brintos / linux-shallow public Read only

1========================2ftrace - Function Tracer3========================4 5Copyright 2008 Red Hat Inc.6 7:Author:   Steven Rostedt <srostedt@redhat.com>8:License:  The GNU Free Documentation License, Version 1.29          (dual licensed under the GPL v2)10:Original Reviewers:  Elias Oltmanns, Randy Dunlap, Andrew Morton,11		      John Kacur, and David Teigland.12 13- Written for: 2.6.28-rc214- Updated for: 3.1015- Updated for: 4.13 - Copyright 2017 VMware Inc. Steven Rostedt16- Converted to rst format - Changbin Du <changbin.du@intel.com>17 18Introduction19------------20 21Ftrace is an internal tracer designed to help out developers and22designers of systems to find what is going on inside the kernel.23It can be used for debugging or analyzing latencies and24performance issues that take place outside of user-space.25 26Although ftrace is typically considered the function tracer, it27is really a framework of several assorted tracing utilities.28There's latency tracing to examine what occurs between interrupts29disabled and enabled, as well as for preemption and from a time30a task is woken to the task is actually scheduled in.31 32One of the most common uses of ftrace is the event tracing.33Throughout the kernel is hundreds of static event points that34can be enabled via the tracefs file system to see what is35going on in certain parts of the kernel.36 37See events.rst for more information.38 39 40Implementation Details41----------------------42 43See Documentation/trace/ftrace-design.rst for details for arch porters and such.44 45 46The File System47---------------48 49Ftrace uses the tracefs file system to hold the control files as50well as the files to display output.51 52When tracefs is configured into the kernel (which selecting any ftrace53option will do) the directory /sys/kernel/tracing will be created. To mount54this directory, you can add to your /etc/fstab file::55 56 tracefs       /sys/kernel/tracing       tracefs defaults        0       057 58Or you can mount it at run time with::59 60 mount -t tracefs nodev /sys/kernel/tracing61 62For quicker access to that directory you may want to make a soft link to63it::64 65 ln -s /sys/kernel/tracing /tracing66 67.. attention::68 69  Before 4.1, all ftrace tracing control files were within the debugfs70  file system, which is typically located at /sys/kernel/debug/tracing.71  For backward compatibility, when mounting the debugfs file system,72  the tracefs file system will be automatically mounted at:73 74  /sys/kernel/debug/tracing75 76  All files located in the tracefs file system will be located in that77  debugfs file system directory as well.78 79.. attention::80 81  Any selected ftrace option will also create the tracefs file system.82  The rest of the document will assume that you are in the ftrace directory83  (cd /sys/kernel/tracing) and will only concentrate on the files within that84  directory and not distract from the content with the extended85  "/sys/kernel/tracing" path name.86 87That's it! (assuming that you have ftrace configured into your kernel)88 89After mounting tracefs you will have access to the control and output files90of ftrace. Here is a list of some of the key files:91 92 93 Note: all time values are in microseconds.94 95  current_tracer:96 97	This is used to set or display the current tracer98	that is configured. Changing the current tracer clears99	the ring buffer content as well as the "snapshot" buffer.100 101  available_tracers:102 103	This holds the different types of tracers that104	have been compiled into the kernel. The105	tracers listed here can be configured by106	echoing their name into current_tracer.107 108  tracing_on:109 110	This sets or displays whether writing to the trace111	ring buffer is enabled. Echo 0 into this file to disable112	the tracer or 1 to enable it. Note, this only disables113	writing to the ring buffer, the tracing overhead may114	still be occurring.115 116	The kernel function tracing_off() can be used within the117	kernel to disable writing to the ring buffer, which will118	set this file to "0". User space can re-enable tracing by119	echoing "1" into the file.120 121	Note, the function and event trigger "traceoff" will also122	set this file to zero and stop tracing. Which can also123	be re-enabled by user space using this file.124 125  trace:126 127	This file holds the output of the trace in a human128	readable format (described below). Opening this file for129	writing with the O_TRUNC flag clears the ring buffer content.130        Note, this file is not a consumer. If tracing is off131        (no tracer running, or tracing_on is zero), it will produce132        the same output each time it is read. When tracing is on,133        it may produce inconsistent results as it tries to read134        the entire buffer without consuming it.135 136  trace_pipe:137 138	The output is the same as the "trace" file but this139	file is meant to be streamed with live tracing.140	Reads from this file will block until new data is141	retrieved.  Unlike the "trace" file, this file is a142	consumer. This means reading from this file causes143	sequential reads to display more current data. Once144	data is read from this file, it is consumed, and145	will not be read again with a sequential read. The146	"trace" file is static, and if the tracer is not147	adding more data, it will display the same148	information every time it is read.149 150  trace_options:151 152	This file lets the user control the amount of data153	that is displayed in one of the above output154	files. Options also exist to modify how a tracer155	or events work (stack traces, timestamps, etc).156 157  options:158 159	This is a directory that has a file for every available160	trace option (also in trace_options). Options may also be set161	or cleared by writing a "1" or "0" respectively into the162	corresponding file with the option name.163 164  tracing_max_latency:165 166	Some of the tracers record the max latency.167	For example, the maximum time that interrupts are disabled.168	The maximum time is saved in this file. The max trace will also be169	stored,	and displayed by "trace". A new max trace will only be170	recorded if the latency is greater than the value in this file171	(in microseconds).172 173	By echoing in a time into this file, no latency will be recorded174	unless it is greater than the time in this file.175 176  tracing_thresh:177 178	Some latency tracers will record a trace whenever the179	latency is greater than the number in this file.180	Only active when the file contains a number greater than 0.181	(in microseconds)182 183  buffer_percent:184 185	This is the watermark for how much the ring buffer needs to be filled186	before a waiter is woken up. That is, if an application calls a187	blocking read syscall on one of the per_cpu trace_pipe_raw files, it188	will block until the given amount of data specified by buffer_percent189	is in the ring buffer before it wakes the reader up. This also190	controls how the splice system calls are blocked on this file::191 192	  0   - means to wake up as soon as there is any data in the ring buffer.193	  50  - means to wake up when roughly half of the ring buffer sub-buffers194	        are full.195	  100 - means to block until the ring buffer is totally full and is196	        about to start overwriting the older data.197 198  buffer_size_kb:199 200	This sets or displays the number of kilobytes each CPU201	buffer holds. By default, the trace buffers are the same size202	for each CPU. The displayed number is the size of the203	CPU buffer and not total size of all buffers. The204	trace buffers are allocated in pages (blocks of memory205	that the kernel uses for allocation, usually 4 KB in size).206	A few extra pages may be allocated to accommodate buffer management207	meta-data. If the last page allocated has room for more bytes208	than requested, the rest of the page will be used,209	making the actual allocation bigger than requested or shown.210	( Note, the size may not be a multiple of the page size211	due to buffer management meta-data. )212 213	Buffer sizes for individual CPUs may vary214	(see "per_cpu/cpu0/buffer_size_kb" below), and if they do215	this file will show "X".216 217  buffer_total_size_kb:218 219	This displays the total combined size of all the trace buffers.220 221  buffer_subbuf_size_kb:222 223	This sets or displays the sub buffer size. The ring buffer is broken up224	into several same size "sub buffers". An event can not be bigger than225	the size of the sub buffer. Normally, the sub buffer is the size of the226	architecture's page (4K on x86). The sub buffer also contains meta data227	at the start which also limits the size of an event.  That means when228	the sub buffer is a page size, no event can be larger than the page229	size minus the sub buffer meta data.230 231	Note, the buffer_subbuf_size_kb is a way for the user to specify the232	minimum size of the subbuffer. The kernel may make it bigger due to the233	implementation details, or simply fail the operation if the kernel can234	not handle the request.235 236	Changing the sub buffer size allows for events to be larger than the237	page size.238 239	Note: When changing the sub-buffer size, tracing is stopped and any240	data in the ring buffer and the snapshot buffer will be discarded.241 242  free_buffer:243 244	If a process is performing tracing, and the ring buffer	should be245	shrunk "freed" when the process is finished, even if it were to be246	killed by a signal, this file can be used for that purpose. On close247	of this file, the ring buffer will be resized to its minimum size.248	Having a process that is tracing also open this file, when the process249	exits its file descriptor for this file will be closed, and in doing so,250	the ring buffer will be "freed".251 252	It may also stop tracing if disable_on_free option is set.253 254  tracing_cpumask:255 256	This is a mask that lets the user only trace on specified CPUs.257	The format is a hex string representing the CPUs.258 259  set_ftrace_filter:260 261	When dynamic ftrace is configured in (see the262	section below "dynamic ftrace"), the code is dynamically263	modified (code text rewrite) to disable calling of the264	function profiler (mcount). This lets tracing be configured265	in with practically no overhead in performance.  This also266	has a side effect of enabling or disabling specific functions267	to be traced. Echoing names of functions into this file268	will limit the trace to only those functions.269	This influences the tracers "function" and "function_graph"270	and thus also function profiling (see "function_profile_enabled").271 272	The functions listed in "available_filter_functions" are what273	can be written into this file.274 275	This interface also allows for commands to be used. See the276	"Filter commands" section for more details.277 278	As a speed up, since processing strings can be quite expensive279	and requires a check of all functions registered to tracing, instead280	an index can be written into this file. A number (starting with "1")281	written will instead select the same corresponding at the line position282	of the "available_filter_functions" file.283 284  set_ftrace_notrace:285 286	This has an effect opposite to that of287	set_ftrace_filter. Any function that is added here will not288	be traced. If a function exists in both set_ftrace_filter289	and set_ftrace_notrace,	the function will _not_ be traced.290 291  set_ftrace_pid:292 293	Have the function tracer only trace the threads whose PID are294	listed in this file.295 296	If the "function-fork" option is set, then when a task whose297	PID is listed in this file forks, the child's PID will298	automatically be added to this file, and the child will be299	traced by the function tracer as well. This option will also300	cause PIDs of tasks that exit to be removed from the file.301 302  set_ftrace_notrace_pid:303 304        Have the function tracer ignore threads whose PID are listed in305        this file.306 307        If the "function-fork" option is set, then when a task whose308	PID is listed in this file forks, the child's PID will309	automatically be added to this file, and the child will not be310	traced by the function tracer as well. This option will also311	cause PIDs of tasks that exit to be removed from the file.312 313        If a PID is in both this file and "set_ftrace_pid", then this314        file takes precedence, and the thread will not be traced.315 316  set_event_pid:317 318	Have the events only trace a task with a PID listed in this file.319	Note, sched_switch and sched_wake_up will also trace events320	listed in this file.321 322	To have the PIDs of children of tasks with their PID in this file323	added on fork, enable the "event-fork" option. That option will also324	cause the PIDs of tasks to be removed from this file when the task325	exits.326 327  set_event_notrace_pid:328 329	Have the events not trace a task with a PID listed in this file.330	Note, sched_switch and sched_wakeup will trace threads not listed331	in this file, even if a thread's PID is in the file if the332        sched_switch or sched_wakeup events also trace a thread that should333        be traced.334 335	To have the PIDs of children of tasks with their PID in this file336	added on fork, enable the "event-fork" option. That option will also337	cause the PIDs of tasks to be removed from this file when the task338	exits.339 340  set_graph_function:341 342	Functions listed in this file will cause the function graph343	tracer to only trace these functions and the functions that344	they call. (See the section "dynamic ftrace" for more details).345	Note, set_ftrace_filter and set_ftrace_notrace still affects346	what functions are being traced.347 348  set_graph_notrace:349 350	Similar to set_graph_function, but will disable function graph351	tracing when the function is hit until it exits the function.352	This makes it possible to ignore tracing functions that are called353	by a specific function.354 355  available_filter_functions:356 357	This lists the functions that ftrace has processed and can trace.358	These are the function names that you can pass to359	"set_ftrace_filter", "set_ftrace_notrace",360	"set_graph_function", or "set_graph_notrace".361	(See the section "dynamic ftrace" below for more details.)362 363  available_filter_functions_addrs:364 365	Similar to available_filter_functions, but with address displayed366	for each function. The displayed address is the patch-site address367	and can differ from /proc/kallsyms address.368 369  dyn_ftrace_total_info:370 371	This file is for debugging purposes. The number of functions that372	have been converted to nops and are available to be traced.373 374  enabled_functions:375 376	This file is more for debugging ftrace, but can also be useful377	in seeing if any function has a callback attached to it.378	Not only does the trace infrastructure use ftrace function379	trace utility, but other subsystems might too. This file380	displays all functions that have a callback attached to them381	as well as the number of callbacks that have been attached.382	Note, a callback may also call multiple functions which will383	not be listed in this count.384 385	If the callback registered to be traced by a function with386	the "save regs" attribute (thus even more overhead), a 'R'387	will be displayed on the same line as the function that388	is returning registers.389 390	If the callback registered to be traced by a function with391	the "ip modify" attribute (thus the regs->ip can be changed),392	an 'I' will be displayed on the same line as the function that393	can be overridden.394 395	If a non ftrace trampoline is attached (BPF) a 'D' will be displayed.396	Note, normal ftrace trampolines can also be attached, but only one397	"direct" trampoline can be attached to a given function at a time.398 399	Some architectures can not call direct trampolines, but instead have400	the ftrace ops function located above the function entry point. In401	such cases an 'O' will be displayed.402 403	If a function had either the "ip modify" or a "direct" call attached to404	it in the past, a 'M' will be shown. This flag is never cleared. It is405	used to know if a function was every modified by the ftrace infrastructure,406	and can be used for debugging.407 408	If the architecture supports it, it will also show what callback409	is being directly called by the function. If the count is greater410	than 1 it most likely will be ftrace_ops_list_func().411 412	If the callback of a function jumps to a trampoline that is413	specific to the callback and which is not the standard trampoline,414	its address will be printed as well as the function that the415	trampoline calls.416 417  touched_functions:418 419	This file contains all the functions that ever had a function callback420	to it via the ftrace infrastructure. It has the same format as421	enabled_functions but shows all functions that have every been422	traced.423 424	To see any function that has every been modified by "ip modify" or a425	direct trampoline, one can perform the following command:426 427	grep ' M ' /sys/kernel/tracing/touched_functions428 429  function_profile_enabled:430 431	When set it will enable all functions with either the function432	tracer, or if configured, the function graph tracer. It will433	keep a histogram of the number of functions that were called434	and if the function graph tracer was configured, it will also keep435	track of the time spent in those functions. The histogram436	content can be displayed in the files:437 438	trace_stat/function<cpu> ( function0, function1, etc).439 440  trace_stat:441 442	A directory that holds different tracing stats.443 444  kprobe_events:445 446	Enable dynamic trace points. See kprobetrace.rst.447 448  kprobe_profile:449 450	Dynamic trace points stats. See kprobetrace.rst.451 452  max_graph_depth:453 454	Used with the function graph tracer. This is the max depth455	it will trace into a function. Setting this to a value of456	one will show only the first kernel function that is called457	from user space.458 459  printk_formats:460 461	This is for tools that read the raw format files. If an event in462	the ring buffer references a string, only a pointer to the string463	is recorded into the buffer and not the string itself. This prevents464	tools from knowing what that string was. This file displays the string465	and address for	the string allowing tools to map the pointers to what466	the strings were.467 468  saved_cmdlines:469 470	Only the pid of the task is recorded in a trace event unless471	the event specifically saves the task comm as well. Ftrace472	makes a cache of pid mappings to comms to try to display473	comms for events. If a pid for a comm is not listed, then474	"<...>" is displayed in the output.475 476	If the option "record-cmd" is set to "0", then comms of tasks477	will not be saved during recording. By default, it is enabled.478 479  saved_cmdlines_size:480 481	By default, 128 comms are saved (see "saved_cmdlines" above). To482	increase or decrease the amount of comms that are cached, echo483	the number of comms to cache into this file.484 485  saved_tgids:486 487	If the option "record-tgid" is set, on each scheduling context switch488	the Task Group ID of a task is saved in a table mapping the PID of489	the thread to its TGID. By default, the "record-tgid" option is490	disabled.491 492  snapshot:493 494	This displays the "snapshot" buffer and also lets the user495	take a snapshot of the current running trace.496	See the "Snapshot" section below for more details.497 498  stack_max_size:499 500	When the stack tracer is activated, this will display the501	maximum stack size it has encountered.502	See the "Stack Trace" section below.503 504  stack_trace:505 506	This displays the stack back trace of the largest stack507	that was encountered when the stack tracer is activated.508	See the "Stack Trace" section below.509 510  stack_trace_filter:511 512	This is similar to "set_ftrace_filter" but it limits what513	functions the stack tracer will check.514 515  trace_clock:516 517	Whenever an event is recorded into the ring buffer, a518	"timestamp" is added. This stamp comes from a specified519	clock. By default, ftrace uses the "local" clock. This520	clock is very fast and strictly per cpu, but on some521	systems it may not be monotonic with respect to other522	CPUs. In other words, the local clocks may not be in sync523	with local clocks on other CPUs.524 525	Usual clocks for tracing::526 527	  # cat trace_clock528	  [local] global counter x86-tsc529 530	The clock with the square brackets around it is the one in effect.531 532	local:533		Default clock, but may not be in sync across CPUs534 535	global:536		This clock is in sync with all CPUs but may537		be a bit slower than the local clock.538 539	counter:540		This is not a clock at all, but literally an atomic541		counter. It counts up one by one, but is in sync542		with all CPUs. This is useful when you need to543		know exactly the order events occurred with respect to544		each other on different CPUs.545 546	uptime:547		This uses the jiffies counter and the time stamp548		is relative to the time since boot up.549 550	perf:551		This makes ftrace use the same clock that perf uses.552		Eventually perf will be able to read ftrace buffers553		and this will help out in interleaving the data.554 555	x86-tsc:556		Architectures may define their own clocks. For557		example, x86 uses its own TSC cycle clock here.558 559	ppc-tb:560		This uses the powerpc timebase register value.561		This is in sync across CPUs and can also be used562		to correlate events across hypervisor/guest if563		tb_offset is known.564 565	mono:566		This uses the fast monotonic clock (CLOCK_MONOTONIC)567		which is monotonic and is subject to NTP rate adjustments.568 569	mono_raw:570		This is the raw monotonic clock (CLOCK_MONOTONIC_RAW)571		which is monotonic but is not subject to any rate adjustments572		and ticks at the same rate as the hardware clocksource.573 574	boot:575		This is the boot clock (CLOCK_BOOTTIME) and is based on the576		fast monotonic clock, but also accounts for time spent in577		suspend. Since the clock access is designed for use in578		tracing in the suspend path, some side effects are possible579		if clock is accessed after the suspend time is accounted before580		the fast mono clock is updated. In this case, the clock update581		appears to happen slightly sooner than it normally would have.582		Also on 32-bit systems, it's possible that the 64-bit boot offset583		sees a partial update. These effects are rare and post584		processing should be able to handle them. See comments in the585		ktime_get_boot_fast_ns() function for more information.586 587	tai:588		This is the tai clock (CLOCK_TAI) and is derived from the wall-589		clock time. However, this clock does not experience590		discontinuities and backwards jumps caused by NTP inserting leap591		seconds. Since the clock access is designed for use in tracing,592		side effects are possible. The clock access may yield wrong593		readouts in case the internal TAI offset is updated e.g., caused594		by setting the system time or using adjtimex() with an offset.595		These effects are rare and post processing should be able to596		handle them. See comments in the ktime_get_tai_fast_ns()597		function for more information.598 599	To set a clock, simply echo the clock name into this file::600 601	  # echo global > trace_clock602 603	Setting a clock clears the ring buffer content as well as the604	"snapshot" buffer.605 606  trace_marker:607 608	This is a very useful file for synchronizing user space609	with events happening in the kernel. Writing strings into610	this file will be written into the ftrace buffer.611 612	It is useful in applications to open this file at the start613	of the application and just reference the file descriptor614	for the file::615 616		void trace_write(const char *fmt, ...)617		{618			va_list ap;619			char buf[256];620			int n;621 622			if (trace_fd < 0)623				return;624 625			va_start(ap, fmt);626			n = vsnprintf(buf, 256, fmt, ap);627			va_end(ap);628 629			write(trace_fd, buf, n);630		}631 632	start::633 634		trace_fd = open("trace_marker", O_WRONLY);635 636	Note: Writing into the trace_marker file can also initiate triggers637	      that are written into /sys/kernel/tracing/events/ftrace/print/trigger638	      See "Event triggers" in Documentation/trace/events.rst and an639              example in Documentation/trace/histogram.rst (Section 3.)640 641  trace_marker_raw:642 643	This is similar to trace_marker above, but is meant for binary data644	to be written to it, where a tool can be used to parse the data645	from trace_pipe_raw.646 647  uprobe_events:648 649	Add dynamic tracepoints in programs.650	See uprobetracer.rst651 652  uprobe_profile:653 654	Uprobe statistics. See uprobetrace.txt655 656  instances:657 658	This is a way to make multiple trace buffers where different659	events can be recorded in different buffers.660	See "Instances" section below.661 662  events:663 664	This is the trace event directory. It holds event tracepoints665	(also known as static tracepoints) that have been compiled666	into the kernel. It shows what event tracepoints exist667	and how they are grouped by system. There are "enable"668	files at various levels that can enable the tracepoints669	when a "1" is written to them.670 671	See events.rst for more information.672 673  set_event:674 675	By echoing in the event into this file, will enable that event.676 677	See events.rst for more information.678 679  available_events:680 681	A list of events that can be enabled in tracing.682 683	See events.rst for more information.684 685  timestamp_mode:686 687	Certain tracers may change the timestamp mode used when688	logging trace events into the event buffer.  Events with689	different modes can coexist within a buffer but the mode in690	effect when an event is logged determines which timestamp mode691	is used for that event.  The default timestamp mode is692	'delta'.693 694	Usual timestamp modes for tracing:695 696	  # cat timestamp_mode697	  [delta] absolute698 699	  The timestamp mode with the square brackets around it is the700	  one in effect.701 702	  delta: Default timestamp mode - timestamp is a delta against703	         a per-buffer timestamp.704 705	  absolute: The timestamp is a full timestamp, not a delta706                 against some other value.  As such it takes up more707                 space and is less efficient.708 709  hwlat_detector:710 711	Directory for the Hardware Latency Detector.712	See "Hardware Latency Detector" section below.713 714  per_cpu:715 716	This is a directory that contains the trace per_cpu information.717 718  per_cpu/cpu0/buffer_size_kb:719 720	The ftrace buffer is defined per_cpu. That is, there's a separate721	buffer for each CPU to allow writes to be done atomically,722	and free from cache bouncing. These buffers may have different723	size buffers. This file is similar to the buffer_size_kb724	file, but it only displays or sets the buffer size for the725	specific CPU. (here cpu0).726 727  per_cpu/cpu0/trace:728 729	This is similar to the "trace" file, but it will only display730	the data specific for the CPU. If written to, it only clears731	the specific CPU buffer.732 733  per_cpu/cpu0/trace_pipe734 735	This is similar to the "trace_pipe" file, and is a consuming736	read, but it will only display (and consume) the data specific737	for the CPU.738 739  per_cpu/cpu0/trace_pipe_raw740 741	For tools that can parse the ftrace ring buffer binary format,742	the trace_pipe_raw file can be used to extract the data743	from the ring buffer directly. With the use of the splice()744	system call, the buffer data can be quickly transferred to745	a file or to the network where a server is collecting the746	data.747 748	Like trace_pipe, this is a consuming reader, where multiple749	reads will always produce different data.750 751  per_cpu/cpu0/snapshot:752 753	This is similar to the main "snapshot" file, but will only754	snapshot the current CPU (if supported). It only displays755	the content of the snapshot for a given CPU, and if756	written to, only clears this CPU buffer.757 758  per_cpu/cpu0/snapshot_raw:759 760	Similar to the trace_pipe_raw, but will read the binary format761	from the snapshot buffer for the given CPU.762 763  per_cpu/cpu0/stats:764 765	This displays certain stats about the ring buffer:766 767	entries:768		The number of events that are still in the buffer.769 770	overrun:771		The number of lost events due to overwriting when772		the buffer was full.773 774	commit overrun:775		Should always be zero.776		This gets set if so many events happened within a nested777		event (ring buffer is re-entrant), that it fills the778		buffer and starts dropping events.779 780	bytes:781		Bytes actually read (not overwritten).782 783	oldest event ts:784		The oldest timestamp in the buffer785 786	now ts:787		The current timestamp788 789	dropped events:790		Events lost due to overwrite option being off.791 792	read events:793		The number of events read.794 795The Tracers796-----------797 798Here is the list of current tracers that may be configured.799 800  "function"801 802	Function call tracer to trace all kernel functions.803 804  "function_graph"805 806	Similar to the function tracer except that the807	function tracer probes the functions on their entry808	whereas the function graph tracer traces on both entry809	and exit of the functions. It then provides the ability810	to draw a graph of function calls similar to C code811	source.812 813  "blk"814 815	The block tracer. The tracer used by the blktrace user816	application.817 818  "hwlat"819 820	The Hardware Latency tracer is used to detect if the hardware821	produces any latency. See "Hardware Latency Detector" section822	below.823 824  "irqsoff"825 826	Traces the areas that disable interrupts and saves827	the trace with the longest max latency.828	See tracing_max_latency. When a new max is recorded,829	it replaces the old trace. It is best to view this830	trace with the latency-format option enabled, which831	happens automatically when the tracer is selected.832 833  "preemptoff"834 835	Similar to irqsoff but traces and records the amount of836	time for which preemption is disabled.837 838  "preemptirqsoff"839 840	Similar to irqsoff and preemptoff, but traces and841	records the largest time for which irqs and/or preemption842	is disabled.843 844  "wakeup"845 846	Traces and records the max latency that it takes for847	the highest priority task to get scheduled after848	it has been woken up.849        Traces all tasks as an average developer would expect.850 851  "wakeup_rt"852 853        Traces and records the max latency that it takes for just854        RT tasks (as the current "wakeup" does). This is useful855        for those interested in wake up timings of RT tasks.856 857  "wakeup_dl"858 859	Traces and records the max latency that it takes for860	a SCHED_DEADLINE task to be woken (as the "wakeup" and861	"wakeup_rt" does).862 863  "mmiotrace"864 865	A special tracer that is used to trace binary module.866	It will trace all the calls that a module makes to the867	hardware. Everything it writes and reads from the I/O868	as well.869 870  "branch"871 872	This tracer can be configured when tracing likely/unlikely873	calls within the kernel. It will trace when a likely and874	unlikely branch is hit and if it was correct in its prediction875	of being correct.876 877  "nop"878 879	This is the "trace nothing" tracer. To remove all880	tracers from tracing simply echo "nop" into881	current_tracer.882 883Error conditions884----------------885 886  For most ftrace commands, failure modes are obvious and communicated887  using standard return codes.888 889  For other more involved commands, extended error information may be890  available via the tracing/error_log file.  For the commands that891  support it, reading the tracing/error_log file after an error will892  display more detailed information about what went wrong, if893  information is available.  The tracing/error_log file is a circular894  error log displaying a small number (currently, 8) of ftrace errors895  for the last (8) failed commands.896 897  The extended error information and usage takes the form shown in898  this example::899 900    # echo xxx > /sys/kernel/tracing/events/sched/sched_wakeup/trigger901    echo: write error: Invalid argument902 903    # cat /sys/kernel/tracing/error_log904    [ 5348.887237] location: error: Couldn't yyy: zzz905      Command: xxx906               ^907    [ 7517.023364] location: error: Bad rrr: sss908      Command: ppp qqq909                   ^910 911  To clear the error log, echo the empty string into it::912 913    # echo > /sys/kernel/tracing/error_log914 915Examples of using the tracer916----------------------------917 918Here are typical examples of using the tracers when controlling919them only with the tracefs interface (without using any920user-land utilities).921 922Output format:923--------------924 925Here is an example of the output format of the file "trace"::926 927  # tracer: function928  #929  # entries-in-buffer/entries-written: 140080/250280   #P:4930  #931  #                              _-----=> irqs-off932  #                             / _----=> need-resched933  #                            | / _---=> hardirq/softirq934  #                            || / _--=> preempt-depth935  #                            ||| /     delay936  #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION937  #              | |       |   ||||       |         |938              bash-1977  [000] .... 17284.993652: sys_close <-system_call_fastpath939              bash-1977  [000] .... 17284.993653: __close_fd <-sys_close940              bash-1977  [000] .... 17284.993653: _raw_spin_lock <-__close_fd941              sshd-1974  [003] .... 17284.993653: __srcu_read_unlock <-fsnotify942              bash-1977  [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock943              bash-1977  [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd944              bash-1977  [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock945              bash-1977  [000] .... 17284.993657: filp_close <-__close_fd946              bash-1977  [000] .... 17284.993657: dnotify_flush <-filp_close947              sshd-1974  [003] .... 17284.993658: sys_select <-system_call_fastpath948              ....949 950A header is printed with the tracer name that is represented by951the trace. In this case the tracer is "function". Then it shows the952number of events in the buffer as well as the total number of entries953that were written. The difference is the number of entries that were954lost due to the buffer filling up (250280 - 140080 = 110200 events955lost).956 957The header explains the content of the events. Task name "bash", the task958PID "1977", the CPU that it was running on "000", the latency format959(explained below), the timestamp in <secs>.<usecs> format, the960function name that was traced "sys_close" and the parent function that961called this function "system_call_fastpath". The timestamp is the time962at which the function was entered.963 964Latency trace format965--------------------966 967When the latency-format option is enabled or when one of the latency968tracers is set, the trace file gives somewhat more information to see969why a latency happened. Here is a typical trace::970 971  # tracer: irqsoff972  #973  # irqsoff latency trace v1.1.5 on 3.8.0-test+974  # --------------------------------------------------------------------975  # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)976  #    -----------------977  #    | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0)978  #    -----------------979  #  => started at: __lock_task_sighand980  #  => ended at:   _raw_spin_unlock_irqrestore981  #982  #983  #                  _------=> CPU#            984  #                 / _-----=> irqs-off        985  #                | / _----=> need-resched    986  #                || / _---=> hardirq/softirq 987  #                ||| / _--=> preempt-depth   988  #                |||| /     delay             989  #  cmd     pid   ||||| time  |   caller      990  #     \   /      |||||  \    |   /           991        ps-6143    2d...    0us!: trace_hardirqs_off <-__lock_task_sighand992        ps-6143    2d..1  259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore993        ps-6143    2d..1  263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore994        ps-6143    2d..1  306us : <stack trace>995   => trace_hardirqs_on_caller996   => trace_hardirqs_on997   => _raw_spin_unlock_irqrestore998   => do_task_stat999   => proc_tgid_stat1000   => proc_single_show1001   => seq_read1002   => vfs_read1003   => sys_read1004   => system_call_fastpath1005 1006 1007This shows that the current tracer is "irqsoff" tracing the time1008for which interrupts were disabled. It gives the trace version (which1009never changes) and the version of the kernel upon which this was executed on1010(3.8). Then it displays the max latency in microseconds (259 us). The number1011of trace entries displayed and the total number (both are four: #4/4).1012VP, KP, SP, and HP are always zero and are reserved for later use.1013#P is the number of online CPUs (#P:4).1014 1015The task is the process that was running when the latency1016occurred. (ps pid: 6143).1017 1018The start and stop (the functions in which the interrupts were1019disabled and enabled respectively) that caused the latencies:1020 1021  - __lock_task_sighand is where the interrupts were disabled.1022  - _raw_spin_unlock_irqrestore is where they were enabled again.1023 1024The next lines after the header are the trace itself. The header1025explains which is which.1026 1027  cmd: The name of the process in the trace.1028 1029  pid: The PID of that process.1030 1031  CPU#: The CPU which the process was running on.1032 1033  irqs-off: 'd' interrupts are disabled. '.' otherwise.1034	.. caution:: If the architecture does not support a way to1035		read the irq flags variable, an 'X' will always1036		be printed here.1037 1038  need-resched:1039	- 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set,1040	- 'n' only TIF_NEED_RESCHED is set,1041	- 'p' only PREEMPT_NEED_RESCHED is set,1042	- '.' otherwise.1043 1044  hardirq/softirq:1045	- 'Z' - NMI occurred inside a hardirq1046	- 'z' - NMI is running1047	- 'H' - hard irq occurred inside a softirq.1048	- 'h' - hard irq is running1049	- 's' - soft irq is running1050	- '.' - normal context.1051 1052  preempt-depth: The level of preempt_disabled1053 1054The above is mostly meaningful for kernel developers.1055 1056  time:1057	When the latency-format option is enabled, the trace file1058	output includes a timestamp relative to the start of the1059	trace. This differs from the output when latency-format1060	is disabled, which includes an absolute timestamp.1061 1062  delay:1063	This is just to help catch your eye a bit better. And1064	needs to be fixed to be only relative to the same CPU.1065	The marks are determined by the difference between this1066	current trace and the next trace.1067 1068	  - '$' - greater than 1 second1069	  - '@' - greater than 100 millisecond1070	  - '*' - greater than 10 millisecond1071	  - '#' - greater than 1000 microsecond1072	  - '!' - greater than 100 microsecond1073	  - '+' - greater than 10 microsecond1074	  - ' ' - less than or equal to 10 microsecond.1075 1076  The rest is the same as the 'trace' file.1077 1078  Note, the latency tracers will usually end with a back trace1079  to easily find where the latency occurred.1080 1081trace_options1082-------------1083 1084The trace_options file (or the options directory) is used to control1085what gets printed in the trace output, or manipulate the tracers.1086To see what is available, simply cat the file::1087 1088  cat trace_options1089	print-parent1090	nosym-offset1091	nosym-addr1092	noverbose1093	noraw1094	nohex1095	nobin1096	noblock1097	nofields1098	trace_printk1099	annotate1100	nouserstacktrace1101	nosym-userobj1102	noprintk-msg-only1103	context-info1104	nolatency-format1105	record-cmd1106	norecord-tgid1107	overwrite1108	nodisable_on_free1109	irq-info1110	markers1111	noevent-fork1112	function-trace1113	nofunction-fork1114	nodisplay-graph1115	nostacktrace1116	nobranch1117 1118To disable one of the options, echo in the option prepended with1119"no"::1120 1121  echo noprint-parent > trace_options1122 1123To enable an option, leave off the "no"::1124 1125  echo sym-offset > trace_options1126 1127Here are the available options:1128 1129  print-parent1130	On function traces, display the calling (parent)1131	function as well as the function being traced.1132	::1133 1134	  print-parent:1135	   bash-4000  [01]  1477.606694: simple_strtoul <-kstrtoul1136 1137	  noprint-parent:1138	   bash-4000  [01]  1477.606694: simple_strtoul1139 1140 1141  sym-offset1142	Display not only the function name, but also the1143	offset in the function. For example, instead of1144	seeing just "ktime_get", you will see1145	"ktime_get+0xb/0x20".1146	::1147 1148	  sym-offset:1149	   bash-4000  [01]  1477.606694: simple_strtoul+0x6/0xa01150 1151  sym-addr1152	This will also display the function address as well1153	as the function name.1154	::1155 1156	  sym-addr:1157	   bash-4000  [01]  1477.606694: simple_strtoul <c0339346>1158 1159  verbose1160	This deals with the trace file when the1161        latency-format option is enabled.1162	::1163 1164	    bash  4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \1165	    (+0.000ms): simple_strtoul (kstrtoul)1166 1167  raw1168	This will display raw numbers. This option is best for1169	use with user applications that can translate the raw1170	numbers better than having it done in the kernel.1171 1172  hex1173	Similar to raw, but the numbers will be in a hexadecimal format.1174 1175  bin1176	This will print out the formats in raw binary.1177 1178  block1179	When set, reading trace_pipe will not block when polled.1180 1181  fields1182	Print the fields as described by their types. This is a better1183	option than using hex, bin or raw, as it gives a better parsing1184	of the content of the event.1185 1186  trace_printk1187	Can disable trace_printk() from writing into the buffer.1188 1189  trace_printk_dest1190	Set to have trace_printk() and similar internal tracing functions1191	write into this instance. Note, only one trace instance can have1192	this set. By setting this flag, it clears the trace_printk_dest flag1193	of the instance that had it set previously. By default, the top1194	level trace has this set, and will get it set again if another1195	instance has it set then clears it.1196 1197	This flag cannot be cleared by the top level instance, as it is the1198	default instance. The only way the top level instance has this flag1199	cleared, is by it being set in another instance.1200 1201  annotate1202	It is sometimes confusing when the CPU buffers are full1203	and one CPU buffer had a lot of events recently, thus1204	a shorter time frame, were another CPU may have only had1205	a few events, which lets it have older events. When1206	the trace is reported, it shows the oldest events first,1207	and it may look like only one CPU ran (the one with the1208	oldest events). When the annotate option is set, it will1209	display when a new CPU buffer started::1210 1211			  <idle>-0     [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on1212			  <idle>-0     [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on1213			  <idle>-0     [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore1214		##### CPU 2 buffer started ####1215			  <idle>-0     [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle1216			  <idle>-0     [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog1217			  <idle>-0     [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock1218 1219  userstacktrace1220	This option changes the trace. It records a1221	stacktrace of the current user space thread after1222	each trace event.1223 1224  sym-userobj1225	when user stacktrace are enabled, look up which1226	object the address belongs to, and print a1227	relative address. This is especially useful when1228	ASLR is on, otherwise you don't get a chance to1229	resolve the address to object/file/line after1230	the app is no longer running1231 1232	The lookup is performed when you read1233	trace,trace_pipe. Example::1234 1235		  a.out-1623  [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+01236		  x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6]1237 1238 1239  printk-msg-only1240	When set, trace_printk()s will only show the format1241	and not their parameters (if trace_bprintk() or1242	trace_bputs() was used to save the trace_printk()).1243 1244  context-info1245	Show only the event data. Hides the comm, PID,1246	timestamp, CPU, and other useful data.1247 1248  latency-format1249	This option changes the trace output. When it is enabled,1250	the trace displays additional information about the1251	latency, as described in "Latency trace format".1252 1253  pause-on-trace1254	When set, opening the trace file for read, will pause1255	writing to the ring buffer (as if tracing_on was set to zero).1256	This simulates the original behavior of the trace file.1257	When the file is closed, tracing will be enabled again.1258 1259  hash-ptr1260        When set, "%p" in the event printk format displays the1261        hashed pointer value instead of real address.1262        This will be useful if you want to find out which hashed1263        value is corresponding to the real value in trace log.1264 1265  record-cmd1266	When any event or tracer is enabled, a hook is enabled1267	in the sched_switch trace point to fill comm cache1268	with mapped pids and comms. But this may cause some1269	overhead, and if you only care about pids, and not the1270	name of the task, disabling this option can lower the1271	impact of tracing. See "saved_cmdlines".1272 1273  record-tgid1274	When any event or tracer is enabled, a hook is enabled1275	in the sched_switch trace point to fill the cache of1276	mapped Thread Group IDs (TGID) mapping to pids. See1277	"saved_tgids".1278 1279  overwrite1280	This controls what happens when the trace buffer is1281	full. If "1" (default), the oldest events are1282	discarded and overwritten. If "0", then the newest1283	events are discarded.1284	(see per_cpu/cpu0/stats for overrun and dropped)1285 1286  disable_on_free1287	When the free_buffer is closed, tracing will1288	stop (tracing_on set to 0).1289 1290  irq-info1291	Shows the interrupt, preempt count, need resched data.1292	When disabled, the trace looks like::1293 1294		# tracer: function1295		#1296		# entries-in-buffer/entries-written: 144405/9452052   #P:41297		#1298		#           TASK-PID   CPU#      TIMESTAMP  FUNCTION1299		#              | |       |          |         |1300			  <idle>-0     [002]  23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up1301			  <idle>-0     [002]  23636.756054: activate_task <-ttwu_do_activate.constprop.891302			  <idle>-0     [002]  23636.756055: enqueue_task <-activate_task1303 1304 1305  markers1306	When set, the trace_marker is writable (only by root).1307	When disabled, the trace_marker will error with EINVAL1308	on write.1309 1310  event-fork1311	When set, tasks with PIDs listed in set_event_pid will have1312	the PIDs of their children added to set_event_pid when those1313	tasks fork. Also, when tasks with PIDs in set_event_pid exit,1314	their PIDs will be removed from the file.1315 1316        This affects PIDs listed in set_event_notrace_pid as well.1317 1318  function-trace1319	The latency tracers will enable function tracing1320	if this option is enabled (default it is). When1321	it is disabled, the latency tracers do not trace1322	functions. This keeps the overhead of the tracer down1323	when performing latency tests.1324 1325  function-fork1326	When set, tasks with PIDs listed in set_ftrace_pid will1327	have the PIDs of their children added to set_ftrace_pid1328	when those tasks fork. Also, when tasks with PIDs in1329	set_ftrace_pid exit, their PIDs will be removed from the1330	file.1331 1332        This affects PIDs in set_ftrace_notrace_pid as well.1333 1334  display-graph1335	When set, the latency tracers (irqsoff, wakeup, etc) will1336	use function graph tracing instead of function tracing.1337 1338  stacktrace1339	When set, a stack trace is recorded after any trace event1340	is recorded.1341 1342  branch1343	Enable branch tracing with the tracer. This enables branch1344	tracer along with the currently set tracer. Enabling this1345	with the "nop" tracer is the same as just enabling the1346	"branch" tracer.1347 1348.. tip:: Some tracers have their own options. They only appear in this1349       file when the tracer is active. They always appear in the1350       options directory.1351 1352 1353Here are the per tracer options:1354 1355Options for function tracer:1356 1357  func_stack_trace1358	When set, a stack trace is recorded after every1359	function that is recorded. NOTE! Limit the functions1360	that are recorded before enabling this, with1361	"set_ftrace_filter" otherwise the system performance1362	will be critically degraded. Remember to disable1363	this option before clearing the function filter.1364 1365Options for function_graph tracer:1366 1367 Since the function_graph tracer has a slightly different output1368 it has its own options to control what is displayed.1369 1370  funcgraph-overrun1371	When set, the "overrun" of the graph stack is1372	displayed after each function traced. The1373	overrun, is when the stack depth of the calls1374	is greater than what is reserved for each task.1375	Each task has a fixed array of functions to1376	trace in the call graph. If the depth of the1377	calls exceeds that, the function is not traced.1378	The overrun is the number of functions missed1379	due to exceeding this array.1380 1381  funcgraph-cpu1382	When set, the CPU number of the CPU where the trace1383	occurred is displayed.1384 1385  funcgraph-overhead1386	When set, if the function takes longer than1387	A certain amount, then a delay marker is1388	displayed. See "delay" above, under the1389	header description.1390 1391  funcgraph-proc1392	Unlike other tracers, the process' command line1393	is not displayed by default, but instead only1394	when a task is traced in and out during a context1395	switch. Enabling this options has the command1396	of each process displayed at every line.1397 1398  funcgraph-duration1399	At the end of each function (the return)1400	the duration of the amount of time in the1401	function is displayed in microseconds.1402 1403  funcgraph-abstime1404	When set, the timestamp is displayed at each line.1405 1406  funcgraph-irqs1407	When disabled, functions that happen inside an1408	interrupt will not be traced.1409 1410  funcgraph-tail1411	When set, the return event will include the function1412	that it represents. By default this is off, and1413	only a closing curly bracket "}" is displayed for1414	the return of a function.1415 1416  funcgraph-retval1417	When set, the return value of each traced function1418	will be printed after an equal sign "=". By default1419	this is off.1420 1421  funcgraph-retval-hex1422	When set, the return value will always be printed1423	in hexadecimal format. If the option is not set and1424	the return value is an error code, it will be printed1425	in signed decimal format; otherwise it will also be1426	printed in hexadecimal format. By default, this option1427	is off.1428 1429  sleep-time1430	When running function graph tracer, to include1431	the time a task schedules out in its function.1432	When enabled, it will account time the task has been1433	scheduled out as part of the function call.1434 1435  graph-time1436	When running function profiler with function graph tracer,1437	to include the time to call nested functions. When this is1438	not set, the time reported for the function will only1439	include the time the function itself executed for, not the1440	time for functions that it called.1441 1442Options for blk tracer:1443 1444  blk_classic1445	Shows a more minimalistic output.1446 1447 1448irqsoff1449-------1450 1451When interrupts are disabled, the CPU can not react to any other1452external event (besides NMIs and SMIs). This prevents the timer1453interrupt from triggering or the mouse interrupt from letting1454the kernel know of a new mouse event. The result is a latency1455with the reaction time.1456 1457The irqsoff tracer tracks the time for which interrupts are1458disabled. When a new maximum latency is hit, the tracer saves1459the trace leading up to that latency point so that every time a1460new maximum is reached, the old saved trace is discarded and the1461new trace is saved.1462 1463To reset the maximum, echo 0 into tracing_max_latency. Here is1464an example::1465 1466  # echo 0 > options/function-trace1467  # echo irqsoff > current_tracer1468  # echo 1 > tracing_on1469  # echo 0 > tracing_max_latency1470  # ls -ltr1471  [...]1472  # echo 0 > tracing_on1473  # cat trace1474  # tracer: irqsoff1475  #1476  # irqsoff latency trace v1.1.5 on 3.8.0-test+1477  # --------------------------------------------------------------------1478  # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)1479  #    -----------------1480  #    | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0)1481  #    -----------------1482  #  => started at: run_timer_softirq1483  #  => ended at:   run_timer_softirq1484  #1485  #1486  #                  _------=> CPU#            1487  #                 / _-----=> irqs-off        1488  #                | / _----=> need-resched    1489  #                || / _---=> hardirq/softirq 1490  #                ||| / _--=> preempt-depth   1491  #                |||| /     delay             1492  #  cmd     pid   ||||| time  |   caller      1493  #     \   /      |||||  \    |   /           1494    <idle>-0       0d.s2    0us+: _raw_spin_lock_irq <-run_timer_softirq1495    <idle>-0       0dNs3   17us : _raw_spin_unlock_irq <-run_timer_softirq1496    <idle>-0       0dNs3   17us+: trace_hardirqs_on <-run_timer_softirq1497    <idle>-0       0dNs3   25us : <stack trace>1498   => _raw_spin_unlock_irq1499   => run_timer_softirq1500   => __do_softirq1501   => call_softirq1502   => do_softirq1503   => irq_exit1504   => smp_apic_timer_interrupt1505   => apic_timer_interrupt1506   => rcu_idle_exit1507   => cpu_idle1508   => rest_init1509   => start_kernel1510   => x86_64_start_reservations1511   => x86_64_start_kernel1512 1513Here we see that we had a latency of 16 microseconds (which is1514very good). The _raw_spin_lock_irq in run_timer_softirq disabled1515interrupts. The difference between the 16 and the displayed1516timestamp 25us occurred because the clock was incremented1517between the time of recording the max latency and the time of1518recording the function that had that latency.1519 1520Note the above example had function-trace not set. If we set1521function-trace, we get a much larger output::1522 1523 with echo 1 > options/function-trace1524 1525  # tracer: irqsoff1526  #1527  # irqsoff latency trace v1.1.5 on 3.8.0-test+1528  # --------------------------------------------------------------------1529  # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)1530  #    -----------------1531  #    | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0)1532  #    -----------------1533  #  => started at: ata_scsi_queuecmd1534  #  => ended at:   ata_scsi_queuecmd1535  #1536  #1537  #                  _------=> CPU#            1538  #                 / _-----=> irqs-off        1539  #                | / _----=> need-resched    1540  #                || / _---=> hardirq/softirq 1541  #                ||| / _--=> preempt-depth   1542  #                |||| /     delay             1543  #  cmd     pid   ||||| time  |   caller      1544  #     \   /      |||||  \    |   /           1545      bash-2042    3d...    0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd1546      bash-2042    3d...    0us : add_preempt_count <-_raw_spin_lock_irqsave1547      bash-2042    3d..1    1us : ata_scsi_find_dev <-ata_scsi_queuecmd1548      bash-2042    3d..1    1us : __ata_scsi_find_dev <-ata_scsi_find_dev1549      bash-2042    3d..1    2us : ata_find_dev.part.14 <-__ata_scsi_find_dev1550      bash-2042    3d..1    2us : ata_qc_new_init <-__ata_scsi_queuecmd1551      bash-2042    3d..1    3us : ata_sg_init <-__ata_scsi_queuecmd1552      bash-2042    3d..1    4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd1553      bash-2042    3d..1    4us : ata_build_rw_tf <-ata_scsi_rw_xlat1554  [...]1555      bash-2042    3d..1   67us : delay_tsc <-__delay1556      bash-2042    3d..1   67us : add_preempt_count <-delay_tsc1557      bash-2042    3d..2   67us : sub_preempt_count <-delay_tsc1558      bash-2042    3d..1   67us : add_preempt_count <-delay_tsc1559      bash-2042    3d..2   68us : sub_preempt_count <-delay_tsc1560      bash-2042    3d..1   68us+: ata_bmdma_start <-ata_bmdma_qc_issue1561      bash-2042    3d..1   71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd1562      bash-2042    3d..1   71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd1563      bash-2042    3d..1   72us+: trace_hardirqs_on <-ata_scsi_queuecmd1564      bash-2042    3d..1  120us : <stack trace>1565   => _raw_spin_unlock_irqrestore1566   => ata_scsi_queuecmd1567   => scsi_dispatch_cmd1568   => scsi_request_fn1569   => __blk_run_queue_uncond1570   => __blk_run_queue1571   => blk_queue_bio1572   => submit_bio_noacct1573   => submit_bio1574   => submit_bh1575   => __ext3_get_inode_loc1576   => ext3_iget1577   => ext3_lookup1578   => lookup_real1579   => __lookup_hash1580   => walk_component1581   => lookup_last1582   => path_lookupat1583   => filename_lookup1584   => user_path_at_empty1585   => user_path_at1586   => vfs_fstatat1587   => vfs_stat1588   => sys_newstat1589   => system_call_fastpath1590 1591 1592Here we traced a 71 microsecond latency. But we also see all the1593functions that were called during that time. Note that by1594enabling function tracing, we incur an added overhead. This1595overhead may extend the latency times. But nevertheless, this1596trace has provided some very helpful debugging information.1597 1598If we prefer function graph output instead of function, we can set1599display-graph option::1600 1601 with echo 1 > options/display-graph1602 1603  # tracer: irqsoff1604  #1605  # irqsoff latency trace v1.1.5 on 4.20.0-rc6+1606  # --------------------------------------------------------------------1607  # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4)1608  #    -----------------1609  #    | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0)1610  #    -----------------1611  #  => started at: free_debug_processing1612  #  => ended at:   return_to_handler1613  #1614  #1615  #                                       _-----=> irqs-off1616  #                                      / _----=> need-resched1617  #                                     | / _---=> hardirq/softirq1618  #                                     || / _--=> preempt-depth1619  #                                     ||| /1620  #   REL TIME      CPU  TASK/PID       ||||     DURATION                  FUNCTION CALLS1621  #      |          |     |    |        ||||      |   |                     |   |   |   |1622          0 us |   0)   bash-1507    |  d... |   0.000 us    |  _raw_spin_lock_irqsave();1623          0 us |   0)   bash-1507    |  d..1 |   0.378 us    |    do_raw_spin_trylock();1624          1 us |   0)   bash-1507    |  d..2 |               |    set_track() {1625          2 us |   0)   bash-1507    |  d..2 |               |      save_stack_trace() {1626          2 us |   0)   bash-1507    |  d..2 |               |        __save_stack_trace() {1627          3 us |   0)   bash-1507    |  d..2 |               |          __unwind_start() {1628          3 us |   0)   bash-1507    |  d..2 |               |            get_stack_info() {1629          3 us |   0)   bash-1507    |  d..2 |   0.351 us    |              in_task_stack();1630          4 us |   0)   bash-1507    |  d..2 |   1.107 us    |            }1631  [...]1632       3750 us |   0)   bash-1507    |  d..1 |   0.516 us    |      do_raw_spin_unlock();1633       3750 us |   0)   bash-1507    |  d..1 |   0.000 us    |  _raw_spin_unlock_irqrestore();1634       3764 us |   0)   bash-1507    |  d..1 |   0.000 us    |  tracer_hardirqs_on();1635      bash-1507    0d..1 3792us : <stack trace>1636   => free_debug_processing1637   => __slab_free1638   => kmem_cache_free1639   => vm_area_free1640   => remove_vma1641   => exit_mmap1642   => mmput1643   => begin_new_exec1644   => load_elf_binary1645   => search_binary_handler1646   => __do_execve_file.isra.321647   => __x64_sys_execve1648   => do_syscall_641649   => entry_SYSCALL_64_after_hwframe1650 1651preemptoff1652----------1653 1654When preemption is disabled, we may be able to receive1655interrupts but the task cannot be preempted and a higher1656priority task must wait for preemption to be enabled again1657before it can preempt a lower priority task.1658 1659The preemptoff tracer traces the places that disable preemption.1660Like the irqsoff tracer, it records the maximum latency for1661which preemption was disabled. The control of preemptoff tracer1662is much like the irqsoff tracer.1663::1664 1665  # echo 0 > options/function-trace1666  # echo preemptoff > current_tracer1667  # echo 1 > tracing_on1668  # echo 0 > tracing_max_latency1669  # ls -ltr1670  [...]1671  # echo 0 > tracing_on1672  # cat trace1673  # tracer: preemptoff1674  #1675  # preemptoff latency trace v1.1.5 on 3.8.0-test+1676  # --------------------------------------------------------------------1677  # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)1678  #    -----------------1679  #    | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0)1680  #    -----------------1681  #  => started at: do_IRQ1682  #  => ended at:   do_IRQ1683  #1684  #1685  #                  _------=> CPU#            1686  #                 / _-----=> irqs-off        1687  #                | / _----=> need-resched    1688  #                || / _---=> hardirq/softirq 1689  #                ||| / _--=> preempt-depth   1690  #                |||| /     delay             1691  #  cmd     pid   ||||| time  |   caller      1692  #     \   /      |||||  \    |   /           1693      sshd-1991    1d.h.    0us+: irq_enter <-do_IRQ1694      sshd-1991    1d..1   46us : irq_exit <-do_IRQ1695      sshd-1991    1d..1   47us+: trace_preempt_on <-do_IRQ1696      sshd-1991    1d..1   52us : <stack trace>1697   => sub_preempt_count1698   => irq_exit1699   => do_IRQ1700   => ret_from_intr1701 1702 1703This has some more changes. Preemption was disabled when an1704interrupt came in (notice the 'h'), and was enabled on exit.1705But we also see that interrupts have been disabled when entering1706the preempt off section and leaving it (the 'd'). We do not know if1707interrupts were enabled in the mean time or shortly after this1708was over.1709::1710 1711  # tracer: preemptoff1712  #1713  # preemptoff latency trace v1.1.5 on 3.8.0-test+1714  # --------------------------------------------------------------------1715  # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)1716  #    -----------------1717  #    | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0)1718  #    -----------------1719  #  => started at: wake_up_new_task1720  #  => ended at:   task_rq_unlock1721  #1722  #1723  #                  _------=> CPU#            1724  #                 / _-----=> irqs-off        1725  #                | / _----=> need-resched    1726  #                || / _---=> hardirq/softirq 1727  #                ||| / _--=> preempt-depth   1728  #                |||| /     delay             1729  #  cmd     pid   ||||| time  |   caller      1730  #     \   /      |||||  \    |   /           1731      bash-1994    1d..1    0us : _raw_spin_lock_irqsave <-wake_up_new_task1732      bash-1994    1d..1    0us : select_task_rq_fair <-select_task_rq1733      bash-1994    1d..1    1us : __rcu_read_lock <-select_task_rq_fair1734      bash-1994    1d..1    1us : source_load <-select_task_rq_fair1735      bash-1994    1d..1    1us : source_load <-select_task_rq_fair1736  [...]1737      bash-1994    1d..1   12us : irq_enter <-smp_apic_timer_interrupt1738      bash-1994    1d..1   12us : rcu_irq_enter <-irq_enter1739      bash-1994    1d..1   13us : add_preempt_count <-irq_enter1740      bash-1994    1d.h1   13us : exit_idle <-smp_apic_timer_interrupt1741      bash-1994    1d.h1   13us : hrtimer_interrupt <-smp_apic_timer_interrupt1742      bash-1994    1d.h1   13us : _raw_spin_lock <-hrtimer_interrupt1743      bash-1994    1d.h1   14us : add_preempt_count <-_raw_spin_lock1744      bash-1994    1d.h2   14us : ktime_get_update_offsets <-hrtimer_interrupt1745  [...]1746      bash-1994    1d.h1   35us : lapic_next_event <-clockevents_program_event1747      bash-1994    1d.h1   35us : irq_exit <-smp_apic_timer_interrupt1748      bash-1994    1d.h1   36us : sub_preempt_count <-irq_exit1749      bash-1994    1d..2   36us : do_softirq <-irq_exit1750      bash-1994    1d..2   36us : __do_softirq <-call_softirq1751      bash-1994    1d..2   36us : __local_bh_disable <-__do_softirq1752      bash-1994    1d.s2   37us : add_preempt_count <-_raw_spin_lock_irq1753      bash-1994    1d.s3   38us : _raw_spin_unlock <-run_timer_softirq1754      bash-1994    1d.s3   39us : sub_preempt_count <-_raw_spin_unlock1755      bash-1994    1d.s2   39us : call_timer_fn <-run_timer_softirq1756  [...]1757      bash-1994    1dNs2   81us : cpu_needs_another_gp <-rcu_process_callbacks1758      bash-1994    1dNs2   82us : __local_bh_enable <-__do_softirq1759      bash-1994    1dNs2   82us : sub_preempt_count <-__local_bh_enable1760      bash-1994    1dN.2   82us : idle_cpu <-irq_exit1761      bash-1994    1dN.2   83us : rcu_irq_exit <-irq_exit1762      bash-1994    1dN.2   83us : sub_preempt_count <-irq_exit1763      bash-1994    1.N.1   84us : _raw_spin_unlock_irqrestore <-task_rq_unlock1764      bash-1994    1.N.1   84us+: trace_preempt_on <-task_rq_unlock1765      bash-1994    1.N.1  104us : <stack trace>1766   => sub_preempt_count1767   => _raw_spin_unlock_irqrestore1768   => task_rq_unlock1769   => wake_up_new_task1770   => do_fork1771   => sys_clone1772   => stub_clone1773 1774 1775The above is an example of the preemptoff trace with1776function-trace set. Here we see that interrupts were not disabled1777the entire time. The irq_enter code lets us know that we entered1778an interrupt 'h'. Before that, the functions being traced still1779show that it is not in an interrupt, but we can see from the1780functions themselves that this is not the case.1781 1782preemptirqsoff1783--------------1784 1785Knowing the locations that have interrupts disabled or1786preemption disabled for the longest times is helpful. But1787sometimes we would like to know when either preemption and/or1788interrupts are disabled.1789 1790Consider the following code::1791 1792    local_irq_disable();1793    call_function_with_irqs_off();1794    preempt_disable();1795    call_function_with_irqs_and_preemption_off();1796    local_irq_enable();1797    call_function_with_preemption_off();1798    preempt_enable();1799 1800The irqsoff tracer will record the total length of1801call_function_with_irqs_off() and1802call_function_with_irqs_and_preemption_off().1803 1804The preemptoff tracer will record the total length of1805call_function_with_irqs_and_preemption_off() and1806call_function_with_preemption_off().1807 1808But neither will trace the time that interrupts and/or1809preemption is disabled. This total time is the time that we can1810not schedule. To record this time, use the preemptirqsoff1811tracer.1812 1813Again, using this trace is much like the irqsoff and preemptoff1814tracers.1815::1816 1817  # echo 0 > options/function-trace1818  # echo preemptirqsoff > current_tracer1819  # echo 1 > tracing_on1820  # echo 0 > tracing_max_latency1821  # ls -ltr1822  [...]1823  # echo 0 > tracing_on1824  # cat trace1825  # tracer: preemptirqsoff1826  #1827  # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+1828  # --------------------------------------------------------------------1829  # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)1830  #    -----------------1831  #    | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0)1832  #    -----------------1833  #  => started at: ata_scsi_queuecmd1834  #  => ended at:   ata_scsi_queuecmd1835  #1836  #1837  #                  _------=> CPU#            1838  #                 / _-----=> irqs-off        1839  #                | / _----=> need-resched    1840  #                || / _---=> hardirq/softirq 1841  #                ||| / _--=> preempt-depth   1842  #                |||| /     delay             1843  #  cmd     pid   ||||| time  |   caller      1844  #     \   /      |||||  \    |   /           1845        ls-2230    3d...    0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd1846        ls-2230    3...1  100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd1847        ls-2230    3...1  101us+: trace_preempt_on <-ata_scsi_queuecmd1848        ls-2230    3...1  111us : <stack trace>1849   => sub_preempt_count1850   => _raw_spin_unlock_irqrestore1851   => ata_scsi_queuecmd1852   => scsi_dispatch_cmd1853   => scsi_request_fn1854   => __blk_run_queue_uncond1855   => __blk_run_queue1856   => blk_queue_bio1857   => submit_bio_noacct1858   => submit_bio1859   => submit_bh1860   => ext3_bread1861   => ext3_dir_bread1862   => htree_dirblock_to_tree1863   => ext3_htree_fill_tree1864   => ext3_readdir1865   => vfs_readdir1866   => sys_getdents1867   => system_call_fastpath1868 1869 1870The trace_hardirqs_off_thunk is called from assembly on x86 when1871interrupts are disabled in the assembly code. Without the1872function tracing, we do not know if interrupts were enabled1873within the preemption points. We do see that it started with1874preemption enabled.1875 1876Here is a trace with function-trace set::1877 1878  # tracer: preemptirqsoff1879  #1880  # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+1881  # --------------------------------------------------------------------1882  # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)1883  #    -----------------1884  #    | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0)1885  #    -----------------1886  #  => started at: schedule1887  #  => ended at:   mutex_unlock1888  #1889  #1890  #                  _------=> CPU#            1891  #                 / _-----=> irqs-off        1892  #                | / _----=> need-resched    1893  #                || / _---=> hardirq/softirq 1894  #                ||| / _--=> preempt-depth   1895  #                |||| /     delay             1896  #  cmd     pid   ||||| time  |   caller      1897  #     \   /      |||||  \    |   /           1898  kworker/-59      3...1    0us : __schedule <-schedule1899  kworker/-59      3d..1    0us : rcu_preempt_qs <-rcu_note_context_switch1900  kworker/-59      3d..1    1us : add_preempt_count <-_raw_spin_lock_irq1901  kworker/-59      3d..2    1us : deactivate_task <-__schedule1902  kworker/-59      3d..2    1us : dequeue_task <-deactivate_task1903  kworker/-59      3d..2    2us : update_rq_clock <-dequeue_task1904  kworker/-59      3d..2    2us : dequeue_task_fair <-dequeue_task1905  kworker/-59      3d..2    2us : update_curr <-dequeue_task_fair1906  kworker/-59      3d..2    2us : update_min_vruntime <-update_curr1907  kworker/-59      3d..2    3us : cpuacct_charge <-update_curr1908  kworker/-59      3d..2    3us : __rcu_read_lock <-cpuacct_charge1909  kworker/-59      3d..2    3us : __rcu_read_unlock <-cpuacct_charge1910  kworker/-59      3d..2    3us : update_cfs_rq_blocked_load <-dequeue_task_fair1911  kworker/-59      3d..2    4us : clear_buddies <-dequeue_task_fair1912  kworker/-59      3d..2    4us : account_entity_dequeue <-dequeue_task_fair1913  kworker/-59      3d..2    4us : update_min_vruntime <-dequeue_task_fair1914  kworker/-59      3d..2    4us : update_cfs_shares <-dequeue_task_fair1915  kworker/-59      3d..2    5us : hrtick_update <-dequeue_task_fair1916  kworker/-59      3d..2    5us : wq_worker_sleeping <-__schedule1917  kworker/-59      3d..2    5us : kthread_data <-wq_worker_sleeping1918  kworker/-59      3d..2    5us : put_prev_task_fair <-__schedule1919  kworker/-59      3d..2    6us : pick_next_task_fair <-pick_next_task1920  kworker/-59      3d..2    6us : clear_buddies <-pick_next_task_fair1921  kworker/-59      3d..2    6us : set_next_entity <-pick_next_task_fair1922  kworker/-59      3d..2    6us : update_stats_wait_end <-set_next_entity1923        ls-2269    3d..2    7us : finish_task_switch <-__schedule1924        ls-2269    3d..2    7us : _raw_spin_unlock_irq <-finish_task_switch1925        ls-2269    3d..2    8us : do_IRQ <-ret_from_intr1926        ls-2269    3d..2    8us : irq_enter <-do_IRQ1927        ls-2269    3d..2    8us : rcu_irq_enter <-irq_enter1928        ls-2269    3d..2    9us : add_preempt_count <-irq_enter1929        ls-2269    3d.h2    9us : exit_idle <-do_IRQ1930  [...]1931        ls-2269    3d.h3   20us : sub_preempt_count <-_raw_spin_unlock1932        ls-2269    3d.h2   20us : irq_exit <-do_IRQ1933        ls-2269    3d.h2   21us : sub_preempt_count <-irq_exit1934        ls-2269    3d..3   21us : do_softirq <-irq_exit1935        ls-2269    3d..3   21us : __do_softirq <-call_softirq1936        ls-2269    3d..3   21us+: __local_bh_disable <-__do_softirq1937        ls-2269    3d.s4   29us : sub_preempt_count <-_local_bh_enable_ip1938        ls-2269    3d.s5   29us : sub_preempt_count <-_local_bh_enable_ip1939        ls-2269    3d.s5   31us : do_IRQ <-ret_from_intr1940        ls-2269    3d.s5   31us : irq_enter <-do_IRQ1941        ls-2269    3d.s5   31us : rcu_irq_enter <-irq_enter1942  [...]1943        ls-2269    3d.s5   31us : rcu_irq_enter <-irq_enter1944        ls-2269    3d.s5   32us : add_preempt_count <-irq_enter1945        ls-2269    3d.H5   32us : exit_idle <-do_IRQ1946        ls-2269    3d.H5   32us : handle_irq <-do_IRQ1947        ls-2269    3d.H5   32us : irq_to_desc <-handle_irq1948        ls-2269    3d.H5   33us : handle_fasteoi_irq <-handle_irq1949  [...]1950        ls-2269    3d.s5  158us : _raw_spin_unlock_irqrestore <-rtl8139_poll1951        ls-2269    3d.s3  158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action1952        ls-2269    3d.s3  159us : __local_bh_enable <-__do_softirq1953        ls-2269    3d.s3  159us : sub_preempt_count <-__local_bh_enable1954        ls-2269    3d..3  159us : idle_cpu <-irq_exit1955        ls-2269    3d..3  159us : rcu_irq_exit <-irq_exit1956        ls-2269    3d..3  160us : sub_preempt_count <-irq_exit1957        ls-2269    3d...  161us : __mutex_unlock_slowpath <-mutex_unlock1958        ls-2269    3d...  162us+: trace_hardirqs_on <-mutex_unlock1959        ls-2269    3d...  186us : <stack trace>1960   => __mutex_unlock_slowpath1961   => mutex_unlock1962   => process_output1963   => n_tty_write1964   => tty_write1965   => vfs_write1966   => sys_write1967   => system_call_fastpath1968 1969This is an interesting trace. It started with kworker running and1970scheduling out and ls taking over. But as soon as ls released the1971rq lock and enabled interrupts (but not preemption) an interrupt1972triggered. When the interrupt finished, it started running softirqs.1973But while the softirq was running, another interrupt triggered.1974When an interrupt is running inside a softirq, the annotation is 'H'.1975 1976 1977wakeup1978------1979 1980One common case that people are interested in tracing is the1981time it takes for a task that is woken to actually wake up.1982Now for non Real-Time tasks, this can be arbitrary. But tracing1983it nonetheless can be interesting. 1984 1985Without function tracing::1986 1987  # echo 0 > options/function-trace1988  # echo wakeup > current_tracer1989  # echo 1 > tracing_on1990  # echo 0 > tracing_max_latency1991  # chrt -f 5 sleep 11992  # echo 0 > tracing_on1993  # cat trace1994  # tracer: wakeup1995  #1996  # wakeup latency trace v1.1.5 on 3.8.0-test+1997  # --------------------------------------------------------------------1998  # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)1999  #    -----------------2000  #    | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0)2001  #    -----------------2002  #2003  #                  _------=> CPU#            2004  #                 / _-----=> irqs-off        2005  #                | / _----=> need-resched    2006  #                || / _---=> hardirq/softirq 2007  #                ||| / _--=> preempt-depth   2008  #                |||| /     delay             2009  #  cmd     pid   ||||| time  |   caller      2010  #     \   /      |||||  \    |   /           2011    <idle>-0       3dNs7    0us :      0:120:R   + [003]   312:100:R kworker/3:1H2012    <idle>-0       3dNs7    1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up2013    <idle>-0       3d..3   15us : __schedule <-schedule2014    <idle>-0       3d..3   15us :      0:120:R ==> [003]   312:100:R kworker/3:1H2015 2016The tracer only traces the highest priority task in the system2017to avoid tracing the normal circumstances. Here we see that2018the kworker with a nice priority of -20 (not very nice), took2019just 15 microseconds from the time it woke up, to the time it2020ran.2021 2022Non Real-Time tasks are not that interesting. A more interesting2023trace is to concentrate only on Real-Time tasks.2024 2025wakeup_rt2026---------2027 2028In a Real-Time environment it is very important to know the2029wakeup time it takes for the highest priority task that is woken2030up to the time that it executes. This is also known as "schedule2031latency". I stress the point that this is about RT tasks. It is2032also important to know the scheduling latency of non-RT tasks,2033but the average schedule latency is better for non-RT tasks.2034Tools like LatencyTop are more appropriate for such2035measurements.2036 2037Real-Time environments are interested in the worst case latency.2038That is the longest latency it takes for something to happen,2039and not the average. We can have a very fast scheduler that may2040only have a large latency once in a while, but that would not2041work well with Real-Time tasks.  The wakeup_rt tracer was designed2042to record the worst case wakeups of RT tasks. Non-RT tasks are2043not recorded because the tracer only records one worst case and2044tracing non-RT tasks that are unpredictable will overwrite the2045worst case latency of RT tasks (just run the normal wakeup2046tracer for a while to see that effect).2047 2048Since this tracer only deals with RT tasks, we will run this2049slightly differently than we did with the previous tracers.2050Instead of performing an 'ls', we will run 'sleep 1' under2051'chrt' which changes the priority of the task.2052::2053 2054  # echo 0 > options/function-trace2055  # echo wakeup_rt > current_tracer2056  # echo 1 > tracing_on2057  # echo 0 > tracing_max_latency2058  # chrt -f 5 sleep 12059  # echo 0 > tracing_on2060  # cat trace2061  # tracer: wakeup2062  #2063  # tracer: wakeup_rt2064  #2065  # wakeup_rt latency trace v1.1.5 on 3.8.0-test+2066  # --------------------------------------------------------------------2067  # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)2068  #    -----------------2069  #    | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5)2070  #    -----------------2071  #2072  #                  _------=> CPU#            2073  #                 / _-----=> irqs-off        2074  #                | / _----=> need-resched    2075  #                || / _---=> hardirq/softirq 2076  #                ||| / _--=> preempt-depth   2077  #                |||| /     delay             2078  #  cmd     pid   ||||| time  |   caller      2079  #     \   /      |||||  \    |   /           2080    <idle>-0       3d.h4    0us :      0:120:R   + [003]  2389: 94:R sleep2081    <idle>-0       3d.h4    1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up2082    <idle>-0       3d..3    5us : __schedule <-schedule2083    <idle>-0       3d..3    5us :      0:120:R ==> [003]  2389: 94:R sleep2084 2085 2086Running this on an idle system, we see that it only took 5 microseconds2087to perform the task switch.  Note, since the trace point in the schedule2088is before the actual "switch", we stop the tracing when the recorded task2089is about to schedule in. This may change if we add a new marker at the2090end of the scheduler.2091 2092Notice that the recorded task is 'sleep' with the PID of 23892093and it has an rt_prio of 5. This priority is user-space priority2094and not the internal kernel priority. The policy is 1 for2095SCHED_FIFO and 2 for SCHED_RR.2096 2097Note, that the trace data shows the internal priority (99 - rtprio).2098::2099 2100  <idle>-0       3d..3    5us :      0:120:R ==> [003]  2389: 94:R sleep2101 2102The 0:120:R means idle was running with a nice priority of 0 (120 - 120)2103and in the running state 'R'. The sleep task was scheduled in with21042389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94)2105and it too is in the running state.2106 2107Doing the same with chrt -r 5 and function-trace set.2108::2109 2110  echo 1 > options/function-trace2111 2112  # tracer: wakeup_rt2113  #2114  # wakeup_rt latency trace v1.1.5 on 3.8.0-test+2115  # --------------------------------------------------------------------2116  # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)2117  #    -----------------2118  #    | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5)2119  #    -----------------2120  #2121  #                  _------=> CPU#            2122  #                 / _-----=> irqs-off        2123  #                | / _----=> need-resched    2124  #                || / _---=> hardirq/softirq 2125  #                ||| / _--=> preempt-depth   2126  #                |||| /     delay             2127  #  cmd     pid   ||||| time  |   caller      2128  #     \   /      |||||  \    |   /           2129    <idle>-0       3d.h4    1us+:      0:120:R   + [003]  2448: 94:R sleep2130    <idle>-0       3d.h4    2us : ttwu_do_activate.constprop.87 <-try_to_wake_up2131    <idle>-0       3d.h3    3us : check_preempt_curr <-ttwu_do_wakeup2132    <idle>-0       3d.h3    3us : resched_curr <-check_preempt_curr2133    <idle>-0       3dNh3    4us : task_woken_rt <-ttwu_do_wakeup2134    <idle>-0       3dNh3    4us : _raw_spin_unlock <-try_to_wake_up2135    <idle>-0       3dNh3    4us : sub_preempt_count <-_raw_spin_unlock2136    <idle>-0       3dNh2    5us : ttwu_stat <-try_to_wake_up2137    <idle>-0       3dNh2    5us : _raw_spin_unlock_irqrestore <-try_to_wake_up2138    <idle>-0       3dNh2    6us : sub_preempt_count <-_raw_spin_unlock_irqrestore2139    <idle>-0       3dNh1    6us : _raw_spin_lock <-__run_hrtimer2140    <idle>-0       3dNh1    6us : add_preempt_count <-_raw_spin_lock2141    <idle>-0       3dNh2    7us : _raw_spin_unlock <-hrtimer_interrupt2142    <idle>-0       3dNh2    7us : sub_preempt_count <-_raw_spin_unlock2143    <idle>-0       3dNh1    7us : tick_program_event <-hrtimer_interrupt2144    <idle>-0       3dNh1    7us : clockevents_program_event <-tick_program_event2145    <idle>-0       3dNh1    8us : ktime_get <-clockevents_program_event2146    <idle>-0       3dNh1    8us : lapic_next_event <-clockevents_program_event2147    <idle>-0       3dNh1    8us : irq_exit <-smp_apic_timer_interrupt2148    <idle>-0       3dNh1    9us : sub_preempt_count <-irq_exit2149    <idle>-0       3dN.2    9us : idle_cpu <-irq_exit2150    <idle>-0       3dN.2    9us : rcu_irq_exit <-irq_exit2151    <idle>-0       3dN.2   10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit2152    <idle>-0       3dN.2   10us : sub_preempt_count <-irq_exit2153    <idle>-0       3.N.1   11us : rcu_idle_exit <-cpu_idle2154    <idle>-0       3dN.1   11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit2155    <idle>-0       3.N.1   11us : tick_nohz_idle_exit <-cpu_idle2156    <idle>-0       3dN.1   12us : menu_hrtimer_cancel <-tick_nohz_idle_exit2157    <idle>-0       3dN.1   12us : ktime_get <-tick_nohz_idle_exit2158    <idle>-0       3dN.1   12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit2159    <idle>-0       3dN.1   13us : cpu_load_update_nohz <-tick_nohz_idle_exit2160    <idle>-0       3dN.1   13us : _raw_spin_lock <-cpu_load_update_nohz2161    <idle>-0       3dN.1   13us : add_preempt_count <-_raw_spin_lock2162    <idle>-0       3dN.2   13us : __cpu_load_update <-cpu_load_update_nohz2163    <idle>-0       3dN.2   14us : sched_avg_update <-__cpu_load_update2164    <idle>-0       3dN.2   14us : _raw_spin_unlock <-cpu_load_update_nohz2165    <idle>-0       3dN.2   14us : sub_preempt_count <-_raw_spin_unlock2166    <idle>-0       3dN.1   15us : calc_load_nohz_stop <-tick_nohz_idle_exit2167    <idle>-0       3dN.1   15us : touch_softlockup_watchdog <-tick_nohz_idle_exit2168    <idle>-0       3dN.1   15us : hrtimer_cancel <-tick_nohz_idle_exit2169    <idle>-0       3dN.1   15us : hrtimer_try_to_cancel <-hrtimer_cancel2170    <idle>-0       3dN.1   16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel2171    <idle>-0       3dN.1   16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.182172    <idle>-0       3dN.1   16us : add_preempt_count <-_raw_spin_lock_irqsave2173    <idle>-0       3dN.2   17us : __remove_hrtimer <-remove_hrtimer.part.162174    <idle>-0       3dN.2   17us : hrtimer_force_reprogram <-__remove_hrtimer2175    <idle>-0       3dN.2   17us : tick_program_event <-hrtimer_force_reprogram2176    <idle>-0       3dN.2   18us : clockevents_program_event <-tick_program_event2177    <idle>-0       3dN.2   18us : ktime_get <-clockevents_program_event2178    <idle>-0       3dN.2   18us : lapic_next_event <-clockevents_program_event2179    <idle>-0       3dN.2   19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel2180    <idle>-0       3dN.2   19us : sub_preempt_count <-_raw_spin_unlock_irqrestore2181    <idle>-0       3dN.1   19us : hrtimer_forward <-tick_nohz_idle_exit2182    <idle>-0       3dN.1   20us : ktime_add_safe <-hrtimer_forward2183    <idle>-0       3dN.1   20us : ktime_add_safe <-hrtimer_forward2184    <idle>-0       3dN.1   20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.112185    <idle>-0       3dN.1   20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns2186    <idle>-0       3dN.1   21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns2187    <idle>-0       3dN.1   21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.182188    <idle>-0       3dN.1   21us : add_preempt_count <-_raw_spin_lock_irqsave2189    <idle>-0       3dN.2   22us : ktime_add_safe <-__hrtimer_start_range_ns2190    <idle>-0       3dN.2   22us : enqueue_hrtimer <-__hrtimer_start_range_ns2191    <idle>-0       3dN.2   22us : tick_program_event <-__hrtimer_start_range_ns2192    <idle>-0       3dN.2   23us : clockevents_program_event <-tick_program_event2193    <idle>-0       3dN.2   23us : ktime_get <-clockevents_program_event2194    <idle>-0       3dN.2   23us : lapic_next_event <-clockevents_program_event2195    <idle>-0       3dN.2   24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns2196    <idle>-0       3dN.2   24us : sub_preempt_count <-_raw_spin_unlock_irqrestore2197    <idle>-0       3dN.1   24us : account_idle_ticks <-tick_nohz_idle_exit2198    <idle>-0       3dN.1   24us : account_idle_time <-account_idle_ticks2199    <idle>-0       3.N.1   25us : sub_preempt_count <-cpu_idle2200    <idle>-0       3.N..   25us : schedule <-cpu_idle2201    <idle>-0       3.N..   25us : __schedule <-preempt_schedule2202    <idle>-0       3.N..   26us : add_preempt_count <-__schedule2203    <idle>-0       3.N.1   26us : rcu_note_context_switch <-__schedule2204    <idle>-0       3.N.1   26us : rcu_sched_qs <-rcu_note_context_switch2205    <idle>-0       3dN.1   27us : rcu_preempt_qs <-rcu_note_context_switch2206    <idle>-0       3.N.1   27us : _raw_spin_lock_irq <-__schedule2207    <idle>-0       3dN.1   27us : add_preempt_count <-_raw_spin_lock_irq2208    <idle>-0       3dN.2   28us : put_prev_task_idle <-__schedule2209    <idle>-0       3dN.2   28us : pick_next_task_stop <-pick_next_task2210    <idle>-0       3dN.2   28us : pick_next_task_rt <-pick_next_task2211    <idle>-0       3dN.2   29us : dequeue_pushable_task <-pick_next_task_rt2212    <idle>-0       3d..3   29us : __schedule <-preempt_schedule2213    <idle>-0       3d..3   30us :      0:120:R ==> [003]  2448: 94:R sleep2214 2215This isn't that big of a trace, even with function tracing enabled,2216so I included the entire trace.2217 2218The interrupt went off while when the system was idle. Somewhere2219before task_woken_rt() was called, the NEED_RESCHED flag was set,2220this is indicated by the first occurrence of the 'N' flag.2221 2222Latency tracing and events2223--------------------------2224As function tracing can induce a much larger latency, but without2225seeing what happens within the latency it is hard to know what2226caused it. There is a middle ground, and that is with enabling2227events.2228::2229 2230  # echo 0 > options/function-trace2231  # echo wakeup_rt > current_tracer2232  # echo 1 > events/enable2233  # echo 1 > tracing_on2234  # echo 0 > tracing_max_latency2235  # chrt -f 5 sleep 12236  # echo 0 > tracing_on2237  # cat trace2238  # tracer: wakeup_rt2239  #2240  # wakeup_rt latency trace v1.1.5 on 3.8.0-test+2241  # --------------------------------------------------------------------2242  # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4)2243  #    -----------------2244  #    | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5)2245  #    -----------------2246  #2247  #                  _------=> CPU#            2248  #                 / _-----=> irqs-off        2249  #                | / _----=> need-resched    2250  #                || / _---=> hardirq/softirq 2251  #                ||| / _--=> preempt-depth   2252  #                |||| /     delay             2253  #  cmd     pid   ||||| time  |   caller      2254  #     \   /      |||||  \    |   /           2255    <idle>-0       2d.h4    0us :      0:120:R   + [002]  5882: 94:R sleep2256    <idle>-0       2d.h4    0us : ttwu_do_activate.constprop.87 <-try_to_wake_up2257    <idle>-0       2d.h4    1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=0022258    <idle>-0       2dNh2    1us : hrtimer_expire_exit: hrtimer=ffff88007796feb82259    <idle>-0       2.N.2    2us : power_end: cpu_id=22260    <idle>-0       2.N.2    3us : cpu_idle: state=4294967295 cpu_id=22261    <idle>-0       2dN.3    4us : hrtimer_cancel: hrtimer=ffff88007d50d5e02262    <idle>-0       2dN.3    4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=343112110000002263    <idle>-0       2.N.2    5us : rcu_utilization: Start context switch2264    <idle>-0       2.N.2    5us : rcu_utilization: End context switch2265    <idle>-0       2d..3    6us : __schedule <-schedule2266    <idle>-0       2d..3    6us :      0:120:R ==> [002]  5882: 94:R sleep2267 2268 2269Hardware Latency Detector2270-------------------------2271 2272The hardware latency detector is executed by enabling the "hwlat" tracer.2273 2274NOTE, this tracer will affect the performance of the system as it will2275periodically make a CPU constantly busy with interrupts disabled.2276::2277 2278  # echo hwlat > current_tracer2279  # sleep 1002280  # cat trace2281  # tracer: hwlat2282  #2283  # entries-in-buffer/entries-written: 13/13   #P:82284  #2285  #                              _-----=> irqs-off2286  #                             / _----=> need-resched2287  #                            | / _---=> hardirq/softirq2288  #                            || / _--=> preempt-depth2289  #                            ||| /     delay2290  #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION2291  #              | |       |   ||||       |         |2292             <...>-1729  [001] d...   678.473449: #1     inner/outer(us):   11/12    ts:1581527483.343962693 count:62293             <...>-1729  [004] d...   689.556542: #2     inner/outer(us):   16/9     ts:1581527494.889008092 count:12294             <...>-1729  [005] d...   714.756290: #3     inner/outer(us):   16/16    ts:1581527519.678961629 count:52295             <...>-1729  [001] d...   718.788247: #4     inner/outer(us):    9/17    ts:1581527523.889012713 count:12296             <...>-1729  [002] d...   719.796341: #5     inner/outer(us):   13/9     ts:1581527524.912872606 count:12297             <...>-1729  [006] d...   844.787091: #6     inner/outer(us):    9/12    ts:1581527649.889048502 count:22298             <...>-1729  [003] d...   849.827033: #7     inner/outer(us):   18/9     ts:1581527654.889013793 count:12299             <...>-1729  [007] d...   853.859002: #8     inner/outer(us):    9/12    ts:1581527658.889065736 count:12300             <...>-1729  [001] d...   855.874978: #9     inner/outer(us):    9/11    ts:1581527660.861991877 count:12301             <...>-1729  [001] d...   863.938932: #10    inner/outer(us):    9/11    ts:1581527668.970010500 count:1 nmi-total:7 nmi-count:12302             <...>-1729  [007] d...   878.050780: #11    inner/outer(us):    9/12    ts:1581527683.385002600 count:1 nmi-total:5 nmi-count:12303             <...>-1729  [007] d...   886.114702: #12    inner/outer(us):    9/12    ts:1581527691.385001600 count:12304 2305 2306The above output is somewhat the same in the header. All events will have2307interrupts disabled 'd'. Under the FUNCTION title there is:2308 2309 #12310	This is the count of events recorded that were greater than the2311	tracing_threshold (See below).2312 2313 inner/outer(us):   11/112314 2315      This shows two numbers as "inner latency" and "outer latency". The test2316      runs in a loop checking a timestamp twice. The latency detected within2317      the two timestamps is the "inner latency" and the latency detected2318      after the previous timestamp and the next timestamp in the loop is2319      the "outer latency".2320 2321 ts:1581527483.3439626932322 2323      The absolute timestamp that the first latency was recorded in the window.2324 2325 count:62326 2327      The number of times a latency was detected during the window.2328 2329 nmi-total:7 nmi-count:12330 2331      On architectures that support it, if an NMI comes in during the2332      test, the time spent in NMI is reported in "nmi-total" (in2333      microseconds).2334 2335      All architectures that have NMIs will show the "nmi-count" if an2336      NMI comes in during the test.2337 2338hwlat files:2339 2340  tracing_threshold2341	This gets automatically set to "10" to represent 102342	microseconds. This is the threshold of latency that2343	needs to be detected before the trace will be recorded.2344 2345	Note, when hwlat tracer is finished (another tracer is2346	written into "current_tracer"), the original value for2347	tracing_threshold is placed back into this file.2348 2349  hwlat_detector/width2350	The length of time the test runs with interrupts disabled.2351 2352  hwlat_detector/window2353	The length of time of the window which the test2354	runs. That is, the test will run for "width"2355	microseconds per "window" microseconds2356 2357  tracing_cpumask2358	When the test is started. A kernel thread is created that2359	runs the test. This thread will alternate between CPUs2360	listed in the tracing_cpumask between each period2361	(one "window"). To limit the test to specific CPUs2362	set the mask in this file to only the CPUs that the test2363	should run on.2364 2365function2366--------2367 2368This tracer is the function tracer. Enabling the function tracer2369can be done from the debug file system. Make sure the2370ftrace_enabled is set; otherwise this tracer is a nop.2371See the "ftrace_enabled" section below.2372::2373 2374  # sysctl kernel.ftrace_enabled=12375  # echo function > current_tracer2376  # echo 1 > tracing_on2377  # usleep 12378  # echo 0 > tracing_on2379  # cat trace2380  # tracer: function2381  #2382  # entries-in-buffer/entries-written: 24799/24799   #P:42383  #2384  #                              _-----=> irqs-off2385  #                             / _----=> need-resched2386  #                            | / _---=> hardirq/softirq2387  #                            || / _--=> preempt-depth2388  #                            ||| /     delay2389  #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION2390  #              | |       |   ||||       |         |2391              bash-1994  [002] ....  3082.063030: mutex_unlock <-rb_simple_write2392              bash-1994  [002] ....  3082.063031: __mutex_unlock_slowpath <-mutex_unlock2393              bash-1994  [002] ....  3082.063031: __fsnotify_parent <-fsnotify_modify2394              bash-1994  [002] ....  3082.063032: fsnotify <-fsnotify_modify2395              bash-1994  [002] ....  3082.063032: __srcu_read_lock <-fsnotify2396              bash-1994  [002] ....  3082.063032: add_preempt_count <-__srcu_read_lock2397              bash-1994  [002] ...1  3082.063032: sub_preempt_count <-__srcu_read_lock2398              bash-1994  [002] ....  3082.063033: __srcu_read_unlock <-fsnotify2399  [...]2400 2401 2402Note: function tracer uses ring buffers to store the above2403entries. The newest data may overwrite the oldest data.2404Sometimes using echo to stop the trace is not sufficient because2405the tracing could have overwritten the data that you wanted to2406record. For this reason, it is sometimes better to disable2407tracing directly from a program. This allows you to stop the2408tracing at the point that you hit the part that you are2409interested in. To disable the tracing directly from a C program,2410something like following code snippet can be used::2411 2412	int trace_fd;2413	[...]2414	int main(int argc, char *argv[]) {2415		[...]2416		trace_fd = open(tracing_file("tracing_on"), O_WRONLY);2417		[...]2418		if (condition_hit()) {2419			write(trace_fd, "0", 1);2420		}2421		[...]2422	}2423 2424 2425Single thread tracing2426---------------------2427 2428By writing into set_ftrace_pid you can trace a2429single thread. For example::2430 2431  # cat set_ftrace_pid2432  no pid2433  # echo 3111 > set_ftrace_pid2434  # cat set_ftrace_pid2435  31112436  # echo function > current_tracer2437  # cat trace | head2438  # tracer: function2439  #2440  #           TASK-PID    CPU#    TIMESTAMP  FUNCTION2441  #              | |       |          |         |2442      yum-updatesd-3111  [003]  1637.254676: finish_task_switch <-thread_return2443      yum-updatesd-3111  [003]  1637.254681: hrtimer_cancel <-schedule_hrtimeout_range2444      yum-updatesd-3111  [003]  1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel2445      yum-updatesd-3111  [003]  1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel2446      yum-updatesd-3111  [003]  1637.254685: fget_light <-do_sys_poll2447      yum-updatesd-3111  [003]  1637.254686: pipe_poll <-do_sys_poll2448  # echo > set_ftrace_pid2449  # cat trace |head2450  # tracer: function2451  #2452  #           TASK-PID    CPU#    TIMESTAMP  FUNCTION2453  #              | |       |          |         |2454  ##### CPU 3 buffer started ####2455      yum-updatesd-3111  [003]  1701.957688: free_poll_entry <-poll_freewait2456      yum-updatesd-3111  [003]  1701.957689: remove_wait_queue <-free_poll_entry2457      yum-updatesd-3111  [003]  1701.957691: fput <-free_poll_entry2458      yum-updatesd-3111  [003]  1701.957692: audit_syscall_exit <-sysret_audit2459      yum-updatesd-3111  [003]  1701.957693: path_put <-audit_syscall_exit2460 2461If you want to trace a function when executing, you could use2462something like this simple program.2463::2464 2465	#include <stdio.h>2466	#include <stdlib.h>2467	#include <sys/types.h>2468	#include <sys/stat.h>2469	#include <fcntl.h>2470	#include <unistd.h>2471	#include <string.h>2472 2473	#define _STR(x) #x2474	#define STR(x) _STR(x)2475	#define MAX_PATH 2562476 2477	const char *find_tracefs(void)2478	{2479	       static char tracefs[MAX_PATH+1];2480	       static int tracefs_found;2481	       char type[100];2482	       FILE *fp;2483 2484	       if (tracefs_found)2485		       return tracefs;2486 2487	       if ((fp = fopen("/proc/mounts","r")) == NULL) {2488		       perror("/proc/mounts");2489		       return NULL;2490	       }2491 2492	       while (fscanf(fp, "%*s %"2493		             STR(MAX_PATH)2494		             "s %99s %*s %*d %*d\n",2495		             tracefs, type) == 2) {2496		       if (strcmp(type, "tracefs") == 0)2497		               break;2498	       }2499	       fclose(fp);2500 2501	       if (strcmp(type, "tracefs") != 0) {2502		       fprintf(stderr, "tracefs not mounted");2503		       return NULL;2504	       }2505 2506	       strcat(tracefs, "/tracing/");2507	       tracefs_found = 1;2508 2509	       return tracefs;2510	}2511 2512	const char *tracing_file(const char *file_name)2513	{2514	       static char trace_file[MAX_PATH+1];2515	       snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name);2516	       return trace_file;2517	}2518 2519	int main (int argc, char **argv)2520	{2521		if (argc < 1)2522		        exit(-1);2523 2524		if (fork() > 0) {2525		        int fd, ffd;2526		        char line[64];2527		        int s;2528 2529		        ffd = open(tracing_file("current_tracer"), O_WRONLY);2530		        if (ffd < 0)2531		                exit(-1);2532		        write(ffd, "nop", 3);2533 2534		        fd = open(tracing_file("set_ftrace_pid"), O_WRONLY);2535		        s = sprintf(line, "%d\n", getpid());2536		        write(fd, line, s);2537 2538		        write(ffd, "function", 8);2539 2540		        close(fd);2541		        close(ffd);2542 2543		        execvp(argv[1], argv+1);2544		}2545 2546		return 0;2547	}2548 2549Or this simple script!2550::2551 2552  #!/bin/bash2553 2554  tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts`2555  echo 0 > $tracefs/tracing_on2556  echo $$ > $tracefs/set_ftrace_pid2557  echo function > $tracefs/current_tracer2558  echo 1 > $tracefs/tracing_on2559  exec "$@"2560 2561 2562function graph tracer2563---------------------------2564 2565This tracer is similar to the function tracer except that it2566probes a function on its entry and its exit. This is done by2567using a dynamically allocated stack of return addresses in each2568task_struct. On function entry the tracer overwrites the return2569address of each function traced to set a custom probe. Thus the2570original return address is stored on the stack of return address2571in the task_struct.2572 2573Probing on both ends of a function leads to special features2574such as:2575 2576- measure of a function's time execution2577- having a reliable call stack to draw function calls graph2578 2579This tracer is useful in several situations:2580 2581- you want to find the reason of a strange kernel behavior and2582  need to see what happens in detail on any areas (or specific2583  ones).2584 2585- you are experiencing weird latencies but it's difficult to2586  find its origin.2587 2588- you want to find quickly which path is taken by a specific2589  function2590 2591- you just want to peek inside a working kernel and want to see2592  what happens there.2593 2594::2595 2596  # tracer: function_graph2597  #2598  # CPU  DURATION                  FUNCTION CALLS2599  # |     |   |                     |   |   |   |2600 2601   0)               |  sys_open() {2602   0)               |    do_sys_open() {2603   0)               |      getname() {2604   0)               |        kmem_cache_alloc() {2605   0)   1.382 us    |          __might_sleep();2606   0)   2.478 us    |        }2607   0)               |        strncpy_from_user() {2608   0)               |          might_fault() {2609   0)   1.389 us    |            __might_sleep();2610   0)   2.553 us    |          }2611   0)   3.807 us    |        }2612   0)   7.876 us    |      }2613   0)               |      alloc_fd() {2614   0)   0.668 us    |        _spin_lock();2615   0)   0.570 us    |        expand_files();2616   0)   0.586 us    |        _spin_unlock();2617 2618 2619There are several columns that can be dynamically2620enabled/disabled. You can use every combination of options you2621want, depending on your needs.2622 2623- The cpu number on which the function executed is default2624  enabled.  It is sometimes better to only trace one cpu (see2625  tracing_cpumask file) or you might sometimes see unordered2626  function calls while cpu tracing switch.2627 2628	- hide: echo nofuncgraph-cpu > trace_options2629	- show: echo funcgraph-cpu > trace_options2630 2631- The duration (function's time of execution) is displayed on2632  the closing bracket line of a function or on the same line2633  than the current function in case of a leaf one. It is default2634  enabled.2635 2636	- hide: echo nofuncgraph-duration > trace_options2637	- show: echo funcgraph-duration > trace_options2638 2639- The overhead field precedes the duration field in case of2640  reached duration thresholds.2641 2642	- hide: echo nofuncgraph-overhead > trace_options2643	- show: echo funcgraph-overhead > trace_options2644	- depends on: funcgraph-duration2645 2646  ie::2647 2648    3) # 1837.709 us |          } /* __switch_to */2649    3)               |          finish_task_switch() {2650    3)   0.313 us    |            _raw_spin_unlock_irq();2651    3)   3.177 us    |          }2652    3) # 1889.063 us |        } /* __schedule */2653    3) ! 140.417 us  |      } /* __schedule */2654    3) # 2034.948 us |    } /* schedule */2655    3) * 33998.59 us |  } /* schedule_preempt_disabled */2656 2657    [...]2658 2659    1)   0.260 us    |              msecs_to_jiffies();2660    1)   0.313 us    |              __rcu_read_unlock();2661    1) + 61.770 us   |            }2662    1) + 64.479 us   |          }2663    1)   0.313 us    |          rcu_bh_qs();2664    1)   0.313 us    |          __local_bh_enable();2665    1) ! 217.240 us  |        }2666    1)   0.365 us    |        idle_cpu();2667    1)               |        rcu_irq_exit() {2668    1)   0.417 us    |          rcu_eqs_enter_common.isra.47();2669    1)   3.125 us    |        }2670    1) ! 227.812 us  |      }2671    1) ! 457.395 us  |    }2672    1) @ 119760.2 us |  }2673 2674    [...]2675 2676    2)               |    handle_IPI() {2677    1)   6.979 us    |                  }2678    2)   0.417 us    |      scheduler_ipi();2679    1)   9.791 us    |                }2680    1) + 12.917 us   |              }2681    2)   3.490 us    |    }2682    1) + 15.729 us   |            }2683    1) + 18.542 us   |          }2684    2) $ 3594274 us  |  }2685 2686Flags::2687 2688  + means that the function exceeded 10 usecs.2689  ! means that the function exceeded 100 usecs.2690  # means that the function exceeded 1000 usecs.2691  * means that the function exceeded 10 msecs.2692  @ means that the function exceeded 100 msecs.2693  $ means that the function exceeded 1 sec.2694 2695 2696- The task/pid field displays the thread cmdline and pid which2697  executed the function. It is default disabled.2698 2699	- hide: echo nofuncgraph-proc > trace_options2700	- show: echo funcgraph-proc > trace_options2701 2702  ie::2703 2704    # tracer: function_graph2705    #2706    # CPU  TASK/PID        DURATION                  FUNCTION CALLS2707    # |    |    |           |   |                     |   |   |   |2708    0)    sh-4802     |               |                  d_free() {2709    0)    sh-4802     |               |                    call_rcu() {2710    0)    sh-4802     |               |                      __call_rcu() {2711    0)    sh-4802     |   0.616 us    |                        rcu_process_gp_end();2712    0)    sh-4802     |   0.586 us    |                        check_for_new_grace_period();2713    0)    sh-4802     |   2.899 us    |                      }2714    0)    sh-4802     |   4.040 us    |                    }2715    0)    sh-4802     |   5.151 us    |                  }2716    0)    sh-4802     | + 49.370 us   |                }2717 2718 2719- The absolute time field is an absolute timestamp given by the2720  system clock since it started. A snapshot of this time is2721  given on each entry/exit of functions2722 2723	- hide: echo nofuncgraph-abstime > trace_options2724	- show: echo funcgraph-abstime > trace_options2725 2726  ie::2727 2728    #2729    #      TIME       CPU  DURATION                  FUNCTION CALLS2730    #       |         |     |   |                     |   |   |   |2731    360.774522 |   1)   0.541 us    |                                          }2732    360.774522 |   1)   4.663 us    |                                        }2733    360.774523 |   1)   0.541 us    |                                        __wake_up_bit();2734    360.774524 |   1)   6.796 us    |                                      }2735    360.774524 |   1)   7.952 us    |                                    }2736    360.774525 |   1)   9.063 us    |                                  }2737    360.774525 |   1)   0.615 us    |                                  journal_mark_dirty();2738    360.774527 |   1)   0.578 us    |                                  __brelse();2739    360.774528 |   1)               |                                  reiserfs_prepare_for_journal() {2740    360.774528 |   1)               |                                    unlock_buffer() {2741    360.774529 |   1)               |                                      wake_up_bit() {2742    360.774529 |   1)               |                                        bit_waitqueue() {2743    360.774530 |   1)   0.594 us    |                                          __phys_addr();2744 2745 2746The function name is always displayed after the closing bracket2747for a function if the start of that function is not in the2748trace buffer.2749 2750Display of the function name after the closing bracket may be2751enabled for functions whose start is in the trace buffer,2752allowing easier searching with grep for function durations.2753It is default disabled.2754 2755	- hide: echo nofuncgraph-tail > trace_options2756	- show: echo funcgraph-tail > trace_options2757 2758  Example with nofuncgraph-tail (default)::2759 2760    0)               |      putname() {2761    0)               |        kmem_cache_free() {2762    0)   0.518 us    |          __phys_addr();2763    0)   1.757 us    |        }2764    0)   2.861 us    |      }2765 2766  Example with funcgraph-tail::2767 2768    0)               |      putname() {2769    0)               |        kmem_cache_free() {2770    0)   0.518 us    |          __phys_addr();2771    0)   1.757 us    |        } /* kmem_cache_free() */2772    0)   2.861 us    |      } /* putname() */2773 2774The return value of each traced function can be displayed after2775an equal sign "=". When encountering system call failures, it2776can be very helpful to quickly locate the function that first2777returns an error code.2778 2779	- hide: echo nofuncgraph-retval > trace_options2780	- show: echo funcgraph-retval > trace_options2781 2782  Example with funcgraph-retval::2783 2784    1)               |    cgroup_migrate() {2785    1)   0.651 us    |      cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */2786    1)               |      cgroup_migrate_execute() {2787    1)               |        cpu_cgroup_can_attach() {2788    1)               |          cgroup_taskset_first() {2789    1)   0.732 us    |            cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */2790    1)   1.232 us    |          } /* cgroup_taskset_first = 0xffff93fc8fb20000 */2791    1)   0.380 us    |          sched_rt_can_attach(); /* = 0x0 */2792    1)   2.335 us    |        } /* cpu_cgroup_can_attach = -22 */2793    1)   4.369 us    |      } /* cgroup_migrate_execute = -22 */2794    1)   7.143 us    |    } /* cgroup_migrate = -22 */2795 2796The above example shows that the function cpu_cgroup_can_attach2797returned the error code -22 firstly, then we can read the code2798of this function to get the root cause.2799 2800When the option funcgraph-retval-hex is not set, the return value can2801be displayed in a smart way. Specifically, if it is an error code,2802it will be printed in signed decimal format, otherwise it will2803printed in hexadecimal format.2804 2805	- smart: echo nofuncgraph-retval-hex > trace_options2806	- hexadecimal: echo funcgraph-retval-hex > trace_options2807 2808  Example with funcgraph-retval-hex::2809 2810    1)               |      cgroup_migrate() {2811    1)   0.651 us    |        cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */2812    1)               |        cgroup_migrate_execute() {2813    1)               |          cpu_cgroup_can_attach() {2814    1)               |            cgroup_taskset_first() {2815    1)   0.732 us    |              cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */2816    1)   1.232 us    |            } /* cgroup_taskset_first = 0xffff93fc8fb20000 */2817    1)   0.380 us    |            sched_rt_can_attach(); /* = 0x0 */2818    1)   2.335 us    |          } /* cpu_cgroup_can_attach = 0xffffffea */2819    1)   4.369 us    |        } /* cgroup_migrate_execute = 0xffffffea */2820    1)   7.143 us    |      } /* cgroup_migrate = 0xffffffea */2821 2822At present, there are some limitations when using the funcgraph-retval2823option, and these limitations will be eliminated in the future:2824 2825- Even if the function return type is void, a return value will still2826  be printed, and you can just ignore it.2827 2828- Even if return values are stored in multiple registers, only the2829  value contained in the first register will be recorded and printed.2830  To illustrate, in the x86 architecture, eax and edx are used to store2831  a 64-bit return value, with the lower 32 bits saved in eax and the2832  upper 32 bits saved in edx. However, only the value stored in eax2833  will be recorded and printed.2834 2835- In certain procedure call standards, such as arm64's AAPCS64, when a2836  type is smaller than a GPR, it is the responsibility of the consumer2837  to perform the narrowing, and the upper bits may contain UNKNOWN values.2838  Therefore, it is advisable to check the code for such cases. For instance,2839  when using a u8 in a 64-bit GPR, bits [63:8] may contain arbitrary values,2840  especially when larger types are truncated, whether explicitly or implicitly.2841  Here are some specific cases to illustrate this point:2842 2843  **Case One**:2844 2845  The function narrow_to_u8 is defined as follows::2846 2847	u8 narrow_to_u8(u64 val)2848	{2849		// implicitly truncated2850		return val;2851	}2852 2853  It may be compiled to::2854 2855	narrow_to_u8:2856		< ... ftrace instrumentation ... >2857		RET2858 2859  If you pass 0x123456789abcdef to this function and want to narrow it,2860  it may be recorded as 0x123456789abcdef instead of 0xef.2861 2862  **Case Two**:2863 2864  The function error_if_not_4g_aligned is defined as follows::2865 2866	int error_if_not_4g_aligned(u64 val)2867	{2868		if (val & GENMASK(31, 0))2869			return -EINVAL;2870 2871		return 0;2872	}2873 2874  It could be compiled to::2875 2876	error_if_not_4g_aligned:2877		CBNZ    w0, .Lnot_aligned2878		RET			// bits [31:0] are zero, bits2879					// [63:32] are UNKNOWN2880	.Lnot_aligned:2881		MOV    x0, #-EINVAL2882		RET2883 2884  When passing 0x2_0000_0000 to it, the return value may be recorded as2885  0x2_0000_0000 instead of 0.2886 2887You can put some comments on specific functions by using2888trace_printk() For example, if you want to put a comment inside2889the __might_sleep() function, you just have to include2890<linux/ftrace.h> and call trace_printk() inside __might_sleep()::2891 2892	trace_printk("I'm a comment!\n")2893 2894will produce::2895 2896   1)               |             __might_sleep() {2897   1)               |                /* I'm a comment! */2898   1)   1.449 us    |             }2899 2900 2901You might find other useful features for this tracer in the2902following "dynamic ftrace" section such as tracing only specific2903functions or tasks.2904 2905dynamic ftrace2906--------------2907 2908If CONFIG_DYNAMIC_FTRACE is set, the system will run with2909virtually no overhead when function tracing is disabled. The way2910this works is the mcount function call (placed at the start of2911every kernel function, produced by the -pg switch in gcc),2912starts of pointing to a simple return. (Enabling FTRACE will2913include the -pg switch in the compiling of the kernel.)2914 2915At compile time every C file object is run through the2916recordmcount program (located in the scripts directory). This2917program will parse the ELF headers in the C object to find all2918the locations in the .text section that call mcount. Starting2919with gcc version 4.6, the -mfentry has been added for x86, which2920calls "__fentry__" instead of "mcount". Which is called before2921the creation of the stack frame.2922 2923Note, not all sections are traced. They may be prevented by either2924a notrace, or blocked another way and all inline functions are not2925traced. Check the "available_filter_functions" file to see what functions2926can be traced.2927 2928A section called "__mcount_loc" is created that holds2929references to all the mcount/fentry call sites in the .text section.2930The recordmcount program re-links this section back into the2931original object. The final linking stage of the kernel will add all these2932references into a single table.2933 2934On boot up, before SMP is initialized, the dynamic ftrace code2935scans this table and updates all the locations into nops. It2936also records the locations, which are added to the2937available_filter_functions list.  Modules are processed as they2938are loaded and before they are executed.  When a module is2939unloaded, it also removes its functions from the ftrace function2940list. This is automatic in the module unload code, and the2941module author does not need to worry about it.2942 2943When tracing is enabled, the process of modifying the function2944tracepoints is dependent on architecture. The old method is to use2945kstop_machine to prevent races with the CPUs executing code being2946modified (which can cause the CPU to do undesirable things, especially2947if the modified code crosses cache (or page) boundaries), and the nops are2948patched back to calls. But this time, they do not call mcount2949(which is just a function stub). They now call into the ftrace2950infrastructure.2951 2952The new method of modifying the function tracepoints is to place2953a breakpoint at the location to be modified, sync all CPUs, modify2954the rest of the instruction not covered by the breakpoint. Sync2955all CPUs again, and then remove the breakpoint with the finished2956version to the ftrace call site.2957 2958Some archs do not even need to monkey around with the synchronization,2959and can just slap the new code on top of the old without any2960problems with other CPUs executing it at the same time.2961 2962One special side-effect to the recording of the functions being2963traced is that we can now selectively choose which functions we2964wish to trace and which ones we want the mcount calls to remain2965as nops.2966 2967Two files are used, one for enabling and one for disabling the2968tracing of specified functions. They are:2969 2970  set_ftrace_filter2971 2972and2973 2974  set_ftrace_notrace2975 2976A list of available functions that you can add to these files is2977listed in:2978 2979   available_filter_functions2980 2981::2982 2983  # cat available_filter_functions2984  put_prev_task_idle2985  kmem_cache_create2986  pick_next_task_rt2987  cpus_read_lock2988  pick_next_task_fair2989  mutex_lock2990  [...]2991 2992If I am only interested in sys_nanosleep and hrtimer_interrupt::2993 2994  # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter2995  # echo function > current_tracer2996  # echo 1 > tracing_on2997  # usleep 12998  # echo 0 > tracing_on2999  # cat trace3000  # tracer: function3001  #3002  # entries-in-buffer/entries-written: 5/5   #P:43003  #3004  #                              _-----=> irqs-off3005  #                             / _----=> need-resched3006  #                            | / _---=> hardirq/softirq3007  #                            || / _--=> preempt-depth3008  #                            ||| /     delay3009  #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION3010  #              | |       |   ||||       |         |3011            usleep-2665  [001] ....  4186.475355: sys_nanosleep <-system_call_fastpath3012            <idle>-0     [001] d.h1  4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt3013            usleep-2665  [001] d.h1  4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt3014            <idle>-0     [003] d.h1  4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt3015            <idle>-0     [002] d.h1  4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt3016 3017To see which functions are being traced, you can cat the file:3018::3019 3020  # cat set_ftrace_filter3021  hrtimer_interrupt3022  sys_nanosleep3023 3024 3025Perhaps this is not enough. The filters also allow glob(7) matching.3026 3027  ``<match>*``3028	will match functions that begin with <match>3029  ``*<match>``3030	will match functions that end with <match>3031  ``*<match>*``3032	will match functions that have <match> in it3033  ``<match1>*<match2>``3034	will match functions that begin with <match1> and end with <match2>3035 3036.. note::3037      It is better to use quotes to enclose the wild cards,3038      otherwise the shell may expand the parameters into names3039      of files in the local directory.3040 3041::3042 3043  # echo 'hrtimer_*' > set_ftrace_filter3044 3045Produces::3046 3047  # tracer: function3048  #3049  # entries-in-buffer/entries-written: 897/897   #P:43050  #3051  #                              _-----=> irqs-off3052  #                             / _----=> need-resched3053  #                            | / _---=> hardirq/softirq3054  #                            || / _--=> preempt-depth3055  #                            ||| /     delay3056  #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION3057  #              | |       |   ||||       |         |3058            <idle>-0     [003] dN.1  4228.547803: hrtimer_cancel <-tick_nohz_idle_exit3059            <idle>-0     [003] dN.1  4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel3060            <idle>-0     [003] dN.2  4228.547805: hrtimer_force_reprogram <-__remove_hrtimer3061            <idle>-0     [003] dN.1  4228.547805: hrtimer_forward <-tick_nohz_idle_exit3062            <idle>-0     [003] dN.1  4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.113063            <idle>-0     [003] d..1  4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt3064            <idle>-0     [003] d..1  4228.547859: hrtimer_start <-__tick_nohz_idle_enter3065            <idle>-0     [003] d..2  4228.547860: hrtimer_force_reprogram <-__rem3066 3067Notice that we lost the sys_nanosleep.3068::3069 3070  # cat set_ftrace_filter3071  hrtimer_run_queues3072  hrtimer_run_pending3073  hrtimer_init3074  hrtimer_cancel3075  hrtimer_try_to_cancel3076  hrtimer_forward3077  hrtimer_start3078  hrtimer_reprogram3079  hrtimer_force_reprogram3080  hrtimer_get_next_event3081  hrtimer_interrupt3082  hrtimer_nanosleep3083  hrtimer_wakeup3084  hrtimer_get_remaining3085  hrtimer_get_res3086  hrtimer_init_sleeper3087 3088 3089This is because the '>' and '>>' act just like they do in bash.3090To rewrite the filters, use '>'3091To append to the filters, use '>>'3092 3093To clear out a filter so that all functions will be recorded3094again::3095 3096 # echo > set_ftrace_filter3097 # cat set_ftrace_filter3098 #3099 3100Again, now we want to append.3101 3102::3103 3104  # echo sys_nanosleep > set_ftrace_filter3105  # cat set_ftrace_filter3106  sys_nanosleep3107  # echo 'hrtimer_*' >> set_ftrace_filter3108  # cat set_ftrace_filter3109  hrtimer_run_queues3110  hrtimer_run_pending3111  hrtimer_init3112  hrtimer_cancel3113  hrtimer_try_to_cancel3114  hrtimer_forward3115  hrtimer_start3116  hrtimer_reprogram3117  hrtimer_force_reprogram3118  hrtimer_get_next_event3119  hrtimer_interrupt3120  sys_nanosleep3121  hrtimer_nanosleep3122  hrtimer_wakeup3123  hrtimer_get_remaining3124  hrtimer_get_res3125  hrtimer_init_sleeper3126 3127 3128The set_ftrace_notrace prevents those functions from being3129traced.3130::3131 3132  # echo '*preempt*' '*lock*' > set_ftrace_notrace3133 3134Produces::3135 3136  # tracer: function3137  #3138  # entries-in-buffer/entries-written: 39608/39608   #P:43139  #3140  #                              _-----=> irqs-off3141  #                             / _----=> need-resched3142  #                            | / _---=> hardirq/softirq3143  #                            || / _--=> preempt-depth3144  #                            ||| /     delay3145  #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION3146  #              | |       |   ||||       |         |3147              bash-1994  [000] ....  4342.324896: file_ra_state_init <-do_dentry_open3148              bash-1994  [000] ....  4342.324897: open_check_o_direct <-do_last3149              bash-1994  [000] ....  4342.324897: ima_file_check <-do_last3150              bash-1994  [000] ....  4342.324898: process_measurement <-ima_file_check3151              bash-1994  [000] ....  4342.324898: ima_get_action <-process_measurement3152              bash-1994  [000] ....  4342.324898: ima_match_policy <-ima_get_action3153              bash-1994  [000] ....  4342.324899: do_truncate <-do_last3154              bash-1994  [000] ....  4342.324899: setattr_should_drop_suidgid <-do_truncate3155              bash-1994  [000] ....  4342.324899: notify_change <-do_truncate3156              bash-1994  [000] ....  4342.324900: current_fs_time <-notify_change3157              bash-1994  [000] ....  4342.324900: current_kernel_time <-current_fs_time3158              bash-1994  [000] ....  4342.324900: timespec_trunc <-current_fs_time3159 3160We can see that there's no more lock or preempt tracing.3161 3162Selecting function filters via index3163------------------------------------3164 3165Because processing of strings is expensive (the address of the function3166needs to be looked up before comparing to the string being passed in),3167an index can be used as well to enable functions. This is useful in the3168case of setting thousands of specific functions at a time. By passing3169in a list of numbers, no string processing will occur. Instead, the function3170at the specific location in the internal array (which corresponds to the3171functions in the "available_filter_functions" file), is selected.3172 3173::3174 3175  # echo 1 > set_ftrace_filter3176 3177Will select the first function listed in "available_filter_functions"3178 3179::3180 3181  # head -1 available_filter_functions3182  trace_initcall_finish_cb3183 3184  # cat set_ftrace_filter3185  trace_initcall_finish_cb3186 3187  # head -50 available_filter_functions | tail -13188  x86_pmu_commit_txn3189 3190  # echo 1 50 > set_ftrace_filter3191  # cat set_ftrace_filter3192  trace_initcall_finish_cb3193  x86_pmu_commit_txn3194 3195Dynamic ftrace with the function graph tracer3196---------------------------------------------3197 3198Although what has been explained above concerns both the3199function tracer and the function-graph-tracer, there are some3200special features only available in the function-graph tracer.3201 3202If you want to trace only one function and all of its children,3203you just have to echo its name into set_graph_function::3204 3205 echo __do_fault > set_graph_function3206 3207will produce the following "expanded" trace of the __do_fault()3208function::3209 3210   0)               |  __do_fault() {3211   0)               |    filemap_fault() {3212   0)               |      find_lock_page() {3213   0)   0.804 us    |        find_get_page();3214   0)               |        __might_sleep() {3215   0)   1.329 us    |        }3216   0)   3.904 us    |      }3217   0)   4.979 us    |    }3218   0)   0.653 us    |    _spin_lock();3219   0)   0.578 us    |    page_add_file_rmap();3220   0)   0.525 us    |    native_set_pte_at();3221   0)   0.585 us    |    _spin_unlock();3222   0)               |    unlock_page() {3223   0)   0.541 us    |      page_waitqueue();3224   0)   0.639 us    |      __wake_up_bit();3225   0)   2.786 us    |    }3226   0) + 14.237 us   |  }3227   0)               |  __do_fault() {3228   0)               |    filemap_fault() {3229   0)               |      find_lock_page() {3230   0)   0.698 us    |        find_get_page();3231   0)               |        __might_sleep() {3232   0)   1.412 us    |        }3233   0)   3.950 us    |      }3234   0)   5.098 us    |    }3235   0)   0.631 us    |    _spin_lock();3236   0)   0.571 us    |    page_add_file_rmap();3237   0)   0.526 us    |    native_set_pte_at();3238   0)   0.586 us    |    _spin_unlock();3239   0)               |    unlock_page() {3240   0)   0.533 us    |      page_waitqueue();3241   0)   0.638 us    |      __wake_up_bit();3242   0)   2.793 us    |    }3243   0) + 14.012 us   |  }3244 3245You can also expand several functions at once::3246 3247 echo sys_open > set_graph_function3248 echo sys_close >> set_graph_function3249 3250Now if you want to go back to trace all functions you can clear3251this special filter via::3252 3253 echo > set_graph_function3254 3255 3256ftrace_enabled3257--------------3258 3259Note, the proc sysctl ftrace_enable is a big on/off switch for the3260function tracer. By default it is enabled (when function tracing is3261enabled in the kernel). If it is disabled, all function tracing is3262disabled. This includes not only the function tracers for ftrace, but3263also for any other uses (perf, kprobes, stack tracing, profiling, etc). It3264cannot be disabled if there is a callback with FTRACE_OPS_FL_PERMANENT set3265registered.3266 3267Please disable this with care.3268 3269This can be disable (and enabled) with::3270 3271  sysctl kernel.ftrace_enabled=03272  sysctl kernel.ftrace_enabled=13273 3274 or3275 3276  echo 0 > /proc/sys/kernel/ftrace_enabled3277  echo 1 > /proc/sys/kernel/ftrace_enabled3278 3279 3280Filter commands3281---------------3282 3283A few commands are supported by the set_ftrace_filter interface.3284Trace commands have the following format::3285 3286  <function>:<command>:<parameter>3287 3288The following commands are supported:3289 3290- mod:3291  This command enables function filtering per module. The3292  parameter defines the module. For example, if only the write*3293  functions in the ext3 module are desired, run:3294 3295   echo 'write*:mod:ext3' > set_ftrace_filter3296 3297  This command interacts with the filter in the same way as3298  filtering based on function names. Thus, adding more functions3299  in a different module is accomplished by appending (>>) to the3300  filter file. Remove specific module functions by prepending3301  '!'::3302 3303   echo '!writeback*:mod:ext3' >> set_ftrace_filter3304 3305  Mod command supports module globbing. Disable tracing for all3306  functions except a specific module::3307 3308   echo '!*:mod:!ext3' >> set_ftrace_filter3309 3310  Disable tracing for all modules, but still trace kernel::3311 3312   echo '!*:mod:*' >> set_ftrace_filter3313 3314  Enable filter only for kernel::3315 3316   echo '*write*:mod:!*' >> set_ftrace_filter3317 3318  Enable filter for module globbing::3319 3320   echo '*write*:mod:*snd*' >> set_ftrace_filter3321 3322- traceon/traceoff:3323  These commands turn tracing on and off when the specified3324  functions are hit. The parameter determines how many times the3325  tracing system is turned on and off. If unspecified, there is3326  no limit. For example, to disable tracing when a schedule bug3327  is hit the first 5 times, run::3328 3329   echo '__schedule_bug:traceoff:5' > set_ftrace_filter3330 3331  To always disable tracing when __schedule_bug is hit::3332 3333   echo '__schedule_bug:traceoff' > set_ftrace_filter3334 3335  These commands are cumulative whether or not they are appended3336  to set_ftrace_filter. To remove a command, prepend it by '!'3337  and drop the parameter::3338 3339   echo '!__schedule_bug:traceoff:0' > set_ftrace_filter3340 3341  The above removes the traceoff command for __schedule_bug3342  that have a counter. To remove commands without counters::3343 3344   echo '!__schedule_bug:traceoff' > set_ftrace_filter3345 3346- snapshot:3347  Will cause a snapshot to be triggered when the function is hit.3348  ::3349 3350   echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter3351 3352  To only snapshot once:3353  ::3354 3355   echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter3356 3357  To remove the above commands::3358 3359   echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter3360   echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter3361 3362- enable_event/disable_event:3363  These commands can enable or disable a trace event. Note, because3364  function tracing callbacks are very sensitive, when these commands3365  are registered, the trace point is activated, but disabled in3366  a "soft" mode. That is, the tracepoint will be called, but3367  just will not be traced. The event tracepoint stays in this mode3368  as long as there's a command that triggers it.3369  ::3370 3371   echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \3372   	 set_ftrace_filter3373 3374  The format is::3375 3376    <function>:enable_event:<system>:<event>[:count]3377    <function>:disable_event:<system>:<event>[:count]3378 3379  To remove the events commands::3380 3381   echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \3382   	 set_ftrace_filter3383   echo '!schedule:disable_event:sched:sched_switch' > \3384   	 set_ftrace_filter3385 3386- dump:3387  When the function is hit, it will dump the contents of the ftrace3388  ring buffer to the console. This is useful if you need to debug3389  something, and want to dump the trace when a certain function3390  is hit. Perhaps it's a function that is called before a triple3391  fault happens and does not allow you to get a regular dump.3392 3393- cpudump:3394  When the function is hit, it will dump the contents of the ftrace3395  ring buffer for the current CPU to the console. Unlike the "dump"3396  command, it only prints out the contents of the ring buffer for the3397  CPU that executed the function that triggered the dump.3398 3399- stacktrace:3400  When the function is hit, a stack trace is recorded.3401 3402trace_pipe3403----------3404 3405The trace_pipe outputs the same content as the trace file, but3406the effect on the tracing is different. Every read from3407trace_pipe is consumed. This means that subsequent reads will be3408different. The trace is live.3409::3410 3411  # echo function > current_tracer3412  # cat trace_pipe > /tmp/trace.out &3413  [1] 41533414  # echo 1 > tracing_on3415  # usleep 13416  # echo 0 > tracing_on3417  # cat trace3418  # tracer: function3419  #3420  # entries-in-buffer/entries-written: 0/0   #P:43421  #3422  #                              _-----=> irqs-off3423  #                             / _----=> need-resched3424  #                            | / _---=> hardirq/softirq3425  #                            || / _--=> preempt-depth3426  #                            ||| /     delay3427  #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION3428  #              | |       |   ||||       |         |3429 3430  #3431  # cat /tmp/trace.out3432             bash-1994  [000] ....  5281.568961: mutex_unlock <-rb_simple_write3433             bash-1994  [000] ....  5281.568963: __mutex_unlock_slowpath <-mutex_unlock3434             bash-1994  [000] ....  5281.568963: __fsnotify_parent <-fsnotify_modify3435             bash-1994  [000] ....  5281.568964: fsnotify <-fsnotify_modify3436             bash-1994  [000] ....  5281.568964: __srcu_read_lock <-fsnotify3437             bash-1994  [000] ....  5281.568964: add_preempt_count <-__srcu_read_lock3438             bash-1994  [000] ...1  5281.568965: sub_preempt_count <-__srcu_read_lock3439             bash-1994  [000] ....  5281.568965: __srcu_read_unlock <-fsnotify3440             bash-1994  [000] ....  5281.568967: sys_dup2 <-system_call_fastpath3441 3442 3443Note, reading the trace_pipe file will block until more input is3444added. This is contrary to the trace file. If any process opened3445the trace file for reading, it will actually disable tracing and3446prevent new entries from being added. The trace_pipe file does3447not have this limitation.3448 3449trace entries3450-------------3451 3452Having too much or not enough data can be troublesome in3453diagnosing an issue in the kernel. The file buffer_size_kb is3454used to modify the size of the internal trace buffers. The3455number listed is the number of entries that can be recorded per3456CPU. To know the full size, multiply the number of possible CPUs3457with the number of entries.3458::3459 3460  # cat buffer_size_kb3461  1408 (units kilobytes)3462 3463Or simply read buffer_total_size_kb3464::3465 3466  # cat buffer_total_size_kb 3467  56323468 3469To modify the buffer, simple echo in a number (in 1024 byte segments).3470::3471 3472  # echo 10000 > buffer_size_kb3473  # cat buffer_size_kb3474  10000 (units kilobytes)3475 3476It will try to allocate as much as possible. If you allocate too3477much, it can cause Out-Of-Memory to trigger.3478::3479 3480  # echo 1000000000000 > buffer_size_kb3481  -bash: echo: write error: Cannot allocate memory3482  # cat buffer_size_kb3483  853484 3485The per_cpu buffers can be changed individually as well:3486::3487 3488  # echo 10000 > per_cpu/cpu0/buffer_size_kb3489  # echo 100 > per_cpu/cpu1/buffer_size_kb3490 3491When the per_cpu buffers are not the same, the buffer_size_kb3492at the top level will just show an X3493::3494 3495  # cat buffer_size_kb3496  X3497 3498This is where the buffer_total_size_kb is useful:3499::3500 3501  # cat buffer_total_size_kb 3502  129163503 3504Writing to the top level buffer_size_kb will reset all the buffers3505to be the same again.3506 3507Snapshot3508--------3509CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature3510available to all non latency tracers. (Latency tracers which3511record max latency, such as "irqsoff" or "wakeup", can't use3512this feature, since those are already using the snapshot3513mechanism internally.)3514 3515Snapshot preserves a current trace buffer at a particular point3516in time without stopping tracing. Ftrace swaps the current3517buffer with a spare buffer, and tracing continues in the new3518current (=previous spare) buffer.3519 3520The following tracefs files in "tracing" are related to this3521feature:3522 3523  snapshot:3524 3525	This is used to take a snapshot and to read the output3526	of the snapshot. Echo 1 into this file to allocate a3527	spare buffer and to take a snapshot (swap), then read3528	the snapshot from this file in the same format as3529	"trace" (described above in the section "The File3530	System"). Both reads snapshot and tracing are executable3531	in parallel. When the spare buffer is allocated, echoing3532	0 frees it, and echoing else (positive) values clear the3533	snapshot contents.3534	More details are shown in the table below.3535 3536	+--------------+------------+------------+------------+3537	|status\\input |     0      |     1      |    else    |3538	+==============+============+============+============+3539	|not allocated |(do nothing)| alloc+swap |(do nothing)|3540	+--------------+------------+------------+------------+3541	|allocated     |    free    |    swap    |   clear    |3542	+--------------+------------+------------+------------+3543 3544Here is an example of using the snapshot feature.3545::3546 3547  # echo 1 > events/sched/enable3548  # echo 1 > snapshot3549  # cat snapshot3550  # tracer: nop3551  #3552  # entries-in-buffer/entries-written: 71/71   #P:83553  #3554  #                              _-----=> irqs-off3555  #                             / _----=> need-resched3556  #                            | / _---=> hardirq/softirq3557  #                            || / _--=> preempt-depth3558  #                            ||| /     delay3559  #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION3560  #              | |       |   ||||       |         |3561            <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=1203562             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=1203563  [...]3564          <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  3565 3566  # cat trace  3567  # tracer: nop3568  #3569  # entries-in-buffer/entries-written: 77/77   #P:83570  #3571  #                              _-----=> irqs-off3572  #                             / _----=> need-resched3573  #                            | / _---=> hardirq/softirq3574  #                            || / _--=> preempt-depth3575  #                            ||| /     delay3576  #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION3577  #              | |       |   ||||       |         |3578            <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=1203579   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=1203580  [...]3581 3582 3583If you try to use this snapshot feature when current tracer is3584one of the latency tracers, you will get the following results.3585::3586 3587  # echo wakeup > current_tracer3588  # echo 1 > snapshot3589  bash: echo: write error: Device or resource busy3590  # cat snapshot3591  cat: snapshot: Device or resource busy3592 3593 3594Instances3595---------3596In the tracefs tracing directory, there is a directory called "instances".3597This directory can have new directories created inside of it using3598mkdir, and removing directories with rmdir. The directory created3599with mkdir in this directory will already contain files and other3600directories after it is created.3601::3602 3603  # mkdir instances/foo3604  # ls instances/foo3605  buffer_size_kb  buffer_total_size_kb  events  free_buffer  per_cpu3606  set_event  snapshot  trace  trace_clock  trace_marker  trace_options3607  trace_pipe  tracing_on3608 3609As you can see, the new directory looks similar to the tracing directory3610itself. In fact, it is very similar, except that the buffer and3611events are agnostic from the main directory, or from any other3612instances that are created.3613 3614The files in the new directory work just like the files with the3615same name in the tracing directory except the buffer that is used3616is a separate and new buffer. The files affect that buffer but do not3617affect the main buffer with the exception of trace_options. Currently,3618the trace_options affect all instances and the top level buffer3619the same, but this may change in future releases. That is, options3620may become specific to the instance they reside in.3621 3622Notice that none of the function tracer files are there, nor is3623current_tracer and available_tracers. This is because the buffers3624can currently only have events enabled for them.3625::3626 3627  # mkdir instances/foo3628  # mkdir instances/bar3629  # mkdir instances/zoot3630  # echo 100000 > buffer_size_kb3631  # echo 1000 > instances/foo/buffer_size_kb3632  # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb3633  # echo function > current_trace3634  # echo 1 > instances/foo/events/sched/sched_wakeup/enable3635  # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable3636  # echo 1 > instances/foo/events/sched/sched_switch/enable3637  # echo 1 > instances/bar/events/irq/enable3638  # echo 1 > instances/zoot/events/syscalls/enable3639  # cat trace_pipe3640  CPU:2 [LOST 11745 EVENTS]3641              bash-2044  [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist3642              bash-2044  [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave3643              bash-2044  [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist3644              bash-2044  [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist3645              bash-2044  [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock3646              bash-2044  [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype3647              bash-2044  [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist3648              bash-2044  [002] d... 10594.481034: zone_statistics <-get_page_from_freelist3649              bash-2044  [002] d... 10594.481034: __inc_zone_state <-zone_statistics3650              bash-2044  [002] d... 10594.481034: __inc_zone_state <-zone_statistics3651              bash-2044  [002] .... 10594.481035: arch_dup_task_struct <-copy_process3652  [...]3653 3654  # cat instances/foo/trace_pipe3655              bash-1998  [000] d..4   136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=0003656              bash-1998  [000] dN.4   136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=0003657            <idle>-0     [003] d.h3   136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=0033658            <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=1203659       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=1203660              bash-1998  [000] d..4   136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=0003661              bash-1998  [000] dN.4   136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=0003662              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=1203663       kworker/0:1-59    [000] d..4   136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=0013664       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=1203665  [...]3666 3667  # cat instances/bar/trace_pipe3668       migration/1-14    [001] d.h3   138.732674: softirq_raise: vec=3 [action=NET_RX]3669            <idle>-0     [001] dNh3   138.732725: softirq_raise: vec=3 [action=NET_RX]3670              bash-1998  [000] d.h1   138.733101: softirq_raise: vec=1 [action=TIMER]3671              bash-1998  [000] d.h1   138.733102: softirq_raise: vec=9 [action=RCU]3672              bash-1998  [000] ..s2   138.733105: softirq_entry: vec=1 [action=TIMER]3673              bash-1998  [000] ..s2   138.733106: softirq_exit: vec=1 [action=TIMER]3674              bash-1998  [000] ..s2   138.733106: softirq_entry: vec=9 [action=RCU]3675              bash-1998  [000] ..s2   138.733109: softirq_exit: vec=9 [action=RCU]3676              sshd-1995  [001] d.h1   138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb43677              sshd-1995  [001] d.h1   138.733280: irq_handler_exit: irq=21 ret=unhandled3678              sshd-1995  [001] d.h1   138.733281: irq_handler_entry: irq=21 name=eth03679              sshd-1995  [001] d.h1   138.733283: irq_handler_exit: irq=21 ret=handled3680  [...]3681 3682  # cat instances/zoot/trace3683  # tracer: nop3684  #3685  # entries-in-buffer/entries-written: 18996/18996   #P:43686  #3687  #                              _-----=> irqs-off3688  #                             / _----=> need-resched3689  #                            | / _---=> hardirq/softirq3690  #                            || / _--=> preempt-depth3691  #                            ||| /     delay3692  #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION3693  #              | |       |   ||||       |         |3694              bash-1998  [000] d...   140.733501: sys_write -> 0x23695              bash-1998  [000] d...   140.733504: sys_dup2(oldfd: a, newfd: 1)3696              bash-1998  [000] d...   140.733506: sys_dup2 -> 0x13697              bash-1998  [000] d...   140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0)3698              bash-1998  [000] d...   140.733509: sys_fcntl -> 0x13699              bash-1998  [000] d...   140.733510: sys_close(fd: a)3700              bash-1998  [000] d...   140.733510: sys_close -> 0x03701              bash-1998  [000] d...   140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8)3702              bash-1998  [000] d...   140.733515: sys_rt_sigprocmask -> 0x03703              bash-1998  [000] d...   140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8)3704              bash-1998  [000] d...   140.733516: sys_rt_sigaction -> 0x03705 3706You can see that the trace of the top most trace buffer shows only3707the function tracing. The foo instance displays wakeups and task3708switches.3709 3710To remove the instances, simply delete their directories:3711::3712 3713  # rmdir instances/foo3714  # rmdir instances/bar3715  # rmdir instances/zoot3716 3717Note, if a process has a trace file open in one of the instance3718directories, the rmdir will fail with EBUSY.3719 3720 3721Stack trace3722-----------3723Since the kernel has a fixed sized stack, it is important not to3724waste it in functions. A kernel developer must be conscious of3725what they allocate on the stack. If they add too much, the system3726can be in danger of a stack overflow, and corruption will occur,3727usually leading to a system panic.3728 3729There are some tools that check this, usually with interrupts3730periodically checking usage. But if you can perform a check3731at every function call that will become very useful. As ftrace provides3732a function tracer, it makes it convenient to check the stack size3733at every function call. This is enabled via the stack tracer.3734 3735CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.3736To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.3737::3738 3739 # echo 1 > /proc/sys/kernel/stack_tracer_enabled3740 3741You can also enable it from the kernel command line to trace3742the stack size of the kernel during boot up, by adding "stacktrace"3743to the kernel command line parameter.3744 3745After running it for a few minutes, the output looks like:3746::3747 3748  # cat stack_max_size3749  29283750 3751  # cat stack_trace3752          Depth    Size   Location    (18 entries)3753          -----    ----   --------3754    0)     2928     224   update_sd_lb_stats+0xbc/0x4ac3755    1)     2704     160   find_busiest_group+0x31/0x1f13756    2)     2544     256   load_balance+0xd9/0x6623757    3)     2288      80   idle_balance+0xbb/0x1303758    4)     2208     128   __schedule+0x26e/0x5b93759    5)     2080      16   schedule+0x64/0x663760    6)     2064     128   schedule_timeout+0x34/0xe03761    7)     1936     112   wait_for_common+0x97/0xf13762    8)     1824      16   wait_for_completion+0x1d/0x1f3763    9)     1808     128   flush_work+0xfe/0x1193764   10)     1680      16   tty_flush_to_ldisc+0x1e/0x203765   11)     1664      48   input_available_p+0x1d/0x5c3766   12)     1616      48   n_tty_poll+0x6d/0x1343767   13)     1568      64   tty_poll+0x64/0x7f3768   14)     1504     880   do_select+0x31e/0x5113769   15)      624     400   core_sys_select+0x177/0x2163770   16)      224      96   sys_select+0x91/0xb93771   17)      128     128   system_call_fastpath+0x16/0x1b3772 3773Note, if -mfentry is being used by gcc, functions get traced before3774they set up the stack frame. This means that leaf level functions3775are not tested by the stack tracer when -mfentry is used.3776 3777Currently, -mfentry is used by gcc 4.6.0 and above on x86 only.3778 3779More3780----3781More details can be found in the source code, in the `kernel/trace/*.c` files.3782