1 ========================
2 ftrace - Function Tracer
3 ========================
5 Copyright 2008 Red Hat Inc.
7 :Author: Steven Rostedt <srostedt@redhat.com>
8 :License: The GNU Free Documentation License, Version 1.2
9 (dual licensed under the GPL v2)
10 :Original Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton,
11 John Kacur, and David Teigland.
13 - Written for: 2.6.28-rc2
15 - Updated for: 4.13 - Copyright 2017 VMware Inc. Steven Rostedt
16 - Converted to rst format - Changbin Du <changbin.du@intel.com>
21 Ftrace is an internal tracer designed to help out developers and
22 designers of systems to find what is going on inside the kernel.
23 It can be used for debugging or analyzing latencies and
24 performance issues that take place outside of user-space.
26 Although ftrace is typically considered the function tracer, it
27 is really a framework of several assorted tracing utilities.
28 There's latency tracing to examine what occurs between interrupts
29 disabled and enabled, as well as for preemption and from a time
30 a task is woken to the task is actually scheduled in.
32 One of the most common uses of ftrace is the event tracing.
33 Throughout the kernel is hundreds of static event points that
34 can be enabled via the tracefs file system to see what is
35 going on in certain parts of the kernel.
37 See events.rst for more information.
40 Implementation Details
41 ----------------------
43 See Documentation/trace/ftrace-design.rst for details for arch porters and such.
49 Ftrace uses the tracefs file system to hold the control files as
50 well as the files to display output.
52 When tracefs is configured into the kernel (which selecting any ftrace
53 option will do) the directory /sys/kernel/tracing will be created. To mount
54 this directory, you can add to your /etc/fstab file::
56 tracefs /sys/kernel/tracing tracefs defaults 0 0
58 Or you can mount it at run time with::
60 mount -t tracefs nodev /sys/kernel/tracing
62 For quicker access to that directory you may want to make a soft link to
65 ln -s /sys/kernel/tracing /tracing
69 Before 4.1, all ftrace tracing control files were within the debugfs
70 file system, which is typically located at /sys/kernel/debug/tracing.
71 For backward compatibility, when mounting the debugfs file system,
72 the tracefs file system will be automatically mounted at:
74 /sys/kernel/debug/tracing
76 All files located in the tracefs file system will be located in that
77 debugfs file system directory as well.
81 Any selected ftrace option will also create the tracefs file system.
82 The rest of the document will assume that you are in the ftrace directory
83 (cd /sys/kernel/tracing) and will only concentrate on the files within that
84 directory and not distract from the content with the extended
85 "/sys/kernel/tracing" path name.
87 That's it! (assuming that you have ftrace configured into your kernel)
89 After mounting tracefs you will have access to the control and output files
90 of ftrace. Here is a list of some of the key files:
93 Note: all time values are in microseconds.
97 This is used to set or display the current tracer
98 that is configured. Changing the current tracer clears
99 the ring buffer content as well as the "snapshot" buffer.
103 This holds the different types of tracers that
104 have been compiled into the kernel. The
105 tracers listed here can be configured by
106 echoing their name into current_tracer.
110 This sets or displays whether writing to the trace
111 ring buffer is enabled. Echo 0 into this file to disable
112 the tracer or 1 to enable it. Note, this only disables
113 writing to the ring buffer, the tracing overhead may
116 The kernel function tracing_off() can be used within the
117 kernel to disable writing to the ring buffer, which will
118 set this file to "0". User space can re-enable tracing by
119 echoing "1" into the file.
121 Note, the function and event trigger "traceoff" will also
122 set this file to zero and stop tracing. Which can also
123 be re-enabled by user space using this file.
127 This file holds the output of the trace in a human
128 readable format (described below). Opening this file for
129 writing with the O_TRUNC flag clears the ring buffer content.
130 Note, this file is not a consumer. If tracing is off
131 (no tracer running, or tracing_on is zero), it will produce
132 the same output each time it is read. When tracing is on,
133 it may produce inconsistent results as it tries to read
134 the entire buffer without consuming it.
138 The output is the same as the "trace" file but this
139 file is meant to be streamed with live tracing.
140 Reads from this file will block until new data is
141 retrieved. Unlike the "trace" file, this file is a
142 consumer. This means reading from this file causes
143 sequential reads to display more current data. Once
144 data is read from this file, it is consumed, and
145 will not be read again with a sequential read. The
146 "trace" file is static, and if the tracer is not
147 adding more data, it will display the same
148 information every time it is read.
152 This file lets the user control the amount of data
153 that is displayed in one of the above output
154 files. Options also exist to modify how a tracer
155 or events work (stack traces, timestamps, etc).
159 This is a directory that has a file for every available
160 trace option (also in trace_options). Options may also be set
161 or cleared by writing a "1" or "0" respectively into the
162 corresponding file with the option name.
166 Some of the tracers record the max latency.
167 For example, the maximum time that interrupts are disabled.
168 The maximum time is saved in this file. The max trace will also be
169 stored, and displayed by "trace". A new max trace will only be
170 recorded if the latency is greater than the value in this file
173 By echoing in a time into this file, no latency will be recorded
174 unless it is greater than the time in this file.
178 Some latency tracers will record a trace whenever the
179 latency is greater than the number in this file.
180 Only active when the file contains a number greater than 0.
185 This is the watermark for how much the ring buffer needs to be filled
186 before a waiter is woken up. That is, if an application calls a
187 blocking read syscall on one of the per_cpu trace_pipe_raw files, it
188 will block until the given amount of data specified by buffer_percent
189 is in the ring buffer before it wakes the reader up. This also
190 controls how the splice system calls are blocked on this file::
192 0 - means to wake up as soon as there is any data in the ring buffer.
193 50 - means to wake up when roughly half of the ring buffer sub-buffers
195 100 - means to block until the ring buffer is totally full and is
196 about to start overwriting the older data.
200 This sets or displays the number of kilobytes each CPU
201 buffer holds. By default, the trace buffers are the same size
202 for each CPU. The displayed number is the size of the
203 CPU buffer and not total size of all buffers. The
204 trace buffers are allocated in pages (blocks of memory
205 that the kernel uses for allocation, usually 4 KB in size).
206 A few extra pages may be allocated to accommodate buffer management
207 meta-data. If the last page allocated has room for more bytes
208 than requested, the rest of the page will be used,
209 making the actual allocation bigger than requested or shown.
210 ( Note, the size may not be a multiple of the page size
211 due to buffer management meta-data. )
213 Buffer sizes for individual CPUs may vary
214 (see "per_cpu/cpu0/buffer_size_kb" below), and if they do
215 this file will show "X".
217 buffer_total_size_kb:
219 This displays the total combined size of all the trace buffers.
221 buffer_subbuf_size_kb:
223 This sets or displays the sub buffer size. The ring buffer is broken up
224 into several same size "sub buffers". An event can not be bigger than
225 the size of the sub buffer. Normally, the sub buffer is the size of the
226 architecture's page (4K on x86). The sub buffer also contains meta data
227 at the start which also limits the size of an event. That means when
228 the sub buffer is a page size, no event can be larger than the page
229 size minus the sub buffer meta data.
231 Note, the buffer_subbuf_size_kb is a way for the user to specify the
232 minimum size of the subbuffer. The kernel may make it bigger due to the
233 implementation details, or simply fail the operation if the kernel can
234 not handle the request.
236 Changing the sub buffer size allows for events to be larger than the
239 Note: When changing the sub-buffer size, tracing is stopped and any
240 data in the ring buffer and the snapshot buffer will be discarded.
244 If a process is performing tracing, and the ring buffer should be
245 shrunk "freed" when the process is finished, even if it were to be
246 killed by a signal, this file can be used for that purpose. On close
247 of this file, the ring buffer will be resized to its minimum size.
248 Having a process that is tracing also open this file, when the process
249 exits its file descriptor for this file will be closed, and in doing so,
250 the ring buffer will be "freed".
252 It may also stop tracing if disable_on_free option is set.
256 This is a mask that lets the user only trace on specified CPUs.
257 The format is a hex string representing the CPUs.
261 When dynamic ftrace is configured in (see the
262 section below "dynamic ftrace"), the code is dynamically
263 modified (code text rewrite) to disable calling of the
264 function profiler (mcount). This lets tracing be configured
265 in with practically no overhead in performance. This also
266 has a side effect of enabling or disabling specific functions
267 to be traced. Echoing names of functions into this file
268 will limit the trace to only those functions.
269 This influences the tracers "function" and "function_graph"
270 and thus also function profiling (see "function_profile_enabled").
272 The functions listed in "available_filter_functions" are what
273 can be written into this file.
275 This interface also allows for commands to be used. See the
276 "Filter commands" section for more details.
278 As a speed up, since processing strings can be quite expensive
279 and requires a check of all functions registered to tracing, instead
280 an index can be written into this file. A number (starting with "1")
281 written will instead select the same corresponding at the line position
282 of the "available_filter_functions" file.
286 This has an effect opposite to that of
287 set_ftrace_filter. Any function that is added here will not
288 be traced. If a function exists in both set_ftrace_filter
289 and set_ftrace_notrace, the function will _not_ be traced.
293 Have the function tracer only trace the threads whose PID are
296 If the "function-fork" option is set, then when a task whose
297 PID is listed in this file forks, the child's PID will
298 automatically be added to this file, and the child will be
299 traced by the function tracer as well. This option will also
300 cause PIDs of tasks that exit to be removed from the file.
302 set_ftrace_notrace_pid:
304 Have the function tracer ignore threads whose PID are listed in
307 If the "function-fork" option is set, then when a task whose
308 PID is listed in this file forks, the child's PID will
309 automatically be added to this file, and the child will not be
310 traced by the function tracer as well. This option will also
311 cause PIDs of tasks that exit to be removed from the file.
313 If a PID is in both this file and "set_ftrace_pid", then this
314 file takes precedence, and the thread will not be traced.
318 Have the events only trace a task with a PID listed in this file.
319 Note, sched_switch and sched_wake_up will also trace events
322 To have the PIDs of children of tasks with their PID in this file
323 added on fork, enable the "event-fork" option. That option will also
324 cause the PIDs of tasks to be removed from this file when the task
327 set_event_notrace_pid:
329 Have the events not trace a task with a PID listed in this file.
330 Note, sched_switch and sched_wakeup will trace threads not listed
331 in this file, even if a thread's PID is in the file if the
332 sched_switch or sched_wakeup events also trace a thread that should
335 To have the PIDs of children of tasks with their PID in this file
336 added on fork, enable the "event-fork" option. That option will also
337 cause the PIDs of tasks to be removed from this file when the task
342 Functions listed in this file will cause the function graph
343 tracer to only trace these functions and the functions that
344 they call. (See the section "dynamic ftrace" for more details).
345 Note, set_ftrace_filter and set_ftrace_notrace still affects
346 what functions are being traced.
350 Similar to set_graph_function, but will disable function graph
351 tracing when the function is hit until it exits the function.
352 This makes it possible to ignore tracing functions that are called
353 by a specific function.
355 available_filter_functions:
357 This lists the functions that ftrace has processed and can trace.
358 These are the function names that you can pass to
359 "set_ftrace_filter", "set_ftrace_notrace",
360 "set_graph_function", or "set_graph_notrace".
361 (See the section "dynamic ftrace" below for more details.)
363 available_filter_functions_addrs:
365 Similar to available_filter_functions, but with address displayed
366 for each function. The displayed address is the patch-site address
367 and can differ from /proc/kallsyms address.
369 dyn_ftrace_total_info:
371 This file is for debugging purposes. The number of functions that
372 have been converted to nops and are available to be traced.
376 This file is more for debugging ftrace, but can also be useful
377 in seeing if any function has a callback attached to it.
378 Not only does the trace infrastructure use ftrace function
379 trace utility, but other subsystems might too. This file
380 displays all functions that have a callback attached to them
381 as well as the number of callbacks that have been attached.
382 Note, a callback may also call multiple functions which will
383 not be listed in this count.
385 If the callback registered to be traced by a function with
386 the "save regs" attribute (thus even more overhead), a 'R'
387 will be displayed on the same line as the function that
388 is returning registers.
390 If the callback registered to be traced by a function with
391 the "ip modify" attribute (thus the regs->ip can be changed),
392 an 'I' will be displayed on the same line as the function that
395 If a non ftrace trampoline is attached (BPF) a 'D' will be displayed.
396 Note, normal ftrace trampolines can also be attached, but only one
397 "direct" trampoline can be attached to a given function at a time.
399 Some architectures can not call direct trampolines, but instead have
400 the ftrace ops function located above the function entry point. In
401 such cases an 'O' will be displayed.
403 If a function had either the "ip modify" or a "direct" call attached to
404 it in the past, a 'M' will be shown. This flag is never cleared. It is
405 used to know if a function was every modified by the ftrace infrastructure,
406 and can be used for debugging.
408 If the architecture supports it, it will also show what callback
409 is being directly called by the function. If the count is greater
410 than 1 it most likely will be ftrace_ops_list_func().
412 If the callback of a function jumps to a trampoline that is
413 specific to the callback and which is not the standard trampoline,
414 its address will be printed as well as the function that the
419 This file contains all the functions that ever had a function callback
420 to it via the ftrace infrastructure. It has the same format as
421 enabled_functions but shows all functions that have every been
424 To see any function that has every been modified by "ip modify" or a
425 direct trampoline, one can perform the following command:
427 grep ' M ' /sys/kernel/tracing/touched_functions
429 function_profile_enabled:
431 When set it will enable all functions with either the function
432 tracer, or if configured, the function graph tracer. It will
433 keep a histogram of the number of functions that were called
434 and if the function graph tracer was configured, it will also keep
435 track of the time spent in those functions. The histogram
436 content can be displayed in the files:
438 trace_stat/function<cpu> ( function0, function1, etc).
442 A directory that holds different tracing stats.
446 Enable dynamic trace points. See kprobetrace.rst.
450 Dynamic trace points stats. See kprobetrace.rst.
454 Used with the function graph tracer. This is the max depth
455 it will trace into a function. Setting this to a value of
456 one will show only the first kernel function that is called
461 This is for tools that read the raw format files. If an event in
462 the ring buffer references a string, only a pointer to the string
463 is recorded into the buffer and not the string itself. This prevents
464 tools from knowing what that string was. This file displays the string
465 and address for the string allowing tools to map the pointers to what
470 Only the pid of the task is recorded in a trace event unless
471 the event specifically saves the task comm as well. Ftrace
472 makes a cache of pid mappings to comms to try to display
473 comms for events. If a pid for a comm is not listed, then
474 "<...>" is displayed in the output.
476 If the option "record-cmd" is set to "0", then comms of tasks
477 will not be saved during recording. By default, it is enabled.
481 By default, 128 comms are saved (see "saved_cmdlines" above). To
482 increase or decrease the amount of comms that are cached, echo
483 the number of comms to cache into this file.
487 If the option "record-tgid" is set, on each scheduling context switch
488 the Task Group ID of a task is saved in a table mapping the PID of
489 the thread to its TGID. By default, the "record-tgid" option is
494 This displays the "snapshot" buffer and also lets the user
495 take a snapshot of the current running trace.
496 See the "Snapshot" section below for more details.
500 When the stack tracer is activated, this will display the
501 maximum stack size it has encountered.
502 See the "Stack Trace" section below.
506 This displays the stack back trace of the largest stack
507 that was encountered when the stack tracer is activated.
508 See the "Stack Trace" section below.
512 This is similar to "set_ftrace_filter" but it limits what
513 functions the stack tracer will check.
517 Whenever an event is recorded into the ring buffer, a
518 "timestamp" is added. This stamp comes from a specified
519 clock. By default, ftrace uses the "local" clock. This
520 clock is very fast and strictly per cpu, but on some
521 systems it may not be monotonic with respect to other
522 CPUs. In other words, the local clocks may not be in sync
523 with local clocks on other CPUs.
525 Usual clocks for tracing::
528 [local] global counter x86-tsc
530 The clock with the square brackets around it is the one in effect.
533 Default clock, but may not be in sync across CPUs
536 This clock is in sync with all CPUs but may
537 be a bit slower than the local clock.
540 This is not a clock at all, but literally an atomic
541 counter. It counts up one by one, but is in sync
542 with all CPUs. This is useful when you need to
543 know exactly the order events occurred with respect to
544 each other on different CPUs.
547 This uses the jiffies counter and the time stamp
548 is relative to the time since boot up.
551 This makes ftrace use the same clock that perf uses.
552 Eventually perf will be able to read ftrace buffers
553 and this will help out in interleaving the data.
556 Architectures may define their own clocks. For
557 example, x86 uses its own TSC cycle clock here.
560 This uses the powerpc timebase register value.
561 This is in sync across CPUs and can also be used
562 to correlate events across hypervisor/guest if
566 This uses the fast monotonic clock (CLOCK_MONOTONIC)
567 which is monotonic and is subject to NTP rate adjustments.
570 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
571 which is monotonic but is not subject to any rate adjustments
572 and ticks at the same rate as the hardware clocksource.
575 This is the boot clock (CLOCK_BOOTTIME) and is based on the
576 fast monotonic clock, but also accounts for time spent in
577 suspend. Since the clock access is designed for use in
578 tracing in the suspend path, some side effects are possible
579 if clock is accessed after the suspend time is accounted before
580 the fast mono clock is updated. In this case, the clock update
581 appears to happen slightly sooner than it normally would have.
582 Also on 32-bit systems, it's possible that the 64-bit boot offset
583 sees a partial update. These effects are rare and post
584 processing should be able to handle them. See comments in the
585 ktime_get_boot_fast_ns() function for more information.
588 This is the tai clock (CLOCK_TAI) and is derived from the wall-
589 clock time. However, this clock does not experience
590 discontinuities and backwards jumps caused by NTP inserting leap
591 seconds. Since the clock access is designed for use in tracing,
592 side effects are possible. The clock access may yield wrong
593 readouts in case the internal TAI offset is updated e.g., caused
594 by setting the system time or using adjtimex() with an offset.
595 These effects are rare and post processing should be able to
596 handle them. See comments in the ktime_get_tai_fast_ns()
597 function for more information.
599 To set a clock, simply echo the clock name into this file::
601 # echo global > trace_clock
603 Setting a clock clears the ring buffer content as well as the
608 This is a very useful file for synchronizing user space
609 with events happening in the kernel. Writing strings into
610 this file will be written into the ftrace buffer.
612 It is useful in applications to open this file at the start
613 of the application and just reference the file descriptor
616 void trace_write(const char *fmt, ...)
626 n = vsnprintf(buf, 256, fmt, ap);
629 write(trace_fd, buf, n);
634 trace_fd = open("trace_marker", O_WRONLY);
636 Note: Writing into the trace_marker file can also initiate triggers
637 that are written into /sys/kernel/tracing/events/ftrace/print/trigger
638 See "Event triggers" in Documentation/trace/events.rst and an
639 example in Documentation/trace/histogram.rst (Section 3.)
643 This is similar to trace_marker above, but is meant for binary data
644 to be written to it, where a tool can be used to parse the data
649 Add dynamic tracepoints in programs.
654 Uprobe statistics. See uprobetrace.txt
658 This is a way to make multiple trace buffers where different
659 events can be recorded in different buffers.
660 See "Instances" section below.
664 This is the trace event directory. It holds event tracepoints
665 (also known as static tracepoints) that have been compiled
666 into the kernel. It shows what event tracepoints exist
667 and how they are grouped by system. There are "enable"
668 files at various levels that can enable the tracepoints
669 when a "1" is written to them.
671 See events.rst for more information.
675 By echoing in the event into this file, will enable that event.
677 See events.rst for more information.
681 A list of events that can be enabled in tracing.
683 See events.rst for more information.
687 Certain tracers may change the timestamp mode used when
688 logging trace events into the event buffer. Events with
689 different modes can coexist within a buffer but the mode in
690 effect when an event is logged determines which timestamp mode
691 is used for that event. The default timestamp mode is
694 Usual timestamp modes for tracing:
699 The timestamp mode with the square brackets around it is the
702 delta: Default timestamp mode - timestamp is a delta against
703 a per-buffer timestamp.
705 absolute: The timestamp is a full timestamp, not a delta
706 against some other value. As such it takes up more
707 space and is less efficient.
711 Directory for the Hardware Latency Detector.
712 See "Hardware Latency Detector" section below.
716 This is a directory that contains the trace per_cpu information.
718 per_cpu/cpu0/buffer_size_kb:
720 The ftrace buffer is defined per_cpu. That is, there's a separate
721 buffer for each CPU to allow writes to be done atomically,
722 and free from cache bouncing. These buffers may have different
723 size buffers. This file is similar to the buffer_size_kb
724 file, but it only displays or sets the buffer size for the
725 specific CPU. (here cpu0).
729 This is similar to the "trace" file, but it will only display
730 the data specific for the CPU. If written to, it only clears
731 the specific CPU buffer.
733 per_cpu/cpu0/trace_pipe
735 This is similar to the "trace_pipe" file, and is a consuming
736 read, but it will only display (and consume) the data specific
739 per_cpu/cpu0/trace_pipe_raw
741 For tools that can parse the ftrace ring buffer binary format,
742 the trace_pipe_raw file can be used to extract the data
743 from the ring buffer directly. With the use of the splice()
744 system call, the buffer data can be quickly transferred to
745 a file or to the network where a server is collecting the
748 Like trace_pipe, this is a consuming reader, where multiple
749 reads will always produce different data.
751 per_cpu/cpu0/snapshot:
753 This is similar to the main "snapshot" file, but will only
754 snapshot the current CPU (if supported). It only displays
755 the content of the snapshot for a given CPU, and if
756 written to, only clears this CPU buffer.
758 per_cpu/cpu0/snapshot_raw:
760 Similar to the trace_pipe_raw, but will read the binary format
761 from the snapshot buffer for the given CPU.
765 This displays certain stats about the ring buffer:
768 The number of events that are still in the buffer.
771 The number of lost events due to overwriting when
775 Should always be zero.
776 This gets set if so many events happened within a nested
777 event (ring buffer is re-entrant), that it fills the
778 buffer and starts dropping events.
781 Bytes actually read (not overwritten).
784 The oldest timestamp in the buffer
787 The current timestamp
790 Events lost due to overwrite option being off.
793 The number of events read.
798 Here is the list of current tracers that may be configured.
802 Function call tracer to trace all kernel functions.
806 Similar to the function tracer except that the
807 function tracer probes the functions on their entry
808 whereas the function graph tracer traces on both entry
809 and exit of the functions. It then provides the ability
810 to draw a graph of function calls similar to C code
815 The block tracer. The tracer used by the blktrace user
820 The Hardware Latency tracer is used to detect if the hardware
821 produces any latency. See "Hardware Latency Detector" section
826 Traces the areas that disable interrupts and saves
827 the trace with the longest max latency.
828 See tracing_max_latency. When a new max is recorded,
829 it replaces the old trace. It is best to view this
830 trace with the latency-format option enabled, which
831 happens automatically when the tracer is selected.
835 Similar to irqsoff but traces and records the amount of
836 time for which preemption is disabled.
840 Similar to irqsoff and preemptoff, but traces and
841 records the largest time for which irqs and/or preemption
846 Traces and records the max latency that it takes for
847 the highest priority task to get scheduled after
848 it has been woken up.
849 Traces all tasks as an average developer would expect.
853 Traces and records the max latency that it takes for just
854 RT tasks (as the current "wakeup" does). This is useful
855 for those interested in wake up timings of RT tasks.
859 Traces and records the max latency that it takes for
860 a SCHED_DEADLINE task to be woken (as the "wakeup" and
865 A special tracer that is used to trace binary module.
866 It will trace all the calls that a module makes to the
867 hardware. Everything it writes and reads from the I/O
872 This tracer can be configured when tracing likely/unlikely
873 calls within the kernel. It will trace when a likely and
874 unlikely branch is hit and if it was correct in its prediction
879 This is the "trace nothing" tracer. To remove all
880 tracers from tracing simply echo "nop" into
886 For most ftrace commands, failure modes are obvious and communicated
887 using standard return codes.
889 For other more involved commands, extended error information may be
890 available via the tracing/error_log file. For the commands that
891 support it, reading the tracing/error_log file after an error will
892 display more detailed information about what went wrong, if
893 information is available. The tracing/error_log file is a circular
894 error log displaying a small number (currently, 8) of ftrace errors
895 for the last (8) failed commands.
897 The extended error information and usage takes the form shown in
900 # echo xxx > /sys/kernel/tracing/events/sched/sched_wakeup/trigger
901 echo: write error: Invalid argument
903 # cat /sys/kernel/tracing/error_log
904 [ 5348.887237] location: error: Couldn't yyy: zzz
907 [ 7517.023364] location: error: Bad rrr: sss
911 To clear the error log, echo the empty string into it::
913 # echo > /sys/kernel/tracing/error_log
915 Examples of using the tracer
916 ----------------------------
918 Here are typical examples of using the tracers when controlling
919 them only with the tracefs interface (without using any
920 user-land utilities).
925 Here is an example of the output format of the file "trace"::
929 # entries-in-buffer/entries-written: 140080/250280 #P:4
932 # / _----=> need-resched
933 # | / _---=> hardirq/softirq
934 # || / _--=> preempt-depth
936 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
938 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
939 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
940 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
941 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
942 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
943 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
944 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
945 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
946 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
947 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
950 A header is printed with the tracer name that is represented by
951 the trace. In this case the tracer is "function". Then it shows the
952 number of events in the buffer as well as the total number of entries
953 that were written. The difference is the number of entries that were
954 lost due to the buffer filling up (250280 - 140080 = 110200 events
957 The header explains the content of the events. Task name "bash", the task
958 PID "1977", the CPU that it was running on "000", the latency format
959 (explained below), the timestamp in <secs>.<usecs> format, the
960 function name that was traced "sys_close" and the parent function that
961 called this function "system_call_fastpath". The timestamp is the time
962 at which the function was entered.
967 When the latency-format option is enabled or when one of the latency
968 tracers is set, the trace file gives somewhat more information to see
969 why a latency happened. Here is a typical trace::
973 # irqsoff latency trace v1.1.5 on 3.8.0-test+
974 # --------------------------------------------------------------------
975 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
977 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
979 # => started at: __lock_task_sighand
980 # => ended at: _raw_spin_unlock_irqrestore
984 # / _-----=> irqs-off
985 # | / _----=> need-resched
986 # || / _---=> hardirq/softirq
987 # ||| / _--=> preempt-depth
989 # cmd pid ||||| time | caller
991 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
992 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
993 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
994 ps-6143 2d..1 306us : <stack trace>
995 => trace_hardirqs_on_caller
997 => _raw_spin_unlock_irqrestore
1004 => system_call_fastpath
1007 This shows that the current tracer is "irqsoff" tracing the time
1008 for which interrupts were disabled. It gives the trace version (which
1009 never changes) and the version of the kernel upon which this was executed on
1010 (3.8). Then it displays the max latency in microseconds (259 us). The number
1011 of trace entries displayed and the total number (both are four: #4/4).
1012 VP, KP, SP, and HP are always zero and are reserved for later use.
1013 #P is the number of online CPUs (#P:4).
1015 The task is the process that was running when the latency
1016 occurred. (ps pid: 6143).
1018 The start and stop (the functions in which the interrupts were
1019 disabled and enabled respectively) that caused the latencies:
1021 - __lock_task_sighand is where the interrupts were disabled.
1022 - _raw_spin_unlock_irqrestore is where they were enabled again.
1024 The next lines after the header are the trace itself. The header
1025 explains which is which.
1027 cmd: The name of the process in the trace.
1029 pid: The PID of that process.
1031 CPU#: The CPU which the process was running on.
1033 irqs-off: 'd' interrupts are disabled. '.' otherwise.
1034 .. caution:: If the architecture does not support a way to
1035 read the irq flags variable, an 'X' will always
1039 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
1040 - 'n' only TIF_NEED_RESCHED is set,
1041 - 'p' only PREEMPT_NEED_RESCHED is set,
1045 - 'Z' - NMI occurred inside a hardirq
1046 - 'z' - NMI is running
1047 - 'H' - hard irq occurred inside a softirq.
1048 - 'h' - hard irq is running
1049 - 's' - soft irq is running
1050 - '.' - normal context.
1052 preempt-depth: The level of preempt_disabled
1054 The above is mostly meaningful for kernel developers.
1057 When the latency-format option is enabled, the trace file
1058 output includes a timestamp relative to the start of the
1059 trace. This differs from the output when latency-format
1060 is disabled, which includes an absolute timestamp.
1063 This is just to help catch your eye a bit better. And
1064 needs to be fixed to be only relative to the same CPU.
1065 The marks are determined by the difference between this
1066 current trace and the next trace.
1068 - '$' - greater than 1 second
1069 - '@' - greater than 100 millisecond
1070 - '*' - greater than 10 millisecond
1071 - '#' - greater than 1000 microsecond
1072 - '!' - greater than 100 microsecond
1073 - '+' - greater than 10 microsecond
1074 - ' ' - less than or equal to 10 microsecond.
1076 The rest is the same as the 'trace' file.
1078 Note, the latency tracers will usually end with a back trace
1079 to easily find where the latency occurred.
1084 The trace_options file (or the options directory) is used to control
1085 what gets printed in the trace output, or manipulate the tracers.
1086 To see what is available, simply cat the file::
1118 To disable one of the options, echo in the option prepended with
1121 echo noprint-parent > trace_options
1123 To enable an option, leave off the "no"::
1125 echo sym-offset > trace_options
1127 Here are the available options:
1130 On function traces, display the calling (parent)
1131 function as well as the function being traced.
1135 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
1138 bash-4000 [01] 1477.606694: simple_strtoul
1142 Display not only the function name, but also the
1143 offset in the function. For example, instead of
1144 seeing just "ktime_get", you will see
1145 "ktime_get+0xb/0x20".
1149 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
1152 This will also display the function address as well
1153 as the function name.
1157 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
1160 This deals with the trace file when the
1161 latency-format option is enabled.
1164 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
1165 (+0.000ms): simple_strtoul (kstrtoul)
1168 This will display raw numbers. This option is best for
1169 use with user applications that can translate the raw
1170 numbers better than having it done in the kernel.
1173 Similar to raw, but the numbers will be in a hexadecimal format.
1176 This will print out the formats in raw binary.
1179 When set, reading trace_pipe will not block when polled.
1182 Print the fields as described by their types. This is a better
1183 option than using hex, bin or raw, as it gives a better parsing
1184 of the content of the event.
1187 Can disable trace_printk() from writing into the buffer.
1190 It is sometimes confusing when the CPU buffers are full
1191 and one CPU buffer had a lot of events recently, thus
1192 a shorter time frame, were another CPU may have only had
1193 a few events, which lets it have older events. When
1194 the trace is reported, it shows the oldest events first,
1195 and it may look like only one CPU ran (the one with the
1196 oldest events). When the annotate option is set, it will
1197 display when a new CPU buffer started::
1199 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
1200 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
1201 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
1202 ##### CPU 2 buffer started ####
1203 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
1204 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
1205 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
1208 This option changes the trace. It records a
1209 stacktrace of the current user space thread after
1213 when user stacktrace are enabled, look up which
1214 object the address belongs to, and print a
1215 relative address. This is especially useful when
1216 ASLR is on, otherwise you don't get a chance to
1217 resolve the address to object/file/line after
1218 the app is no longer running
1220 The lookup is performed when you read
1221 trace,trace_pipe. Example::
1223 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
1224 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
1228 When set, trace_printk()s will only show the format
1229 and not their parameters (if trace_bprintk() or
1230 trace_bputs() was used to save the trace_printk()).
1233 Show only the event data. Hides the comm, PID,
1234 timestamp, CPU, and other useful data.
1237 This option changes the trace output. When it is enabled,
1238 the trace displays additional information about the
1239 latency, as described in "Latency trace format".
1242 When set, opening the trace file for read, will pause
1243 writing to the ring buffer (as if tracing_on was set to zero).
1244 This simulates the original behavior of the trace file.
1245 When the file is closed, tracing will be enabled again.
1248 When set, "%p" in the event printk format displays the
1249 hashed pointer value instead of real address.
1250 This will be useful if you want to find out which hashed
1251 value is corresponding to the real value in trace log.
1254 When any event or tracer is enabled, a hook is enabled
1255 in the sched_switch trace point to fill comm cache
1256 with mapped pids and comms. But this may cause some
1257 overhead, and if you only care about pids, and not the
1258 name of the task, disabling this option can lower the
1259 impact of tracing. See "saved_cmdlines".
1262 When any event or tracer is enabled, a hook is enabled
1263 in the sched_switch trace point to fill the cache of
1264 mapped Thread Group IDs (TGID) mapping to pids. See
1268 This controls what happens when the trace buffer is
1269 full. If "1" (default), the oldest events are
1270 discarded and overwritten. If "0", then the newest
1271 events are discarded.
1272 (see per_cpu/cpu0/stats for overrun and dropped)
1275 When the free_buffer is closed, tracing will
1276 stop (tracing_on set to 0).
1279 Shows the interrupt, preempt count, need resched data.
1280 When disabled, the trace looks like::
1284 # entries-in-buffer/entries-written: 144405/9452052 #P:4
1286 # TASK-PID CPU# TIMESTAMP FUNCTION
1288 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
1289 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
1290 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
1294 When set, the trace_marker is writable (only by root).
1295 When disabled, the trace_marker will error with EINVAL
1299 When set, tasks with PIDs listed in set_event_pid will have
1300 the PIDs of their children added to set_event_pid when those
1301 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
1302 their PIDs will be removed from the file.
1304 This affects PIDs listed in set_event_notrace_pid as well.
1307 The latency tracers will enable function tracing
1308 if this option is enabled (default it is). When
1309 it is disabled, the latency tracers do not trace
1310 functions. This keeps the overhead of the tracer down
1311 when performing latency tests.
1314 When set, tasks with PIDs listed in set_ftrace_pid will
1315 have the PIDs of their children added to set_ftrace_pid
1316 when those tasks fork. Also, when tasks with PIDs in
1317 set_ftrace_pid exit, their PIDs will be removed from the
1320 This affects PIDs in set_ftrace_notrace_pid as well.
1323 When set, the latency tracers (irqsoff, wakeup, etc) will
1324 use function graph tracing instead of function tracing.
1327 When set, a stack trace is recorded after any trace event
1331 Enable branch tracing with the tracer. This enables branch
1332 tracer along with the currently set tracer. Enabling this
1333 with the "nop" tracer is the same as just enabling the
1336 .. tip:: Some tracers have their own options. They only appear in this
1337 file when the tracer is active. They always appear in the
1341 Here are the per tracer options:
1343 Options for function tracer:
1346 When set, a stack trace is recorded after every
1347 function that is recorded. NOTE! Limit the functions
1348 that are recorded before enabling this, with
1349 "set_ftrace_filter" otherwise the system performance
1350 will be critically degraded. Remember to disable
1351 this option before clearing the function filter.
1353 Options for function_graph tracer:
1355 Since the function_graph tracer has a slightly different output
1356 it has its own options to control what is displayed.
1359 When set, the "overrun" of the graph stack is
1360 displayed after each function traced. The
1361 overrun, is when the stack depth of the calls
1362 is greater than what is reserved for each task.
1363 Each task has a fixed array of functions to
1364 trace in the call graph. If the depth of the
1365 calls exceeds that, the function is not traced.
1366 The overrun is the number of functions missed
1367 due to exceeding this array.
1370 When set, the CPU number of the CPU where the trace
1371 occurred is displayed.
1374 When set, if the function takes longer than
1375 A certain amount, then a delay marker is
1376 displayed. See "delay" above, under the
1380 Unlike other tracers, the process' command line
1381 is not displayed by default, but instead only
1382 when a task is traced in and out during a context
1383 switch. Enabling this options has the command
1384 of each process displayed at every line.
1387 At the end of each function (the return)
1388 the duration of the amount of time in the
1389 function is displayed in microseconds.
1392 When set, the timestamp is displayed at each line.
1395 When disabled, functions that happen inside an
1396 interrupt will not be traced.
1399 When set, the return event will include the function
1400 that it represents. By default this is off, and
1401 only a closing curly bracket "}" is displayed for
1402 the return of a function.
1405 When set, the return value of each traced function
1406 will be printed after an equal sign "=". By default
1409 funcgraph-retval-hex
1410 When set, the return value will always be printed
1411 in hexadecimal format. If the option is not set and
1412 the return value is an error code, it will be printed
1413 in signed decimal format; otherwise it will also be
1414 printed in hexadecimal format. By default, this option
1418 When running function graph tracer, to include
1419 the time a task schedules out in its function.
1420 When enabled, it will account time the task has been
1421 scheduled out as part of the function call.
1424 When running function profiler with function graph tracer,
1425 to include the time to call nested functions. When this is
1426 not set, the time reported for the function will only
1427 include the time the function itself executed for, not the
1428 time for functions that it called.
1430 Options for blk tracer:
1433 Shows a more minimalistic output.
1439 When interrupts are disabled, the CPU can not react to any other
1440 external event (besides NMIs and SMIs). This prevents the timer
1441 interrupt from triggering or the mouse interrupt from letting
1442 the kernel know of a new mouse event. The result is a latency
1443 with the reaction time.
1445 The irqsoff tracer tracks the time for which interrupts are
1446 disabled. When a new maximum latency is hit, the tracer saves
1447 the trace leading up to that latency point so that every time a
1448 new maximum is reached, the old saved trace is discarded and the
1451 To reset the maximum, echo 0 into tracing_max_latency. Here is
1454 # echo 0 > options/function-trace
1455 # echo irqsoff > current_tracer
1456 # echo 1 > tracing_on
1457 # echo 0 > tracing_max_latency
1460 # echo 0 > tracing_on
1464 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1465 # --------------------------------------------------------------------
1466 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1468 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
1470 # => started at: run_timer_softirq
1471 # => ended at: run_timer_softirq
1475 # / _-----=> irqs-off
1476 # | / _----=> need-resched
1477 # || / _---=> hardirq/softirq
1478 # ||| / _--=> preempt-depth
1480 # cmd pid ||||| time | caller
1482 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
1483 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
1484 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
1485 <idle>-0 0dNs3 25us : <stack trace>
1486 => _raw_spin_unlock_irq
1487 => run_timer_softirq
1492 => smp_apic_timer_interrupt
1493 => apic_timer_interrupt
1498 => x86_64_start_reservations
1499 => x86_64_start_kernel
1501 Here we see that we had a latency of 16 microseconds (which is
1502 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
1503 interrupts. The difference between the 16 and the displayed
1504 timestamp 25us occurred because the clock was incremented
1505 between the time of recording the max latency and the time of
1506 recording the function that had that latency.
1508 Note the above example had function-trace not set. If we set
1509 function-trace, we get a much larger output::
1511 with echo 1 > options/function-trace
1515 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1516 # --------------------------------------------------------------------
1517 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1519 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1521 # => started at: ata_scsi_queuecmd
1522 # => ended at: ata_scsi_queuecmd
1526 # / _-----=> irqs-off
1527 # | / _----=> need-resched
1528 # || / _---=> hardirq/softirq
1529 # ||| / _--=> preempt-depth
1531 # cmd pid ||||| time | caller
1533 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1534 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1535 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1536 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1537 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1538 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1539 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1540 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1541 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1543 bash-2042 3d..1 67us : delay_tsc <-__delay
1544 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1545 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1546 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1547 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1548 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1549 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1550 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1551 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1552 bash-2042 3d..1 120us : <stack trace>
1553 => _raw_spin_unlock_irqrestore
1554 => ata_scsi_queuecmd
1555 => scsi_dispatch_cmd
1557 => __blk_run_queue_uncond
1560 => submit_bio_noacct
1563 => __ext3_get_inode_loc
1572 => user_path_at_empty
1577 => system_call_fastpath
1580 Here we traced a 71 microsecond latency. But we also see all the
1581 functions that were called during that time. Note that by
1582 enabling function tracing, we incur an added overhead. This
1583 overhead may extend the latency times. But nevertheless, this
1584 trace has provided some very helpful debugging information.
1586 If we prefer function graph output instead of function, we can set
1587 display-graph option::
1589 with echo 1 > options/display-graph
1593 # irqsoff latency trace v1.1.5 on 4.20.0-rc6+
1594 # --------------------------------------------------------------------
1595 # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4)
1597 # | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0)
1599 # => started at: free_debug_processing
1600 # => ended at: return_to_handler
1604 # / _----=> need-resched
1605 # | / _---=> hardirq/softirq
1606 # || / _--=> preempt-depth
1608 # REL TIME CPU TASK/PID |||| DURATION FUNCTION CALLS
1609 # | | | | |||| | | | | | |
1610 0 us | 0) bash-1507 | d... | 0.000 us | _raw_spin_lock_irqsave();
1611 0 us | 0) bash-1507 | d..1 | 0.378 us | do_raw_spin_trylock();
1612 1 us | 0) bash-1507 | d..2 | | set_track() {
1613 2 us | 0) bash-1507 | d..2 | | save_stack_trace() {
1614 2 us | 0) bash-1507 | d..2 | | __save_stack_trace() {
1615 3 us | 0) bash-1507 | d..2 | | __unwind_start() {
1616 3 us | 0) bash-1507 | d..2 | | get_stack_info() {
1617 3 us | 0) bash-1507 | d..2 | 0.351 us | in_task_stack();
1618 4 us | 0) bash-1507 | d..2 | 1.107 us | }
1620 3750 us | 0) bash-1507 | d..1 | 0.516 us | do_raw_spin_unlock();
1621 3750 us | 0) bash-1507 | d..1 | 0.000 us | _raw_spin_unlock_irqrestore();
1622 3764 us | 0) bash-1507 | d..1 | 0.000 us | tracer_hardirqs_on();
1623 bash-1507 0d..1 3792us : <stack trace>
1624 => free_debug_processing
1633 => search_binary_handler
1634 => __do_execve_file.isra.32
1637 => entry_SYSCALL_64_after_hwframe
1642 When preemption is disabled, we may be able to receive
1643 interrupts but the task cannot be preempted and a higher
1644 priority task must wait for preemption to be enabled again
1645 before it can preempt a lower priority task.
1647 The preemptoff tracer traces the places that disable preemption.
1648 Like the irqsoff tracer, it records the maximum latency for
1649 which preemption was disabled. The control of preemptoff tracer
1650 is much like the irqsoff tracer.
1653 # echo 0 > options/function-trace
1654 # echo preemptoff > current_tracer
1655 # echo 1 > tracing_on
1656 # echo 0 > tracing_max_latency
1659 # echo 0 > tracing_on
1661 # tracer: preemptoff
1663 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1664 # --------------------------------------------------------------------
1665 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1667 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1669 # => started at: do_IRQ
1670 # => ended at: do_IRQ
1674 # / _-----=> irqs-off
1675 # | / _----=> need-resched
1676 # || / _---=> hardirq/softirq
1677 # ||| / _--=> preempt-depth
1679 # cmd pid ||||| time | caller
1681 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1682 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1683 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1684 sshd-1991 1d..1 52us : <stack trace>
1685 => sub_preempt_count
1691 This has some more changes. Preemption was disabled when an
1692 interrupt came in (notice the 'h'), and was enabled on exit.
1693 But we also see that interrupts have been disabled when entering
1694 the preempt off section and leaving it (the 'd'). We do not know if
1695 interrupts were enabled in the mean time or shortly after this
1699 # tracer: preemptoff
1701 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1702 # --------------------------------------------------------------------
1703 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1705 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1707 # => started at: wake_up_new_task
1708 # => ended at: task_rq_unlock
1712 # / _-----=> irqs-off
1713 # | / _----=> need-resched
1714 # || / _---=> hardirq/softirq
1715 # ||| / _--=> preempt-depth
1717 # cmd pid ||||| time | caller
1719 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1720 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1721 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1722 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1723 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1725 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1726 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1727 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1728 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1729 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1730 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1731 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1732 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1734 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1735 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1736 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1737 bash-1994 1d..2 36us : do_softirq <-irq_exit
1738 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1739 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1740 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1741 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1742 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1743 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1745 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1746 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1747 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1748 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1749 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1750 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1751 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1752 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1753 bash-1994 1.N.1 104us : <stack trace>
1754 => sub_preempt_count
1755 => _raw_spin_unlock_irqrestore
1763 The above is an example of the preemptoff trace with
1764 function-trace set. Here we see that interrupts were not disabled
1765 the entire time. The irq_enter code lets us know that we entered
1766 an interrupt 'h'. Before that, the functions being traced still
1767 show that it is not in an interrupt, but we can see from the
1768 functions themselves that this is not the case.
1773 Knowing the locations that have interrupts disabled or
1774 preemption disabled for the longest times is helpful. But
1775 sometimes we would like to know when either preemption and/or
1776 interrupts are disabled.
1778 Consider the following code::
1780 local_irq_disable();
1781 call_function_with_irqs_off();
1783 call_function_with_irqs_and_preemption_off();
1785 call_function_with_preemption_off();
1788 The irqsoff tracer will record the total length of
1789 call_function_with_irqs_off() and
1790 call_function_with_irqs_and_preemption_off().
1792 The preemptoff tracer will record the total length of
1793 call_function_with_irqs_and_preemption_off() and
1794 call_function_with_preemption_off().
1796 But neither will trace the time that interrupts and/or
1797 preemption is disabled. This total time is the time that we can
1798 not schedule. To record this time, use the preemptirqsoff
1801 Again, using this trace is much like the irqsoff and preemptoff
1805 # echo 0 > options/function-trace
1806 # echo preemptirqsoff > current_tracer
1807 # echo 1 > tracing_on
1808 # echo 0 > tracing_max_latency
1811 # echo 0 > tracing_on
1813 # tracer: preemptirqsoff
1815 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1816 # --------------------------------------------------------------------
1817 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1819 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1821 # => started at: ata_scsi_queuecmd
1822 # => ended at: ata_scsi_queuecmd
1826 # / _-----=> irqs-off
1827 # | / _----=> need-resched
1828 # || / _---=> hardirq/softirq
1829 # ||| / _--=> preempt-depth
1831 # cmd pid ||||| time | caller
1833 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1834 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1835 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1836 ls-2230 3...1 111us : <stack trace>
1837 => sub_preempt_count
1838 => _raw_spin_unlock_irqrestore
1839 => ata_scsi_queuecmd
1840 => scsi_dispatch_cmd
1842 => __blk_run_queue_uncond
1845 => submit_bio_noacct
1850 => htree_dirblock_to_tree
1851 => ext3_htree_fill_tree
1855 => system_call_fastpath
1858 The trace_hardirqs_off_thunk is called from assembly on x86 when
1859 interrupts are disabled in the assembly code. Without the
1860 function tracing, we do not know if interrupts were enabled
1861 within the preemption points. We do see that it started with
1864 Here is a trace with function-trace set::
1866 # tracer: preemptirqsoff
1868 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1869 # --------------------------------------------------------------------
1870 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1872 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1874 # => started at: schedule
1875 # => ended at: mutex_unlock
1879 # / _-----=> irqs-off
1880 # | / _----=> need-resched
1881 # || / _---=> hardirq/softirq
1882 # ||| / _--=> preempt-depth
1884 # cmd pid ||||| time | caller
1886 kworker/-59 3...1 0us : __schedule <-schedule
1887 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1888 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1889 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1890 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1891 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1892 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1893 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1894 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1895 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1896 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1897 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1898 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1899 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1900 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1901 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1902 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1903 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1904 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1905 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1906 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1907 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1908 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1909 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1910 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1911 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1912 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1913 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1914 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1915 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1916 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1917 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1919 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1920 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1921 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1922 ls-2269 3d..3 21us : do_softirq <-irq_exit
1923 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1924 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1925 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1926 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1927 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1928 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1929 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1931 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1932 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1933 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1934 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1935 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1936 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1938 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1939 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1940 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1941 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1942 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1943 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1944 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1945 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1946 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1947 ls-2269 3d... 186us : <stack trace>
1948 => __mutex_unlock_slowpath
1955 => system_call_fastpath
1957 This is an interesting trace. It started with kworker running and
1958 scheduling out and ls taking over. But as soon as ls released the
1959 rq lock and enabled interrupts (but not preemption) an interrupt
1960 triggered. When the interrupt finished, it started running softirqs.
1961 But while the softirq was running, another interrupt triggered.
1962 When an interrupt is running inside a softirq, the annotation is 'H'.
1968 One common case that people are interested in tracing is the
1969 time it takes for a task that is woken to actually wake up.
1970 Now for non Real-Time tasks, this can be arbitrary. But tracing
1971 it none the less can be interesting.
1973 Without function tracing::
1975 # echo 0 > options/function-trace
1976 # echo wakeup > current_tracer
1977 # echo 1 > tracing_on
1978 # echo 0 > tracing_max_latency
1980 # echo 0 > tracing_on
1984 # wakeup latency trace v1.1.5 on 3.8.0-test+
1985 # --------------------------------------------------------------------
1986 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1988 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1992 # / _-----=> irqs-off
1993 # | / _----=> need-resched
1994 # || / _---=> hardirq/softirq
1995 # ||| / _--=> preempt-depth
1997 # cmd pid ||||| time | caller
1999 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
2000 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
2001 <idle>-0 3d..3 15us : __schedule <-schedule
2002 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
2004 The tracer only traces the highest priority task in the system
2005 to avoid tracing the normal circumstances. Here we see that
2006 the kworker with a nice priority of -20 (not very nice), took
2007 just 15 microseconds from the time it woke up, to the time it
2010 Non Real-Time tasks are not that interesting. A more interesting
2011 trace is to concentrate only on Real-Time tasks.
2016 In a Real-Time environment it is very important to know the
2017 wakeup time it takes for the highest priority task that is woken
2018 up to the time that it executes. This is also known as "schedule
2019 latency". I stress the point that this is about RT tasks. It is
2020 also important to know the scheduling latency of non-RT tasks,
2021 but the average schedule latency is better for non-RT tasks.
2022 Tools like LatencyTop are more appropriate for such
2025 Real-Time environments are interested in the worst case latency.
2026 That is the longest latency it takes for something to happen,
2027 and not the average. We can have a very fast scheduler that may
2028 only have a large latency once in a while, but that would not
2029 work well with Real-Time tasks. The wakeup_rt tracer was designed
2030 to record the worst case wakeups of RT tasks. Non-RT tasks are
2031 not recorded because the tracer only records one worst case and
2032 tracing non-RT tasks that are unpredictable will overwrite the
2033 worst case latency of RT tasks (just run the normal wakeup
2034 tracer for a while to see that effect).
2036 Since this tracer only deals with RT tasks, we will run this
2037 slightly differently than we did with the previous tracers.
2038 Instead of performing an 'ls', we will run 'sleep 1' under
2039 'chrt' which changes the priority of the task.
2042 # echo 0 > options/function-trace
2043 # echo wakeup_rt > current_tracer
2044 # echo 1 > tracing_on
2045 # echo 0 > tracing_max_latency
2047 # echo 0 > tracing_on
2053 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
2054 # --------------------------------------------------------------------
2055 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
2057 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
2061 # / _-----=> irqs-off
2062 # | / _----=> need-resched
2063 # || / _---=> hardirq/softirq
2064 # ||| / _--=> preempt-depth
2066 # cmd pid ||||| time | caller
2068 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
2069 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
2070 <idle>-0 3d..3 5us : __schedule <-schedule
2071 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
2074 Running this on an idle system, we see that it only took 5 microseconds
2075 to perform the task switch. Note, since the trace point in the schedule
2076 is before the actual "switch", we stop the tracing when the recorded task
2077 is about to schedule in. This may change if we add a new marker at the
2078 end of the scheduler.
2080 Notice that the recorded task is 'sleep' with the PID of 2389
2081 and it has an rt_prio of 5. This priority is user-space priority
2082 and not the internal kernel priority. The policy is 1 for
2083 SCHED_FIFO and 2 for SCHED_RR.
2085 Note, that the trace data shows the internal priority (99 - rtprio).
2088 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
2090 The 0:120:R means idle was running with a nice priority of 0 (120 - 120)
2091 and in the running state 'R'. The sleep task was scheduled in with
2092 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
2093 and it too is in the running state.
2095 Doing the same with chrt -r 5 and function-trace set.
2098 echo 1 > options/function-trace
2102 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
2103 # --------------------------------------------------------------------
2104 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
2106 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
2110 # / _-----=> irqs-off
2111 # | / _----=> need-resched
2112 # || / _---=> hardirq/softirq
2113 # ||| / _--=> preempt-depth
2115 # cmd pid ||||| time | caller
2117 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
2118 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
2119 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
2120 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
2121 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
2122 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
2123 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
2124 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
2125 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
2126 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2127 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
2128 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
2129 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
2130 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
2131 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
2132 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
2133 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
2134 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
2135 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
2136 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
2137 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
2138 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
2139 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
2140 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
2141 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
2142 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
2143 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
2144 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
2145 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
2146 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
2147 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
2148 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
2149 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
2150 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
2151 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
2152 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
2153 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
2154 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit
2155 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
2156 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
2157 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
2158 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
2159 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
2160 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
2161 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
2162 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
2163 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
2164 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
2165 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
2166 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
2167 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
2168 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2169 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
2170 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
2171 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
2172 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2173 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
2174 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
2175 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
2176 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
2177 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
2178 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
2179 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
2180 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
2181 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
2182 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
2183 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
2184 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
2185 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
2186 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
2187 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
2188 <idle>-0 3.N.. 25us : schedule <-cpu_idle
2189 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
2190 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
2191 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
2192 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
2193 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
2194 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
2195 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
2196 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
2197 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
2198 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
2199 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
2200 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
2201 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
2203 This isn't that big of a trace, even with function tracing enabled,
2204 so I included the entire trace.
2206 The interrupt went off while when the system was idle. Somewhere
2207 before task_woken_rt() was called, the NEED_RESCHED flag was set,
2208 this is indicated by the first occurrence of the 'N' flag.
2210 Latency tracing and events
2211 --------------------------
2212 As function tracing can induce a much larger latency, but without
2213 seeing what happens within the latency it is hard to know what
2214 caused it. There is a middle ground, and that is with enabling
2218 # echo 0 > options/function-trace
2219 # echo wakeup_rt > current_tracer
2220 # echo 1 > events/enable
2221 # echo 1 > tracing_on
2222 # echo 0 > tracing_max_latency
2224 # echo 0 > tracing_on
2228 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
2229 # --------------------------------------------------------------------
2230 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
2232 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
2236 # / _-----=> irqs-off
2237 # | / _----=> need-resched
2238 # || / _---=> hardirq/softirq
2239 # ||| / _--=> preempt-depth
2241 # cmd pid ||||| time | caller
2243 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
2244 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
2245 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
2246 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
2247 <idle>-0 2.N.2 2us : power_end: cpu_id=2
2248 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
2249 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
2250 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
2251 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
2252 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
2253 <idle>-0 2d..3 6us : __schedule <-schedule
2254 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
2257 Hardware Latency Detector
2258 -------------------------
2260 The hardware latency detector is executed by enabling the "hwlat" tracer.
2262 NOTE, this tracer will affect the performance of the system as it will
2263 periodically make a CPU constantly busy with interrupts disabled.
2266 # echo hwlat > current_tracer
2271 # entries-in-buffer/entries-written: 13/13 #P:8
2274 # / _----=> need-resched
2275 # | / _---=> hardirq/softirq
2276 # || / _--=> preempt-depth
2278 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2280 <...>-1729 [001] d... 678.473449: #1 inner/outer(us): 11/12 ts:1581527483.343962693 count:6
2281 <...>-1729 [004] d... 689.556542: #2 inner/outer(us): 16/9 ts:1581527494.889008092 count:1
2282 <...>-1729 [005] d... 714.756290: #3 inner/outer(us): 16/16 ts:1581527519.678961629 count:5
2283 <...>-1729 [001] d... 718.788247: #4 inner/outer(us): 9/17 ts:1581527523.889012713 count:1
2284 <...>-1729 [002] d... 719.796341: #5 inner/outer(us): 13/9 ts:1581527524.912872606 count:1
2285 <...>-1729 [006] d... 844.787091: #6 inner/outer(us): 9/12 ts:1581527649.889048502 count:2
2286 <...>-1729 [003] d... 849.827033: #7 inner/outer(us): 18/9 ts:1581527654.889013793 count:1
2287 <...>-1729 [007] d... 853.859002: #8 inner/outer(us): 9/12 ts:1581527658.889065736 count:1
2288 <...>-1729 [001] d... 855.874978: #9 inner/outer(us): 9/11 ts:1581527660.861991877 count:1
2289 <...>-1729 [001] d... 863.938932: #10 inner/outer(us): 9/11 ts:1581527668.970010500 count:1 nmi-total:7 nmi-count:1
2290 <...>-1729 [007] d... 878.050780: #11 inner/outer(us): 9/12 ts:1581527683.385002600 count:1 nmi-total:5 nmi-count:1
2291 <...>-1729 [007] d... 886.114702: #12 inner/outer(us): 9/12 ts:1581527691.385001600 count:1
2294 The above output is somewhat the same in the header. All events will have
2295 interrupts disabled 'd'. Under the FUNCTION title there is:
2298 This is the count of events recorded that were greater than the
2299 tracing_threshold (See below).
2301 inner/outer(us): 11/11
2303 This shows two numbers as "inner latency" and "outer latency". The test
2304 runs in a loop checking a timestamp twice. The latency detected within
2305 the two timestamps is the "inner latency" and the latency detected
2306 after the previous timestamp and the next timestamp in the loop is
2307 the "outer latency".
2309 ts:1581527483.343962693
2311 The absolute timestamp that the first latency was recorded in the window.
2315 The number of times a latency was detected during the window.
2317 nmi-total:7 nmi-count:1
2319 On architectures that support it, if an NMI comes in during the
2320 test, the time spent in NMI is reported in "nmi-total" (in
2323 All architectures that have NMIs will show the "nmi-count" if an
2324 NMI comes in during the test.
2329 This gets automatically set to "10" to represent 10
2330 microseconds. This is the threshold of latency that
2331 needs to be detected before the trace will be recorded.
2333 Note, when hwlat tracer is finished (another tracer is
2334 written into "current_tracer"), the original value for
2335 tracing_threshold is placed back into this file.
2337 hwlat_detector/width
2338 The length of time the test runs with interrupts disabled.
2340 hwlat_detector/window
2341 The length of time of the window which the test
2342 runs. That is, the test will run for "width"
2343 microseconds per "window" microseconds
2346 When the test is started. A kernel thread is created that
2347 runs the test. This thread will alternate between CPUs
2348 listed in the tracing_cpumask between each period
2349 (one "window"). To limit the test to specific CPUs
2350 set the mask in this file to only the CPUs that the test
2356 This tracer is the function tracer. Enabling the function tracer
2357 can be done from the debug file system. Make sure the
2358 ftrace_enabled is set; otherwise this tracer is a nop.
2359 See the "ftrace_enabled" section below.
2362 # sysctl kernel.ftrace_enabled=1
2363 # echo function > current_tracer
2364 # echo 1 > tracing_on
2366 # echo 0 > tracing_on
2370 # entries-in-buffer/entries-written: 24799/24799 #P:4
2373 # / _----=> need-resched
2374 # | / _---=> hardirq/softirq
2375 # || / _--=> preempt-depth
2377 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2379 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
2380 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
2381 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
2382 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
2383 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
2384 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
2385 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
2386 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
2390 Note: function tracer uses ring buffers to store the above
2391 entries. The newest data may overwrite the oldest data.
2392 Sometimes using echo to stop the trace is not sufficient because
2393 the tracing could have overwritten the data that you wanted to
2394 record. For this reason, it is sometimes better to disable
2395 tracing directly from a program. This allows you to stop the
2396 tracing at the point that you hit the part that you are
2397 interested in. To disable the tracing directly from a C program,
2398 something like following code snippet can be used::
2402 int main(int argc, char *argv[]) {
2404 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
2406 if (condition_hit()) {
2407 write(trace_fd, "0", 1);
2413 Single thread tracing
2414 ---------------------
2416 By writing into set_ftrace_pid you can trace a
2417 single thread. For example::
2419 # cat set_ftrace_pid
2421 # echo 3111 > set_ftrace_pid
2422 # cat set_ftrace_pid
2424 # echo function > current_tracer
2428 # TASK-PID CPU# TIMESTAMP FUNCTION
2430 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
2431 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
2432 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
2433 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
2434 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
2435 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
2436 # echo > set_ftrace_pid
2440 # TASK-PID CPU# TIMESTAMP FUNCTION
2442 ##### CPU 3 buffer started ####
2443 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
2444 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
2445 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
2446 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
2447 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
2449 If you want to trace a function when executing, you could use
2450 something like this simple program.
2455 #include <sys/types.h>
2456 #include <sys/stat.h>
2462 #define STR(x) _STR(x)
2463 #define MAX_PATH 256
2465 const char *find_tracefs(void)
2467 static char tracefs[MAX_PATH+1];
2468 static int tracefs_found;
2475 if ((fp = fopen("/proc/mounts","r")) == NULL) {
2476 perror("/proc/mounts");
2480 while (fscanf(fp, "%*s %"
2482 "s %99s %*s %*d %*d\n",
2483 tracefs, type) == 2) {
2484 if (strcmp(type, "tracefs") == 0)
2489 if (strcmp(type, "tracefs") != 0) {
2490 fprintf(stderr, "tracefs not mounted");
2494 strcat(tracefs, "/tracing/");
2500 const char *tracing_file(const char *file_name)
2502 static char trace_file[MAX_PATH+1];
2503 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name);
2507 int main (int argc, char **argv)
2517 ffd = open(tracing_file("current_tracer"), O_WRONLY);
2520 write(ffd, "nop", 3);
2522 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
2523 s = sprintf(line, "%d\n", getpid());
2526 write(ffd, "function", 8);
2531 execvp(argv[1], argv+1);
2537 Or this simple script!
2542 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts`
2543 echo 0 > $tracefs/tracing_on
2544 echo $$ > $tracefs/set_ftrace_pid
2545 echo function > $tracefs/current_tracer
2546 echo 1 > $tracefs/tracing_on
2550 function graph tracer
2551 ---------------------------
2553 This tracer is similar to the function tracer except that it
2554 probes a function on its entry and its exit. This is done by
2555 using a dynamically allocated stack of return addresses in each
2556 task_struct. On function entry the tracer overwrites the return
2557 address of each function traced to set a custom probe. Thus the
2558 original return address is stored on the stack of return address
2561 Probing on both ends of a function leads to special features
2564 - measure of a function's time execution
2565 - having a reliable call stack to draw function calls graph
2567 This tracer is useful in several situations:
2569 - you want to find the reason of a strange kernel behavior and
2570 need to see what happens in detail on any areas (or specific
2573 - you are experiencing weird latencies but it's difficult to
2576 - you want to find quickly which path is taken by a specific
2579 - you just want to peek inside a working kernel and want to see
2584 # tracer: function_graph
2586 # CPU DURATION FUNCTION CALLS
2590 0) | do_sys_open() {
2592 0) | kmem_cache_alloc() {
2593 0) 1.382 us | __might_sleep();
2595 0) | strncpy_from_user() {
2596 0) | might_fault() {
2597 0) 1.389 us | __might_sleep();
2602 0) 0.668 us | _spin_lock();
2603 0) 0.570 us | expand_files();
2604 0) 0.586 us | _spin_unlock();
2607 There are several columns that can be dynamically
2608 enabled/disabled. You can use every combination of options you
2609 want, depending on your needs.
2611 - The cpu number on which the function executed is default
2612 enabled. It is sometimes better to only trace one cpu (see
2613 tracing_cpumask file) or you might sometimes see unordered
2614 function calls while cpu tracing switch.
2616 - hide: echo nofuncgraph-cpu > trace_options
2617 - show: echo funcgraph-cpu > trace_options
2619 - The duration (function's time of execution) is displayed on
2620 the closing bracket line of a function or on the same line
2621 than the current function in case of a leaf one. It is default
2624 - hide: echo nofuncgraph-duration > trace_options
2625 - show: echo funcgraph-duration > trace_options
2627 - The overhead field precedes the duration field in case of
2628 reached duration thresholds.
2630 - hide: echo nofuncgraph-overhead > trace_options
2631 - show: echo funcgraph-overhead > trace_options
2632 - depends on: funcgraph-duration
2636 3) # 1837.709 us | } /* __switch_to */
2637 3) | finish_task_switch() {
2638 3) 0.313 us | _raw_spin_unlock_irq();
2640 3) # 1889.063 us | } /* __schedule */
2641 3) ! 140.417 us | } /* __schedule */
2642 3) # 2034.948 us | } /* schedule */
2643 3) * 33998.59 us | } /* schedule_preempt_disabled */
2647 1) 0.260 us | msecs_to_jiffies();
2648 1) 0.313 us | __rcu_read_unlock();
2651 1) 0.313 us | rcu_bh_qs();
2652 1) 0.313 us | __local_bh_enable();
2654 1) 0.365 us | idle_cpu();
2655 1) | rcu_irq_exit() {
2656 1) 0.417 us | rcu_eqs_enter_common.isra.47();
2660 1) @ 119760.2 us | }
2666 2) 0.417 us | scheduler_ipi();
2676 + means that the function exceeded 10 usecs.
2677 ! means that the function exceeded 100 usecs.
2678 # means that the function exceeded 1000 usecs.
2679 * means that the function exceeded 10 msecs.
2680 @ means that the function exceeded 100 msecs.
2681 $ means that the function exceeded 1 sec.
2684 - The task/pid field displays the thread cmdline and pid which
2685 executed the function. It is default disabled.
2687 - hide: echo nofuncgraph-proc > trace_options
2688 - show: echo funcgraph-proc > trace_options
2692 # tracer: function_graph
2694 # CPU TASK/PID DURATION FUNCTION CALLS
2696 0) sh-4802 | | d_free() {
2697 0) sh-4802 | | call_rcu() {
2698 0) sh-4802 | | __call_rcu() {
2699 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2700 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2701 0) sh-4802 | 2.899 us | }
2702 0) sh-4802 | 4.040 us | }
2703 0) sh-4802 | 5.151 us | }
2704 0) sh-4802 | + 49.370 us | }
2707 - The absolute time field is an absolute timestamp given by the
2708 system clock since it started. A snapshot of this time is
2709 given on each entry/exit of functions
2711 - hide: echo nofuncgraph-abstime > trace_options
2712 - show: echo funcgraph-abstime > trace_options
2717 # TIME CPU DURATION FUNCTION CALLS
2719 360.774522 | 1) 0.541 us | }
2720 360.774522 | 1) 4.663 us | }
2721 360.774523 | 1) 0.541 us | __wake_up_bit();
2722 360.774524 | 1) 6.796 us | }
2723 360.774524 | 1) 7.952 us | }
2724 360.774525 | 1) 9.063 us | }
2725 360.774525 | 1) 0.615 us | journal_mark_dirty();
2726 360.774527 | 1) 0.578 us | __brelse();
2727 360.774528 | 1) | reiserfs_prepare_for_journal() {
2728 360.774528 | 1) | unlock_buffer() {
2729 360.774529 | 1) | wake_up_bit() {
2730 360.774529 | 1) | bit_waitqueue() {
2731 360.774530 | 1) 0.594 us | __phys_addr();
2734 The function name is always displayed after the closing bracket
2735 for a function if the start of that function is not in the
2738 Display of the function name after the closing bracket may be
2739 enabled for functions whose start is in the trace buffer,
2740 allowing easier searching with grep for function durations.
2741 It is default disabled.
2743 - hide: echo nofuncgraph-tail > trace_options
2744 - show: echo funcgraph-tail > trace_options
2746 Example with nofuncgraph-tail (default)::
2749 0) | kmem_cache_free() {
2750 0) 0.518 us | __phys_addr();
2754 Example with funcgraph-tail::
2757 0) | kmem_cache_free() {
2758 0) 0.518 us | __phys_addr();
2759 0) 1.757 us | } /* kmem_cache_free() */
2760 0) 2.861 us | } /* putname() */
2762 The return value of each traced function can be displayed after
2763 an equal sign "=". When encountering system call failures, it
2764 can be very helpful to quickly locate the function that first
2765 returns an error code.
2767 - hide: echo nofuncgraph-retval > trace_options
2768 - show: echo funcgraph-retval > trace_options
2770 Example with funcgraph-retval::
2772 1) | cgroup_migrate() {
2773 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */
2774 1) | cgroup_migrate_execute() {
2775 1) | cpu_cgroup_can_attach() {
2776 1) | cgroup_taskset_first() {
2777 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */
2778 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */
2779 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */
2780 1) 2.335 us | } /* cpu_cgroup_can_attach = -22 */
2781 1) 4.369 us | } /* cgroup_migrate_execute = -22 */
2782 1) 7.143 us | } /* cgroup_migrate = -22 */
2784 The above example shows that the function cpu_cgroup_can_attach
2785 returned the error code -22 firstly, then we can read the code
2786 of this function to get the root cause.
2788 When the option funcgraph-retval-hex is not set, the return value can
2789 be displayed in a smart way. Specifically, if it is an error code,
2790 it will be printed in signed decimal format, otherwise it will
2791 printed in hexadecimal format.
2793 - smart: echo nofuncgraph-retval-hex > trace_options
2794 - hexadecimal: echo funcgraph-retval-hex > trace_options
2796 Example with funcgraph-retval-hex::
2798 1) | cgroup_migrate() {
2799 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */
2800 1) | cgroup_migrate_execute() {
2801 1) | cpu_cgroup_can_attach() {
2802 1) | cgroup_taskset_first() {
2803 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */
2804 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */
2805 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */
2806 1) 2.335 us | } /* cpu_cgroup_can_attach = 0xffffffea */
2807 1) 4.369 us | } /* cgroup_migrate_execute = 0xffffffea */
2808 1) 7.143 us | } /* cgroup_migrate = 0xffffffea */
2810 At present, there are some limitations when using the funcgraph-retval
2811 option, and these limitations will be eliminated in the future:
2813 - Even if the function return type is void, a return value will still
2814 be printed, and you can just ignore it.
2816 - Even if return values are stored in multiple registers, only the
2817 value contained in the first register will be recorded and printed.
2818 To illustrate, in the x86 architecture, eax and edx are used to store
2819 a 64-bit return value, with the lower 32 bits saved in eax and the
2820 upper 32 bits saved in edx. However, only the value stored in eax
2821 will be recorded and printed.
2823 - In certain procedure call standards, such as arm64's AAPCS64, when a
2824 type is smaller than a GPR, it is the responsibility of the consumer
2825 to perform the narrowing, and the upper bits may contain UNKNOWN values.
2826 Therefore, it is advisable to check the code for such cases. For instance,
2827 when using a u8 in a 64-bit GPR, bits [63:8] may contain arbitrary values,
2828 especially when larger types are truncated, whether explicitly or implicitly.
2829 Here are some specific cases to illustrate this point:
2833 The function narrow_to_u8 is defined as follows::
2835 u8 narrow_to_u8(u64 val)
2837 // implicitly truncated
2841 It may be compiled to::
2844 < ... ftrace instrumentation ... >
2847 If you pass 0x123456789abcdef to this function and want to narrow it,
2848 it may be recorded as 0x123456789abcdef instead of 0xef.
2852 The function error_if_not_4g_aligned is defined as follows::
2854 int error_if_not_4g_aligned(u64 val)
2856 if (val & GENMASK(31, 0))
2862 It could be compiled to::
2864 error_if_not_4g_aligned:
2865 CBNZ w0, .Lnot_aligned
2866 RET // bits [31:0] are zero, bits
2867 // [63:32] are UNKNOWN
2872 When passing 0x2_0000_0000 to it, the return value may be recorded as
2873 0x2_0000_0000 instead of 0.
2875 You can put some comments on specific functions by using
2876 trace_printk() For example, if you want to put a comment inside
2877 the __might_sleep() function, you just have to include
2878 <linux/ftrace.h> and call trace_printk() inside __might_sleep()::
2880 trace_printk("I'm a comment!\n")
2884 1) | __might_sleep() {
2885 1) | /* I'm a comment! */
2889 You might find other useful features for this tracer in the
2890 following "dynamic ftrace" section such as tracing only specific
2896 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2897 virtually no overhead when function tracing is disabled. The way
2898 this works is the mcount function call (placed at the start of
2899 every kernel function, produced by the -pg switch in gcc),
2900 starts of pointing to a simple return. (Enabling FTRACE will
2901 include the -pg switch in the compiling of the kernel.)
2903 At compile time every C file object is run through the
2904 recordmcount program (located in the scripts directory). This
2905 program will parse the ELF headers in the C object to find all
2906 the locations in the .text section that call mcount. Starting
2907 with gcc version 4.6, the -mfentry has been added for x86, which
2908 calls "__fentry__" instead of "mcount". Which is called before
2909 the creation of the stack frame.
2911 Note, not all sections are traced. They may be prevented by either
2912 a notrace, or blocked another way and all inline functions are not
2913 traced. Check the "available_filter_functions" file to see what functions
2916 A section called "__mcount_loc" is created that holds
2917 references to all the mcount/fentry call sites in the .text section.
2918 The recordmcount program re-links this section back into the
2919 original object. The final linking stage of the kernel will add all these
2920 references into a single table.
2922 On boot up, before SMP is initialized, the dynamic ftrace code
2923 scans this table and updates all the locations into nops. It
2924 also records the locations, which are added to the
2925 available_filter_functions list. Modules are processed as they
2926 are loaded and before they are executed. When a module is
2927 unloaded, it also removes its functions from the ftrace function
2928 list. This is automatic in the module unload code, and the
2929 module author does not need to worry about it.
2931 When tracing is enabled, the process of modifying the function
2932 tracepoints is dependent on architecture. The old method is to use
2933 kstop_machine to prevent races with the CPUs executing code being
2934 modified (which can cause the CPU to do undesirable things, especially
2935 if the modified code crosses cache (or page) boundaries), and the nops are
2936 patched back to calls. But this time, they do not call mcount
2937 (which is just a function stub). They now call into the ftrace
2940 The new method of modifying the function tracepoints is to place
2941 a breakpoint at the location to be modified, sync all CPUs, modify
2942 the rest of the instruction not covered by the breakpoint. Sync
2943 all CPUs again, and then remove the breakpoint with the finished
2944 version to the ftrace call site.
2946 Some archs do not even need to monkey around with the synchronization,
2947 and can just slap the new code on top of the old without any
2948 problems with other CPUs executing it at the same time.
2950 One special side-effect to the recording of the functions being
2951 traced is that we can now selectively choose which functions we
2952 wish to trace and which ones we want the mcount calls to remain
2955 Two files are used, one for enabling and one for disabling the
2956 tracing of specified functions. They are:
2964 A list of available functions that you can add to these files is
2967 available_filter_functions
2971 # cat available_filter_functions
2980 If I am only interested in sys_nanosleep and hrtimer_interrupt::
2982 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2983 # echo function > current_tracer
2984 # echo 1 > tracing_on
2986 # echo 0 > tracing_on
2990 # entries-in-buffer/entries-written: 5/5 #P:4
2993 # / _----=> need-resched
2994 # | / _---=> hardirq/softirq
2995 # || / _--=> preempt-depth
2997 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2999 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
3000 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
3001 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
3002 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
3003 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
3005 To see which functions are being traced, you can cat the file:
3008 # cat set_ftrace_filter
3013 Perhaps this is not enough. The filters also allow glob(7) matching.
3016 will match functions that begin with <match>
3018 will match functions that end with <match>
3020 will match functions that have <match> in it
3021 ``<match1>*<match2>``
3022 will match functions that begin with <match1> and end with <match2>
3025 It is better to use quotes to enclose the wild cards,
3026 otherwise the shell may expand the parameters into names
3027 of files in the local directory.
3031 # echo 'hrtimer_*' > set_ftrace_filter
3037 # entries-in-buffer/entries-written: 897/897 #P:4
3040 # / _----=> need-resched
3041 # | / _---=> hardirq/softirq
3042 # || / _--=> preempt-depth
3044 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3046 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
3047 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
3048 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
3049 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
3050 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
3051 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
3052 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
3053 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
3055 Notice that we lost the sys_nanosleep.
3058 # cat set_ftrace_filter
3063 hrtimer_try_to_cancel
3067 hrtimer_force_reprogram
3068 hrtimer_get_next_event
3072 hrtimer_get_remaining
3074 hrtimer_init_sleeper
3077 This is because the '>' and '>>' act just like they do in bash.
3078 To rewrite the filters, use '>'
3079 To append to the filters, use '>>'
3081 To clear out a filter so that all functions will be recorded
3084 # echo > set_ftrace_filter
3085 # cat set_ftrace_filter
3088 Again, now we want to append.
3092 # echo sys_nanosleep > set_ftrace_filter
3093 # cat set_ftrace_filter
3095 # echo 'hrtimer_*' >> set_ftrace_filter
3096 # cat set_ftrace_filter
3101 hrtimer_try_to_cancel
3105 hrtimer_force_reprogram
3106 hrtimer_get_next_event
3111 hrtimer_get_remaining
3113 hrtimer_init_sleeper
3116 The set_ftrace_notrace prevents those functions from being
3120 # echo '*preempt*' '*lock*' > set_ftrace_notrace
3126 # entries-in-buffer/entries-written: 39608/39608 #P:4
3129 # / _----=> need-resched
3130 # | / _---=> hardirq/softirq
3131 # || / _--=> preempt-depth
3133 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3135 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
3136 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
3137 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
3138 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
3139 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
3140 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
3141 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
3142 bash-1994 [000] .... 4342.324899: setattr_should_drop_suidgid <-do_truncate
3143 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
3144 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
3145 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
3146 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
3148 We can see that there's no more lock or preempt tracing.
3150 Selecting function filters via index
3151 ------------------------------------
3153 Because processing of strings is expensive (the address of the function
3154 needs to be looked up before comparing to the string being passed in),
3155 an index can be used as well to enable functions. This is useful in the
3156 case of setting thousands of specific functions at a time. By passing
3157 in a list of numbers, no string processing will occur. Instead, the function
3158 at the specific location in the internal array (which corresponds to the
3159 functions in the "available_filter_functions" file), is selected.
3163 # echo 1 > set_ftrace_filter
3165 Will select the first function listed in "available_filter_functions"
3169 # head -1 available_filter_functions
3170 trace_initcall_finish_cb
3172 # cat set_ftrace_filter
3173 trace_initcall_finish_cb
3175 # head -50 available_filter_functions | tail -1
3178 # echo 1 50 > set_ftrace_filter
3179 # cat set_ftrace_filter
3180 trace_initcall_finish_cb
3183 Dynamic ftrace with the function graph tracer
3184 ---------------------------------------------
3186 Although what has been explained above concerns both the
3187 function tracer and the function-graph-tracer, there are some
3188 special features only available in the function-graph tracer.
3190 If you want to trace only one function and all of its children,
3191 you just have to echo its name into set_graph_function::
3193 echo __do_fault > set_graph_function
3195 will produce the following "expanded" trace of the __do_fault()
3199 0) | filemap_fault() {
3200 0) | find_lock_page() {
3201 0) 0.804 us | find_get_page();
3202 0) | __might_sleep() {
3206 0) 0.653 us | _spin_lock();
3207 0) 0.578 us | page_add_file_rmap();
3208 0) 0.525 us | native_set_pte_at();
3209 0) 0.585 us | _spin_unlock();
3210 0) | unlock_page() {
3211 0) 0.541 us | page_waitqueue();
3212 0) 0.639 us | __wake_up_bit();
3216 0) | filemap_fault() {
3217 0) | find_lock_page() {
3218 0) 0.698 us | find_get_page();
3219 0) | __might_sleep() {
3223 0) 0.631 us | _spin_lock();
3224 0) 0.571 us | page_add_file_rmap();
3225 0) 0.526 us | native_set_pte_at();
3226 0) 0.586 us | _spin_unlock();
3227 0) | unlock_page() {
3228 0) 0.533 us | page_waitqueue();
3229 0) 0.638 us | __wake_up_bit();
3233 You can also expand several functions at once::
3235 echo sys_open > set_graph_function
3236 echo sys_close >> set_graph_function
3238 Now if you want to go back to trace all functions you can clear
3239 this special filter via::
3241 echo > set_graph_function
3247 Note, the proc sysctl ftrace_enable is a big on/off switch for the
3248 function tracer. By default it is enabled (when function tracing is
3249 enabled in the kernel). If it is disabled, all function tracing is
3250 disabled. This includes not only the function tracers for ftrace, but
3251 also for any other uses (perf, kprobes, stack tracing, profiling, etc). It
3252 cannot be disabled if there is a callback with FTRACE_OPS_FL_PERMANENT set
3255 Please disable this with care.
3257 This can be disable (and enabled) with::
3259 sysctl kernel.ftrace_enabled=0
3260 sysctl kernel.ftrace_enabled=1
3264 echo 0 > /proc/sys/kernel/ftrace_enabled
3265 echo 1 > /proc/sys/kernel/ftrace_enabled
3271 A few commands are supported by the set_ftrace_filter interface.
3272 Trace commands have the following format::
3274 <function>:<command>:<parameter>
3276 The following commands are supported:
3279 This command enables function filtering per module. The
3280 parameter defines the module. For example, if only the write*
3281 functions in the ext3 module are desired, run:
3283 echo 'write*:mod:ext3' > set_ftrace_filter
3285 This command interacts with the filter in the same way as
3286 filtering based on function names. Thus, adding more functions
3287 in a different module is accomplished by appending (>>) to the
3288 filter file. Remove specific module functions by prepending
3291 echo '!writeback*:mod:ext3' >> set_ftrace_filter
3293 Mod command supports module globbing. Disable tracing for all
3294 functions except a specific module::
3296 echo '!*:mod:!ext3' >> set_ftrace_filter
3298 Disable tracing for all modules, but still trace kernel::
3300 echo '!*:mod:*' >> set_ftrace_filter
3302 Enable filter only for kernel::
3304 echo '*write*:mod:!*' >> set_ftrace_filter
3306 Enable filter for module globbing::
3308 echo '*write*:mod:*snd*' >> set_ftrace_filter
3311 These commands turn tracing on and off when the specified
3312 functions are hit. The parameter determines how many times the
3313 tracing system is turned on and off. If unspecified, there is
3314 no limit. For example, to disable tracing when a schedule bug
3315 is hit the first 5 times, run::
3317 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
3319 To always disable tracing when __schedule_bug is hit::
3321 echo '__schedule_bug:traceoff' > set_ftrace_filter
3323 These commands are cumulative whether or not they are appended
3324 to set_ftrace_filter. To remove a command, prepend it by '!'
3325 and drop the parameter::
3327 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
3329 The above removes the traceoff command for __schedule_bug
3330 that have a counter. To remove commands without counters::
3332 echo '!__schedule_bug:traceoff' > set_ftrace_filter
3335 Will cause a snapshot to be triggered when the function is hit.
3338 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
3340 To only snapshot once:
3343 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
3345 To remove the above commands::
3347 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
3348 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
3350 - enable_event/disable_event:
3351 These commands can enable or disable a trace event. Note, because
3352 function tracing callbacks are very sensitive, when these commands
3353 are registered, the trace point is activated, but disabled in
3354 a "soft" mode. That is, the tracepoint will be called, but
3355 just will not be traced. The event tracepoint stays in this mode
3356 as long as there's a command that triggers it.
3359 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
3364 <function>:enable_event:<system>:<event>[:count]
3365 <function>:disable_event:<system>:<event>[:count]
3367 To remove the events commands::
3369 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
3371 echo '!schedule:disable_event:sched:sched_switch' > \
3375 When the function is hit, it will dump the contents of the ftrace
3376 ring buffer to the console. This is useful if you need to debug
3377 something, and want to dump the trace when a certain function
3378 is hit. Perhaps it's a function that is called before a triple
3379 fault happens and does not allow you to get a regular dump.
3382 When the function is hit, it will dump the contents of the ftrace
3383 ring buffer for the current CPU to the console. Unlike the "dump"
3384 command, it only prints out the contents of the ring buffer for the
3385 CPU that executed the function that triggered the dump.
3388 When the function is hit, a stack trace is recorded.
3393 The trace_pipe outputs the same content as the trace file, but
3394 the effect on the tracing is different. Every read from
3395 trace_pipe is consumed. This means that subsequent reads will be
3396 different. The trace is live.
3399 # echo function > current_tracer
3400 # cat trace_pipe > /tmp/trace.out &
3402 # echo 1 > tracing_on
3404 # echo 0 > tracing_on
3408 # entries-in-buffer/entries-written: 0/0 #P:4
3411 # / _----=> need-resched
3412 # | / _---=> hardirq/softirq
3413 # || / _--=> preempt-depth
3415 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3419 # cat /tmp/trace.out
3420 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
3421 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
3422 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
3423 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
3424 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
3425 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
3426 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
3427 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
3428 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
3431 Note, reading the trace_pipe file will block until more input is
3432 added. This is contrary to the trace file. If any process opened
3433 the trace file for reading, it will actually disable tracing and
3434 prevent new entries from being added. The trace_pipe file does
3435 not have this limitation.
3440 Having too much or not enough data can be troublesome in
3441 diagnosing an issue in the kernel. The file buffer_size_kb is
3442 used to modify the size of the internal trace buffers. The
3443 number listed is the number of entries that can be recorded per
3444 CPU. To know the full size, multiply the number of possible CPUs
3445 with the number of entries.
3448 # cat buffer_size_kb
3449 1408 (units kilobytes)
3451 Or simply read buffer_total_size_kb
3454 # cat buffer_total_size_kb
3457 To modify the buffer, simple echo in a number (in 1024 byte segments).
3460 # echo 10000 > buffer_size_kb
3461 # cat buffer_size_kb
3462 10000 (units kilobytes)
3464 It will try to allocate as much as possible. If you allocate too
3465 much, it can cause Out-Of-Memory to trigger.
3468 # echo 1000000000000 > buffer_size_kb
3469 -bash: echo: write error: Cannot allocate memory
3470 # cat buffer_size_kb
3473 The per_cpu buffers can be changed individually as well:
3476 # echo 10000 > per_cpu/cpu0/buffer_size_kb
3477 # echo 100 > per_cpu/cpu1/buffer_size_kb
3479 When the per_cpu buffers are not the same, the buffer_size_kb
3480 at the top level will just show an X
3483 # cat buffer_size_kb
3486 This is where the buffer_total_size_kb is useful:
3489 # cat buffer_total_size_kb
3492 Writing to the top level buffer_size_kb will reset all the buffers
3493 to be the same again.
3497 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
3498 available to all non latency tracers. (Latency tracers which
3499 record max latency, such as "irqsoff" or "wakeup", can't use
3500 this feature, since those are already using the snapshot
3501 mechanism internally.)
3503 Snapshot preserves a current trace buffer at a particular point
3504 in time without stopping tracing. Ftrace swaps the current
3505 buffer with a spare buffer, and tracing continues in the new
3506 current (=previous spare) buffer.
3508 The following tracefs files in "tracing" are related to this
3513 This is used to take a snapshot and to read the output
3514 of the snapshot. Echo 1 into this file to allocate a
3515 spare buffer and to take a snapshot (swap), then read
3516 the snapshot from this file in the same format as
3517 "trace" (described above in the section "The File
3518 System"). Both reads snapshot and tracing are executable
3519 in parallel. When the spare buffer is allocated, echoing
3520 0 frees it, and echoing else (positive) values clear the
3522 More details are shown in the table below.
3524 +--------------+------------+------------+------------+
3525 |status\\input | 0 | 1 | else |
3526 +==============+============+============+============+
3527 |not allocated |(do nothing)| alloc+swap |(do nothing)|
3528 +--------------+------------+------------+------------+
3529 |allocated | free | swap | clear |
3530 +--------------+------------+------------+------------+
3532 Here is an example of using the snapshot feature.
3535 # echo 1 > events/sched/enable
3540 # entries-in-buffer/entries-written: 71/71 #P:8
3543 # / _----=> need-resched
3544 # | / _---=> hardirq/softirq
3545 # || / _--=> preempt-depth
3547 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3549 <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
3550 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
3552 <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
3557 # entries-in-buffer/entries-written: 77/77 #P:8
3560 # / _----=> need-resched
3561 # | / _---=> hardirq/softirq
3562 # || / _--=> preempt-depth
3564 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3566 <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
3567 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
3571 If you try to use this snapshot feature when current tracer is
3572 one of the latency tracers, you will get the following results.
3575 # echo wakeup > current_tracer
3577 bash: echo: write error: Device or resource busy
3579 cat: snapshot: Device or resource busy
3584 In the tracefs tracing directory, there is a directory called "instances".
3585 This directory can have new directories created inside of it using
3586 mkdir, and removing directories with rmdir. The directory created
3587 with mkdir in this directory will already contain files and other
3588 directories after it is created.
3591 # mkdir instances/foo
3593 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
3594 set_event snapshot trace trace_clock trace_marker trace_options
3595 trace_pipe tracing_on
3597 As you can see, the new directory looks similar to the tracing directory
3598 itself. In fact, it is very similar, except that the buffer and
3599 events are agnostic from the main directory, or from any other
3600 instances that are created.
3602 The files in the new directory work just like the files with the
3603 same name in the tracing directory except the buffer that is used
3604 is a separate and new buffer. The files affect that buffer but do not
3605 affect the main buffer with the exception of trace_options. Currently,
3606 the trace_options affect all instances and the top level buffer
3607 the same, but this may change in future releases. That is, options
3608 may become specific to the instance they reside in.
3610 Notice that none of the function tracer files are there, nor is
3611 current_tracer and available_tracers. This is because the buffers
3612 can currently only have events enabled for them.
3615 # mkdir instances/foo
3616 # mkdir instances/bar
3617 # mkdir instances/zoot
3618 # echo 100000 > buffer_size_kb
3619 # echo 1000 > instances/foo/buffer_size_kb
3620 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
3621 # echo function > current_trace
3622 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
3623 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
3624 # echo 1 > instances/foo/events/sched/sched_switch/enable
3625 # echo 1 > instances/bar/events/irq/enable
3626 # echo 1 > instances/zoot/events/syscalls/enable
3628 CPU:2 [LOST 11745 EVENTS]
3629 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
3630 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
3631 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
3632 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
3633 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
3634 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
3635 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
3636 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
3637 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3638 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
3639 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
3642 # cat instances/foo/trace_pipe
3643 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3644 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3645 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
3646 <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
3647 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
3648 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
3649 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
3650 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
3651 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
3652 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
3655 # cat instances/bar/trace_pipe
3656 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
3657 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
3658 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
3659 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
3660 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
3661 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
3662 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
3663 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
3664 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
3665 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
3666 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
3667 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
3670 # cat instances/zoot/trace
3673 # entries-in-buffer/entries-written: 18996/18996 #P:4
3676 # / _----=> need-resched
3677 # | / _---=> hardirq/softirq
3678 # || / _--=> preempt-depth
3680 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
3682 bash-1998 [000] d... 140.733501: sys_write -> 0x2
3683 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
3684 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
3685 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
3686 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
3687 bash-1998 [000] d... 140.733510: sys_close(fd: a)
3688 bash-1998 [000] d... 140.733510: sys_close -> 0x0
3689 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
3690 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
3691 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
3692 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
3694 You can see that the trace of the top most trace buffer shows only
3695 the function tracing. The foo instance displays wakeups and task
3698 To remove the instances, simply delete their directories:
3701 # rmdir instances/foo
3702 # rmdir instances/bar
3703 # rmdir instances/zoot
3705 Note, if a process has a trace file open in one of the instance
3706 directories, the rmdir will fail with EBUSY.
3711 Since the kernel has a fixed sized stack, it is important not to
3712 waste it in functions. A kernel developer must be conscious of
3713 what they allocate on the stack. If they add too much, the system
3714 can be in danger of a stack overflow, and corruption will occur,
3715 usually leading to a system panic.
3717 There are some tools that check this, usually with interrupts
3718 periodically checking usage. But if you can perform a check
3719 at every function call that will become very useful. As ftrace provides
3720 a function tracer, it makes it convenient to check the stack size
3721 at every function call. This is enabled via the stack tracer.
3723 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
3724 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
3727 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
3729 You can also enable it from the kernel command line to trace
3730 the stack size of the kernel during boot up, by adding "stacktrace"
3731 to the kernel command line parameter.
3733 After running it for a few minutes, the output looks like:
3736 # cat stack_max_size
3740 Depth Size Location (18 entries)
3742 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
3743 1) 2704 160 find_busiest_group+0x31/0x1f1
3744 2) 2544 256 load_balance+0xd9/0x662
3745 3) 2288 80 idle_balance+0xbb/0x130
3746 4) 2208 128 __schedule+0x26e/0x5b9
3747 5) 2080 16 schedule+0x64/0x66
3748 6) 2064 128 schedule_timeout+0x34/0xe0
3749 7) 1936 112 wait_for_common+0x97/0xf1
3750 8) 1824 16 wait_for_completion+0x1d/0x1f
3751 9) 1808 128 flush_work+0xfe/0x119
3752 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
3753 11) 1664 48 input_available_p+0x1d/0x5c
3754 12) 1616 48 n_tty_poll+0x6d/0x134
3755 13) 1568 64 tty_poll+0x64/0x7f
3756 14) 1504 880 do_select+0x31e/0x511
3757 15) 624 400 core_sys_select+0x177/0x216
3758 16) 224 96 sys_select+0x91/0xb9
3759 17) 128 128 system_call_fastpath+0x16/0x1b
3761 Note, if -mfentry is being used by gcc, functions get traced before
3762 they set up the stack frame. This means that leaf level functions
3763 are not tested by the stack tracer when -mfentry is used.
3765 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
3769 More details can be found in the source code, in the `kernel/trace/*.c` files.