1 ftrace - Function Tracer
2 ========================
4 Copyright 2008 Red Hat Inc.
5 Author: Steven Rostedt <srostedt@redhat.com>
6 License: The GNU Free Documentation License, Version 1.2
7 (dual licensed under the GPL v2)
8 Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton,
9 John Kacur, and David Teigland.
10 Written for: 2.6.28-rc2
16 Ftrace is an internal tracer designed to help out developers and
17 designers of systems to find what is going on inside the kernel.
18 It can be used for debugging or analyzing latencies and
19 performance issues that take place outside of user-space.
21 Although ftrace is typically considered the function tracer, it
22 is really a frame work of several assorted tracing utilities.
23 There's latency tracing to examine what occurs between interrupts
24 disabled and enabled, as well as for preemption and from a time
25 a task is woken to the task is actually scheduled in.
27 One of the most common uses of ftrace is the event tracing.
28 Through out the kernel is hundreds of static event points that
29 can be enabled via the debugfs file system to see what is
30 going on in certain parts of the kernel.
33 Implementation Details
34 ----------------------
36 See ftrace-design.txt for details for arch porters and such.
42 Ftrace uses the debugfs file system to hold the control files as
43 well as the files to display output.
45 When debugfs is configured into the kernel (which selecting any ftrace
46 option will do) the directory /sys/kernel/debug will be created. To mount
47 this directory, you can add to your /etc/fstab file:
49 debugfs /sys/kernel/debug debugfs defaults 0 0
51 Or you can mount it at run time with:
53 mount -t debugfs nodev /sys/kernel/debug
55 For quicker access to that directory you may want to make a soft link to
58 ln -s /sys/kernel/debug /debug
60 Any selected ftrace option will also create a directory called tracing
61 within the debugfs. The rest of the document will assume that you are in
62 the ftrace directory (cd /sys/kernel/debug/tracing) and will only concentrate
63 on the files within that directory and not distract from the content with
64 the extended "/sys/kernel/debug/tracing" path name.
66 That's it! (assuming that you have ftrace configured into your kernel)
68 After mounting debugfs, you can see a directory called
69 "tracing". This directory contains the control and output files
70 of ftrace. Here is a list of some of the key files:
73 Note: all time values are in microseconds.
77 This is used to set or display the current tracer
82 This holds the different types of tracers that
83 have been compiled into the kernel. The
84 tracers listed here can be configured by
85 echoing their name into current_tracer.
89 This sets or displays whether writing to the trace
90 ring buffer is enabled. Echo 0 into this file to disable
91 the tracer or 1 to enable it. Note, this only disables
92 writing to the ring buffer, the tracing overhead may
97 This file holds the output of the trace in a human
98 readable format (described below).
102 The output is the same as the "trace" file but this
103 file is meant to be streamed with live tracing.
104 Reads from this file will block until new data is
105 retrieved. Unlike the "trace" file, this file is a
106 consumer. This means reading from this file causes
107 sequential reads to display more current data. Once
108 data is read from this file, it is consumed, and
109 will not be read again with a sequential read. The
110 "trace" file is static, and if the tracer is not
111 adding more data, it will display the same
112 information every time it is read.
116 This file lets the user control the amount of data
117 that is displayed in one of the above output
118 files. Options also exist to modify how a tracer
119 or events work (stack traces, timestamps, etc).
123 This is a directory that has a file for every available
124 trace option (also in trace_options). Options may also be set
125 or cleared by writing a "1" or "0" respectively into the
126 corresponding file with the option name.
130 Some of the tracers record the max latency.
131 For example, the time interrupts are disabled.
132 This time is saved in this file. The max trace
133 will also be stored, and displayed by "trace".
134 A new max trace will only be recorded if the
135 latency is greater than the value in this
136 file. (in microseconds)
140 Some latency tracers will record a trace whenever the
141 latency is greater than the number in this file.
142 Only active when the file contains a number greater than 0.
147 This sets or displays the number of kilobytes each CPU
148 buffer holds. By default, the trace buffers are the same size
149 for each CPU. The displayed number is the size of the
150 CPU buffer and not total size of all buffers. The
151 trace buffers are allocated in pages (blocks of memory
152 that the kernel uses for allocation, usually 4 KB in size).
153 If the last page allocated has room for more bytes
154 than requested, the rest of the page will be used,
155 making the actual allocation bigger than requested.
156 ( Note, the size may not be a multiple of the page size
157 due to buffer management meta-data. )
159 buffer_total_size_kb:
161 This displays the total combined size of all the trace buffers.
165 If a process is performing the tracing, and the ring buffer
166 should be shrunk "freed" when the process is finished, even
167 if it were to be killed by a signal, this file can be used
168 for that purpose. On close of this file, the ring buffer will
169 be resized to its minimum size. Having a process that is tracing
170 also open this file, when the process exits its file descriptor
171 for this file will be closed, and in doing so, the ring buffer
174 It may also stop tracing if disable_on_free option is set.
178 This is a mask that lets the user only trace
179 on specified CPUs. The format is a hex string
180 representing the CPUs.
184 When dynamic ftrace is configured in (see the
185 section below "dynamic ftrace"), the code is dynamically
186 modified (code text rewrite) to disable calling of the
187 function profiler (mcount). This lets tracing be configured
188 in with practically no overhead in performance. This also
189 has a side effect of enabling or disabling specific functions
190 to be traced. Echoing names of functions into this file
191 will limit the trace to only those functions.
193 This interface also allows for commands to be used. See the
194 "Filter commands" section for more details.
198 This has an effect opposite to that of
199 set_ftrace_filter. Any function that is added here will not
200 be traced. If a function exists in both set_ftrace_filter
201 and set_ftrace_notrace, the function will _not_ be traced.
205 Have the function tracer only trace a single thread.
209 Have the events only trace a task with a PID listed in this file.
210 Note, sched_switch and sched_wake_up will also trace events
215 Set a "trigger" function where tracing should start
216 with the function graph tracer (See the section
217 "dynamic ftrace" for more details).
219 available_filter_functions:
221 This lists the functions that ftrace
222 has processed and can trace. These are the function
223 names that you can pass to "set_ftrace_filter" or
224 "set_ftrace_notrace". (See the section "dynamic ftrace"
225 below for more details.)
229 This file is more for debugging ftrace, but can also be useful
230 in seeing if any function has a callback attached to it.
231 Not only does the trace infrastructure use ftrace function
232 trace utility, but other subsystems might too. This file
233 displays all functions that have a callback attached to them
234 as well as the number of callbacks that have been attached.
235 Note, a callback may also call multiple functions which will
236 not be listed in this count.
238 If the callback registered to be traced by a function with
239 the "save regs" attribute (thus even more overhead), a 'R'
240 will be displayed on the same line as the function that
241 is returning registers.
243 If the callback registered to be traced by a function with
244 the "ip modify" attribute (thus the regs->ip can be changed),
245 an 'I' will be displayed on the same line as the function that
248 function_profile_enabled:
250 When set it will enable all functions with either the function
251 tracer, or if enabled, the function graph tracer. It will
252 keep a histogram of the number of functions that were called
253 and if run with the function graph tracer, it will also keep
254 track of the time spent in those functions. The histogram
255 content can be displayed in the files:
257 trace_stats/function<cpu> ( function0, function1, etc).
261 A directory that holds different tracing stats.
265 Enable dynamic trace points. See kprobetrace.txt.
269 Dynamic trace points stats. See kprobetrace.txt.
273 Used with the function graph tracer. This is the max depth
274 it will trace into a function. Setting this to a value of
275 one will show only the first kernel function that is called
280 This is for tools that read the raw format files. If an event in
281 the ring buffer references a string (currently only trace_printk()
282 does this), only a pointer to the string is recorded into the buffer
283 and not the string itself. This prevents tools from knowing what
284 that string was. This file displays the string and address for
285 the string allowing tools to map the pointers to what the
290 Only the pid of the task is recorded in a trace event unless
291 the event specifically saves the task comm as well. Ftrace
292 makes a cache of pid mappings to comms to try to display
293 comms for events. If a pid for a comm is not listed, then
294 "<...>" is displayed in the output.
298 This displays the "snapshot" buffer and also lets the user
299 take a snapshot of the current running trace.
300 See the "Snapshot" section below for more details.
304 When the stack tracer is activated, this will display the
305 maximum stack size it has encountered.
306 See the "Stack Trace" section below.
310 This displays the stack back trace of the largest stack
311 that was encountered when the stack tracer is activated.
312 See the "Stack Trace" section below.
316 This is similar to "set_ftrace_filter" but it limits what
317 functions the stack tracer will check.
321 Whenever an event is recorded into the ring buffer, a
322 "timestamp" is added. This stamp comes from a specified
323 clock. By default, ftrace uses the "local" clock. This
324 clock is very fast and strictly per cpu, but on some
325 systems it may not be monotonic with respect to other
326 CPUs. In other words, the local clocks may not be in sync
327 with local clocks on other CPUs.
329 Usual clocks for tracing:
332 [local] global counter x86-tsc
334 local: Default clock, but may not be in sync across CPUs
336 global: This clock is in sync with all CPUs but may
337 be a bit slower than the local clock.
339 counter: This is not a clock at all, but literally an atomic
340 counter. It counts up one by one, but is in sync
341 with all CPUs. This is useful when you need to
342 know exactly the order events occurred with respect to
343 each other on different CPUs.
345 uptime: This uses the jiffies counter and the time stamp
346 is relative to the time since boot up.
348 perf: This makes ftrace use the same clock that perf uses.
349 Eventually perf will be able to read ftrace buffers
350 and this will help out in interleaving the data.
352 x86-tsc: Architectures may define their own clocks. For
353 example, x86 uses its own TSC cycle clock here.
355 ppc-tb: This uses the powerpc timebase register value.
356 This is in sync across CPUs and can also be used
357 to correlate events across hypervisor/guest if
360 mono: This uses the fast monotonic clock (CLOCK_MONOTONIC)
361 which is monotonic and is subject to NTP rate adjustments.
364 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)
365 which is montonic but is not subject to any rate adjustments
366 and ticks at the same rate as the hardware clocksource.
368 boot: This is the boot clock (CLOCK_BOOTTIME) and is based on the
369 fast monotonic clock, but also accounts for time spent in
370 suspend. Since the clock access is designed for use in
371 tracing in the suspend path, some side effects are possible
372 if clock is accessed after the suspend time is accounted before
373 the fast mono clock is updated. In this case, the clock update
374 appears to happen slightly sooner than it normally would have.
375 Also on 32-bit systems, its possible that the 64-bit boot offset
376 sees a partial update. These effects are rare and post
377 processing should be able to handle them. See comments on
378 ktime_get_boot_fast_ns function for more information.
380 To set a clock, simply echo the clock name into this file.
382 echo global > trace_clock
386 This is a very useful file for synchronizing user space
387 with events happening in the kernel. Writing strings into
388 this file will be written into the ftrace buffer.
390 It is useful in applications to open this file at the start
391 of the application and just reference the file descriptor
394 void trace_write(const char *fmt, ...)
404 n = vsnprintf(buf, 256, fmt, ap);
407 write(trace_fd, buf, n);
412 trace_fd = open("trace_marker", WR_ONLY);
416 Add dynamic tracepoints in programs.
421 Uprobe statistics. See uprobetrace.txt
425 This is a way to make multiple trace buffers where different
426 events can be recorded in different buffers.
427 See "Instances" section below.
431 This is the trace event directory. It holds event tracepoints
432 (also known as static tracepoints) that have been compiled
433 into the kernel. It shows what event tracepoints exist
434 and how they are grouped by system. There are "enable"
435 files at various levels that can enable the tracepoints
436 when a "1" is written to them.
438 See events.txt for more information.
442 This is a directory that contains the trace per_cpu information.
444 per_cpu/cpu0/buffer_size_kb:
446 The ftrace buffer is defined per_cpu. That is, there's a separate
447 buffer for each CPU to allow writes to be done atomically,
448 and free from cache bouncing. These buffers may have different
449 size buffers. This file is similar to the buffer_size_kb
450 file, but it only displays or sets the buffer size for the
451 specific CPU. (here cpu0).
455 This is similar to the "trace" file, but it will only display
456 the data specific for the CPU. If written to, it only clears
457 the specific CPU buffer.
459 per_cpu/cpu0/trace_pipe
461 This is similar to the "trace_pipe" file, and is a consuming
462 read, but it will only display (and consume) the data specific
465 per_cpu/cpu0/trace_pipe_raw
467 For tools that can parse the ftrace ring buffer binary format,
468 the trace_pipe_raw file can be used to extract the data
469 from the ring buffer directly. With the use of the splice()
470 system call, the buffer data can be quickly transferred to
471 a file or to the network where a server is collecting the
474 Like trace_pipe, this is a consuming reader, where multiple
475 reads will always produce different data.
477 per_cpu/cpu0/snapshot:
479 This is similar to the main "snapshot" file, but will only
480 snapshot the current CPU (if supported). It only displays
481 the content of the snapshot for a given CPU, and if
482 written to, only clears this CPU buffer.
484 per_cpu/cpu0/snapshot_raw:
486 Similar to the trace_pipe_raw, but will read the binary format
487 from the snapshot buffer for the given CPU.
491 This displays certain stats about the ring buffer:
493 entries: The number of events that are still in the buffer.
495 overrun: The number of lost events due to overwriting when
498 commit overrun: Should always be zero.
499 This gets set if so many events happened within a nested
500 event (ring buffer is re-entrant), that it fills the
501 buffer and starts dropping events.
503 bytes: Bytes actually read (not overwritten).
505 oldest event ts: The oldest timestamp in the buffer
507 now ts: The current timestamp
509 dropped events: Events lost due to overwrite option being off.
511 read events: The number of events read.
516 Here is the list of current tracers that may be configured.
520 Function call tracer to trace all kernel functions.
524 Similar to the function tracer except that the
525 function tracer probes the functions on their entry
526 whereas the function graph tracer traces on both entry
527 and exit of the functions. It then provides the ability
528 to draw a graph of function calls similar to C code
533 Traces the areas that disable interrupts and saves
534 the trace with the longest max latency.
535 See tracing_max_latency. When a new max is recorded,
536 it replaces the old trace. It is best to view this
537 trace with the latency-format option enabled.
541 Similar to irqsoff but traces and records the amount of
542 time for which preemption is disabled.
546 Similar to irqsoff and preemptoff, but traces and
547 records the largest time for which irqs and/or preemption
552 Traces and records the max latency that it takes for
553 the highest priority task to get scheduled after
554 it has been woken up.
555 Traces all tasks as an average developer would expect.
559 Traces and records the max latency that it takes for just
560 RT tasks (as the current "wakeup" does). This is useful
561 for those interested in wake up timings of RT tasks.
565 This is the "trace nothing" tracer. To remove all
566 tracers from tracing simply echo "nop" into
570 Examples of using the tracer
571 ----------------------------
573 Here are typical examples of using the tracers when controlling
574 them only with the debugfs interface (without using any
575 user-land utilities).
580 Here is an example of the output format of the file "trace"
585 # entries-in-buffer/entries-written: 140080/250280 #P:4
588 # / _----=> need-resched
589 # | / _---=> hardirq/softirq
590 # || / _--=> preempt-depth
592 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
594 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
595 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
596 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
597 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
598 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
599 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
600 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
601 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
602 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
603 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
606 A header is printed with the tracer name that is represented by
607 the trace. In this case the tracer is "function". Then it shows the
608 number of events in the buffer as well as the total number of entries
609 that were written. The difference is the number of entries that were
610 lost due to the buffer filling up (250280 - 140080 = 110200 events
613 The header explains the content of the events. Task name "bash", the task
614 PID "1977", the CPU that it was running on "000", the latency format
615 (explained below), the timestamp in <secs>.<usecs> format, the
616 function name that was traced "sys_close" and the parent function that
617 called this function "system_call_fastpath". The timestamp is the time
618 at which the function was entered.
623 When the latency-format option is enabled or when one of the latency
624 tracers is set, the trace file gives somewhat more information to see
625 why a latency happened. Here is a typical trace.
629 # irqsoff latency trace v1.1.5 on 3.8.0-test+
630 # --------------------------------------------------------------------
631 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
633 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
635 # => started at: __lock_task_sighand
636 # => ended at: _raw_spin_unlock_irqrestore
640 # / _-----=> irqs-off
641 # | / _----=> need-resched
642 # || / _---=> hardirq/softirq
643 # ||| / _--=> preempt-depth
645 # cmd pid ||||| time | caller
647 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
648 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
649 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
650 ps-6143 2d..1 306us : <stack trace>
651 => trace_hardirqs_on_caller
653 => _raw_spin_unlock_irqrestore
660 => system_call_fastpath
663 This shows that the current tracer is "irqsoff" tracing the time
664 for which interrupts were disabled. It gives the trace version (which
665 never changes) and the version of the kernel upon which this was executed on
666 (3.10). Then it displays the max latency in microseconds (259 us). The number
667 of trace entries displayed and the total number (both are four: #4/4).
668 VP, KP, SP, and HP are always zero and are reserved for later use.
669 #P is the number of online CPUs (#P:4).
671 The task is the process that was running when the latency
672 occurred. (ps pid: 6143).
674 The start and stop (the functions in which the interrupts were
675 disabled and enabled respectively) that caused the latencies:
677 __lock_task_sighand is where the interrupts were disabled.
678 _raw_spin_unlock_irqrestore is where they were enabled again.
680 The next lines after the header are the trace itself. The header
681 explains which is which.
683 cmd: The name of the process in the trace.
685 pid: The PID of that process.
687 CPU#: The CPU which the process was running on.
689 irqs-off: 'd' interrupts are disabled. '.' otherwise.
690 Note: If the architecture does not support a way to
691 read the irq flags variable, an 'X' will always
695 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
696 'n' only TIF_NEED_RESCHED is set,
697 'p' only PREEMPT_NEED_RESCHED is set,
701 'H' - hard irq occurred inside a softirq.
702 'h' - hard irq is running
703 's' - soft irq is running
704 '.' - normal context.
706 preempt-depth: The level of preempt_disabled
708 The above is mostly meaningful for kernel developers.
710 time: When the latency-format option is enabled, the trace file
711 output includes a timestamp relative to the start of the
712 trace. This differs from the output when latency-format
713 is disabled, which includes an absolute timestamp.
715 delay: This is just to help catch your eye a bit better. And
716 needs to be fixed to be only relative to the same CPU.
717 The marks are determined by the difference between this
718 current trace and the next trace.
719 '$' - greater than 1 second
720 '@' - greater than 100 milisecond
721 '*' - greater than 10 milisecond
722 '#' - greater than 1000 microsecond
723 '!' - greater than 100 microsecond
724 '+' - greater than 10 microsecond
725 ' ' - less than or equal to 10 microsecond.
727 The rest is the same as the 'trace' file.
729 Note, the latency tracers will usually end with a back trace
730 to easily find where the latency occurred.
735 The trace_options file (or the options directory) is used to control
736 what gets printed in the trace output, or manipulate the tracers.
737 To see what is available, simply cat the file:
767 To disable one of the options, echo in the option prepended with
770 echo noprint-parent > trace_options
772 To enable an option, leave off the "no".
774 echo sym-offset > trace_options
776 Here are the available options:
778 print-parent - On function traces, display the calling (parent)
779 function as well as the function being traced.
782 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
785 bash-4000 [01] 1477.606694: simple_strtoul
788 sym-offset - Display not only the function name, but also the
789 offset in the function. For example, instead of
790 seeing just "ktime_get", you will see
791 "ktime_get+0xb/0x20".
794 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
796 sym-addr - this will also display the function address as well
797 as the function name.
800 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
802 verbose - This deals with the trace file when the
803 latency-format option is enabled.
805 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
806 (+0.000ms): simple_strtoul (kstrtoul)
808 raw - This will display raw numbers. This option is best for
809 use with user applications that can translate the raw
810 numbers better than having it done in the kernel.
812 hex - Similar to raw, but the numbers will be in a hexadecimal
815 bin - This will print out the formats in raw binary.
817 block - When set, reading trace_pipe will not block when polled.
819 stacktrace - This is one of the options that changes the trace
820 itself. When a trace is recorded, so is the stack
821 of functions. This allows for back traces of
824 trace_printk - Can disable trace_printk() from writing into the buffer.
826 branch - Enable branch tracing with the tracer.
828 annotate - It is sometimes confusing when the CPU buffers are full
829 and one CPU buffer had a lot of events recently, thus
830 a shorter time frame, were another CPU may have only had
831 a few events, which lets it have older events. When
832 the trace is reported, it shows the oldest events first,
833 and it may look like only one CPU ran (the one with the
834 oldest events). When the annotate option is set, it will
835 display when a new CPU buffer started:
837 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
838 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
839 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
840 ##### CPU 2 buffer started ####
841 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
842 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
843 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
845 userstacktrace - This option changes the trace. It records a
846 stacktrace of the current userspace thread.
848 sym-userobj - when user stacktrace are enabled, look up which
849 object the address belongs to, and print a
850 relative address. This is especially useful when
851 ASLR is on, otherwise you don't get a chance to
852 resolve the address to object/file/line after
853 the app is no longer running
855 The lookup is performed when you read
856 trace,trace_pipe. Example:
858 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
859 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
862 printk-msg-only - When set, trace_printk()s will only show the format
863 and not their parameters (if trace_bprintk() or
864 trace_bputs() was used to save the trace_printk()).
866 context-info - Show only the event data. Hides the comm, PID,
867 timestamp, CPU, and other useful data.
869 latency-format - This option changes the trace. When
870 it is enabled, the trace displays
871 additional information about the
872 latencies, as described in "Latency
875 sleep-time - When running function graph tracer, to include
876 the time a task schedules out in its function.
877 When enabled, it will account time the task has been
878 scheduled out as part of the function call.
880 graph-time - When running function graph tracer, to include the
881 time to call nested functions. When this is not set,
882 the time reported for the function will only include
883 the time the function itself executed for, not the time
884 for functions that it called.
886 record-cmd - When any event or tracer is enabled, a hook is enabled
887 in the sched_switch trace point to fill comm cache
888 with mapped pids and comms. But this may cause some
889 overhead, and if you only care about pids, and not the
890 name of the task, disabling this option can lower the
893 overwrite - This controls what happens when the trace buffer is
894 full. If "1" (default), the oldest events are
895 discarded and overwritten. If "0", then the newest
896 events are discarded.
897 (see per_cpu/cpu0/stats for overrun and dropped)
899 disable_on_free - When the free_buffer is closed, tracing will
900 stop (tracing_on set to 0).
902 irq-info - Shows the interrupt, preempt count, need resched data.
903 When disabled, the trace looks like:
907 # entries-in-buffer/entries-written: 144405/9452052 #P:4
909 # TASK-PID CPU# TIMESTAMP FUNCTION
911 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
912 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
913 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
916 markers - When set, the trace_marker is writable (only by root).
917 When disabled, the trace_marker will error with EINVAL
921 function-trace - The latency tracers will enable function tracing
922 if this option is enabled (default it is). When
923 it is disabled, the latency tracers do not trace
924 functions. This keeps the overhead of the tracer down
925 when performing latency tests.
927 Note: Some tracers have their own options. They only appear
928 when the tracer is active.
935 When interrupts are disabled, the CPU can not react to any other
936 external event (besides NMIs and SMIs). This prevents the timer
937 interrupt from triggering or the mouse interrupt from letting
938 the kernel know of a new mouse event. The result is a latency
939 with the reaction time.
941 The irqsoff tracer tracks the time for which interrupts are
942 disabled. When a new maximum latency is hit, the tracer saves
943 the trace leading up to that latency point so that every time a
944 new maximum is reached, the old saved trace is discarded and the
947 To reset the maximum, echo 0 into tracing_max_latency. Here is
950 # echo 0 > options/function-trace
951 # echo irqsoff > current_tracer
952 # echo 1 > tracing_on
953 # echo 0 > tracing_max_latency
956 # echo 0 > tracing_on
960 # irqsoff latency trace v1.1.5 on 3.8.0-test+
961 # --------------------------------------------------------------------
962 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
964 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
966 # => started at: run_timer_softirq
967 # => ended at: run_timer_softirq
971 # / _-----=> irqs-off
972 # | / _----=> need-resched
973 # || / _---=> hardirq/softirq
974 # ||| / _--=> preempt-depth
976 # cmd pid ||||| time | caller
978 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
979 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
980 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
981 <idle>-0 0dNs3 25us : <stack trace>
982 => _raw_spin_unlock_irq
988 => smp_apic_timer_interrupt
989 => apic_timer_interrupt
994 => x86_64_start_reservations
995 => x86_64_start_kernel
997 Here we see that that we had a latency of 16 microseconds (which is
998 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
999 interrupts. The difference between the 16 and the displayed
1000 timestamp 25us occurred because the clock was incremented
1001 between the time of recording the max latency and the time of
1002 recording the function that had that latency.
1004 Note the above example had function-trace not set. If we set
1005 function-trace, we get a much larger output:
1007 with echo 1 > options/function-trace
1011 # irqsoff latency trace v1.1.5 on 3.8.0-test+
1012 # --------------------------------------------------------------------
1013 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1015 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
1017 # => started at: ata_scsi_queuecmd
1018 # => ended at: ata_scsi_queuecmd
1022 # / _-----=> irqs-off
1023 # | / _----=> need-resched
1024 # || / _---=> hardirq/softirq
1025 # ||| / _--=> preempt-depth
1027 # cmd pid ||||| time | caller
1029 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1030 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1031 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1032 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1033 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1034 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1035 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1036 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1037 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1039 bash-2042 3d..1 67us : delay_tsc <-__delay
1040 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1041 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1042 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1043 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1044 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1045 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1046 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1047 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1048 bash-2042 3d..1 120us : <stack trace>
1049 => _raw_spin_unlock_irqrestore
1050 => ata_scsi_queuecmd
1051 => scsi_dispatch_cmd
1053 => __blk_run_queue_uncond
1056 => generic_make_request
1059 => __ext3_get_inode_loc
1068 => user_path_at_empty
1073 => system_call_fastpath
1076 Here we traced a 71 microsecond latency. But we also see all the
1077 functions that were called during that time. Note that by
1078 enabling function tracing, we incur an added overhead. This
1079 overhead may extend the latency times. But nevertheless, this
1080 trace has provided some very helpful debugging information.
1086 When preemption is disabled, we may be able to receive
1087 interrupts but the task cannot be preempted and a higher
1088 priority task must wait for preemption to be enabled again
1089 before it can preempt a lower priority task.
1091 The preemptoff tracer traces the places that disable preemption.
1092 Like the irqsoff tracer, it records the maximum latency for
1093 which preemption was disabled. The control of preemptoff tracer
1094 is much like the irqsoff tracer.
1096 # echo 0 > options/function-trace
1097 # echo preemptoff > current_tracer
1098 # echo 1 > tracing_on
1099 # echo 0 > tracing_max_latency
1102 # echo 0 > tracing_on
1104 # tracer: preemptoff
1106 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1107 # --------------------------------------------------------------------
1108 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1110 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1112 # => started at: do_IRQ
1113 # => ended at: do_IRQ
1117 # / _-----=> irqs-off
1118 # | / _----=> need-resched
1119 # || / _---=> hardirq/softirq
1120 # ||| / _--=> preempt-depth
1122 # cmd pid ||||| time | caller
1124 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1125 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1126 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1127 sshd-1991 1d..1 52us : <stack trace>
1128 => sub_preempt_count
1134 This has some more changes. Preemption was disabled when an
1135 interrupt came in (notice the 'h'), and was enabled on exit.
1136 But we also see that interrupts have been disabled when entering
1137 the preempt off section and leaving it (the 'd'). We do not know if
1138 interrupts were enabled in the mean time or shortly after this
1141 # tracer: preemptoff
1143 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1144 # --------------------------------------------------------------------
1145 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1147 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1149 # => started at: wake_up_new_task
1150 # => ended at: task_rq_unlock
1154 # / _-----=> irqs-off
1155 # | / _----=> need-resched
1156 # || / _---=> hardirq/softirq
1157 # ||| / _--=> preempt-depth
1159 # cmd pid ||||| time | caller
1161 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1162 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1163 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1164 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1165 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1167 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1168 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1169 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1170 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1171 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1172 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1173 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1174 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1176 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1177 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1178 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1179 bash-1994 1d..2 36us : do_softirq <-irq_exit
1180 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1181 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1182 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1183 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1184 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1185 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1187 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1188 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1189 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1190 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1191 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1192 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1193 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1194 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1195 bash-1994 1.N.1 104us : <stack trace>
1196 => sub_preempt_count
1197 => _raw_spin_unlock_irqrestore
1205 The above is an example of the preemptoff trace with
1206 function-trace set. Here we see that interrupts were not disabled
1207 the entire time. The irq_enter code lets us know that we entered
1208 an interrupt 'h'. Before that, the functions being traced still
1209 show that it is not in an interrupt, but we can see from the
1210 functions themselves that this is not the case.
1215 Knowing the locations that have interrupts disabled or
1216 preemption disabled for the longest times is helpful. But
1217 sometimes we would like to know when either preemption and/or
1218 interrupts are disabled.
1220 Consider the following code:
1222 local_irq_disable();
1223 call_function_with_irqs_off();
1225 call_function_with_irqs_and_preemption_off();
1227 call_function_with_preemption_off();
1230 The irqsoff tracer will record the total length of
1231 call_function_with_irqs_off() and
1232 call_function_with_irqs_and_preemption_off().
1234 The preemptoff tracer will record the total length of
1235 call_function_with_irqs_and_preemption_off() and
1236 call_function_with_preemption_off().
1238 But neither will trace the time that interrupts and/or
1239 preemption is disabled. This total time is the time that we can
1240 not schedule. To record this time, use the preemptirqsoff
1243 Again, using this trace is much like the irqsoff and preemptoff
1246 # echo 0 > options/function-trace
1247 # echo preemptirqsoff > current_tracer
1248 # echo 1 > tracing_on
1249 # echo 0 > tracing_max_latency
1252 # echo 0 > tracing_on
1254 # tracer: preemptirqsoff
1256 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1257 # --------------------------------------------------------------------
1258 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1260 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1262 # => started at: ata_scsi_queuecmd
1263 # => ended at: ata_scsi_queuecmd
1267 # / _-----=> irqs-off
1268 # | / _----=> need-resched
1269 # || / _---=> hardirq/softirq
1270 # ||| / _--=> preempt-depth
1272 # cmd pid ||||| time | caller
1274 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1275 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1276 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1277 ls-2230 3...1 111us : <stack trace>
1278 => sub_preempt_count
1279 => _raw_spin_unlock_irqrestore
1280 => ata_scsi_queuecmd
1281 => scsi_dispatch_cmd
1283 => __blk_run_queue_uncond
1286 => generic_make_request
1291 => htree_dirblock_to_tree
1292 => ext3_htree_fill_tree
1296 => system_call_fastpath
1299 The trace_hardirqs_off_thunk is called from assembly on x86 when
1300 interrupts are disabled in the assembly code. Without the
1301 function tracing, we do not know if interrupts were enabled
1302 within the preemption points. We do see that it started with
1305 Here is a trace with function-trace set:
1307 # tracer: preemptirqsoff
1309 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1310 # --------------------------------------------------------------------
1311 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1313 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1315 # => started at: schedule
1316 # => ended at: mutex_unlock
1320 # / _-----=> irqs-off
1321 # | / _----=> need-resched
1322 # || / _---=> hardirq/softirq
1323 # ||| / _--=> preempt-depth
1325 # cmd pid ||||| time | caller
1327 kworker/-59 3...1 0us : __schedule <-schedule
1328 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1329 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1330 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1331 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1332 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1333 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1334 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1335 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1336 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1337 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1338 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1339 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1340 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1341 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1342 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1343 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1344 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1345 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1346 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1347 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1348 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1349 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1350 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1351 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1352 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1353 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1354 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1355 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1356 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1357 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1358 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1360 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1361 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1362 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1363 ls-2269 3d..3 21us : do_softirq <-irq_exit
1364 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1365 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1366 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1367 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1368 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1369 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1370 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1372 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1373 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1374 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1375 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1376 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1377 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1379 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1380 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1381 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1382 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1383 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1384 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1385 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1386 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1387 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1388 ls-2269 3d... 186us : <stack trace>
1389 => __mutex_unlock_slowpath
1396 => system_call_fastpath
1398 This is an interesting trace. It started with kworker running and
1399 scheduling out and ls taking over. But as soon as ls released the
1400 rq lock and enabled interrupts (but not preemption) an interrupt
1401 triggered. When the interrupt finished, it started running softirqs.
1402 But while the softirq was running, another interrupt triggered.
1403 When an interrupt is running inside a softirq, the annotation is 'H'.
1409 One common case that people are interested in tracing is the
1410 time it takes for a task that is woken to actually wake up.
1411 Now for non Real-Time tasks, this can be arbitrary. But tracing
1412 it none the less can be interesting.
1414 Without function tracing:
1416 # echo 0 > options/function-trace
1417 # echo wakeup > current_tracer
1418 # echo 1 > tracing_on
1419 # echo 0 > tracing_max_latency
1421 # echo 0 > tracing_on
1425 # wakeup latency trace v1.1.5 on 3.8.0-test+
1426 # --------------------------------------------------------------------
1427 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1429 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1433 # / _-----=> irqs-off
1434 # | / _----=> need-resched
1435 # || / _---=> hardirq/softirq
1436 # ||| / _--=> preempt-depth
1438 # cmd pid ||||| time | caller
1440 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1441 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1442 <idle>-0 3d..3 15us : __schedule <-schedule
1443 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1445 The tracer only traces the highest priority task in the system
1446 to avoid tracing the normal circumstances. Here we see that
1447 the kworker with a nice priority of -20 (not very nice), took
1448 just 15 microseconds from the time it woke up, to the time it
1451 Non Real-Time tasks are not that interesting. A more interesting
1452 trace is to concentrate only on Real-Time tasks.
1457 In a Real-Time environment it is very important to know the
1458 wakeup time it takes for the highest priority task that is woken
1459 up to the time that it executes. This is also known as "schedule
1460 latency". I stress the point that this is about RT tasks. It is
1461 also important to know the scheduling latency of non-RT tasks,
1462 but the average schedule latency is better for non-RT tasks.
1463 Tools like LatencyTop are more appropriate for such
1466 Real-Time environments are interested in the worst case latency.
1467 That is the longest latency it takes for something to happen,
1468 and not the average. We can have a very fast scheduler that may
1469 only have a large latency once in a while, but that would not
1470 work well with Real-Time tasks. The wakeup_rt tracer was designed
1471 to record the worst case wakeups of RT tasks. Non-RT tasks are
1472 not recorded because the tracer only records one worst case and
1473 tracing non-RT tasks that are unpredictable will overwrite the
1474 worst case latency of RT tasks (just run the normal wakeup
1475 tracer for a while to see that effect).
1477 Since this tracer only deals with RT tasks, we will run this
1478 slightly differently than we did with the previous tracers.
1479 Instead of performing an 'ls', we will run 'sleep 1' under
1480 'chrt' which changes the priority of the task.
1482 # echo 0 > options/function-trace
1483 # echo wakeup_rt > current_tracer
1484 # echo 1 > tracing_on
1485 # echo 0 > tracing_max_latency
1487 # echo 0 > tracing_on
1493 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1494 # --------------------------------------------------------------------
1495 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1497 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1501 # / _-----=> irqs-off
1502 # | / _----=> need-resched
1503 # || / _---=> hardirq/softirq
1504 # ||| / _--=> preempt-depth
1506 # cmd pid ||||| time | caller
1508 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1509 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1510 <idle>-0 3d..3 5us : __schedule <-schedule
1511 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1514 Running this on an idle system, we see that it only took 5 microseconds
1515 to perform the task switch. Note, since the trace point in the schedule
1516 is before the actual "switch", we stop the tracing when the recorded task
1517 is about to schedule in. This may change if we add a new marker at the
1518 end of the scheduler.
1520 Notice that the recorded task is 'sleep' with the PID of 2389
1521 and it has an rt_prio of 5. This priority is user-space priority
1522 and not the internal kernel priority. The policy is 1 for
1523 SCHED_FIFO and 2 for SCHED_RR.
1525 Note, that the trace data shows the internal priority (99 - rtprio).
1527 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1529 The 0:120:R means idle was running with a nice priority of 0 (120 - 20)
1530 and in the running state 'R'. The sleep task was scheduled in with
1531 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1532 and it too is in the running state.
1534 Doing the same with chrt -r 5 and function-trace set.
1536 echo 1 > options/function-trace
1540 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1541 # --------------------------------------------------------------------
1542 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1544 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
1548 # / _-----=> irqs-off
1549 # | / _----=> need-resched
1550 # || / _---=> hardirq/softirq
1551 # ||| / _--=> preempt-depth
1553 # cmd pid ||||| time | caller
1555 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
1556 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1557 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
1558 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
1559 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
1560 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
1561 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
1562 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
1563 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
1564 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1565 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
1566 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
1567 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
1568 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
1569 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
1570 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
1571 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
1572 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
1573 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
1574 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
1575 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
1576 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
1577 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
1578 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
1579 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
1580 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
1581 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
1582 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
1583 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
1584 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
1585 <idle>-0 3dN.1 13us : update_cpu_load_nohz <-tick_nohz_idle_exit
1586 <idle>-0 3dN.1 13us : _raw_spin_lock <-update_cpu_load_nohz
1587 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
1588 <idle>-0 3dN.2 13us : __update_cpu_load <-update_cpu_load_nohz
1589 <idle>-0 3dN.2 14us : sched_avg_update <-__update_cpu_load
1590 <idle>-0 3dN.2 14us : _raw_spin_unlock <-update_cpu_load_nohz
1591 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
1592 <idle>-0 3dN.1 15us : calc_load_exit_idle <-tick_nohz_idle_exit
1593 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
1594 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
1595 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
1596 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
1597 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1598 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
1599 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
1600 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
1601 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
1602 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
1603 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
1604 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
1605 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
1606 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1607 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
1608 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1609 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1610 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
1611 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
1612 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
1613 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1614 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
1615 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
1616 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
1617 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
1618 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
1619 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
1620 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
1621 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
1622 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1623 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
1624 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
1625 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
1626 <idle>-0 3.N.. 25us : schedule <-cpu_idle
1627 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
1628 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
1629 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
1630 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
1631 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
1632 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
1633 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
1634 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
1635 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
1636 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
1637 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
1638 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
1639 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
1641 This isn't that big of a trace, even with function tracing enabled,
1642 so I included the entire trace.
1644 The interrupt went off while when the system was idle. Somewhere
1645 before task_woken_rt() was called, the NEED_RESCHED flag was set,
1646 this is indicated by the first occurrence of the 'N' flag.
1648 Latency tracing and events
1649 --------------------------
1650 As function tracing can induce a much larger latency, but without
1651 seeing what happens within the latency it is hard to know what
1652 caused it. There is a middle ground, and that is with enabling
1655 # echo 0 > options/function-trace
1656 # echo wakeup_rt > current_tracer
1657 # echo 1 > events/enable
1658 # echo 1 > tracing_on
1659 # echo 0 > tracing_max_latency
1661 # echo 0 > tracing_on
1665 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1666 # --------------------------------------------------------------------
1667 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1669 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
1673 # / _-----=> irqs-off
1674 # | / _----=> need-resched
1675 # || / _---=> hardirq/softirq
1676 # ||| / _--=> preempt-depth
1678 # cmd pid ||||| time | caller
1680 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
1681 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1682 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
1683 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
1684 <idle>-0 2.N.2 2us : power_end: cpu_id=2
1685 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
1686 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
1687 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
1688 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
1689 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
1690 <idle>-0 2d..3 6us : __schedule <-schedule
1691 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
1697 This tracer is the function tracer. Enabling the function tracer
1698 can be done from the debug file system. Make sure the
1699 ftrace_enabled is set; otherwise this tracer is a nop.
1700 See the "ftrace_enabled" section below.
1702 # sysctl kernel.ftrace_enabled=1
1703 # echo function > current_tracer
1704 # echo 1 > tracing_on
1706 # echo 0 > tracing_on
1710 # entries-in-buffer/entries-written: 24799/24799 #P:4
1713 # / _----=> need-resched
1714 # | / _---=> hardirq/softirq
1715 # || / _--=> preempt-depth
1717 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
1719 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
1720 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
1721 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
1722 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
1723 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
1724 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
1725 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
1726 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
1730 Note: function tracer uses ring buffers to store the above
1731 entries. The newest data may overwrite the oldest data.
1732 Sometimes using echo to stop the trace is not sufficient because
1733 the tracing could have overwritten the data that you wanted to
1734 record. For this reason, it is sometimes better to disable
1735 tracing directly from a program. This allows you to stop the
1736 tracing at the point that you hit the part that you are
1737 interested in. To disable the tracing directly from a C program,
1738 something like following code snippet can be used:
1742 int main(int argc, char *argv[]) {
1744 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
1746 if (condition_hit()) {
1747 write(trace_fd, "0", 1);
1753 Single thread tracing
1754 ---------------------
1756 By writing into set_ftrace_pid you can trace a
1757 single thread. For example:
1759 # cat set_ftrace_pid
1761 # echo 3111 > set_ftrace_pid
1762 # cat set_ftrace_pid
1764 # echo function > current_tracer
1768 # TASK-PID CPU# TIMESTAMP FUNCTION
1770 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
1771 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
1772 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
1773 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
1774 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
1775 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
1776 # echo > set_ftrace_pid
1780 # TASK-PID CPU# TIMESTAMP FUNCTION
1782 ##### CPU 3 buffer started ####
1783 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
1784 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
1785 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
1786 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
1787 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
1789 If you want to trace a function when executing, you could use
1790 something like this simple program:
1794 #include <sys/types.h>
1795 #include <sys/stat.h>
1801 #define STR(x) _STR(x)
1802 #define MAX_PATH 256
1804 const char *find_debugfs(void)
1806 static char debugfs[MAX_PATH+1];
1807 static int debugfs_found;
1814 if ((fp = fopen("/proc/mounts","r")) == NULL) {
1815 perror("/proc/mounts");
1819 while (fscanf(fp, "%*s %"
1821 "s %99s %*s %*d %*d\n",
1822 debugfs, type) == 2) {
1823 if (strcmp(type, "debugfs") == 0)
1828 if (strcmp(type, "debugfs") != 0) {
1829 fprintf(stderr, "debugfs not mounted");
1833 strcat(debugfs, "/tracing/");
1839 const char *tracing_file(const char *file_name)
1841 static char trace_file[MAX_PATH+1];
1842 snprintf(trace_file, MAX_PATH, "%s/%s", find_debugfs(), file_name);
1846 int main (int argc, char **argv)
1856 ffd = open(tracing_file("current_tracer"), O_WRONLY);
1859 write(ffd, "nop", 3);
1861 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
1862 s = sprintf(line, "%d\n", getpid());
1865 write(ffd, "function", 8);
1870 execvp(argv[1], argv+1);
1876 Or this simple script!
1881 debugfs=`sed -ne 's/^debugfs \(.*\) debugfs.*/\1/p' /proc/mounts`
1882 echo nop > $debugfs/tracing/current_tracer
1883 echo 0 > $debugfs/tracing/tracing_on
1884 echo $$ > $debugfs/tracing/set_ftrace_pid
1885 echo function > $debugfs/tracing/current_tracer
1886 echo 1 > $debugfs/tracing/tracing_on
1891 function graph tracer
1892 ---------------------------
1894 This tracer is similar to the function tracer except that it
1895 probes a function on its entry and its exit. This is done by
1896 using a dynamically allocated stack of return addresses in each
1897 task_struct. On function entry the tracer overwrites the return
1898 address of each function traced to set a custom probe. Thus the
1899 original return address is stored on the stack of return address
1902 Probing on both ends of a function leads to special features
1905 - measure of a function's time execution
1906 - having a reliable call stack to draw function calls graph
1908 This tracer is useful in several situations:
1910 - you want to find the reason of a strange kernel behavior and
1911 need to see what happens in detail on any areas (or specific
1914 - you are experiencing weird latencies but it's difficult to
1917 - you want to find quickly which path is taken by a specific
1920 - you just want to peek inside a working kernel and want to see
1923 # tracer: function_graph
1925 # CPU DURATION FUNCTION CALLS
1929 0) | do_sys_open() {
1931 0) | kmem_cache_alloc() {
1932 0) 1.382 us | __might_sleep();
1934 0) | strncpy_from_user() {
1935 0) | might_fault() {
1936 0) 1.389 us | __might_sleep();
1941 0) 0.668 us | _spin_lock();
1942 0) 0.570 us | expand_files();
1943 0) 0.586 us | _spin_unlock();
1946 There are several columns that can be dynamically
1947 enabled/disabled. You can use every combination of options you
1948 want, depending on your needs.
1950 - The cpu number on which the function executed is default
1951 enabled. It is sometimes better to only trace one cpu (see
1952 tracing_cpu_mask file) or you might sometimes see unordered
1953 function calls while cpu tracing switch.
1955 hide: echo nofuncgraph-cpu > trace_options
1956 show: echo funcgraph-cpu > trace_options
1958 - The duration (function's time of execution) is displayed on
1959 the closing bracket line of a function or on the same line
1960 than the current function in case of a leaf one. It is default
1963 hide: echo nofuncgraph-duration > trace_options
1964 show: echo funcgraph-duration > trace_options
1966 - The overhead field precedes the duration field in case of
1967 reached duration thresholds.
1969 hide: echo nofuncgraph-overhead > trace_options
1970 show: echo funcgraph-overhead > trace_options
1971 depends on: funcgraph-duration
1975 3) # 1837.709 us | } /* __switch_to */
1976 3) | finish_task_switch() {
1977 3) 0.313 us | _raw_spin_unlock_irq();
1979 3) # 1889.063 us | } /* __schedule */
1980 3) ! 140.417 us | } /* __schedule */
1981 3) # 2034.948 us | } /* schedule */
1982 3) * 33998.59 us | } /* schedule_preempt_disabled */
1986 1) 0.260 us | msecs_to_jiffies();
1987 1) 0.313 us | __rcu_read_unlock();
1990 1) 0.313 us | rcu_bh_qs();
1991 1) 0.313 us | __local_bh_enable();
1993 1) 0.365 us | idle_cpu();
1994 1) | rcu_irq_exit() {
1995 1) 0.417 us | rcu_eqs_enter_common.isra.47();
1999 1) @ 119760.2 us | }
2005 2) 0.417 us | scheduler_ipi();
2013 + means that the function exceeded 10 usecs.
2014 ! means that the function exceeded 100 usecs.
2015 # means that the function exceeded 1000 usecs.
2016 * means that the function exceeded 10 msecs.
2017 @ means that the function exceeded 100 msecs.
2018 $ means that the function exceeded 1 sec.
2021 - The task/pid field displays the thread cmdline and pid which
2022 executed the function. It is default disabled.
2024 hide: echo nofuncgraph-proc > trace_options
2025 show: echo funcgraph-proc > trace_options
2029 # tracer: function_graph
2031 # CPU TASK/PID DURATION FUNCTION CALLS
2033 0) sh-4802 | | d_free() {
2034 0) sh-4802 | | call_rcu() {
2035 0) sh-4802 | | __call_rcu() {
2036 0) sh-4802 | 0.616 us | rcu_process_gp_end();
2037 0) sh-4802 | 0.586 us | check_for_new_grace_period();
2038 0) sh-4802 | 2.899 us | }
2039 0) sh-4802 | 4.040 us | }
2040 0) sh-4802 | 5.151 us | }
2041 0) sh-4802 | + 49.370 us | }
2044 - The absolute time field is an absolute timestamp given by the
2045 system clock since it started. A snapshot of this time is
2046 given on each entry/exit of functions
2048 hide: echo nofuncgraph-abstime > trace_options
2049 show: echo funcgraph-abstime > trace_options
2054 # TIME CPU DURATION FUNCTION CALLS
2056 360.774522 | 1) 0.541 us | }
2057 360.774522 | 1) 4.663 us | }
2058 360.774523 | 1) 0.541 us | __wake_up_bit();
2059 360.774524 | 1) 6.796 us | }
2060 360.774524 | 1) 7.952 us | }
2061 360.774525 | 1) 9.063 us | }
2062 360.774525 | 1) 0.615 us | journal_mark_dirty();
2063 360.774527 | 1) 0.578 us | __brelse();
2064 360.774528 | 1) | reiserfs_prepare_for_journal() {
2065 360.774528 | 1) | unlock_buffer() {
2066 360.774529 | 1) | wake_up_bit() {
2067 360.774529 | 1) | bit_waitqueue() {
2068 360.774530 | 1) 0.594 us | __phys_addr();
2071 The function name is always displayed after the closing bracket
2072 for a function if the start of that function is not in the
2075 Display of the function name after the closing bracket may be
2076 enabled for functions whose start is in the trace buffer,
2077 allowing easier searching with grep for function durations.
2078 It is default disabled.
2080 hide: echo nofuncgraph-tail > trace_options
2081 show: echo funcgraph-tail > trace_options
2083 Example with nofuncgraph-tail (default):
2085 0) | kmem_cache_free() {
2086 0) 0.518 us | __phys_addr();
2090 Example with funcgraph-tail:
2092 0) | kmem_cache_free() {
2093 0) 0.518 us | __phys_addr();
2094 0) 1.757 us | } /* kmem_cache_free() */
2095 0) 2.861 us | } /* putname() */
2097 You can put some comments on specific functions by using
2098 trace_printk() For example, if you want to put a comment inside
2099 the __might_sleep() function, you just have to include
2100 <linux/ftrace.h> and call trace_printk() inside __might_sleep()
2102 trace_printk("I'm a comment!\n")
2106 1) | __might_sleep() {
2107 1) | /* I'm a comment! */
2111 You can disable the hierarchical function call formatting and instead print a
2112 flat list of function entry and return events. This uses the format described
2113 in the Output Formatting section and respects all the trace options that
2114 control that formatting. Hierarchical formatting is the default.
2116 hierachical: echo nofuncgraph-flat > trace_options
2117 flat: echo funcgraph-flat > trace_options
2121 # tracer: function_graph
2123 # entries-in-buffer/entries-written: 68355/68355 #P:2
2126 # / _----=> need-resched
2127 # | / _---=> hardirq/softirq
2128 # || / _--=> preempt-depth
2130 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2132 sh-1806 [001] d... 198.843443: graph_ent: func=_raw_spin_lock
2133 sh-1806 [001] d... 198.843445: graph_ent: func=__raw_spin_lock
2134 sh-1806 [001] d..1 198.843447: graph_ret: func=__raw_spin_lock
2135 sh-1806 [001] d..1 198.843449: graph_ret: func=_raw_spin_lock
2136 sh-1806 [001] d..1 198.843451: graph_ent: func=_raw_spin_unlock_irqrestore
2137 sh-1806 [001] d... 198.843453: graph_ret: func=_raw_spin_unlock_irqrestore
2140 You might find other useful features for this tracer in the
2141 following "dynamic ftrace" section such as tracing only specific
2147 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2148 virtually no overhead when function tracing is disabled. The way
2149 this works is the mcount function call (placed at the start of
2150 every kernel function, produced by the -pg switch in gcc),
2151 starts of pointing to a simple return. (Enabling FTRACE will
2152 include the -pg switch in the compiling of the kernel.)
2154 At compile time every C file object is run through the
2155 recordmcount program (located in the scripts directory). This
2156 program will parse the ELF headers in the C object to find all
2157 the locations in the .text section that call mcount. (Note, only
2158 white listed .text sections are processed, since processing other
2159 sections like .init.text may cause races due to those sections
2160 being freed unexpectedly).
2162 A new section called "__mcount_loc" is created that holds
2163 references to all the mcount call sites in the .text section.
2164 The recordmcount program re-links this section back into the
2165 original object. The final linking stage of the kernel will add all these
2166 references into a single table.
2168 On boot up, before SMP is initialized, the dynamic ftrace code
2169 scans this table and updates all the locations into nops. It
2170 also records the locations, which are added to the
2171 available_filter_functions list. Modules are processed as they
2172 are loaded and before they are executed. When a module is
2173 unloaded, it also removes its functions from the ftrace function
2174 list. This is automatic in the module unload code, and the
2175 module author does not need to worry about it.
2177 When tracing is enabled, the process of modifying the function
2178 tracepoints is dependent on architecture. The old method is to use
2179 kstop_machine to prevent races with the CPUs executing code being
2180 modified (which can cause the CPU to do undesirable things, especially
2181 if the modified code crosses cache (or page) boundaries), and the nops are
2182 patched back to calls. But this time, they do not call mcount
2183 (which is just a function stub). They now call into the ftrace
2186 The new method of modifying the function tracepoints is to place
2187 a breakpoint at the location to be modified, sync all CPUs, modify
2188 the rest of the instruction not covered by the breakpoint. Sync
2189 all CPUs again, and then remove the breakpoint with the finished
2190 version to the ftrace call site.
2192 Some archs do not even need to monkey around with the synchronization,
2193 and can just slap the new code on top of the old without any
2194 problems with other CPUs executing it at the same time.
2196 One special side-effect to the recording of the functions being
2197 traced is that we can now selectively choose which functions we
2198 wish to trace and which ones we want the mcount calls to remain
2201 Two files are used, one for enabling and one for disabling the
2202 tracing of specified functions. They are:
2210 A list of available functions that you can add to these files is
2213 available_filter_functions
2215 # cat available_filter_functions
2224 If I am only interested in sys_nanosleep and hrtimer_interrupt:
2226 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2227 # echo function > current_tracer
2228 # echo 1 > tracing_on
2230 # echo 0 > tracing_on
2234 # entries-in-buffer/entries-written: 5/5 #P:4
2237 # / _----=> need-resched
2238 # | / _---=> hardirq/softirq
2239 # || / _--=> preempt-depth
2241 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2243 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2244 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2245 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2246 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2247 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2249 To see which functions are being traced, you can cat the file:
2251 # cat set_ftrace_filter
2256 Perhaps this is not enough. The filters also allow simple wild
2257 cards. Only the following are currently available
2259 <match>* - will match functions that begin with <match>
2260 *<match> - will match functions that end with <match>
2261 *<match>* - will match functions that have <match> in it
2263 These are the only wild cards which are supported.
2265 <match>*<match> will not work.
2267 Note: It is better to use quotes to enclose the wild cards,
2268 otherwise the shell may expand the parameters into names
2269 of files in the local directory.
2271 # echo 'hrtimer_*' > set_ftrace_filter
2277 # entries-in-buffer/entries-written: 897/897 #P:4
2280 # / _----=> need-resched
2281 # | / _---=> hardirq/softirq
2282 # || / _--=> preempt-depth
2284 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2286 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2287 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2288 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2289 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2290 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2291 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2292 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2293 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2295 Notice that we lost the sys_nanosleep.
2297 # cat set_ftrace_filter
2302 hrtimer_try_to_cancel
2306 hrtimer_force_reprogram
2307 hrtimer_get_next_event
2311 hrtimer_get_remaining
2313 hrtimer_init_sleeper
2316 This is because the '>' and '>>' act just like they do in bash.
2317 To rewrite the filters, use '>'
2318 To append to the filters, use '>>'
2320 To clear out a filter so that all functions will be recorded
2323 # echo > set_ftrace_filter
2324 # cat set_ftrace_filter
2327 Again, now we want to append.
2329 # echo sys_nanosleep > set_ftrace_filter
2330 # cat set_ftrace_filter
2332 # echo 'hrtimer_*' >> set_ftrace_filter
2333 # cat set_ftrace_filter
2338 hrtimer_try_to_cancel
2342 hrtimer_force_reprogram
2343 hrtimer_get_next_event
2348 hrtimer_get_remaining
2350 hrtimer_init_sleeper
2353 The set_ftrace_notrace prevents those functions from being
2356 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2362 # entries-in-buffer/entries-written: 39608/39608 #P:4
2365 # / _----=> need-resched
2366 # | / _---=> hardirq/softirq
2367 # || / _--=> preempt-depth
2369 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2371 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2372 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2373 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2374 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2375 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2376 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2377 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2378 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2379 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2380 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2381 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2382 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2384 We can see that there's no more lock or preempt tracing.
2387 Dynamic ftrace with the function graph tracer
2388 ---------------------------------------------
2390 Although what has been explained above concerns both the
2391 function tracer and the function-graph-tracer, there are some
2392 special features only available in the function-graph tracer.
2394 If you want to trace only one function and all of its children,
2395 you just have to echo its name into set_graph_function:
2397 echo __do_fault > set_graph_function
2399 will produce the following "expanded" trace of the __do_fault()
2403 0) | filemap_fault() {
2404 0) | find_lock_page() {
2405 0) 0.804 us | find_get_page();
2406 0) | __might_sleep() {
2410 0) 0.653 us | _spin_lock();
2411 0) 0.578 us | page_add_file_rmap();
2412 0) 0.525 us | native_set_pte_at();
2413 0) 0.585 us | _spin_unlock();
2414 0) | unlock_page() {
2415 0) 0.541 us | page_waitqueue();
2416 0) 0.639 us | __wake_up_bit();
2420 0) | filemap_fault() {
2421 0) | find_lock_page() {
2422 0) 0.698 us | find_get_page();
2423 0) | __might_sleep() {
2427 0) 0.631 us | _spin_lock();
2428 0) 0.571 us | page_add_file_rmap();
2429 0) 0.526 us | native_set_pte_at();
2430 0) 0.586 us | _spin_unlock();
2431 0) | unlock_page() {
2432 0) 0.533 us | page_waitqueue();
2433 0) 0.638 us | __wake_up_bit();
2437 You can also expand several functions at once:
2439 echo sys_open > set_graph_function
2440 echo sys_close >> set_graph_function
2442 Now if you want to go back to trace all functions you can clear
2443 this special filter via:
2445 echo > set_graph_function
2451 Note, the proc sysctl ftrace_enable is a big on/off switch for the
2452 function tracer. By default it is enabled (when function tracing is
2453 enabled in the kernel). If it is disabled, all function tracing is
2454 disabled. This includes not only the function tracers for ftrace, but
2455 also for any other uses (perf, kprobes, stack tracing, profiling, etc).
2457 Please disable this with care.
2459 This can be disable (and enabled) with:
2461 sysctl kernel.ftrace_enabled=0
2462 sysctl kernel.ftrace_enabled=1
2466 echo 0 > /proc/sys/kernel/ftrace_enabled
2467 echo 1 > /proc/sys/kernel/ftrace_enabled
2473 A few commands are supported by the set_ftrace_filter interface.
2474 Trace commands have the following format:
2476 <function>:<command>:<parameter>
2478 The following commands are supported:
2481 This command enables function filtering per module. The
2482 parameter defines the module. For example, if only the write*
2483 functions in the ext3 module are desired, run:
2485 echo 'write*:mod:ext3' > set_ftrace_filter
2487 This command interacts with the filter in the same way as
2488 filtering based on function names. Thus, adding more functions
2489 in a different module is accomplished by appending (>>) to the
2490 filter file. Remove specific module functions by prepending
2493 echo '!writeback*:mod:ext3' >> set_ftrace_filter
2495 Mod command supports module globbing. Disable tracing for all
2496 functions except a specific module:
2498 echo '!*:mod:!ext3' >> set_ftrace_filter
2500 Disable tracing for all modules, but still trace kernel:
2502 echo '!*:mod:*' >> set_ftrace_filter
2504 Enable filter only for kernel:
2506 echo '*write*:mod:!*' >> set_ftrace_filter
2508 Enable filter for module globbing:
2510 echo '*write*:mod:*snd*' >> set_ftrace_filter
2513 These commands turn tracing on and off when the specified
2514 functions are hit. The parameter determines how many times the
2515 tracing system is turned on and off. If unspecified, there is
2516 no limit. For example, to disable tracing when a schedule bug
2517 is hit the first 5 times, run:
2519 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
2521 To always disable tracing when __schedule_bug is hit:
2523 echo '__schedule_bug:traceoff' > set_ftrace_filter
2525 These commands are cumulative whether or not they are appended
2526 to set_ftrace_filter. To remove a command, prepend it by '!'
2527 and drop the parameter:
2529 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
2531 The above removes the traceoff command for __schedule_bug
2532 that have a counter. To remove commands without counters:
2534 echo '!__schedule_bug:traceoff' > set_ftrace_filter
2537 Will cause a snapshot to be triggered when the function is hit.
2539 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
2541 To only snapshot once:
2543 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
2545 To remove the above commands:
2547 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
2548 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
2550 - enable_event/disable_event
2551 These commands can enable or disable a trace event. Note, because
2552 function tracing callbacks are very sensitive, when these commands
2553 are registered, the trace point is activated, but disabled in
2554 a "soft" mode. That is, the tracepoint will be called, but
2555 just will not be traced. The event tracepoint stays in this mode
2556 as long as there's a command that triggers it.
2558 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
2563 <function>:enable_event:<system>:<event>[:count]
2564 <function>:disable_event:<system>:<event>[:count]
2566 To remove the events commands:
2569 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
2571 echo '!schedule:disable_event:sched:sched_switch' > \
2575 When the function is hit, it will dump the contents of the ftrace
2576 ring buffer to the console. This is useful if you need to debug
2577 something, and want to dump the trace when a certain function
2578 is hit. Perhaps its a function that is called before a tripple
2579 fault happens and does not allow you to get a regular dump.
2582 When the function is hit, it will dump the contents of the ftrace
2583 ring buffer for the current CPU to the console. Unlike the "dump"
2584 command, it only prints out the contents of the ring buffer for the
2585 CPU that executed the function that triggered the dump.
2590 The trace_pipe outputs the same content as the trace file, but
2591 the effect on the tracing is different. Every read from
2592 trace_pipe is consumed. This means that subsequent reads will be
2593 different. The trace is live.
2595 # echo function > current_tracer
2596 # cat trace_pipe > /tmp/trace.out &
2598 # echo 1 > tracing_on
2600 # echo 0 > tracing_on
2604 # entries-in-buffer/entries-written: 0/0 #P:4
2607 # / _----=> need-resched
2608 # | / _---=> hardirq/softirq
2609 # || / _--=> preempt-depth
2611 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2615 # cat /tmp/trace.out
2616 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
2617 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
2618 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
2619 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
2620 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
2621 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
2622 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
2623 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
2624 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
2627 Note, reading the trace_pipe file will block until more input is
2633 Having too much or not enough data can be troublesome in
2634 diagnosing an issue in the kernel. The file buffer_size_kb is
2635 used to modify the size of the internal trace buffers. The
2636 number listed is the number of entries that can be recorded per
2637 CPU. To know the full size, multiply the number of possible CPUs
2638 with the number of entries.
2640 # cat buffer_size_kb
2641 1408 (units kilobytes)
2643 Or simply read buffer_total_size_kb
2645 # cat buffer_total_size_kb
2648 To modify the buffer, simple echo in a number (in 1024 byte segments).
2650 # echo 10000 > buffer_size_kb
2651 # cat buffer_size_kb
2652 10000 (units kilobytes)
2654 It will try to allocate as much as possible. If you allocate too
2655 much, it can cause Out-Of-Memory to trigger.
2657 # echo 1000000000000 > buffer_size_kb
2658 -bash: echo: write error: Cannot allocate memory
2659 # cat buffer_size_kb
2662 The per_cpu buffers can be changed individually as well:
2664 # echo 10000 > per_cpu/cpu0/buffer_size_kb
2665 # echo 100 > per_cpu/cpu1/buffer_size_kb
2667 When the per_cpu buffers are not the same, the buffer_size_kb
2668 at the top level will just show an X
2670 # cat buffer_size_kb
2673 This is where the buffer_total_size_kb is useful:
2675 # cat buffer_total_size_kb
2678 Writing to the top level buffer_size_kb will reset all the buffers
2679 to be the same again.
2683 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
2684 available to all non latency tracers. (Latency tracers which
2685 record max latency, such as "irqsoff" or "wakeup", can't use
2686 this feature, since those are already using the snapshot
2687 mechanism internally.)
2689 Snapshot preserves a current trace buffer at a particular point
2690 in time without stopping tracing. Ftrace swaps the current
2691 buffer with a spare buffer, and tracing continues in the new
2692 current (=previous spare) buffer.
2694 The following debugfs files in "tracing" are related to this
2699 This is used to take a snapshot and to read the output
2700 of the snapshot. Echo 1 into this file to allocate a
2701 spare buffer and to take a snapshot (swap), then read
2702 the snapshot from this file in the same format as
2703 "trace" (described above in the section "The File
2704 System"). Both reads snapshot and tracing are executable
2705 in parallel. When the spare buffer is allocated, echoing
2706 0 frees it, and echoing else (positive) values clear the
2708 More details are shown in the table below.
2710 status\input | 0 | 1 | else |
2711 --------------+------------+------------+------------+
2712 not allocated |(do nothing)| alloc+swap |(do nothing)|
2713 --------------+------------+------------+------------+
2714 allocated | free | swap | clear |
2715 --------------+------------+------------+------------+
2717 Here is an example of using the snapshot feature.
2719 # echo 1 > events/sched/enable
2724 # entries-in-buffer/entries-written: 71/71 #P:8
2727 # / _----=> need-resched
2728 # | / _---=> hardirq/softirq
2729 # || / _--=> preempt-depth
2731 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2733 <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
2734 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
2736 <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
2741 # entries-in-buffer/entries-written: 77/77 #P:8
2744 # / _----=> need-resched
2745 # | / _---=> hardirq/softirq
2746 # || / _--=> preempt-depth
2748 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2750 <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
2751 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
2755 If you try to use this snapshot feature when current tracer is
2756 one of the latency tracers, you will get the following results.
2758 # echo wakeup > current_tracer
2760 bash: echo: write error: Device or resource busy
2762 cat: snapshot: Device or resource busy
2767 In the debugfs tracing directory is a directory called "instances".
2768 This directory can have new directories created inside of it using
2769 mkdir, and removing directories with rmdir. The directory created
2770 with mkdir in this directory will already contain files and other
2771 directories after it is created.
2773 # mkdir instances/foo
2775 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
2776 set_event snapshot trace trace_clock trace_marker trace_options
2777 trace_pipe tracing_on
2779 As you can see, the new directory looks similar to the tracing directory
2780 itself. In fact, it is very similar, except that the buffer and
2781 events are agnostic from the main director, or from any other
2782 instances that are created.
2784 The files in the new directory work just like the files with the
2785 same name in the tracing directory except the buffer that is used
2786 is a separate and new buffer. The files affect that buffer but do not
2787 affect the main buffer with the exception of trace_options. Currently,
2788 the trace_options affect all instances and the top level buffer
2789 the same, but this may change in future releases. That is, options
2790 may become specific to the instance they reside in.
2792 Notice that none of the function tracer files are there, nor is
2793 current_tracer and available_tracers. This is because the buffers
2794 can currently only have events enabled for them.
2796 # mkdir instances/foo
2797 # mkdir instances/bar
2798 # mkdir instances/zoot
2799 # echo 100000 > buffer_size_kb
2800 # echo 1000 > instances/foo/buffer_size_kb
2801 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
2802 # echo function > current_trace
2803 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
2804 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
2805 # echo 1 > instances/foo/events/sched/sched_switch/enable
2806 # echo 1 > instances/bar/events/irq/enable
2807 # echo 1 > instances/zoot/events/syscalls/enable
2809 CPU:2 [LOST 11745 EVENTS]
2810 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
2811 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
2812 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
2813 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
2814 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
2815 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
2816 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
2817 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
2818 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
2819 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
2820 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
2823 # cat instances/foo/trace_pipe
2824 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
2825 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
2826 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
2827 <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
2828 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
2829 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
2830 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
2831 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
2832 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
2833 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
2836 # cat instances/bar/trace_pipe
2837 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
2838 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
2839 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
2840 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
2841 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
2842 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
2843 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
2844 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
2845 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
2846 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
2847 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
2848 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
2851 # cat instances/zoot/trace
2854 # entries-in-buffer/entries-written: 18996/18996 #P:4
2857 # / _----=> need-resched
2858 # | / _---=> hardirq/softirq
2859 # || / _--=> preempt-depth
2861 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2863 bash-1998 [000] d... 140.733501: sys_write -> 0x2
2864 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
2865 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
2866 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
2867 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
2868 bash-1998 [000] d... 140.733510: sys_close(fd: a)
2869 bash-1998 [000] d... 140.733510: sys_close -> 0x0
2870 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
2871 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
2872 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
2873 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
2875 You can see that the trace of the top most trace buffer shows only
2876 the function tracing. The foo instance displays wakeups and task
2879 To remove the instances, simply delete their directories:
2881 # rmdir instances/foo
2882 # rmdir instances/bar
2883 # rmdir instances/zoot
2885 Note, if a process has a trace file open in one of the instance
2886 directories, the rmdir will fail with EBUSY.
2891 Since the kernel has a fixed sized stack, it is important not to
2892 waste it in functions. A kernel developer must be conscience of
2893 what they allocate on the stack. If they add too much, the system
2894 can be in danger of a stack overflow, and corruption will occur,
2895 usually leading to a system panic.
2897 There are some tools that check this, usually with interrupts
2898 periodically checking usage. But if you can perform a check
2899 at every function call that will become very useful. As ftrace provides
2900 a function tracer, it makes it convenient to check the stack size
2901 at every function call. This is enabled via the stack tracer.
2903 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
2904 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
2906 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
2908 You can also enable it from the kernel command line to trace
2909 the stack size of the kernel during boot up, by adding "stacktrace"
2910 to the kernel command line parameter.
2912 After running it for a few minutes, the output looks like:
2914 # cat stack_max_size
2918 Depth Size Location (18 entries)
2920 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
2921 1) 2704 160 find_busiest_group+0x31/0x1f1
2922 2) 2544 256 load_balance+0xd9/0x662
2923 3) 2288 80 idle_balance+0xbb/0x130
2924 4) 2208 128 __schedule+0x26e/0x5b9
2925 5) 2080 16 schedule+0x64/0x66
2926 6) 2064 128 schedule_timeout+0x34/0xe0
2927 7) 1936 112 wait_for_common+0x97/0xf1
2928 8) 1824 16 wait_for_completion+0x1d/0x1f
2929 9) 1808 128 flush_work+0xfe/0x119
2930 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
2931 11) 1664 48 input_available_p+0x1d/0x5c
2932 12) 1616 48 n_tty_poll+0x6d/0x134
2933 13) 1568 64 tty_poll+0x64/0x7f
2934 14) 1504 880 do_select+0x31e/0x511
2935 15) 624 400 core_sys_select+0x177/0x216
2936 16) 224 96 sys_select+0x91/0xb9
2937 17) 128 128 system_call_fastpath+0x16/0x1b
2939 Note, if -mfentry is being used by gcc, functions get traced before
2940 they set up the stack frame. This means that leaf level functions
2941 are not tested by the stack tracer when -mfentry is used.
2943 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
2947 More details can be found in the source code, in the
2948 kernel/trace/*.c files.
OSDN Git Service
