1======================== 2ftrace - Function Tracer 3======================== 4 5Copyright 2008 Red Hat Inc. 6 7:Author: Steven Rostedt <srostedt@redhat.com> 8:License: The GNU Free Documentation License, Version 1.2 9 (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-rc2 14- Updated for: 3.10 15- Updated for: 4.13 - Copyright 2017 VMware Inc. Steven Rostedt 16- Converted to rst format - Changbin Du <changbin.du@intel.com> 17 18Introduction 19------------ 20 21Ftrace is an internal tracer designed to help out developers and 22designers of systems to find what is going on inside the kernel. 23It can be used for debugging or analyzing latencies and 24performance issues that take place outside of user-space. 25 26Although ftrace is typically considered the function tracer, it 27is really a framework of several assorted tracing utilities. 28There's latency tracing to examine what occurs between interrupts 29disabled and enabled, as well as for preemption and from a time 30a 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 are hundreds of static event points that 34can be enabled via the tracefs file system to see what is 35going on in certain parts of the kernel. 36 37See events.rst for more information. 38 39 40Implementation Details 41---------------------- 42 43See Documentation/trace/ftrace-design.rst for details for arch porters and such. 44 45 46The File System 47--------------- 48 49Ftrace uses the tracefs file system to hold the control files as 50well as the files to display output. 51 52When tracefs is configured into the kernel (which selecting any ftrace 53option will do) the directory /sys/kernel/tracing will be created. To mount 54this directory, you can add to your /etc/fstab file:: 55 56 tracefs /sys/kernel/tracing tracefs defaults 0 0 57 58Or you can mount it at run time with:: 59 60 mount -t tracefs nodev /sys/kernel/tracing 61 62For quicker access to that directory you may want to make a soft link to 63it:: 64 65 ln -s /sys/kernel/tracing /tracing 66 67.. attention:: 68 69 Before 4.1, all ftrace tracing control files were within the debugfs 70 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/tracing 75 76 All files located in the tracefs file system will be located in that 77 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 directory 83 (cd /sys/kernel/tracing) and will only concentrate on the files within that 84 directory and not distract from the content with the extended 85 "/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 files 90of 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 tracer 98 that is configured. Changing the current tracer clears 99 the ring buffer content as well as the "snapshot" buffer. 100 101 available_tracers: 102 103 This holds the different types of tracers that 104 have been compiled into the kernel. The 105 tracers listed here can be configured by 106 echoing their name into current_tracer. 107 108 tracing_on: 109 110 This sets or displays whether writing to the trace 111 ring buffer is enabled. Echo 0 into this file to disable 112 the tracer or 1 to enable it. Note, this only disables 113 writing to the ring buffer, the tracing overhead may 114 still be occurring. 115 116 The kernel function tracing_off() can be used within the 117 kernel to disable writing to the ring buffer, which will 118 set this file to "0". User space can re-enable tracing by 119 echoing "1" into the file. 120 121 Note, the function and event trigger "traceoff" will also 122 set this file to zero and stop tracing. Which can also 123 be re-enabled by user space using this file. 124 125 trace: 126 127 This file holds the output of the trace in a human 128 readable format (described below). Opening this file for 129 writing with the O_TRUNC flag clears the ring buffer content. 130 Note, this file is not a consumer. If tracing is off 131 (no tracer running, or tracing_on is zero), it will produce 132 the same output each time it is read. When tracing is on, 133 it may produce inconsistent results as it tries to read 134 the entire buffer without consuming it. 135 136 trace_pipe: 137 138 The output is the same as the "trace" file but this 139 file is meant to be streamed with live tracing. 140 Reads from this file will block until new data is 141 retrieved. Unlike the "trace" file, this file is a 142 consumer. This means reading from this file causes 143 sequential reads to display more current data. Once 144 data is read from this file, it is consumed, and 145 will not be read again with a sequential read. The 146 "trace" file is static, and if the tracer is not 147 adding more data, it will display the same 148 information every time it is read. 149 150 trace_options: 151 152 This file lets the user control the amount of data 153 that is displayed in one of the above output 154 files. Options also exist to modify how a tracer 155 or events work (stack traces, timestamps, etc). 156 157 options: 158 159 This is a directory that has a file for every available 160 trace option (also in trace_options). Options may also be set 161 or cleared by writing a "1" or "0" respectively into the 162 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 be 169 stored, and displayed by "trace". A new max trace will only be 170 recorded if the latency is greater than the value in this file 171 (in microseconds). 172 173 By echoing in a time into this file, no latency will be recorded 174 unless it is greater than the time in this file. 175 176 tracing_thresh: 177 178 Some latency tracers will record a trace whenever the 179 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 filled 186 before a waiter is woken up. That is, if an application calls a 187 blocking read syscall on one of the per_cpu trace_pipe_raw files, it 188 will block until the given amount of data specified by buffer_percent 189 is in the ring buffer before it wakes the reader up. This also 190 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-buffers 194 are full. 195 100 - means to block until the ring buffer is totally full and is 196 about to start overwriting the older data. 197 198 buffer_size_kb: 199 200 This sets or displays the number of kilobytes each CPU 201 buffer holds. By default, the trace buffers are the same size 202 for each CPU. The displayed number is the size of the 203 CPU buffer and not total size of all buffers. The 204 trace buffers are allocated in pages (blocks of memory 205 that the kernel uses for allocation, usually 4 KB in size). 206 A few extra pages may be allocated to accommodate buffer management 207 meta-data. If the last page allocated has room for more bytes 208 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 size 211 due to buffer management meta-data. ) 212 213 Buffer sizes for individual CPUs may vary 214 (see "per_cpu/cpu0/buffer_size_kb" below), and if they do 215 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 up 224 into several same size "sub buffers". An event can not be bigger than 225 the size of the sub buffer. Normally, the sub buffer is the size of the 226 architecture's page (4K on x86). The sub buffer also contains meta data 227 at the start which also limits the size of an event. That means when 228 the sub buffer is a page size, no event can be larger than the page 229 size minus the sub buffer meta data. 230 231 Note, the buffer_subbuf_size_kb is a way for the user to specify the 232 minimum size of the subbuffer. The kernel may make it bigger due to the 233 implementation details, or simply fail the operation if the kernel can 234 not handle the request. 235 236 Changing the sub buffer size allows for events to be larger than the 237 page size. 238 239 Note: When changing the sub-buffer size, tracing is stopped and any 240 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 be 245 shrunk "freed" when the process is finished, even if it were to be 246 killed by a signal, this file can be used for that purpose. On close 247 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 process 249 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 the 262 section below "dynamic ftrace"), the code is dynamically 263 modified (code text rewrite) to disable calling of the 264 function profiler (mcount). This lets tracing be configured 265 in with practically no overhead in performance. This also 266 has a side effect of enabling or disabling specific functions 267 to be traced. Echoing names of functions into this file 268 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 what 273 can be written into this file. 274 275 This interface also allows for commands to be used. See the 276 "Filter commands" section for more details. 277 278 As a speed up, since processing strings can be quite expensive 279 and requires a check of all functions registered to tracing, instead 280 an index can be written into this file. A number (starting with "1") 281 written will instead select the same corresponding at the line position 282 of the "available_filter_functions" file. 283 284 set_ftrace_notrace: 285 286 This has an effect opposite to that of 287 set_ftrace_filter. Any function that is added here will not 288 be traced. If a function exists in both set_ftrace_filter 289 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 are 294 listed in this file. 295 296 If the "function-fork" option is set, then when a task whose 297 PID is listed in this file forks, the child's PID will 298 automatically be added to this file, and the child will be 299 traced by the function tracer as well. This option will also 300 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 in 305 this file. 306 307 If the "function-fork" option is set, then when a task whose 308 PID is listed in this file forks, the child's PID will 309 automatically be added to this file, and the child will not be 310 traced by the function tracer as well. This option will also 311 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 this 314 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 events 320 listed in this file. 321 322 To have the PIDs of children of tasks with their PID in this file 323 added on fork, enable the "event-fork" option. That option will also 324 cause the PIDs of tasks to be removed from this file when the task 325 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 listed 331 in this file, even if a thread's PID is in the file if the 332 sched_switch or sched_wakeup events also trace a thread that should 333 be traced. 334 335 To have the PIDs of children of tasks with their PID in this file 336 added on fork, enable the "event-fork" option. That option will also 337 cause the PIDs of tasks to be removed from this file when the task 338 exits. 339 340 set_graph_function: 341 342 Functions listed in this file will cause the function graph 343 tracer to only trace these functions and the functions that 344 they call. (See the section "dynamic ftrace" for more details). 345 Note, set_ftrace_filter and set_ftrace_notrace still affects 346 what functions are being traced. 347 348 set_graph_notrace: 349 350 Similar to set_graph_function, but will disable function graph 351 tracing when the function is hit until it exits the function. 352 This makes it possible to ignore tracing functions that are called 353 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 to 359 "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 displayed 366 for each function. The displayed address is the patch-site address 367 and can differ from /proc/kallsyms address. 368 369 syscall_user_buf_size: 370 371 Some system call trace events will record the data from a user 372 space address that one of the parameters point to. The amount of 373 data per event is limited. This file holds the max number of bytes 374 that will be recorded into the ring buffer to hold this data. 375 The max value is currently 165. 376 377 dyn_ftrace_total_info: 378 379 This file is for debugging purposes. The number of functions that 380 have been converted to nops and are available to be traced. 381 382 enabled_functions: 383 384 This file is more for debugging ftrace, but can also be useful 385 in seeing if any function has a callback attached to it. 386 Not only does the trace infrastructure use ftrace function 387 trace utility, but other subsystems might too. This file 388 displays all functions that have a callback attached to them 389 as well as the number of callbacks that have been attached. 390 Note, a callback may also call multiple functions which will 391 not be listed in this count. 392 393 If the callback registered to be traced by a function with 394 the "save regs" attribute (thus even more overhead), an 'R' 395 will be displayed on the same line as the function that 396 is returning registers. 397 398 If the callback registered to be traced by a function with 399 the "ip modify" attribute (thus the regs->ip can be changed), 400 an 'I' will be displayed on the same line as the function that 401 can be overridden. 402 403 If a non-ftrace trampoline is attached (BPF) a 'D' will be displayed. 404 Note, normal ftrace trampolines can also be attached, but only one 405 "direct" trampoline can be attached to a given function at a time. 406 407 Some architectures can not call direct trampolines, but instead have 408 the ftrace ops function located above the function entry point. In 409 such cases an 'O' will be displayed. 410 411 If a function had either the "ip modify" or a "direct" call attached to 412 it in the past, a 'M' will be shown. This flag is never cleared. It is 413 used to know if a function was ever modified by the ftrace infrastructure, 414 and can be used for debugging. 415 416 If the architecture supports it, it will also show what callback 417 is being directly called by the function. If the count is greater 418 than 1 it most likely will be ftrace_ops_list_func(). 419 420 If the callback of a function jumps to a trampoline that is 421 specific to the callback and which is not the standard trampoline, 422 its address will be printed as well as the function that the 423 trampoline calls. 424 425 touched_functions: 426 427 This file contains all the functions that ever had a function callback 428 to it via the ftrace infrastructure. It has the same format as 429 enabled_functions but shows all functions that have ever been 430 traced. 431 432 To see any function that has every been modified by "ip modify" or a 433 direct trampoline, one can perform the following command: 434 435 grep ' M ' /sys/kernel/tracing/touched_functions 436 437 function_profile_enabled: 438 439 When set it will enable all functions with either the function 440 tracer, or if configured, the function graph tracer. It will 441 keep a histogram of the number of functions that were called 442 and if the function graph tracer was configured, it will also keep 443 track of the time spent in those functions. The histogram 444 content can be displayed in the files: 445 446 trace_stat/function<cpu> ( function0, function1, etc). 447 448 trace_stat: 449 450 A directory that holds different tracing stats. 451 452 kprobe_events: 453 454 Enable dynamic trace points. See kprobetrace.rst. 455 456 kprobe_profile: 457 458 Dynamic trace points stats. See kprobetrace.rst. 459 460 max_graph_depth: 461 462 Used with the function graph tracer. This is the max depth 463 it will trace into a function. Setting this to a value of 464 one will show only the first kernel function that is called 465 from user space. 466 467 printk_formats: 468 469 This is for tools that read the raw format files. If an event in 470 the ring buffer references a string, only a pointer to the string 471 is recorded into the buffer and not the string itself. This prevents 472 tools from knowing what that string was. This file displays the string 473 and address for the string allowing tools to map the pointers to what 474 the strings were. 475 476 saved_cmdlines: 477 478 Only the pid of the task is recorded in a trace event unless 479 the event specifically saves the task comm as well. Ftrace 480 makes a cache of pid mappings to comms to try to display 481 comms for events. If a pid for a comm is not listed, then 482 "<...>" is displayed in the output. 483 484 If the option "record-cmd" is set to "0", then comms of tasks 485 will not be saved during recording. By default, it is enabled. 486 487 saved_cmdlines_size: 488 489 By default, 128 comms are saved (see "saved_cmdlines" above). To 490 increase or decrease the amount of comms that are cached, echo 491 the number of comms to cache into this file. 492 493 saved_tgids: 494 495 If the option "record-tgid" is set, on each scheduling context switch 496 the Task Group ID of a task is saved in a table mapping the PID of 497 the thread to its TGID. By default, the "record-tgid" option is 498 disabled. 499 500 snapshot: 501 502 This displays the "snapshot" buffer and also lets the user 503 take a snapshot of the current running trace. 504 See the "Snapshot" section below for more details. 505 506 stack_max_size: 507 508 When the stack tracer is activated, this will display the 509 maximum stack size it has encountered. 510 See the "Stack Trace" section below. 511 512 stack_trace: 513 514 This displays the stack back trace of the largest stack 515 that was encountered when the stack tracer is activated. 516 See the "Stack Trace" section below. 517 518 stack_trace_filter: 519 520 This is similar to "set_ftrace_filter" but it limits what 521 functions the stack tracer will check. 522 523 trace_clock: 524 525 Whenever an event is recorded into the ring buffer, a 526 "timestamp" is added. This stamp comes from a specified 527 clock. By default, ftrace uses the "local" clock. This 528 clock is very fast and strictly per CPU, but on some 529 systems it may not be monotonic with respect to other 530 CPUs. In other words, the local clocks may not be in sync 531 with local clocks on other CPUs. 532 533 Usual clocks for tracing:: 534 535 # cat trace_clock 536 [local] global counter x86-tsc 537 538 The clock with the square brackets around it is the one in effect. 539 540 local: 541 Default clock, but may not be in sync across CPUs 542 543 global: 544 This clock is in sync with all CPUs but may 545 be a bit slower than the local clock. 546 547 counter: 548 This is not a clock at all, but literally an atomic 549 counter. It counts up one by one, but is in sync 550 with all CPUs. This is useful when you need to 551 know exactly the order events occurred with respect to 552 each other on different CPUs. 553 554 uptime: 555 This uses the jiffies counter and the time stamp 556 is relative to the time since boot up. 557 558 perf: 559 This makes ftrace use the same clock that perf uses. 560 Eventually perf will be able to read ftrace buffers 561 and this will help out in interleaving the data. 562 563 x86-tsc: 564 Architectures may define their own clocks. For 565 example, x86 uses its own TSC cycle clock here. 566 567 ppc-tb: 568 This uses the powerpc timebase register value. 569 This is in sync across CPUs and can also be used 570 to correlate events across hypervisor/guest if 571 tb_offset is known. 572 573 s390-tod: 574 This uses the s390 TOD clock value. This clock is usually in 575 sync across virtual machines and STP-enabled machines. 576 577 mono: 578 This uses the fast monotonic clock (CLOCK_MONOTONIC) 579 which is monotonic and is subject to NTP rate adjustments. 580 581 mono_raw: 582 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW) 583 which is monotonic but is not subject to any rate adjustments 584 and ticks at the same rate as the hardware clocksource. 585 586 boot: 587 This is the boot clock (CLOCK_BOOTTIME) and is based on the 588 fast monotonic clock, but also accounts for time spent in 589 suspend. Since the clock access is designed for use in 590 tracing in the suspend path, some side effects are possible 591 if clock is accessed after the suspend time is accounted before 592 the fast mono clock is updated. In this case, the clock update 593 appears to happen slightly sooner than it normally would have. 594 Also on 32-bit systems, it's possible that the 64-bit boot offset 595 sees a partial update. These effects are rare and post 596 processing should be able to handle them. See comments in the 597 ktime_get_boot_fast_ns() function for more information. 598 599 tai: 600 This is the tai clock (CLOCK_TAI) and is derived from the wall- 601 clock time. However, this clock does not experience 602 discontinuities and backwards jumps caused by NTP inserting leap 603 seconds. Since the clock access is designed for use in tracing, 604 side effects are possible. The clock access may yield wrong 605 readouts in case the internal TAI offset is updated e.g., caused 606 by setting the system time or using adjtimex() with an offset. 607 These effects are rare and post processing should be able to 608 handle them. See comments in the ktime_get_tai_fast_ns() 609 function for more information. 610 611 To set a clock, simply echo the clock name into this file:: 612 613 # echo global > trace_clock 614 615 Setting a clock clears the ring buffer content as well as the 616 "snapshot" buffer. 617 618 trace_marker: 619 620 This is a very useful file for synchronizing user space 621 with events happening in the kernel. Writing strings into 622 this file will be written into the ftrace buffer. 623 624 It is useful in applications to open this file at the start 625 of the application and just reference the file descriptor 626 for the file:: 627 628 void trace_write(const char *fmt, ...) 629 { 630 va_list ap; 631 char buf[256]; 632 int n; 633 634 if (trace_fd < 0) 635 return; 636 637 va_start(ap, fmt); 638 n = vsnprintf(buf, 256, fmt, ap); 639 va_end(ap); 640 641 write(trace_fd, buf, n); 642 } 643 644 start:: 645 646 trace_fd = open("trace_marker", O_WRONLY); 647 648 Note: Writing into the trace_marker file can also initiate triggers 649 that are written into /sys/kernel/tracing/events/ftrace/print/trigger 650 See "Event triggers" in Documentation/trace/events.rst and an 651 example in Documentation/trace/histogram.rst (Section 3.) 652 653 trace_marker_raw: 654 655 This is similar to trace_marker above, but is meant for binary data 656 to be written to it, where a tool can be used to parse the data 657 from trace_pipe_raw. 658 659 uprobe_events: 660 661 Add dynamic tracepoints in programs. 662 See uprobetracer.rst 663 664 uprobe_profile: 665 666 Uprobe statistics. See uprobetrace.txt 667 668 instances: 669 670 This is a way to make multiple trace buffers where different 671 events can be recorded in different buffers. 672 See "Instances" section below. 673 674 events: 675 676 This is the trace event directory. It holds event tracepoints 677 (also known as static tracepoints) that have been compiled 678 into the kernel. It shows what event tracepoints exist 679 and how they are grouped by system. There are "enable" 680 files at various levels that can enable the tracepoints 681 when a "1" is written to them. 682 683 See events.rst for more information. 684 685 set_event: 686 687 By echoing in the event into this file, will enable that event. 688 689 See events.rst for more information. 690 691 show_event_filters: 692 693 A list of events that have filters. This shows the 694 system/event pair along with the filter that is attached to 695 the event. 696 697 See events.rst for more information. 698 699 show_event_triggers: 700 701 A list of events that have triggers. This shows the 702 system/event pair along with the trigger that is attached to 703 the event. 704 705 See events.rst for more information. 706 707 available_events: 708 709 A list of events that can be enabled in tracing. 710 711 See events.rst for more information. 712 713 timestamp_mode: 714 715 Certain tracers may change the timestamp mode used when 716 logging trace events into the event buffer. Events with 717 different modes can coexist within a buffer but the mode in 718 effect when an event is logged determines which timestamp mode 719 is used for that event. The default timestamp mode is 720 'delta'. 721 722 Usual timestamp modes for tracing: 723 724 # cat timestamp_mode 725 [delta] absolute 726 727 The timestamp mode with the square brackets around it is the 728 one in effect. 729 730 delta: Default timestamp mode - timestamp is a delta against 731 a per-buffer timestamp. 732 733 absolute: The timestamp is a full timestamp, not a delta 734 against some other value. As such it takes up more 735 space and is less efficient. 736 737 hwlat_detector: 738 739 Directory for the Hardware Latency Detector. 740 See "Hardware Latency Detector" section below. 741 742 per_cpu: 743 744 This is a directory that contains the trace per_cpu information. 745 746 per_cpu/cpu0/buffer_size_kb: 747 748 The ftrace buffer is defined per_cpu. That is, there's a separate 749 buffer for each CPU to allow writes to be done atomically, 750 and free from cache bouncing. These buffers may have different 751 size buffers. This file is similar to the buffer_size_kb 752 file, but it only displays or sets the buffer size for the 753 specific CPU. (here cpu0). 754 755 per_cpu/cpu0/trace: 756 757 This is similar to the "trace" file, but it will only display 758 the data specific for the CPU. If written to, it only clears 759 the specific CPU buffer. 760 761 per_cpu/cpu0/trace_pipe 762 763 This is similar to the "trace_pipe" file, and is a consuming 764 read, but it will only display (and consume) the data specific 765 for the CPU. 766 767 per_cpu/cpu0/trace_pipe_raw 768 769 For tools that can parse the ftrace ring buffer binary format, 770 the trace_pipe_raw file can be used to extract the data 771 from the ring buffer directly. With the use of the splice() 772 system call, the buffer data can be quickly transferred to 773 a file or to the network where a server is collecting the 774 data. 775 776 Like trace_pipe, this is a consuming reader, where multiple 777 reads will always produce different data. 778 779 per_cpu/cpu0/snapshot: 780 781 This is similar to the main "snapshot" file, but will only 782 snapshot the current CPU (if supported). It only displays 783 the content of the snapshot for a given CPU, and if 784 written to, only clears this CPU buffer. 785 786 per_cpu/cpu0/snapshot_raw: 787 788 Similar to the trace_pipe_raw, but will read the binary format 789 from the snapshot buffer for the given CPU. 790 791 per_cpu/cpu0/stats: 792 793 This displays certain stats about the ring buffer: 794 795 entries: 796 The number of events that are still in the buffer. 797 798 overrun: 799 The number of lost events due to overwriting when 800 the buffer was full. 801 802 commit overrun: 803 Should always be zero. 804 This gets set if so many events happened within a nested 805 event (ring buffer is re-entrant), that it fills the 806 buffer and starts dropping events. 807 808 bytes: 809 Bytes actually read (not overwritten). 810 811 oldest event ts: 812 The oldest timestamp in the buffer 813 814 now ts: 815 The current timestamp 816 817 dropped events: 818 Events lost due to overwrite option being off. 819 820 read events: 821 The number of events read. 822 823The Tracers 824----------- 825 826Here is the list of current tracers that may be configured. 827 828 "function" 829 830 Function call tracer to trace all kernel functions. 831 832 "function_graph" 833 834 Similar to the function tracer except that the 835 function tracer probes the functions on their entry 836 whereas the function graph tracer traces on both entry 837 and exit of the functions. It then provides the ability 838 to draw a graph of function calls similar to C code 839 source. 840 841 Note that the function graph calculates the timings of when the 842 function starts and returns internally and for each instance. If 843 there are two instances that run function graph tracer and traces 844 the same functions, the length of the timings may be slightly off as 845 each read the timestamp separately and not at the same time. 846 847 "blk" 848 849 The block tracer. The tracer used by the blktrace user 850 application. 851 852 "hwlat" 853 854 The Hardware Latency tracer is used to detect if the hardware 855 produces any latency. See "Hardware Latency Detector" section 856 below. 857 858 "irqsoff" 859 860 Traces the areas that disable interrupts and saves 861 the trace with the longest max latency. 862 See tracing_max_latency. When a new max is recorded, 863 it replaces the old trace. It is best to view this 864 trace with the latency-format option enabled, which 865 happens automatically when the tracer is selected. 866 867 "preemptoff" 868 869 Similar to irqsoff but traces and records the amount of 870 time for which preemption is disabled. 871 872 "preemptirqsoff" 873 874 Similar to irqsoff and preemptoff, but traces and 875 records the largest time for which irqs and/or preemption 876 is disabled. 877 878 "wakeup" 879 880 Traces and records the max latency that it takes for 881 the highest priority task to get scheduled after 882 it has been woken up. 883 Traces all tasks as an average developer would expect. 884 885 "wakeup_rt" 886 887 Traces and records the max latency that it takes for just 888 RT tasks (as the current "wakeup" does). This is useful 889 for those interested in wake up timings of RT tasks. 890 891 "wakeup_dl" 892 893 Traces and records the max latency that it takes for 894 a SCHED_DEADLINE task to be woken (as the "wakeup" and 895 "wakeup_rt" does). 896 897 "mmiotrace" 898 899 A special tracer that is used to trace binary modules. 900 It will trace all the calls that a module makes to the 901 hardware. Everything it writes and reads from the I/O 902 as well. 903 904 "branch" 905 906 This tracer can be configured when tracing likely/unlikely 907 calls within the kernel. It will trace when a likely and 908 unlikely branch is hit and if it was correct in its prediction 909 of being correct. 910 911 "nop" 912 913 This is the "trace nothing" tracer. To remove all 914 tracers from tracing simply echo "nop" into 915 current_tracer. 916 917Error conditions 918---------------- 919 920 For most ftrace commands, failure modes are obvious and communicated 921 using standard return codes. 922 923 For other more involved commands, extended error information may be 924 available via the tracing/error_log file. For the commands that 925 support it, reading the tracing/error_log file after an error will 926 display more detailed information about what went wrong, if 927 information is available. The tracing/error_log file is a circular 928 error log displaying a small number (currently, 8) of ftrace errors 929 for the last (8) failed commands. 930 931 The extended error information and usage takes the form shown in 932 this example:: 933 934 # echo xxx > /sys/kernel/tracing/events/sched/sched_wakeup/trigger 935 echo: write error: Invalid argument 936 937 # cat /sys/kernel/tracing/error_log 938 [ 5348.887237] location: error: Couldn't yyy: zzz 939 Command: xxx 940 ^ 941 [ 7517.023364] location: error: Bad rrr: sss 942 Command: ppp qqq 943 ^ 944 945 To clear the error log, echo the empty string into it:: 946 947 # echo > /sys/kernel/tracing/error_log 948 949Examples of using the tracer 950---------------------------- 951 952Here are typical examples of using the tracers when controlling 953them only with the tracefs interface (without using any 954user-land utilities). 955 956Output format: 957-------------- 958 959Here is an example of the output format of the file "trace":: 960 961 # tracer: function 962 # 963 # entries-in-buffer/entries-written: 140080/250280 #P:4 964 # 965 # _-----=> irqs-off 966 # / _----=> need-resched 967 # | / _---=> hardirq/softirq 968 # || / _--=> preempt-depth 969 # ||| / delay 970 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 971 # | | | |||| | | 972 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath 973 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close 974 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd 975 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify 976 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock 977 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd 978 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock 979 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd 980 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close 981 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath 982 .... 983 984A header is printed with the tracer name that is represented by 985the trace. In this case the tracer is "function". Then it shows the 986number of events in the buffer as well as the total number of entries 987that were written. The difference is the number of entries that were 988lost due to the buffer filling up (250280 - 140080 = 110200 events 989lost). 990 991The header explains the content of the events. Task name "bash", the task 992PID "1977", the CPU that it was running on "000", the latency format 993(explained below), the timestamp in <secs>.<usecs> format, the 994function name that was traced "sys_close" and the parent function that 995called this function "system_call_fastpath". The timestamp is the time 996at which the function was entered. 997 998Latency trace format 999-------------------- 1000 1001When the latency-format option is enabled or when one of the latency 1002tracers is set, the trace file gives somewhat more information to see 1003why a latency happened. Here is a typical trace:: 1004 1005 # tracer: irqsoff 1006 # 1007 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1008 # -------------------------------------------------------------------- 1009 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1010 # ----------------- 1011 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0) 1012 # ----------------- 1013 # => started at: __lock_task_sighand 1014 # => ended at: _raw_spin_unlock_irqrestore 1015 # 1016 # 1017 # _------=> CPU# 1018 # / _-----=> irqs-off 1019 # | / _----=> need-resched 1020 # || / _---=> hardirq/softirq 1021 # ||| / _--=> preempt-depth 1022 # |||| / delay 1023 # cmd pid ||||| time | caller 1024 # \ / ||||| \ | / 1025 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand 1026 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore 1027 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore 1028 ps-6143 2d..1 306us : <stack trace> 1029 => trace_hardirqs_on_caller 1030 => trace_hardirqs_on 1031 => _raw_spin_unlock_irqrestore 1032 => do_task_stat 1033 => proc_tgid_stat 1034 => proc_single_show 1035 => seq_read 1036 => vfs_read 1037 => sys_read 1038 => system_call_fastpath 1039 1040 1041This shows that the current tracer is "irqsoff" tracing the time 1042for which interrupts were disabled. It gives the trace version (which 1043never changes) and the version of the kernel upon which this was executed on 1044(3.8). Then it displays the max latency in microseconds (259 us). The number 1045of trace entries displayed and the total number (both are four: #4/4). 1046VP, KP, SP, and HP are always zero and are reserved for later use. 1047#P is the number of online CPUs (#P:4). 1048 1049The task is the process that was running when the latency 1050occurred. (ps pid: 6143). 1051 1052The start and stop (the functions in which the interrupts were 1053disabled and enabled respectively) that caused the latencies: 1054 1055 - __lock_task_sighand is where the interrupts were disabled. 1056 - _raw_spin_unlock_irqrestore is where they were enabled again. 1057 1058The next lines after the header are the trace itself. The header 1059explains which is which. 1060 1061 cmd: The name of the process in the trace. 1062 1063 pid: The PID of that process. 1064 1065 CPU#: The CPU which the process was running on. 1066 1067 irqs-off: 'd' interrupts are disabled. '.' otherwise. 1068 1069 need-resched: 1070 - 'B' all, TIF_NEED_RESCHED, PREEMPT_NEED_RESCHED and TIF_RESCHED_LAZY is set, 1071 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set, 1072 - 'n' only TIF_NEED_RESCHED is set, 1073 - 'p' only PREEMPT_NEED_RESCHED is set, 1074 - 'L' both PREEMPT_NEED_RESCHED and TIF_RESCHED_LAZY is set, 1075 - 'b' both TIF_NEED_RESCHED and TIF_RESCHED_LAZY is set, 1076 - 'l' only TIF_RESCHED_LAZY is set 1077 - '.' otherwise. 1078 1079 hardirq/softirq: 1080 - 'Z' - NMI occurred inside a hardirq 1081 - 'z' - NMI is running 1082 - 'H' - hard irq occurred inside a softirq. 1083 - 'h' - hard irq is running 1084 - 's' - soft irq is running 1085 - '.' - normal context. 1086 1087 preempt-depth: The level of preempt_disabled 1088 1089The above is mostly meaningful for kernel developers. 1090 1091 time: 1092 When the latency-format option is enabled, the trace file 1093 output includes a timestamp relative to the start of the 1094 trace. This differs from the output when latency-format 1095 is disabled, which includes an absolute timestamp. 1096 1097 delay: 1098 This is just to help catch your eye a bit better. And 1099 needs to be fixed to be only relative to the same CPU. 1100 The marks are determined by the difference between this 1101 current trace and the next trace. 1102 1103 - '$' - greater than 1 second 1104 - '@' - greater than 100 millisecond 1105 - '*' - greater than 10 millisecond 1106 - '#' - greater than 1000 microsecond 1107 - '!' - greater than 100 microsecond 1108 - '+' - greater than 10 microsecond 1109 - ' ' - less than or equal to 10 microsecond. 1110 1111 The rest is the same as the 'trace' file. 1112 1113 Note, the latency tracers will usually end with a back trace 1114 to easily find where the latency occurred. 1115 1116trace_options 1117------------- 1118 1119The trace_options file (or the options directory) is used to control 1120what gets printed in the trace output, or manipulate the tracers. 1121To see what is available, simply cat the file:: 1122 1123 cat trace_options 1124 print-parent 1125 nosym-offset 1126 nosym-addr 1127 noverbose 1128 noraw 1129 nohex 1130 nobin 1131 noblock 1132 nofields 1133 trace_printk 1134 annotate 1135 nouserstacktrace 1136 nosym-userobj 1137 noprintk-msg-only 1138 context-info 1139 nolatency-format 1140 record-cmd 1141 norecord-tgid 1142 overwrite 1143 nodisable_on_free 1144 irq-info 1145 markers 1146 noevent-fork 1147 function-trace 1148 nofunction-fork 1149 nodisplay-graph 1150 nostacktrace 1151 nobranch 1152 1153To disable one of the options, echo in the option prepended with 1154"no":: 1155 1156 echo noprint-parent > trace_options 1157 1158To enable an option, leave off the "no":: 1159 1160 echo sym-offset > trace_options 1161 1162Here are the available options: 1163 1164 print-parent 1165 On function traces, display the calling (parent) 1166 function as well as the function being traced. 1167 :: 1168 1169 print-parent: 1170 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul 1171 1172 noprint-parent: 1173 bash-4000 [01] 1477.606694: simple_strtoul 1174 1175 1176 sym-offset 1177 Display not only the function name, but also the 1178 offset in the function. For example, instead of 1179 seeing just "ktime_get", you will see 1180 "ktime_get+0xb/0x20". 1181 :: 1182 1183 sym-offset: 1184 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0 1185 1186 sym-addr 1187 This will also display the function address as well 1188 as the function name. 1189 :: 1190 1191 sym-addr: 1192 bash-4000 [01] 1477.606694: simple_strtoul <c0339346> 1193 1194 verbose 1195 This deals with the trace file when the 1196 latency-format option is enabled. 1197 :: 1198 1199 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \ 1200 (+0.000ms): simple_strtoul (kstrtoul) 1201 1202 raw 1203 This will display raw numbers. This option is best for 1204 use with user applications that can translate the raw 1205 numbers better than having it done in the kernel. 1206 1207 hex 1208 Similar to raw, but the numbers will be in a hexadecimal format. 1209 1210 bin 1211 This will print out the formats in raw binary. 1212 1213 block 1214 When set, reading trace_pipe will not block when polled. 1215 1216 fields 1217 Print the fields as described by their types. This is a better 1218 option than using hex, bin or raw, as it gives a better parsing 1219 of the content of the event. 1220 1221 trace_printk 1222 Can disable trace_printk() from writing into the buffer. 1223 1224 trace_printk_dest 1225 Set to have trace_printk() and similar internal tracing functions 1226 write into this instance. Note, only one trace instance can have 1227 this set. By setting this flag, it clears the trace_printk_dest flag 1228 of the instance that had it set previously. By default, the top 1229 level trace has this set, and will get it set again if another 1230 instance has it set then clears it. 1231 1232 This flag cannot be cleared by the top level instance, as it is the 1233 default instance. The only way the top level instance has this flag 1234 cleared, is by it being set in another instance. 1235 1236 copy_trace_marker 1237 If there are applications that hard code writing into the top level 1238 trace_marker file (/sys/kernel/tracing/trace_marker or trace_marker_raw), 1239 and the tooling would like it to go into an instance, this option can 1240 be used. Create an instance and set this option, and then all writes 1241 into the top level trace_marker file will also be redirected into this 1242 instance. 1243 1244 Note, by default this option is set for the top level instance. If it 1245 is disabled, then writes to the trace_marker or trace_marker_raw files 1246 will not be written into the top level file. If no instance has this 1247 option set, then a write will error with the errno of ENODEV. 1248 1249 annotate 1250 It is sometimes confusing when the CPU buffers are full 1251 and one CPU buffer had a lot of events recently, thus 1252 a shorter time frame, were another CPU may have only had 1253 a few events, which lets it have older events. When 1254 the trace is reported, it shows the oldest events first, 1255 and it may look like only one CPU ran (the one with the 1256 oldest events). When the annotate option is set, it will 1257 display when a new CPU buffer started:: 1258 1259 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on 1260 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on 1261 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore 1262 ##### CPU 2 buffer started #### 1263 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle 1264 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog 1265 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock 1266 1267 userstacktrace 1268 This option changes the trace. It records a 1269 stacktrace of the current user space thread after 1270 each trace event. 1271 1272 sym-userobj 1273 when user stacktrace are enabled, look up which 1274 object the address belongs to, and print a 1275 relative address. This is especially useful when 1276 ASLR is on, otherwise you don't get a chance to 1277 resolve the address to object/file/line after 1278 the app is no longer running 1279 1280 The lookup is performed when you read 1281 trace,trace_pipe. Example:: 1282 1283 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0 1284 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6] 1285 1286 1287 printk-msg-only 1288 When set, trace_printk()s will only show the format 1289 and not their parameters (if trace_bprintk() or 1290 trace_bputs() was used to save the trace_printk()). 1291 1292 context-info 1293 Show only the event data. Hides the comm, PID, 1294 timestamp, CPU, and other useful data. 1295 1296 latency-format 1297 This option changes the trace output. When it is enabled, 1298 the trace displays additional information about the 1299 latency, as described in "Latency trace format". 1300 1301 pause-on-trace 1302 When set, opening the trace file for read, will pause 1303 writing to the ring buffer (as if tracing_on was set to zero). 1304 This simulates the original behavior of the trace file. 1305 When the file is closed, tracing will be enabled again. 1306 1307 hash-ptr 1308 When set, "%p" in the event printk format displays the 1309 hashed pointer value instead of real address. 1310 This will be useful if you want to find out which hashed 1311 value is corresponding to the real value in trace log. 1312 1313 bitmask-list 1314 When enabled, bitmasks are displayed as a human-readable list of 1315 ranges (e.g., 0,2-5,7) using the printk "%*pbl" format specifier. 1316 When disabled (the default), bitmasks are displayed in the 1317 traditional hexadecimal bitmap representation. The list format is 1318 particularly useful for tracing CPU masks and other large bitmasks 1319 where individual bit positions are more meaningful than their 1320 hexadecimal encoding. 1321 1322 record-cmd 1323 When any event or tracer is enabled, a hook is enabled 1324 in the sched_switch trace point to fill comm cache 1325 with mapped pids and comms. But this may cause some 1326 overhead, and if you only care about pids, and not the 1327 name of the task, disabling this option can lower the 1328 impact of tracing. See "saved_cmdlines". 1329 1330 record-tgid 1331 When any event or tracer is enabled, a hook is enabled 1332 in the sched_switch trace point to fill the cache of 1333 mapped Thread Group IDs (TGID) mapping to pids. See 1334 "saved_tgids". 1335 1336 overwrite 1337 This controls what happens when the trace buffer is 1338 full. If "1" (default), the oldest events are 1339 discarded and overwritten. If "0", then the newest 1340 events are discarded. 1341 (see per_cpu/cpu0/stats for overrun and dropped) 1342 1343 disable_on_free 1344 When the free_buffer is closed, tracing will 1345 stop (tracing_on set to 0). 1346 1347 irq-info 1348 Shows the interrupt, preempt count, need resched data. 1349 When disabled, the trace looks like:: 1350 1351 # tracer: function 1352 # 1353 # entries-in-buffer/entries-written: 144405/9452052 #P:4 1354 # 1355 # TASK-PID CPU# TIMESTAMP FUNCTION 1356 # | | | | | 1357 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up 1358 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89 1359 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task 1360 1361 1362 markers 1363 When set, the trace_marker is writable (only by root). 1364 When disabled, the trace_marker will error with EINVAL 1365 on write. 1366 1367 event-fork 1368 When set, tasks with PIDs listed in set_event_pid will have 1369 the PIDs of their children added to set_event_pid when those 1370 tasks fork. Also, when tasks with PIDs in set_event_pid exit, 1371 their PIDs will be removed from the file. 1372 1373 This affects PIDs listed in set_event_notrace_pid as well. 1374 1375 function-trace 1376 The latency tracers will enable function tracing 1377 if this option is enabled (default it is). When 1378 it is disabled, the latency tracers do not trace 1379 functions. This keeps the overhead of the tracer down 1380 when performing latency tests. 1381 1382 function-fork 1383 When set, tasks with PIDs listed in set_ftrace_pid will 1384 have the PIDs of their children added to set_ftrace_pid 1385 when those tasks fork. Also, when tasks with PIDs in 1386 set_ftrace_pid exit, their PIDs will be removed from the 1387 file. 1388 1389 This affects PIDs in set_ftrace_notrace_pid as well. 1390 1391 display-graph 1392 When set, the latency tracers (irqsoff, wakeup, etc) will 1393 use function graph tracing instead of function tracing. 1394 1395 stacktrace 1396 When set, a stack trace is recorded after any trace event 1397 is recorded. 1398 1399 branch 1400 Enable branch tracing with the tracer. This enables branch 1401 tracer along with the currently set tracer. Enabling this 1402 with the "nop" tracer is the same as just enabling the 1403 "branch" tracer. 1404 1405.. tip:: Some tracers have their own options. They only appear in this 1406 file when the tracer is active. They always appear in the 1407 options directory. 1408 1409 1410Here are the per tracer options: 1411 1412Options for function tracer: 1413 1414 func_stack_trace 1415 When set, a stack trace is recorded after every 1416 function that is recorded. NOTE! Limit the functions 1417 that are recorded before enabling this, with 1418 "set_ftrace_filter" otherwise the system performance 1419 will be critically degraded. Remember to disable 1420 this option before clearing the function filter. 1421 1422Options for function_graph tracer: 1423 1424 Since the function_graph tracer has a slightly different output 1425 it has its own options to control what is displayed. 1426 1427 funcgraph-overrun 1428 When set, the "overrun" of the graph stack is 1429 displayed after each function traced. The 1430 overrun, is when the stack depth of the calls 1431 is greater than what is reserved for each task. 1432 Each task has a fixed array of functions to 1433 trace in the call graph. If the depth of the 1434 calls exceeds that, the function is not traced. 1435 The overrun is the number of functions missed 1436 due to exceeding this array. 1437 1438 funcgraph-cpu 1439 When set, the CPU number of the CPU where the trace 1440 occurred is displayed. 1441 1442 funcgraph-overhead 1443 When set, if the function takes longer than 1444 A certain amount, then a delay marker is 1445 displayed. See "delay" above, under the 1446 header description. 1447 1448 funcgraph-proc 1449 Unlike other tracers, the process' command line 1450 is not displayed by default, but instead only 1451 when a task is traced in and out during a context 1452 switch. Enabling this options has the command 1453 of each process displayed at every line. 1454 1455 funcgraph-duration 1456 At the end of each function (the return) 1457 the duration of the amount of time in the 1458 function is displayed in microseconds. 1459 1460 funcgraph-abstime 1461 When set, the timestamp is displayed at each line. 1462 1463 funcgraph-irqs 1464 When disabled, functions that happen inside an 1465 interrupt will not be traced. 1466 1467 funcgraph-tail 1468 When set, the return event will include the function 1469 that it represents. By default this is off, and 1470 only a closing curly bracket "}" is displayed for 1471 the return of a function. 1472 1473 funcgraph-retval 1474 When set, the return value of each traced function 1475 will be printed after an equal sign "=". By default 1476 this is off. 1477 1478 funcgraph-retval-hex 1479 When set, the return value will always be printed 1480 in hexadecimal format. If the option is not set and 1481 the return value is an error code, it will be printed 1482 in signed decimal format; otherwise it will also be 1483 printed in hexadecimal format. By default, this option 1484 is off. 1485 1486 sleep-time 1487 When running function graph tracer, to include 1488 the time a task schedules out in its function. 1489 When enabled, it will account time the task has been 1490 scheduled out as part of the function call. 1491 1492 graph-time 1493 When running function profiler with function graph tracer, 1494 to include the time to call nested functions. When this is 1495 not set, the time reported for the function will only 1496 include the time the function itself executed for, not the 1497 time for functions that it called. 1498 1499Options for blk tracer: 1500 1501 blk_classic 1502 Shows a more minimalistic output. 1503 1504 1505irqsoff 1506------- 1507 1508When interrupts are disabled, the CPU can not react to any other 1509external event (besides NMIs and SMIs). This prevents the timer 1510interrupt from triggering or the mouse interrupt from letting 1511the kernel know of a new mouse event. The result is a latency 1512with the reaction time. 1513 1514The irqsoff tracer tracks the time for which interrupts are 1515disabled. When a new maximum latency is hit, the tracer saves 1516the trace leading up to that latency point so that every time a 1517new maximum is reached, the old saved trace is discarded and the 1518new trace is saved. 1519 1520To reset the maximum, echo 0 into tracing_max_latency. Here is 1521an example:: 1522 1523 # echo 0 > options/function-trace 1524 # echo irqsoff > current_tracer 1525 # echo 1 > tracing_on 1526 # echo 0 > tracing_max_latency 1527 # ls -ltr 1528 [...] 1529 # echo 0 > tracing_on 1530 # cat trace 1531 # tracer: irqsoff 1532 # 1533 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1534 # -------------------------------------------------------------------- 1535 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1536 # ----------------- 1537 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0) 1538 # ----------------- 1539 # => started at: run_timer_softirq 1540 # => ended at: run_timer_softirq 1541 # 1542 # 1543 # _------=> CPU# 1544 # / _-----=> irqs-off 1545 # | / _----=> need-resched 1546 # || / _---=> hardirq/softirq 1547 # ||| / _--=> preempt-depth 1548 # |||| / delay 1549 # cmd pid ||||| time | caller 1550 # \ / ||||| \ | / 1551 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq 1552 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq 1553 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq 1554 <idle>-0 0dNs3 25us : <stack trace> 1555 => _raw_spin_unlock_irq 1556 => run_timer_softirq 1557 => __do_softirq 1558 => call_softirq 1559 => do_softirq 1560 => irq_exit 1561 => smp_apic_timer_interrupt 1562 => apic_timer_interrupt 1563 => rcu_idle_exit 1564 => cpu_idle 1565 => rest_init 1566 => start_kernel 1567 => x86_64_start_reservations 1568 => x86_64_start_kernel 1569 1570Here we see that we had a latency of 16 microseconds (which is 1571very good). The _raw_spin_lock_irq in run_timer_softirq disabled 1572interrupts. The difference between the 16 and the displayed 1573timestamp 25us occurred because the clock was incremented 1574between the time of recording the max latency and the time of 1575recording the function that had that latency. 1576 1577Note the above example had function-trace not set. If we set 1578function-trace, we get a much larger output:: 1579 1580 with echo 1 > options/function-trace 1581 1582 # tracer: irqsoff 1583 # 1584 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1585 # -------------------------------------------------------------------- 1586 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1587 # ----------------- 1588 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0) 1589 # ----------------- 1590 # => started at: ata_scsi_queuecmd 1591 # => ended at: ata_scsi_queuecmd 1592 # 1593 # 1594 # _------=> CPU# 1595 # / _-----=> irqs-off 1596 # | / _----=> need-resched 1597 # || / _---=> hardirq/softirq 1598 # ||| / _--=> preempt-depth 1599 # |||| / delay 1600 # cmd pid ||||| time | caller 1601 # \ / ||||| \ | / 1602 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd 1603 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave 1604 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd 1605 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev 1606 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev 1607 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd 1608 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd 1609 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd 1610 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat 1611 [...] 1612 bash-2042 3d..1 67us : delay_tsc <-__delay 1613 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc 1614 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc 1615 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc 1616 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc 1617 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue 1618 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1619 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1620 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd 1621 bash-2042 3d..1 120us : <stack trace> 1622 => _raw_spin_unlock_irqrestore 1623 => ata_scsi_queuecmd 1624 => scsi_dispatch_cmd 1625 => scsi_request_fn 1626 => __blk_run_queue_uncond 1627 => __blk_run_queue 1628 => blk_queue_bio 1629 => submit_bio_noacct 1630 => submit_bio 1631 => bh_submit 1632 => __ext3_get_inode_loc 1633 => ext3_iget 1634 => ext3_lookup 1635 => lookup_real 1636 => __lookup_hash 1637 => walk_component 1638 => lookup_last 1639 => path_lookupat 1640 => filename_lookup 1641 => user_path_at_empty 1642 => user_path_at 1643 => vfs_fstatat 1644 => vfs_stat 1645 => sys_newstat 1646 => system_call_fastpath 1647 1648 1649Here we traced a 71 microsecond latency. But we also see all the 1650functions that were called during that time. Note that by 1651enabling function tracing, we incur an added overhead. This 1652overhead may extend the latency times. But nevertheless, this 1653trace has provided some very helpful debugging information. 1654 1655If we prefer function graph output instead of function, we can set 1656display-graph option:: 1657 1658 with echo 1 > options/display-graph 1659 1660 # tracer: irqsoff 1661 # 1662 # irqsoff latency trace v1.1.5 on 4.20.0-rc6+ 1663 # -------------------------------------------------------------------- 1664 # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4) 1665 # ----------------- 1666 # | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0) 1667 # ----------------- 1668 # => started at: free_debug_processing 1669 # => ended at: return_to_handler 1670 # 1671 # 1672 # _-----=> irqs-off 1673 # / _----=> need-resched 1674 # | / _---=> hardirq/softirq 1675 # || / _--=> preempt-depth 1676 # ||| / 1677 # REL TIME CPU TASK/PID |||| DURATION FUNCTION CALLS 1678 # | | | | |||| | | | | | | 1679 0 us | 0) bash-1507 | d... | 0.000 us | _raw_spin_lock_irqsave(); 1680 0 us | 0) bash-1507 | d..1 | 0.378 us | do_raw_spin_trylock(); 1681 1 us | 0) bash-1507 | d..2 | | set_track() { 1682 2 us | 0) bash-1507 | d..2 | | save_stack_trace() { 1683 2 us | 0) bash-1507 | d..2 | | __save_stack_trace() { 1684 3 us | 0) bash-1507 | d..2 | | __unwind_start() { 1685 3 us | 0) bash-1507 | d..2 | | get_stack_info() { 1686 3 us | 0) bash-1507 | d..2 | 0.351 us | in_task_stack(); 1687 4 us | 0) bash-1507 | d..2 | 1.107 us | } 1688 [...] 1689 3750 us | 0) bash-1507 | d..1 | 0.516 us | do_raw_spin_unlock(); 1690 3750 us | 0) bash-1507 | d..1 | 0.000 us | _raw_spin_unlock_irqrestore(); 1691 3764 us | 0) bash-1507 | d..1 | 0.000 us | tracer_hardirqs_on(); 1692 bash-1507 0d..1 3792us : <stack trace> 1693 => free_debug_processing 1694 => __slab_free 1695 => kmem_cache_free 1696 => vm_area_free 1697 => remove_vma 1698 => exit_mmap 1699 => mmput 1700 => begin_new_exec 1701 => load_elf_binary 1702 => search_binary_handler 1703 => __do_execve_file.isra.32 1704 => __x64_sys_execve 1705 => do_syscall_64 1706 => entry_SYSCALL_64_after_hwframe 1707 1708preemptoff 1709---------- 1710 1711When preemption is disabled, we may be able to receive 1712interrupts but the task cannot be preempted and a higher 1713priority task must wait for preemption to be enabled again 1714before it can preempt a lower priority task. 1715 1716The preemptoff tracer traces the places that disable preemption. 1717Like the irqsoff tracer, it records the maximum latency for 1718which preemption was disabled. The control of preemptoff tracer 1719is much like the irqsoff tracer. 1720:: 1721 1722 # echo 0 > options/function-trace 1723 # echo preemptoff > current_tracer 1724 # echo 1 > tracing_on 1725 # echo 0 > tracing_max_latency 1726 # ls -ltr 1727 [...] 1728 # echo 0 > tracing_on 1729 # cat trace 1730 # tracer: preemptoff 1731 # 1732 # preemptoff latency trace v1.1.5 on 3.8.0-test+ 1733 # -------------------------------------------------------------------- 1734 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1735 # ----------------- 1736 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0) 1737 # ----------------- 1738 # => started at: do_IRQ 1739 # => ended at: do_IRQ 1740 # 1741 # 1742 # _------=> CPU# 1743 # / _-----=> irqs-off 1744 # | / _----=> need-resched 1745 # || / _---=> hardirq/softirq 1746 # ||| / _--=> preempt-depth 1747 # |||| / delay 1748 # cmd pid ||||| time | caller 1749 # \ / ||||| \ | / 1750 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ 1751 sshd-1991 1d..1 46us : irq_exit <-do_IRQ 1752 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ 1753 sshd-1991 1d..1 52us : <stack trace> 1754 => sub_preempt_count 1755 => irq_exit 1756 => do_IRQ 1757 => ret_from_intr 1758 1759 1760This has some more changes. Preemption was disabled when an 1761interrupt came in (notice the 'h'), and was enabled on exit. 1762But we also see that interrupts have been disabled when entering 1763the preempt off section and leaving it (the 'd'). We do not know if 1764interrupts were enabled in the mean time or shortly after this 1765was over. 1766:: 1767 1768 # tracer: preemptoff 1769 # 1770 # preemptoff latency trace v1.1.5 on 3.8.0-test+ 1771 # -------------------------------------------------------------------- 1772 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1773 # ----------------- 1774 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0) 1775 # ----------------- 1776 # => started at: wake_up_new_task 1777 # => ended at: task_rq_unlock 1778 # 1779 # 1780 # _------=> CPU# 1781 # / _-----=> irqs-off 1782 # | / _----=> need-resched 1783 # || / _---=> hardirq/softirq 1784 # ||| / _--=> preempt-depth 1785 # |||| / delay 1786 # cmd pid ||||| time | caller 1787 # \ / ||||| \ | / 1788 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task 1789 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq 1790 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair 1791 bash-1994 1d..1 1us : source_load <-select_task_rq_fair 1792 bash-1994 1d..1 1us : source_load <-select_task_rq_fair 1793 [...] 1794 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt 1795 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter 1796 bash-1994 1d..1 13us : add_preempt_count <-irq_enter 1797 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt 1798 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt 1799 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt 1800 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock 1801 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt 1802 [...] 1803 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event 1804 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt 1805 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit 1806 bash-1994 1d..2 36us : do_softirq <-irq_exit 1807 bash-1994 1d..2 36us : __do_softirq <-call_softirq 1808 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq 1809 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq 1810 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq 1811 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock 1812 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq 1813 [...] 1814 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks 1815 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq 1816 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable 1817 bash-1994 1dN.2 82us : idle_cpu <-irq_exit 1818 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit 1819 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit 1820 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock 1821 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock 1822 bash-1994 1.N.1 104us : <stack trace> 1823 => sub_preempt_count 1824 => _raw_spin_unlock_irqrestore 1825 => task_rq_unlock 1826 => wake_up_new_task 1827 => do_fork 1828 => sys_clone 1829 => stub_clone 1830 1831 1832The above is an example of the preemptoff trace with 1833function-trace set. Here we see that interrupts were not disabled 1834the entire time. The irq_enter code lets us know that we entered 1835an interrupt 'h'. Before that, the functions being traced still 1836show that it is not in an interrupt, but we can see from the 1837functions themselves that this is not the case. 1838 1839preemptirqsoff 1840-------------- 1841 1842Knowing the locations that have interrupts disabled or 1843preemption disabled for the longest times is helpful. But 1844sometimes we would like to know when either preemption and/or 1845interrupts are disabled. 1846 1847Consider the following code:: 1848 1849 local_irq_disable(); 1850 call_function_with_irqs_off(); 1851 preempt_disable(); 1852 call_function_with_irqs_and_preemption_off(); 1853 local_irq_enable(); 1854 call_function_with_preemption_off(); 1855 preempt_enable(); 1856 1857The irqsoff tracer will record the total length of 1858call_function_with_irqs_off() and 1859call_function_with_irqs_and_preemption_off(). 1860 1861The preemptoff tracer will record the total length of 1862call_function_with_irqs_and_preemption_off() and 1863call_function_with_preemption_off(). 1864 1865But neither will trace the time that interrupts and/or 1866preemption is disabled. This total time is the time that we can 1867not schedule. To record this time, use the preemptirqsoff 1868tracer. 1869 1870Again, using this trace is much like the irqsoff and preemptoff 1871tracers. 1872:: 1873 1874 # echo 0 > options/function-trace 1875 # echo preemptirqsoff > current_tracer 1876 # echo 1 > tracing_on 1877 # echo 0 > tracing_max_latency 1878 # ls -ltr 1879 [...] 1880 # echo 0 > tracing_on 1881 # cat trace 1882 # tracer: preemptirqsoff 1883 # 1884 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+ 1885 # -------------------------------------------------------------------- 1886 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1887 # ----------------- 1888 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0) 1889 # ----------------- 1890 # => started at: ata_scsi_queuecmd 1891 # => ended at: ata_scsi_queuecmd 1892 # 1893 # 1894 # _------=> CPU# 1895 # / _-----=> irqs-off 1896 # | / _----=> need-resched 1897 # || / _---=> hardirq/softirq 1898 # ||| / _--=> preempt-depth 1899 # |||| / delay 1900 # cmd pid ||||| time | caller 1901 # \ / ||||| \ | / 1902 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd 1903 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1904 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd 1905 ls-2230 3...1 111us : <stack trace> 1906 => sub_preempt_count 1907 => _raw_spin_unlock_irqrestore 1908 => ata_scsi_queuecmd 1909 => scsi_dispatch_cmd 1910 => scsi_request_fn 1911 => __blk_run_queue_uncond 1912 => __blk_run_queue 1913 => blk_queue_bio 1914 => submit_bio_noacct 1915 => submit_bio 1916 => bh_submit 1917 => ext3_bread 1918 => ext3_dir_bread 1919 => htree_dirblock_to_tree 1920 => ext3_htree_fill_tree 1921 => ext3_readdir 1922 => vfs_readdir 1923 => sys_getdents 1924 => system_call_fastpath 1925 1926 1927The trace_hardirqs_off_thunk is called from assembly on x86 when 1928interrupts are disabled in the assembly code. Without the 1929function tracing, we do not know if interrupts were enabled 1930within the preemption points. We do see that it started with 1931preemption enabled. 1932 1933Here is a trace with function-trace set:: 1934 1935 # tracer: preemptirqsoff 1936 # 1937 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+ 1938 # -------------------------------------------------------------------- 1939 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1940 # ----------------- 1941 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0) 1942 # ----------------- 1943 # => started at: schedule 1944 # => ended at: mutex_unlock 1945 # 1946 # 1947 # _------=> CPU# 1948 # / _-----=> irqs-off 1949 # | / _----=> need-resched 1950 # || / _---=> hardirq/softirq 1951 # ||| / _--=> preempt-depth 1952 # |||| / delay 1953 # cmd pid ||||| time | caller 1954 # \ / ||||| \ | / 1955 kworker/-59 3...1 0us : __schedule <-schedule 1956 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch 1957 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq 1958 kworker/-59 3d..2 1us : deactivate_task <-__schedule 1959 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task 1960 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task 1961 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task 1962 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair 1963 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr 1964 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr 1965 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge 1966 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge 1967 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair 1968 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair 1969 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair 1970 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair 1971 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair 1972 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair 1973 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule 1974 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping 1975 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule 1976 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task 1977 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair 1978 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair 1979 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity 1980 ls-2269 3d..2 7us : finish_task_switch <-__schedule 1981 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch 1982 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr 1983 ls-2269 3d..2 8us : irq_enter <-do_IRQ 1984 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter 1985 ls-2269 3d..2 9us : add_preempt_count <-irq_enter 1986 ls-2269 3d.h2 9us : exit_idle <-do_IRQ 1987 [...] 1988 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock 1989 ls-2269 3d.h2 20us : irq_exit <-do_IRQ 1990 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit 1991 ls-2269 3d..3 21us : do_softirq <-irq_exit 1992 ls-2269 3d..3 21us : __do_softirq <-call_softirq 1993 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq 1994 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip 1995 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip 1996 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr 1997 ls-2269 3d.s5 31us : irq_enter <-do_IRQ 1998 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter 1999 [...] 2000 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter 2001 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter 2002 ls-2269 3d.H5 32us : exit_idle <-do_IRQ 2003 ls-2269 3d.H5 32us : handle_irq <-do_IRQ 2004 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq 2005 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq 2006 [...] 2007 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll 2008 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action 2009 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq 2010 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable 2011 ls-2269 3d..3 159us : idle_cpu <-irq_exit 2012 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit 2013 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit 2014 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock 2015 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock 2016 ls-2269 3d... 186us : <stack trace> 2017 => __mutex_unlock_slowpath 2018 => mutex_unlock 2019 => process_output 2020 => n_tty_write 2021 => tty_write 2022 => vfs_write 2023 => sys_write 2024 => system_call_fastpath 2025 2026This is an interesting trace. It started with kworker running and 2027scheduling out and ls taking over. But as soon as ls released the 2028rq lock and enabled interrupts (but not preemption) an interrupt 2029triggered. When the interrupt finished, it started running softirqs. 2030But while the softirq was running, another interrupt triggered. 2031When an interrupt is running inside a softirq, the annotation is 'H'. 2032 2033 2034wakeup 2035------ 2036 2037One common case that people are interested in tracing is the 2038time it takes for a task that is woken to actually wake up. 2039Now for non Real-Time tasks, this can be arbitrary. But tracing 2040it nonetheless can be interesting. 2041 2042Without function tracing:: 2043 2044 # echo 0 > options/function-trace 2045 # echo wakeup > current_tracer 2046 # echo 1 > tracing_on 2047 # echo 0 > tracing_max_latency 2048 # chrt -f 5 sleep 1 2049 # echo 0 > tracing_on 2050 # cat trace 2051 # tracer: wakeup 2052 # 2053 # wakeup latency trace v1.1.5 on 3.8.0-test+ 2054 # -------------------------------------------------------------------- 2055 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2056 # ----------------- 2057 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0) 2058 # ----------------- 2059 # 2060 # _------=> CPU# 2061 # / _-----=> irqs-off 2062 # | / _----=> need-resched 2063 # || / _---=> hardirq/softirq 2064 # ||| / _--=> preempt-depth 2065 # |||| / delay 2066 # cmd pid ||||| time | caller 2067 # \ / ||||| \ | / 2068 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H 2069 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up 2070 <idle>-0 3d..3 15us : __schedule <-schedule 2071 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H 2072 2073The tracer only traces the highest priority task in the system 2074to avoid tracing the normal circumstances. Here we see that 2075the kworker with a nice priority of -20 (not very nice), took 2076just 15 microseconds from the time it woke up, to the time it 2077ran. 2078 2079Non Real-Time tasks are not that interesting. A more interesting 2080trace is to concentrate only on Real-Time tasks. 2081 2082wakeup_rt 2083--------- 2084 2085In a Real-Time environment it is very important to know the 2086wakeup time it takes for the highest priority task that is woken 2087up to the time that it executes. This is also known as "schedule 2088latency". I stress the point that this is about RT tasks. It is 2089also important to know the scheduling latency of non-RT tasks, 2090but the average schedule latency is better for non-RT tasks. 2091Tools like LatencyTop are more appropriate for such 2092measurements. 2093 2094Real-Time environments are interested in the worst case latency. 2095That is the longest latency it takes for something to happen, 2096and not the average. We can have a very fast scheduler that may 2097only have a large latency once in a while, but that would not 2098work well with Real-Time tasks. The wakeup_rt tracer was designed 2099to record the worst case wakeups of RT tasks. Non-RT tasks are 2100not recorded because the tracer only records one worst case and 2101tracing non-RT tasks that are unpredictable will overwrite the 2102worst case latency of RT tasks (just run the normal wakeup 2103tracer for a while to see that effect). 2104 2105Since this tracer only deals with RT tasks, we will run this 2106slightly differently than we did with the previous tracers. 2107Instead of performing an 'ls', we will run 'sleep 1' under 2108'chrt' which changes the priority of the task. 2109:: 2110 2111 # echo 0 > options/function-trace 2112 # echo wakeup_rt > current_tracer 2113 # echo 1 > tracing_on 2114 # echo 0 > tracing_max_latency 2115 # chrt -f 5 sleep 1 2116 # echo 0 > tracing_on 2117 # cat trace 2118 # tracer: wakeup 2119 # 2120 # tracer: wakeup_rt 2121 # 2122 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2123 # -------------------------------------------------------------------- 2124 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2125 # ----------------- 2126 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5) 2127 # ----------------- 2128 # 2129 # _------=> CPU# 2130 # / _-----=> irqs-off 2131 # | / _----=> need-resched 2132 # || / _---=> hardirq/softirq 2133 # ||| / _--=> preempt-depth 2134 # |||| / delay 2135 # cmd pid ||||| time | caller 2136 # \ / ||||| \ | / 2137 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep 2138 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up 2139 <idle>-0 3d..3 5us : __schedule <-schedule 2140 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep 2141 2142 2143Running this on an idle system, we see that it only took 5 microseconds 2144to perform the task switch. Note, since the trace point in the schedule 2145is before the actual "switch", we stop the tracing when the recorded task 2146is about to schedule in. This may change if we add a new marker at the 2147end of the scheduler. 2148 2149Notice that the recorded task is 'sleep' with the PID of 2389 2150and it has an rt_prio of 5. This priority is user-space priority 2151and not the internal kernel priority. The policy is 1 for 2152SCHED_FIFO and 2 for SCHED_RR. 2153 2154Note, that the trace data shows the internal priority (99 - rtprio). 2155:: 2156 2157 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep 2158 2159The 0:120:R means idle was running with a nice priority of 0 (120 - 120) 2160and in the running state 'R'. The sleep task was scheduled in with 21612389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94) 2162and it too is in the running state. 2163 2164Doing the same with chrt -r 5 and function-trace set. 2165:: 2166 2167 echo 1 > options/function-trace 2168 2169 # tracer: wakeup_rt 2170 # 2171 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2172 # -------------------------------------------------------------------- 2173 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2174 # ----------------- 2175 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5) 2176 # ----------------- 2177 # 2178 # _------=> CPU# 2179 # / _-----=> irqs-off 2180 # | / _----=> need-resched 2181 # || / _---=> hardirq/softirq 2182 # ||| / _--=> preempt-depth 2183 # |||| / delay 2184 # cmd pid ||||| time | caller 2185 # \ / ||||| \ | / 2186 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep 2187 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up 2188 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup 2189 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr 2190 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup 2191 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up 2192 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock 2193 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up 2194 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up 2195 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2196 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer 2197 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock 2198 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt 2199 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock 2200 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt 2201 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event 2202 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event 2203 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event 2204 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt 2205 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit 2206 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit 2207 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit 2208 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit 2209 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit 2210 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle 2211 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit 2212 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle 2213 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit 2214 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit 2215 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit 2216 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit 2217 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz 2218 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock 2219 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz 2220 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update 2221 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz 2222 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock 2223 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit 2224 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit 2225 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit 2226 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel 2227 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel 2228 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18 2229 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave 2230 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16 2231 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer 2232 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram 2233 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event 2234 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event 2235 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event 2236 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel 2237 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2238 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit 2239 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward 2240 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward 2241 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11 2242 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns 2243 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns 2244 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18 2245 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave 2246 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns 2247 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns 2248 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns 2249 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event 2250 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event 2251 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event 2252 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns 2253 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2254 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit 2255 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks 2256 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle 2257 <idle>-0 3.N.. 25us : schedule <-cpu_idle 2258 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule 2259 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule 2260 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule 2261 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch 2262 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch 2263 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule 2264 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq 2265 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule 2266 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task 2267 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task 2268 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt 2269 <idle>-0 3d..3 29us : __schedule <-preempt_schedule 2270 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep 2271 2272This isn't that big of a trace, even with function tracing enabled, 2273so I included the entire trace. 2274 2275The interrupt went off while when the system was idle. Somewhere 2276before task_woken_rt() was called, the NEED_RESCHED flag was set, 2277this is indicated by the first occurrence of the 'N' flag. 2278 2279Latency tracing and events 2280-------------------------- 2281As function tracing can induce a much larger latency, but without 2282seeing what happens within the latency it is hard to know what 2283caused it. There is a middle ground, and that is with enabling 2284events. 2285:: 2286 2287 # echo 0 > options/function-trace 2288 # echo wakeup_rt > current_tracer 2289 # echo 1 > events/enable 2290 # echo 1 > tracing_on 2291 # echo 0 > tracing_max_latency 2292 # chrt -f 5 sleep 1 2293 # echo 0 > tracing_on 2294 # cat trace 2295 # tracer: wakeup_rt 2296 # 2297 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2298 # -------------------------------------------------------------------- 2299 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2300 # ----------------- 2301 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5) 2302 # ----------------- 2303 # 2304 # _------=> CPU# 2305 # / _-----=> irqs-off 2306 # | / _----=> need-resched 2307 # || / _---=> hardirq/softirq 2308 # ||| / _--=> preempt-depth 2309 # |||| / delay 2310 # cmd pid ||||| time | caller 2311 # \ / ||||| \ | / 2312 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep 2313 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up 2314 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002 2315 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8 2316 <idle>-0 2.N.2 2us : power_end: cpu_id=2 2317 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2 2318 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0 2319 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000 2320 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch 2321 <idle>-0 2.N.2 5us : rcu_utilization: End context switch 2322 <idle>-0 2d..3 6us : __schedule <-schedule 2323 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep 2324 2325 2326Hardware Latency Detector 2327------------------------- 2328 2329The hardware latency detector is executed by enabling the "hwlat" tracer. 2330 2331NOTE, this tracer will affect the performance of the system as it will 2332periodically make a CPU constantly busy with interrupts disabled. 2333:: 2334 2335 # echo hwlat > current_tracer 2336 # sleep 100 2337 # cat trace 2338 # tracer: hwlat 2339 # 2340 # entries-in-buffer/entries-written: 13/13 #P:8 2341 # 2342 # _-----=> irqs-off 2343 # / _----=> need-resched 2344 # | / _---=> hardirq/softirq 2345 # || / _--=> preempt-depth 2346 # ||| / delay 2347 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2348 # | | | |||| | | 2349 <...>-1729 [001] d... 678.473449: #1 inner/outer(us): 11/12 ts:1581527483.343962693 count:6 2350 <...>-1729 [004] d... 689.556542: #2 inner/outer(us): 16/9 ts:1581527494.889008092 count:1 2351 <...>-1729 [005] d... 714.756290: #3 inner/outer(us): 16/16 ts:1581527519.678961629 count:5 2352 <...>-1729 [001] d... 718.788247: #4 inner/outer(us): 9/17 ts:1581527523.889012713 count:1 2353 <...>-1729 [002] d... 719.796341: #5 inner/outer(us): 13/9 ts:1581527524.912872606 count:1 2354 <...>-1729 [006] d... 844.787091: #6 inner/outer(us): 9/12 ts:1581527649.889048502 count:2 2355 <...>-1729 [003] d... 849.827033: #7 inner/outer(us): 18/9 ts:1581527654.889013793 count:1 2356 <...>-1729 [007] d... 853.859002: #8 inner/outer(us): 9/12 ts:1581527658.889065736 count:1 2357 <...>-1729 [001] d... 855.874978: #9 inner/outer(us): 9/11 ts:1581527660.861991877 count:1 2358 <...>-1729 [001] d... 863.938932: #10 inner/outer(us): 9/11 ts:1581527668.970010500 count:1 nmi-total:7 nmi-count:1 2359 <...>-1729 [007] d... 878.050780: #11 inner/outer(us): 9/12 ts:1581527683.385002600 count:1 nmi-total:5 nmi-count:1 2360 <...>-1729 [007] d... 886.114702: #12 inner/outer(us): 9/12 ts:1581527691.385001600 count:1 2361 2362 2363The above output is somewhat the same in the header. All events will have 2364interrupts disabled 'd'. Under the FUNCTION title there is: 2365 2366 #1 2367 This is the count of events recorded that were greater than the 2368 tracing_threshold (See below). 2369 2370 inner/outer(us): 11/11 2371 2372 This shows two numbers as "inner latency" and "outer latency". The test 2373 runs in a loop checking a timestamp twice. The latency detected within 2374 the two timestamps is the "inner latency" and the latency detected 2375 after the previous timestamp and the next timestamp in the loop is 2376 the "outer latency". 2377 2378 ts:1581527483.343962693 2379 2380 The absolute timestamp that the first latency was recorded in the window. 2381 2382 count:6 2383 2384 The number of times a latency was detected during the window. 2385 2386 nmi-total:7 nmi-count:1 2387 2388 On architectures that support it, if an NMI comes in during the 2389 test, the time spent in NMI is reported in "nmi-total" (in 2390 microseconds). 2391 2392 All architectures that have NMIs will show the "nmi-count" if an 2393 NMI comes in during the test. 2394 2395hwlat files: 2396 2397 tracing_threshold 2398 This gets automatically set to "10" to represent 10 2399 microseconds. This is the threshold of latency that 2400 needs to be detected before the trace will be recorded. 2401 2402 Note, when hwlat tracer is finished (another tracer is 2403 written into "current_tracer"), the original value for 2404 tracing_threshold is placed back into this file. 2405 2406 hwlat_detector/width 2407 The length of time the test runs with interrupts disabled. 2408 2409 hwlat_detector/window 2410 The length of time of the window which the test 2411 runs. That is, the test will run for "width" 2412 microseconds per "window" microseconds 2413 2414 tracing_cpumask 2415 When the test is started. A kernel thread is created that 2416 runs the test. This thread will alternate between CPUs 2417 listed in the tracing_cpumask between each period 2418 (one "window"). To limit the test to specific CPUs 2419 set the mask in this file to only the CPUs that the test 2420 should run on. 2421 2422function 2423-------- 2424 2425This tracer is the function tracer. Enabling the function tracer 2426can be done from the debug file system. Make sure the 2427ftrace_enabled is set; otherwise this tracer is a nop. 2428See the "ftrace_enabled" section below. 2429:: 2430 2431 # sysctl kernel.ftrace_enabled=1 2432 # echo function > current_tracer 2433 # echo 1 > tracing_on 2434 # usleep 1 2435 # echo 0 > tracing_on 2436 # cat trace 2437 # tracer: function 2438 # 2439 # entries-in-buffer/entries-written: 24799/24799 #P:4 2440 # 2441 # _-----=> irqs-off 2442 # / _----=> need-resched 2443 # | / _---=> hardirq/softirq 2444 # || / _--=> preempt-depth 2445 # ||| / delay 2446 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2447 # | | | |||| | | 2448 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write 2449 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock 2450 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify 2451 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify 2452 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify 2453 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock 2454 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock 2455 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify 2456 [...] 2457 2458 2459Note: function tracer uses ring buffers to store the above 2460entries. The newest data may overwrite the oldest data. 2461Sometimes using echo to stop the trace is not sufficient because 2462the tracing could have overwritten the data that you wanted to 2463record. For this reason, it is sometimes better to disable 2464tracing directly from a program. This allows you to stop the 2465tracing at the point that you hit the part that you are 2466interested in. To disable the tracing directly from a C program, 2467something like following code snippet can be used:: 2468 2469 int trace_fd; 2470 [...] 2471 int main(int argc, char *argv[]) { 2472 [...] 2473 trace_fd = open(tracing_file("tracing_on"), O_WRONLY); 2474 [...] 2475 if (condition_hit()) { 2476 write(trace_fd, "0", 1); 2477 } 2478 [...] 2479 } 2480 2481 2482Single thread tracing 2483--------------------- 2484 2485By writing into set_ftrace_pid you can trace a 2486single thread. For example:: 2487 2488 # cat set_ftrace_pid 2489 no pid 2490 # echo 3111 > set_ftrace_pid 2491 # cat set_ftrace_pid 2492 3111 2493 # echo function > current_tracer 2494 # cat trace | head 2495 # tracer: function 2496 # 2497 # TASK-PID CPU# TIMESTAMP FUNCTION 2498 # | | | | | 2499 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return 2500 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range 2501 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel 2502 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel 2503 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll 2504 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll 2505 # echo > set_ftrace_pid 2506 # cat trace |head 2507 # tracer: function 2508 # 2509 # TASK-PID CPU# TIMESTAMP FUNCTION 2510 # | | | | | 2511 ##### CPU 3 buffer started #### 2512 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait 2513 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry 2514 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry 2515 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit 2516 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit 2517 2518If you want to trace a function when executing, you could use 2519something like this simple program. 2520:: 2521 2522 #include <stdio.h> 2523 #include <stdlib.h> 2524 #include <sys/types.h> 2525 #include <sys/stat.h> 2526 #include <fcntl.h> 2527 #include <unistd.h> 2528 #include <string.h> 2529 2530 #define _STR(x) #x 2531 #define STR(x) _STR(x) 2532 #define MAX_PATH 256 2533 2534 const char *find_tracefs(void) 2535 { 2536 static char tracefs[MAX_PATH+1]; 2537 static int tracefs_found; 2538 char type[100]; 2539 FILE *fp; 2540 2541 if (tracefs_found) 2542 return tracefs; 2543 2544 if ((fp = fopen("/proc/mounts","r")) == NULL) { 2545 perror("/proc/mounts"); 2546 return NULL; 2547 } 2548 2549 while (fscanf(fp, "%*s %" 2550 STR(MAX_PATH) 2551 "s %99s %*s %*d %*d\n", 2552 tracefs, type) == 2) { 2553 if (strcmp(type, "tracefs") == 0) 2554 break; 2555 } 2556 fclose(fp); 2557 2558 if (strcmp(type, "tracefs") != 0) { 2559 fprintf(stderr, "tracefs not mounted"); 2560 return NULL; 2561 } 2562 2563 strcat(tracefs, "/tracing/"); 2564 tracefs_found = 1; 2565 2566 return tracefs; 2567 } 2568 2569 const char *tracing_file(const char *file_name) 2570 { 2571 static char trace_file[MAX_PATH+1]; 2572 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name); 2573 return trace_file; 2574 } 2575 2576 int main (int argc, char **argv) 2577 { 2578 if (argc < 1) 2579 exit(-1); 2580 2581 if (fork() > 0) { 2582 int fd, ffd; 2583 char line[64]; 2584 int s; 2585 2586 ffd = open(tracing_file("current_tracer"), O_WRONLY); 2587 if (ffd < 0) 2588 exit(-1); 2589 write(ffd, "nop", 3); 2590 2591 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY); 2592 s = sprintf(line, "%d\n", getpid()); 2593 write(fd, line, s); 2594 2595 write(ffd, "function", 8); 2596 2597 close(fd); 2598 close(ffd); 2599 2600 execvp(argv[1], argv+1); 2601 } 2602 2603 return 0; 2604 } 2605 2606Or this simple script! 2607:: 2608 2609 #!/bin/bash 2610 2611 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts` 2612 echo 0 > $tracefs/tracing_on 2613 echo $$ > $tracefs/set_ftrace_pid 2614 echo function > $tracefs/current_tracer 2615 echo 1 > $tracefs/tracing_on 2616 exec "$@" 2617 2618 2619function graph tracer 2620--------------------------- 2621 2622This tracer is similar to the function tracer except that it 2623probes a function on its entry and its exit. This is done by 2624using a dynamically allocated stack of return addresses in each 2625task_struct. On function entry the tracer overwrites the return 2626address of each function traced to set a custom probe. Thus the 2627original return address is stored on the stack of return address 2628in the task_struct. 2629 2630Probing on both ends of a function leads to special features 2631such as: 2632 2633- measure of a function's time execution 2634- having a reliable call stack to draw function calls graph 2635 2636This tracer is useful in several situations: 2637 2638- you want to find the reason of a strange kernel behavior and 2639 need to see what happens in detail on any areas (or specific 2640 ones). 2641 2642- you are experiencing weird latencies but it's difficult to 2643 find its origin. 2644 2645- you want to find quickly which path is taken by a specific 2646 function 2647 2648- you just want to peek inside a working kernel and want to see 2649 what happens there. 2650 2651:: 2652 2653 # tracer: function_graph 2654 # 2655 # CPU DURATION FUNCTION CALLS 2656 # | | | | | | | 2657 2658 0) | sys_open() { 2659 0) | do_sys_open() { 2660 0) | getname() { 2661 0) | kmem_cache_alloc() { 2662 0) 1.382 us | __might_sleep(); 2663 0) 2.478 us | } 2664 0) | strncpy_from_user() { 2665 0) | might_fault() { 2666 0) 1.389 us | __might_sleep(); 2667 0) 2.553 us | } 2668 0) 3.807 us | } 2669 0) 7.876 us | } 2670 0) | alloc_fd() { 2671 0) 0.668 us | _spin_lock(); 2672 0) 0.570 us | expand_files(); 2673 0) 0.586 us | _spin_unlock(); 2674 2675 2676There are several columns that can be dynamically 2677enabled/disabled. You can use every combination of options you 2678want, depending on your needs. 2679 2680- The cpu number on which the function executed is default 2681 enabled. It is sometimes better to only trace one cpu (see 2682 tracing_cpumask file) or you might sometimes see unordered 2683 function calls while cpu tracing switch. 2684 2685 - hide: echo nofuncgraph-cpu > trace_options 2686 - show: echo funcgraph-cpu > trace_options 2687 2688- The duration (function's time of execution) is displayed on 2689 the closing bracket line of a function or on the same line 2690 than the current function in case of a leaf one. It is default 2691 enabled. 2692 2693 - hide: echo nofuncgraph-duration > trace_options 2694 - show: echo funcgraph-duration > trace_options 2695 2696- The overhead field precedes the duration field in case of 2697 reached duration thresholds. 2698 2699 - hide: echo nofuncgraph-overhead > trace_options 2700 - show: echo funcgraph-overhead > trace_options 2701 - depends on: funcgraph-duration 2702 2703 ie:: 2704 2705 3) # 1837.709 us | } /* __switch_to */ 2706 3) | finish_task_switch() { 2707 3) 0.313 us | _raw_spin_unlock_irq(); 2708 3) 3.177 us | } 2709 3) # 1889.063 us | } /* __schedule */ 2710 3) ! 140.417 us | } /* __schedule */ 2711 3) # 2034.948 us | } /* schedule */ 2712 3) * 33998.59 us | } /* schedule_preempt_disabled */ 2713 2714 [...] 2715 2716 1) 0.260 us | msecs_to_jiffies(); 2717 1) 0.313 us | __rcu_read_unlock(); 2718 1) + 61.770 us | } 2719 1) + 64.479 us | } 2720 1) 0.313 us | rcu_bh_qs(); 2721 1) 0.313 us | __local_bh_enable(); 2722 1) ! 217.240 us | } 2723 1) 0.365 us | idle_cpu(); 2724 1) | rcu_irq_exit() { 2725 1) 0.417 us | rcu_eqs_enter_common.isra.47(); 2726 1) 3.125 us | } 2727 1) ! 227.812 us | } 2728 1) ! 457.395 us | } 2729 1) @ 119760.2 us | } 2730 2731 [...] 2732 2733 2) | handle_IPI() { 2734 1) 6.979 us | } 2735 2) 0.417 us | scheduler_ipi(); 2736 1) 9.791 us | } 2737 1) + 12.917 us | } 2738 2) 3.490 us | } 2739 1) + 15.729 us | } 2740 1) + 18.542 us | } 2741 2) $ 3594274 us | } 2742 2743Flags:: 2744 2745 + means that the function exceeded 10 usecs. 2746 ! means that the function exceeded 100 usecs. 2747 # means that the function exceeded 1000 usecs. 2748 * means that the function exceeded 10 msecs. 2749 @ means that the function exceeded 100 msecs. 2750 $ means that the function exceeded 1 sec. 2751 2752 2753- The task/pid field displays the thread cmdline and pid which 2754 executed the function. It is default disabled. 2755 2756 - hide: echo nofuncgraph-proc > trace_options 2757 - show: echo funcgraph-proc > trace_options 2758 2759 ie:: 2760 2761 # tracer: function_graph 2762 # 2763 # CPU TASK/PID DURATION FUNCTION CALLS 2764 # | | | | | | | | | 2765 0) sh-4802 | | d_free() { 2766 0) sh-4802 | | call_rcu() { 2767 0) sh-4802 | | __call_rcu() { 2768 0) sh-4802 | 0.616 us | rcu_process_gp_end(); 2769 0) sh-4802 | 0.586 us | check_for_new_grace_period(); 2770 0) sh-4802 | 2.899 us | } 2771 0) sh-4802 | 4.040 us | } 2772 0) sh-4802 | 5.151 us | } 2773 0) sh-4802 | + 49.370 us | } 2774 2775 2776- The absolute time field is an absolute timestamp given by the 2777 system clock since it started. A snapshot of this time is 2778 given on each entry/exit of functions 2779 2780 - hide: echo nofuncgraph-abstime > trace_options 2781 - show: echo funcgraph-abstime > trace_options 2782 2783 ie:: 2784 2785 # 2786 # TIME CPU DURATION FUNCTION CALLS 2787 # | | | | | | | | 2788 360.774522 | 1) 0.541 us | } 2789 360.774522 | 1) 4.663 us | } 2790 360.774523 | 1) 0.541 us | __wake_up_bit(); 2791 360.774524 | 1) 6.796 us | } 2792 360.774524 | 1) 7.952 us | } 2793 360.774525 | 1) 9.063 us | } 2794 360.774525 | 1) 0.615 us | journal_mark_dirty(); 2795 360.774527 | 1) 0.578 us | __brelse(); 2796 360.774528 | 1) | reiserfs_prepare_for_journal() { 2797 360.774528 | 1) | unlock_buffer() { 2798 360.774529 | 1) | wake_up_bit() { 2799 360.774529 | 1) | bit_waitqueue() { 2800 360.774530 | 1) 0.594 us | __phys_addr(); 2801 2802 2803The function name is always displayed after the closing bracket 2804for a function if the start of that function is not in the 2805trace buffer. 2806 2807Display of the function name after the closing bracket may be 2808enabled for functions whose start is in the trace buffer, 2809allowing easier searching with grep for function durations. 2810It is default disabled. 2811 2812 - hide: echo nofuncgraph-tail > trace_options 2813 - show: echo funcgraph-tail > trace_options 2814 2815 Example with nofuncgraph-tail (default):: 2816 2817 0) | putname() { 2818 0) | kmem_cache_free() { 2819 0) 0.518 us | __phys_addr(); 2820 0) 1.757 us | } 2821 0) 2.861 us | } 2822 2823 Example with funcgraph-tail:: 2824 2825 0) | putname() { 2826 0) | kmem_cache_free() { 2827 0) 0.518 us | __phys_addr(); 2828 0) 1.757 us | } /* kmem_cache_free() */ 2829 0) 2.861 us | } /* putname() */ 2830 2831The return value of each traced function can be displayed after 2832an equal sign "=". When encountering system call failures, it 2833can be very helpful to quickly locate the function that first 2834returns an error code. 2835 2836 - hide: echo nofuncgraph-retval > trace_options 2837 - show: echo funcgraph-retval > trace_options 2838 2839 Example with funcgraph-retval:: 2840 2841 1) | cgroup_migrate() { 2842 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */ 2843 1) | cgroup_migrate_execute() { 2844 1) | cpu_cgroup_can_attach() { 2845 1) | cgroup_taskset_first() { 2846 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */ 2847 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */ 2848 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */ 2849 1) 2.335 us | } /* cpu_cgroup_can_attach = -22 */ 2850 1) 4.369 us | } /* cgroup_migrate_execute = -22 */ 2851 1) 7.143 us | } /* cgroup_migrate = -22 */ 2852 2853The above example shows that the function cpu_cgroup_can_attach 2854returned the error code -22 firstly, then we can read the code 2855of this function to get the root cause. 2856 2857When the option funcgraph-retval-hex is not set, the return value can 2858be displayed in a smart way. Specifically, if it is an error code, 2859it will be printed in signed decimal format, otherwise it will 2860printed in hexadecimal format. 2861 2862 - smart: echo nofuncgraph-retval-hex > trace_options 2863 - hexadecimal: echo funcgraph-retval-hex > trace_options 2864 2865 Example with funcgraph-retval-hex:: 2866 2867 1) | cgroup_migrate() { 2868 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */ 2869 1) | cgroup_migrate_execute() { 2870 1) | cpu_cgroup_can_attach() { 2871 1) | cgroup_taskset_first() { 2872 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */ 2873 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */ 2874 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */ 2875 1) 2.335 us | } /* cpu_cgroup_can_attach = 0xffffffea */ 2876 1) 4.369 us | } /* cgroup_migrate_execute = 0xffffffea */ 2877 1) 7.143 us | } /* cgroup_migrate = 0xffffffea */ 2878 2879At present, there are some limitations when using the funcgraph-retval 2880option, and these limitations will be eliminated in the future: 2881 2882- Even if the function return type is void, a return value will still 2883 be printed, and you can just ignore it. 2884 2885- Even if return values are stored in multiple registers, only the 2886 value contained in the first register will be recorded and printed. 2887 To illustrate, in the x86 architecture, eax and edx are used to store 2888 a 64-bit return value, with the lower 32 bits saved in eax and the 2889 upper 32 bits saved in edx. However, only the value stored in eax 2890 will be recorded and printed. 2891 2892- In certain procedure call standards, such as arm64's AAPCS64, when a 2893 type is smaller than a GPR, it is the responsibility of the consumer 2894 to perform the narrowing, and the upper bits may contain UNKNOWN values. 2895 Therefore, it is advisable to check the code for such cases. For instance, 2896 when using a u8 in a 64-bit GPR, bits [63:8] may contain arbitrary values, 2897 especially when larger types are truncated, whether explicitly or implicitly. 2898 Here are some specific cases to illustrate this point: 2899 2900 **Case One**: 2901 2902 The function narrow_to_u8 is defined as follows:: 2903 2904 u8 narrow_to_u8(u64 val) 2905 { 2906 // implicitly truncated 2907 return val; 2908 } 2909 2910 It may be compiled to:: 2911 2912 narrow_to_u8: 2913 < ... ftrace instrumentation ... > 2914 RET 2915 2916 If you pass 0x123456789abcdef to this function and want to narrow it, 2917 it may be recorded as 0x123456789abcdef instead of 0xef. 2918 2919 **Case Two**: 2920 2921 The function error_if_not_4g_aligned is defined as follows:: 2922 2923 int error_if_not_4g_aligned(u64 val) 2924 { 2925 if (val & GENMASK(31, 0)) 2926 return -EINVAL; 2927 2928 return 0; 2929 } 2930 2931 It could be compiled to:: 2932 2933 error_if_not_4g_aligned: 2934 CBNZ w0, .Lnot_aligned 2935 RET // bits [31:0] are zero, bits 2936 // [63:32] are UNKNOWN 2937 .Lnot_aligned: 2938 MOV x0, #-EINVAL 2939 RET 2940 2941 When passing 0x2_0000_0000 to it, the return value may be recorded as 2942 0x2_0000_0000 instead of 0. 2943 2944You can put some comments on specific functions by using 2945trace_printk() For example, if you want to put a comment inside 2946the __might_sleep() function, you just have to include 2947<linux/ftrace.h> and call trace_printk() inside __might_sleep():: 2948 2949 trace_printk("I'm a comment!\n") 2950 2951will produce:: 2952 2953 1) | __might_sleep() { 2954 1) | /* I'm a comment! */ 2955 1) 1.449 us | } 2956 2957 2958You might find other useful features for this tracer in the 2959following "dynamic ftrace" section such as tracing only specific 2960functions or tasks. 2961 2962dynamic ftrace 2963-------------- 2964 2965If CONFIG_DYNAMIC_FTRACE is set, the system will run with 2966virtually no overhead when function tracing is disabled. The way 2967this works is the mcount function call (placed at the start of 2968every kernel function, produced by the -pg switch in gcc), 2969starts of pointing to a simple return. (Enabling FTRACE will 2970include the -pg switch in the compiling of the kernel.) 2971 2972At compile time every C file object is run through the 2973recordmcount program (located in the scripts directory). This 2974program will parse the ELF headers in the C object to find all 2975the locations in the .text section that call mcount. Starting 2976with gcc version 4.6, the -mfentry has been added for x86, which 2977calls "__fentry__" instead of "mcount". Which is called before 2978the creation of the stack frame. 2979 2980Note, not all sections are traced. They may be prevented by either 2981a notrace, or blocked another way and all inline functions are not 2982traced. Check the "available_filter_functions" file to see what functions 2983can be traced. 2984 2985A section called "__mcount_loc" is created that holds 2986references to all the mcount/fentry call sites in the .text section. 2987The recordmcount program re-links this section back into the 2988original object. The final linking stage of the kernel will add all these 2989references into a single table. 2990 2991On boot up, before SMP is initialized, the dynamic ftrace code 2992scans this table and updates all the locations into nops. It 2993also records the locations, which are added to the 2994available_filter_functions list. Modules are processed as they 2995are loaded and before they are executed. When a module is 2996unloaded, it also removes its functions from the ftrace function 2997list. This is automatic in the module unload code, and the 2998module author does not need to worry about it. 2999 3000When tracing is enabled, the process of modifying the function 3001tracepoints is dependent on architecture. The old method is to use 3002kstop_machine to prevent races with the CPUs executing code being 3003modified (which can cause the CPU to do undesirable things, especially 3004if the modified code crosses cache (or page) boundaries), and the nops are 3005patched back to calls. But this time, they do not call mcount 3006(which is just a function stub). They now call into the ftrace 3007infrastructure. 3008 3009The new method of modifying the function tracepoints is to place 3010a breakpoint at the location to be modified, sync all CPUs, modify 3011the rest of the instruction not covered by the breakpoint. Sync 3012all CPUs again, and then remove the breakpoint with the finished 3013version to the ftrace call site. 3014 3015Some archs do not even need to monkey around with the synchronization, 3016and can just slap the new code on top of the old without any 3017problems with other CPUs executing it at the same time. 3018 3019One special side-effect to the recording of the functions being 3020traced is that we can now selectively choose which functions we 3021wish to trace and which ones we want the mcount calls to remain 3022as nops. 3023 3024Two files are used, one for enabling and one for disabling the 3025tracing of specified functions. They are: 3026 3027 set_ftrace_filter 3028 3029and 3030 3031 set_ftrace_notrace 3032 3033A list of available functions that you can add to these files is 3034listed in: 3035 3036 available_filter_functions 3037 3038:: 3039 3040 # cat available_filter_functions 3041 put_prev_task_idle 3042 kmem_cache_create 3043 pick_next_task_rt 3044 cpus_read_lock 3045 pick_next_task_fair 3046 mutex_lock 3047 [...] 3048 3049If I am only interested in sys_nanosleep and hrtimer_interrupt:: 3050 3051 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter 3052 # echo function > current_tracer 3053 # echo 1 > tracing_on 3054 # usleep 1 3055 # echo 0 > tracing_on 3056 # cat trace 3057 # tracer: function 3058 # 3059 # entries-in-buffer/entries-written: 5/5 #P:4 3060 # 3061 # _-----=> irqs-off 3062 # / _----=> need-resched 3063 # | / _---=> hardirq/softirq 3064 # || / _--=> preempt-depth 3065 # ||| / delay 3066 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3067 # | | | |||| | | 3068 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath 3069 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt 3070 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt 3071 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt 3072 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt 3073 3074To see which functions are being traced, you can cat the file: 3075:: 3076 3077 # cat set_ftrace_filter 3078 hrtimer_interrupt 3079 sys_nanosleep 3080 3081 3082Perhaps this is not enough. The filters also allow glob(7) matching. 3083 3084 ``<match>*`` 3085 will match functions that begin with <match> 3086 ``*<match>`` 3087 will match functions that end with <match> 3088 ``*<match>*`` 3089 will match functions that have <match> in it 3090 ``<match1>*<match2>`` 3091 will match functions that begin with <match1> and end with <match2> 3092 3093.. note:: 3094 It is better to use quotes to enclose the wild cards, 3095 otherwise the shell may expand the parameters into names 3096 of files in the local directory. 3097 3098:: 3099 3100 # echo 'hrtimer_*' > set_ftrace_filter 3101 3102Produces:: 3103 3104 # tracer: function 3105 # 3106 # entries-in-buffer/entries-written: 897/897 #P:4 3107 # 3108 # _-----=> irqs-off 3109 # / _----=> need-resched 3110 # | / _---=> hardirq/softirq 3111 # || / _--=> preempt-depth 3112 # ||| / delay 3113 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3114 # | | | |||| | | 3115 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit 3116 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel 3117 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer 3118 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit 3119 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11 3120 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt 3121 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter 3122 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem 3123 3124Notice that we lost the sys_nanosleep. 3125:: 3126 3127 # cat set_ftrace_filter 3128 hrtimer_run_queues 3129 hrtimer_run_pending 3130 hrtimer_setup 3131 hrtimer_cancel 3132 hrtimer_try_to_cancel 3133 hrtimer_forward 3134 hrtimer_start 3135 hrtimer_reprogram 3136 hrtimer_force_reprogram 3137 hrtimer_get_next_event 3138 hrtimer_interrupt 3139 hrtimer_nanosleep 3140 hrtimer_wakeup 3141 hrtimer_get_remaining 3142 hrtimer_get_res 3143 hrtimer_init_sleeper 3144 3145 3146This is because the '>' and '>>' act just like they do in bash. 3147To rewrite the filters, use '>' 3148To append to the filters, use '>>' 3149 3150To clear out a filter so that all functions will be recorded 3151again:: 3152 3153 # echo > set_ftrace_filter 3154 # cat set_ftrace_filter 3155 # 3156 3157Again, now we want to append. 3158 3159:: 3160 3161 # echo sys_nanosleep > set_ftrace_filter 3162 # cat set_ftrace_filter 3163 sys_nanosleep 3164 # echo 'hrtimer_*' >> set_ftrace_filter 3165 # cat set_ftrace_filter 3166 hrtimer_run_queues 3167 hrtimer_run_pending 3168 hrtimer_setup 3169 hrtimer_cancel 3170 hrtimer_try_to_cancel 3171 hrtimer_forward 3172 hrtimer_start 3173 hrtimer_reprogram 3174 hrtimer_force_reprogram 3175 hrtimer_get_next_event 3176 hrtimer_interrupt 3177 sys_nanosleep 3178 hrtimer_nanosleep 3179 hrtimer_wakeup 3180 hrtimer_get_remaining 3181 hrtimer_get_res 3182 hrtimer_init_sleeper 3183 3184 3185The set_ftrace_notrace prevents those functions from being 3186traced. 3187:: 3188 3189 # echo '*preempt*' '*lock*' > set_ftrace_notrace 3190 3191Produces:: 3192 3193 # tracer: function 3194 # 3195 # entries-in-buffer/entries-written: 39608/39608 #P:4 3196 # 3197 # _-----=> irqs-off 3198 # / _----=> need-resched 3199 # | / _---=> hardirq/softirq 3200 # || / _--=> preempt-depth 3201 # ||| / delay 3202 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3203 # | | | |||| | | 3204 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open 3205 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last 3206 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last 3207 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check 3208 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement 3209 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action 3210 bash-1994 [000] .... 4342.324899: do_truncate <-do_last 3211 bash-1994 [000] .... 4342.324899: setattr_should_drop_suidgid <-do_truncate 3212 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate 3213 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change 3214 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time 3215 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time 3216 3217We can see that there's no more lock or preempt tracing. 3218 3219Selecting function filters via index 3220------------------------------------ 3221 3222Because processing of strings is expensive (the address of the function 3223needs to be looked up before comparing to the string being passed in), 3224an index can be used as well to enable functions. This is useful in the 3225case of setting thousands of specific functions at a time. By passing 3226in a list of numbers, no string processing will occur. Instead, the function 3227at the specific location in the internal array (which corresponds to the 3228functions in the "available_filter_functions" file), is selected. 3229 3230:: 3231 3232 # echo 1 > set_ftrace_filter 3233 3234Will select the first function listed in "available_filter_functions" 3235 3236:: 3237 3238 # head -1 available_filter_functions 3239 trace_initcall_finish_cb 3240 3241 # cat set_ftrace_filter 3242 trace_initcall_finish_cb 3243 3244 # head -50 available_filter_functions | tail -1 3245 x86_pmu_commit_txn 3246 3247 # echo 1 50 > set_ftrace_filter 3248 # cat set_ftrace_filter 3249 trace_initcall_finish_cb 3250 x86_pmu_commit_txn 3251 3252Dynamic ftrace with the function graph tracer 3253--------------------------------------------- 3254 3255Although what has been explained above concerns both the 3256function tracer and the function-graph-tracer, there are some 3257special features only available in the function-graph tracer. 3258 3259If you want to trace only one function and all of its children, 3260you just have to echo its name into set_graph_function:: 3261 3262 echo __do_fault > set_graph_function 3263 3264will produce the following "expanded" trace of the __do_fault() 3265function:: 3266 3267 0) | __do_fault() { 3268 0) | filemap_fault() { 3269 0) | find_lock_page() { 3270 0) 0.804 us | find_get_page(); 3271 0) | __might_sleep() { 3272 0) 1.329 us | } 3273 0) 3.904 us | } 3274 0) 4.979 us | } 3275 0) 0.653 us | _spin_lock(); 3276 0) 0.578 us | page_add_file_rmap(); 3277 0) 0.525 us | native_set_pte_at(); 3278 0) 0.585 us | _spin_unlock(); 3279 0) | unlock_page() { 3280 0) 0.541 us | page_waitqueue(); 3281 0) 0.639 us | __wake_up_bit(); 3282 0) 2.786 us | } 3283 0) + 14.237 us | } 3284 0) | __do_fault() { 3285 0) | filemap_fault() { 3286 0) | find_lock_page() { 3287 0) 0.698 us | find_get_page(); 3288 0) | __might_sleep() { 3289 0) 1.412 us | } 3290 0) 3.950 us | } 3291 0) 5.098 us | } 3292 0) 0.631 us | _spin_lock(); 3293 0) 0.571 us | page_add_file_rmap(); 3294 0) 0.526 us | native_set_pte_at(); 3295 0) 0.586 us | _spin_unlock(); 3296 0) | unlock_page() { 3297 0) 0.533 us | page_waitqueue(); 3298 0) 0.638 us | __wake_up_bit(); 3299 0) 2.793 us | } 3300 0) + 14.012 us | } 3301 3302You can also expand several functions at once:: 3303 3304 echo sys_open > set_graph_function 3305 echo sys_close >> set_graph_function 3306 3307Now if you want to go back to trace all functions you can clear 3308this special filter via:: 3309 3310 echo > set_graph_function 3311 3312 3313ftrace_enabled 3314-------------- 3315 3316Note, the proc sysctl ftrace_enable is a big on/off switch for the 3317function tracer. By default it is enabled (when function tracing is 3318enabled in the kernel). If it is disabled, all function tracing is 3319disabled. This includes not only the function tracers for ftrace, but 3320also for any other uses (perf, kprobes, stack tracing, profiling, etc). It 3321cannot be disabled if there is a callback with FTRACE_OPS_FL_PERMANENT set 3322registered. 3323 3324Please disable this with care. 3325 3326This can be disable (and enabled) with:: 3327 3328 sysctl kernel.ftrace_enabled=0 3329 sysctl kernel.ftrace_enabled=1 3330 3331 or 3332 3333 echo 0 > /proc/sys/kernel/ftrace_enabled 3334 echo 1 > /proc/sys/kernel/ftrace_enabled 3335 3336 3337Filter commands 3338--------------- 3339 3340A few commands are supported by the set_ftrace_filter interface. 3341Trace commands have the following format:: 3342 3343 <function>:<command>:<parameter> 3344 3345The following commands are supported: 3346 3347- mod: 3348 This command enables function filtering per module. The 3349 parameter defines the module. For example, if only the write* 3350 functions in the ext3 module are desired, run: 3351 3352 echo 'write*:mod:ext3' > set_ftrace_filter 3353 3354 This command interacts with the filter in the same way as 3355 filtering based on function names. Thus, adding more functions 3356 in a different module is accomplished by appending (>>) to the 3357 filter file. Remove specific module functions by prepending 3358 '!':: 3359 3360 echo '!writeback*:mod:ext3' >> set_ftrace_filter 3361 3362 Mod command supports module globbing. Disable tracing for all 3363 functions except a specific module:: 3364 3365 echo '!*:mod:!ext3' >> set_ftrace_filter 3366 3367 Disable tracing for all modules, but still trace kernel:: 3368 3369 echo '!*:mod:*' >> set_ftrace_filter 3370 3371 Enable filter only for kernel:: 3372 3373 echo '*write*:mod:!*' >> set_ftrace_filter 3374 3375 Enable filter for module globbing:: 3376 3377 echo '*write*:mod:*snd*' >> set_ftrace_filter 3378 3379- traceon/traceoff: 3380 These commands turn tracing on and off when the specified 3381 functions are hit. The parameter determines how many times the 3382 tracing system is turned on and off. If unspecified, there is 3383 no limit. For example, to disable tracing when a schedule bug 3384 is hit the first 5 times, run:: 3385 3386 echo '__schedule_bug:traceoff:5' > set_ftrace_filter 3387 3388 To always disable tracing when __schedule_bug is hit:: 3389 3390 echo '__schedule_bug:traceoff' > set_ftrace_filter 3391 3392 These commands are cumulative whether or not they are appended 3393 to set_ftrace_filter. To remove a command, prepend it by '!' 3394 and drop the parameter:: 3395 3396 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter 3397 3398 The above removes the traceoff command for __schedule_bug 3399 that have a counter. To remove commands without counters:: 3400 3401 echo '!__schedule_bug:traceoff' > set_ftrace_filter 3402 3403- snapshot: 3404 Will cause a snapshot to be triggered when the function is hit. 3405 :: 3406 3407 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter 3408 3409 To only snapshot once: 3410 :: 3411 3412 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter 3413 3414 To remove the above commands:: 3415 3416 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter 3417 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter 3418 3419- enable_event/disable_event: 3420 These commands can enable or disable a trace event. Note, because 3421 function tracing callbacks are very sensitive, when these commands 3422 are registered, the trace point is activated, but disabled in 3423 a "soft" mode. That is, the tracepoint will be called, but 3424 just will not be traced. The event tracepoint stays in this mode 3425 as long as there's a command that triggers it. 3426 :: 3427 3428 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \ 3429 set_ftrace_filter 3430 3431 The format is:: 3432 3433 <function>:enable_event:<system>:<event>[:count] 3434 <function>:disable_event:<system>:<event>[:count] 3435 3436 To remove the events commands:: 3437 3438 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \ 3439 set_ftrace_filter 3440 echo '!schedule:disable_event:sched:sched_switch' > \ 3441 set_ftrace_filter 3442 3443- dump: 3444 When the function is hit, it will dump the contents of the ftrace 3445 ring buffer to the console. This is useful if you need to debug 3446 something, and want to dump the trace when a certain function 3447 is hit. Perhaps it's a function that is called before a triple 3448 fault happens and does not allow you to get a regular dump. 3449 3450- cpudump: 3451 When the function is hit, it will dump the contents of the ftrace 3452 ring buffer for the current CPU to the console. Unlike the "dump" 3453 command, it only prints out the contents of the ring buffer for the 3454 CPU that executed the function that triggered the dump. 3455 3456- stacktrace: 3457 When the function is hit, a stack trace is recorded. 3458 3459trace_pipe 3460---------- 3461 3462The trace_pipe outputs the same content as the trace file, but 3463the effect on the tracing is different. Every read from 3464trace_pipe is consumed. This means that subsequent reads will be 3465different. The trace is live. 3466:: 3467 3468 # echo function > current_tracer 3469 # cat trace_pipe > /tmp/trace.out & 3470 [1] 4153 3471 # echo 1 > tracing_on 3472 # usleep 1 3473 # echo 0 > tracing_on 3474 # cat trace 3475 # tracer: function 3476 # 3477 # entries-in-buffer/entries-written: 0/0 #P:4 3478 # 3479 # _-----=> irqs-off 3480 # / _----=> need-resched 3481 # | / _---=> hardirq/softirq 3482 # || / _--=> preempt-depth 3483 # ||| / delay 3484 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3485 # | | | |||| | | 3486 3487 # 3488 # cat /tmp/trace.out 3489 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write 3490 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock 3491 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify 3492 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify 3493 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify 3494 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock 3495 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock 3496 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify 3497 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath 3498 3499 3500Note, reading the trace_pipe file will block until more input is 3501added. This is contrary to the trace file. If any process opened 3502the trace file for reading, it will actually disable tracing and 3503prevent new entries from being added. The trace_pipe file does 3504not have this limitation. 3505 3506trace entries 3507------------- 3508 3509Having too much or not enough data can be troublesome in 3510diagnosing an issue in the kernel. The file buffer_size_kb is 3511used to modify the size of the internal trace buffers. The 3512number listed is the number of entries that can be recorded per 3513CPU. To know the full size, multiply the number of possible CPUs 3514with the number of entries. 3515:: 3516 3517 # cat buffer_size_kb 3518 1408 (units kilobytes) 3519 3520Or simply read buffer_total_size_kb 3521:: 3522 3523 # cat buffer_total_size_kb 3524 5632 3525 3526To modify the buffer, simple echo in a number (in 1024 byte segments). 3527:: 3528 3529 # echo 10000 > buffer_size_kb 3530 # cat buffer_size_kb 3531 10000 (units kilobytes) 3532 3533It will try to allocate as much as possible. If you allocate too 3534much, it can cause Out-Of-Memory to trigger. 3535:: 3536 3537 # echo 1000000000000 > buffer_size_kb 3538 -bash: echo: write error: Cannot allocate memory 3539 # cat buffer_size_kb 3540 85 3541 3542The per_cpu buffers can be changed individually as well: 3543:: 3544 3545 # echo 10000 > per_cpu/cpu0/buffer_size_kb 3546 # echo 100 > per_cpu/cpu1/buffer_size_kb 3547 3548When the per_cpu buffers are not the same, the buffer_size_kb 3549at the top level will just show an X 3550:: 3551 3552 # cat buffer_size_kb 3553 X 3554 3555This is where the buffer_total_size_kb is useful: 3556:: 3557 3558 # cat buffer_total_size_kb 3559 12916 3560 3561Writing to the top level buffer_size_kb will reset all the buffers 3562to be the same again. 3563 3564Snapshot 3565-------- 3566CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature 3567available to all non latency tracers. (Latency tracers which 3568record max latency, such as "irqsoff" or "wakeup", can't use 3569this feature, since those are already using the snapshot 3570mechanism internally.) 3571 3572Snapshot preserves a current trace buffer at a particular point 3573in time without stopping tracing. Ftrace swaps the current 3574buffer with a spare buffer, and tracing continues in the new 3575current (=previous spare) buffer. 3576 3577The following tracefs files in "tracing" are related to this 3578feature: 3579 3580 snapshot: 3581 3582 This is used to take a snapshot and to read the output 3583 of the snapshot. Echo 1 into this file to allocate a 3584 spare buffer and to take a snapshot (swap), then read 3585 the snapshot from this file in the same format as 3586 "trace" (described above in the section "The File 3587 System"). Both reads snapshot and tracing are executable 3588 in parallel. When the spare buffer is allocated, echoing 3589 0 frees it, and echoing else (positive) values clear the 3590 snapshot contents. 3591 More details are shown in the table below. 3592 3593 +--------------+------------+------------+------------+ 3594 |status\\input | 0 | 1 | else | 3595 +==============+============+============+============+ 3596 |not allocated |(do nothing)| alloc+swap |(do nothing)| 3597 +--------------+------------+------------+------------+ 3598 |allocated | free | swap | clear | 3599 +--------------+------------+------------+------------+ 3600 3601Here is an example of using the snapshot feature. 3602:: 3603 3604 # echo 1 > events/sched/enable 3605 # echo 1 > snapshot 3606 # cat snapshot 3607 # tracer: nop 3608 # 3609 # entries-in-buffer/entries-written: 71/71 #P:8 3610 # 3611 # _-----=> irqs-off 3612 # / _----=> need-resched 3613 # | / _---=> hardirq/softirq 3614 # || / _--=> preempt-depth 3615 # ||| / delay 3616 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3617 # | | | |||| | | 3618 <idle>-0 [005] d... 2440.603828: sched_switch: prev_comm=swapper/5 prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=snapshot-test-2 next_pid=2242 next_prio=120 3619 sleep-2242 [005] d... 2440.603846: sched_switch: prev_comm=snapshot-test-2 prev_pid=2242 prev_prio=120 prev_state=R ==> next_comm=kworker/5:1 next_pid=60 next_prio=120 3620 [...] 3621 <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 3622 3623 # cat trace 3624 # tracer: nop 3625 # 3626 # entries-in-buffer/entries-written: 77/77 #P:8 3627 # 3628 # _-----=> irqs-off 3629 # / _----=> need-resched 3630 # | / _---=> hardirq/softirq 3631 # || / _--=> preempt-depth 3632 # ||| / delay 3633 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3634 # | | | |||| | | 3635 <idle>-0 [007] d... 2440.707395: sched_switch: prev_comm=swapper/7 prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=snapshot-test-2 next_pid=2243 next_prio=120 3636 snapshot-test-2-2229 [002] d... 2440.707438: sched_switch: prev_comm=snapshot-test-2 prev_pid=2229 prev_prio=120 prev_state=S ==> next_comm=swapper/2 next_pid=0 next_prio=120 3637 [...] 3638 3639 3640If you try to use this snapshot feature when current tracer is 3641one of the latency tracers, you will get the following results. 3642:: 3643 3644 # echo wakeup > current_tracer 3645 # echo 1 > snapshot 3646 bash: echo: write error: Device or resource busy 3647 # cat snapshot 3648 cat: snapshot: Device or resource busy 3649 3650 3651Instances 3652--------- 3653In the tracefs tracing directory, there is a directory called "instances". 3654This directory can have new directories created inside of it using 3655mkdir, and removing directories with rmdir. The directory created 3656with mkdir in this directory will already contain files and other 3657directories after it is created. 3658:: 3659 3660 # mkdir instances/foo 3661 # ls instances/foo 3662 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu 3663 set_event snapshot trace trace_clock trace_marker trace_options 3664 trace_pipe tracing_on 3665 3666As you can see, the new directory looks similar to the tracing directory 3667itself. In fact, it is very similar, except that the buffer and 3668events are agnostic from the main directory, or from any other 3669instances that are created. 3670 3671The files in the new directory work just like the files with the 3672same name in the tracing directory except the buffer that is used 3673is a separate and new buffer. The files affect that buffer but do not 3674affect the main buffer with the exception of trace_options. Currently, 3675the trace_options affect all instances and the top level buffer 3676the same, but this may change in future releases. That is, options 3677may become specific to the instance they reside in. 3678 3679Notice that none of the function tracer files are there, nor is 3680current_tracer and available_tracers. This is because the buffers 3681can currently only have events enabled for them. 3682:: 3683 3684 # mkdir instances/foo 3685 # mkdir instances/bar 3686 # mkdir instances/zoot 3687 # echo 100000 > buffer_size_kb 3688 # echo 1000 > instances/foo/buffer_size_kb 3689 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb 3690 # echo function > current_trace 3691 # echo 1 > instances/foo/events/sched/sched_wakeup/enable 3692 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable 3693 # echo 1 > instances/foo/events/sched/sched_switch/enable 3694 # echo 1 > instances/bar/events/irq/enable 3695 # echo 1 > instances/zoot/events/syscalls/enable 3696 # cat trace_pipe 3697 CPU:2 [LOST 11745 EVENTS] 3698 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist 3699 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave 3700 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist 3701 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist 3702 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock 3703 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype 3704 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist 3705 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist 3706 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics 3707 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics 3708 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process 3709 [...] 3710 3711 # cat instances/foo/trace_pipe 3712 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000 3713 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000 3714 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003 3715 <idle>-0 [003] d..3 136.676909: sched_switch: prev_comm=swapper/3 prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=rcu_preempt next_pid=9 next_prio=120 3716 rcu_preempt-9 [003] d..3 136.676916: sched_switch: prev_comm=rcu_preempt prev_pid=9 prev_prio=120 prev_state=S ==> next_comm=swapper/3 next_pid=0 next_prio=120 3717 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000 3718 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000 3719 bash-1998 [000] d..3 136.677018: sched_switch: prev_comm=bash prev_pid=1998 prev_prio=120 prev_state=R+ ==> next_comm=kworker/0:1 next_pid=59 next_prio=120 3720 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001 3721 kworker/0:1-59 [000] d..3 136.677025: sched_switch: prev_comm=kworker/0:1 prev_pid=59 prev_prio=120 prev_state=S ==> next_comm=bash next_pid=1998 next_prio=120 3722 [...] 3723 3724 # cat instances/bar/trace_pipe 3725 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX] 3726 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX] 3727 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER] 3728 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU] 3729 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER] 3730 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER] 3731 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU] 3732 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU] 3733 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4 3734 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled 3735 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0 3736 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled 3737 [...] 3738 3739 # cat instances/zoot/trace 3740 # tracer: nop 3741 # 3742 # entries-in-buffer/entries-written: 18996/18996 #P:4 3743 # 3744 # _-----=> irqs-off 3745 # / _----=> need-resched 3746 # | / _---=> hardirq/softirq 3747 # || / _--=> preempt-depth 3748 # ||| / delay 3749 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3750 # | | | |||| | | 3751 bash-1998 [000] d... 140.733501: sys_write -> 0x2 3752 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1) 3753 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1 3754 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0) 3755 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1 3756 bash-1998 [000] d... 140.733510: sys_close(fd: a) 3757 bash-1998 [000] d... 140.733510: sys_close -> 0x0 3758 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8) 3759 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0 3760 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8) 3761 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0 3762 3763You can see that the trace of the top most trace buffer shows only 3764the function tracing. The foo instance displays wakeups and task 3765switches. 3766 3767To remove the instances, simply delete their directories: 3768:: 3769 3770 # rmdir instances/foo 3771 # rmdir instances/bar 3772 # rmdir instances/zoot 3773 3774Note, if a process has a trace file open in one of the instance 3775directories, the rmdir will fail with EBUSY. 3776 3777 3778Stack trace 3779----------- 3780Since the kernel has a fixed sized stack, it is important not to 3781waste it in functions. A kernel developer must be conscious of 3782what they allocate on the stack. If they add too much, the system 3783can be in danger of a stack overflow, and corruption will occur, 3784usually leading to a system panic. 3785 3786There are some tools that check this, usually with interrupts 3787periodically checking usage. But if you can perform a check 3788at every function call that will become very useful. As ftrace provides 3789a function tracer, it makes it convenient to check the stack size 3790at every function call. This is enabled via the stack tracer. 3791 3792CONFIG_STACK_TRACER enables the ftrace stack tracing functionality. 3793To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled. 3794:: 3795 3796 # echo 1 > /proc/sys/kernel/stack_tracer_enabled 3797 3798You can also enable it from the kernel command line to trace 3799the stack size of the kernel during boot up, by adding "stacktrace" 3800to the kernel command line parameter. 3801 3802After running it for a few minutes, the output looks like: 3803:: 3804 3805 # cat stack_max_size 3806 2928 3807 3808 # cat stack_trace 3809 Depth Size Location (18 entries) 3810 ----- ---- -------- 3811 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac 3812 1) 2704 160 find_busiest_group+0x31/0x1f1 3813 2) 2544 256 load_balance+0xd9/0x662 3814 3) 2288 80 idle_balance+0xbb/0x130 3815 4) 2208 128 __schedule+0x26e/0x5b9 3816 5) 2080 16 schedule+0x64/0x66 3817 6) 2064 128 schedule_timeout+0x34/0xe0 3818 7) 1936 112 wait_for_common+0x97/0xf1 3819 8) 1824 16 wait_for_completion+0x1d/0x1f 3820 9) 1808 128 flush_work+0xfe/0x119 3821 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20 3822 11) 1664 48 input_available_p+0x1d/0x5c 3823 12) 1616 48 n_tty_poll+0x6d/0x134 3824 13) 1568 64 tty_poll+0x64/0x7f 3825 14) 1504 880 do_select+0x31e/0x511 3826 15) 624 400 core_sys_select+0x177/0x216 3827 16) 224 96 sys_select+0x91/0xb9 3828 17) 128 128 system_call_fastpath+0x16/0x1b 3829 3830Note, if -mfentry is being used by gcc, functions get traced before 3831they set up the stack frame. This means that leaf level functions 3832are not tested by the stack tracer when -mfentry is used. 3833 3834Currently, -mfentry is used by gcc 4.6.0 and above on x86 only. 3835 3836More 3837---- 3838More details can be found in the source code, in the `kernel/trace/*.c` files. 3839