshd-1991 1d..1 52us : <stack trace>
1436 => sub_preempt_count
1442 This has some more changes. Preemption was disabled when an
1443 interrupt came in (notice the 'h'), and was enabled on exit.
1444 But we also see that interrupts have been disabled when entering
1445 the preempt off section and leaving it (the 'd'). We do not know if
1446 interrupts were enabled in the mean time or shortly after this
1450 # tracer: preemptoff
1452 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1453 # --------------------------------------------------------------------
1454 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1456 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1458 # => started at: wake_up_new_task
1459 # => ended at: task_rq_unlock
1463 # / _-----=> irqs-off
1464 # | / _----=> need-resched
1465 # || / _---=> hardirq/softirq
1466 # ||| / _--=> preempt-depth
1468 # cmd pid ||||| time | caller
1470 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1471 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1472 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1473 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1474 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1476 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1477 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1478 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1479 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1480 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1481 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1482 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1483 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1485 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1486 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1487 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1488 bash-1994 1d..2 36us : do_softirq <-irq_exit
1489 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1490 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1491 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1492 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1493 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1494 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1496 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1497 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1498 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1499 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1500 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1501 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1502 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1503 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1504 bash-1994 1.N.1 104us : <stack trace>
1505 => sub_preempt_count
1506 => _raw_spin_unlock_irqrestore
1514 The above is an example of the preemptoff trace with
1515 function-trace set. Here we see that interrupts were not disabled
1516 the entire time. The irq_enter code lets us know that we entered
1517 an interrupt 'h'. Before that, the functions being traced still
1518 show that it is not in an interrupt, but we can see from the
1519 functions themselves that this is not the case.
1524 Knowing the locations that have interrupts disabled or
1525 preemption disabled for the longest times is helpful. But
1526 sometimes we would like to know when either preemption and/or
1527 interrupts are disabled.
1529 Consider the following code::
1531 local_irq_disable();
1532 call_function_with_irqs_off();
1534 call_function_with_irqs_and_preemption_off();
1536 call_function_with_preemption_off();
1539 The irqsoff tracer will record the total length of
1540 call_function_with_irqs_off() and
1541 call_function_with_irqs_and_preemption_off().
1543 The preemptoff tracer will record the total length of
1544 call_function_with_irqs_and_preemption_off() and
1545 call_function_with_preemption_off().
1547 But neither will trace the time that interrupts and/or
1548 preemption is disabled. This total time is the time that we can
1549 not schedule. To record this time, use the preemptirqsoff
1552 Again, using this trace is much like the irqsoff and preemptoff
1556 # echo 0 > options/function-trace
1557 # echo preemptirqsoff > current_tracer
1558 # echo 1 > tracing_on
1559 # echo 0 > tracing_max_latency
1562 # echo 0 > tracing_on
1564 # tracer: preemptirqsoff
1566 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1567 # --------------------------------------------------------------------
1568 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1570 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1572 # => started at: ata_scsi_queuecmd
1573 # => ended at: ata_scsi_queuecmd
1577 # / _-----=> irqs-off
1578 # | / _----=> need-resched
1579 # || / _---=> hardirq/softirq
1580 # ||| / _--=> preempt-depth
1582 # cmd pid ||||| time | caller
1584 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1585 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1586 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1587 ls-2230 3...1 111us : <stack trace>
1588 => sub_preempt_count
1589 => _raw_spin_unlock_irqrestore
1590 => ata_scsi_queuecmd
1591 => scsi_dispatch_cmd
1593 => __blk_run_queue_uncond
1596 => generic_make_request
1601 => htree_dirblock_to_tree
1602 => ext3_htree_fill_tree
1606 => system_call_fastpath
1609 The trace_hardirqs_off_thunk is called from assembly on x86 when
1610 interrupts are disabled in the assembly code. Without the
1611 function tracing, we do not know if interrupts were enabled
1612 within the preemption points. We do see that it started with
1615 Here is a trace with function-trace set::
1617 # tracer: preemptirqsoff
1619 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1620 # --------------------------------------------------------------------
1621 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1623 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1625 # => started at: schedule
1626 # => ended at: mutex_unlock
1630 # / _-----=> irqs-off
1631 # | / _----=> need-resched
1632 # || / _---=> hardirq/softirq
1633 # ||| / _--=> preempt-depth
1635 # cmd pid ||||| time | caller
1637 kworker/-59 3...1 0us : __schedule <-schedule
1638 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1639 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1640 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1641 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1642 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1643 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1644 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1645 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1646 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1647 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1648 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1649 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1650 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1651 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1652 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1653 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1654 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1655 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1656 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1657 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1658 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1659 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1660 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1661 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1662 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1663 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1664 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1665 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1666 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1667 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1668 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1670 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1671 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1672 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1673 ls-2269 3d..3 21us : do_softirq <-irq_exit
1674 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1675 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1676 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1677 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1678 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1679 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1680 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1682 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1683 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1684 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1685 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1686 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1687 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1689 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1690 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1691 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1692 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1693 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1694 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1695 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1696 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1697 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1698 ls-2269 3d... 186us : <stack trace>
1699 => __mutex_unlock_slowpath
1706 => system_call_fastpath
1708 This is an interesting trace. It started with kworker running and
1709 scheduling out and ls taking over. But as soon as ls released the
1710 rq lock and enabled interrupts (but not preemption) an interrupt
1711 triggered. When the interrupt finished, it started running softirqs.
1712 But while the softirq was running, another interrupt triggered.
1713 When an interrupt is running inside a softirq, the annotation is 'H'.
1719 One common case that people are interested in tracing is the
1720 time it takes for a task that is woken to actually wake up.
1721 Now for non Real-Time tasks, this can be arbitrary. But tracing
1722 it none the less can be interesting.
1724 Without function tracing::
1726 # echo 0 > options/function-trace
1727 # echo wakeup > current_tracer
1728 # echo 1 > tracing_on
1729 # echo 0 > tracing_max_latency
1731 # echo 0 > tracing_on
1735 # wakeup latency trace v1.1.5 on 3.8.0-test+
1736 # --------------------------------------------------------------------
1737 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1739 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1743 # / _-----=> irqs-off
1744 # | / _----=> need-resched
1745 # || / _---=> hardirq/softirq
1746 # ||| / _--=> preempt-depth
1748 # cmd pid ||||| time | caller
1750 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1751 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1752 <idle>-0 3d..3 15us : __schedule <-schedule
1753 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1755 The tracer only traces the highest priority task in the system
1756 to avoid tracing the normal circumstances. Here we see that
1757 the kworker with a nice priority of -20 (not very nice), took
1758 just 15 microseconds from the time it woke up, to the time it
1761 Non Real-Time tasks are not that interesting. A more interesting
1762 trace is to concentrate only on Real-Time tasks.
1767 In a Real-Time environment it is very important to know the
1768 wakeup time it takes for the highest priority task that is woken
1769 up to the time that it executes. This is also known as "schedule
1770 latency". I stress the point that this is about RT tasks. It is
1771 also important to know the scheduling latency of non-RT tasks,
1772 but the average schedule latency is better for non-RT tasks.
1773 Tools like LatencyTop are more appropriate for such
1776 Real-Time environments are interested in the worst case latency.
1777 That is the longest latency it takes for something to happen,
1778 and not the average. We can have a very fast scheduler that may
1779 only have a large latency once in a while, but that would not
1780 work well with Real-Time tasks. The wakeup_rt tracer was designed
1781 to record the worst case wakeups of RT tasks. Non-RT tasks are
1782 not recorded because the tracer only records one worst case and
1783 tracing non-RT tasks that are unpredictable will overwrite the
1784 worst case latency of RT tasks (just run the normal wakeup
1785 tracer for a while to see that effect).
1787 Since this tracer only deals with RT tasks, we will run this
1788 slightly differently than we did with the previous tracers.
1789 Instead of performing an 'ls', we will run 'sleep 1' under
1790 'chrt' which changes the priority of the task.
1793 # echo 0 > options/function-trace
1794 # echo wakeup_rt > current_tracer
1795 # echo 1 > tracing_on
1796 # echo 0 > tracing_max_latency
1798 # echo 0 > tracing_on
1804 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1805 # --------------------------------------------------------------------
1806 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1808 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1812 # / _-----=> irqs-off
1813 # | / _----=> need-resched
1814 # || / _---=> hardirq/softirq
1815 # ||| / _--=> preempt-depth
1817 # cmd pid ||||| time | caller
1819 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1820 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1821 <idle>-0 3d..3 5us : __schedule <-schedule
1822 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1825 Running this on an idle system, we see that it only took 5 microseconds
1826 to perform the task switch. Note, since the trace point in the schedule
1827 is before the actual "switch", we stop the tracing when the recorded task
1828 is about to schedule in. This may change if we add a new marker at the
1829 end of the scheduler.
1831 Notice that the recorded task is 'sleep' with the PID of 2389
1832 and it has an rt_prio of 5. This priority is user-space priority
1833 and not the internal kernel priority. The policy is 1 for
1834 SCHED_FIFO and 2 for SCHED_RR.
1836 Note, that the trace data shows the internal priority (99 - rtprio).
1839 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1841 The 0:120:R means idle was running with a nice priority of 0 (120 - 120)
1842 and in the running state 'R'. The sleep task was scheduled in with
1843 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1844 and it too is in the running state.
1846 Doing the same with chrt -r 5 and function-trace set.
1849 echo 1 > options/function-trace
1853 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1854 # --------------------------------------------------------------------
1855 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1857 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
1861 # / _-----=> irqs-off
1862 # | / _----=> need-resched
1863 # || / _---=> hardirq/softirq
1864 # ||| / _--=> preempt-depth
1866 # cmd pid ||||| time | caller
1868 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
1869 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1870 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
1871 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
1872 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
1873 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
1874 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
1875 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
1876 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
1877 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1878 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
1879 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
1880 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
1881 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
1882 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
1883 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
1884 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
1885 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
1886 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
1887 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
1888 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
1889 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
1890 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
1891 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
1892 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
1893 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
1894 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
1895 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
1896 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
1897 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
1898 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
1899 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
1900 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
1901 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
1902 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
1903 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
1904 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
1905 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit
1906 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
1907 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
1908 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
1909 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
1910 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1911 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
1912 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
1913 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
1914 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
1915 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
1916 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
1917 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
1918 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
1919 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1920 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
1921 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1922 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1923 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
1924 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
1925 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
1926 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1927 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
1928 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
1929 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
1930 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
1931 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
1932 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
1933 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
1934 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
1935 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1936 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
1937 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
1938 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
1939 <idle>-0 3.N.. 25us : schedule <-cpu_idle
1940 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
1941 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
1942 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
1943 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
1944 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
1945 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
1946 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
1947 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
1948 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
1949 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
1950 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
1951 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
1952 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
1954 This isn't that big of a trace, even with function tracing enabled,
1955 so I included the entire trace.
1957 The interrupt went off while when the system was idle. Somewhere
1958 before task_woken_rt() was called, the NEED_RESCHED flag was set,
1959 this is indicated by the first occurrence of the 'N' flag.
1961 Latency tracing and events
1962 --------------------------
1963 As function tracing can induce a much larger latency, but without
1964 seeing what happens within the latency it is hard to know what
1965 caused it. There is a middle ground, and that is with enabling
1969 # echo 0 > options/function-trace
1970 # echo wakeup_rt > current_tracer
1971 # echo 1 > events/enable
1972 # echo 1 > tracing_on
1973 # echo 0 > tracing_max_latency
1975 # echo 0 > tracing_on
1979 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1980 # --------------------------------------------------------------------
1981 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1983 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
1987 # / _-----=> irqs-off
1988 # | / _----=> need-resched
1989 # || / _---=> hardirq/softirq
1990 # ||| / _--=> preempt-depth
1992 # cmd pid ||||| time | caller
1994 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
1995 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1996 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
1997 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
1998 <idle>-0 2.N.2 2us : power_end: cpu_id=2
1999 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
2000 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
2001 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
2002 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
2003 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
2004 <idle>-0 2d..3 6us : __schedule <-schedule
2005 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
2008 Hardware Latency Detector
2009 -------------------------
2011 The hardware latency detector is executed by enabling the "hwlat" tracer.
2013 NOTE, this tracer will affect the performance of the system as it will
2014 periodically make a CPU constantly busy with interrupts disabled.
2017 # echo hwlat > current_tracer
2023 # / _----=> need-resched
2024 # | / _---=> hardirq/softirq
2025 # || / _--=> preempt-depth
2027 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2029 <...>-3638 [001] d... 19452.055471: #1 inner/outer(us): 12/14 ts:1499801089.066141940
2030 <...>-3638 [003] d... 19454.071354: #2 inner/outer(us): 11/9 ts:1499801091.082164365
2031 <...>-3638 [002] dn.. 19461.126852: #3 inner/outer(us): 12/9 ts:1499801098.138150062
2032 <...>-3638 [001] d... 19488.340960: #4 inner/outer(us): 8/12 ts:1499801125.354139633
2033 <...>-3638 [003] d... 19494.388553: #5 inner/outer(us): 8/12 ts:1499801131.402150961
2034 <...>-3638 [003] d... 19501.283419: #6 inner/outer(us): 0/12 ts:1499801138.297435289 nmi-total:4 nmi-count:1
2037 The above output is somewhat the same in the header. All events will have
2038 interrupts disabled 'd'. Under the FUNCTION title there is:
2041 This is the count of events recorded that were greater than the
2042 tracing_threshold (See below).
2044 inner/outer(us): 12/14
2046 This shows two numbers as "inner latency" and "outer latency". The test
2047 runs in a loop checking a timestamp twice. The latency detected within
2048 the two timestamps is the "inner latency" and the latency detected
2049 after the previous timestamp and the next timestamp in the loop is
2050 the "outer latency".
2052 ts:1499801089.066141940
2054 The absolute timestamp that the event happened.
2056 nmi-total:4 nmi-count:1
2058 On architectures that support it, if an NMI comes in during the
2059 test, the time spent in NMI is reported in "nmi-total" (in
2062 All architectures that have NMIs will show the "nmi-count" if an
2063 NMI comes in during the test.
2068 This gets automatically set to "10" to represent 10
2069 microseconds. This is the threshold of latency that
2070 needs to be detected before the trace will be recorded.
2072 Note, when hwlat tracer is finished (another tracer is
2073 written into "current_tracer"), the original value for
2074 tracing_threshold is placed back into this file.
2076 hwlat_detector/width
2077 The length of time the test runs with interrupts disabled.
2079 hwlat_detector/window
2080 The length of time of the window which the test
2081 runs. That is, the test will run for "width"
2082 microseconds per "window" microseconds
2085 When the test is started. A kernel thread is created that
2086 runs the test. This thread will alternate between CPUs
2087 listed in the tracing_cpumask between each period
2088 (one "window"). To limit the test to specific CPUs
2089 set the mask in this file to only the CPUs that the test
2095 This tracer is the function tracer. Enabling the function tracer
2096 can be done from the debug file system. Make sure the
2097 ftrace_enabled is set; otherwise this tracer is a nop.
2098 See the "ftrace_enabled" section below.
2101 # sysctl kernel.ftrace_enabled=1
2102 # echo function > current_tracer
2103 # echo 1 > tracing_on
2105 # echo 0 > tracing_on
2109 # entries-in-buffer/entries-written: 24799/24799 #P:4
2112 # / _----=> need-resched
2113 # | / _---=> hardirq/softirq
2114 # || / _--=> preempt-depth
2116 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2118 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
2119 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
2120 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
2121 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
2122 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
2123 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
2124 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
2125 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
2129 Note: function tracer uses ring buffers to store the above
2130 entries. The newest data may overwrite the oldest data.
2131 Sometimes using echo to stop the trace is not sufficient because
2132 the tracing could have overwritten the data that you wanted to
2133 record. For this reason, it is sometimes better to disable
2134 tracing directly from a program. This allows you to stop the
2135 tracing at the point that you hit the part that you are
2136 interested in. To disable the tracing directly from a C program,
2137 something like following code snippet can be used::
2141 int main(int argc, char *argv[]) {
2143 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
2145 if (condition_hit()) {
2146 write(trace_fd, "0", 1);
2152 Single thread tracing
2153 ---------------------
2155 By writing into set_ftrace_pid you can trace a
2156 single thread. For example::
2158 # cat set_ftrace_pid
2160 # echo 3111 > set_ftrace_pid
2161 # cat set_ftrace_pid
2163 # echo function > current_tracer
2167 # TASK-PID CPU# TIMESTAMP FUNCTION
2169 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
2170 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
2171 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
2172 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
2173 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
2174 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
2175 # echo > set_ftrace_pid
2179 # TASK-PID CPU# TIMESTAMP FUNCTION
2181 ##### CPU 3 buffer started ####
2182 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
2183 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
2184 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
2185 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
2186 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
2188 If you want to trace a function when executing, you could use
2189 something like this simple program.
2194 #include <sys/types.h>
2195 #include <sys/stat.h>
2201 #define STR(x) _STR(x)
2202 #define MAX_PATH 256
2204 const char *find_tracefs(void)
2206 static char tracefs[MAX_PATH+1];
2207 static int tracefs_found;
2214 if ((fp = fopen("/proc/mounts","r")) == NULL) {
2215 perror("/proc/mounts");
2219 while (fscanf(fp, "%*s %"
2221 "s %99s %*s %*d %*d\n",
2222 tracefs, type) == 2) {
2223 if (strcmp(type, "tracefs") == 0)
2228 if (strcmp(type, "tracefs") != 0) {
2229 fprintf(stderr, "tracefs not mounted");
2233 strcat(tracefs, "/tracing/");
2239 const char *tracing_file(const char *file_name)
2241 static char trace_file[MAX_PATH+1];
2242 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name);
2246 int main (int argc, char **argv)
2256 ffd = open(tracing_file("current_tracer"), O_WRONLY);
2259 write(ffd, "nop", 3);
2261 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
2262 s = sprintf(line, "%d\n", getpid());
2265 write(ffd, "function", 8);
2270 execvp(argv[1], argv+1);
2276 Or this simple script!
2281 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts`
2282 echo nop > $tracefs/tracing/current_tracer
2283 echo 0 > $tracefs/tracing/tracing_on
2284 echo $$ > $tracefs/tracing/set_ftrace_pid
2285 echo function > $tracefs/tracing/current_tracer
2286 echo 1 > $tracefs/tracing/tracing_on
2290 function graph tracer
2291 ---------------------------
2293 This tracer is similar to the function tracer except that it
2294 probes a function on its entry and its exit. This is done by
2295 using a dynamically allocated stack of return addresses in each
2296 task_struct. On function entry the tracer overwrites the return
2297 address of each function traced to set a custom probe. Thus the
2298 original return address is stored on the stack of return address
2301 Probing on both ends of a function leads to special features
2304 - measure of a function's time execution
2305 - having a reliable call stack to draw function calls graph
2307 This tracer is useful in several situations:
2309 - you want to find the reason of a strange kernel behavior and
2310 need to see what happens in detail on any areas (or specific
2313 - you are experiencing weird latencies but it's difficult to
2316 - you want to find quickly which path is taken by a specific
2319 - you just want to peek inside a working kernel and want to see
2324 # tracer: function_graph
2326 # CPU DURATION FUNCTION CALLS
2330 0) | do_sys_open() {
2332 0) | kmem_cache_alloc() {
2333 0) 1.382 us | __might_sleep();
2335 0) | strncpy_from_user() {
2336 0) | might_fault() {
2337 0) 1.389 us | __might_sleep();
2342 0) 0.668 us | _spin_lock();
2343 0) 0.570 us | expand_files();
2344 0) 0.586 us | _spin_unlock();
2347 There are several columns that can be dynamically
2348 enabled/disabled. You can use every combination of options you
2349 want, depending on your needs.
2351 - The cpu number on which the function executed is default
2352 enabled. It is sometimes better to only trace one cpu (see
2353 tracing_cpu_mask file) or you might sometimes see unordered
2354 function calls while cpu tracing switch.
2356 - hide: echo nofuncgraph-cpu > trace_options
2357 - show: echo funcgraph-cpu > trace_options
2359 - The duration (function's time of execution) is displayed on
2360 the closing bracket line of a function or on the same line
2361 than the current function in case of a leaf one. It is default
2364 - hide: echo nofuncgraph-duration > trace_options
2365 - show: echo funcgraph-duration > trace_options
2367 - The overhead field precedes the duration field in case of
2368 reached duration thresholds.
2370 - hide: echo nofuncgraph-overhead > trace_options
2371 - show: echo funcgraph-overhead > trace_options
2372 - depends on: funcgraph-duration
2376 3) # 1837.709 us | } /* __switch_to */
2377 3) | finish_task_switch() {
2378 3) 0.313 us | _raw_spin_unlock_irq();
2380 3) # 1889.063 us | } /* __schedule */
2381 3) ! 140.417 us | } /* __schedule */
2382 3) # 2034.948 us | } /* schedule */
2383 3) * 33998.59 us | } /* schedule_preempt_disabled */
2387 1) 0.260 us | msecs_to_jiffies();
2388 1) 0.313 us | __rcu_read_unlock();
2391 1) 0.313 us | rcu_bh_qs();
2392 1) 0.313 us | __local_bh_enable();
2394 1) 0.365 us | idle_cpu();
2395 1) | rcu_irq_exit() {
2396 1) 0.417 us | rcu_eqs_enter_common.isra.47();
2400 1) @ 119760.2 us | }
2406 2) 0.417 us | scheduler_ipi();
2416 + means that the function exceeded 10 usecs.
2417 ! means that the function exceeded 100 usecs.
2418 # means that the function exceeded 1000 usecs.
2419 * means that the function exceeded 10 msecs.
2420 @ means that the function exceeded 100 msecs.
2421 $ means that the function exceeded 1 sec.
2424 - The task/pid field displays the thread cmdline and pid which
2425 executed the function. It is default disabled.
2427 - hide: echo nofuncgraph-proc > trace_options
2428 - show: echo funcgraph-proc > trace_options
2432 # tracer: function_graph
2434 # CPU TASK/PID DURATION FUNCTION CALLS
2436 0) sh-4802 | | d_free() {
2437 0) sh-4802 | | call_rcu() {
2438 0) sh-4802 | | __call_rcu() {
2439 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2440 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2441 0) sh-4802 | 2.899 us | }
2442 0) sh-4802 | 4.040 us | }
2443 0) sh-4802 | 5.151 us | }
2444 0) sh-4802 | + 49.370 us | }
2447 - The absolute time field is an absolute timestamp given by the
2448 system clock since it started. A snapshot of this time is
2449 given on each entry/exit of functions
2451 - hide: echo nofuncgraph-abstime > trace_options
2452 - show: echo funcgraph-abstime > trace_options
2457 # TIME CPU DURATION FUNCTION CALLS
2459 360.774522 | 1) 0.541 us | }
2460 360.774522 | 1) 4.663 us | }
2461 360.774523 | 1) 0.541 us | __wake_up_bit();
2462 360.774524 | 1) 6.796 us | }
2463 360.774524 | 1) 7.952 us | }
2464 360.774525 | 1) 9.063 us | }
2465 360.774525 | 1) 0.615 us | journal_mark_dirty();
2466 360.774527 | 1) 0.578 us | __brelse();
2467 360.774528 | 1) | reiserfs_prepare_for_journal() {
2468 360.774528 | 1) | unlock_buffer() {
2469 360.774529 | 1) | wake_up_bit() {
2470 360.774529 | 1) | bit_waitqueue() {
2471 360.774530 | 1) 0.594 us | __phys_addr();
2474 The function name is always displayed after the closing bracket
2475 for a function if the start of that function is not in the
2478 Display of the function name after the closing bracket may be
2479 enabled for functions whose start is in the trace buffer,
2480 allowing easier searching with grep for function durations.
2481 It is default disabled.
2483 - hide: echo nofuncgraph-tail > trace_options
2484 - show: echo funcgraph-tail > trace_options
2486 Example with nofuncgraph-tail (default)::
2489 0) | kmem_cache_free() {
2490 0) 0.518 us | __phys_addr();
2494 Example with funcgraph-tail::
2497 0) | kmem_cache_free() {
2498 0) 0.518 us | __phys_addr();
2499 0) 1.757 us | } /* kmem_cache_free() */
2500 0) 2.861 us | } /* putname() */
2502 You can put some comments on specific functions by using
2503 trace_printk() For example, if you want to put a comment inside
2504 the __might_sleep() function, you just have to include
2505 <linux/ftrace.h> and call trace_printk() inside __might_sleep()::
2507 trace_printk("I'm a comment!\n")
2511 1) | __might_sleep() {
2512 1) | /* I'm a comment! */
2516 You might find other useful features for this tracer in the
2517 following "dynamic ftrace" section such as tracing only specific
2523 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2524 virtually no overhead when function tracing is disabled. The way
2525 this works is the mcount function call (placed at the start of
2526 every kernel function, produced by the -pg switch in gcc),
2527 starts of pointing to a simple return. (Enabling FTRACE will
2528 include the -pg switch in the compiling of the kernel.)
2530 At compile time every C file object is run through the
2531 recordmcount program (located in the scripts directory). This
2532 program will parse the ELF headers in the C object to find all
2533 the locations in the .text section that call mcount. Starting
2534 with gcc verson 4.6, the -mfentry has been added for x86, which
2535 calls "__fentry__" instead of "mcount". Which is called before
2536 the creation of the stack frame.
2538 Note, not all sections are traced. They may be prevented by either
2539 a notrace, or blocked another way and all inline functions are not
2540 traced. Check the "available_filter_functions" file to see what functions
2543 A section called "__mcount_loc" is created that holds
2544 references to all the mcount/fentry call sites in the .text section.
2545 The recordmcount program re-links this section back into the
2546 original object. The final linking stage of the kernel will add all these
2547 references into a single table.
2549 On boot up, before SMP is initialized, the dynamic ftrace code
2550 scans this table and updates all the locations into nops. It
2551 also records the locations, which are added to the
2552 available_filter_functions list. Modules are processed as they
2553 are loaded and before they are executed. When a module is
2554 unloaded, it also removes its functions from the ftrace function
2555 list. This is automatic in the module unload code, and the
2556 module author does not need to worry about it.
2558 When tracing is enabled, the process of modifying the function
2559 tracepoints is dependent on architecture. The old method is to use
2560 kstop_machine to prevent races with the CPUs executing code being
2561 modified (which can cause the CPU to do undesirable things, especially
2562 if the modified code crosses cache (or page) boundaries), and the nops are
2563 patched back to calls. But this time, they do not call mcount
2564 (which is just a function stub). They now call into the ftrace
2567 The new method of modifying the function tracepoints is to place
2568 a breakpoint at the location to be modified, sync all CPUs, modify
2569 the rest of the instruction not covered by the breakpoint. Sync
2570 all CPUs again, and then remove the breakpoint with the finished
2571 version to the ftrace call site.
2573 Some archs do not even need to monkey around with the synchronization,
2574 and can just slap the new code on top of the old without any
2575 problems with other CPUs executing it at the same time.
2577 One special side-effect to the recording of the functions being
2578 traced is that we can now selectively choose which functions we
2579 wish to trace and which ones we want the mcount calls to remain
2582 Two files are used, one for enabling and one for disabling the
2583 tracing of specified functions. They are:
2591 A list of available functions that you can add to these files is
2594 available_filter_functions
2598 # cat available_filter_functions
2607 If I am only interested in sys_nanosleep and hrtimer_interrupt::
2609 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2610 # echo function > current_tracer
2611 # echo 1 > tracing_on
2613 # echo 0 > tracing_on
2617 # entries-in-buffer/entries-written: 5/5 #P:4
2620 # / _----=> need-resched
2621 # | / _---=> hardirq/softirq
2622 # || / _--=> preempt-depth
2624 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2626 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2627 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2628 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2629 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2630 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2632 To see which functions are being traced, you can cat the file:
2635 # cat set_ftrace_filter
2640 Perhaps this is not enough. The filters also allow glob(7) matching.
2643 will match functions that begin with <match>
2645 will match functions that end with <match>
2647 will match functions that have <match> in it
2648 ``<match1>*<match2>``
2649 will match functions that begin with <match1> and end with <match2>
2652 It is better to use quotes to enclose the wild cards,
2653 otherwise the shell may expand the parameters into names
2654 of files in the local directory.
2658 # echo 'hrtimer_*' > set_ftrace_filter
2664 # entries-in-buffer/entries-written: 897/897 #P:4
2667 # / _----=> need-resched
2668 # | / _---=> hardirq/softirq
2669 # || / _--=> preempt-depth
2671 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2673 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2674 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2675 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2676 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2677 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2678 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2679 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2680 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2682 Notice that we lost the sys_nanosleep.
2685 # cat set_ftrace_filter
2690 hrtimer_try_to_cancel
2694 hrtimer_force_reprogram
2695 hrtimer_get_next_event
2699 hrtimer_get_remaining
2701 hrtimer_init_sleeper
2704 This is because the '>' and '>>' act just like they do in bash.
2705 To rewrite the filters, use '>'
2706 To append to the filters, use '>>'
2708 To clear out a filter so that all functions will be recorded
2711 # echo > set_ftrace_filter
2712 # cat set_ftrace_filter
2715 Again, now we want to append.
2719 # echo sys_nanosleep > set_ftrace_filter
2720 # cat set_ftrace_filter
2722 # echo 'hrtimer_*' >> set_ftrace_filter
2723 # cat set_ftrace_filter
2728 hrtimer_try_to_cancel
2732 hrtimer_force_reprogram
2733 hrtimer_get_next_event
2738 hrtimer_get_remaining
2740 hrtimer_init_sleeper
2743 The set_ftrace_notrace prevents those functions from being
2747 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2753 # entries-in-buffer/entries-written: 39608/39608 #P:4
2756 # / _----=> need-resched
2757 # | / _---=> hardirq/softirq
2758 # || / _--=> preempt-depth
2760 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2762 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2763 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2764 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2765 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2766 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2767 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2768 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2769 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2770 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2771 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2772 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2773 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2775 We can see that there's no more lock or preempt tracing.
2778 Dynamic ftrace with the function graph tracer
2779 ---------------------------------------------
2781 Although what has been explained above concerns both the
2782 function tracer and the function-graph-tracer, there are some
2783 special features only available in the function-graph tracer.
2785 If you want to trace only one function and all of its children,
2786 you just have to echo its name into set_graph_function::
2788 echo __do_fault > set_graph_function
2790 will produce the following "expanded" trace of the __do_fault()
2794 0) | filemap_fault() {
2795 0) | find_lock_page() {
2796 0) 0.804 us | find_get_page();
2797 0) | __might_sleep() {
2801 0) 0.653 us | _spin_lock();
2802 0) 0.578 us | page_add_file_rmap();
2803 0) 0.525 us | native_set_pte_at();
2804 0) 0.585 us | _spin_unlock();
2805 0) | unlock_page() {
2806 0) 0.541 us | page_waitqueue();
2807 0) 0.639 us | __wake_up_bit();
2811 0) | filemap_fault() {
2812 0) | find_lock_page() {
2813 0) 0.698 us | find_get_page();
2814 0) | __might_sleep() {
2818 0) 0.631 us | _spin_lock();
2819 0) 0.571 us | page_add_file_rmap();
2820 0) 0.526 us | native_set_pte_at();
2821 0) 0.586 us | _spin_unlock();
2822 0) | unlock_page() {
2823 0) 0.533 us | page_waitqueue();
2824 0) 0.638 us | __wake_up_bit();
2828 You can also expand several functions at once::
2830 echo sys_open > set_graph_function
2831 echo sys_close >> set_graph_function
2833 Now if you want to go back to trace all functions you can clear
2834 this special filter via::
2836 echo > set_graph_function
2842 Note, the proc sysctl ftrace_enable is a big on/off switch for the
2843 function tracer. By default it is enabled (when function tracing is
2844 enabled in the kernel). If it is disabled, all function tracing is
2845 disabled. This includes not only the function tracers for ftrace, but
2846 also for any other uses (perf, kprobes, stack tracing, profiling, etc).
2848 Please disable this with care.
2850 This can be disable (and enabled) with::
2852 sysctl kernel.ftrace_enabled=0
2853 sysctl kernel.ftrace_enabled=1
2857 echo 0 > /proc/sys/kernel/ftrace_enabled
2858 echo 1 > /proc/sys/kernel/ftrace_enabled
2864 A few commands are supported by the set_ftrace_filter interface.
2865 Trace commands have the following format::
2867 <function>:<command>:<parameter>
2869 The following commands are supported:
2872 This command enables function filtering per module. The
2873 parameter defines the module. For example, if only the write*
2874 functions in the ext3 module are desired, run:
2876 echo 'write*:mod:ext3' > set_ftrace_filter
2878 This command interacts with the filter in the same way as
2879 filtering based on function names. Thus, adding more functions
2880 in a different module is accomplished by appending (>>) to the
2881 filter file. Remove specific module functions by prepending
2884 echo '!writeback*:mod:ext3' >> set_ftrace_filter
2886 Mod command supports module globbing. Disable tracing for all
2887 functions except a specific module::
2889 echo '!*:mod:!ext3' >> set_ftrace_filter
2891 Disable tracing for all modules, but still trace kernel::
2893 echo '!*:mod:*' >> set_ftrace_filter
2895 Enable filter only for kernel::
2897 echo '*write*:mod:!*' >> set_ftrace_filter
2899 Enable filter for module globbing::
2901 echo '*write*:mod:*snd*' >> set_ftrace_filter
2904 These commands turn tracing on and off when the specified
2905 functions are hit. The parameter determines how many times the
2906 tracing system is turned on and off. If unspecified, there is
2907 no limit. For example, to disable tracing when a schedule bug
2908 is hit the first 5 times, run::
2910 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
2912 To always disable tracing when __schedule_bug is hit::
2914 echo '__schedule_bug:traceoff' > set_ftrace_filter
2916 These commands are cumulative whether or not they are appended
2917 to set_ftrace_filter. To remove a command, prepend it by '!'
2918 and drop the parameter::
2920 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
2922 The above removes the traceoff command for __schedule_bug
2923 that have a counter. To remove commands without counters::
2925 echo '!__schedule_bug:traceoff' > set_ftrace_filter
2928 Will cause a snapshot to be triggered when the function is hit.
2931 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
2933 To only snapshot once:
2936 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
2938 To remove the above commands::
2940 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
2941 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
2943 - enable_event/disable_event:
2944 These commands can enable or disable a trace event. Note, because
2945 function tracing callbacks are very sensitive, when these commands
2946 are registered, the trace point is activated, but disabled in
2947 a "soft" mode. That is, the tracepoint will be called, but
2948 just will not be traced. The event tracepoint stays in this mode
2949 as long as there's a command that triggers it.
2952 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
2957 <function>:enable_event:<system>:<event>[:count]
2958 <function>:disable_event:<system>:<event>[:count]
2960 To remove the events commands::
2962 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
2964 echo '!schedule:disable_event:sched:sched_switch' > \
2968 When the function is hit, it will dump the contents of the ftrace
2969 ring buffer to the console. This is useful if you need to debug
2970 something, and want to dump the trace when a certain function
2971 is hit. Perhaps its a function that is called before a tripple
2972 fault happens and does not allow you to get a regular dump.
2975 When the function is hit, it will dump the contents of the ftrace
2976 ring buffer for the current CPU to the console. Unlike the "dump"
2977 command, it only prints out the contents of the ring buffer for the
2978 CPU that executed the function that triggered the dump.
2983 The trace_pipe outputs the same content as the trace file, but
2984 the effect on the tracing is different. Every read from
2985 trace_pipe is consumed. This means that subsequent reads will be
2986 different. The trace is live.
2989 # echo function > current_tracer
2990 # cat trace_pipe > /tmp/trace.out &
2992 # echo 1 > tracing_on
2994 # echo 0 > tracing_on
2998 # entries-in-buffer/entries-written: 0/0 #P:4
3001 # / _----=> need-resched
3002 # | / _---=> hardirq/softirq
3003 # || / _--=> preempt-depth
3005 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3009 # cat /tmp/trace.out
3010 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
3011 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
3012 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
3013 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
3014 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
3015 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
3016 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
3017 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
3018 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
3021 Note, reading the trace_pipe file will block until more input is
3027 Having too much or not enough data can be troublesome in
3028 diagnosing an issue in the kernel. The file buffer_size_kb is
3029 used to modify the size of the internal trace buffers. The
3030 number listed is the number of entries that can be recorded per
3031 CPU. To know the full size, multiply the number of possible CPUs
3032 with the number of entries.
3035 # cat buffer_size_kb
3036 1408 (units kilobytes)
3038 Or simply read buffer_total_size_kb
3041 # cat buffer_total_size_kb
3044 To modify the buffer, simple echo in a number (in 1024 byte segments).
3047 # echo 10000 > buffer_size_kb
3048 # cat buffer_size_kb
3049 10000 (units kilobytes)
3051 It will try to allocate as much as possible. If you allocate too
3052 much, it can cause Out-Of-Memory to trigger.
3055 # echo 1000000000000 > buffer_size_kb
3056 -bash: echo: write error: Cannot allocate memory
3057 # cat buffer_size_kb
3060 The per_cpu buffers can be changed individually as well:
3063 # echo 10000 > per_cpu/cpu0/buffer_size_kb
3064 # echo 100 > per_cpu/cpu1/buffer_size_kb
3066 When the per_cpu buffers are not the same, the buffer_size_kb
3067 at the top level will just show an X
3070 # cat buffer_size_kb
3073 This is where the buffer_total_size_kb is useful:
3076 # cat buffer_total_size_kb
3079 Writing to the top level buffer_size_kb will reset all the buffers
3080 to be the same again.
3084 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
3085 available to all non latency tracers. (Latency tracers which
3086 record max latency, such as "irqsoff" or "wakeup", can't use
3087 this feature, since those are already using the snapshot
3088 mechanism internally.)
3090 Snapshot preserves a current trace buffer at a particular point
3091 in time without stopping tracing. Ftrace swaps the current
3092 buffer with a spare buffer, and tracing continues in the new
3093 current (=previous spare) buffer.
3095 The following tracefs files in "tracing" are related to this
3100 This is used to take a snapshot and to read the output
3101 of the snapshot. Echo 1 into this file to allocate a
3102 spare buffer and to take a snapshot (swap), then read
3103 the snapshot from this file in the same format as
3104 "trace" (described above in the section "The File
3105 System"). Both reads snapshot and tracing are executable
3106 in parallel. When the spare buffer is allocated, echoing
3107 0 frees it, and echoing else (positive) values clear the
3109 More details are shown in the table below.
3111 +--------------+------------+------------+------------+
3112 |status\\input | 0 | 1 | else |
3113 +==============+============+============+============+
3114 |not allocated |(do nothing)| alloc+swap |(do nothing)|
3115 +--------------+------------+------------+------------+
3116 |allocated | free | swap | clear |
3117 +--------------+------------+------------+------------+
3119 Here is an example of using the snapshot feature.
3122 # echo 1 > events/sched/enable
3127 # entries-in-buffer/entries-written: 71/71 #P:8
3130 # / _----=> need-resched
3131 # | / _---=> hardirq/softirq
3132 # || / _--=> preempt-depth
3134 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3136 <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
3137 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
3139 <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
3144 # entries-in-buffer/entries-written: 77/77 #P:8
3147 # / _----=> need-resched
3148 # | / _---=> hardirq/softirq
3149 # || / _--=> preempt-depth
3151 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3153 <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
3154 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
3158 If you try to use this snapshot feature when current tracer is
3159 one of the latency tracers, you will get the following results.
3162 # echo wakeup > current_tracer
3164 bash: echo: write error: Device or resource busy
3166 cat: snapshot: Device or resource busy
3171 In the tracefs tracing directory is a directory called "instances".
3172 This directory can have new directories created inside of it using
3173 mkdir, and removing directories with rmdir. The directory created
3174 with mkdir in this directory will already contain files and other
3175 directories after it is created.
3178 # mkdir instances/foo
3180 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
3181 set_event snapshot trace trace_clock trace_marker trace_options
3182 trace_pipe tracing_on
3184 As you can see, the new directory looks similar to the tracing directory
3185 itself. In fact, it is very similar, except that the buffer and
3186 events are agnostic from the main director, or from any other
3187 instances that are created.
3189 The files in the new directory work just like the files with the
3190 same name in the tracing directory except the buffer that is used
3191 is a separate and new buffer. The files affect that buffer but do not
3192 affect the main buffer with the exception of trace_options. Currently,
3193 the trace_options affect all instances and the top level buffer
3194 the same, but this may change in future releases. That is, options
3195 may become specific to the instance they reside in.
3197 Notice that none of the function tracer files are there, nor is
3198 current_tracer and available_tracers. This is because the buffers
3199 can currently only have events enabled for them.
3202 # mkdir instances/foo
3203 # mkdir instances/bar
3204 # mkdir instances/zoot
3205 # echo 100000 > buffer_size_kb
3206 # echo 1000 > instances/foo/buffer_size_kb
3207 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
3208 # echo function > current_trace
3209 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
3210 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
3211 # echo 1 > instances/foo/events/sched/sched_switch/enable
3212 # echo 1 > instances/bar/events/irq/enable
3213 # echo 1 > instances/zoot/events/syscalls/enable
3215 CPU:2 [LOST 11745 EVENTS]
3216 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
3217 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
3218 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
3219 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
3220 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
3221 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
3222 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
3223 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
3224 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3225 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3226 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
3229 # cat instances/foo/trace_pipe
3230 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3231 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3232 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
3233 <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
3234 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
3235 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3236 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3237 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
3238 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
3239 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
3242 # cat instances/bar/trace_pipe
3243 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
3244 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
3245 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
3246 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
3247 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
3248 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
3249 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
3250 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
3251 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
3252 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
3253 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
3254 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
3257 # cat instances/zoot/trace
3260 # entries-in-buffer/entries-written: 18996/18996 #P:4
3263 # / _----=> need-resched
3264 # | / _---=> hardirq/softirq
3265 # || / _--=> preempt-depth
3267 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3269 bash-1998 [000] d... 140.733501: sys_write -> 0x2
3270 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
3271 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
3272 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
3273 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
3274 bash-1998 [000] d... 140.733510: sys_close(fd: a)
3275 bash-1998 [000] d... 140.733510: sys_close -> 0x0
3276 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
3277 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
3278 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
3279 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
3281 You can see that the trace of the top most trace buffer shows only
3282 the function tracing. The foo instance displays wakeups and task
3285 To remove the instances, simply delete their directories:
3288 # rmdir instances/foo
3289 # rmdir instances/bar
3290 # rmdir instances/zoot
3292 Note, if a process has a trace file open in one of the instance
3293 directories, the rmdir will fail with EBUSY.
3298 Since the kernel has a fixed sized stack, it is important not to
3299 waste it in functions. A kernel developer must be conscience of
3300 what they allocate on the stack. If they add too much, the system
3301 can be in danger of a stack overflow, and corruption will occur,
3302 usually leading to a system panic.
3304 There are some tools that check this, usually with interrupts
3305 periodically checking usage. But if you can perform a check
3306 at every function call that will become very useful. As ftrace provides
3307 a function tracer, it makes it convenient to check the stack size
3308 at every function call. This is enabled via the stack tracer.
3310 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
3311 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
3314 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
3316 You can also enable it from the kernel command line to trace
3317 the stack size of the kernel during boot up, by adding "stacktrace"
3318 to the kernel command line parameter.
3320 After running it for a few minutes, the output looks like:
3323 # cat stack_max_size
3327 Depth Size Location (18 entries)
3329 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
3330 1) 2704 160 find_busiest_group+0x31/0x1f1
3331 2) 2544 256 load_balance+0xd9/0x662
3332 3) 2288 80 idle_balance+0xbb/0x130
3333 4) 2208 128 __schedule+0x26e/0x5b9
3334 5) 2080 16 schedule+0x64/0x66
3335 6) 2064 128 schedule_timeout+0x34/0xe0
3336 7) 1936 112 wait_for_common+0x97/0xf1
3337 8) 1824 16 wait_for_completion+0x1d/0x1f
3338 9) 1808 128 flush_work+0xfe/0x119
3339 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
3340 11) 1664 48 input_available_p+0x1d/0x5c
3341 12) 1616 48 n_tty_poll+0x6d/0x134
3342 13) 1568 64 tty_poll+0x64/0x7f
3343 14) 1504 880 do_select+0x31e/0x511
3344 15) 624 400 core_sys_select+0x177/0x216
3345 16) 224 96 sys_select+0x91/0xb9
3346 17) 128 128 system_call_fastpath+0x16/0x1b
3348 Note, if -mfentry is being used by gcc, functions get traced before
3349 they set up the stack frame. This means that leaf level functions
3350 are not tested by the stack tracer when -mfentry is used.
3352 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
3356 More details can be found in the source code, in the `kernel/trace/*.c` files.
projects / linux-2.6-microblaze.git / blob