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 mono: This uses the fast monotonic clock (CLOCK_MONOTONIC)
366 which is monotonic and is subject to NTP rate adjustments.
369 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
370 which is montonic but is not subject to any rate adjustments
371 and ticks at the same rate as the hardware clocksource.
373 boot: This is the boot clock (CLOCK_BOOTTIME) and is based on the
374 fast monotonic clock, but also accounts for time spent in
375 suspend. Since the clock access is designed for use in
376 tracing in the suspend path, some side effects are possible
377 if clock is accessed after the suspend time is accounted before
378 the fast mono clock is updated. In this case, the clock update
379 appears to happen slightly sooner than it normally would have.
380 Also on 32-bit systems, it's possible that the 64-bit boot offset
381 sees a partial update. These effects are rare and post
382 processing should be able to handle them. See comments in the
383 ktime_get_boot_fast_ns() function for more information.
385 To set a clock, simply echo the clock name into this file.
387 echo global > trace_clock
391 This is a very useful file for synchronizing user space
392 with events happening in the kernel. Writing strings into
393 this file will be written into the ftrace buffer.
395 It is useful in applications to open this file at the start
396 of the application and just reference the file descriptor
399 void trace_write(const char *fmt, ...)
409 n = vsnprintf(buf, 256, fmt, ap);
412 write(trace_fd, buf, n);
417 trace_fd = open("trace_marker", WR_ONLY);
421 Add dynamic tracepoints in programs.
426 Uprobe statistics. See uprobetrace.txt
430 This is a way to make multiple trace buffers where different
431 events can be recorded in different buffers.
432 See "Instances" section below.
436 This is the trace event directory. It holds event tracepoints
437 (also known as static tracepoints) that have been compiled
438 into the kernel. It shows what event tracepoints exist
439 and how they are grouped by system. There are "enable"
440 files at various levels that can enable the tracepoints
441 when a "1" is written to them.
443 See events.txt for more information.
447 This is a directory that contains the trace per_cpu information.
449 per_cpu/cpu0/buffer_size_kb:
451 The ftrace buffer is defined per_cpu. That is, there's a separate
452 buffer for each CPU to allow writes to be done atomically,
453 and free from cache bouncing. These buffers may have different
454 size buffers. This file is similar to the buffer_size_kb
455 file, but it only displays or sets the buffer size for the
456 specific CPU. (here cpu0).
460 This is similar to the "trace" file, but it will only display
461 the data specific for the CPU. If written to, it only clears
462 the specific CPU buffer.
464 per_cpu/cpu0/trace_pipe
466 This is similar to the "trace_pipe" file, and is a consuming
467 read, but it will only display (and consume) the data specific
470 per_cpu/cpu0/trace_pipe_raw
472 For tools that can parse the ftrace ring buffer binary format,
473 the trace_pipe_raw file can be used to extract the data
474 from the ring buffer directly. With the use of the splice()
475 system call, the buffer data can be quickly transferred to
476 a file or to the network where a server is collecting the
479 Like trace_pipe, this is a consuming reader, where multiple
480 reads will always produce different data.
482 per_cpu/cpu0/snapshot:
484 This is similar to the main "snapshot" file, but will only
485 snapshot the current CPU (if supported). It only displays
486 the content of the snapshot for a given CPU, and if
487 written to, only clears this CPU buffer.
489 per_cpu/cpu0/snapshot_raw:
491 Similar to the trace_pipe_raw, but will read the binary format
492 from the snapshot buffer for the given CPU.
496 This displays certain stats about the ring buffer:
498 entries: The number of events that are still in the buffer.
500 overrun: The number of lost events due to overwriting when
503 commit overrun: Should always be zero.
504 This gets set if so many events happened within a nested
505 event (ring buffer is re-entrant), that it fills the
506 buffer and starts dropping events.
508 bytes: Bytes actually read (not overwritten).
510 oldest event ts: The oldest timestamp in the buffer
512 now ts: The current timestamp
514 dropped events: Events lost due to overwrite option being off.
516 read events: The number of events read.
521 Here is the list of current tracers that may be configured.
525 Function call tracer to trace all kernel functions.
529 Similar to the function tracer except that the
530 function tracer probes the functions on their entry
531 whereas the function graph tracer traces on both entry
532 and exit of the functions. It then provides the ability
533 to draw a graph of function calls similar to C code
538 Traces the areas that disable interrupts and saves
539 the trace with the longest max latency.
540 See tracing_max_latency. When a new max is recorded,
541 it replaces the old trace. It is best to view this
542 trace with the latency-format option enabled.
546 Similar to irqsoff but traces and records the amount of
547 time for which preemption is disabled.
551 Similar to irqsoff and preemptoff, but traces and
552 records the largest time for which irqs and/or preemption
557 Traces and records the max latency that it takes for
558 the highest priority task to get scheduled after
559 it has been woken up.
560 Traces all tasks as an average developer would expect.
564 Traces and records the max latency that it takes for just
565 RT tasks (as the current "wakeup" does). This is useful
566 for those interested in wake up timings of RT tasks.
570 This is the "trace nothing" tracer. To remove all
571 tracers from tracing simply echo "nop" into
575 Examples of using the tracer
576 ----------------------------
578 Here are typical examples of using the tracers when controlling
579 them only with the debugfs interface (without using any
580 user-land utilities).
585 Here is an example of the output format of the file "trace"
590 # entries-in-buffer/entries-written: 140080/250280 #P:4
593 # / _----=> need-resched
594 # | / _---=> hardirq/softirq
595 # || / _--=> preempt-depth
597 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
599 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
600 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
601 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
602 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
603 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
604 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
605 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
606 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
607 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
608 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
611 A header is printed with the tracer name that is represented by
612 the trace. In this case the tracer is "function". Then it shows the
613 number of events in the buffer as well as the total number of entries
614 that were written. The difference is the number of entries that were
615 lost due to the buffer filling up (250280 - 140080 = 110200 events
618 The header explains the content of the events. Task name "bash", the task
619 PID "1977", the CPU that it was running on "000", the latency format
620 (explained below), the timestamp in <secs>.<usecs> format, the
621 function name that was traced "sys_close" and the parent function that
622 called this function "system_call_fastpath". The timestamp is the time
623 at which the function was entered.
628 When the latency-format option is enabled or when one of the latency
629 tracers is set, the trace file gives somewhat more information to see
630 why a latency happened. Here is a typical trace.
634 # irqsoff latency trace v1.1.5 on 3.8.0-test+
635 # --------------------------------------------------------------------
636 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
638 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
640 # => started at: __lock_task_sighand
641 # => ended at: _raw_spin_unlock_irqrestore
645 # / _-----=> irqs-off
646 # | / _----=> need-resched
647 # || / _---=> hardirq/softirq
648 # ||| / _--=> preempt-depth
650 # cmd pid ||||| time | caller
652 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
653 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
654 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
655 ps-6143 2d..1 306us : <stack trace>
656 => trace_hardirqs_on_caller
658 => _raw_spin_unlock_irqrestore
665 => system_call_fastpath
668 This shows that the current tracer is "irqsoff" tracing the time
669 for which interrupts were disabled. It gives the trace version (which
670 never changes) and the version of the kernel upon which this was executed on
671 (3.10). Then it displays the max latency in microseconds (259 us). The number
672 of trace entries displayed and the total number (both are four: #4/4).
673 VP, KP, SP, and HP are always zero and are reserved for later use.
674 #P is the number of online CPUs (#P:4).
676 The task is the process that was running when the latency
677 occurred. (ps pid: 6143).
679 The start and stop (the functions in which the interrupts were
680 disabled and enabled respectively) that caused the latencies:
682 __lock_task_sighand is where the interrupts were disabled.
683 _raw_spin_unlock_irqrestore is where they were enabled again.
685 The next lines after the header are the trace itself. The header
686 explains which is which.
688 cmd: The name of the process in the trace.
690 pid: The PID of that process.
692 CPU#: The CPU which the process was running on.
694 irqs-off: 'd' interrupts are disabled. '.' otherwise.
695 Note: If the architecture does not support a way to
696 read the irq flags variable, an 'X' will always
700 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
701 'n' only TIF_NEED_RESCHED is set,
702 'p' only PREEMPT_NEED_RESCHED is set,
706 'H' - hard irq occurred inside a softirq.
707 'h' - hard irq is running
708 's' - soft irq is running
709 '.' - normal context.
711 preempt-depth: The level of preempt_disabled
713 The above is mostly meaningful for kernel developers.
715 time: When the latency-format option is enabled, the trace file
716 output includes a timestamp relative to the start of the
717 trace. This differs from the output when latency-format
718 is disabled, which includes an absolute timestamp.
720 delay: This is just to help catch your eye a bit better. And
721 needs to be fixed to be only relative to the same CPU.
722 The marks are determined by the difference between this
723 current trace and the next trace.
724 '$' - greater than 1 second
725 '@' - greater than 100 milisecond
726 '*' - greater than 10 milisecond
727 '#' - greater than 1000 microsecond
728 '!' - greater than 100 microsecond
729 '+' - greater than 10 microsecond
730 ' ' - less than or equal to 10 microsecond.
732 The rest is the same as the 'trace' file.
734 Note, the latency tracers will usually end with a back trace
735 to easily find where the latency occurred.
740 The trace_options file (or the options directory) is used to control
741 what gets printed in the trace output, or manipulate the tracers.
742 To see what is available, simply cat the file:
773 To disable one of the options, echo in the option prepended with
776 echo noprint-parent > trace_options
778 To enable an option, leave off the "no".
780 echo sym-offset > trace_options
782 Here are the available options:
784 print-parent - On function traces, display the calling (parent)
785 function as well as the function being traced.
788 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
791 bash-4000 [01] 1477.606694: simple_strtoul
794 sym-offset - Display not only the function name, but also the
795 offset in the function. For example, instead of
796 seeing just "ktime_get", you will see
797 "ktime_get+0xb/0x20".
800 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
802 sym-addr - this will also display the function address as well
803 as the function name.
806 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
808 verbose - This deals with the trace file when the
809 latency-format option is enabled.
811 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
812 (+0.000ms): simple_strtoul (kstrtoul)
814 raw - This will display raw numbers. This option is best for
815 use with user applications that can translate the raw
816 numbers better than having it done in the kernel.
818 hex - Similar to raw, but the numbers will be in a hexadecimal
821 bin - This will print out the formats in raw binary.
823 block - When set, reading trace_pipe will not block when polled.
825 trace_printk - Can disable trace_printk() from writing into the buffer.
827 branch - Enable branch tracing with the tracer.
829 annotate - It is sometimes confusing when the CPU buffers are full
830 and one CPU buffer had a lot of events recently, thus
831 a shorter time frame, were another CPU may have only had
832 a few events, which lets it have older events. When
833 the trace is reported, it shows the oldest events first,
834 and it may look like only one CPU ran (the one with the
835 oldest events). When the annotate option is set, it will
836 display when a new CPU buffer started:
838 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
839 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
840 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
841 ##### CPU 2 buffer started ####
842 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
843 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
844 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
846 userstacktrace - This option changes the trace. It records a
847 stacktrace of the current userspace thread.
849 sym-userobj - when user stacktrace are enabled, look up which
850 object the address belongs to, and print a
851 relative address. This is especially useful when
852 ASLR is on, otherwise you don't get a chance to
853 resolve the address to object/file/line after
854 the app is no longer running
856 The lookup is performed when you read
857 trace,trace_pipe. Example:
859 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
860 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
863 printk-msg-only - When set, trace_printk()s will only show the format
864 and not their parameters (if trace_bprintk() or
865 trace_bputs() was used to save the trace_printk()).
867 context-info - Show only the event data. Hides the comm, PID,
868 timestamp, CPU, and other useful data.
870 latency-format - This option changes the trace. When
871 it is enabled, the trace displays
872 additional information about the
873 latencies, as described in "Latency
876 sleep-time - When running function graph tracer, to include
877 the time a task schedules out in its function.
878 When enabled, it will account time the task has been
879 scheduled out as part of the function call.
881 graph-time - When running function profiler with function graph tracer,
882 to include the time to call nested functions. When this is
883 not set, the time reported for the function will only
884 include the time the function itself executed for, not the
885 time for functions that it called.
887 record-cmd - When any event or tracer is enabled, a hook is enabled
888 in the sched_switch trace point to fill comm cache
889 with mapped pids and comms. But this may cause some
890 overhead, and if you only care about pids, and not the
891 name of the task, disabling this option can lower the
894 overwrite - This controls what happens when the trace buffer is
895 full. If "1" (default), the oldest events are
896 discarded and overwritten. If "0", then the newest
897 events are discarded.
898 (see per_cpu/cpu0/stats for overrun and dropped)
900 disable_on_free - When the free_buffer is closed, tracing will
901 stop (tracing_on set to 0).
903 irq-info - Shows the interrupt, preempt count, need resched data.
904 When disabled, the trace looks like:
908 # entries-in-buffer/entries-written: 144405/9452052 #P:4
910 # TASK-PID CPU# TIMESTAMP FUNCTION
912 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
913 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
914 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
917 markers - When set, the trace_marker is writable (only by root).
918 When disabled, the trace_marker will error with EINVAL
921 event-fork - When set, tasks with PIDs listed in set_event_pid will have
922 the PIDs of their children added to set_event_pid when those
923 tasks fork. Also, when tasks with PIDs in set_event_pid exit,
924 their PIDs will be removed from the file.
926 function-trace - The latency tracers will enable function tracing
927 if this option is enabled (default it is). When
928 it is disabled, the latency tracers do not trace
929 functions. This keeps the overhead of the tracer down
930 when performing latency tests.
932 display-graph - When set, the latency tracers (irqsoff, wakeup, etc) will
933 use function graph tracing instead of function tracing.
935 stacktrace - This is one of the options that changes the trace
936 itself. When a trace is recorded, so is the stack
937 of functions. This allows for back traces of
940 Note: Some tracers have their own options. They only appear in this
941 file when the tracer is active. They always appear in the
949 When interrupts are disabled, the CPU can not react to any other
950 external event (besides NMIs and SMIs). This prevents the timer
951 interrupt from triggering or the mouse interrupt from letting
952 the kernel know of a new mouse event. The result is a latency
953 with the reaction time.
955 The irqsoff tracer tracks the time for which interrupts are
956 disabled. When a new maximum latency is hit, the tracer saves
957 the trace leading up to that latency point so that every time a
958 new maximum is reached, the old saved trace is discarded and the
961 To reset the maximum, echo 0 into tracing_max_latency. Here is
964 # echo 0 > options/function-trace
965 # echo irqsoff > current_tracer
966 # echo 1 > tracing_on
967 # echo 0 > tracing_max_latency
970 # echo 0 > tracing_on
974 # irqsoff latency trace v1.1.5 on 3.8.0-test+
975 # --------------------------------------------------------------------
976 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
978 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
980 # => started at: run_timer_softirq
981 # => ended at: run_timer_softirq
985 # / _-----=> irqs-off
986 # | / _----=> need-resched
987 # || / _---=> hardirq/softirq
988 # ||| / _--=> preempt-depth
990 # cmd pid ||||| time | caller
992 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
993 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
994 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
995 <idle>-0 0dNs3 25us : <stack trace>
996 => _raw_spin_unlock_irq
1002 => smp_apic_timer_interrupt
1003 => apic_timer_interrupt
1008 => x86_64_start_reservations
1009 => x86_64_start_kernel
1011 Here we see that that we had a latency of 16 microseconds (which is
1012 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
1013 interrupts. The difference between the 16 and the displayed
1014 timestamp 25us occurred because the clock was incremented
1015 between the time of recording the max latency and the time of
1016 recording the function that had that latency.
1018 Note the above example had function-trace not set. If we set
1019 function-trace, we get a much larger output:
1021 with echo 1 > options/function-trace
1025 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1026 # --------------------------------------------------------------------
1027 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1029 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1031 # => started at: ata_scsi_queuecmd
1032 # => ended at: ata_scsi_queuecmd
1036 # / _-----=> irqs-off
1037 # | / _----=> need-resched
1038 # || / _---=> hardirq/softirq
1039 # ||| / _--=> preempt-depth
1041 # cmd pid ||||| time | caller
1043 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1044 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1045 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1046 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1047 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1048 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1049 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1050 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1051 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1053 bash-2042 3d..1 67us : delay_tsc <-__delay
1054 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1055 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1056 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1057 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1058 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1059 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1060 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1061 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1062 bash-2042 3d..1 120us : <stack trace>
1063 => _raw_spin_unlock_irqrestore
1064 => ata_scsi_queuecmd
1065 => scsi_dispatch_cmd
1067 => __blk_run_queue_uncond
1070 => generic_make_request
1073 => __ext3_get_inode_loc
1082 => user_path_at_empty
1087 => system_call_fastpath
1090 Here we traced a 71 microsecond latency. But we also see all the
1091 functions that were called during that time. Note that by
1092 enabling function tracing, we incur an added overhead. This
1093 overhead may extend the latency times. But nevertheless, this
1094 trace has provided some very helpful debugging information.
1100 When preemption is disabled, we may be able to receive
1101 interrupts but the task cannot be preempted and a higher
1102 priority task must wait for preemption to be enabled again
1103 before it can preempt a lower priority task.
1105 The preemptoff tracer traces the places that disable preemption.
1106 Like the irqsoff tracer, it records the maximum latency for
1107 which preemption was disabled. The control of preemptoff tracer
1108 is much like the irqsoff tracer.
1110 # echo 0 > options/function-trace
1111 # echo preemptoff > current_tracer
1112 # echo 1 > tracing_on
1113 # echo 0 > tracing_max_latency
1116 # echo 0 > tracing_on
1118 # tracer: preemptoff
1120 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1121 # --------------------------------------------------------------------
1122 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1124 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1126 # => started at: do_IRQ
1127 # => ended at: do_IRQ
1131 # / _-----=> irqs-off
1132 # | / _----=> need-resched
1133 # || / _---=> hardirq/softirq
1134 # ||| / _--=> preempt-depth
1136 # cmd pid ||||| time | caller
1138 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1139 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1140 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1141 sshd-1991 1d..1 52us : <stack trace>
1142 => sub_preempt_count
1148 This has some more changes. Preemption was disabled when an
1149 interrupt came in (notice the 'h'), and was enabled on exit.
1150 But we also see that interrupts have been disabled when entering
1151 the preempt off section and leaving it (the 'd'). We do not know if
1152 interrupts were enabled in the mean time or shortly after this
1155 # tracer: preemptoff
1157 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1158 # --------------------------------------------------------------------
1159 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1161 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1163 # => started at: wake_up_new_task
1164 # => ended at: task_rq_unlock
1168 # / _-----=> irqs-off
1169 # | / _----=> need-resched
1170 # || / _---=> hardirq/softirq
1171 # ||| / _--=> preempt-depth
1173 # cmd pid ||||| time | caller
1175 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1176 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1177 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1178 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1179 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1181 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1182 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1183 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1184 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1185 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1186 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1187 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1188 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1190 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1191 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1192 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1193 bash-1994 1d..2 36us : do_softirq <-irq_exit
1194 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1195 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1196 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1197 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1198 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1199 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1201 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1202 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1203 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1204 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1205 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1206 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1207 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1208 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1209 bash-1994 1.N.1 104us : <stack trace>
1210 => sub_preempt_count
1211 => _raw_spin_unlock_irqrestore
1219 The above is an example of the preemptoff trace with
1220 function-trace set. Here we see that interrupts were not disabled
1221 the entire time. The irq_enter code lets us know that we entered
1222 an interrupt 'h'. Before that, the functions being traced still
1223 show that it is not in an interrupt, but we can see from the
1224 functions themselves that this is not the case.
1229 Knowing the locations that have interrupts disabled or
1230 preemption disabled for the longest times is helpful. But
1231 sometimes we would like to know when either preemption and/or
1232 interrupts are disabled.
1234 Consider the following code:
1236 local_irq_disable();
1237 call_function_with_irqs_off();
1239 call_function_with_irqs_and_preemption_off();
1241 call_function_with_preemption_off();
1244 The irqsoff tracer will record the total length of
1245 call_function_with_irqs_off() and
1246 call_function_with_irqs_and_preemption_off().
1248 The preemptoff tracer will record the total length of
1249 call_function_with_irqs_and_preemption_off() and
1250 call_function_with_preemption_off().
1252 But neither will trace the time that interrupts and/or
1253 preemption is disabled. This total time is the time that we can
1254 not schedule. To record this time, use the preemptirqsoff
1257 Again, using this trace is much like the irqsoff and preemptoff
1260 # echo 0 > options/function-trace
1261 # echo preemptirqsoff > current_tracer
1262 # echo 1 > tracing_on
1263 # echo 0 > tracing_max_latency
1266 # echo 0 > tracing_on
1268 # tracer: preemptirqsoff
1270 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1271 # --------------------------------------------------------------------
1272 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1274 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1276 # => started at: ata_scsi_queuecmd
1277 # => ended at: ata_scsi_queuecmd
1281 # / _-----=> irqs-off
1282 # | / _----=> need-resched
1283 # || / _---=> hardirq/softirq
1284 # ||| / _--=> preempt-depth
1286 # cmd pid ||||| time | caller
1288 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1289 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1290 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1291 ls-2230 3...1 111us : <stack trace>
1292 => sub_preempt_count
1293 => _raw_spin_unlock_irqrestore
1294 => ata_scsi_queuecmd
1295 => scsi_dispatch_cmd
1297 => __blk_run_queue_uncond
1300 => generic_make_request
1305 => htree_dirblock_to_tree
1306 => ext3_htree_fill_tree
1310 => system_call_fastpath
1313 The trace_hardirqs_off_thunk is called from assembly on x86 when
1314 interrupts are disabled in the assembly code. Without the
1315 function tracing, we do not know if interrupts were enabled
1316 within the preemption points. We do see that it started with
1319 Here is a trace with function-trace set:
1321 # tracer: preemptirqsoff
1323 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1324 # --------------------------------------------------------------------
1325 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1327 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1329 # => started at: schedule
1330 # => ended at: mutex_unlock
1334 # / _-----=> irqs-off
1335 # | / _----=> need-resched
1336 # || / _---=> hardirq/softirq
1337 # ||| / _--=> preempt-depth
1339 # cmd pid ||||| time | caller
1341 kworker/-59 3...1 0us : __schedule <-schedule
1342 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1343 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1344 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1345 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1346 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1347 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1348 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1349 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1350 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1351 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1352 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1353 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1354 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1355 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1356 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1357 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1358 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1359 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1360 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1361 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1362 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1363 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1364 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1365 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1366 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1367 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1368 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1369 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1370 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1371 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1372 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1374 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1375 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1376 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1377 ls-2269 3d..3 21us : do_softirq <-irq_exit
1378 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1379 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1380 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1381 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1382 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1383 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1384 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1386 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1387 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1388 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1389 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1390 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1391 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1393 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1394 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1395 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1396 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1397 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1398 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1399 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1400 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1401 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1402 ls-2269 3d... 186us : <stack trace>
1403 => __mutex_unlock_slowpath
1410 => system_call_fastpath
1412 This is an interesting trace. It started with kworker running and
1413 scheduling out and ls taking over. But as soon as ls released the
1414 rq lock and enabled interrupts (but not preemption) an interrupt
1415 triggered. When the interrupt finished, it started running softirqs.
1416 But while the softirq was running, another interrupt triggered.
1417 When an interrupt is running inside a softirq, the annotation is 'H'.
1423 One common case that people are interested in tracing is the
1424 time it takes for a task that is woken to actually wake up.
1425 Now for non Real-Time tasks, this can be arbitrary. But tracing
1426 it none the less can be interesting.
1428 Without function tracing:
1430 # echo 0 > options/function-trace
1431 # echo wakeup > current_tracer
1432 # echo 1 > tracing_on
1433 # echo 0 > tracing_max_latency
1435 # echo 0 > tracing_on
1439 # wakeup latency trace v1.1.5 on 3.8.0-test+
1440 # --------------------------------------------------------------------
1441 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1443 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1447 # / _-----=> irqs-off
1448 # | / _----=> need-resched
1449 # || / _---=> hardirq/softirq
1450 # ||| / _--=> preempt-depth
1452 # cmd pid ||||| time | caller
1454 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1455 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1456 <idle>-0 3d..3 15us : __schedule <-schedule
1457 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1459 The tracer only traces the highest priority task in the system
1460 to avoid tracing the normal circumstances. Here we see that
1461 the kworker with a nice priority of -20 (not very nice), took
1462 just 15 microseconds from the time it woke up, to the time it
1465 Non Real-Time tasks are not that interesting. A more interesting
1466 trace is to concentrate only on Real-Time tasks.
1471 In a Real-Time environment it is very important to know the
1472 wakeup time it takes for the highest priority task that is woken
1473 up to the time that it executes. This is also known as "schedule
1474 latency". I stress the point that this is about RT tasks. It is
1475 also important to know the scheduling latency of non-RT tasks,
1476 but the average schedule latency is better for non-RT tasks.
1477 Tools like LatencyTop are more appropriate for such
1480 Real-Time environments are interested in the worst case latency.
1481 That is the longest latency it takes for something to happen,
1482 and not the average. We can have a very fast scheduler that may
1483 only have a large latency once in a while, but that would not
1484 work well with Real-Time tasks. The wakeup_rt tracer was designed
1485 to record the worst case wakeups of RT tasks. Non-RT tasks are
1486 not recorded because the tracer only records one worst case and
1487 tracing non-RT tasks that are unpredictable will overwrite the
1488 worst case latency of RT tasks (just run the normal wakeup
1489 tracer for a while to see that effect).
1491 Since this tracer only deals with RT tasks, we will run this
1492 slightly differently than we did with the previous tracers.
1493 Instead of performing an 'ls', we will run 'sleep 1' under
1494 'chrt' which changes the priority of the task.
1496 # echo 0 > options/function-trace
1497 # echo wakeup_rt > current_tracer
1498 # echo 1 > tracing_on
1499 # echo 0 > tracing_max_latency
1501 # echo 0 > tracing_on
1507 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1508 # --------------------------------------------------------------------
1509 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1511 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1515 # / _-----=> irqs-off
1516 # | / _----=> need-resched
1517 # || / _---=> hardirq/softirq
1518 # ||| / _--=> preempt-depth
1520 # cmd pid ||||| time | caller
1522 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1523 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1524 <idle>-0 3d..3 5us : __schedule <-schedule
1525 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1528 Running this on an idle system, we see that it only took 5 microseconds
1529 to perform the task switch. Note, since the trace point in the schedule
1530 is before the actual "switch", we stop the tracing when the recorded task
1531 is about to schedule in. This may change if we add a new marker at the
1532 end of the scheduler.
1534 Notice that the recorded task is 'sleep' with the PID of 2389
1535 and it has an rt_prio of 5. This priority is user-space priority
1536 and not the internal kernel priority. The policy is 1 for
1537 SCHED_FIFO and 2 for SCHED_RR.
1539 Note, that the trace data shows the internal priority (99 - rtprio).
1541 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1543 The 0:120:R means idle was running with a nice priority of 0 (120 - 20)
1544 and in the running state 'R'. The sleep task was scheduled in with
1545 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1546 and it too is in the running state.
1548 Doing the same with chrt -r 5 and function-trace set.
1550 echo 1 > options/function-trace
1554 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1555 # --------------------------------------------------------------------
1556 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1558 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
1562 # / _-----=> irqs-off
1563 # | / _----=> need-resched
1564 # || / _---=> hardirq/softirq
1565 # ||| / _--=> preempt-depth
1567 # cmd pid ||||| time | caller
1569 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
1570 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1571 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
1572 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
1573 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
1574 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
1575 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
1576 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
1577 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
1578 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1579 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
1580 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
1581 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
1582 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
1583 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
1584 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
1585 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
1586 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
1587 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
1588 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
1589 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
1590 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
1591 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
1592 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
1593 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
1594 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
1595 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
1596 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
1597 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
1598 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
1599 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit
1600 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz
1601 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
1602 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz
1603 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update
1604 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz
1605 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
1606 <idle>-0 3dN.1 15us : calc_load_exit_idle <-tick_nohz_idle_exit
1607 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
1608 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
1609 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
1610 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
1611 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1612 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
1613 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
1614 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
1615 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
1616 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
1617 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
1618 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
1619 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
1620 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1621 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
1622 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1623 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1624 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
1625 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
1626 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
1627 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1628 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
1629 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
1630 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
1631 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
1632 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
1633 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
1634 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
1635 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
1636 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1637 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
1638 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
1639 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
1640 <idle>-0 3.N.. 25us : schedule <-cpu_idle
1641 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
1642 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
1643 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
1644 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
1645 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
1646 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
1647 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
1648 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
1649 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
1650 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
1651 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
1652 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
1653 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
1655 This isn't that big of a trace, even with function tracing enabled,
1656 so I included the entire trace.
1658 The interrupt went off while when the system was idle. Somewhere
1659 before task_woken_rt() was called, the NEED_RESCHED flag was set,
1660 this is indicated by the first occurrence of the 'N' flag.
1662 Latency tracing and events
1663 --------------------------
1664 As function tracing can induce a much larger latency, but without
1665 seeing what happens within the latency it is hard to know what
1666 caused it. There is a middle ground, and that is with enabling
1669 # echo 0 > options/function-trace
1670 # echo wakeup_rt > current_tracer
1671 # echo 1 > events/enable
1672 # echo 1 > tracing_on
1673 # echo 0 > tracing_max_latency
1675 # echo 0 > tracing_on
1679 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1680 # --------------------------------------------------------------------
1681 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1683 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
1687 # / _-----=> irqs-off
1688 # | / _----=> need-resched
1689 # || / _---=> hardirq/softirq
1690 # ||| / _--=> preempt-depth
1692 # cmd pid ||||| time | caller
1694 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
1695 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1696 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
1697 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
1698 <idle>-0 2.N.2 2us : power_end: cpu_id=2
1699 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
1700 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
1701 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
1702 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
1703 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
1704 <idle>-0 2d..3 6us : __schedule <-schedule
1705 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
1711 This tracer is the function tracer. Enabling the function tracer
1712 can be done from the debug file system. Make sure the
1713 ftrace_enabled is set; otherwise this tracer is a nop.
1714 See the "ftrace_enabled" section below.
1716 # sysctl kernel.ftrace_enabled=1
1717 # echo function > current_tracer
1718 # echo 1 > tracing_on
1720 # echo 0 > tracing_on
1724 # entries-in-buffer/entries-written: 24799/24799 #P:4
1727 # / _----=> need-resched
1728 # | / _---=> hardirq/softirq
1729 # || / _--=> preempt-depth
1731 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
1733 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
1734 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
1735 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
1736 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
1737 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
1738 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
1739 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
1740 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
1744 Note: function tracer uses ring buffers to store the above
1745 entries. The newest data may overwrite the oldest data.
1746 Sometimes using echo to stop the trace is not sufficient because
1747 the tracing could have overwritten the data that you wanted to
1748 record. For this reason, it is sometimes better to disable
1749 tracing directly from a program. This allows you to stop the
1750 tracing at the point that you hit the part that you are
1751 interested in. To disable the tracing directly from a C program,
1752 something like following code snippet can be used:
1756 int main(int argc, char *argv[]) {
1758 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
1760 if (condition_hit()) {
1761 write(trace_fd, "0", 1);
1767 Single thread tracing
1768 ---------------------
1770 By writing into set_ftrace_pid you can trace a
1771 single thread. For example:
1773 # cat set_ftrace_pid
1775 # echo 3111 > set_ftrace_pid
1776 # cat set_ftrace_pid
1778 # echo function > current_tracer
1782 # TASK-PID CPU# TIMESTAMP FUNCTION
1784 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
1785 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
1786 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
1787 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
1788 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
1789 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
1790 # echo > set_ftrace_pid
1794 # TASK-PID CPU# TIMESTAMP FUNCTION
1796 ##### CPU 3 buffer started ####
1797 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
1798 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
1799 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
1800 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
1801 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
1803 If you want to trace a function when executing, you could use
1804 something like this simple program:
1808 #include <sys/types.h>
1809 #include <sys/stat.h>
1815 #define STR(x) _STR(x)
1816 #define MAX_PATH 256
1818 const char *find_debugfs(void)
1820 static char debugfs[MAX_PATH+1];
1821 static int debugfs_found;
1828 if ((fp = fopen("/proc/mounts","r")) == NULL) {
1829 perror("/proc/mounts");
1833 while (fscanf(fp, "%*s %"
1835 "s %99s %*s %*d %*d\n",
1836 debugfs, type) == 2) {
1837 if (strcmp(type, "debugfs") == 0)
1842 if (strcmp(type, "debugfs") != 0) {
1843 fprintf(stderr, "debugfs not mounted");
1847 strcat(debugfs, "/tracing/");
1853 const char *tracing_file(const char *file_name)
1855 static char trace_file[MAX_PATH+1];
1856 snprintf(trace_file, MAX_PATH, "%s/%s", find_debugfs(), file_name);
1860 int main (int argc, char **argv)
1870 ffd = open(tracing_file("current_tracer"), O_WRONLY);
1873 write(ffd, "nop", 3);
1875 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
1876 s = sprintf(line, "%d\n", getpid());
1879 write(ffd, "function", 8);
1884 execvp(argv[1], argv+1);
1890 Or this simple script!
1895 debugfs=`sed -ne 's/^debugfs \(.*\) debugfs.*/\1/p' /proc/mounts`
1896 echo nop > $debugfs/tracing/current_tracer
1897 echo 0 > $debugfs/tracing/tracing_on
1898 echo $$ > $debugfs/tracing/set_ftrace_pid
1899 echo function > $debugfs/tracing/current_tracer
1900 echo 1 > $debugfs/tracing/tracing_on
1905 function graph tracer
1906 ---------------------------
1908 This tracer is similar to the function tracer except that it
1909 probes a function on its entry and its exit. This is done by
1910 using a dynamically allocated stack of return addresses in each
1911 task_struct. On function entry the tracer overwrites the return
1912 address of each function traced to set a custom probe. Thus the
1913 original return address is stored on the stack of return address
1916 Probing on both ends of a function leads to special features
1919 - measure of a function's time execution
1920 - having a reliable call stack to draw function calls graph
1922 This tracer is useful in several situations:
1924 - you want to find the reason of a strange kernel behavior and
1925 need to see what happens in detail on any areas (or specific
1928 - you are experiencing weird latencies but it's difficult to
1931 - you want to find quickly which path is taken by a specific
1934 - you just want to peek inside a working kernel and want to see
1937 # tracer: function_graph
1939 # CPU DURATION FUNCTION CALLS
1943 0) | do_sys_open() {
1945 0) | kmem_cache_alloc() {
1946 0) 1.382 us | __might_sleep();
1948 0) | strncpy_from_user() {
1949 0) | might_fault() {
1950 0) 1.389 us | __might_sleep();
1955 0) 0.668 us | _spin_lock();
1956 0) 0.570 us | expand_files();
1957 0) 0.586 us | _spin_unlock();
1960 There are several columns that can be dynamically
1961 enabled/disabled. You can use every combination of options you
1962 want, depending on your needs.
1964 - The cpu number on which the function executed is default
1965 enabled. It is sometimes better to only trace one cpu (see
1966 tracing_cpu_mask file) or you might sometimes see unordered
1967 function calls while cpu tracing switch.
1969 hide: echo nofuncgraph-cpu > trace_options
1970 show: echo funcgraph-cpu > trace_options
1972 - The duration (function's time of execution) is displayed on
1973 the closing bracket line of a function or on the same line
1974 than the current function in case of a leaf one. It is default
1977 hide: echo nofuncgraph-duration > trace_options
1978 show: echo funcgraph-duration > trace_options
1980 - The overhead field precedes the duration field in case of
1981 reached duration thresholds.
1983 hide: echo nofuncgraph-overhead > trace_options
1984 show: echo funcgraph-overhead > trace_options
1985 depends on: funcgraph-duration
1989 3) # 1837.709 us | } /* __switch_to */
1990 3) | finish_task_switch() {
1991 3) 0.313 us | _raw_spin_unlock_irq();
1993 3) # 1889.063 us | } /* __schedule */
1994 3) ! 140.417 us | } /* __schedule */
1995 3) # 2034.948 us | } /* schedule */
1996 3) * 33998.59 us | } /* schedule_preempt_disabled */
2000 1) 0.260 us | msecs_to_jiffies();
2001 1) 0.313 us | __rcu_read_unlock();
2004 1) 0.313 us | rcu_bh_qs();
2005 1) 0.313 us | __local_bh_enable();
2007 1) 0.365 us | idle_cpu();
2008 1) | rcu_irq_exit() {
2009 1) 0.417 us | rcu_eqs_enter_common.isra.47();
2013 1) @ 119760.2 us | }
2019 2) 0.417 us | scheduler_ipi();
2027 + means that the function exceeded 10 usecs.
2028 ! means that the function exceeded 100 usecs.
2029 # means that the function exceeded 1000 usecs.
2030 * means that the function exceeded 10 msecs.
2031 @ means that the function exceeded 100 msecs.
2032 $ means that the function exceeded 1 sec.
2035 - The task/pid field displays the thread cmdline and pid which
2036 executed the function. It is default disabled.
2038 hide: echo nofuncgraph-proc > trace_options
2039 show: echo funcgraph-proc > trace_options
2043 # tracer: function_graph
2045 # CPU TASK/PID DURATION FUNCTION CALLS
2047 0) sh-4802 | | d_free() {
2048 0) sh-4802 | | call_rcu() {
2049 0) sh-4802 | | __call_rcu() {
2050 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2051 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2052 0) sh-4802 | 2.899 us | }
2053 0) sh-4802 | 4.040 us | }
2054 0) sh-4802 | 5.151 us | }
2055 0) sh-4802 | + 49.370 us | }
2058 - The absolute time field is an absolute timestamp given by the
2059 system clock since it started. A snapshot of this time is
2060 given on each entry/exit of functions
2062 hide: echo nofuncgraph-abstime > trace_options
2063 show: echo funcgraph-abstime > trace_options
2068 # TIME CPU DURATION FUNCTION CALLS
2070 360.774522 | 1) 0.541 us | }
2071 360.774522 | 1) 4.663 us | }
2072 360.774523 | 1) 0.541 us | __wake_up_bit();
2073 360.774524 | 1) 6.796 us | }
2074 360.774524 | 1) 7.952 us | }
2075 360.774525 | 1) 9.063 us | }
2076 360.774525 | 1) 0.615 us | journal_mark_dirty();
2077 360.774527 | 1) 0.578 us | __brelse();
2078 360.774528 | 1) | reiserfs_prepare_for_journal() {
2079 360.774528 | 1) | unlock_buffer() {
2080 360.774529 | 1) | wake_up_bit() {
2081 360.774529 | 1) | bit_waitqueue() {
2082 360.774530 | 1) 0.594 us | __phys_addr();
2085 The function name is always displayed after the closing bracket
2086 for a function if the start of that function is not in the
2089 Display of the function name after the closing bracket may be
2090 enabled for functions whose start is in the trace buffer,
2091 allowing easier searching with grep for function durations.
2092 It is default disabled.
2094 hide: echo nofuncgraph-tail > trace_options
2095 show: echo funcgraph-tail > trace_options
2097 Example with nofuncgraph-tail (default):
2099 0) | kmem_cache_free() {
2100 0) 0.518 us | __phys_addr();
2104 Example with funcgraph-tail:
2106 0) | kmem_cache_free() {
2107 0) 0.518 us | __phys_addr();
2108 0) 1.757 us | } /* kmem_cache_free() */
2109 0) 2.861 us | } /* putname() */
2111 You can put some comments on specific functions by using
2112 trace_printk() For example, if you want to put a comment inside
2113 the __might_sleep() function, you just have to include
2114 <linux/ftrace.h> and call trace_printk() inside __might_sleep()
2116 trace_printk("I'm a comment!\n")
2120 1) | __might_sleep() {
2121 1) | /* I'm a comment! */
2125 You might find other useful features for this tracer in the
2126 following "dynamic ftrace" section such as tracing only specific
2132 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2133 virtually no overhead when function tracing is disabled. The way
2134 this works is the mcount function call (placed at the start of
2135 every kernel function, produced by the -pg switch in gcc),
2136 starts of pointing to a simple return. (Enabling FTRACE will
2137 include the -pg switch in the compiling of the kernel.)
2139 At compile time every C file object is run through the
2140 recordmcount program (located in the scripts directory). This
2141 program will parse the ELF headers in the C object to find all
2142 the locations in the .text section that call mcount. (Note, only
2143 white listed .text sections are processed, since processing other
2144 sections like .init.text may cause races due to those sections
2145 being freed unexpectedly).
2147 A new section called "__mcount_loc" is created that holds
2148 references to all the mcount call sites in the .text section.
2149 The recordmcount program re-links this section back into the
2150 original object. The final linking stage of the kernel will add all these
2151 references into a single table.
2153 On boot up, before SMP is initialized, the dynamic ftrace code
2154 scans this table and updates all the locations into nops. It
2155 also records the locations, which are added to the
2156 available_filter_functions list. Modules are processed as they
2157 are loaded and before they are executed. When a module is
2158 unloaded, it also removes its functions from the ftrace function
2159 list. This is automatic in the module unload code, and the
2160 module author does not need to worry about it.
2162 When tracing is enabled, the process of modifying the function
2163 tracepoints is dependent on architecture. The old method is to use
2164 kstop_machine to prevent races with the CPUs executing code being
2165 modified (which can cause the CPU to do undesirable things, especially
2166 if the modified code crosses cache (or page) boundaries), and the nops are
2167 patched back to calls. But this time, they do not call mcount
2168 (which is just a function stub). They now call into the ftrace
2171 The new method of modifying the function tracepoints is to place
2172 a breakpoint at the location to be modified, sync all CPUs, modify
2173 the rest of the instruction not covered by the breakpoint. Sync
2174 all CPUs again, and then remove the breakpoint with the finished
2175 version to the ftrace call site.
2177 Some archs do not even need to monkey around with the synchronization,
2178 and can just slap the new code on top of the old without any
2179 problems with other CPUs executing it at the same time.
2181 One special side-effect to the recording of the functions being
2182 traced is that we can now selectively choose which functions we
2183 wish to trace and which ones we want the mcount calls to remain
2186 Two files are used, one for enabling and one for disabling the
2187 tracing of specified functions. They are:
2195 A list of available functions that you can add to these files is
2198 available_filter_functions
2200 # cat available_filter_functions
2209 If I am only interested in sys_nanosleep and hrtimer_interrupt:
2211 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2212 # echo function > current_tracer
2213 # echo 1 > tracing_on
2215 # echo 0 > tracing_on
2219 # entries-in-buffer/entries-written: 5/5 #P:4
2222 # / _----=> need-resched
2223 # | / _---=> hardirq/softirq
2224 # || / _--=> preempt-depth
2226 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2228 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2229 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2230 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2231 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2232 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2234 To see which functions are being traced, you can cat the file:
2236 # cat set_ftrace_filter
2241 Perhaps this is not enough. The filters also allow simple wild
2242 cards. Only the following are currently available
2244 <match>* - will match functions that begin with <match>
2245 *<match> - will match functions that end with <match>
2246 *<match>* - will match functions that have <match> in it
2248 These are the only wild cards which are supported.
2250 <match>*<match> will not work.
2252 Note: It is better to use quotes to enclose the wild cards,
2253 otherwise the shell may expand the parameters into names
2254 of files in the local directory.
2256 # echo 'hrtimer_*' > set_ftrace_filter
2262 # entries-in-buffer/entries-written: 897/897 #P:4
2265 # / _----=> need-resched
2266 # | / _---=> hardirq/softirq
2267 # || / _--=> preempt-depth
2269 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2271 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2272 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2273 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2274 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2275 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2276 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2277 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2278 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2280 Notice that we lost the sys_nanosleep.
2282 # cat set_ftrace_filter
2287 hrtimer_try_to_cancel
2291 hrtimer_force_reprogram
2292 hrtimer_get_next_event
2296 hrtimer_get_remaining
2298 hrtimer_init_sleeper
2301 This is because the '>' and '>>' act just like they do in bash.
2302 To rewrite the filters, use '>'
2303 To append to the filters, use '>>'
2305 To clear out a filter so that all functions will be recorded
2308 # echo > set_ftrace_filter
2309 # cat set_ftrace_filter
2312 Again, now we want to append.
2314 # echo sys_nanosleep > set_ftrace_filter
2315 # cat set_ftrace_filter
2317 # echo 'hrtimer_*' >> set_ftrace_filter
2318 # cat set_ftrace_filter
2323 hrtimer_try_to_cancel
2327 hrtimer_force_reprogram
2328 hrtimer_get_next_event
2333 hrtimer_get_remaining
2335 hrtimer_init_sleeper
2338 The set_ftrace_notrace prevents those functions from being
2341 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2347 # entries-in-buffer/entries-written: 39608/39608 #P:4
2350 # / _----=> need-resched
2351 # | / _---=> hardirq/softirq
2352 # || / _--=> preempt-depth
2354 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2356 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2357 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2358 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2359 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2360 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2361 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2362 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2363 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2364 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2365 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2366 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2367 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2369 We can see that there's no more lock or preempt tracing.
2372 Dynamic ftrace with the function graph tracer
2373 ---------------------------------------------
2375 Although what has been explained above concerns both the
2376 function tracer and the function-graph-tracer, there are some
2377 special features only available in the function-graph tracer.
2379 If you want to trace only one function and all of its children,
2380 you just have to echo its name into set_graph_function:
2382 echo __do_fault > set_graph_function
2384 will produce the following "expanded" trace of the __do_fault()
2388 0) | filemap_fault() {
2389 0) | find_lock_page() {
2390 0) 0.804 us | find_get_page();
2391 0) | __might_sleep() {
2395 0) 0.653 us | _spin_lock();
2396 0) 0.578 us | page_add_file_rmap();
2397 0) 0.525 us | native_set_pte_at();
2398 0) 0.585 us | _spin_unlock();
2399 0) | unlock_page() {
2400 0) 0.541 us | page_waitqueue();
2401 0) 0.639 us | __wake_up_bit();
2405 0) | filemap_fault() {
2406 0) | find_lock_page() {
2407 0) 0.698 us | find_get_page();
2408 0) | __might_sleep() {
2412 0) 0.631 us | _spin_lock();
2413 0) 0.571 us | page_add_file_rmap();
2414 0) 0.526 us | native_set_pte_at();
2415 0) 0.586 us | _spin_unlock();
2416 0) | unlock_page() {
2417 0) 0.533 us | page_waitqueue();
2418 0) 0.638 us | __wake_up_bit();
2422 You can also expand several functions at once:
2424 echo sys_open > set_graph_function
2425 echo sys_close >> set_graph_function
2427 Now if you want to go back to trace all functions you can clear
2428 this special filter via:
2430 echo > set_graph_function
2436 Note, the proc sysctl ftrace_enable is a big on/off switch for the
2437 function tracer. By default it is enabled (when function tracing is
2438 enabled in the kernel). If it is disabled, all function tracing is
2439 disabled. This includes not only the function tracers for ftrace, but
2440 also for any other uses (perf, kprobes, stack tracing, profiling, etc).
2442 Please disable this with care.
2444 This can be disable (and enabled) with:
2446 sysctl kernel.ftrace_enabled=0
2447 sysctl kernel.ftrace_enabled=1
2451 echo 0 > /proc/sys/kernel/ftrace_enabled
2452 echo 1 > /proc/sys/kernel/ftrace_enabled
2458 A few commands are supported by the set_ftrace_filter interface.
2459 Trace commands have the following format:
2461 <function>:<command>:<parameter>
2463 The following commands are supported:
2466 This command enables function filtering per module. The
2467 parameter defines the module. For example, if only the write*
2468 functions in the ext3 module are desired, run:
2470 echo 'write*:mod:ext3' > set_ftrace_filter
2472 This command interacts with the filter in the same way as
2473 filtering based on function names. Thus, adding more functions
2474 in a different module is accomplished by appending (>>) to the
2475 filter file. Remove specific module functions by prepending
2478 echo '!writeback*:mod:ext3' >> set_ftrace_filter
2480 Mod command supports module globbing. Disable tracing for all
2481 functions except a specific module:
2483 echo '!*:mod:!ext3' >> set_ftrace_filter
2485 Disable tracing for all modules, but still trace kernel:
2487 echo '!*:mod:*' >> set_ftrace_filter
2489 Enable filter only for kernel:
2491 echo '*write*:mod:!*' >> set_ftrace_filter
2493 Enable filter for module globbing:
2495 echo '*write*:mod:*snd*' >> set_ftrace_filter
2498 These commands turn tracing on and off when the specified
2499 functions are hit. The parameter determines how many times the
2500 tracing system is turned on and off. If unspecified, there is
2501 no limit. For example, to disable tracing when a schedule bug
2502 is hit the first 5 times, run:
2504 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
2506 To always disable tracing when __schedule_bug is hit:
2508 echo '__schedule_bug:traceoff' > set_ftrace_filter
2510 These commands are cumulative whether or not they are appended
2511 to set_ftrace_filter. To remove a command, prepend it by '!'
2512 and drop the parameter:
2514 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
2516 The above removes the traceoff command for __schedule_bug
2517 that have a counter. To remove commands without counters:
2519 echo '!__schedule_bug:traceoff' > set_ftrace_filter
2522 Will cause a snapshot to be triggered when the function is hit.
2524 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
2526 To only snapshot once:
2528 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
2530 To remove the above commands:
2532 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
2533 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
2535 - enable_event/disable_event
2536 These commands can enable or disable a trace event. Note, because
2537 function tracing callbacks are very sensitive, when these commands
2538 are registered, the trace point is activated, but disabled in
2539 a "soft" mode. That is, the tracepoint will be called, but
2540 just will not be traced. The event tracepoint stays in this mode
2541 as long as there's a command that triggers it.
2543 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
2548 <function>:enable_event:<system>:<event>[:count]
2549 <function>:disable_event:<system>:<event>[:count]
2551 To remove the events commands:
2554 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
2556 echo '!schedule:disable_event:sched:sched_switch' > \
2560 When the function is hit, it will dump the contents of the ftrace
2561 ring buffer to the console. This is useful if you need to debug
2562 something, and want to dump the trace when a certain function
2563 is hit. Perhaps its a function that is called before a tripple
2564 fault happens and does not allow you to get a regular dump.
2567 When the function is hit, it will dump the contents of the ftrace
2568 ring buffer for the current CPU to the console. Unlike the "dump"
2569 command, it only prints out the contents of the ring buffer for the
2570 CPU that executed the function that triggered the dump.
2575 The trace_pipe outputs the same content as the trace file, but
2576 the effect on the tracing is different. Every read from
2577 trace_pipe is consumed. This means that subsequent reads will be
2578 different. The trace is live.
2580 # echo function > current_tracer
2581 # cat trace_pipe > /tmp/trace.out &
2583 # echo 1 > tracing_on
2585 # echo 0 > tracing_on
2589 # entries-in-buffer/entries-written: 0/0 #P:4
2592 # / _----=> need-resched
2593 # | / _---=> hardirq/softirq
2594 # || / _--=> preempt-depth
2596 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2600 # cat /tmp/trace.out
2601 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
2602 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
2603 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
2604 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
2605 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
2606 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
2607 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
2608 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
2609 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
2612 Note, reading the trace_pipe file will block until more input is
2618 Having too much or not enough data can be troublesome in
2619 diagnosing an issue in the kernel. The file buffer_size_kb is
2620 used to modify the size of the internal trace buffers. The
2621 number listed is the number of entries that can be recorded per
2622 CPU. To know the full size, multiply the number of possible CPUs
2623 with the number of entries.
2625 # cat buffer_size_kb
2626 1408 (units kilobytes)
2628 Or simply read buffer_total_size_kb
2630 # cat buffer_total_size_kb
2633 To modify the buffer, simple echo in a number (in 1024 byte segments).
2635 # echo 10000 > buffer_size_kb
2636 # cat buffer_size_kb
2637 10000 (units kilobytes)
2639 It will try to allocate as much as possible. If you allocate too
2640 much, it can cause Out-Of-Memory to trigger.
2642 # echo 1000000000000 > buffer_size_kb
2643 -bash: echo: write error: Cannot allocate memory
2644 # cat buffer_size_kb
2647 The per_cpu buffers can be changed individually as well:
2649 # echo 10000 > per_cpu/cpu0/buffer_size_kb
2650 # echo 100 > per_cpu/cpu1/buffer_size_kb
2652 When the per_cpu buffers are not the same, the buffer_size_kb
2653 at the top level will just show an X
2655 # cat buffer_size_kb
2658 This is where the buffer_total_size_kb is useful:
2660 # cat buffer_total_size_kb
2663 Writing to the top level buffer_size_kb will reset all the buffers
2664 to be the same again.
2668 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
2669 available to all non latency tracers. (Latency tracers which
2670 record max latency, such as "irqsoff" or "wakeup", can't use
2671 this feature, since those are already using the snapshot
2672 mechanism internally.)
2674 Snapshot preserves a current trace buffer at a particular point
2675 in time without stopping tracing. Ftrace swaps the current
2676 buffer with a spare buffer, and tracing continues in the new
2677 current (=previous spare) buffer.
2679 The following debugfs files in "tracing" are related to this
2684 This is used to take a snapshot and to read the output
2685 of the snapshot. Echo 1 into this file to allocate a
2686 spare buffer and to take a snapshot (swap), then read
2687 the snapshot from this file in the same format as
2688 "trace" (described above in the section "The File
2689 System"). Both reads snapshot and tracing are executable
2690 in parallel. When the spare buffer is allocated, echoing
2691 0 frees it, and echoing else (positive) values clear the
2693 More details are shown in the table below.
2695 status\input | 0 | 1 | else |
2696 --------------+------------+------------+------------+
2697 not allocated |(do nothing)| alloc+swap |(do nothing)|
2698 --------------+------------+------------+------------+
2699 allocated | free | swap | clear |
2700 --------------+------------+------------+------------+
2702 Here is an example of using the snapshot feature.
2704 # echo 1 > events/sched/enable
2709 # entries-in-buffer/entries-written: 71/71 #P:8
2712 # / _----=> need-resched
2713 # | / _---=> hardirq/softirq
2714 # || / _--=> preempt-depth
2716 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2718 <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
2719 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
2721 <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
2726 # entries-in-buffer/entries-written: 77/77 #P:8
2729 # / _----=> need-resched
2730 # | / _---=> hardirq/softirq
2731 # || / _--=> preempt-depth
2733 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2735 <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
2736 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
2740 If you try to use this snapshot feature when current tracer is
2741 one of the latency tracers, you will get the following results.
2743 # echo wakeup > current_tracer
2745 bash: echo: write error: Device or resource busy
2747 cat: snapshot: Device or resource busy
2752 In the debugfs tracing directory is a directory called "instances".
2753 This directory can have new directories created inside of it using
2754 mkdir, and removing directories with rmdir. The directory created
2755 with mkdir in this directory will already contain files and other
2756 directories after it is created.
2758 # mkdir instances/foo
2760 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
2761 set_event snapshot trace trace_clock trace_marker trace_options
2762 trace_pipe tracing_on
2764 As you can see, the new directory looks similar to the tracing directory
2765 itself. In fact, it is very similar, except that the buffer and
2766 events are agnostic from the main director, or from any other
2767 instances that are created.
2769 The files in the new directory work just like the files with the
2770 same name in the tracing directory except the buffer that is used
2771 is a separate and new buffer. The files affect that buffer but do not
2772 affect the main buffer with the exception of trace_options. Currently,
2773 the trace_options affect all instances and the top level buffer
2774 the same, but this may change in future releases. That is, options
2775 may become specific to the instance they reside in.
2777 Notice that none of the function tracer files are there, nor is
2778 current_tracer and available_tracers. This is because the buffers
2779 can currently only have events enabled for them.
2781 # mkdir instances/foo
2782 # mkdir instances/bar
2783 # mkdir instances/zoot
2784 # echo 100000 > buffer_size_kb
2785 # echo 1000 > instances/foo/buffer_size_kb
2786 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
2787 # echo function > current_trace
2788 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
2789 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
2790 # echo 1 > instances/foo/events/sched/sched_switch/enable
2791 # echo 1 > instances/bar/events/irq/enable
2792 # echo 1 > instances/zoot/events/syscalls/enable
2794 CPU:2 [LOST 11745 EVENTS]
2795 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
2796 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
2797 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
2798 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
2799 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
2800 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
2801 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
2802 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
2803 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
2804 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
2805 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
2808 # cat instances/foo/trace_pipe
2809 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
2810 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
2811 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
2812 <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
2813 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
2814 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
2815 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
2816 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
2817 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
2818 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
2821 # cat instances/bar/trace_pipe
2822 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
2823 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
2824 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
2825 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
2826 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
2827 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
2828 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
2829 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
2830 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
2831 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
2832 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
2833 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
2836 # cat instances/zoot/trace
2839 # entries-in-buffer/entries-written: 18996/18996 #P:4
2842 # / _----=> need-resched
2843 # | / _---=> hardirq/softirq
2844 # || / _--=> preempt-depth
2846 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2848 bash-1998 [000] d... 140.733501: sys_write -> 0x2
2849 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
2850 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
2851 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
2852 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
2853 bash-1998 [000] d... 140.733510: sys_close(fd: a)
2854 bash-1998 [000] d... 140.733510: sys_close -> 0x0
2855 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
2856 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
2857 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
2858 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
2860 You can see that the trace of the top most trace buffer shows only
2861 the function tracing. The foo instance displays wakeups and task
2864 To remove the instances, simply delete their directories:
2866 # rmdir instances/foo
2867 # rmdir instances/bar
2868 # rmdir instances/zoot
2870 Note, if a process has a trace file open in one of the instance
2871 directories, the rmdir will fail with EBUSY.
2876 Since the kernel has a fixed sized stack, it is important not to
2877 waste it in functions. A kernel developer must be conscience of
2878 what they allocate on the stack. If they add too much, the system
2879 can be in danger of a stack overflow, and corruption will occur,
2880 usually leading to a system panic.
2882 There are some tools that check this, usually with interrupts
2883 periodically checking usage. But if you can perform a check
2884 at every function call that will become very useful. As ftrace provides
2885 a function tracer, it makes it convenient to check the stack size
2886 at every function call. This is enabled via the stack tracer.
2888 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
2889 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
2891 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
2893 You can also enable it from the kernel command line to trace
2894 the stack size of the kernel during boot up, by adding "stacktrace"
2895 to the kernel command line parameter.
2897 After running it for a few minutes, the output looks like:
2899 # cat stack_max_size
2903 Depth Size Location (18 entries)
2905 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
2906 1) 2704 160 find_busiest_group+0x31/0x1f1
2907 2) 2544 256 load_balance+0xd9/0x662
2908 3) 2288 80 idle_balance+0xbb/0x130
2909 4) 2208 128 __schedule+0x26e/0x5b9
2910 5) 2080 16 schedule+0x64/0x66
2911 6) 2064 128 schedule_timeout+0x34/0xe0
2912 7) 1936 112 wait_for_common+0x97/0xf1
2913 8) 1824 16 wait_for_completion+0x1d/0x1f
2914 9) 1808 128 flush_work+0xfe/0x119
2915 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
2916 11) 1664 48 input_available_p+0x1d/0x5c
2917 12) 1616 48 n_tty_poll+0x6d/0x134
2918 13) 1568 64 tty_poll+0x64/0x7f
2919 14) 1504 880 do_select+0x31e/0x511
2920 15) 624 400 core_sys_select+0x177/0x216
2921 16) 224 96 sys_select+0x91/0xb9
2922 17) 128 128 system_call_fastpath+0x16/0x1b
2924 Note, if -mfentry is being used by gcc, functions get traced before
2925 they set up the stack frame. This means that leaf level functions
2926 are not tested by the stack tracer when -mfentry is used.
2928 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
2932 More details can be found in the source code, in the
2933 kernel/trace/*.c files.
projects / karo-tx-linux.git / blob