| blob | 4da42616939f05002062117be457699bb8543d69 |
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
11 Updated for: 3.10
13 Introduction
14 ------------
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.
39 The File System
40 ---------------
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
56 it:
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.
75 current_tracer:
77 This is used to set or display the current tracer
78 that is configured.
80 available_tracers:
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.
87 tracing_on:
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
93 still be occurring.
95 trace:
97 This file holds the output of the trace in a human
98 readable format (described below).
100 trace_pipe:
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,they will display the same
112 information every time they are read.
114 trace_options:
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).
121 options:
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.
128 tracing_max_latency:
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)
138 tracing_thresh:
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.
143 (in microseconds)
145 buffer_size_kb:
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.
163 free_buffer:
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
172 will be "freed".
174 It may also stop tracing if disable_on_free option is set.
176 tracing_cpumask:
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.
182 set_ftrace_filter:
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.
196 set_ftrace_notrace:
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.
203 set_ftrace_pid:
205 Have the function tracer only trace a single thread.
207 set_graph_function:
209 Set a "trigger" function where tracing should start
210 with the function graph tracer (See the section
211 "dynamic ftrace" for more details).
213 available_filter_functions:
215 This lists the functions that ftrace
216 has processed and can trace. These are the function
217 names that you can pass to "set_ftrace_filter" or
218 "set_ftrace_notrace". (See the section "dynamic ftrace"
219 below for more details.)
221 enabled_functions:
223 This file is more for debugging ftrace, but can also be useful
224 in seeing if any function has a callback attached to it.
225 Not only does the trace infrastructure use ftrace function
226 trace utility, but other subsystems might too. This file
227 displays all functions that have a callback attached to them
228 as well as the number of callbacks that have been attached.
229 Note, a callback may also call multiple functions which will
230 not be listed in this count.
232 If the callback registered to be traced by a function with
233 the "save regs" attribute (thus even more overhead), a 'R'
234 will be displayed on the same line as the function that
235 is returning registers.
237 function_profile_enabled:
239 When set it will enable all functions with either the function
240 tracer, or if enabled, the function graph tracer. It will
241 keep a histogram of the number of functions that were called
242 and if run with the function graph tracer, it will also keep
243 track of the time spent in those functions. The histogram
244 content can be displayed in the files:
246 trace_stats/function<cpu> ( function0, function1, etc).
248 trace_stats:
250 A directory that holds different tracing stats.
252 kprobe_events:
254 Enable dynamic trace points. See kprobetrace.txt.
256 kprobe_profile:
258 Dynamic trace points stats. See kprobetrace.txt.
260 max_graph_depth:
262 Used with the function graph tracer. This is the max depth
263 it will trace into a function. Setting this to a value of
264 one will show only the first kernel function that is called
265 from user space.
267 printk_formats:
269 This is for tools that read the raw format files. If an event in
270 the ring buffer references a string (currently only trace_printk()
271 does this), only a pointer to the string is recorded into the buffer
272 and not the string itself. This prevents tools from knowing what
273 that string was. This file displays the string and address for
274 the string allowing tools to map the pointers to what the
275 strings were.
277 saved_cmdlines:
279 Only the pid of the task is recorded in a trace event unless
280 the event specifically saves the task comm as well. Ftrace
281 makes a cache of pid mappings to comms to try to display
282 comms for events. If a pid for a comm is not listed, then
283 "<...>" is displayed in the output.
285 snapshot:
287 This displays the "snapshot" buffer and also lets the user
288 take a snapshot of the current running trace.
289 See the "Snapshot" section below for more details.
291 stack_max_size:
293 When the stack tracer is activated, this will display the
294 maximum stack size it has encountered.
295 See the "Stack Trace" section below.
297 stack_trace:
299 This displays the stack back trace of the largest stack
300 that was encountered when the stack tracer is activated.
301 See the "Stack Trace" section below.
303 stack_trace_filter:
305 This is similar to "set_ftrace_filter" but it limits what
306 functions the stack tracer will check.
308 trace_clock:
310 Whenever an event is recorded into the ring buffer, a
311 "timestamp" is added. This stamp comes from a specified
312 clock. By default, ftrace uses the "local" clock. This
313 clock is very fast and strictly per cpu, but on some
314 systems it may not be monotonic with respect to other
315 CPUs. In other words, the local clocks may not be in sync
316 with local clocks on other CPUs.
318 Usual clocks for tracing:
320 # cat trace_clock
321 [local] global counter x86-tsc
323 local: Default clock, but may not be in sync across CPUs
325 global: This clock is in sync with all CPUs but may
326 be a bit slower than the local clock.
328 counter: This is not a clock at all, but literally an atomic
329 counter. It counts up one by one, but is in sync
330 with all CPUs. This is useful when you need to
331 know exactly the order events occurred with respect to
332 each other on different CPUs.
334 uptime: This uses the jiffies counter and the time stamp
335 is relative to the time since boot up.
337 perf: This makes ftrace use the same clock that perf uses.
338 Eventually perf will be able to read ftrace buffers
339 and this will help out in interleaving the data.
341 x86-tsc: Architectures may define their own clocks. For
342 example, x86 uses its own TSC cycle clock here.
344 To set a clock, simply echo the clock name into this file.
346 echo global > trace_clock
348 trace_marker:
350 This is a very useful file for synchronizing user space
351 with events happening in the kernel. Writing strings into
352 this file will be written into the ftrace buffer.
354 It is useful in applications to open this file at the start
355 of the application and just reference the file descriptor
356 for the file.
358 void trace_write(const char *fmt, ...)
359 {
360 va_list ap;
361 char buf[256];
362 int n;
364 if (trace_fd < 0)
365 return;
367 va_start(ap, fmt);
368 n = vsnprintf(buf, 256, fmt, ap);
369 va_end(ap);
371 write(trace_fd, buf, n);
372 }
374 start:
376 trace_fd = open("trace_marker", WR_ONLY);
378 uprobe_events:
380 Add dynamic tracepoints in programs.
381 See uprobetracer.txt
383 uprobe_profile:
385 Uprobe statistics. See uprobetrace.txt
387 instances:
389 This is a way to make multiple trace buffers where different
390 events can be recorded in different buffers.
391 See "Instances" section below.
393 events:
395 This is the trace event directory. It holds event tracepoints
396 (also known as static tracepoints) that have been compiled
397 into the kernel. It shows what event tracepoints exist
398 and how they are grouped by system. There are "enable"
399 files at various levels that can enable the tracepoints
400 when a "1" is written to them.
402 See events.txt for more information.
404 per_cpu:
406 This is a directory that contains the trace per_cpu information.
408 per_cpu/cpu0/buffer_size_kb:
410 The ftrace buffer is defined per_cpu. That is, there's a separate
411 buffer for each CPU to allow writes to be done atomically,
412 and free from cache bouncing. These buffers may have different
413 size buffers. This file is similar to the buffer_size_kb
414 file, but it only displays or sets the buffer size for the
415 specific CPU. (here cpu0).
417 per_cpu/cpu0/trace:
419 This is similar to the "trace" file, but it will only display
420 the data specific for the CPU. If written to, it only clears
421 the specific CPU buffer.
423 per_cpu/cpu0/trace_pipe
425 This is similar to the "trace_pipe" file, and is a consuming
426 read, but it will only display (and consume) the data specific
427 for the CPU.
429 per_cpu/cpu0/trace_pipe_raw
431 For tools that can parse the ftrace ring buffer binary format,
432 the trace_pipe_raw file can be used to extract the data
433 from the ring buffer directly. With the use of the splice()
434 system call, the buffer data can be quickly transferred to
435 a file or to the network where a server is collecting the
436 data.
438 Like trace_pipe, this is a consuming reader, where multiple
439 reads will always produce different data.
441 per_cpu/cpu0/snapshot:
443 This is similar to the main "snapshot" file, but will only
444 snapshot the current CPU (if supported). It only displays
445 the content of the snapshot for a given CPU, and if
446 written to, only clears this CPU buffer.
448 per_cpu/cpu0/snapshot_raw:
450 Similar to the trace_pipe_raw, but will read the binary format
451 from the snapshot buffer for the given CPU.
453 per_cpu/cpu0/stats:
455 This displays certain stats about the ring buffer:
457 entries: The number of events that are still in the buffer.
459 overrun: The number of lost events due to overwriting when
460 the buffer was full.
462 commit overrun: Should always be zero.
463 This gets set if so many events happened within a nested
464 event (ring buffer is re-entrant), that it fills the
465 buffer and starts dropping events.
467 bytes: Bytes actually read (not overwritten).
469 oldest event ts: The oldest timestamp in the buffer
471 now ts: The current timestamp
473 dropped events: Events lost due to overwrite option being off.
475 read events: The number of events read.
477 The Tracers
478 -----------
480 Here is the list of current tracers that may be configured.
482 "function"
484 Function call tracer to trace all kernel functions.
486 "function_graph"
488 Similar to the function tracer except that the
489 function tracer probes the functions on their entry
490 whereas the function graph tracer traces on both entry
491 and exit of the functions. It then provides the ability
492 to draw a graph of function calls similar to C code
493 source.
495 "irqsoff"
497 Traces the areas that disable interrupts and saves
498 the trace with the longest max latency.
499 See tracing_max_latency. When a new max is recorded,
500 it replaces the old trace. It is best to view this
501 trace with the latency-format option enabled.
503 "preemptoff"
505 Similar to irqsoff but traces and records the amount of
506 time for which preemption is disabled.
508 "preemptirqsoff"
510 Similar to irqsoff and preemptoff, but traces and
511 records the largest time for which irqs and/or preemption
512 is disabled.
514 "wakeup"
516 Traces and records the max latency that it takes for
517 the highest priority task to get scheduled after
518 it has been woken up.
519 Traces all tasks as an average developer would expect.
521 "wakeup_rt"
523 Traces and records the max latency that it takes for just
524 RT tasks (as the current "wakeup" does). This is useful
525 for those interested in wake up timings of RT tasks.
527 "nop"
529 This is the "trace nothing" tracer. To remove all
530 tracers from tracing simply echo "nop" into
531 current_tracer.
534 Examples of using the tracer
535 ----------------------------
537 Here are typical examples of using the tracers when controlling
538 them only with the debugfs interface (without using any
539 user-land utilities).
541 Output format:
542 --------------
544 Here is an example of the output format of the file "trace"
546 --------
547 # tracer: function
548 #
549 # entries-in-buffer/entries-written: 140080/250280 #P:4
550 #
551 # _-----=> irqs-off
552 # / _----=> need-resched
553 # | / _---=> hardirq/softirq
554 # || / _--=> preempt-depth
555 # ||| / delay
556 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
557 # | | | |||| | |
558 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
559 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
560 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
561 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
562 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
563 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
564 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
565 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
566 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
567 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
568 --------
570 A header is printed with the tracer name that is represented by
571 the trace. In this case the tracer is "function". Then it shows the
572 number of events in the buffer as well as the total number of entries
573 that were written. The difference is the number of entries that were
574 lost due to the buffer filling up (250280 - 140080 = 110200 events
575 lost).
577 The header explains the content of the events. Task name "bash", the task
578 PID "1977", the CPU that it was running on "000", the latency format
579 (explained below), the timestamp in <secs>.<usecs> format, the
580 function name that was traced "sys_close" and the parent function that
581 called this function "system_call_fastpath". The timestamp is the time
582 at which the function was entered.
584 Latency trace format
585 --------------------
587 When the latency-format option is enabled or when one of the latency
588 tracers is set, the trace file gives somewhat more information to see
589 why a latency happened. Here is a typical trace.
591 # tracer: irqsoff
592 #
593 # irqsoff latency trace v1.1.5 on 3.8.0-test+
594 # --------------------------------------------------------------------
595 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
596 # -----------------
597 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
598 # -----------------
599 # => started at: __lock_task_sighand
600 # => ended at: _raw_spin_unlock_irqrestore
601 #
602 #
603 # _------=> CPU#
604 # / _-----=> irqs-off
605 # | / _----=> need-resched
606 # || / _---=> hardirq/softirq
607 # ||| / _--=> preempt-depth
608 # |||| / delay
609 # cmd pid ||||| time | caller
610 # \ / ||||| \ | /
611 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
612 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
613 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
614 ps-6143 2d..1 306us : <stack trace>
615 => trace_hardirqs_on_caller
616 => trace_hardirqs_on
617 => _raw_spin_unlock_irqrestore
618 => do_task_stat
619 => proc_tgid_stat
620 => proc_single_show
621 => seq_read
622 => vfs_read
623 => sys_read
624 => system_call_fastpath
627 This shows that the current tracer is "irqsoff" tracing the time
628 for which interrupts were disabled. It gives the trace version (which
629 never changes) and the version of the kernel upon which this was executed on
630 (3.10). Then it displays the max latency in microseconds (259 us). The number
631 of trace entries displayed and the total number (both are four: #4/4).
632 VP, KP, SP, and HP are always zero and are reserved for later use.
633 #P is the number of online CPUs (#P:4).
635 The task is the process that was running when the latency
636 occurred. (ps pid: 6143).
638 The start and stop (the functions in which the interrupts were
639 disabled and enabled respectively) that caused the latencies:
641 __lock_task_sighand is where the interrupts were disabled.
642 _raw_spin_unlock_irqrestore is where they were enabled again.
644 The next lines after the header are the trace itself. The header
645 explains which is which.
647 cmd: The name of the process in the trace.
649 pid: The PID of that process.
651 CPU#: The CPU which the process was running on.
653 irqs-off: 'd' interrupts are disabled. '.' otherwise.
654 Note: If the architecture does not support a way to
655 read the irq flags variable, an 'X' will always
656 be printed here.
658 need-resched:
659 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
660 'n' only TIF_NEED_RESCHED is set,
661 'p' only PREEMPT_NEED_RESCHED is set,
662 '.' otherwise.
664 hardirq/softirq:
665 'H' - hard irq occurred inside a softirq.
666 'h' - hard irq is running
667 's' - soft irq is running
668 '.' - normal context.
670 preempt-depth: The level of preempt_disabled
672 The above is mostly meaningful for kernel developers.
674 time: When the latency-format option is enabled, the trace file
675 output includes a timestamp relative to the start of the
676 trace. This differs from the output when latency-format
677 is disabled, which includes an absolute timestamp.
679 delay: This is just to help catch your eye a bit better. And
680 needs to be fixed to be only relative to the same CPU.
681 The marks are determined by the difference between this
682 current trace and the next trace.
683 '!' - greater than preempt_mark_thresh (default 100)
684 '+' - greater than 1 microsecond
685 ' ' - less than or equal to 1 microsecond.
687 The rest is the same as the 'trace' file.
689 Note, the latency tracers will usually end with a back trace
690 to easily find where the latency occurred.
692 trace_options
693 -------------
695 The trace_options file (or the options directory) is used to control
696 what gets printed in the trace output, or manipulate the tracers.
697 To see what is available, simply cat the file:
699 cat trace_options
700 print-parent
701 nosym-offset
702 nosym-addr
703 noverbose
704 noraw
705 nohex
706 nobin
707 noblock
708 nostacktrace
709 trace_printk
710 noftrace_preempt
711 nobranch
712 annotate
713 nouserstacktrace
714 nosym-userobj
715 noprintk-msg-only
716 context-info
717 latency-format
718 sleep-time
719 graph-time
720 record-cmd
721 overwrite
722 nodisable_on_free
723 irq-info
724 markers
725 function-trace
727 To disable one of the options, echo in the option prepended with
728 "no".
730 echo noprint-parent > trace_options
732 To enable an option, leave off the "no".
734 echo sym-offset > trace_options
736 Here are the available options:
738 print-parent - On function traces, display the calling (parent)
739 function as well as the function being traced.
741 print-parent:
742 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
744 noprint-parent:
745 bash-4000 [01] 1477.606694: simple_strtoul
748 sym-offset - Display not only the function name, but also the
749 offset in the function. For example, instead of
750 seeing just "ktime_get", you will see
751 "ktime_get+0xb/0x20".
753 sym-offset:
754 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
756 sym-addr - this will also display the function address as well
757 as the function name.
759 sym-addr:
760 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
762 verbose - This deals with the trace file when the
763 latency-format option is enabled.
765 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
766 (+0.000ms): simple_strtoul (kstrtoul)
768 raw - This will display raw numbers. This option is best for
769 use with user applications that can translate the raw
770 numbers better than having it done in the kernel.
772 hex - Similar to raw, but the numbers will be in a hexadecimal
773 format.
775 bin - This will print out the formats in raw binary.
777 block - When set, reading trace_pipe will not block when polled.
779 stacktrace - This is one of the options that changes the trace
780 itself. When a trace is recorded, so is the stack
781 of functions. This allows for back traces of
782 trace sites.
784 trace_printk - Can disable trace_printk() from writing into the buffer.
786 branch - Enable branch tracing with the tracer.
788 annotate - It is sometimes confusing when the CPU buffers are full
789 and one CPU buffer had a lot of events recently, thus
790 a shorter time frame, were another CPU may have only had
791 a few events, which lets it have older events. When
792 the trace is reported, it shows the oldest events first,
793 and it may look like only one CPU ran (the one with the
794 oldest events). When the annotate option is set, it will
795 display when a new CPU buffer started:
797 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
798 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
799 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
800 ##### CPU 2 buffer started ####
801 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
802 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
803 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
805 userstacktrace - This option changes the trace. It records a
806 stacktrace of the current userspace thread.
808 sym-userobj - when user stacktrace are enabled, look up which
809 object the address belongs to, and print a
810 relative address. This is especially useful when
811 ASLR is on, otherwise you don't get a chance to
812 resolve the address to object/file/line after
813 the app is no longer running
815 The lookup is performed when you read
816 trace,trace_pipe. Example:
818 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
819 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
822 printk-msg-only - When set, trace_printk()s will only show the format
823 and not their parameters (if trace_bprintk() or
824 trace_bputs() was used to save the trace_printk()).
826 context-info - Show only the event data. Hides the comm, PID,
827 timestamp, CPU, and other useful data.
829 latency-format - This option changes the trace. When
830 it is enabled, the trace displays
831 additional information about the
832 latencies, as described in "Latency
833 trace format".
835 sleep-time - When running function graph tracer, to include
836 the time a task schedules out in its function.
837 When enabled, it will account time the task has been
838 scheduled out as part of the function call.
840 graph-time - When running function graph tracer, to include the
841 time to call nested functions. When this is not set,
842 the time reported for the function will only include
843 the time the function itself executed for, not the time
844 for functions that it called.
846 record-cmd - When any event or tracer is enabled, a hook is enabled
847 in the sched_switch trace point to fill comm cache
848 with mapped pids and comms. But this may cause some
849 overhead, and if you only care about pids, and not the
850 name of the task, disabling this option can lower the
851 impact of tracing.
853 overwrite - This controls what happens when the trace buffer is
854 full. If "1" (default), the oldest events are
855 discarded and overwritten. If "0", then the newest
856 events are discarded.
857 (see per_cpu/cpu0/stats for overrun and dropped)
859 disable_on_free - When the free_buffer is closed, tracing will
860 stop (tracing_on set to 0).
862 irq-info - Shows the interrupt, preempt count, need resched data.
863 When disabled, the trace looks like:
865 # tracer: function
866 #
867 # entries-in-buffer/entries-written: 144405/9452052 #P:4
868 #
869 # TASK-PID CPU# TIMESTAMP FUNCTION
870 # | | | | |
871 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
872 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
873 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
876 markers - When set, the trace_marker is writable (only by root).
877 When disabled, the trace_marker will error with EINVAL
878 on write.
881 function-trace - The latency tracers will enable function tracing
882 if this option is enabled (default it is). When
883 it is disabled, the latency tracers do not trace
884 functions. This keeps the overhead of the tracer down
885 when performing latency tests.
887 Note: Some tracers have their own options. They only appear
888 when the tracer is active.
892 irqsoff
893 -------
895 When interrupts are disabled, the CPU can not react to any other
896 external event (besides NMIs and SMIs). This prevents the timer
897 interrupt from triggering or the mouse interrupt from letting
898 the kernel know of a new mouse event. The result is a latency
899 with the reaction time.
901 The irqsoff tracer tracks the time for which interrupts are
902 disabled. When a new maximum latency is hit, the tracer saves
903 the trace leading up to that latency point so that every time a
904 new maximum is reached, the old saved trace is discarded and the
905 new trace is saved.
907 To reset the maximum, echo 0 into tracing_max_latency. Here is
908 an example:
910 # echo 0 > options/function-trace
911 # echo irqsoff > current_tracer
912 # echo 1 > tracing_on
913 # echo 0 > tracing_max_latency
914 # ls -ltr
915 [...]
916 # echo 0 > tracing_on
917 # cat trace
918 # tracer: irqsoff
919 #
920 # irqsoff latency trace v1.1.5 on 3.8.0-test+
921 # --------------------------------------------------------------------
922 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
923 # -----------------
924 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
925 # -----------------
926 # => started at: run_timer_softirq
927 # => ended at: run_timer_softirq
928 #
929 #
930 # _------=> CPU#
931 # / _-----=> irqs-off
932 # | / _----=> need-resched
933 # || / _---=> hardirq/softirq
934 # ||| / _--=> preempt-depth
935 # |||| / delay
936 # cmd pid ||||| time | caller
937 # \ / ||||| \ | /
938 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
939 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
940 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
941 <idle>-0 0dNs3 25us : <stack trace>
942 => _raw_spin_unlock_irq
943 => run_timer_softirq
944 => __do_softirq
945 => call_softirq
946 => do_softirq
947 => irq_exit
948 => smp_apic_timer_interrupt
949 => apic_timer_interrupt
950 => rcu_idle_exit
951 => cpu_idle
952 => rest_init
953 => start_kernel
954 => x86_64_start_reservations
955 => x86_64_start_kernel
957 Here we see that that we had a latency of 16 microseconds (which is
958 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
959 interrupts. The difference between the 16 and the displayed
960 timestamp 25us occurred because the clock was incremented
961 between the time of recording the max latency and the time of
962 recording the function that had that latency.
964 Note the above example had function-trace not set. If we set
965 function-trace, we get a much larger output:
967 with echo 1 > options/function-trace
969 # tracer: irqsoff
970 #
971 # irqsoff latency trace v1.1.5 on 3.8.0-test+
972 # --------------------------------------------------------------------
973 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
974 # -----------------
975 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
976 # -----------------
977 # => started at: ata_scsi_queuecmd
978 # => ended at: ata_scsi_queuecmd
979 #
980 #
981 # _------=> CPU#
982 # / _-----=> irqs-off
983 # | / _----=> need-resched
984 # || / _---=> hardirq/softirq
985 # ||| / _--=> preempt-depth
986 # |||| / delay
987 # cmd pid ||||| time | caller
988 # \ / ||||| \ | /
989 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
990 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
991 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
992 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
993 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
994 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
995 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
996 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
997 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
998 [...]
999 bash-2042 3d..1 67us : delay_tsc <-__delay
1000 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1001 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1002 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1003 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1004 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1005 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1006 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1007 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1008 bash-2042 3d..1 120us : <stack trace>
1009 => _raw_spin_unlock_irqrestore
1010 => ata_scsi_queuecmd
1011 => scsi_dispatch_cmd
1012 => scsi_request_fn
1013 => __blk_run_queue_uncond
1014 => __blk_run_queue
1015 => blk_queue_bio
1016 => generic_make_request
1017 => submit_bio
1018 => submit_bh
1019 => __ext3_get_inode_loc
1020 => ext3_iget
1021 => ext3_lookup
1022 => lookup_real
1023 => __lookup_hash
1024 => walk_component
1025 => lookup_last
1026 => path_lookupat
1027 => filename_lookup
1028 => user_path_at_empty
1029 => user_path_at
1030 => vfs_fstatat
1031 => vfs_stat
1032 => sys_newstat
1033 => system_call_fastpath
1036 Here we traced a 71 microsecond latency. But we also see all the
1037 functions that were called during that time. Note that by
1038 enabling function tracing, we incur an added overhead. This
1039 overhead may extend the latency times. But nevertheless, this
1040 trace has provided some very helpful debugging information.
1043 preemptoff
1044 ----------
1046 When preemption is disabled, we may be able to receive
1047 interrupts but the task cannot be preempted and a higher
1048 priority task must wait for preemption to be enabled again
1049 before it can preempt a lower priority task.
1051 The preemptoff tracer traces the places that disable preemption.
1052 Like the irqsoff tracer, it records the maximum latency for
1053 which preemption was disabled. The control of preemptoff tracer
1054 is much like the irqsoff tracer.
1056 # echo 0 > options/function-trace
1057 # echo preemptoff > current_tracer
1058 # echo 1 > tracing_on
1059 # echo 0 > tracing_max_latency
1060 # ls -ltr
1061 [...]
1062 # echo 0 > tracing_on
1063 # cat trace
1064 # tracer: preemptoff
1065 #
1066 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1067 # --------------------------------------------------------------------
1068 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1069 # -----------------
1070 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1071 # -----------------
1072 # => started at: do_IRQ
1073 # => ended at: do_IRQ
1074 #
1075 #
1076 # _------=> CPU#
1077 # / _-----=> irqs-off
1078 # | / _----=> need-resched
1079 # || / _---=> hardirq/softirq
1080 # ||| / _--=> preempt-depth
1081 # |||| / delay
1082 # cmd pid ||||| time | caller
1083 # \ / ||||| \ | /
1084 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1085 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1086 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1087 sshd-1991 1d..1 52us : <stack trace>
1088 => sub_preempt_count
1089 => irq_exit
1090 => do_IRQ
1091 => ret_from_intr
1094 This has some more changes. Preemption was disabled when an
1095 interrupt came in (notice the 'h'), and was enabled on exit.
1096 But we also see that interrupts have been disabled when entering
1097 the preempt off section and leaving it (the 'd'). We do not know if
1098 interrupts were enabled in the mean time or shortly after this
1099 was over.
1101 # tracer: preemptoff
1102 #
1103 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1104 # --------------------------------------------------------------------
1105 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1106 # -----------------
1107 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1108 # -----------------
1109 # => started at: wake_up_new_task
1110 # => ended at: task_rq_unlock
1111 #
1112 #
1113 # _------=> CPU#
1114 # / _-----=> irqs-off
1115 # | / _----=> need-resched
1116 # || / _---=> hardirq/softirq
1117 # ||| / _--=> preempt-depth
1118 # |||| / delay
1119 # cmd pid ||||| time | caller
1120 # \ / ||||| \ | /
1121 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1122 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1123 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1124 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1125 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1126 [...]
1127 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1128 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1129 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1130 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1131 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1132 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1133 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1134 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1135 [...]
1136 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1137 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1138 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1139 bash-1994 1d..2 36us : do_softirq <-irq_exit
1140 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1141 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1142 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1143 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1144 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1145 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1146 [...]
1147 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1148 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1149 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1150 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1151 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1152 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1153 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1154 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1155 bash-1994 1.N.1 104us : <stack trace>
1156 => sub_preempt_count
1157 => _raw_spin_unlock_irqrestore
1158 => task_rq_unlock
1159 => wake_up_new_task
1160 => do_fork
1161 => sys_clone
1162 => stub_clone
1165 The above is an example of the preemptoff trace with
1166 function-trace set. Here we see that interrupts were not disabled
1167 the entire time. The irq_enter code lets us know that we entered
1168 an interrupt 'h'. Before that, the functions being traced still
1169 show that it is not in an interrupt, but we can see from the
1170 functions themselves that this is not the case.
1172 preemptirqsoff
1173 --------------
1175 Knowing the locations that have interrupts disabled or
1176 preemption disabled for the longest times is helpful. But
1177 sometimes we would like to know when either preemption and/or
1178 interrupts are disabled.
1180 Consider the following code:
1182 local_irq_disable();
1183 call_function_with_irqs_off();
1184 preempt_disable();
1185 call_function_with_irqs_and_preemption_off();
1186 local_irq_enable();
1187 call_function_with_preemption_off();
1188 preempt_enable();
1190 The irqsoff tracer will record the total length of
1191 call_function_with_irqs_off() and
1192 call_function_with_irqs_and_preemption_off().
1194 The preemptoff tracer will record the total length of
1195 call_function_with_irqs_and_preemption_off() and
1196 call_function_with_preemption_off().
1198 But neither will trace the time that interrupts and/or
1199 preemption is disabled. This total time is the time that we can
1200 not schedule. To record this time, use the preemptirqsoff
1201 tracer.
1203 Again, using this trace is much like the irqsoff and preemptoff
1204 tracers.
1206 # echo 0 > options/function-trace
1207 # echo preemptirqsoff > current_tracer
1208 # echo 1 > tracing_on
1209 # echo 0 > tracing_max_latency
1210 # ls -ltr
1211 [...]
1212 # echo 0 > tracing_on
1213 # cat trace
1214 # tracer: preemptirqsoff
1215 #
1216 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1217 # --------------------------------------------------------------------
1218 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1219 # -----------------
1220 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1221 # -----------------
1222 # => started at: ata_scsi_queuecmd
1223 # => ended at: ata_scsi_queuecmd
1224 #
1225 #
1226 # _------=> CPU#
1227 # / _-----=> irqs-off
1228 # | / _----=> need-resched
1229 # || / _---=> hardirq/softirq
1230 # ||| / _--=> preempt-depth
1231 # |||| / delay
1232 # cmd pid ||||| time | caller
1233 # \ / ||||| \ | /
1234 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1235 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1236 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1237 ls-2230 3...1 111us : <stack trace>
1238 => sub_preempt_count
1239 => _raw_spin_unlock_irqrestore
1240 => ata_scsi_queuecmd
1241 => scsi_dispatch_cmd
1242 => scsi_request_fn
1243 => __blk_run_queue_uncond
1244 => __blk_run_queue
1245 => blk_queue_bio
1246 => generic_make_request
1247 => submit_bio
1248 => submit_bh
1249 => ext3_bread
1250 => ext3_dir_bread
1251 => htree_dirblock_to_tree
1252 => ext3_htree_fill_tree
1253 => ext3_readdir
1254 => vfs_readdir
1255 => sys_getdents
1256 => system_call_fastpath
1259 The trace_hardirqs_off_thunk is called from assembly on x86 when
1260 interrupts are disabled in the assembly code. Without the
1261 function tracing, we do not know if interrupts were enabled
1262 within the preemption points. We do see that it started with
1263 preemption enabled.
1265 Here is a trace with function-trace set:
1267 # tracer: preemptirqsoff
1268 #
1269 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1270 # --------------------------------------------------------------------
1271 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1272 # -----------------
1273 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1274 # -----------------
1275 # => started at: schedule
1276 # => ended at: mutex_unlock
1277 #
1278 #
1279 # _------=> CPU#
1280 # / _-----=> irqs-off
1281 # | / _----=> need-resched
1282 # || / _---=> hardirq/softirq
1283 # ||| / _--=> preempt-depth
1284 # |||| / delay
1285 # cmd pid ||||| time | caller
1286 # \ / ||||| \ | /
1287 kworker/-59 3...1 0us : __schedule <-schedule
1288 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1289 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1290 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1291 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1292 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1293 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1294 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1295 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1296 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1297 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1298 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1299 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1300 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1301 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1302 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1303 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1304 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1305 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1306 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1307 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1308 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1309 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1310 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1311 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1312 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1313 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1314 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1315 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1316 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1317 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1318 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1319 [...]
1320 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1321 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1322 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1323 ls-2269 3d..3 21us : do_softirq <-irq_exit
1324 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1325 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1326 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1327 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1328 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1329 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1330 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1331 [...]
1332 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1333 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1334 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1335 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1336 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1337 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1338 [...]
1339 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1340 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1341 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1342 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1343 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1344 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1345 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1346 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1347 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1348 ls-2269 3d... 186us : <stack trace>
1349 => __mutex_unlock_slowpath
1350 => mutex_unlock
1351 => process_output
1352 => n_tty_write
1353 => tty_write
1354 => vfs_write
1355 => sys_write
1356 => system_call_fastpath
1358 This is an interesting trace. It started with kworker running and
1359 scheduling out and ls taking over. But as soon as ls released the
1360 rq lock and enabled interrupts (but not preemption) an interrupt
1361 triggered. When the interrupt finished, it started running softirqs.
1362 But while the softirq was running, another interrupt triggered.
1363 When an interrupt is running inside a softirq, the annotation is 'H'.
1366 wakeup
1367 ------
1369 One common case that people are interested in tracing is the
1370 time it takes for a task that is woken to actually wake up.
1371 Now for non Real-Time tasks, this can be arbitrary. But tracing
1372 it none the less can be interesting.
1374 Without function tracing:
1376 # echo 0 > options/function-trace
1377 # echo wakeup > current_tracer
1378 # echo 1 > tracing_on
1379 # echo 0 > tracing_max_latency
1380 # chrt -f 5 sleep 1
1381 # echo 0 > tracing_on
1382 # cat trace
1383 # tracer: wakeup
1384 #
1385 # wakeup latency trace v1.1.5 on 3.8.0-test+
1386 # --------------------------------------------------------------------
1387 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1388 # -----------------
1389 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1390 # -----------------
1391 #
1392 # _------=> CPU#
1393 # / _-----=> irqs-off
1394 # | / _----=> need-resched
1395 # || / _---=> hardirq/softirq
1396 # ||| / _--=> preempt-depth
1397 # |||| / delay
1398 # cmd pid ||||| time | caller
1399 # \ / ||||| \ | /
1400 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1401 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1402 <idle>-0 3d..3 15us : __schedule <-schedule
1403 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1405 The tracer only traces the highest priority task in the system
1406 to avoid tracing the normal circumstances. Here we see that
1407 the kworker with a nice priority of -20 (not very nice), took
1408 just 15 microseconds from the time it woke up, to the time it
1409 ran.
1411 Non Real-Time tasks are not that interesting. A more interesting
1412 trace is to concentrate only on Real-Time tasks.
1414 wakeup_rt
1415 ---------
1417 In a Real-Time environment it is very important to know the
1418 wakeup time it takes for the highest priority task that is woken
1419 up to the time that it executes. This is also known as "schedule
1420 latency". I stress the point that this is about RT tasks. It is
1421 also important to know the scheduling latency of non-RT tasks,
1422 but the average schedule latency is better for non-RT tasks.
1423 Tools like LatencyTop are more appropriate for such
1424 measurements.
1426 Real-Time environments are interested in the worst case latency.
1427 That is the longest latency it takes for something to happen,
1428 and not the average. We can have a very fast scheduler that may
1429 only have a large latency once in a while, but that would not
1430 work well with Real-Time tasks. The wakeup_rt tracer was designed
1431 to record the worst case wakeups of RT tasks. Non-RT tasks are
1432 not recorded because the tracer only records one worst case and
1433 tracing non-RT tasks that are unpredictable will overwrite the
1434 worst case latency of RT tasks (just run the normal wakeup
1435 tracer for a while to see that effect).
1437 Since this tracer only deals with RT tasks, we will run this
1438 slightly differently than we did with the previous tracers.
1439 Instead of performing an 'ls', we will run 'sleep 1' under
1440 'chrt' which changes the priority of the task.
1442 # echo 0 > options/function-trace
1443 # echo wakeup_rt > current_tracer
1444 # echo 1 > tracing_on
1445 # echo 0 > tracing_max_latency
1446 # chrt -f 5 sleep 1
1447 # echo 0 > tracing_on
1448 # cat trace
1449 # tracer: wakeup
1450 #
1451 # tracer: wakeup_rt
1452 #
1453 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1454 # --------------------------------------------------------------------
1455 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1456 # -----------------
1457 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1458 # -----------------
1459 #
1460 # _------=> CPU#
1461 # / _-----=> irqs-off
1462 # | / _----=> need-resched
1463 # || / _---=> hardirq/softirq
1464 # ||| / _--=> preempt-depth
1465 # |||| / delay
1466 # cmd pid ||||| time | caller
1467 # \ / ||||| \ | /
1468 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1469 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1470 <idle>-0 3d..3 5us : __schedule <-schedule
1471 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1474 Running this on an idle system, we see that it only took 5 microseconds
1475 to perform the task switch. Note, since the trace point in the schedule
1476 is before the actual "switch", we stop the tracing when the recorded task
1477 is about to schedule in. This may change if we add a new marker at the
1478 end of the scheduler.
1480 Notice that the recorded task is 'sleep' with the PID of 2389
1481 and it has an rt_prio of 5. This priority is user-space priority
1482 and not the internal kernel priority. The policy is 1 for
1483 SCHED_FIFO and 2 for SCHED_RR.
1485 Note, that the trace data shows the internal priority (99 - rtprio).
1487 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1489 The 0:120:R means idle was running with a nice priority of 0 (120 - 20)
1490 and in the running state 'R'. The sleep task was scheduled in with
1491 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1492 and it too is in the running state.
1494 Doing the same with chrt -r 5 and function-trace set.
1496 echo 1 > options/function-trace
1498 # tracer: wakeup_rt
1499 #
1500 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1501 # --------------------------------------------------------------------
1502 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1503 # -----------------
1504 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
1505 # -----------------
1506 #
1507 # _------=> CPU#
1508 # / _-----=> irqs-off
1509 # | / _----=> need-resched
1510 # || / _---=> hardirq/softirq
1511 # ||| / _--=> preempt-depth
1512 # |||| / delay
1513 # cmd pid ||||| time | caller
1514 # \ / ||||| \ | /
1515 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
1516 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1517 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
1518 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
1519 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
1520 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
1521 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
1522 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
1523 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
1524 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1525 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
1526 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
1527 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
1528 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
1529 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
1530 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
1531 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
1532 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
1533 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
1534 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
1535 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
1536 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
1537 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
1538 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
1539 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
1540 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
1541 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
1542 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
1543 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
1544 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
1545 <idle>-0 3dN.1 13us : update_cpu_load_nohz <-tick_nohz_idle_exit
1546 <idle>-0 3dN.1 13us : _raw_spin_lock <-update_cpu_load_nohz
1547 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
1548 <idle>-0 3dN.2 13us : __update_cpu_load <-update_cpu_load_nohz
1549 <idle>-0 3dN.2 14us : sched_avg_update <-__update_cpu_load
1550 <idle>-0 3dN.2 14us : _raw_spin_unlock <-update_cpu_load_nohz
1551 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
1552 <idle>-0 3dN.1 15us : calc_load_exit_idle <-tick_nohz_idle_exit
1553 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
1554 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
1555 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
1556 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
1557 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1558 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
1559 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
1560 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
1561 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
1562 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
1563 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
1564 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
1565 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
1566 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1567 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
1568 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1569 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1570 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
1571 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
1572 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
1573 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1574 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
1575 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
1576 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
1577 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
1578 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
1579 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
1580 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
1581 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
1582 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1583 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
1584 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
1585 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
1586 <idle>-0 3.N.. 25us : schedule <-cpu_idle
1587 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
1588 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
1589 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
1590 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
1591 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
1592 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
1593 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
1594 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
1595 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
1596 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
1597 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
1598 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
1599 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
1601 This isn't that big of a trace, even with function tracing enabled,
1602 so I included the entire trace.
1604 The interrupt went off while when the system was idle. Somewhere
1605 before task_woken_rt() was called, the NEED_RESCHED flag was set,
1606 this is indicated by the first occurrence of the 'N' flag.
1608 Latency tracing and events
1609 --------------------------
1610 As function tracing can induce a much larger latency, but without
1611 seeing what happens within the latency it is hard to know what
1612 caused it. There is a middle ground, and that is with enabling
1613 events.
1615 # echo 0 > options/function-trace
1616 # echo wakeup_rt > current_tracer
1617 # echo 1 > events/enable
1618 # echo 1 > tracing_on
1619 # echo 0 > tracing_max_latency
1620 # chrt -f 5 sleep 1
1621 # echo 0 > tracing_on
1622 # cat trace
1623 # tracer: wakeup_rt
1624 #
1625 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1626 # --------------------------------------------------------------------
1627 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1628 # -----------------
1629 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
1630 # -----------------
1631 #
1632 # _------=> CPU#
1633 # / _-----=> irqs-off
1634 # | / _----=> need-resched
1635 # || / _---=> hardirq/softirq
1636 # ||| / _--=> preempt-depth
1637 # |||| / delay
1638 # cmd pid ||||| time | caller
1639 # \ / ||||| \ | /
1640 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
1641 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1642 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
1643 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
1644 <idle>-0 2.N.2 2us : power_end: cpu_id=2
1645 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
1646 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
1647 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
1648 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
1649 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
1650 <idle>-0 2d..3 6us : __schedule <-schedule
1651 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
1654 function
1655 --------
1657 This tracer is the function tracer. Enabling the function tracer
1658 can be done from the debug file system. Make sure the
1659 ftrace_enabled is set; otherwise this tracer is a nop.
1660 See the "ftrace_enabled" section below.
1662 # sysctl kernel.ftrace_enabled=1
1663 # echo function > current_tracer
1664 # echo 1 > tracing_on
1665 # usleep 1
1666 # echo 0 > tracing_on
1667 # cat trace
1668 # tracer: function
1669 #
1670 # entries-in-buffer/entries-written: 24799/24799 #P:4
1671 #
1672 # _-----=> irqs-off
1673 # / _----=> need-resched
1674 # | / _---=> hardirq/softirq
1675 # || / _--=> preempt-depth
1676 # ||| / delay
1677 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
1678 # | | | |||| | |
1679 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
1680 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
1681 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
1682 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
1683 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
1684 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
1685 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
1686 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
1687 [...]
1690 Note: function tracer uses ring buffers to store the above
1691 entries. The newest data may overwrite the oldest data.
1692 Sometimes using echo to stop the trace is not sufficient because
1693 the tracing could have overwritten the data that you wanted to
1694 record. For this reason, it is sometimes better to disable
1695 tracing directly from a program. This allows you to stop the
1696 tracing at the point that you hit the part that you are
1697 interested in. To disable the tracing directly from a C program,
1698 something like following code snippet can be used:
1700 int trace_fd;
1701 [...]
1702 int main(int argc, char *argv[]) {
1703 [...]
1704 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
1705 [...]
1706 if (condition_hit()) {
1707 write(trace_fd, "0", 1);
1708 }
1709 [...]
1710 }
1713 Single thread tracing
1714 ---------------------
1716 By writing into set_ftrace_pid you can trace a
1717 single thread. For example:
1719 # cat set_ftrace_pid
1720 no pid
1721 # echo 3111 > set_ftrace_pid
1722 # cat set_ftrace_pid
1723 3111
1724 # echo function > current_tracer
1725 # cat trace | head
1726 # tracer: function
1727 #
1728 # TASK-PID CPU# TIMESTAMP FUNCTION
1729 # | | | | |
1730 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
1731 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
1732 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
1733 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
1734 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
1735 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
1736 # echo -1 > set_ftrace_pid
1737 # cat trace |head
1738 # tracer: function
1739 #
1740 # TASK-PID CPU# TIMESTAMP FUNCTION
1741 # | | | | |
1742 ##### CPU 3 buffer started ####
1743 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
1744 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
1745 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
1746 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
1747 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
1749 If you want to trace a function when executing, you could use
1750 something like this simple program:
1752 #include <stdio.h>
1753 #include <stdlib.h>
1754 #include <sys/types.h>
1755 #include <sys/stat.h>
1756 #include <fcntl.h>
1757 #include <unistd.h>
1758 #include <string.h>
1760 #define _STR(x) #x
1761 #define STR(x) _STR(x)
1762 #define MAX_PATH 256
1764 const char *find_debugfs(void)
1765 {
1766 static char debugfs[MAX_PATH+1];
1767 static int debugfs_found;
1768 char type[100];
1769 FILE *fp;
1771 if (debugfs_found)
1772 return debugfs;
1774 if ((fp = fopen("/proc/mounts","r")) == NULL) {
1775 perror("/proc/mounts");
1776 return NULL;
1777 }
1779 while (fscanf(fp, "%*s %"
1780 STR(MAX_PATH)
1781 "s %99s %*s %*d %*d\n",
1782 debugfs, type) == 2) {
1783 if (strcmp(type, "debugfs") == 0)
1784 break;
1785 }
1786 fclose(fp);
1788 if (strcmp(type, "debugfs") != 0) {
1789 fprintf(stderr, "debugfs not mounted");
1790 return NULL;
1791 }
1793 strcat(debugfs, "/tracing/");
1794 debugfs_found = 1;
1796 return debugfs;
1797 }
1799 const char *tracing_file(const char *file_name)
1800 {
1801 static char trace_file[MAX_PATH+1];
1802 snprintf(trace_file, MAX_PATH, "%s/%s", find_debugfs(), file_name);
1803 return trace_file;
1804 }
1806 int main (int argc, char **argv)
1807 {
1808 if (argc < 1)
1809 exit(-1);
1811 if (fork() > 0) {
1812 int fd, ffd;
1813 char line[64];
1814 int s;
1816 ffd = open(tracing_file("current_tracer"), O_WRONLY);
1817 if (ffd < 0)
1818 exit(-1);
1819 write(ffd, "nop", 3);
1821 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
1822 s = sprintf(line, "%d\n", getpid());
1823 write(fd, line, s);
1825 write(ffd, "function", 8);
1827 close(fd);
1828 close(ffd);
1830 execvp(argv[1], argv+1);
1831 }
1833 return 0;
1834 }
1836 Or this simple script!
1838 ------
1839 #!/bin/bash
1841 debugfs=`sed -ne 's/^debugfs \(.*\) debugfs.*/\1/p' /proc/mounts`
1842 echo nop > $debugfs/tracing/current_tracer
1843 echo 0 > $debugfs/tracing/tracing_on
1844 echo $$ > $debugfs/tracing/set_ftrace_pid
1845 echo function > $debugfs/tracing/current_tracer
1846 echo 1 > $debugfs/tracing/tracing_on
1847 exec "$@"
1848 ------
1851 function graph tracer
1852 ---------------------------
1854 This tracer is similar to the function tracer except that it
1855 probes a function on its entry and its exit. This is done by
1856 using a dynamically allocated stack of return addresses in each
1857 task_struct. On function entry the tracer overwrites the return
1858 address of each function traced to set a custom probe. Thus the
1859 original return address is stored on the stack of return address
1860 in the task_struct.
1862 Probing on both ends of a function leads to special features
1863 such as:
1865 - measure of a function's time execution
1866 - having a reliable call stack to draw function calls graph
1868 This tracer is useful in several situations:
1870 - you want to find the reason of a strange kernel behavior and
1871 need to see what happens in detail on any areas (or specific
1872 ones).
1874 - you are experiencing weird latencies but it's difficult to
1875 find its origin.
1877 - you want to find quickly which path is taken by a specific
1878 function
1880 - you just want to peek inside a working kernel and want to see
1881 what happens there.
1883 # tracer: function_graph
1884 #
1885 # CPU DURATION FUNCTION CALLS
1886 # | | | | | | |
1888 0) | sys_open() {
1889 0) | do_sys_open() {
1890 0) | getname() {
1891 0) | kmem_cache_alloc() {
1892 0) 1.382 us | __might_sleep();
1893 0) 2.478 us | }
1894 0) | strncpy_from_user() {
1895 0) | might_fault() {
1896 0) 1.389 us | __might_sleep();
1897 0) 2.553 us | }
1898 0) 3.807 us | }
1899 0) 7.876 us | }
1900 0) | alloc_fd() {
1901 0) 0.668 us | _spin_lock();
1902 0) 0.570 us | expand_files();
1903 0) 0.586 us | _spin_unlock();
1906 There are several columns that can be dynamically
1907 enabled/disabled. You can use every combination of options you
1908 want, depending on your needs.
1910 - The cpu number on which the function executed is default
1911 enabled. It is sometimes better to only trace one cpu (see
1912 tracing_cpu_mask file) or you might sometimes see unordered
1913 function calls while cpu tracing switch.
1915 hide: echo nofuncgraph-cpu > trace_options
1916 show: echo funcgraph-cpu > trace_options
1918 - The duration (function's time of execution) is displayed on
1919 the closing bracket line of a function or on the same line
1920 than the current function in case of a leaf one. It is default
1921 enabled.
1923 hide: echo nofuncgraph-duration > trace_options
1924 show: echo funcgraph-duration > trace_options
1926 - The overhead field precedes the duration field in case of
1927 reached duration thresholds.
1929 hide: echo nofuncgraph-overhead > trace_options
1930 show: echo funcgraph-overhead > trace_options
1931 depends on: funcgraph-duration
1933 ie:
1935 0) | up_write() {
1936 0) 0.646 us | _spin_lock_irqsave();
1937 0) 0.684 us | _spin_unlock_irqrestore();
1938 0) 3.123 us | }
1939 0) 0.548 us | fput();
1940 0) + 58.628 us | }
1942 [...]
1944 0) | putname() {
1945 0) | kmem_cache_free() {
1946 0) 0.518 us | __phys_addr();
1947 0) 1.757 us | }
1948 0) 2.861 us | }
1949 0) ! 115.305 us | }
1950 0) ! 116.402 us | }
1952 + means that the function exceeded 10 usecs.
1953 ! means that the function exceeded 100 usecs.
1956 - The task/pid field displays the thread cmdline and pid which
1957 executed the function. It is default disabled.
1959 hide: echo nofuncgraph-proc > trace_options
1960 show: echo funcgraph-proc > trace_options
1962 ie:
1964 # tracer: function_graph
1965 #
1966 # CPU TASK/PID DURATION FUNCTION CALLS
1967 # | | | | | | | | |
1968 0) sh-4802 | | d_free() {
1969 0) sh-4802 | | call_rcu() {
1970 0) sh-4802 | | __call_rcu() {
1971 0) sh-4802 | 0.616 us | rcu_process_gp_end();
1972 0) sh-4802 | 0.586 us | check_for_new_grace_period();
1973 0) sh-4802 | 2.899 us | }
1974 0) sh-4802 | 4.040 us | }
1975 0) sh-4802 | 5.151 us | }
1976 0) sh-4802 | + 49.370 us | }
1979 - The absolute time field is an absolute timestamp given by the
1980 system clock since it started. A snapshot of this time is
1981 given on each entry/exit of functions
1983 hide: echo nofuncgraph-abstime > trace_options
1984 show: echo funcgraph-abstime > trace_options
1986 ie:
1988 #
1989 # TIME CPU DURATION FUNCTION CALLS
1990 # | | | | | | | |
1991 360.774522 | 1) 0.541 us | }
1992 360.774522 | 1) 4.663 us | }
1993 360.774523 | 1) 0.541 us | __wake_up_bit();
1994 360.774524 | 1) 6.796 us | }
1995 360.774524 | 1) 7.952 us | }
1996 360.774525 | 1) 9.063 us | }
1997 360.774525 | 1) 0.615 us | journal_mark_dirty();
1998 360.774527 | 1) 0.578 us | __brelse();
1999 360.774528 | 1) | reiserfs_prepare_for_journal() {
2000 360.774528 | 1) | unlock_buffer() {
2001 360.774529 | 1) | wake_up_bit() {
2002 360.774529 | 1) | bit_waitqueue() {
2003 360.774530 | 1) 0.594 us | __phys_addr();
2006 The function name is always displayed after the closing bracket
2007 for a function if the start of that function is not in the
2008 trace buffer.
2010 Display of the function name after the closing bracket may be
2011 enabled for functions whose start is in the trace buffer,
2012 allowing easier searching with grep for function durations.
2013 It is default disabled.
2015 hide: echo nofuncgraph-tail > trace_options
2016 show: echo funcgraph-tail > trace_options
2018 Example with nofuncgraph-tail (default):
2019 0) | putname() {
2020 0) | kmem_cache_free() {
2021 0) 0.518 us | __phys_addr();
2022 0) 1.757 us | }
2023 0) 2.861 us | }
2025 Example with funcgraph-tail:
2026 0) | putname() {
2027 0) | kmem_cache_free() {
2028 0) 0.518 us | __phys_addr();
2029 0) 1.757 us | } /* kmem_cache_free() */
2030 0) 2.861 us | } /* putname() */
2032 You can put some comments on specific functions by using
2033 trace_printk() For example, if you want to put a comment inside
2034 the __might_sleep() function, you just have to include
2035 <linux/ftrace.h> and call trace_printk() inside __might_sleep()
2037 trace_printk("I'm a comment!\n")
2039 will produce:
2041 1) | __might_sleep() {
2042 1) | /* I'm a comment! */
2043 1) 1.449 us | }
2046 You might find other useful features for this tracer in the
2047 following "dynamic ftrace" section such as tracing only specific
2048 functions or tasks.
2050 dynamic ftrace
2051 --------------
2053 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2054 virtually no overhead when function tracing is disabled. The way
2055 this works is the mcount function call (placed at the start of
2056 every kernel function, produced by the -pg switch in gcc),
2057 starts of pointing to a simple return. (Enabling FTRACE will
2058 include the -pg switch in the compiling of the kernel.)
2060 At compile time every C file object is run through the
2061 recordmcount program (located in the scripts directory). This
2062 program will parse the ELF headers in the C object to find all
2063 the locations in the .text section that call mcount. (Note, only
2064 white listed .text sections are processed, since processing other
2065 sections like .init.text may cause races due to those sections
2066 being freed unexpectedly).
2068 A new section called "__mcount_loc" is created that holds
2069 references to all the mcount call sites in the .text section.
2070 The recordmcount program re-links this section back into the
2071 original object. The final linking stage of the kernel will add all these
2072 references into a single table.
2074 On boot up, before SMP is initialized, the dynamic ftrace code
2075 scans this table and updates all the locations into nops. It
2076 also records the locations, which are added to the
2077 available_filter_functions list. Modules are processed as they
2078 are loaded and before they are executed. When a module is
2079 unloaded, it also removes its functions from the ftrace function
2080 list. This is automatic in the module unload code, and the
2081 module author does not need to worry about it.
2083 When tracing is enabled, the process of modifying the function
2084 tracepoints is dependent on architecture. The old method is to use
2085 kstop_machine to prevent races with the CPUs executing code being
2086 modified (which can cause the CPU to do undesirable things, especially
2087 if the modified code crosses cache (or page) boundaries), and the nops are
2088 patched back to calls. But this time, they do not call mcount
2089 (which is just a function stub). They now call into the ftrace
2090 infrastructure.
2092 The new method of modifying the function tracepoints is to place
2093 a breakpoint at the location to be modified, sync all CPUs, modify
2094 the rest of the instruction not covered by the breakpoint. Sync
2095 all CPUs again, and then remove the breakpoint with the finished
2096 version to the ftrace call site.
2098 Some archs do not even need to monkey around with the synchronization,
2099 and can just slap the new code on top of the old without any
2100 problems with other CPUs executing it at the same time.
2102 One special side-effect to the recording of the functions being
2103 traced is that we can now selectively choose which functions we
2104 wish to trace and which ones we want the mcount calls to remain
2105 as nops.
2107 Two files are used, one for enabling and one for disabling the
2108 tracing of specified functions. They are:
2110 set_ftrace_filter
2112 and
2114 set_ftrace_notrace
2116 A list of available functions that you can add to these files is
2117 listed in:
2119 available_filter_functions
2121 # cat available_filter_functions
2122 put_prev_task_idle
2123 kmem_cache_create
2124 pick_next_task_rt
2125 get_online_cpus
2126 pick_next_task_fair
2127 mutex_lock
2128 [...]
2130 If I am only interested in sys_nanosleep and hrtimer_interrupt:
2132 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2133 # echo function > current_tracer
2134 # echo 1 > tracing_on
2135 # usleep 1
2136 # echo 0 > tracing_on
2137 # cat trace
2138 # tracer: function
2139 #
2140 # entries-in-buffer/entries-written: 5/5 #P:4
2141 #
2142 # _-----=> irqs-off
2143 # / _----=> need-resched
2144 # | / _---=> hardirq/softirq
2145 # || / _--=> preempt-depth
2146 # ||| / delay
2147 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2148 # | | | |||| | |
2149 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2150 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2151 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2152 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2153 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2155 To see which functions are being traced, you can cat the file:
2157 # cat set_ftrace_filter
2158 hrtimer_interrupt
2159 sys_nanosleep
2162 Perhaps this is not enough. The filters also allow simple wild
2163 cards. Only the following are currently available
2165 <match>* - will match functions that begin with <match>
2166 *<match> - will match functions that end with <match>
2167 *<match>* - will match functions that have <match> in it
2169 These are the only wild cards which are supported.
2171 <match>*<match> will not work.
2173 Note: It is better to use quotes to enclose the wild cards,
2174 otherwise the shell may expand the parameters into names
2175 of files in the local directory.
2177 # echo 'hrtimer_*' > set_ftrace_filter
2179 Produces:
2181 # tracer: function
2182 #
2183 # entries-in-buffer/entries-written: 897/897 #P:4
2184 #
2185 # _-----=> irqs-off
2186 # / _----=> need-resched
2187 # | / _---=> hardirq/softirq
2188 # || / _--=> preempt-depth
2189 # ||| / delay
2190 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2191 # | | | |||| | |
2192 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2193 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2194 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2195 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2196 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2197 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2198 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2199 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2201 Notice that we lost the sys_nanosleep.
2203 # cat set_ftrace_filter
2204 hrtimer_run_queues
2205 hrtimer_run_pending
2206 hrtimer_init
2207 hrtimer_cancel
2208 hrtimer_try_to_cancel
2209 hrtimer_forward
2210 hrtimer_start
2211 hrtimer_reprogram
2212 hrtimer_force_reprogram
2213 hrtimer_get_next_event
2214 hrtimer_interrupt
2215 hrtimer_nanosleep
2216 hrtimer_wakeup
2217 hrtimer_get_remaining
2218 hrtimer_get_res
2219 hrtimer_init_sleeper
2222 This is because the '>' and '>>' act just like they do in bash.
2223 To rewrite the filters, use '>'
2224 To append to the filters, use '>>'
2226 To clear out a filter so that all functions will be recorded
2227 again:
2229 # echo > set_ftrace_filter
2230 # cat set_ftrace_filter
2231 #
2233 Again, now we want to append.
2235 # echo sys_nanosleep > set_ftrace_filter
2236 # cat set_ftrace_filter
2237 sys_nanosleep
2238 # echo 'hrtimer_*' >> set_ftrace_filter
2239 # cat set_ftrace_filter
2240 hrtimer_run_queues
2241 hrtimer_run_pending
2242 hrtimer_init
2243 hrtimer_cancel
2244 hrtimer_try_to_cancel
2245 hrtimer_forward
2246 hrtimer_start
2247 hrtimer_reprogram
2248 hrtimer_force_reprogram
2249 hrtimer_get_next_event
2250 hrtimer_interrupt
2251 sys_nanosleep
2252 hrtimer_nanosleep
2253 hrtimer_wakeup
2254 hrtimer_get_remaining
2255 hrtimer_get_res
2256 hrtimer_init_sleeper
2259 The set_ftrace_notrace prevents those functions from being
2260 traced.
2262 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2264 Produces:
2266 # tracer: function
2267 #
2268 # entries-in-buffer/entries-written: 39608/39608 #P:4
2269 #
2270 # _-----=> irqs-off
2271 # / _----=> need-resched
2272 # | / _---=> hardirq/softirq
2273 # || / _--=> preempt-depth
2274 # ||| / delay
2275 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2276 # | | | |||| | |
2277 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2278 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2279 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2280 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2281 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2282 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2283 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2284 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2285 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2286 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2287 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2288 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2290 We can see that there's no more lock or preempt tracing.
2293 Dynamic ftrace with the function graph tracer
2294 ---------------------------------------------
2296 Although what has been explained above concerns both the
2297 function tracer and the function-graph-tracer, there are some
2298 special features only available in the function-graph tracer.
2300 If you want to trace only one function and all of its children,
2301 you just have to echo its name into set_graph_function:
2303 echo __do_fault > set_graph_function
2305 will produce the following "expanded" trace of the __do_fault()
2306 function:
2308 0) | __do_fault() {
2309 0) | filemap_fault() {
2310 0) | find_lock_page() {
2311 0) 0.804 us | find_get_page();
2312 0) | __might_sleep() {
2313 0) 1.329 us | }
2314 0) 3.904 us | }
2315 0) 4.979 us | }
2316 0) 0.653 us | _spin_lock();
2317 0) 0.578 us | page_add_file_rmap();
2318 0) 0.525 us | native_set_pte_at();
2319 0) 0.585 us | _spin_unlock();
2320 0) | unlock_page() {
2321 0) 0.541 us | page_waitqueue();
2322 0) 0.639 us | __wake_up_bit();
2323 0) 2.786 us | }
2324 0) + 14.237 us | }
2325 0) | __do_fault() {
2326 0) | filemap_fault() {
2327 0) | find_lock_page() {
2328 0) 0.698 us | find_get_page();
2329 0) | __might_sleep() {
2330 0) 1.412 us | }
2331 0) 3.950 us | }
2332 0) 5.098 us | }
2333 0) 0.631 us | _spin_lock();
2334 0) 0.571 us | page_add_file_rmap();
2335 0) 0.526 us | native_set_pte_at();
2336 0) 0.586 us | _spin_unlock();
2337 0) | unlock_page() {
2338 0) 0.533 us | page_waitqueue();
2339 0) 0.638 us | __wake_up_bit();
2340 0) 2.793 us | }
2341 0) + 14.012 us | }
2343 You can also expand several functions at once:
2345 echo sys_open > set_graph_function
2346 echo sys_close >> set_graph_function
2348 Now if you want to go back to trace all functions you can clear
2349 this special filter via:
2351 echo > set_graph_function
2354 ftrace_enabled
2355 --------------
2357 Note, the proc sysctl ftrace_enable is a big on/off switch for the
2358 function tracer. By default it is enabled (when function tracing is
2359 enabled in the kernel). If it is disabled, all function tracing is
2360 disabled. This includes not only the function tracers for ftrace, but
2361 also for any other uses (perf, kprobes, stack tracing, profiling, etc).
2363 Please disable this with care.
2365 This can be disable (and enabled) with:
2367 sysctl kernel.ftrace_enabled=0
2368 sysctl kernel.ftrace_enabled=1
2370 or
2372 echo 0 > /proc/sys/kernel/ftrace_enabled
2373 echo 1 > /proc/sys/kernel/ftrace_enabled
2376 Filter commands
2377 ---------------
2379 A few commands are supported by the set_ftrace_filter interface.
2380 Trace commands have the following format:
2382 <function>:<command>:<parameter>
2384 The following commands are supported:
2386 - mod
2387 This command enables function filtering per module. The
2388 parameter defines the module. For example, if only the write*
2389 functions in the ext3 module are desired, run:
2391 echo 'write*:mod:ext3' > set_ftrace_filter
2393 This command interacts with the filter in the same way as
2394 filtering based on function names. Thus, adding more functions
2395 in a different module is accomplished by appending (>>) to the
2396 filter file. Remove specific module functions by prepending
2397 '!':
2399 echo '!writeback*:mod:ext3' >> set_ftrace_filter
2401 - traceon/traceoff
2402 These commands turn tracing on and off when the specified
2403 functions are hit. The parameter determines how many times the
2404 tracing system is turned on and off. If unspecified, there is
2405 no limit. For example, to disable tracing when a schedule bug
2406 is hit the first 5 times, run:
2408 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
2410 To always disable tracing when __schedule_bug is hit:
2412 echo '__schedule_bug:traceoff' > set_ftrace_filter
2414 These commands are cumulative whether or not they are appended
2415 to set_ftrace_filter. To remove a command, prepend it by '!'
2416 and drop the parameter:
2418 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
2420 The above removes the traceoff command for __schedule_bug
2421 that have a counter. To remove commands without counters:
2423 echo '!__schedule_bug:traceoff' > set_ftrace_filter
2425 - snapshot
2426 Will cause a snapshot to be triggered when the function is hit.
2428 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
2430 To only snapshot once:
2432 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
2434 To remove the above commands:
2436 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
2437 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
2439 - enable_event/disable_event
2440 These commands can enable or disable a trace event. Note, because
2441 function tracing callbacks are very sensitive, when these commands
2442 are registered, the trace point is activated, but disabled in
2443 a "soft" mode. That is, the tracepoint will be called, but
2444 just will not be traced. The event tracepoint stays in this mode
2445 as long as there's a command that triggers it.
2447 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
2448 set_ftrace_filter
2450 The format is:
2452 <function>:enable_event:<system>:<event>[:count]
2453 <function>:disable_event:<system>:<event>[:count]
2455 To remove the events commands:
2458 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
2459 set_ftrace_filter
2460 echo '!schedule:disable_event:sched:sched_switch' > \
2461 set_ftrace_filter
2463 - dump
2464 When the function is hit, it will dump the contents of the ftrace
2465 ring buffer to the console. This is useful if you need to debug
2466 something, and want to dump the trace when a certain function
2467 is hit. Perhaps its a function that is called before a tripple
2468 fault happens and does not allow you to get a regular dump.
2470 - cpudump
2471 When the function is hit, it will dump the contents of the ftrace
2472 ring buffer for the current CPU to the console. Unlike the "dump"
2473 command, it only prints out the contents of the ring buffer for the
2474 CPU that executed the function that triggered the dump.
2476 trace_pipe
2477 ----------
2479 The trace_pipe outputs the same content as the trace file, but
2480 the effect on the tracing is different. Every read from
2481 trace_pipe is consumed. This means that subsequent reads will be
2482 different. The trace is live.
2484 # echo function > current_tracer
2485 # cat trace_pipe > /tmp/trace.out &
2486 [1] 4153
2487 # echo 1 > tracing_on
2488 # usleep 1
2489 # echo 0 > tracing_on
2490 # cat trace
2491 # tracer: function
2492 #
2493 # entries-in-buffer/entries-written: 0/0 #P:4
2494 #
2495 # _-----=> irqs-off
2496 # / _----=> need-resched
2497 # | / _---=> hardirq/softirq
2498 # || / _--=> preempt-depth
2499 # ||| / delay
2500 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2501 # | | | |||| | |
2503 #
2504 # cat /tmp/trace.out
2505 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
2506 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
2507 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
2508 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
2509 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
2510 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
2511 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
2512 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
2513 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
2516 Note, reading the trace_pipe file will block until more input is
2517 added.
2519 trace entries
2520 -------------
2522 Having too much or not enough data can be troublesome in
2523 diagnosing an issue in the kernel. The file buffer_size_kb is
2524 used to modify the size of the internal trace buffers. The
2525 number listed is the number of entries that can be recorded per
2526 CPU. To know the full size, multiply the number of possible CPUs
2527 with the number of entries.
2529 # cat buffer_size_kb
2530 1408 (units kilobytes)
2532 Or simply read buffer_total_size_kb
2534 # cat buffer_total_size_kb
2535 5632
2537 To modify the buffer, simple echo in a number (in 1024 byte segments).
2539 # echo 10000 > buffer_size_kb
2540 # cat buffer_size_kb
2541 10000 (units kilobytes)
2543 It will try to allocate as much as possible. If you allocate too
2544 much, it can cause Out-Of-Memory to trigger.
2546 # echo 1000000000000 > buffer_size_kb
2547 -bash: echo: write error: Cannot allocate memory
2548 # cat buffer_size_kb
2549 85
2551 The per_cpu buffers can be changed individually as well:
2553 # echo 10000 > per_cpu/cpu0/buffer_size_kb
2554 # echo 100 > per_cpu/cpu1/buffer_size_kb
2556 When the per_cpu buffers are not the same, the buffer_size_kb
2557 at the top level will just show an X
2559 # cat buffer_size_kb
2560 X
2562 This is where the buffer_total_size_kb is useful:
2564 # cat buffer_total_size_kb
2565 12916
2567 Writing to the top level buffer_size_kb will reset all the buffers
2568 to be the same again.
2570 Snapshot
2571 --------
2572 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
2573 available to all non latency tracers. (Latency tracers which
2574 record max latency, such as "irqsoff" or "wakeup", can't use
2575 this feature, since those are already using the snapshot
2576 mechanism internally.)
2578 Snapshot preserves a current trace buffer at a particular point
2579 in time without stopping tracing. Ftrace swaps the current
2580 buffer with a spare buffer, and tracing continues in the new
2581 current (=previous spare) buffer.
2583 The following debugfs files in "tracing" are related to this
2584 feature:
2586 snapshot:
2588 This is used to take a snapshot and to read the output
2589 of the snapshot. Echo 1 into this file to allocate a
2590 spare buffer and to take a snapshot (swap), then read
2591 the snapshot from this file in the same format as
2592 "trace" (described above in the section "The File
2593 System"). Both reads snapshot and tracing are executable
2594 in parallel. When the spare buffer is allocated, echoing
2595 0 frees it, and echoing else (positive) values clear the
2596 snapshot contents.
2597 More details are shown in the table below.
2599 status\input | 0 | 1 | else |
2600 --------------+------------+------------+------------+
2601 not allocated |(do nothing)| alloc+swap |(do nothing)|
2602 --------------+------------+------------+------------+
2603 allocated | free | swap | clear |
2604 --------------+------------+------------+------------+
2606 Here is an example of using the snapshot feature.
2608 # echo 1 > events/sched/enable
2609 # echo 1 > snapshot
2610 # cat snapshot
2611 # tracer: nop
2612 #
2613 # entries-in-buffer/entries-written: 71/71 #P:8
2614 #
2615 # _-----=> irqs-off
2616 # / _----=> need-resched
2617 # | / _---=> hardirq/softirq
2618 # || / _--=> preempt-depth
2619 # ||| / delay
2620 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2621 # | | | |||| | |
2622 <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
2623 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
2624 [...]
2625 <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
2627 # cat trace
2628 # tracer: nop
2629 #
2630 # entries-in-buffer/entries-written: 77/77 #P:8
2631 #
2632 # _-----=> irqs-off
2633 # / _----=> need-resched
2634 # | / _---=> hardirq/softirq
2635 # || / _--=> preempt-depth
2636 # ||| / delay
2637 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2638 # | | | |||| | |
2639 <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
2640 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
2641 [...]
2644 If you try to use this snapshot feature when current tracer is
2645 one of the latency tracers, you will get the following results.
2647 # echo wakeup > current_tracer
2648 # echo 1 > snapshot
2649 bash: echo: write error: Device or resource busy
2650 # cat snapshot
2651 cat: snapshot: Device or resource busy
2654 Instances
2655 ---------
2656 In the debugfs tracing directory is a directory called "instances".
2657 This directory can have new directories created inside of it using
2658 mkdir, and removing directories with rmdir. The directory created
2659 with mkdir in this directory will already contain files and other
2660 directories after it is created.
2662 # mkdir instances/foo
2663 # ls instances/foo
2664 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
2665 set_event snapshot trace trace_clock trace_marker trace_options
2666 trace_pipe tracing_on
2668 As you can see, the new directory looks similar to the tracing directory
2669 itself. In fact, it is very similar, except that the buffer and
2670 events are agnostic from the main director, or from any other
2671 instances that are created.
2673 The files in the new directory work just like the files with the
2674 same name in the tracing directory except the buffer that is used
2675 is a separate and new buffer. The files affect that buffer but do not
2676 affect the main buffer with the exception of trace_options. Currently,
2677 the trace_options affect all instances and the top level buffer
2678 the same, but this may change in future releases. That is, options
2679 may become specific to the instance they reside in.
2681 Notice that none of the function tracer files are there, nor is
2682 current_tracer and available_tracers. This is because the buffers
2683 can currently only have events enabled for them.
2685 # mkdir instances/foo
2686 # mkdir instances/bar
2687 # mkdir instances/zoot
2688 # echo 100000 > buffer_size_kb
2689 # echo 1000 > instances/foo/buffer_size_kb
2690 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
2691 # echo function > current_trace
2692 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
2693 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
2694 # echo 1 > instances/foo/events/sched/sched_switch/enable
2695 # echo 1 > instances/bar/events/irq/enable
2696 # echo 1 > instances/zoot/events/syscalls/enable
2697 # cat trace_pipe
2698 CPU:2 [LOST 11745 EVENTS]
2699 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
2700 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
2701 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
2702 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
2703 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
2704 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
2705 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
2706 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
2707 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
2708 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
2709 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
2710 [...]
2712 # cat instances/foo/trace_pipe
2713 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
2714 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
2715 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
2716 <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
2717 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
2718 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
2719 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
2720 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
2721 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
2722 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
2723 [...]
2725 # cat instances/bar/trace_pipe
2726 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
2727 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
2728 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
2729 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
2730 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
2731 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
2732 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
2733 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
2734 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
2735 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
2736 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
2737 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
2738 [...]
2740 # cat instances/zoot/trace
2741 # tracer: nop
2742 #
2743 # entries-in-buffer/entries-written: 18996/18996 #P:4
2744 #
2745 # _-----=> irqs-off
2746 # / _----=> need-resched
2747 # | / _---=> hardirq/softirq
2748 # || / _--=> preempt-depth
2749 # ||| / delay
2750 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2751 # | | | |||| | |
2752 bash-1998 [000] d... 140.733501: sys_write -> 0x2
2753 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
2754 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
2755 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
2756 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
2757 bash-1998 [000] d... 140.733510: sys_close(fd: a)
2758 bash-1998 [000] d... 140.733510: sys_close -> 0x0
2759 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
2760 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
2761 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
2762 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
2764 You can see that the trace of the top most trace buffer shows only
2765 the function tracing. The foo instance displays wakeups and task
2766 switches.
2768 To remove the instances, simply delete their directories:
2770 # rmdir instances/foo
2771 # rmdir instances/bar
2772 # rmdir instances/zoot
2774 Note, if a process has a trace file open in one of the instance
2775 directories, the rmdir will fail with EBUSY.
2778 Stack trace
2779 -----------
2780 Since the kernel has a fixed sized stack, it is important not to
2781 waste it in functions. A kernel developer must be conscience of
2782 what they allocate on the stack. If they add too much, the system
2783 can be in danger of a stack overflow, and corruption will occur,
2784 usually leading to a system panic.
2786 There are some tools that check this, usually with interrupts
2787 periodically checking usage. But if you can perform a check
2788 at every function call that will become very useful. As ftrace provides
2789 a function tracer, it makes it convenient to check the stack size
2790 at every function call. This is enabled via the stack tracer.
2792 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
2793 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
2795 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
2797 You can also enable it from the kernel command line to trace
2798 the stack size of the kernel during boot up, by adding "stacktrace"
2799 to the kernel command line parameter.
2801 After running it for a few minutes, the output looks like:
2803 # cat stack_max_size
2804 2928
2806 # cat stack_trace
2807 Depth Size Location (18 entries)
2808 ----- ---- --------
2809 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
2810 1) 2704 160 find_busiest_group+0x31/0x1f1
2811 2) 2544 256 load_balance+0xd9/0x662
2812 3) 2288 80 idle_balance+0xbb/0x130
2813 4) 2208 128 __schedule+0x26e/0x5b9
2814 5) 2080 16 schedule+0x64/0x66
2815 6) 2064 128 schedule_timeout+0x34/0xe0
2816 7) 1936 112 wait_for_common+0x97/0xf1
2817 8) 1824 16 wait_for_completion+0x1d/0x1f
2818 9) 1808 128 flush_work+0xfe/0x119
2819 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
2820 11) 1664 48 input_available_p+0x1d/0x5c
2821 12) 1616 48 n_tty_poll+0x6d/0x134
2822 13) 1568 64 tty_poll+0x64/0x7f
2823 14) 1504 880 do_select+0x31e/0x511
2824 15) 624 400 core_sys_select+0x177/0x216
2825 16) 224 96 sys_select+0x91/0xb9
2826 17) 128 128 system_call_fastpath+0x16/0x1b
2828 Note, if -mfentry is being used by gcc, functions get traced before
2829 they set up the stack frame. This means that leaf level functions
2830 are not tested by the stack tracer when -mfentry is used.
2832 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
2834 ---------
2836 More details can be found in the source code, in the
2837 kernel/trace/*.c files.