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 Set a "trigger" function where tracing should start
210 with the function graph tracer (See the section
211 "dynamic ftrace" for more details).
213 available_filter_functions:
215 This lists the functions that ftrace
216 has processed and can trace. These are the function
217 names that you can pass to "set_ftrace_filter" or
218 "set_ftrace_notrace". (See the section "dynamic ftrace"
219 below for more details.)
223 This file is more for debugging ftrace, but can also be useful
224 in seeing if any function has a callback attached to it.
225 Not only does the trace infrastructure use ftrace function
226 trace utility, but other subsystems might too. This file
227 displays all functions that have a callback attached to them
228 as well as the number of callbacks that have been attached.
229 Note, a callback may also call multiple functions which will
230 not be listed in this count.
232 If the callback registered to be traced by a function with
233 the "save regs" attribute (thus even more overhead), a 'R'
234 will be displayed on the same line as the function that
235 is returning registers.
237 If the callback registered to be traced by a function with
238 the "ip modify" attribute (thus the regs->ip can be changed),
239 an 'I' will be displayed on the same line as the function that
242 function_profile_enabled:
244 When set it will enable all functions with either the function
245 tracer, or if enabled, the function graph tracer. It will
246 keep a histogram of the number of functions that were called
247 and if run with the function graph tracer, it will also keep
248 track of the time spent in those functions. The histogram
249 content can be displayed in the files:
251 trace_stats/function<cpu> ( function0, function1, etc).
255 A directory that holds different tracing stats.
259 Enable dynamic trace points. See kprobetrace.txt.
263 Dynamic trace points stats. See kprobetrace.txt.
267 Used with the function graph tracer. This is the max depth
268 it will trace into a function. Setting this to a value of
269 one will show only the first kernel function that is called
274 This is for tools that read the raw format files. If an event in
275 the ring buffer references a string (currently only trace_printk()
276 does this), only a pointer to the string is recorded into the buffer
277 and not the string itself. This prevents tools from knowing what
278 that string was. This file displays the string and address for
279 the string allowing tools to map the pointers to what the
284 Only the pid of the task is recorded in a trace event unless
285 the event specifically saves the task comm as well. Ftrace
286 makes a cache of pid mappings to comms to try to display
287 comms for events. If a pid for a comm is not listed, then
288 "<...>" is displayed in the output.
292 This displays the "snapshot" buffer and also lets the user
293 take a snapshot of the current running trace.
294 See the "Snapshot" section below for more details.
298 When the stack tracer is activated, this will display the
299 maximum stack size it has encountered.
300 See the "Stack Trace" section below.
304 This displays the stack back trace of the largest stack
305 that was encountered when the stack tracer is activated.
306 See the "Stack Trace" section below.
310 This is similar to "set_ftrace_filter" but it limits what
311 functions the stack tracer will check.
315 Whenever an event is recorded into the ring buffer, a
316 "timestamp" is added. This stamp comes from a specified
317 clock. By default, ftrace uses the "local" clock. This
318 clock is very fast and strictly per cpu, but on some
319 systems it may not be monotonic with respect to other
320 CPUs. In other words, the local clocks may not be in sync
321 with local clocks on other CPUs.
323 Usual clocks for tracing:
326 [local] global counter x86-tsc
328 local: Default clock, but may not be in sync across CPUs
330 global: This clock is in sync with all CPUs but may
331 be a bit slower than the local clock.
333 counter: This is not a clock at all, but literally an atomic
334 counter. It counts up one by one, but is in sync
335 with all CPUs. This is useful when you need to
336 know exactly the order events occurred with respect to
337 each other on different CPUs.
339 uptime: This uses the jiffies counter and the time stamp
340 is relative to the time since boot up.
342 perf: This makes ftrace use the same clock that perf uses.
343 Eventually perf will be able to read ftrace buffers
344 and this will help out in interleaving the data.
346 x86-tsc: Architectures may define their own clocks. For
347 example, x86 uses its own TSC cycle clock here.
349 ppc-tb: This uses the powerpc timebase register value.
350 This is in sync across CPUs and can also be used
351 to correlate events across hypervisor/guest if
354 To set a clock, simply echo the clock name into this file.
356 echo global > trace_clock
360 This is a very useful file for synchronizing user space
361 with events happening in the kernel. Writing strings into
362 this file will be written into the ftrace buffer.
364 It is useful in applications to open this file at the start
365 of the application and just reference the file descriptor
368 void trace_write(const char *fmt, ...)
378 n = vsnprintf(buf, 256, fmt, ap);
381 write(trace_fd, buf, n);
386 trace_fd = open("trace_marker", WR_ONLY);
390 Add dynamic tracepoints in programs.
395 Uprobe statistics. See uprobetrace.txt
399 This is a way to make multiple trace buffers where different
400 events can be recorded in different buffers.
401 See "Instances" section below.
405 This is the trace event directory. It holds event tracepoints
406 (also known as static tracepoints) that have been compiled
407 into the kernel. It shows what event tracepoints exist
408 and how they are grouped by system. There are "enable"
409 files at various levels that can enable the tracepoints
410 when a "1" is written to them.
412 See events.txt for more information.
416 This is a directory that contains the trace per_cpu information.
418 per_cpu/cpu0/buffer_size_kb:
420 The ftrace buffer is defined per_cpu. That is, there's a separate
421 buffer for each CPU to allow writes to be done atomically,
422 and free from cache bouncing. These buffers may have different
423 size buffers. This file is similar to the buffer_size_kb
424 file, but it only displays or sets the buffer size for the
425 specific CPU. (here cpu0).
429 This is similar to the "trace" file, but it will only display
430 the data specific for the CPU. If written to, it only clears
431 the specific CPU buffer.
433 per_cpu/cpu0/trace_pipe
435 This is similar to the "trace_pipe" file, and is a consuming
436 read, but it will only display (and consume) the data specific
439 per_cpu/cpu0/trace_pipe_raw
441 For tools that can parse the ftrace ring buffer binary format,
442 the trace_pipe_raw file can be used to extract the data
443 from the ring buffer directly. With the use of the splice()
444 system call, the buffer data can be quickly transferred to
445 a file or to the network where a server is collecting the
448 Like trace_pipe, this is a consuming reader, where multiple
449 reads will always produce different data.
451 per_cpu/cpu0/snapshot:
453 This is similar to the main "snapshot" file, but will only
454 snapshot the current CPU (if supported). It only displays
455 the content of the snapshot for a given CPU, and if
456 written to, only clears this CPU buffer.
458 per_cpu/cpu0/snapshot_raw:
460 Similar to the trace_pipe_raw, but will read the binary format
461 from the snapshot buffer for the given CPU.
465 This displays certain stats about the ring buffer:
467 entries: The number of events that are still in the buffer.
469 overrun: The number of lost events due to overwriting when
472 commit overrun: Should always be zero.
473 This gets set if so many events happened within a nested
474 event (ring buffer is re-entrant), that it fills the
475 buffer and starts dropping events.
477 bytes: Bytes actually read (not overwritten).
479 oldest event ts: The oldest timestamp in the buffer
481 now ts: The current timestamp
483 dropped events: Events lost due to overwrite option being off.
485 read events: The number of events read.
490 Here is the list of current tracers that may be configured.
494 Function call tracer to trace all kernel functions.
498 Similar to the function tracer except that the
499 function tracer probes the functions on their entry
500 whereas the function graph tracer traces on both entry
501 and exit of the functions. It then provides the ability
502 to draw a graph of function calls similar to C code
507 Traces the areas that disable interrupts and saves
508 the trace with the longest max latency.
509 See tracing_max_latency. When a new max is recorded,
510 it replaces the old trace. It is best to view this
511 trace with the latency-format option enabled.
515 Similar to irqsoff but traces and records the amount of
516 time for which preemption is disabled.
520 Similar to irqsoff and preemptoff, but traces and
521 records the largest time for which irqs and/or preemption
526 Traces and records the max latency that it takes for
527 the highest priority task to get scheduled after
528 it has been woken up.
529 Traces all tasks as an average developer would expect.
533 Traces and records the max latency that it takes for just
534 RT tasks (as the current "wakeup" does). This is useful
535 for those interested in wake up timings of RT tasks.
539 This is the "trace nothing" tracer. To remove all
540 tracers from tracing simply echo "nop" into
544 Examples of using the tracer
545 ----------------------------
547 Here are typical examples of using the tracers when controlling
548 them only with the debugfs interface (without using any
549 user-land utilities).
554 Here is an example of the output format of the file "trace"
559 # entries-in-buffer/entries-written: 140080/250280 #P:4
562 # / _----=> need-resched
563 # | / _---=> hardirq/softirq
564 # || / _--=> preempt-depth
566 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
568 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath
569 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close
570 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd
571 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify
572 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock
573 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd
574 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock
575 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd
576 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close
577 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath
580 A header is printed with the tracer name that is represented by
581 the trace. In this case the tracer is "function". Then it shows the
582 number of events in the buffer as well as the total number of entries
583 that were written. The difference is the number of entries that were
584 lost due to the buffer filling up (250280 - 140080 = 110200 events
587 The header explains the content of the events. Task name "bash", the task
588 PID "1977", the CPU that it was running on "000", the latency format
589 (explained below), the timestamp in <secs>.<usecs> format, the
590 function name that was traced "sys_close" and the parent function that
591 called this function "system_call_fastpath". The timestamp is the time
592 at which the function was entered.
597 When the latency-format option is enabled or when one of the latency
598 tracers is set, the trace file gives somewhat more information to see
599 why a latency happened. Here is a typical trace.
603 # irqsoff latency trace v1.1.5 on 3.8.0-test+
604 # --------------------------------------------------------------------
605 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
607 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)
609 # => started at: __lock_task_sighand
610 # => ended at: _raw_spin_unlock_irqrestore
614 # / _-----=> irqs-off
615 # | / _----=> need-resched
616 # || / _---=> hardirq/softirq
617 # ||| / _--=> preempt-depth
619 # cmd pid ||||| time | caller
621 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand
622 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore
623 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore
624 ps-6143 2d..1 306us : <stack trace>
625 => trace_hardirqs_on_caller
627 => _raw_spin_unlock_irqrestore
634 => system_call_fastpath
637 This shows that the current tracer is "irqsoff" tracing the time
638 for which interrupts were disabled. It gives the trace version (which
639 never changes) and the version of the kernel upon which this was executed on
640 (3.10). Then it displays the max latency in microseconds (259 us). The number
641 of trace entries displayed and the total number (both are four: #4/4).
642 VP, KP, SP, and HP are always zero and are reserved for later use.
643 #P is the number of online CPUs (#P:4).
645 The task is the process that was running when the latency
646 occurred. (ps pid: 6143).
648 The start and stop (the functions in which the interrupts were
649 disabled and enabled respectively) that caused the latencies:
651 __lock_task_sighand is where the interrupts were disabled.
652 _raw_spin_unlock_irqrestore is where they were enabled again.
654 The next lines after the header are the trace itself. The header
655 explains which is which.
657 cmd: The name of the process in the trace.
659 pid: The PID of that process.
661 CPU#: The CPU which the process was running on.
663 irqs-off: 'd' interrupts are disabled. '.' otherwise.
664 Note: If the architecture does not support a way to
665 read the irq flags variable, an 'X' will always
669 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,
670 'n' only TIF_NEED_RESCHED is set,
671 'p' only PREEMPT_NEED_RESCHED is set,
675 'H' - hard irq occurred inside a softirq.
676 'h' - hard irq is running
677 's' - soft irq is running
678 '.' - normal context.
680 preempt-depth: The level of preempt_disabled
682 The above is mostly meaningful for kernel developers.
684 time: When the latency-format option is enabled, the trace file
685 output includes a timestamp relative to the start of the
686 trace. This differs from the output when latency-format
687 is disabled, which includes an absolute timestamp.
689 delay: This is just to help catch your eye a bit better. And
690 needs to be fixed to be only relative to the same CPU.
691 The marks are determined by the difference between this
692 current trace and the next trace.
693 '$' - greater than 1 second
694 '#' - greater than 1000 microsecond
695 '!' - greater than 100 microsecond
696 '+' - greater than 10 microsecond
697 ' ' - less than or equal to 10 microsecond.
699 The rest is the same as the 'trace' file.
701 Note, the latency tracers will usually end with a back trace
702 to easily find where the latency occurred.
707 The trace_options file (or the options directory) is used to control
708 what gets printed in the trace output, or manipulate the tracers.
709 To see what is available, simply cat the file:
739 To disable one of the options, echo in the option prepended with
742 echo noprint-parent > trace_options
744 To enable an option, leave off the "no".
746 echo sym-offset > trace_options
748 Here are the available options:
750 print-parent - On function traces, display the calling (parent)
751 function as well as the function being traced.
754 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul
757 bash-4000 [01] 1477.606694: simple_strtoul
760 sym-offset - Display not only the function name, but also the
761 offset in the function. For example, instead of
762 seeing just "ktime_get", you will see
763 "ktime_get+0xb/0x20".
766 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0
768 sym-addr - this will also display the function address as well
769 as the function name.
772 bash-4000 [01] 1477.606694: simple_strtoul <c0339346>
774 verbose - This deals with the trace file when the
775 latency-format option is enabled.
777 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \
778 (+0.000ms): simple_strtoul (kstrtoul)
780 raw - This will display raw numbers. This option is best for
781 use with user applications that can translate the raw
782 numbers better than having it done in the kernel.
784 hex - Similar to raw, but the numbers will be in a hexadecimal
787 bin - This will print out the formats in raw binary.
789 block - When set, reading trace_pipe will not block when polled.
791 stacktrace - This is one of the options that changes the trace
792 itself. When a trace is recorded, so is the stack
793 of functions. This allows for back traces of
796 trace_printk - Can disable trace_printk() from writing into the buffer.
798 branch - Enable branch tracing with the tracer.
800 annotate - It is sometimes confusing when the CPU buffers are full
801 and one CPU buffer had a lot of events recently, thus
802 a shorter time frame, were another CPU may have only had
803 a few events, which lets it have older events. When
804 the trace is reported, it shows the oldest events first,
805 and it may look like only one CPU ran (the one with the
806 oldest events). When the annotate option is set, it will
807 display when a new CPU buffer started:
809 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on
810 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on
811 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore
812 ##### CPU 2 buffer started ####
813 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle
814 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog
815 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock
817 userstacktrace - This option changes the trace. It records a
818 stacktrace of the current userspace thread.
820 sym-userobj - when user stacktrace are enabled, look up which
821 object the address belongs to, and print a
822 relative address. This is especially useful when
823 ASLR is on, otherwise you don't get a chance to
824 resolve the address to object/file/line after
825 the app is no longer running
827 The lookup is performed when you read
828 trace,trace_pipe. Example:
830 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0
831 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]
834 printk-msg-only - When set, trace_printk()s will only show the format
835 and not their parameters (if trace_bprintk() or
836 trace_bputs() was used to save the trace_printk()).
838 context-info - Show only the event data. Hides the comm, PID,
839 timestamp, CPU, and other useful data.
841 latency-format - This option changes the trace. When
842 it is enabled, the trace displays
843 additional information about the
844 latencies, as described in "Latency
847 sleep-time - When running function graph tracer, to include
848 the time a task schedules out in its function.
849 When enabled, it will account time the task has been
850 scheduled out as part of the function call.
852 graph-time - When running function graph tracer, to include the
853 time to call nested functions. When this is not set,
854 the time reported for the function will only include
855 the time the function itself executed for, not the time
856 for functions that it called.
858 record-cmd - When any event or tracer is enabled, a hook is enabled
859 in the sched_switch trace point to fill comm cache
860 with mapped pids and comms. But this may cause some
861 overhead, and if you only care about pids, and not the
862 name of the task, disabling this option can lower the
865 overwrite - This controls what happens when the trace buffer is
866 full. If "1" (default), the oldest events are
867 discarded and overwritten. If "0", then the newest
868 events are discarded.
869 (see per_cpu/cpu0/stats for overrun and dropped)
871 disable_on_free - When the free_buffer is closed, tracing will
872 stop (tracing_on set to 0).
874 irq-info - Shows the interrupt, preempt count, need resched data.
875 When disabled, the trace looks like:
879 # entries-in-buffer/entries-written: 144405/9452052 #P:4
881 # TASK-PID CPU# TIMESTAMP FUNCTION
883 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up
884 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89
885 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task
888 markers - When set, the trace_marker is writable (only by root).
889 When disabled, the trace_marker will error with EINVAL
893 function-trace - The latency tracers will enable function tracing
894 if this option is enabled (default it is). When
895 it is disabled, the latency tracers do not trace
896 functions. This keeps the overhead of the tracer down
897 when performing latency tests.
899 Note: Some tracers have their own options. They only appear
900 when the tracer is active.
907 When interrupts are disabled, the CPU can not react to any other
908 external event (besides NMIs and SMIs). This prevents the timer
909 interrupt from triggering or the mouse interrupt from letting
910 the kernel know of a new mouse event. The result is a latency
911 with the reaction time.
913 The irqsoff tracer tracks the time for which interrupts are
914 disabled. When a new maximum latency is hit, the tracer saves
915 the trace leading up to that latency point so that every time a
916 new maximum is reached, the old saved trace is discarded and the
919 To reset the maximum, echo 0 into tracing_max_latency. Here is
922 # echo 0 > options/function-trace
923 # echo irqsoff > current_tracer
924 # echo 1 > tracing_on
925 # echo 0 > tracing_max_latency
928 # echo 0 > tracing_on
932 # irqsoff latency trace v1.1.5 on 3.8.0-test+
933 # --------------------------------------------------------------------
934 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
936 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)
938 # => started at: run_timer_softirq
939 # => ended at: run_timer_softirq
943 # / _-----=> irqs-off
944 # | / _----=> need-resched
945 # || / _---=> hardirq/softirq
946 # ||| / _--=> preempt-depth
948 # cmd pid ||||| time | caller
950 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq
951 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq
952 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq
953 <idle>-0 0dNs3 25us : <stack trace>
954 => _raw_spin_unlock_irq
960 => smp_apic_timer_interrupt
961 => apic_timer_interrupt
966 => x86_64_start_reservations
967 => x86_64_start_kernel
969 Here we see that that we had a latency of 16 microseconds (which is
970 very good). The _raw_spin_lock_irq in run_timer_softirq disabled
971 interrupts. The difference between the 16 and the displayed
972 timestamp 25us occurred because the clock was incremented
973 between the time of recording the max latency and the time of
974 recording the function that had that latency.
976 Note the above example had function-trace not set. If we set
977 function-trace, we get a much larger output:
979 with echo 1 > options/function-trace
983 # irqsoff latency trace v1.1.5 on 3.8.0-test+
984 # --------------------------------------------------------------------
985 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
987 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)
989 # => started at: ata_scsi_queuecmd
990 # => ended at: ata_scsi_queuecmd
994 # / _-----=> irqs-off
995 # | / _----=> need-resched
996 # || / _---=> hardirq/softirq
997 # ||| / _--=> preempt-depth
999 # cmd pid ||||| time | caller
1001 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1002 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave
1003 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd
1004 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev
1005 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev
1006 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd
1007 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd
1008 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd
1009 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat
1011 bash-2042 3d..1 67us : delay_tsc <-__delay
1012 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1013 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc
1014 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc
1015 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc
1016 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue
1017 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1018 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1019 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd
1020 bash-2042 3d..1 120us : <stack trace>
1021 => _raw_spin_unlock_irqrestore
1022 => ata_scsi_queuecmd
1023 => scsi_dispatch_cmd
1025 => __blk_run_queue_uncond
1028 => generic_make_request
1031 => __ext3_get_inode_loc
1040 => user_path_at_empty
1045 => system_call_fastpath
1048 Here we traced a 71 microsecond latency. But we also see all the
1049 functions that were called during that time. Note that by
1050 enabling function tracing, we incur an added overhead. This
1051 overhead may extend the latency times. But nevertheless, this
1052 trace has provided some very helpful debugging information.
1058 When preemption is disabled, we may be able to receive
1059 interrupts but the task cannot be preempted and a higher
1060 priority task must wait for preemption to be enabled again
1061 before it can preempt a lower priority task.
1063 The preemptoff tracer traces the places that disable preemption.
1064 Like the irqsoff tracer, it records the maximum latency for
1065 which preemption was disabled. The control of preemptoff tracer
1066 is much like the irqsoff tracer.
1068 # echo 0 > options/function-trace
1069 # echo preemptoff > current_tracer
1070 # echo 1 > tracing_on
1071 # echo 0 > tracing_max_latency
1074 # echo 0 > tracing_on
1076 # tracer: preemptoff
1078 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1079 # --------------------------------------------------------------------
1080 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1082 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)
1084 # => started at: do_IRQ
1085 # => ended at: do_IRQ
1089 # / _-----=> irqs-off
1090 # | / _----=> need-resched
1091 # || / _---=> hardirq/softirq
1092 # ||| / _--=> preempt-depth
1094 # cmd pid ||||| time | caller
1096 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ
1097 sshd-1991 1d..1 46us : irq_exit <-do_IRQ
1098 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ
1099 sshd-1991 1d..1 52us : <stack trace>
1100 => sub_preempt_count
1106 This has some more changes. Preemption was disabled when an
1107 interrupt came in (notice the 'h'), and was enabled on exit.
1108 But we also see that interrupts have been disabled when entering
1109 the preempt off section and leaving it (the 'd'). We do not know if
1110 interrupts were enabled in the mean time or shortly after this
1113 # tracer: preemptoff
1115 # preemptoff latency trace v1.1.5 on 3.8.0-test+
1116 # --------------------------------------------------------------------
1117 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1119 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)
1121 # => started at: wake_up_new_task
1122 # => ended at: task_rq_unlock
1126 # / _-----=> irqs-off
1127 # | / _----=> need-resched
1128 # || / _---=> hardirq/softirq
1129 # ||| / _--=> preempt-depth
1131 # cmd pid ||||| time | caller
1133 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task
1134 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq
1135 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair
1136 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1137 bash-1994 1d..1 1us : source_load <-select_task_rq_fair
1139 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt
1140 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter
1141 bash-1994 1d..1 13us : add_preempt_count <-irq_enter
1142 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt
1143 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt
1144 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt
1145 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock
1146 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt
1148 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event
1149 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt
1150 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit
1151 bash-1994 1d..2 36us : do_softirq <-irq_exit
1152 bash-1994 1d..2 36us : __do_softirq <-call_softirq
1153 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq
1154 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq
1155 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq
1156 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock
1157 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq
1159 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks
1160 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq
1161 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable
1162 bash-1994 1dN.2 82us : idle_cpu <-irq_exit
1163 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit
1164 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit
1165 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock
1166 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock
1167 bash-1994 1.N.1 104us : <stack trace>
1168 => sub_preempt_count
1169 => _raw_spin_unlock_irqrestore
1177 The above is an example of the preemptoff trace with
1178 function-trace set. Here we see that interrupts were not disabled
1179 the entire time. The irq_enter code lets us know that we entered
1180 an interrupt 'h'. Before that, the functions being traced still
1181 show that it is not in an interrupt, but we can see from the
1182 functions themselves that this is not the case.
1187 Knowing the locations that have interrupts disabled or
1188 preemption disabled for the longest times is helpful. But
1189 sometimes we would like to know when either preemption and/or
1190 interrupts are disabled.
1192 Consider the following code:
1194 local_irq_disable();
1195 call_function_with_irqs_off();
1197 call_function_with_irqs_and_preemption_off();
1199 call_function_with_preemption_off();
1202 The irqsoff tracer will record the total length of
1203 call_function_with_irqs_off() and
1204 call_function_with_irqs_and_preemption_off().
1206 The preemptoff tracer will record the total length of
1207 call_function_with_irqs_and_preemption_off() and
1208 call_function_with_preemption_off().
1210 But neither will trace the time that interrupts and/or
1211 preemption is disabled. This total time is the time that we can
1212 not schedule. To record this time, use the preemptirqsoff
1215 Again, using this trace is much like the irqsoff and preemptoff
1218 # echo 0 > options/function-trace
1219 # echo preemptirqsoff > current_tracer
1220 # echo 1 > tracing_on
1221 # echo 0 > tracing_max_latency
1224 # echo 0 > tracing_on
1226 # tracer: preemptirqsoff
1228 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1229 # --------------------------------------------------------------------
1230 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1232 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)
1234 # => started at: ata_scsi_queuecmd
1235 # => ended at: ata_scsi_queuecmd
1239 # / _-----=> irqs-off
1240 # | / _----=> need-resched
1241 # || / _---=> hardirq/softirq
1242 # ||| / _--=> preempt-depth
1244 # cmd pid ||||| time | caller
1246 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd
1247 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd
1248 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd
1249 ls-2230 3...1 111us : <stack trace>
1250 => sub_preempt_count
1251 => _raw_spin_unlock_irqrestore
1252 => ata_scsi_queuecmd
1253 => scsi_dispatch_cmd
1255 => __blk_run_queue_uncond
1258 => generic_make_request
1263 => htree_dirblock_to_tree
1264 => ext3_htree_fill_tree
1268 => system_call_fastpath
1271 The trace_hardirqs_off_thunk is called from assembly on x86 when
1272 interrupts are disabled in the assembly code. Without the
1273 function tracing, we do not know if interrupts were enabled
1274 within the preemption points. We do see that it started with
1277 Here is a trace with function-trace set:
1279 # tracer: preemptirqsoff
1281 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+
1282 # --------------------------------------------------------------------
1283 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1285 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)
1287 # => started at: schedule
1288 # => ended at: mutex_unlock
1292 # / _-----=> irqs-off
1293 # | / _----=> need-resched
1294 # || / _---=> hardirq/softirq
1295 # ||| / _--=> preempt-depth
1297 # cmd pid ||||| time | caller
1299 kworker/-59 3...1 0us : __schedule <-schedule
1300 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch
1301 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq
1302 kworker/-59 3d..2 1us : deactivate_task <-__schedule
1303 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task
1304 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task
1305 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task
1306 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair
1307 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr
1308 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr
1309 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge
1310 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge
1311 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair
1312 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair
1313 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair
1314 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair
1315 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair
1316 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair
1317 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule
1318 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping
1319 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule
1320 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task
1321 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair
1322 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair
1323 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity
1324 ls-2269 3d..2 7us : finish_task_switch <-__schedule
1325 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch
1326 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr
1327 ls-2269 3d..2 8us : irq_enter <-do_IRQ
1328 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter
1329 ls-2269 3d..2 9us : add_preempt_count <-irq_enter
1330 ls-2269 3d.h2 9us : exit_idle <-do_IRQ
1332 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock
1333 ls-2269 3d.h2 20us : irq_exit <-do_IRQ
1334 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit
1335 ls-2269 3d..3 21us : do_softirq <-irq_exit
1336 ls-2269 3d..3 21us : __do_softirq <-call_softirq
1337 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq
1338 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip
1339 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip
1340 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr
1341 ls-2269 3d.s5 31us : irq_enter <-do_IRQ
1342 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1344 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter
1345 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter
1346 ls-2269 3d.H5 32us : exit_idle <-do_IRQ
1347 ls-2269 3d.H5 32us : handle_irq <-do_IRQ
1348 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq
1349 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq
1351 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll
1352 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action
1353 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq
1354 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable
1355 ls-2269 3d..3 159us : idle_cpu <-irq_exit
1356 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit
1357 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit
1358 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock
1359 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock
1360 ls-2269 3d... 186us : <stack trace>
1361 => __mutex_unlock_slowpath
1368 => system_call_fastpath
1370 This is an interesting trace. It started with kworker running and
1371 scheduling out and ls taking over. But as soon as ls released the
1372 rq lock and enabled interrupts (but not preemption) an interrupt
1373 triggered. When the interrupt finished, it started running softirqs.
1374 But while the softirq was running, another interrupt triggered.
1375 When an interrupt is running inside a softirq, the annotation is 'H'.
1381 One common case that people are interested in tracing is the
1382 time it takes for a task that is woken to actually wake up.
1383 Now for non Real-Time tasks, this can be arbitrary. But tracing
1384 it none the less can be interesting.
1386 Without function tracing:
1388 # echo 0 > options/function-trace
1389 # echo wakeup > current_tracer
1390 # echo 1 > tracing_on
1391 # echo 0 > tracing_max_latency
1393 # echo 0 > tracing_on
1397 # wakeup latency trace v1.1.5 on 3.8.0-test+
1398 # --------------------------------------------------------------------
1399 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1401 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)
1405 # / _-----=> irqs-off
1406 # | / _----=> need-resched
1407 # || / _---=> hardirq/softirq
1408 # ||| / _--=> preempt-depth
1410 # cmd pid ||||| time | caller
1412 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H
1413 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1414 <idle>-0 3d..3 15us : __schedule <-schedule
1415 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H
1417 The tracer only traces the highest priority task in the system
1418 to avoid tracing the normal circumstances. Here we see that
1419 the kworker with a nice priority of -20 (not very nice), took
1420 just 15 microseconds from the time it woke up, to the time it
1423 Non Real-Time tasks are not that interesting. A more interesting
1424 trace is to concentrate only on Real-Time tasks.
1429 In a Real-Time environment it is very important to know the
1430 wakeup time it takes for the highest priority task that is woken
1431 up to the time that it executes. This is also known as "schedule
1432 latency". I stress the point that this is about RT tasks. It is
1433 also important to know the scheduling latency of non-RT tasks,
1434 but the average schedule latency is better for non-RT tasks.
1435 Tools like LatencyTop are more appropriate for such
1438 Real-Time environments are interested in the worst case latency.
1439 That is the longest latency it takes for something to happen,
1440 and not the average. We can have a very fast scheduler that may
1441 only have a large latency once in a while, but that would not
1442 work well with Real-Time tasks. The wakeup_rt tracer was designed
1443 to record the worst case wakeups of RT tasks. Non-RT tasks are
1444 not recorded because the tracer only records one worst case and
1445 tracing non-RT tasks that are unpredictable will overwrite the
1446 worst case latency of RT tasks (just run the normal wakeup
1447 tracer for a while to see that effect).
1449 Since this tracer only deals with RT tasks, we will run this
1450 slightly differently than we did with the previous tracers.
1451 Instead of performing an 'ls', we will run 'sleep 1' under
1452 'chrt' which changes the priority of the task.
1454 # echo 0 > options/function-trace
1455 # echo wakeup_rt > current_tracer
1456 # echo 1 > tracing_on
1457 # echo 0 > tracing_max_latency
1459 # echo 0 > tracing_on
1465 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1466 # --------------------------------------------------------------------
1467 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1469 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)
1473 # / _-----=> irqs-off
1474 # | / _----=> need-resched
1475 # || / _---=> hardirq/softirq
1476 # ||| / _--=> preempt-depth
1478 # cmd pid ||||| time | caller
1480 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep
1481 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up
1482 <idle>-0 3d..3 5us : __schedule <-schedule
1483 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1486 Running this on an idle system, we see that it only took 5 microseconds
1487 to perform the task switch. Note, since the trace point in the schedule
1488 is before the actual "switch", we stop the tracing when the recorded task
1489 is about to schedule in. This may change if we add a new marker at the
1490 end of the scheduler.
1492 Notice that the recorded task is 'sleep' with the PID of 2389
1493 and it has an rt_prio of 5. This priority is user-space priority
1494 and not the internal kernel priority. The policy is 1 for
1495 SCHED_FIFO and 2 for SCHED_RR.
1497 Note, that the trace data shows the internal priority (99 - rtprio).
1499 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep
1501 The 0:120:R means idle was running with a nice priority of 0 (120 - 20)
1502 and in the running state 'R'. The sleep task was scheduled in with
1503 2389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)
1504 and it too is in the running state.
1506 Doing the same with chrt -r 5 and function-trace set.
1508 echo 1 > options/function-trace
1512 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1513 # --------------------------------------------------------------------
1514 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1516 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)
1520 # / _-----=> irqs-off
1521 # | / _----=> need-resched
1522 # || / _---=> hardirq/softirq
1523 # ||| / _--=> preempt-depth
1525 # cmd pid ||||| time | caller
1527 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep
1528 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1529 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup
1530 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr
1531 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup
1532 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up
1533 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock
1534 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up
1535 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up
1536 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1537 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer
1538 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock
1539 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt
1540 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock
1541 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt
1542 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event
1543 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event
1544 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event
1545 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt
1546 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit
1547 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit
1548 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit
1549 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit
1550 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit
1551 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle
1552 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit
1553 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle
1554 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit
1555 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit
1556 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit
1557 <idle>-0 3dN.1 13us : update_cpu_load_nohz <-tick_nohz_idle_exit
1558 <idle>-0 3dN.1 13us : _raw_spin_lock <-update_cpu_load_nohz
1559 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock
1560 <idle>-0 3dN.2 13us : __update_cpu_load <-update_cpu_load_nohz
1561 <idle>-0 3dN.2 14us : sched_avg_update <-__update_cpu_load
1562 <idle>-0 3dN.2 14us : _raw_spin_unlock <-update_cpu_load_nohz
1563 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock
1564 <idle>-0 3dN.1 15us : calc_load_exit_idle <-tick_nohz_idle_exit
1565 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit
1566 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit
1567 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel
1568 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel
1569 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1570 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave
1571 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16
1572 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer
1573 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram
1574 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event
1575 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event
1576 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event
1577 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel
1578 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1579 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit
1580 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1581 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward
1582 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
1583 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns
1584 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns
1585 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18
1586 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave
1587 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns
1588 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns
1589 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns
1590 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event
1591 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event
1592 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event
1593 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns
1594 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore
1595 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit
1596 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks
1597 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle
1598 <idle>-0 3.N.. 25us : schedule <-cpu_idle
1599 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule
1600 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule
1601 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule
1602 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch
1603 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch
1604 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule
1605 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq
1606 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule
1607 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task
1608 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task
1609 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt
1610 <idle>-0 3d..3 29us : __schedule <-preempt_schedule
1611 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep
1613 This isn't that big of a trace, even with function tracing enabled,
1614 so I included the entire trace.
1616 The interrupt went off while when the system was idle. Somewhere
1617 before task_woken_rt() was called, the NEED_RESCHED flag was set,
1618 this is indicated by the first occurrence of the 'N' flag.
1620 Latency tracing and events
1621 --------------------------
1622 As function tracing can induce a much larger latency, but without
1623 seeing what happens within the latency it is hard to know what
1624 caused it. There is a middle ground, and that is with enabling
1627 # echo 0 > options/function-trace
1628 # echo wakeup_rt > current_tracer
1629 # echo 1 > events/enable
1630 # echo 1 > tracing_on
1631 # echo 0 > tracing_max_latency
1633 # echo 0 > tracing_on
1637 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+
1638 # --------------------------------------------------------------------
1639 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)
1641 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)
1645 # / _-----=> irqs-off
1646 # | / _----=> need-resched
1647 # || / _---=> hardirq/softirq
1648 # ||| / _--=> preempt-depth
1650 # cmd pid ||||| time | caller
1652 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep
1653 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up
1654 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002
1655 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8
1656 <idle>-0 2.N.2 2us : power_end: cpu_id=2
1657 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2
1658 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0
1659 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000
1660 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch
1661 <idle>-0 2.N.2 5us : rcu_utilization: End context switch
1662 <idle>-0 2d..3 6us : __schedule <-schedule
1663 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep
1669 This tracer is the function tracer. Enabling the function tracer
1670 can be done from the debug file system. Make sure the
1671 ftrace_enabled is set; otherwise this tracer is a nop.
1672 See the "ftrace_enabled" section below.
1674 # sysctl kernel.ftrace_enabled=1
1675 # echo function > current_tracer
1676 # echo 1 > tracing_on
1678 # echo 0 > tracing_on
1682 # entries-in-buffer/entries-written: 24799/24799 #P:4
1685 # / _----=> need-resched
1686 # | / _---=> hardirq/softirq
1687 # || / _--=> preempt-depth
1689 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
1691 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write
1692 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock
1693 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify
1694 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify
1695 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify
1696 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock
1697 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock
1698 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify
1702 Note: function tracer uses ring buffers to store the above
1703 entries. The newest data may overwrite the oldest data.
1704 Sometimes using echo to stop the trace is not sufficient because
1705 the tracing could have overwritten the data that you wanted to
1706 record. For this reason, it is sometimes better to disable
1707 tracing directly from a program. This allows you to stop the
1708 tracing at the point that you hit the part that you are
1709 interested in. To disable the tracing directly from a C program,
1710 something like following code snippet can be used:
1714 int main(int argc, char *argv[]) {
1716 trace_fd = open(tracing_file("tracing_on"), O_WRONLY);
1718 if (condition_hit()) {
1719 write(trace_fd, "0", 1);
1725 Single thread tracing
1726 ---------------------
1728 By writing into set_ftrace_pid you can trace a
1729 single thread. For example:
1731 # cat set_ftrace_pid
1733 # echo 3111 > set_ftrace_pid
1734 # cat set_ftrace_pid
1736 # echo function > current_tracer
1740 # TASK-PID CPU# TIMESTAMP FUNCTION
1742 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return
1743 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range
1744 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel
1745 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel
1746 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll
1747 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll
1748 # echo > set_ftrace_pid
1752 # TASK-PID CPU# TIMESTAMP FUNCTION
1754 ##### CPU 3 buffer started ####
1755 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait
1756 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry
1757 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry
1758 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit
1759 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit
1761 If you want to trace a function when executing, you could use
1762 something like this simple program:
1766 #include <sys/types.h>
1767 #include <sys/stat.h>
1773 #define STR(x) _STR(x)
1774 #define MAX_PATH 256
1776 const char *find_debugfs(void)
1778 static char debugfs[MAX_PATH+1];
1779 static int debugfs_found;
1786 if ((fp = fopen("/proc/mounts","r")) == NULL) {
1787 perror("/proc/mounts");
1791 while (fscanf(fp, "%*s %"
1793 "s %99s %*s %*d %*d\n",
1794 debugfs, type) == 2) {
1795 if (strcmp(type, "debugfs") == 0)
1800 if (strcmp(type, "debugfs") != 0) {
1801 fprintf(stderr, "debugfs not mounted");
1805 strcat(debugfs, "/tracing/");
1811 const char *tracing_file(const char *file_name)
1813 static char trace_file[MAX_PATH+1];
1814 snprintf(trace_file, MAX_PATH, "%s/%s", find_debugfs(), file_name);
1818 int main (int argc, char **argv)
1828 ffd = open(tracing_file("current_tracer"), O_WRONLY);
1831 write(ffd, "nop", 3);
1833 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);
1834 s = sprintf(line, "%d\n", getpid());
1837 write(ffd, "function", 8);
1842 execvp(argv[1], argv+1);
1848 Or this simple script!
1853 debugfs=`sed -ne 's/^debugfs \(.*\) debugfs.*/\1/p' /proc/mounts`
1854 echo nop > $debugfs/tracing/current_tracer
1855 echo 0 > $debugfs/tracing/tracing_on
1856 echo $$ > $debugfs/tracing/set_ftrace_pid
1857 echo function > $debugfs/tracing/current_tracer
1858 echo 1 > $debugfs/tracing/tracing_on
1863 function graph tracer
1864 ---------------------------
1866 This tracer is similar to the function tracer except that it
1867 probes a function on its entry and its exit. This is done by
1868 using a dynamically allocated stack of return addresses in each
1869 task_struct. On function entry the tracer overwrites the return
1870 address of each function traced to set a custom probe. Thus the
1871 original return address is stored on the stack of return address
1874 Probing on both ends of a function leads to special features
1877 - measure of a function's time execution
1878 - having a reliable call stack to draw function calls graph
1880 This tracer is useful in several situations:
1882 - you want to find the reason of a strange kernel behavior and
1883 need to see what happens in detail on any areas (or specific
1886 - you are experiencing weird latencies but it's difficult to
1889 - you want to find quickly which path is taken by a specific
1892 - you just want to peek inside a working kernel and want to see
1895 # tracer: function_graph
1897 # CPU DURATION FUNCTION CALLS
1901 0) | do_sys_open() {
1903 0) | kmem_cache_alloc() {
1904 0) 1.382 us | __might_sleep();
1906 0) | strncpy_from_user() {
1907 0) | might_fault() {
1908 0) 1.389 us | __might_sleep();
1913 0) 0.668 us | _spin_lock();
1914 0) 0.570 us | expand_files();
1915 0) 0.586 us | _spin_unlock();
1918 There are several columns that can be dynamically
1919 enabled/disabled. You can use every combination of options you
1920 want, depending on your needs.
1922 - The cpu number on which the function executed is default
1923 enabled. It is sometimes better to only trace one cpu (see
1924 tracing_cpu_mask file) or you might sometimes see unordered
1925 function calls while cpu tracing switch.
1927 hide: echo nofuncgraph-cpu > trace_options
1928 show: echo funcgraph-cpu > trace_options
1930 - The duration (function's time of execution) is displayed on
1931 the closing bracket line of a function or on the same line
1932 than the current function in case of a leaf one. It is default
1935 hide: echo nofuncgraph-duration > trace_options
1936 show: echo funcgraph-duration > trace_options
1938 - The overhead field precedes the duration field in case of
1939 reached duration thresholds.
1941 hide: echo nofuncgraph-overhead > trace_options
1942 show: echo funcgraph-overhead > trace_options
1943 depends on: funcgraph-duration
1948 0) 0.646 us | _spin_lock_irqsave();
1949 0) 0.684 us | _spin_unlock_irqrestore();
1951 0) 0.548 us | fput();
1957 0) | kmem_cache_free() {
1958 0) 0.518 us | __phys_addr();
1964 + means that the function exceeded 10 usecs.
1965 ! means that the function exceeded 100 usecs.
1966 # means that the function exceeded 1000 usecs.
1967 $ means that the function exceeded 1 sec.
1970 - The task/pid field displays the thread cmdline and pid which
1971 executed the function. It is default disabled.
1973 hide: echo nofuncgraph-proc > trace_options
1974 show: echo funcgraph-proc > trace_options
1978 # tracer: function_graph
1980 # CPU TASK/PID DURATION FUNCTION CALLS
1982 0) sh-4802 | | d_free() {
1983 0) sh-4802 | | call_rcu() {
1984 0) sh-4802 | | __call_rcu() {
1985 0) sh-4802 | 0.616 us | rcu_process_gp_end();
1986 0) sh-4802 | 0.586 us | check_for_new_grace_period();
1987 0) sh-4802 | 2.899 us | }
1988 0) sh-4802 | 4.040 us | }
1989 0) sh-4802 | 5.151 us | }
1990 0) sh-4802 | + 49.370 us | }
1993 - The absolute time field is an absolute timestamp given by the
1994 system clock since it started. A snapshot of this time is
1995 given on each entry/exit of functions
1997 hide: echo nofuncgraph-abstime > trace_options
1998 show: echo funcgraph-abstime > trace_options
2003 # TIME CPU DURATION FUNCTION CALLS
2005 360.774522 | 1) 0.541 us | }
2006 360.774522 | 1) 4.663 us | }
2007 360.774523 | 1) 0.541 us | __wake_up_bit();
2008 360.774524 | 1) 6.796 us | }
2009 360.774524 | 1) 7.952 us | }
2010 360.774525 | 1) 9.063 us | }
2011 360.774525 | 1) 0.615 us | journal_mark_dirty();
2012 360.774527 | 1) 0.578 us | __brelse();
2013 360.774528 | 1) | reiserfs_prepare_for_journal() {
2014 360.774528 | 1) | unlock_buffer() {
2015 360.774529 | 1) | wake_up_bit() {
2016 360.774529 | 1) | bit_waitqueue() {
2017 360.774530 | 1) 0.594 us | __phys_addr();
2020 The function name is always displayed after the closing bracket
2021 for a function if the start of that function is not in the
2024 Display of the function name after the closing bracket may be
2025 enabled for functions whose start is in the trace buffer,
2026 allowing easier searching with grep for function durations.
2027 It is default disabled.
2029 hide: echo nofuncgraph-tail > trace_options
2030 show: echo funcgraph-tail > trace_options
2032 Example with nofuncgraph-tail (default):
2034 0) | kmem_cache_free() {
2035 0) 0.518 us | __phys_addr();
2039 Example with funcgraph-tail:
2041 0) | kmem_cache_free() {
2042 0) 0.518 us | __phys_addr();
2043 0) 1.757 us | } /* kmem_cache_free() */
2044 0) 2.861 us | } /* putname() */
2046 You can put some comments on specific functions by using
2047 trace_printk() For example, if you want to put a comment inside
2048 the __might_sleep() function, you just have to include
2049 <linux/ftrace.h> and call trace_printk() inside __might_sleep()
2051 trace_printk("I'm a comment!\n")
2055 1) | __might_sleep() {
2056 1) | /* I'm a comment! */
2060 You might find other useful features for this tracer in the
2061 following "dynamic ftrace" section such as tracing only specific
2067 If CONFIG_DYNAMIC_FTRACE is set, the system will run with
2068 virtually no overhead when function tracing is disabled. The way
2069 this works is the mcount function call (placed at the start of
2070 every kernel function, produced by the -pg switch in gcc),
2071 starts of pointing to a simple return. (Enabling FTRACE will
2072 include the -pg switch in the compiling of the kernel.)
2074 At compile time every C file object is run through the
2075 recordmcount program (located in the scripts directory). This
2076 program will parse the ELF headers in the C object to find all
2077 the locations in the .text section that call mcount. (Note, only
2078 white listed .text sections are processed, since processing other
2079 sections like .init.text may cause races due to those sections
2080 being freed unexpectedly).
2082 A new section called "__mcount_loc" is created that holds
2083 references to all the mcount call sites in the .text section.
2084 The recordmcount program re-links this section back into the
2085 original object. The final linking stage of the kernel will add all these
2086 references into a single table.
2088 On boot up, before SMP is initialized, the dynamic ftrace code
2089 scans this table and updates all the locations into nops. It
2090 also records the locations, which are added to the
2091 available_filter_functions list. Modules are processed as they
2092 are loaded and before they are executed. When a module is
2093 unloaded, it also removes its functions from the ftrace function
2094 list. This is automatic in the module unload code, and the
2095 module author does not need to worry about it.
2097 When tracing is enabled, the process of modifying the function
2098 tracepoints is dependent on architecture. The old method is to use
2099 kstop_machine to prevent races with the CPUs executing code being
2100 modified (which can cause the CPU to do undesirable things, especially
2101 if the modified code crosses cache (or page) boundaries), and the nops are
2102 patched back to calls. But this time, they do not call mcount
2103 (which is just a function stub). They now call into the ftrace
2106 The new method of modifying the function tracepoints is to place
2107 a breakpoint at the location to be modified, sync all CPUs, modify
2108 the rest of the instruction not covered by the breakpoint. Sync
2109 all CPUs again, and then remove the breakpoint with the finished
2110 version to the ftrace call site.
2112 Some archs do not even need to monkey around with the synchronization,
2113 and can just slap the new code on top of the old without any
2114 problems with other CPUs executing it at the same time.
2116 One special side-effect to the recording of the functions being
2117 traced is that we can now selectively choose which functions we
2118 wish to trace and which ones we want the mcount calls to remain
2121 Two files are used, one for enabling and one for disabling the
2122 tracing of specified functions. They are:
2130 A list of available functions that you can add to these files is
2133 available_filter_functions
2135 # cat available_filter_functions
2144 If I am only interested in sys_nanosleep and hrtimer_interrupt:
2146 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter
2147 # echo function > current_tracer
2148 # echo 1 > tracing_on
2150 # echo 0 > tracing_on
2154 # entries-in-buffer/entries-written: 5/5 #P:4
2157 # / _----=> need-resched
2158 # | / _---=> hardirq/softirq
2159 # || / _--=> preempt-depth
2161 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2163 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath
2164 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt
2165 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2166 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt
2167 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt
2169 To see which functions are being traced, you can cat the file:
2171 # cat set_ftrace_filter
2176 Perhaps this is not enough. The filters also allow simple wild
2177 cards. Only the following are currently available
2179 <match>* - will match functions that begin with <match>
2180 *<match> - will match functions that end with <match>
2181 *<match>* - will match functions that have <match> in it
2183 These are the only wild cards which are supported.
2185 <match>*<match> will not work.
2187 Note: It is better to use quotes to enclose the wild cards,
2188 otherwise the shell may expand the parameters into names
2189 of files in the local directory.
2191 # echo 'hrtimer_*' > set_ftrace_filter
2197 # entries-in-buffer/entries-written: 897/897 #P:4
2200 # / _----=> need-resched
2201 # | / _---=> hardirq/softirq
2202 # || / _--=> preempt-depth
2204 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2206 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit
2207 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel
2208 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer
2209 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit
2210 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11
2211 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt
2212 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter
2213 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem
2215 Notice that we lost the sys_nanosleep.
2217 # cat set_ftrace_filter
2222 hrtimer_try_to_cancel
2226 hrtimer_force_reprogram
2227 hrtimer_get_next_event
2231 hrtimer_get_remaining
2233 hrtimer_init_sleeper
2236 This is because the '>' and '>>' act just like they do in bash.
2237 To rewrite the filters, use '>'
2238 To append to the filters, use '>>'
2240 To clear out a filter so that all functions will be recorded
2243 # echo > set_ftrace_filter
2244 # cat set_ftrace_filter
2247 Again, now we want to append.
2249 # echo sys_nanosleep > set_ftrace_filter
2250 # cat set_ftrace_filter
2252 # echo 'hrtimer_*' >> set_ftrace_filter
2253 # cat set_ftrace_filter
2258 hrtimer_try_to_cancel
2262 hrtimer_force_reprogram
2263 hrtimer_get_next_event
2268 hrtimer_get_remaining
2270 hrtimer_init_sleeper
2273 The set_ftrace_notrace prevents those functions from being
2276 # echo '*preempt*' '*lock*' > set_ftrace_notrace
2282 # entries-in-buffer/entries-written: 39608/39608 #P:4
2285 # / _----=> need-resched
2286 # | / _---=> hardirq/softirq
2287 # || / _--=> preempt-depth
2289 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2291 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open
2292 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last
2293 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last
2294 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check
2295 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement
2296 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action
2297 bash-1994 [000] .... 4342.324899: do_truncate <-do_last
2298 bash-1994 [000] .... 4342.324899: should_remove_suid <-do_truncate
2299 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate
2300 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change
2301 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time
2302 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time
2304 We can see that there's no more lock or preempt tracing.
2307 Dynamic ftrace with the function graph tracer
2308 ---------------------------------------------
2310 Although what has been explained above concerns both the
2311 function tracer and the function-graph-tracer, there are some
2312 special features only available in the function-graph tracer.
2314 If you want to trace only one function and all of its children,
2315 you just have to echo its name into set_graph_function:
2317 echo __do_fault > set_graph_function
2319 will produce the following "expanded" trace of the __do_fault()
2323 0) | filemap_fault() {
2324 0) | find_lock_page() {
2325 0) 0.804 us | find_get_page();
2326 0) | __might_sleep() {
2330 0) 0.653 us | _spin_lock();
2331 0) 0.578 us | page_add_file_rmap();
2332 0) 0.525 us | native_set_pte_at();
2333 0) 0.585 us | _spin_unlock();
2334 0) | unlock_page() {
2335 0) 0.541 us | page_waitqueue();
2336 0) 0.639 us | __wake_up_bit();
2340 0) | filemap_fault() {
2341 0) | find_lock_page() {
2342 0) 0.698 us | find_get_page();
2343 0) | __might_sleep() {
2347 0) 0.631 us | _spin_lock();
2348 0) 0.571 us | page_add_file_rmap();
2349 0) 0.526 us | native_set_pte_at();
2350 0) 0.586 us | _spin_unlock();
2351 0) | unlock_page() {
2352 0) 0.533 us | page_waitqueue();
2353 0) 0.638 us | __wake_up_bit();
2357 You can also expand several functions at once:
2359 echo sys_open > set_graph_function
2360 echo sys_close >> set_graph_function
2362 Now if you want to go back to trace all functions you can clear
2363 this special filter via:
2365 echo > set_graph_function
2371 Note, the proc sysctl ftrace_enable is a big on/off switch for the
2372 function tracer. By default it is enabled (when function tracing is
2373 enabled in the kernel). If it is disabled, all function tracing is
2374 disabled. This includes not only the function tracers for ftrace, but
2375 also for any other uses (perf, kprobes, stack tracing, profiling, etc).
2377 Please disable this with care.
2379 This can be disable (and enabled) with:
2381 sysctl kernel.ftrace_enabled=0
2382 sysctl kernel.ftrace_enabled=1
2386 echo 0 > /proc/sys/kernel/ftrace_enabled
2387 echo 1 > /proc/sys/kernel/ftrace_enabled
2393 A few commands are supported by the set_ftrace_filter interface.
2394 Trace commands have the following format:
2396 <function>:<command>:<parameter>
2398 The following commands are supported:
2401 This command enables function filtering per module. The
2402 parameter defines the module. For example, if only the write*
2403 functions in the ext3 module are desired, run:
2405 echo 'write*:mod:ext3' > set_ftrace_filter
2407 This command interacts with the filter in the same way as
2408 filtering based on function names. Thus, adding more functions
2409 in a different module is accomplished by appending (>>) to the
2410 filter file. Remove specific module functions by prepending
2413 echo '!writeback*:mod:ext3' >> set_ftrace_filter
2416 These commands turn tracing on and off when the specified
2417 functions are hit. The parameter determines how many times the
2418 tracing system is turned on and off. If unspecified, there is
2419 no limit. For example, to disable tracing when a schedule bug
2420 is hit the first 5 times, run:
2422 echo '__schedule_bug:traceoff:5' > set_ftrace_filter
2424 To always disable tracing when __schedule_bug is hit:
2426 echo '__schedule_bug:traceoff' > set_ftrace_filter
2428 These commands are cumulative whether or not they are appended
2429 to set_ftrace_filter. To remove a command, prepend it by '!'
2430 and drop the parameter:
2432 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter
2434 The above removes the traceoff command for __schedule_bug
2435 that have a counter. To remove commands without counters:
2437 echo '!__schedule_bug:traceoff' > set_ftrace_filter
2440 Will cause a snapshot to be triggered when the function is hit.
2442 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter
2444 To only snapshot once:
2446 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter
2448 To remove the above commands:
2450 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter
2451 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter
2453 - enable_event/disable_event
2454 These commands can enable or disable a trace event. Note, because
2455 function tracing callbacks are very sensitive, when these commands
2456 are registered, the trace point is activated, but disabled in
2457 a "soft" mode. That is, the tracepoint will be called, but
2458 just will not be traced. The event tracepoint stays in this mode
2459 as long as there's a command that triggers it.
2461 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \
2466 <function>:enable_event:<system>:<event>[:count]
2467 <function>:disable_event:<system>:<event>[:count]
2469 To remove the events commands:
2472 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \
2474 echo '!schedule:disable_event:sched:sched_switch' > \
2478 When the function is hit, it will dump the contents of the ftrace
2479 ring buffer to the console. This is useful if you need to debug
2480 something, and want to dump the trace when a certain function
2481 is hit. Perhaps its a function that is called before a tripple
2482 fault happens and does not allow you to get a regular dump.
2485 When the function is hit, it will dump the contents of the ftrace
2486 ring buffer for the current CPU to the console. Unlike the "dump"
2487 command, it only prints out the contents of the ring buffer for the
2488 CPU that executed the function that triggered the dump.
2493 The trace_pipe outputs the same content as the trace file, but
2494 the effect on the tracing is different. Every read from
2495 trace_pipe is consumed. This means that subsequent reads will be
2496 different. The trace is live.
2498 # echo function > current_tracer
2499 # cat trace_pipe > /tmp/trace.out &
2501 # echo 1 > tracing_on
2503 # echo 0 > tracing_on
2507 # entries-in-buffer/entries-written: 0/0 #P:4
2510 # / _----=> need-resched
2511 # | / _---=> hardirq/softirq
2512 # || / _--=> preempt-depth
2514 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2518 # cat /tmp/trace.out
2519 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write
2520 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock
2521 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify
2522 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify
2523 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify
2524 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock
2525 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock
2526 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify
2527 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath
2530 Note, reading the trace_pipe file will block until more input is
2536 Having too much or not enough data can be troublesome in
2537 diagnosing an issue in the kernel. The file buffer_size_kb is
2538 used to modify the size of the internal trace buffers. The
2539 number listed is the number of entries that can be recorded per
2540 CPU. To know the full size, multiply the number of possible CPUs
2541 with the number of entries.
2543 # cat buffer_size_kb
2544 1408 (units kilobytes)
2546 Or simply read buffer_total_size_kb
2548 # cat buffer_total_size_kb
2551 To modify the buffer, simple echo in a number (in 1024 byte segments).
2553 # echo 10000 > buffer_size_kb
2554 # cat buffer_size_kb
2555 10000 (units kilobytes)
2557 It will try to allocate as much as possible. If you allocate too
2558 much, it can cause Out-Of-Memory to trigger.
2560 # echo 1000000000000 > buffer_size_kb
2561 -bash: echo: write error: Cannot allocate memory
2562 # cat buffer_size_kb
2565 The per_cpu buffers can be changed individually as well:
2567 # echo 10000 > per_cpu/cpu0/buffer_size_kb
2568 # echo 100 > per_cpu/cpu1/buffer_size_kb
2570 When the per_cpu buffers are not the same, the buffer_size_kb
2571 at the top level will just show an X
2573 # cat buffer_size_kb
2576 This is where the buffer_total_size_kb is useful:
2578 # cat buffer_total_size_kb
2581 Writing to the top level buffer_size_kb will reset all the buffers
2582 to be the same again.
2586 CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature
2587 available to all non latency tracers. (Latency tracers which
2588 record max latency, such as "irqsoff" or "wakeup", can't use
2589 this feature, since those are already using the snapshot
2590 mechanism internally.)
2592 Snapshot preserves a current trace buffer at a particular point
2593 in time without stopping tracing. Ftrace swaps the current
2594 buffer with a spare buffer, and tracing continues in the new
2595 current (=previous spare) buffer.
2597 The following debugfs files in "tracing" are related to this
2602 This is used to take a snapshot and to read the output
2603 of the snapshot. Echo 1 into this file to allocate a
2604 spare buffer and to take a snapshot (swap), then read
2605 the snapshot from this file in the same format as
2606 "trace" (described above in the section "The File
2607 System"). Both reads snapshot and tracing are executable
2608 in parallel. When the spare buffer is allocated, echoing
2609 0 frees it, and echoing else (positive) values clear the
2611 More details are shown in the table below.
2613 status\input | 0 | 1 | else |
2614 --------------+------------+------------+------------+
2615 not allocated |(do nothing)| alloc+swap |(do nothing)|
2616 --------------+------------+------------+------------+
2617 allocated | free | swap | clear |
2618 --------------+------------+------------+------------+
2620 Here is an example of using the snapshot feature.
2622 # echo 1 > events/sched/enable
2627 # entries-in-buffer/entries-written: 71/71 #P:8
2630 # / _----=> need-resched
2631 # | / _---=> hardirq/softirq
2632 # || / _--=> preempt-depth
2634 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2636 <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
2637 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
2639 <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
2644 # entries-in-buffer/entries-written: 77/77 #P:8
2647 # / _----=> need-resched
2648 # | / _---=> hardirq/softirq
2649 # || / _--=> preempt-depth
2651 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2653 <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
2654 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
2658 If you try to use this snapshot feature when current tracer is
2659 one of the latency tracers, you will get the following results.
2661 # echo wakeup > current_tracer
2663 bash: echo: write error: Device or resource busy
2665 cat: snapshot: Device or resource busy
2670 In the debugfs tracing directory is a directory called "instances".
2671 This directory can have new directories created inside of it using
2672 mkdir, and removing directories with rmdir. The directory created
2673 with mkdir in this directory will already contain files and other
2674 directories after it is created.
2676 # mkdir instances/foo
2678 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu
2679 set_event snapshot trace trace_clock trace_marker trace_options
2680 trace_pipe tracing_on
2682 As you can see, the new directory looks similar to the tracing directory
2683 itself. In fact, it is very similar, except that the buffer and
2684 events are agnostic from the main director, or from any other
2685 instances that are created.
2687 The files in the new directory work just like the files with the
2688 same name in the tracing directory except the buffer that is used
2689 is a separate and new buffer. The files affect that buffer but do not
2690 affect the main buffer with the exception of trace_options. Currently,
2691 the trace_options affect all instances and the top level buffer
2692 the same, but this may change in future releases. That is, options
2693 may become specific to the instance they reside in.
2695 Notice that none of the function tracer files are there, nor is
2696 current_tracer and available_tracers. This is because the buffers
2697 can currently only have events enabled for them.
2699 # mkdir instances/foo
2700 # mkdir instances/bar
2701 # mkdir instances/zoot
2702 # echo 100000 > buffer_size_kb
2703 # echo 1000 > instances/foo/buffer_size_kb
2704 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb
2705 # echo function > current_trace
2706 # echo 1 > instances/foo/events/sched/sched_wakeup/enable
2707 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable
2708 # echo 1 > instances/foo/events/sched/sched_switch/enable
2709 # echo 1 > instances/bar/events/irq/enable
2710 # echo 1 > instances/zoot/events/syscalls/enable
2712 CPU:2 [LOST 11745 EVENTS]
2713 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist
2714 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave
2715 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist
2716 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist
2717 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock
2718 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype
2719 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist
2720 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist
2721 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
2722 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics
2723 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process
2726 # cat instances/foo/trace_pipe
2727 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
2728 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
2729 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003
2730 <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
2731 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
2732 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000
2733 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000
2734 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
2735 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001
2736 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
2739 # cat instances/bar/trace_pipe
2740 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX]
2741 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX]
2742 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER]
2743 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU]
2744 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER]
2745 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER]
2746 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU]
2747 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU]
2748 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4
2749 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled
2750 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0
2751 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled
2754 # cat instances/zoot/trace
2757 # entries-in-buffer/entries-written: 18996/18996 #P:4
2760 # / _----=> need-resched
2761 # | / _---=> hardirq/softirq
2762 # || / _--=> preempt-depth
2764 # TASK-PID CPU# |||| TIMESTAMP FUNCTION
2766 bash-1998 [000] d... 140.733501: sys_write -> 0x2
2767 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1)
2768 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1
2769 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)
2770 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1
2771 bash-1998 [000] d... 140.733510: sys_close(fd: a)
2772 bash-1998 [000] d... 140.733510: sys_close -> 0x0
2773 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)
2774 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0
2775 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)
2776 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0
2778 You can see that the trace of the top most trace buffer shows only
2779 the function tracing. The foo instance displays wakeups and task
2782 To remove the instances, simply delete their directories:
2784 # rmdir instances/foo
2785 # rmdir instances/bar
2786 # rmdir instances/zoot
2788 Note, if a process has a trace file open in one of the instance
2789 directories, the rmdir will fail with EBUSY.
2794 Since the kernel has a fixed sized stack, it is important not to
2795 waste it in functions. A kernel developer must be conscience of
2796 what they allocate on the stack. If they add too much, the system
2797 can be in danger of a stack overflow, and corruption will occur,
2798 usually leading to a system panic.
2800 There are some tools that check this, usually with interrupts
2801 periodically checking usage. But if you can perform a check
2802 at every function call that will become very useful. As ftrace provides
2803 a function tracer, it makes it convenient to check the stack size
2804 at every function call. This is enabled via the stack tracer.
2806 CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.
2807 To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.
2809 # echo 1 > /proc/sys/kernel/stack_tracer_enabled
2811 You can also enable it from the kernel command line to trace
2812 the stack size of the kernel during boot up, by adding "stacktrace"
2813 to the kernel command line parameter.
2815 After running it for a few minutes, the output looks like:
2817 # cat stack_max_size
2821 Depth Size Location (18 entries)
2823 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac
2824 1) 2704 160 find_busiest_group+0x31/0x1f1
2825 2) 2544 256 load_balance+0xd9/0x662
2826 3) 2288 80 idle_balance+0xbb/0x130
2827 4) 2208 128 __schedule+0x26e/0x5b9
2828 5) 2080 16 schedule+0x64/0x66
2829 6) 2064 128 schedule_timeout+0x34/0xe0
2830 7) 1936 112 wait_for_common+0x97/0xf1
2831 8) 1824 16 wait_for_completion+0x1d/0x1f
2832 9) 1808 128 flush_work+0xfe/0x119
2833 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20
2834 11) 1664 48 input_available_p+0x1d/0x5c
2835 12) 1616 48 n_tty_poll+0x6d/0x134
2836 13) 1568 64 tty_poll+0x64/0x7f
2837 14) 1504 880 do_select+0x31e/0x511
2838 15) 624 400 core_sys_select+0x177/0x216
2839 16) 224 96 sys_select+0x91/0xb9
2840 17) 128 128 system_call_fastpath+0x16/0x1b
2842 Note, if -mfentry is being used by gcc, functions get traced before
2843 they set up the stack frame. This means that leaf level functions
2844 are not tested by the stack tracer when -mfentry is used.
2846 Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.
2850 More details can be found in the source code, in the
2851 kernel/trace/*.c files.
projects / deliverable / linux.git / blob