1 ftrace - Function Tracer
2 ========================
4 Copyright 2008 Red Hat Inc.
5 Author: Steven Rostedt <srostedt@redhat.com>
6 License: The GNU Free Documentation License, Version 1.2
7 (dual licensed under the GPL v2)
8 Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton,
9 John Kacur, and David Teigland.
10 Written for: 2.6.28-rc2
16 Ftrace is an internal tracer designed to help out developers and
17 designers of systems to find what is going on inside the kernel.
18 It can be used for debugging or analyzing latencies and
19 performance issues that take place outside of user-space.
21 Although ftrace is typically considered the function tracer, it
22 is really a frame work of several assorted tracing utilities.
23 There's latency tracing to examine what occurs between interrupts
24 disabled and enabled, as well as for preemption and from a time
25 a task is woken to the task is actually scheduled in.
27 One of the most common uses of ftrace is the event tracing.
28 Through out the kernel is hundreds of static event points that
29 can be enabled via the debugfs file system to see what is
30 going on in certain parts of the kernel.
33 Implementation Details
34 ----------------------
36 See ftrace-design.txt for details for arch porters and such.
42 Ftrace uses the debugfs file system to hold the control files as
43 well as the files to display output.
45 When debugfs is configured into the kernel (which selecting any ftrace
46 option will do) the directory /sys/kernel/debug will be created. To mount
47 this directory, you can add to your /etc/fstab file:
49 debugfs /sys/kernel/debug debugfs defaults 0 0
51 Or you can mount it at run time with:
53 mount -t debugfs nodev /sys/kernel/debug
55 For quicker access to that directory you may want to make a soft link to
58 ln -s /sys/kernel/debug /debug
60 Any selected ftrace option will also create a directory called tracing
61 within the debugfs. The rest of the document will assume that you are in
62 the ftrace directory (cd /sys/kernel/debug/tracing) and will only concentrate
63 on the files within that directory and not distract from the content with
64 the extended "/sys/kernel/debug/tracing" path name.
66 That's it! (assuming that you have ftrace configured into your kernel)
68 After mounting debugfs, you can see a directory called
69 "tracing". This directory contains the control and output files
70 of ftrace. Here is a list of some of the key files:
73 Note: all time values are in microseconds.
77 This is used to set or display the current tracer
82 This holds the different types of tracers that
83 have been compiled into the kernel. The
84 tracers listed here can be configured by
85 echoing their name into current_tracer.
89 This sets or displays whether writing to the trace
90 ring buffer is enabled. Echo 0 into this file to disable
91 the tracer or 1 to enable it. Note, this only disables
92 writing to the ring buffer, the tracing overhead may
97 This file holds the output of the trace in a human
98 readable format (described below).
102 The output is the same as the "trace" file but this
103 file is meant to be streamed with live tracing.
104 Reads from this file will block until new data is
105 retrieved. Unlike the "trace" file, this file is a
106 consumer. This means reading from this file causes
107 sequential reads to display more current data. Once
108 data is read from this file, it is consumed, and
109 will not be read again with a sequential read. The
110 "trace" file is static, and if the tracer is not
111 adding more data, it will display the same
112 information every time it is read.
116 This file lets the user control the amount of data
117 that is displayed in one of the above output
118 files. Options also exist to modify how a tracer
119 or events work (stack traces, timestamps, etc).
123 This is a directory that has a file for every available
124 trace option (also in trace_options). Options may also be set
125 or cleared by writing a "1" or "0" respectively into the
126 corresponding file with the option name.
130 Some of the tracers record the max latency.
131 For example, the time interrupts are disabled.
132 This time is saved in this file. The max trace
133 will also be stored, and displayed by "trace".
134 A new max trace will only be recorded if the
135 latency is greater than the value in this
136 file. (in microseconds)
140 Some latency tracers will record a trace whenever the
141 latency is greater than the number in this file.
142 Only active when the file contains a number greater than 0.
147 This sets or displays the number of kilobytes each CPU
148 buffer holds. By default, the trace buffers are the same size
149 for each CPU. The displayed number is the size of the
150 CPU buffer and not total size of all buffers. The
151 trace buffers are allocated in pages (blocks of memory
152 that the kernel uses for allocation, usually 4 KB in size).
153 If the last page allocated has room for more bytes
154 than requested, the rest of the page will be used,
155 making the actual allocation bigger than requested.
156 ( Note, the size may not be a multiple of the page size
157 due to buffer management meta-data. )
159 buffer_total_size_kb:
161 This displays the total combined size of all the trace buffers.
165 If a process is performing the tracing, and the ring buffer
166 should be shrunk "freed" when the process is finished, even
167 if it were to be killed by a signal, this file can be used
168 for that purpose. On close of this file, the ring buffer will
169 be resized to its minimum size. Having a process that is tracing
170 also open this file, when the process exits its file descriptor
171 for this file will be closed, and in doing so, the ring buffer
174 It may also stop tracing if disable_on_free option is set.
178 This is a mask that lets the user only trace
179 on specified CPUs. The format is a hex string
180 representing the CPUs.
184 When dynamic ftrace is configured in (see the
185 section below "dynamic ftrace"), the code is dynamically
186 modified (code text rewrite) to disable calling of the
187 function profiler (mcount). This lets tracing be configured
188 in with practically no overhead in performance. This also
189 has a side effect of enabling or disabling specific functions
190 to be traced. Echoing names of functions into this file
191 will limit the trace to only those functions.
193 This interface also allows for commands to be used. See the
194 "Filter commands" section for more details.
198 This has an effect opposite to that of
199 set_ftrace_filter. Any function that is added here will not
200 be traced. If a function exists in both set_ftrace_filter
201 and set_ftrace_notrace, the function will _not_ be traced.
205 Have the function tracer only trace a single thread.
209 Have the events only trace a task with a PID listed in this file.
210 Note, sched_switch and sched_wake_up will also trace events
213 To have the PIDs of children of tasks with their PID in this file
214 added on fork, enable the "event-fork" option. That option will also
215 cause the PIDs of tasks to be removed from this file when the task
220 Set a "trigger" function where tracing should start
221 with the function graph tracer (See the section
222 "dynamic ftrace" for more details).
224 available_filter_functions:
226 This lists the functions that ftrace
227 has processed and can trace. These are the function
228 names that you can pass to "set_ftrace_filter" or
229 "set_ftrace_notrace". (See the section "dynamic ftrace"
230 below for more details.)
234 This file is more for debugging ftrace, but can also be useful
235 in seeing if any function has a callback attached to it.
236 Not only does the trace infrastructure use ftrace function
237 trace utility, but other subsystems might too. This file
238 displays all functions that have a callback attached to them
239 as well as the number of callbacks that have been attached.
240 Note, a callback may also call multiple functions which will
241 not be listed in this count.
243 If the callback registered to be traced by a function with
244 the "save regs" attribute (thus even more overhead), a 'R'
245 will be displayed on the same line as the function that
246 is returning registers.
248 If the callback registered to be traced by a function with
249 the "ip modify" attribute (thus the regs->ip can be changed),
250 an 'I' will be displayed on the same line as the function that
253 function_profile_enabled:
255 When set it will enable all functions with either the function
256 tracer, or if enabled, the function graph tracer. It will
257 keep a histogram of the number of functions that were called
258 and if run with the function graph tracer, it will also keep
259 track of the time spent in those functions. The histogram
260 content can be displayed in the files:
262 trace_stats/function<cpu> ( function0, function1, etc).
266 A directory that holds different tracing stats.
270 Enable dynamic trace points. See kprobetrace.txt.
274 Dynamic trace points stats. See kprobetrace.txt.
278 Used with the function graph tracer. This is the max depth
279 it will trace into a function. Setting this to a value of
280 one will show only the first kernel function that is called
285 This is for tools that read the raw format files. If an event in
286 the ring buffer references a string (currently only trace_printk()
287 does this), only a pointer to the string is recorded into the buffer
288 and not the string itself. This prevents tools from knowing what
289 that string was. This file displays the string and address for
290 the string allowing tools to map the pointers to what the
295 Only the pid of the task is recorded in a trace event unless
296 the event specifically saves the task comm as well. Ftrace
297 makes a cache of pid mappings to comms to try to display
298 comms for events. If a pid for a comm is not listed, then
299 "<...>" is displayed in the output.
303 This displays the "snapshot" buffer and also lets the user
304 take a snapshot of the current running trace.
305 See the "Snapshot" section below for more details.
309 When the stack tracer is activated, this will display the
310 maximum stack size it has encountered.
311 See the "Stack Trace" section below.
315 This displays the stack back trace of the largest stack
316 that was encountered when the stack tracer is activated.
317 See the "Stack Trace" section below.
321 This is similar to "set_ftrace_filter" but it limits what
322 functions the stack tracer will check.
326 Whenever an event is recorded into the ring buffer, a
327 "timestamp" is added. This stamp comes from a specified
328 clock. By default, ftrace uses the "local" clock. This
329 clock is very fast and strictly per cpu, but on some
330 systems it may not be monotonic with respect to other
331 CPUs. In other words, the local clocks may not be in sync
332 with local clocks on other CPUs.
334 Usual clocks for tracing:
337 [local] global counter x86-tsc
339 local: Default clock, but may not be in sync across CPUs
341 global: This clock is in sync with all CPUs but may
342 be a bit slower than the local clock.
344 counter: This is not a clock at all, but literally an atomic
345 counter. It counts up one by one, but is in sync
346 with all CPUs. This is useful when you need to
347 know exactly the order events occurred with respect to
348 each other on different CPUs.
350 uptime: This uses the jiffies counter and the time stamp
351 is relative to the time since boot up.
353 perf: This makes ftrace use the same clock that perf uses.
354 Eventually perf will be able to read ftrace buffers
355 and this will help out in interleaving the data.
357 x86-tsc: Architectures may define their own clocks. For
358 example, x86 uses its own TSC cycle clock here.
360 ppc-tb: This uses the powerpc timebase register value.
361 This is in sync across CPUs and can also be used
362 to correlate events across hypervisor/guest if
365 To set a clock, simply echo the clock name into this file.
367 echo global > trace_clock
371 This is a very useful file for synchronizing user space
372 with events happening in the kernel. Writing strings into
373 this file will be written into the ftrace buffer.
375 It is useful in applications to open this file at the start
376 of the application and just reference the file descriptor
379 void trace_write(const char *fmt, ...)
389 n = vsnprintf(buf, 256, fmt, ap);
392 write(trace_fd, buf, n);
397 trace_fd = open("trace_marker", WR_ONLY);
401 Add dynamic tracepoints in programs.
406 Uprobe statistics. See uprobetrace.txt
410 This is a way to make multiple trace buffers where different
411 events can be recorded in different buffers.
412 See "Instances" section below.
416 This is the trace event directory. It holds event tracepoints
417 (also known as static tracepoints) that have been compiled
418 into the kernel. It shows what event tracepoints exist
419 and how they are grouped by system. There are "enable"
420 files at various levels that can enable the tracepoints
421 when a "1" is written to them.
423 See events.txt for more information.
427 This is a directory that contains the trace per_cpu information.
429 per_cpu/cpu0/buffer_size_kb:
431 The ftrace buffer is defined per_cpu. That is, there's a separate
432 buffer for each CPU to allow writes to be done atomically,
433 and free from cache bouncing. These buffers may have different
434 size buffers. This file is similar to the buffer_size_kb
435 file, but it only displays or sets the buffer size for the
436 specific CPU. (here cpu0).
440 This is similar to the "trace" file, but it will only display
441 the data specific for the CPU. If written to, it only clears
442 the specific CPU buffer.
444 per_cpu/cpu0/trace_pipe
446 This is similar to the "trace_pipe" file, and is a consuming
447 read, but it will only display (and consume) the data specific
450 per_cpu/cpu0/trace_pipe_raw
452 For tools that can parse the ftrace ring buffer binary format,
453 the trace_pipe_raw file can be used to extract the data
454 from the ring buffer directly. With the use of the splice()
455 system call, the buffer data can be quickly transferred to
456 a file or to the network where a server is collecting the
459 Like trace_pipe, this is a consuming reader, where multiple
460 reads will always produce different data.
462 per_cpu/cpu0/snapshot:
464 This is similar to the main "snapshot" file, but will only
465 snapshot the current CPU (if supported). It only displays
466 the content of the snapshot for a given CPU, and if
467 written to, only clears this CPU buffer.
469 per_cpu/cpu0/snapshot_raw:
471 Similar to the trace_pipe_raw, but will read the binary format
472 from the snapshot buffer for the given CPU.
476 This displays certain stats about the ring buffer:
478 entries: The number of events that are still in the buffer.
480 overrun: The number of lost events due to overwriting when
483 commit overrun: Should always be zero.
484 This gets set if so many events happened within a nested
485 event (ring buffer is re-entrant), that it fills the
486 buffer and starts dropping events.
488 bytes: Bytes actually read (not overwritten).
490 oldest event ts: The oldest timestamp in the buffer
492 now ts: The current timestamp
494 dropped events: Events lost due to overwrite option being off.
496 read events: The number of events read.
501 Here is the list of current tracers that may be configured.
505 Function call tracer to trace all kernel functions.
509 Similar to the function tracer except that the
510 function tracer probes the functions on their entry
511 whereas the function graph tracer traces on both entry
512 and exit of the functions. It then provides the ability
513 to draw a graph of function calls similar to C code
518 Traces the areas that disable interrupts and saves
519 the trace with the longest max latency.
520 See tracing_max_latency. When a new max is recorded,
521 it replaces the old trace. It is best to view this
522 trace with the latency-format option enabled.
526 Similar to irqsoff but traces and records the amount of
527 time for which preemption is disabled.
531 Similar to irqsoff and preemptoff, but traces and
532 records the largest time for which irqs and/or preemption
537 Traces and records the max latency that it takes for
538 the highest priority task to get scheduled after
539 it has been woken up.
540 Traces all tasks as an average developer would expect.
544 Traces and records the max latency that it takes for just
545 RT tasks (as the current "wakeup" does). This is useful
546 for those interested in wake up timings of RT tasks.
550 This is the "trace nothing" tracer. To remove all
551 tracers from tracing simply echo "nop" into
555 Examples of using the tracer
556 ----------------------------
558 Here are typical examples of using the tracers when controlling
559 them only with the debugfs interface (without using any
560 user-land utilities).
565 Here is an example of the output format of the file "trace"
570 # entries-in-buffer/entries-written: 140080/250280 #P:4
573 # / _----=> need-resched
574 # | / _---=> hardirq/softirq
575 # || / _--=> preempt-depth
577 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
579 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
580 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
581 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
582 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
583 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
584 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
585 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
586 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
587 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
588 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
591 A header is printed with the tracer name that is represented by
592 the trace. In this case the tracer is "function". Then it shows the
593 number of events in the buffer as well as the total number of entries
594 that were written. The difference is the number of entries that were
595 lost due to the buffer filling up (250280 - 140080 = 110200 events
598 The header explains the content of the events. Task name "bash", the task
599 PID "1977", the CPU that it was running on "000", the latency format
600 (explained below), the timestamp in <secs>.<usecs> format, the
601 function name that was traced "sys_close" and the parent function that
602 called this function "system_call_fastpath". The timestamp is the time
603 at which the function was entered.
608 When the latency-format option is enabled or when one of the latency
609 tracers is set, the trace file gives somewhat more information to see
610 why a latency happened. Here is a typical trace.
614 # irqsoff latency trace v1.1.5 on 3.8.0-test+
615 # --------------------------------------------------------------------
616 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
618 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
620 # => started at: __lock_task_sighand
621 # => ended at: _raw_spin_unlock_irqrestore
625 # / _-----=> irqs-off
626 # | / _----=> need-resched
627 # || / _---=> hardirq/softirq
628 # ||| / _--=> preempt-depth
630 # cmd pid ||||| time | caller
632 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
633 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
634 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
635 ps-6143 2d..1 306us : <stack trace>
636 => trace_hardirqs_on_caller
638 => _raw_spin_unlock_irqrestore
645 => system_call_fastpath
648 This shows that the current tracer is "irqsoff" tracing the time
649 for which interrupts were disabled. It gives the trace version (which
650 never changes) and the version of the kernel upon which this was executed on
651 (3.10). Then it displays the max latency in microseconds (259 us). The number
652 of trace entries displayed and the total number (both are four: #4/4).
653 VP, KP, SP, and HP are always zero and are reserved for later use.
654 #P is the number of online CPUs (#P:4).
656 The task is the process that was running when the latency
657 occurred. (ps pid: 6143).
659 The start and stop (the functions in which the interrupts were
660 disabled and enabled respectively) that caused the latencies:
662 __lock_task_sighand is where the interrupts were disabled.
663 _raw_spin_unlock_irqrestore is where they were enabled again.
665 The next lines after the header are the trace itself. The header
666 explains which is which.
668 cmd: The name of the process in the trace.
670 pid: The PID of that process.
672 CPU#: The CPU which the process was running on.
674 irqs-off: 'd' interrupts are disabled. '.' otherwise.
675 Note: If the architecture does not support a way to
676 read the irq flags variable, an 'X' will always
680 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
681 'n' only TIF_NEED_RESCHED is set,
682 'p' only PREEMPT_NEED_RESCHED is set,
686 'H' - hard irq occurred inside a softirq.
687 'h' - hard irq is running
688 's' - soft irq is running
689 '.' - normal context.
691 preempt-depth: The level of preempt_disabled
693 The above is mostly meaningful for kernel developers.
695 time: When the latency-format option is enabled, the trace file
696 output includes a timestamp relative to the start of the
697 trace. This differs from the output when latency-format
698 is disabled, which includes an absolute timestamp.
700 delay: This is just to help catch your eye a bit better. And
701 needs to be fixed to be only relative to the same CPU.
702 The marks are determined by the difference between this
703 current trace and the next trace.
704 '$' - greater than 1 second
705 '@' - greater than 100 milisecond
706 '*' - greater than 10 milisecond
707 '#' - greater than 1000 microsecond
708 '!' - greater than 100 microsecond
709 '+' - greater than 10 microsecond
710 ' ' - less than or equal to 10 microsecond.
712 The rest is the same as the 'trace' file.
714 Note, the latency tracers will usually end with a back trace
715 to easily find where the latency occurred.
720 The trace_options file (or the options directory) is used to control
721 what gets printed in the trace output, or manipulate the tracers.
722 To see what is available, simply cat the file:
753 To disable one of the options, echo in the option prepended with
756 echo noprint-parent > trace_options
758 To enable an option, leave off the "no".
760 echo sym-offset > trace_options
762 Here are the available options:
764 print-parent - On function traces, display the calling (parent)
765 function as well as the function being traced.
768 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
771 bash-4000 [01] 1477.606694: simple_strtoul
774 sym-offset - Display not only the function name, but also the
775 offset in the function. For example, instead of
776 seeing just "ktime_get", you will see
777 "ktime_get+0xb/0x20".
780 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
782 sym-addr - this will also display the function address as well
783 as the function name.
786 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
788 verbose - This deals with the trace file when the
789 latency-format option is enabled.
791 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
792 (+0.000ms): simple_strtoul (kstrtoul)
794 raw - This will display raw numbers. This option is best for
795 use with user applications that can translate the raw
796 numbers better than having it done in the kernel.
798 hex - Similar to raw, but the numbers will be in a hexadecimal
801 bin - This will print out the formats in raw binary.
803 block - When set, reading trace_pipe will not block when polled.
805 trace_printk - Can disable trace_printk() from writing into the buffer.
807 branch - Enable branch tracing with the tracer.
809 annotate - It is sometimes confusing when the CPU buffers are full
810 and one CPU buffer had a lot of events recently, thus
811 a shorter time frame, were another CPU may have only had
812 a few events, which lets it have older events. When
813 the trace is reported, it shows the oldest events first,
814 and it may look like only one CPU ran (the one with the
815 oldest events). When the annotate option is set, it will
816 display when a new CPU buffer started:
818 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
819 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
820 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
821 ##### CPU 2 buffer started ####
822 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
823 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
824 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
826 userstacktrace - This option changes the trace. It records a
827 stacktrace of the current userspace thread.
829 sym-userobj - when user stacktrace are enabled, look up which
830 object the address belongs to, and print a
831 relative address. This is especially useful when
832 ASLR is on, otherwise you don't get a chance to
833 resolve the address to object/file/line after
834 the app is no longer running
836 The lookup is performed when you read
837 trace,trace_pipe. Example:
839 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
840 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
843 printk-msg-only - When set, trace_printk()s will only show the format
844 and not their parameters (if trace_bprintk() or
845 trace_bputs() was used to save the trace_printk()).
847 context-info - Show only the event data. Hides the comm, PID,
848 timestamp, CPU, and other useful data.
850 latency-format - This option changes the trace. When
851 it is enabled, the trace displays
852 additional information about the
853 latencies, as described in "Latency
856 sleep-time - When running function graph tracer, to include
857 the time a task schedules out in its function.
858 When enabled, it will account time the task has been
859 scheduled out as part of the function call.
861 graph-time - When running function profiler with function graph tracer,
862 to include the time to call nested functions. When this is
863 not set, the time reported for the function will only
864 include the time the function itself executed for, not the
865 time for functions that it called.
867 record-cmd - When any event or tracer is enabled, a hook is enabled
868 in the sched_switch trace point to fill comm cache
869 with mapped pids and comms. But this may cause some
870 overhead, and if you only care about pids, and not the
871 name of the task, disabling this option can lower the
874 overwrite - This controls what happens when the trace buffer is
875 full. If "1" (default), the oldest events are
876 discarded and overwritten. If "0", then the newest
877 events are discarded.
878 (see per_cpu/cpu0/stats for overrun and dropped)
880 disable_on_free - When the free_buffer is closed, tracing will
881 stop (tracing_on set to 0).
883 irq-info - Shows the interrupt, preempt count, need resched data.
884 When disabled, the trace looks like:
888 # entries-in-buffer/entries-written: 144405/9452052 #P:4
890 # TASK-PID CPU# TIMESTAMP FUNCTION
892 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
893 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
894 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
897 markers - When set, the trace_marker is writable (only by root).
898 When disabled, the trace_marker will error with EINVAL
901 event-fork - When set, tasks with PIDs listed in set_event_pid will have
902 the PIDs of their children added to set_event_pid when those
903 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
904 their PIDs will be removed from the file.
906 function-trace - The latency tracers will enable function tracing
907 if this option is enabled (default it is). When
908 it is disabled, the latency tracers do not trace
909 functions. This keeps the overhead of the tracer down
910 when performing latency tests.
912 display-graph - When set, the latency tracers (irqsoff, wakeup, etc) will
913 use function graph tracing instead of function tracing.
915 stacktrace - This is one of the options that changes the trace
916 itself. When a trace is recorded, so is the stack
917 of functions. This allows for back traces of
920 Note: Some tracers have their own options. They only appear in this
921 file when the tracer is active. They always appear in the
929 When interrupts are disabled, the CPU can not react to any other
930 external event (besides NMIs and SMIs). This prevents the timer
931 interrupt from triggering or the mouse interrupt from letting
932 the kernel know of a new mouse event. The result is a latency
933 with the reaction time.
935 The irqsoff tracer tracks the time for which interrupts are
936 disabled. When a new maximum latency is hit, the tracer saves
937 the trace leading up to that latency point so that every time a
938 new maximum is reached, the old saved trace is discarded and the
941 To reset the maximum, echo 0 into tracing_max_latency. Here is
944 # echo 0 > options/function-trace
945 # echo irqsoff > current_tracer
946 # echo 1 > tracing_on
947 # echo 0 > tracing_max_latency
950 # echo 0 > tracing_on
954 # irqsoff latency trace v1.1.5 on 3.8.0-test+
955 # --------------------------------------------------------------------
956 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
958 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
960 # => started at: run_timer_softirq
961 # => ended at: run_timer_softirq
965 # / _-----=> irqs-off
966 # | / _----=> need-resched
967 # || / _---=> hardirq/softirq
968 # ||| / _--=> preempt-depth
970 # cmd pid ||||| time | caller
972 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
973 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
974 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
975 <idle>-0 0dNs3 25us : <stack trace>
976 => _raw_spin_unlock_irq
982 => smp_apic_timer_interrupt
983 => apic_timer_interrupt
988 => x86_64_start_reservations
989 => x86_64_start_kernel
991 Here we see that that we had a latency of 16 microseconds (which is
992 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
993 interrupts. The difference between the 16 and the displayed
994 timestamp 25us occurred because the clock was incremented
995 between the time of recording the max latency and the time of
996 recording the function that had that latency.
998 Note the above example had function-trace not set. If we set
999 function-trace, we get a much larger output:
1001 with echo 1 > options/function-trace
1005 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1006 # --------------------------------------------------------------------
1007 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1009 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1011 # => started at: ata_scsi_queuecmd
1012 # => ended at: ata_scsi_queuecmd
1016 # / _-----=> irqs-off
1017 # | / _----=> need-resched
1018 # || / _---=> hardirq/softirq
1019 # ||| / _--=> preempt-depth
1021 # cmd pid ||||| time | caller
1023 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1024 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1025 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1026 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1027 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1028 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1029 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1030 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1031 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1033 bash-2042 3d..1 67us : delay_tsc <-__delay
1034 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1035 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1036 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1037 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1038 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1039 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1040 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1041 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1042 bash-2042 3d..1 120us : <stack trace>
1043 => _raw_spin_unlock_irqrestore
1044 => ata_scsi_queuecmd
1045 => scsi_dispatch_cmd
1047 => __blk_run_queue_uncond
1050 => generic_make_request
1053 => __ext3_get_inode_loc
1062 => user_path_at_empty
1067 => system_call_fastpath
1070 Here we traced a 71 microsecond latency. But we also see all the
1071 functions that were called during that time. Note that by
1072 enabling function tracing, we incur an added overhead. This
1073 overhead may extend the latency times. But nevertheless, this
1074 trace has provided some very helpful debugging information.
1080 When preemption is disabled, we may be able to receive
1081 interrupts but the task cannot be preempted and a higher
1082 priority task must wait for preemption to be enabled again
1083 before it can preempt a lower priority task.
1085 The preemptoff tracer traces the places that disable preemption.
1086 Like the irqsoff tracer, it records the maximum latency for
1087 which preemption was disabled. The control of preemptoff tracer
1088 is much like the irqsoff tracer.
1090 # echo 0 > options/function-trace
1091 # echo preemptoff > current_tracer
1092 # echo 1 > tracing_on
1093 # echo 0 > tracing_max_latency
1096 # echo 0 > tracing_on
1098 # tracer: preemptoff
1100 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1101 # --------------------------------------------------------------------
1102 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1104 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1106 # => started at: do_IRQ
1107 # => ended at: do_IRQ
1111 # / _-----=> irqs-off
1112 # | / _----=> need-resched
1113 # || / _---=> hardirq/softirq
1114 # ||| / _--=> preempt-depth
1116 # cmd pid ||||| time | caller
1118 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1119 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1120 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1121 sshd-1991 1d..1 52us : <stack trace>
1122 => sub_preempt_count
1128 This has some more changes. Preemption was disabled when an
1129 interrupt came in (notice the 'h'), and was enabled on exit.
1130 But we also see that interrupts have been disabled when entering
1131 the preempt off section and leaving it (the 'd'). We do not know if
1132 interrupts were enabled in the mean time or shortly after this
1135 # tracer: preemptoff
1137 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1138 # --------------------------------------------------------------------
1139 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1141 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1143 # => started at: wake_up_new_task
1144 # => ended at: task_rq_unlock
1148 # / _-----=> irqs-off
1149 # | / _----=> need-resched
1150 # || / _---=> hardirq/softirq
1151 # ||| / _--=> preempt-depth
1153 # cmd pid ||||| time | caller
1155 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1156 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1157 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1158 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1159 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1161 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1162 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1163 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1164 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1165 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1166 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1167 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1168 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1170 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1171 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1172 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1173 bash-1994 1d..2 36us : do_softirq <-irq_exit
1174 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1175 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1176 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1177 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1178 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1179 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1181 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1182 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1183 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1184 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1185 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1186 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1187 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1188 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1189 bash-1994 1.N.1 104us : <stack trace>
1190 => sub_preempt_count
1191 => _raw_spin_unlock_irqrestore
1199 The above is an example of the preemptoff trace with
1200 function-trace set. Here we see that interrupts were not disabled
1201 the entire time. The irq_enter code lets us know that we entered
1202 an interrupt 'h'. Before that, the functions being traced still
1203 show that it is not in an interrupt, but we can see from the
1204 functions themselves that this is not the case.
1209 Knowing the locations that have interrupts disabled or
1210 preemption disabled for the longest times is helpful. But
1211 sometimes we would like to know when either preemption and/or
1212 interrupts are disabled.
1214 Consider the following code:
1216 local_irq_disable();
1217 call_function_with_irqs_off();
1219 call_function_with_irqs_and_preemption_off();
1221 call_function_with_preemption_off();
1224 The irqsoff tracer will record the total length of
1225 call_function_with_irqs_off() and
1226 call_function_with_irqs_and_preemption_off().
1228 The preemptoff tracer will record the total length of
1229 call_function_with_irqs_and_preemption_off() and
1230 call_function_with_preemption_off().
1232 But neither will trace the time that interrupts and/or
1233 preemption is disabled. This total time is the time that we can
1234 not schedule. To record this time, use the preemptirqsoff
1237 Again, using this trace is much like the irqsoff and preemptoff
1240 # echo 0 > options/function-trace
1241 # echo preemptirqsoff > current_tracer
1242 # echo 1 > tracing_on
1243 # echo 0 > tracing_max_latency
1246 # echo 0 > tracing_on
1248 # tracer: preemptirqsoff
1250 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1251 # --------------------------------------------------------------------
1252 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1254 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1256 # => started at: ata_scsi_queuecmd
1257 # => ended at: ata_scsi_queuecmd
1261 # / _-----=> irqs-off
1262 # | / _----=> need-resched
1263 # || / _---=> hardirq/softirq
1264 # ||| / _--=> preempt-depth
1266 # cmd pid ||||| time | caller
1268 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1269 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1270 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1271 ls-2230 3...1 111us : <stack trace>
1272 => sub_preempt_count
1273 => _raw_spin_unlock_irqrestore
1274 => ata_scsi_queuecmd
1275 => scsi_dispatch_cmd
1277 => __blk_run_queue_uncond
1280 => generic_make_request
1285 => htree_dirblock_to_tree
1286 => ext3_htree_fill_tree
1290 => system_call_fastpath
1293 The trace_hardirqs_off_thunk is called from assembly on x86 when
1294 interrupts are disabled in the assembly code. Without the
1295 function tracing, we do not know if interrupts were enabled
1296 within the preemption points. We do see that it started with
1299 Here is a trace with function-trace set:
1301 # tracer: preemptirqsoff
1303 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1304 # --------------------------------------------------------------------
1305 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1307 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1309 # => started at: schedule
1310 # => ended at: mutex_unlock
1314 # / _-----=> irqs-off
1315 # | / _----=> need-resched
1316 # || / _---=> hardirq/softirq
1317 # ||| / _--=> preempt-depth
1319 # cmd pid ||||| time | caller
1321 kworker/-59 3...1 0us : __schedule <-schedule
1322 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1323 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1324 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1325 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1326 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1327 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1328 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1329 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1330 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1331 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1332 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1333 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1334 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1335 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1336 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1337 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1338 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1339 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1340 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1341 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1342 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1343 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1344 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1345 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1346 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1347 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1348 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1349 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1350 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1351 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1352 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1354 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1355 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1356 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1357 ls-2269 3d..3 21us : do_softirq <-irq_exit
1358 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1359 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1360 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1361 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1362 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1363 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1364 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1366 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1367 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1368 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1369 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1370 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1371 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1373 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1374 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1375 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1376 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1377 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1378 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1379 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1380 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1381 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1382 ls-2269 3d... 186us : <stack trace>
1383 => __mutex_unlock_slowpath
1390 => system_call_fastpath
1392 This is an interesting trace. It started with kworker running and
1393 scheduling out and ls taking over. But as soon as ls released the
1394 rq lock and enabled interrupts (but not preemption) an interrupt
1395 triggered. When the interrupt finished, it started running softirqs.
1396 But while the softirq was running, another interrupt triggered.
1397 When an interrupt is running inside a softirq, the annotation is 'H'.
1403 One common case that people are interested in tracing is the
1404 time it takes for a task that is woken to actually wake up.
1405 Now for non Real-Time tasks, this can be arbitrary. But tracing
1406 it none the less can be interesting.
1408 Without function tracing:
1410 # echo 0 > options/function-trace
1411 # echo wakeup > current_tracer
1412 # echo 1 > tracing_on
1413 # echo 0 > tracing_max_latency
1415 # echo 0 > tracing_on
1419 # wakeup latency trace v1.1.5 on 3.8.0-test+
1420 # --------------------------------------------------------------------
1421 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1423 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1427 # / _-----=> irqs-off
1428 # | / _----=> need-resched
1429 # || / _---=> hardirq/softirq
1430 # ||| / _--=> preempt-depth
1432 # cmd pid ||||| time | caller
1434 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1435 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1436 <idle>-0 3d..3 15us : __schedule <-schedule
1437 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1439 The tracer only traces the highest priority task in the system
1440 to avoid tracing the normal circumstances. Here we see that
1441 the kworker with a nice priority of -20 (not very nice), took
1442 just 15 microseconds from the time it woke up, to the time it
1445 Non Real-Time tasks are not that interesting. A more interesting
1446 trace is to concentrate only on Real-Time tasks.
1451 In a Real-Time environment it is very important to know the
1452 wakeup time it takes for the highest priority task that is woken
1453 up to the time that it executes. This is also known as "schedule
1454 latency". I stress the point that this is about RT tasks. It is
1455 also important to know the scheduling latency of non-RT tasks,
1456 but the average schedule latency is better for non-RT tasks.
1457 Tools like LatencyTop are more appropriate for such
1460 Real-Time environments are interested in the worst case latency.
1461 That is the longest latency it takes for something to happen,
1462 and not the average. We can have a very fast scheduler that may
1463 only have a large latency once in a while, but that would not
1464 work well with Real-Time tasks. The wakeup_rt tracer was designed
1465 to record the worst case wakeups of RT tasks. Non-RT tasks are
1466 not recorded because the tracer only records one worst case and
1467 tracing non-RT tasks that are unpredictable will overwrite the
1468 worst case latency of RT tasks (just run the normal wakeup
1469 tracer for a while to see that effect).
1471 Since this tracer only deals with RT tasks, we will run this
1472 slightly differently than we did with the previous tracers.
1473 Instead of performing an 'ls', we will run 'sleep 1' under
1474 'chrt' which changes the priority of the task.
1476 # echo 0 > options/function-trace
1477 # echo wakeup_rt > current_tracer
1478 # echo 1 > tracing_on
1479 # echo 0 > tracing_max_latency
1481 # echo 0 > tracing_on
1487 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1488 # --------------------------------------------------------------------
1489 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1491 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1495 # / _-----=> irqs-off
1496 # | / _----=> need-resched
1497 # || / _---=> hardirq/softirq
1498 # ||| / _--=> preempt-depth
1500 # cmd pid ||||| time | caller
1502 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1503 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1504 <idle>-0 3d..3 5us : __schedule <-schedule
1505 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1508 Running this on an idle system, we see that it only took 5 microseconds
1509 to perform the task switch. Note, since the trace point in the schedule
1510 is before the actual "switch", we stop the tracing when the recorded task
1511 is about to schedule in. This may change if we add a new marker at the
1512 end of the scheduler.
1514 Notice that the recorded task is 'sleep' with the PID of 2389
1515 and it has an rt_prio of 5. This priority is user-space priority
1516 and not the internal kernel priority. The policy is 1 for
1517 SCHED_FIFO and 2 for SCHED_RR.
1519 Note, that the trace data shows the internal priority (99 - rtprio).
1521 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1523 The 0:120:R means idle was running with a nice priority of 0 (120 - 20)
1524 and in the running state 'R'. The sleep task was scheduled in with
1525 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1526 and it too is in the running state.
1528 Doing the same with chrt -r 5 and function-trace set.
1530 echo 1 > options/function-trace
1534 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1535 # --------------------------------------------------------------------
1536 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1538 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
1542 # / _-----=> irqs-off
1543 # | / _----=> need-resched
1544 # || / _---=> hardirq/softirq
1545 # ||| / _--=> preempt-depth
1547 # cmd pid ||||| time | caller
1549 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
1550 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1551 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
1552 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
1553 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
1554 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
1555 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
1556 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
1557 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
1558 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1559 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
1560 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
1561 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
1562 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
1563 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
1564 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
1565 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
1566 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
1567 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
1568 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
1569 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
1570 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
1571 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
1572 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
1573 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
1574 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
1575 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
1576 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
1577 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
1578 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
1579 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
1580 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
1581 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
1582 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
1583 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
1584 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
1585 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
1586 <idle>-0 3dN.1 15us : calc_load_exit_idle <-tick_nohz_idle_exit
1587 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
1588 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
1589 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
1590 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
1591 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1592 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
1593 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
1594 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
1595 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
1596 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
1597 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
1598 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
1599 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
1600 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1601 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
1602 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1603 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1604 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
1605 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
1606 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
1607 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1608 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
1609 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
1610 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
1611 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
1612 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
1613 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
1614 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
1615 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
1616 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1617 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
1618 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
1619 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
1620 <idle>-0 3.N.. 25us : schedule <-cpu_idle
1621 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
1622 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
1623 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
1624 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
1625 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
1626 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
1627 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
1628 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
1629 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
1630 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
1631 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
1632 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
1633 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
1635 This isn't that big of a trace, even with function tracing enabled,
1636 so I included the entire trace.
1638 The interrupt went off while when the system was idle. Somewhere
1639 before task_woken_rt() was called, the NEED_RESCHED flag was set,
1640 this is indicated by the first occurrence of the 'N' flag.
1642 Latency tracing and events
1643 --------------------------
1644 As function tracing can induce a much larger latency, but without
1645 seeing what happens within the latency it is hard to know what
1646 caused it. There is a middle ground, and that is with enabling
1649 # echo 0 > options/function-trace
1650 # echo wakeup_rt > current_tracer
1651 # echo 1 > events/enable
1652 # echo 1 > tracing_on
1653 # echo 0 > tracing_max_latency
1655 # echo 0 > tracing_on
1659 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1660 # --------------------------------------------------------------------
1661 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1663 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
1667 # / _-----=> irqs-off
1668 # | / _----=> need-resched
1669 # || / _---=> hardirq/softirq
1670 # ||| / _--=> preempt-depth
1672 # cmd pid ||||| time | caller
1674 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
1675 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1676 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
1677 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
1678 <idle>-0 2.N.2 2us : power_end: cpu_id=2
1679 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
1680 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
1681 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
1682 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
1683 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
1684 <idle>-0 2d..3 6us : __schedule <-schedule
1685 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
1691 This tracer is the function tracer. Enabling the function tracer
1692 can be done from the debug file system. Make sure the
1693 ftrace_enabled is set; otherwise this tracer is a nop.
1694 See the "ftrace_enabled" section below.
1696 # sysctl kernel.ftrace_enabled=1
1697 # echo function > current_tracer
1698 # echo 1 > tracing_on
1700 # echo 0 > tracing_on
1704 # entries-in-buffer/entries-written: 24799/24799 #P:4
1707 # / _----=> need-resched
1708 # | / _---=> hardirq/softirq
1709 # || / _--=> preempt-depth
1711 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
1713 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
1714 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
1715 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
1716 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
1717 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
1718 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
1719 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
1720 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
1724 Note: function tracer uses ring buffers to store the above
1725 entries. The newest data may overwrite the oldest data.
1726 Sometimes using echo to stop the trace is not sufficient because
1727 the tracing could have overwritten the data that you wanted to
1728 record. For this reason, it is sometimes better to disable
1729 tracing directly from a program. This allows you to stop the
1730 tracing at the point that you hit the part that you are
1731 interested in. To disable the tracing directly from a C program,
1732 something like following code snippet can be used:
1736 int main(int argc, char *argv[]) {
1738 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
1740 if (condition_hit()) {
1741 write(trace_fd, "0", 1);
1747 Single thread tracing
1748 ---------------------
1750 By writing into set_ftrace_pid you can trace a
1751 single thread. For example:
1753 # cat set_ftrace_pid
1755 # echo 3111 > set_ftrace_pid
1756 # cat set_ftrace_pid
1758 # echo function > current_tracer
1762 # TASK-PID CPU# TIMESTAMP FUNCTION
1764 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
1765 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
1766 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
1767 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
1768 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
1769 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
1770 # echo > set_ftrace_pid
1774 # TASK-PID CPU# TIMESTAMP FUNCTION
1776 ##### CPU 3 buffer started ####
1777 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
1778 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
1779 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
1780 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
1781 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
1783 If you want to trace a function when executing, you could use
1784 something like this simple program:
1788 #include <sys/types.h>
1789 #include <sys/stat.h>
1795 #define STR(x) _STR(x)
1796 #define MAX_PATH 256
1798 const char *find_debugfs(void)
1800 static char debugfs[MAX_PATH+1];
1801 static int debugfs_found;
1808 if ((fp = fopen("/proc/mounts","r")) == NULL) {
1809 perror("/proc/mounts");
1813 while (fscanf(fp, "%*s %"
1815 "s %99s %*s %*d %*d\n",
1816 debugfs, type) == 2) {
1817 if (strcmp(type, "debugfs") == 0)
1822 if (strcmp(type, "debugfs") != 0) {
1823 fprintf(stderr, "debugfs not mounted");
1827 strcat(debugfs, "/tracing/");
1833 const char *tracing_file(const char *file_name)
1835 static char trace_file[MAX_PATH+1];
1836 snprintf(trace_file, MAX_PATH, "%s/%s", find_debugfs(), file_name);
1840 int main (int argc, char **argv)
1850 ffd = open(tracing_file("current_tracer"), O_WRONLY);
1853 write(ffd, "nop", 3);
1855 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
1856 s = sprintf(line, "%d\n", getpid());
1859 write(ffd, "function", 8);
1864 execvp(argv[1], argv+1);
1870 Or this simple script!
1875 debugfs=`sed -ne 's/^debugfs \(.*\) debugfs.*/\1/p' /proc/mounts`
1876 echo nop > $debugfs/tracing/current_tracer
1877 echo 0 > $debugfs/tracing/tracing_on
1878 echo $$ > $debugfs/tracing/set_ftrace_pid
1879 echo function > $debugfs/tracing/current_tracer
1880 echo 1 > $debugfs/tracing/tracing_on
1885 function graph tracer
1886 ---------------------------
1888 This tracer is similar to the function tracer except that it
1889 probes a function on its entry and its exit. This is done by
1890 using a dynamically allocated stack of return addresses in each
1891 task_struct. On function entry the tracer overwrites the return
1892 address of each function traced to set a custom probe. Thus the
1893 original return address is stored on the stack of return address
1896 Probing on both ends of a function leads to special features
1899 - measure of a function's time execution
1900 - having a reliable call stack to draw function calls graph
1902 This tracer is useful in several situations:
1904 - you want to find the reason of a strange kernel behavior and
1905 need to see what happens in detail on any areas (or specific
1908 - you are experiencing weird latencies but it's difficult to
1911 - you want to find quickly which path is taken by a specific
1914 - you just want to peek inside a working kernel and want to see
1917 # tracer: function_graph
1919 # CPU DURATION FUNCTION CALLS
1923 0) | do_sys_open() {
1925 0) | kmem_cache_alloc() {
1926 0) 1.382 us | __might_sleep();
1928 0) | strncpy_from_user() {
1929 0) | might_fault() {
1930 0) 1.389 us | __might_sleep();
1935 0) 0.668 us | _spin_lock();
1936 0) 0.570 us | expand_files();
1937 0) 0.586 us | _spin_unlock();
1940 There are several columns that can be dynamically
1941 enabled/disabled. You can use every combination of options you
1942 want, depending on your needs.
1944 - The cpu number on which the function executed is default
1945 enabled. It is sometimes better to only trace one cpu (see
1946 tracing_cpu_mask file) or you might sometimes see unordered
1947 function calls while cpu tracing switch.
1949 hide: echo nofuncgraph-cpu > trace_options
1950 show: echo funcgraph-cpu > trace_options
1952 - The duration (function's time of execution) is displayed on
1953 the closing bracket line of a function or on the same line
1954 than the current function in case of a leaf one. It is default
1957 hide: echo nofuncgraph-duration > trace_options
1958 show: echo funcgraph-duration > trace_options
1960 - The overhead field precedes the duration field in case of
1961 reached duration thresholds.
1963 hide: echo nofuncgraph-overhead > trace_options
1964 show: echo funcgraph-overhead > trace_options
1965 depends on: funcgraph-duration
1969 3) # 1837.709 us | } /* __switch_to */
1970 3) | finish_task_switch() {
1971 3) 0.313 us | _raw_spin_unlock_irq();
1973 3) # 1889.063 us | } /* __schedule */
1974 3) ! 140.417 us | } /* __schedule */
1975 3) # 2034.948 us | } /* schedule */
1976 3) * 33998.59 us | } /* schedule_preempt_disabled */
1980 1) 0.260 us | msecs_to_jiffies();
1981 1) 0.313 us | __rcu_read_unlock();
1984 1) 0.313 us | rcu_bh_qs();
1985 1) 0.313 us | __local_bh_enable();
1987 1) 0.365 us | idle_cpu();
1988 1) | rcu_irq_exit() {
1989 1) 0.417 us | rcu_eqs_enter_common.isra.47();
1993 1) @ 119760.2 us | }
1999 2) 0.417 us | scheduler_ipi();
2007 + means that the function exceeded 10 usecs.
2008 ! means that the function exceeded 100 usecs.
2009 # means that the function exceeded 1000 usecs.
2010 * means that the function exceeded 10 msecs.
2011 @ means that the function exceeded 100 msecs.
2012 $ means that the function exceeded 1 sec.
2015 - The task/pid field displays the thread cmdline and pid which
2016 executed the function. It is default disabled.
2018 hide: echo nofuncgraph-proc > trace_options
2019 show: echo funcgraph-proc > trace_options
2023 # tracer: function_graph
2025 # CPU TASK/PID DURATION FUNCTION CALLS
2027 0) sh-4802 | | d_free() {
2028 0) sh-4802 | | call_rcu() {
2029 0) sh-4802 | | __call_rcu() {
2030 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2031 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2032 0) sh-4802 | 2.899 us | }
2033 0) sh-4802 | 4.040 us | }
2034 0) sh-4802 | 5.151 us | }
2035 0) sh-4802 | + 49.370 us | }
2038 - The absolute time field is an absolute timestamp given by the
2039 system clock since it started. A snapshot of this time is
2040 given on each entry/exit of functions
2042 hide: echo nofuncgraph-abstime > trace_options
2043 show: echo funcgraph-abstime > trace_options
2048 # TIME CPU DURATION FUNCTION CALLS
2050 360.774522 | 1) 0.541 us | }
2051 360.774522 | 1) 4.663 us | }
2052 360.774523 | 1) 0.541 us | __wake_up_bit();
2053 360.774524 | 1) 6.796 us | }
2054 360.774524 | 1) 7.952 us | }
2055 360.774525 | 1) 9.063 us | }
2056 360.774525 | 1) 0.615 us | journal_mark_dirty();
2057 360.774527 | 1) 0.578 us | __brelse();
2058 360.774528 | 1) | reiserfs_prepare_for_journal() {
2059 360.774528 | 1) | unlock_buffer() {
2060 360.774529 | 1) | wake_up_bit() {
2061 360.774529 | 1) | bit_waitqueue() {
2062 360.774530 | 1) 0.594 us | __phys_addr();
2065 The function name is always displayed after the closing bracket
2066 for a function if the start of that function is not in the
2069 Display of the function name after the closing bracket may be
2070 enabled for functions whose start is in the trace buffer,
2071 allowing easier searching with grep for function durations.
2072 It is default disabled.
2074 hide: echo nofuncgraph-tail > trace_options
2075 show: echo funcgraph-tail > trace_options
2077 Example with nofuncgraph-tail (default):
2079 0) | kmem_cache_free() {
2080 0) 0.518 us | __phys_addr();
2084 Example with funcgraph-tail:
2086 0) | kmem_cache_free() {
2087 0) 0.518 us | __phys_addr();
2088 0) 1.757 us | } /* kmem_cache_free() */
2089 0) 2.861 us | } /* putname() */
2091 You can put some comments on specific functions by using
2092 trace_printk() For example, if you want to put a comment inside
2093 the __might_sleep() function, you just have to include
2094 <linux/ftrace.h> and call trace_printk() inside __might_sleep()
2096 trace_printk("I'm a comment!\n")
2100 1) | __might_sleep() {
2101 1) | /* I'm a comment! */
2105 You might find other useful features for this tracer in the
2106 following "dynamic ftrace" section such as tracing only specific
2112 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2113 virtually no overhead when function tracing is disabled. The way
2114 this works is the mcount function call (placed at the start of
2115 every kernel function, produced by the -pg switch in gcc),
2116 starts of pointing to a simple return. (Enabling FTRACE will
2117 include the -pg switch in the compiling of the kernel.)
2119 At compile time every C file object is run through the
2120 recordmcount program (located in the scripts directory). This
2121 program will parse the ELF headers in the C object to find all
2122 the locations in the .text section that call mcount. (Note, only
2123 white listed .text sections are processed, since processing other
2124 sections like .init.text may cause races due to those sections
2125 being freed unexpectedly).
2127 A new section called "__mcount_loc" is created that holds
2128 references to all the mcount call sites in the .text section.
2129 The recordmcount program re-links this section back into the
2130 original object. The final linking stage of the kernel will add all these
2131 references into a single table.
2133 On boot up, before SMP is initialized, the dynamic ftrace code
2134 scans this table and updates all the locations into nops. It
2135 also records the locations, which are added to the
2136 available_filter_functions list. Modules are processed as they
2137 are loaded and before they are executed. When a module is
2138 unloaded, it also removes its functions from the ftrace function
2139 list. This is automatic in the module unload code, and the
2140 module author does not need to worry about it.
2142 When tracing is enabled, the process of modifying the function
2143 tracepoints is dependent on architecture. The old method is to use
2144 kstop_machine to prevent races with the CPUs executing code being
2145 modified (which can cause the CPU to do undesirable things, especially
2146 if the modified code crosses cache (or page) boundaries), and the nops are
2147 patched back to calls. But this time, they do not call mcount
2148 (which is just a function stub). They now call into the ftrace
2151 The new method of modifying the function tracepoints is to place
2152 a breakpoint at the location to be modified, sync all CPUs, modify
2153 the rest of the instruction not covered by the breakpoint. Sync
2154 all CPUs again, and then remove the breakpoint with the finished
2155 version to the ftrace call site.
2157 Some archs do not even need to monkey around with the synchronization,
2158 and can just slap the new code on top of the old without any
2159 problems with other CPUs executing it at the same time.
2161 One special side-effect to the recording of the functions being
2162 traced is that we can now selectively choose which functions we
2163 wish to trace and which ones we want the mcount calls to remain
2166 Two files are used, one for enabling and one for disabling the
2167 tracing of specified functions. They are:
2175 A list of available functions that you can add to these files is
2178 available_filter_functions
2180 # cat available_filter_functions
2189 If I am only interested in sys_nanosleep and hrtimer_interrupt:
2191 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2192 # echo function > current_tracer
2193 # echo 1 > tracing_on
2195 # echo 0 > tracing_on
2199 # entries-in-buffer/entries-written: 5/5 #P:4
2202 # / _----=> need-resched
2203 # | / _---=> hardirq/softirq
2204 # || / _--=> preempt-depth
2206 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2208 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2209 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2210 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2211 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2212 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2214 To see which functions are being traced, you can cat the file:
2216 # cat set_ftrace_filter
2221 Perhaps this is not enough. The filters also allow simple wild
2222 cards. Only the following are currently available
2224 <match>* - will match functions that begin with <match>
2225 *<match> - will match functions that end with <match>
2226 *<match>* - will match functions that have <match> in it
2228 These are the only wild cards which are supported.
2230 <match>*<match> will not work.
2232 Note: It is better to use quotes to enclose the wild cards,
2233 otherwise the shell may expand the parameters into names
2234 of files in the local directory.
2236 # echo 'hrtimer_*' > set_ftrace_filter
2242 # entries-in-buffer/entries-written: 897/897 #P:4
2245 # / _----=> need-resched
2246 # | / _---=> hardirq/softirq
2247 # || / _--=> preempt-depth
2249 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2251 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2252 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2253 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2254 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2255 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2256 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2257 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2258 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2260 Notice that we lost the sys_nanosleep.
2262 # cat set_ftrace_filter
2267 hrtimer_try_to_cancel
2271 hrtimer_force_reprogram
2272 hrtimer_get_next_event
2276 hrtimer_get_remaining
2278 hrtimer_init_sleeper
2281 This is because the '>' and '>>' act just like they do in bash.
2282 To rewrite the filters, use '>'
2283 To append to the filters, use '>>'
2285 To clear out a filter so that all functions will be recorded
2288 # echo > set_ftrace_filter
2289 # cat set_ftrace_filter
2292 Again, now we want to append.
2294 # echo sys_nanosleep > set_ftrace_filter
2295 # cat set_ftrace_filter
2297 # echo 'hrtimer_*' >> set_ftrace_filter
2298 # cat set_ftrace_filter
2303 hrtimer_try_to_cancel
2307 hrtimer_force_reprogram
2308 hrtimer_get_next_event
2313 hrtimer_get_remaining
2315 hrtimer_init_sleeper
2318 The set_ftrace_notrace prevents those functions from being
2321 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2327 # entries-in-buffer/entries-written: 39608/39608 #P:4
2330 # / _----=> need-resched
2331 # | / _---=> hardirq/softirq
2332 # || / _--=> preempt-depth
2334 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2336 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2337 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2338 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2339 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2340 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2341 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2342 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2343 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2344 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2345 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2346 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2347 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2349 We can see that there's no more lock or preempt tracing.
2352 Dynamic ftrace with the function graph tracer
2353 ---------------------------------------------
2355 Although what has been explained above concerns both the
2356 function tracer and the function-graph-tracer, there are some
2357 special features only available in the function-graph tracer.
2359 If you want to trace only one function and all of its children,
2360 you just have to echo its name into set_graph_function:
2362 echo __do_fault > set_graph_function
2364 will produce the following "expanded" trace of the __do_fault()
2368 0) | filemap_fault() {
2369 0) | find_lock_page() {
2370 0) 0.804 us | find_get_page();
2371 0) | __might_sleep() {
2375 0) 0.653 us | _spin_lock();
2376 0) 0.578 us | page_add_file_rmap();
2377 0) 0.525 us | native_set_pte_at();
2378 0) 0.585 us | _spin_unlock();
2379 0) | unlock_page() {
2380 0) 0.541 us | page_waitqueue();
2381 0) 0.639 us | __wake_up_bit();
2385 0) | filemap_fault() {
2386 0) | find_lock_page() {
2387 0) 0.698 us | find_get_page();
2388 0) | __might_sleep() {
2392 0) 0.631 us | _spin_lock();
2393 0) 0.571 us | page_add_file_rmap();
2394 0) 0.526 us | native_set_pte_at();
2395 0) 0.586 us | _spin_unlock();
2396 0) | unlock_page() {
2397 0) 0.533 us | page_waitqueue();
2398 0) 0.638 us | __wake_up_bit();
2402 You can also expand several functions at once:
2404 echo sys_open > set_graph_function
2405 echo sys_close >> set_graph_function
2407 Now if you want to go back to trace all functions you can clear
2408 this special filter via:
2410 echo > set_graph_function
2416 Note, the proc sysctl ftrace_enable is a big on/off switch for the
2417 function tracer. By default it is enabled (when function tracing is
2418 enabled in the kernel). If it is disabled, all function tracing is
2419 disabled. This includes not only the function tracers for ftrace, but
2420 also for any other uses (perf, kprobes, stack tracing, profiling, etc).
2422 Please disable this with care.
2424 This can be disable (and enabled) with:
2426 sysctl kernel.ftrace_enabled=0
2427 sysctl kernel.ftrace_enabled=1
2431 echo 0 > /proc/sys/kernel/ftrace_enabled
2432 echo 1 > /proc/sys/kernel/ftrace_enabled
2438 A few commands are supported by the set_ftrace_filter interface.
2439 Trace commands have the following format:
2441 <function>:<command>:<parameter>
2443 The following commands are supported:
2446 This command enables function filtering per module. The
2447 parameter defines the module. For example, if only the write*
2448 functions in the ext3 module are desired, run:
2450 echo 'write*:mod:ext3' > set_ftrace_filter
2452 This command interacts with the filter in the same way as
2453 filtering based on function names. Thus, adding more functions
2454 in a different module is accomplished by appending (>>) to the
2455 filter file. Remove specific module functions by prepending
2458 echo '!writeback*:mod:ext3' >> set_ftrace_filter
2460 Mod command supports module globbing. Disable tracing for all
2461 functions except a specific module:
2463 echo '!*:mod:!ext3' >> set_ftrace_filter
2465 Disable tracing for all modules, but still trace kernel:
2467 echo '!*:mod:*' >> set_ftrace_filter
2469 Enable filter only for kernel:
2471 echo '*write*:mod:!*' >> set_ftrace_filter
2473 Enable filter for module globbing:
2475 echo '*write*:mod:*snd*' >> set_ftrace_filter
2478 These commands turn tracing on and off when the specified
2479 functions are hit. The parameter determines how many times the
2480 tracing system is turned on and off. If unspecified, there is
2481 no limit. For example, to disable tracing when a schedule bug
2482 is hit the first 5 times, run:
2484 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
2486 To always disable tracing when __schedule_bug is hit:
2488 echo '__schedule_bug:traceoff' > set_ftrace_filter
2490 These commands are cumulative whether or not they are appended
2491 to set_ftrace_filter. To remove a command, prepend it by '!'
2492 and drop the parameter:
2494 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
2496 The above removes the traceoff command for __schedule_bug
2497 that have a counter. To remove commands without counters:
2499 echo '!__schedule_bug:traceoff' > set_ftrace_filter
2502 Will cause a snapshot to be triggered when the function is hit.
2504 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
2506 To only snapshot once:
2508 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
2510 To remove the above commands:
2512 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
2513 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
2515 - enable_event/disable_event
2516 These commands can enable or disable a trace event. Note, because
2517 function tracing callbacks are very sensitive, when these commands
2518 are registered, the trace point is activated, but disabled in
2519 a "soft" mode. That is, the tracepoint will be called, but
2520 just will not be traced. The event tracepoint stays in this mode
2521 as long as there's a command that triggers it.
2523 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
2528 <function>:enable_event:<system>:<event>[:count]
2529 <function>:disable_event:<system>:<event>[:count]
2531 To remove the events commands:
2534 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
2536 echo '!schedule:disable_event:sched:sched_switch' > \
2540 When the function is hit, it will dump the contents of the ftrace
2541 ring buffer to the console. This is useful if you need to debug
2542 something, and want to dump the trace when a certain function
2543 is hit. Perhaps its a function that is called before a tripple
2544 fault happens and does not allow you to get a regular dump.
2547 When the function is hit, it will dump the contents of the ftrace
2548 ring buffer for the current CPU to the console. Unlike the "dump"
2549 command, it only prints out the contents of the ring buffer for the
2550 CPU that executed the function that triggered the dump.
2555 The trace_pipe outputs the same content as the trace file, but
2556 the effect on the tracing is different. Every read from
2557 trace_pipe is consumed. This means that subsequent reads will be
2558 different. The trace is live.
2560 # echo function > current_tracer
2561 # cat trace_pipe > /tmp/trace.out &
2563 # echo 1 > tracing_on
2565 # echo 0 > tracing_on
2569 # entries-in-buffer/entries-written: 0/0 #P:4
2572 # / _----=> need-resched
2573 # | / _---=> hardirq/softirq
2574 # || / _--=> preempt-depth
2576 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2580 # cat /tmp/trace.out
2581 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
2582 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
2583 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
2584 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
2585 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
2586 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
2587 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
2588 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
2589 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
2592 Note, reading the trace_pipe file will block until more input is
2598 Having too much or not enough data can be troublesome in
2599 diagnosing an issue in the kernel. The file buffer_size_kb is
2600 used to modify the size of the internal trace buffers. The
2601 number listed is the number of entries that can be recorded per
2602 CPU. To know the full size, multiply the number of possible CPUs
2603 with the number of entries.
2605 # cat buffer_size_kb
2606 1408 (units kilobytes)
2608 Or simply read buffer_total_size_kb
2610 # cat buffer_total_size_kb
2613 To modify the buffer, simple echo in a number (in 1024 byte segments).
2615 # echo 10000 > buffer_size_kb
2616 # cat buffer_size_kb
2617 10000 (units kilobytes)
2619 It will try to allocate as much as possible. If you allocate too
2620 much, it can cause Out-Of-Memory to trigger.
2622 # echo 1000000000000 > buffer_size_kb
2623 -bash: echo: write error: Cannot allocate memory
2624 # cat buffer_size_kb
2627 The per_cpu buffers can be changed individually as well:
2629 # echo 10000 > per_cpu/cpu0/buffer_size_kb
2630 # echo 100 > per_cpu/cpu1/buffer_size_kb
2632 When the per_cpu buffers are not the same, the buffer_size_kb
2633 at the top level will just show an X
2635 # cat buffer_size_kb
2638 This is where the buffer_total_size_kb is useful:
2640 # cat buffer_total_size_kb
2643 Writing to the top level buffer_size_kb will reset all the buffers
2644 to be the same again.
2648 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
2649 available to all non latency tracers. (Latency tracers which
2650 record max latency, such as "irqsoff" or "wakeup", can't use
2651 this feature, since those are already using the snapshot
2652 mechanism internally.)
2654 Snapshot preserves a current trace buffer at a particular point
2655 in time without stopping tracing. Ftrace swaps the current
2656 buffer with a spare buffer, and tracing continues in the new
2657 current (=previous spare) buffer.
2659 The following debugfs files in "tracing" are related to this
2664 This is used to take a snapshot and to read the output
2665 of the snapshot. Echo 1 into this file to allocate a
2666 spare buffer and to take a snapshot (swap), then read
2667 the snapshot from this file in the same format as
2668 "trace" (described above in the section "The File
2669 System"). Both reads snapshot and tracing are executable
2670 in parallel. When the spare buffer is allocated, echoing
2671 0 frees it, and echoing else (positive) values clear the
2673 More details are shown in the table below.
2675 status\input | 0 | 1 | else |
2676 --------------+------------+------------+------------+
2677 not allocated |(do nothing)| alloc+swap |(do nothing)|
2678 --------------+------------+------------+------------+
2679 allocated | free | swap | clear |
2680 --------------+------------+------------+------------+
2682 Here is an example of using the snapshot feature.
2684 # echo 1 > events/sched/enable
2689 # entries-in-buffer/entries-written: 71/71 #P:8
2692 # / _----=> need-resched
2693 # | / _---=> hardirq/softirq
2694 # || / _--=> preempt-depth
2696 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2698 <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
2699 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
2701 <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
2706 # entries-in-buffer/entries-written: 77/77 #P:8
2709 # / _----=> need-resched
2710 # | / _---=> hardirq/softirq
2711 # || / _--=> preempt-depth
2713 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2715 <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
2716 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
2720 If you try to use this snapshot feature when current tracer is
2721 one of the latency tracers, you will get the following results.
2723 # echo wakeup > current_tracer
2725 bash: echo: write error: Device or resource busy
2727 cat: snapshot: Device or resource busy
2732 In the debugfs tracing directory is a directory called "instances".
2733 This directory can have new directories created inside of it using
2734 mkdir, and removing directories with rmdir. The directory created
2735 with mkdir in this directory will already contain files and other
2736 directories after it is created.
2738 # mkdir instances/foo
2740 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
2741 set_event snapshot trace trace_clock trace_marker trace_options
2742 trace_pipe tracing_on
2744 As you can see, the new directory looks similar to the tracing directory
2745 itself. In fact, it is very similar, except that the buffer and
2746 events are agnostic from the main director, or from any other
2747 instances that are created.
2749 The files in the new directory work just like the files with the
2750 same name in the tracing directory except the buffer that is used
2751 is a separate and new buffer. The files affect that buffer but do not
2752 affect the main buffer with the exception of trace_options. Currently,
2753 the trace_options affect all instances and the top level buffer
2754 the same, but this may change in future releases. That is, options
2755 may become specific to the instance they reside in.
2757 Notice that none of the function tracer files are there, nor is
2758 current_tracer and available_tracers. This is because the buffers
2759 can currently only have events enabled for them.
2761 # mkdir instances/foo
2762 # mkdir instances/bar
2763 # mkdir instances/zoot
2764 # echo 100000 > buffer_size_kb
2765 # echo 1000 > instances/foo/buffer_size_kb
2766 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
2767 # echo function > current_trace
2768 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
2769 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
2770 # echo 1 > instances/foo/events/sched/sched_switch/enable
2771 # echo 1 > instances/bar/events/irq/enable
2772 # echo 1 > instances/zoot/events/syscalls/enable
2774 CPU:2 [LOST 11745 EVENTS]
2775 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
2776 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
2777 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
2778 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
2779 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
2780 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
2781 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
2782 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
2783 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
2784 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
2785 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
2788 # cat instances/foo/trace_pipe
2789 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
2790 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
2791 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
2792 <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
2793 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
2794 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
2795 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
2796 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
2797 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
2798 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
2801 # cat instances/bar/trace_pipe
2802 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
2803 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
2804 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
2805 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
2806 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
2807 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
2808 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
2809 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
2810 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
2811 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
2812 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
2813 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
2816 # cat instances/zoot/trace
2819 # entries-in-buffer/entries-written: 18996/18996 #P:4
2822 # / _----=> need-resched
2823 # | / _---=> hardirq/softirq
2824 # || / _--=> preempt-depth
2826 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2828 bash-1998 [000] d... 140.733501: sys_write -> 0x2
2829 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
2830 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
2831 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
2832 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
2833 bash-1998 [000] d... 140.733510: sys_close(fd: a)
2834 bash-1998 [000] d... 140.733510: sys_close -> 0x0
2835 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
2836 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
2837 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
2838 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
2840 You can see that the trace of the top most trace buffer shows only
2841 the function tracing. The foo instance displays wakeups and task
2844 To remove the instances, simply delete their directories:
2846 # rmdir instances/foo
2847 # rmdir instances/bar
2848 # rmdir instances/zoot
2850 Note, if a process has a trace file open in one of the instance
2851 directories, the rmdir will fail with EBUSY.
2856 Since the kernel has a fixed sized stack, it is important not to
2857 waste it in functions. A kernel developer must be conscience of
2858 what they allocate on the stack. If they add too much, the system
2859 can be in danger of a stack overflow, and corruption will occur,
2860 usually leading to a system panic.
2862 There are some tools that check this, usually with interrupts
2863 periodically checking usage. But if you can perform a check
2864 at every function call that will become very useful. As ftrace provides
2865 a function tracer, it makes it convenient to check the stack size
2866 at every function call. This is enabled via the stack tracer.
2868 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
2869 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
2871 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
2873 You can also enable it from the kernel command line to trace
2874 the stack size of the kernel during boot up, by adding "stacktrace"
2875 to the kernel command line parameter.
2877 After running it for a few minutes, the output looks like:
2879 # cat stack_max_size
2883 Depth Size Location (18 entries)
2885 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
2886 1) 2704 160 find_busiest_group+0x31/0x1f1
2887 2) 2544 256 load_balance+0xd9/0x662
2888 3) 2288 80 idle_balance+0xbb/0x130
2889 4) 2208 128 __schedule+0x26e/0x5b9
2890 5) 2080 16 schedule+0x64/0x66
2891 6) 2064 128 schedule_timeout+0x34/0xe0
2892 7) 1936 112 wait_for_common+0x97/0xf1
2893 8) 1824 16 wait_for_completion+0x1d/0x1f
2894 9) 1808 128 flush_work+0xfe/0x119
2895 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
2896 11) 1664 48 input_available_p+0x1d/0x5c
2897 12) 1616 48 n_tty_poll+0x6d/0x134
2898 13) 1568 64 tty_poll+0x64/0x7f
2899 14) 1504 880 do_select+0x31e/0x511
2900 15) 624 400 core_sys_select+0x177/0x216
2901 16) 224 96 sys_select+0x91/0xb9
2902 17) 128 128 system_call_fastpath+0x16/0x1b
2904 Note, if -mfentry is being used by gcc, functions get traced before
2905 they set up the stack frame. This means that leaf level functions
2906 are not tested by the stack tracer when -mfentry is used.
2908 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
2912 More details can be found in the source code, in the
2913 kernel/trace/*.c files.
projects / cascardo / linux.git / blob