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 is 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_size_kb: 184 185 This sets or displays the number of kilobytes each CPU 186 buffer holds. By default, the trace buffers are the same size 187 for each CPU. The displayed number is the size of the 188 CPU buffer and not total size of all buffers. The 189 trace buffers are allocated in pages (blocks of memory 190 that the kernel uses for allocation, usually 4 KB in size). 191 A few extra pages may be allocated to accommodate buffer management 192 meta-data. If the last page allocated has room for more bytes 193 than requested, the rest of the page will be used, 194 making the actual allocation bigger than requested or shown. 195 ( Note, the size may not be a multiple of the page size 196 due to buffer management meta-data. ) 197 198 Buffer sizes for individual CPUs may vary 199 (see "per_cpu/cpu0/buffer_size_kb" below), and if they do 200 this file will show "X". 201 202 buffer_total_size_kb: 203 204 This displays the total combined size of all the trace buffers. 205 206 free_buffer: 207 208 If a process is performing tracing, and the ring buffer should be 209 shrunk "freed" when the process is finished, even if it were to be 210 killed by a signal, this file can be used for that purpose. On close 211 of this file, the ring buffer will be resized to its minimum size. 212 Having a process that is tracing also open this file, when the process 213 exits its file descriptor for this file will be closed, and in doing so, 214 the ring buffer will be "freed". 215 216 It may also stop tracing if disable_on_free option is set. 217 218 tracing_cpumask: 219 220 This is a mask that lets the user only trace on specified CPUs. 221 The format is a hex string representing the CPUs. 222 223 set_ftrace_filter: 224 225 When dynamic ftrace is configured in (see the 226 section below "dynamic ftrace"), the code is dynamically 227 modified (code text rewrite) to disable calling of the 228 function profiler (mcount). This lets tracing be configured 229 in with practically no overhead in performance. This also 230 has a side effect of enabling or disabling specific functions 231 to be traced. Echoing names of functions into this file 232 will limit the trace to only those functions. 233 This influences the tracers "function" and "function_graph" 234 and thus also function profiling (see "function_profile_enabled"). 235 236 The functions listed in "available_filter_functions" are what 237 can be written into this file. 238 239 This interface also allows for commands to be used. See the 240 "Filter commands" section for more details. 241 242 As a speed up, since processing strings can be quite expensive 243 and requires a check of all functions registered to tracing, instead 244 an index can be written into this file. A number (starting with "1") 245 written will instead select the same corresponding at the line position 246 of the "available_filter_functions" file. 247 248 set_ftrace_notrace: 249 250 This has an effect opposite to that of 251 set_ftrace_filter. Any function that is added here will not 252 be traced. If a function exists in both set_ftrace_filter 253 and set_ftrace_notrace, the function will _not_ be traced. 254 255 set_ftrace_pid: 256 257 Have the function tracer only trace the threads whose PID are 258 listed in this file. 259 260 If the "function-fork" option is set, then when a task whose 261 PID is listed in this file forks, the child's PID will 262 automatically be added to this file, and the child will be 263 traced by the function tracer as well. This option will also 264 cause PIDs of tasks that exit to be removed from the file. 265 266 set_ftrace_notrace_pid: 267 268 Have the function tracer ignore threads whose PID are listed in 269 this file. 270 271 If the "function-fork" option is set, then when a task whose 272 PID is listed in this file forks, the child's PID will 273 automatically be added to this file, and the child will not be 274 traced by the function tracer as well. This option will also 275 cause PIDs of tasks that exit to be removed from the file. 276 277 If a PID is in both this file and "set_ftrace_pid", then this 278 file takes precedence, and the thread will not be traced. 279 280 set_event_pid: 281 282 Have the events only trace a task with a PID listed in this file. 283 Note, sched_switch and sched_wake_up will also trace events 284 listed in this file. 285 286 To have the PIDs of children of tasks with their PID in this file 287 added on fork, enable the "event-fork" option. That option will also 288 cause the PIDs of tasks to be removed from this file when the task 289 exits. 290 291 set_event_notrace_pid: 292 293 Have the events not trace a task with a PID listed in this file. 294 Note, sched_switch and sched_wakeup will trace threads not listed 295 in this file, even if a thread's PID is in the file if the 296 sched_switch or sched_wakeup events also trace a thread that should 297 be traced. 298 299 To have the PIDs of children of tasks with their PID in this file 300 added on fork, enable the "event-fork" option. That option will also 301 cause the PIDs of tasks to be removed from this file when the task 302 exits. 303 304 set_graph_function: 305 306 Functions listed in this file will cause the function graph 307 tracer to only trace these functions and the functions that 308 they call. (See the section "dynamic ftrace" for more details). 309 Note, set_ftrace_filter and set_ftrace_notrace still affects 310 what functions are being traced. 311 312 set_graph_notrace: 313 314 Similar to set_graph_function, but will disable function graph 315 tracing when the function is hit until it exits the function. 316 This makes it possible to ignore tracing functions that are called 317 by a specific function. 318 319 available_filter_functions: 320 321 This lists the functions that ftrace has processed and can trace. 322 These are the function names that you can pass to 323 "set_ftrace_filter", "set_ftrace_notrace", 324 "set_graph_function", or "set_graph_notrace". 325 (See the section "dynamic ftrace" below for more details.) 326 327 dyn_ftrace_total_info: 328 329 This file is for debugging purposes. The number of functions that 330 have been converted to nops and are available to be traced. 331 332 enabled_functions: 333 334 This file is more for debugging ftrace, but can also be useful 335 in seeing if any function has a callback attached to it. 336 Not only does the trace infrastructure use ftrace function 337 trace utility, but other subsystems might too. This file 338 displays all functions that have a callback attached to them 339 as well as the number of callbacks that have been attached. 340 Note, a callback may also call multiple functions which will 341 not be listed in this count. 342 343 If the callback registered to be traced by a function with 344 the "save regs" attribute (thus even more overhead), a 'R' 345 will be displayed on the same line as the function that 346 is returning registers. 347 348 If the callback registered to be traced by a function with 349 the "ip modify" attribute (thus the regs->ip can be changed), 350 an 'I' will be displayed on the same line as the function that 351 can be overridden. 352 353 If a non ftrace trampoline is attached (BPF) a 'D' will be displayed. 354 Note, normal ftrace trampolines can also be attached, but only one 355 "direct" trampoline can be attached to a given function at a time. 356 357 Some architectures can not call direct trampolines, but instead have 358 the ftrace ops function located above the function entry point. In 359 such cases an 'O' will be displayed. 360 361 If a function had either the "ip modify" or a "direct" call attached to 362 it in the past, a 'M' will be shown. This flag is never cleared. It is 363 used to know if a function was every modified by the ftrace infrastructure, 364 and can be used for debugging. 365 366 If the architecture supports it, it will also show what callback 367 is being directly called by the function. If the count is greater 368 than 1 it most likely will be ftrace_ops_list_func(). 369 370 If the callback of a function jumps to a trampoline that is 371 specific to the callback and which is not the standard trampoline, 372 its address will be printed as well as the function that the 373 trampoline calls. 374 375 touched_functions: 376 377 This file contains all the functions that ever had a function callback 378 to it via the ftrace infrastructure. It has the same format as 379 enabled_functions but shows all functions that have every been 380 traced. 381 382 To see any function that has every been modified by "ip modify" or a 383 direct trampoline, one can perform the following command: 384 385 grep ' M ' /sys/kernel/tracing/touched_functions 386 387 function_profile_enabled: 388 389 When set it will enable all functions with either the function 390 tracer, or if configured, the function graph tracer. It will 391 keep a histogram of the number of functions that were called 392 and if the function graph tracer was configured, it will also keep 393 track of the time spent in those functions. The histogram 394 content can be displayed in the files: 395 396 trace_stat/function<cpu> ( function0, function1, etc). 397 398 trace_stat: 399 400 A directory that holds different tracing stats. 401 402 kprobe_events: 403 404 Enable dynamic trace points. See kprobetrace.rst. 405 406 kprobe_profile: 407 408 Dynamic trace points stats. See kprobetrace.rst. 409 410 max_graph_depth: 411 412 Used with the function graph tracer. This is the max depth 413 it will trace into a function. Setting this to a value of 414 one will show only the first kernel function that is called 415 from user space. 416 417 printk_formats: 418 419 This is for tools that read the raw format files. If an event in 420 the ring buffer references a string, only a pointer to the string 421 is recorded into the buffer and not the string itself. This prevents 422 tools from knowing what that string was. This file displays the string 423 and address for the string allowing tools to map the pointers to what 424 the strings were. 425 426 saved_cmdlines: 427 428 Only the pid of the task is recorded in a trace event unless 429 the event specifically saves the task comm as well. Ftrace 430 makes a cache of pid mappings to comms to try to display 431 comms for events. If a pid for a comm is not listed, then 432 "<...>" is displayed in the output. 433 434 If the option "record-cmd" is set to "0", then comms of tasks 435 will not be saved during recording. By default, it is enabled. 436 437 saved_cmdlines_size: 438 439 By default, 128 comms are saved (see "saved_cmdlines" above). To 440 increase or decrease the amount of comms that are cached, echo 441 the number of comms to cache into this file. 442 443 saved_tgids: 444 445 If the option "record-tgid" is set, on each scheduling context switch 446 the Task Group ID of a task is saved in a table mapping the PID of 447 the thread to its TGID. By default, the "record-tgid" option is 448 disabled. 449 450 snapshot: 451 452 This displays the "snapshot" buffer and also lets the user 453 take a snapshot of the current running trace. 454 See the "Snapshot" section below for more details. 455 456 stack_max_size: 457 458 When the stack tracer is activated, this will display the 459 maximum stack size it has encountered. 460 See the "Stack Trace" section below. 461 462 stack_trace: 463 464 This displays the stack back trace of the largest stack 465 that was encountered when the stack tracer is activated. 466 See the "Stack Trace" section below. 467 468 stack_trace_filter: 469 470 This is similar to "set_ftrace_filter" but it limits what 471 functions the stack tracer will check. 472 473 trace_clock: 474 475 Whenever an event is recorded into the ring buffer, a 476 "timestamp" is added. This stamp comes from a specified 477 clock. By default, ftrace uses the "local" clock. This 478 clock is very fast and strictly per cpu, but on some 479 systems it may not be monotonic with respect to other 480 CPUs. In other words, the local clocks may not be in sync 481 with local clocks on other CPUs. 482 483 Usual clocks for tracing:: 484 485 # cat trace_clock 486 [local] global counter x86-tsc 487 488 The clock with the square brackets around it is the one in effect. 489 490 local: 491 Default clock, but may not be in sync across CPUs 492 493 global: 494 This clock is in sync with all CPUs but may 495 be a bit slower than the local clock. 496 497 counter: 498 This is not a clock at all, but literally an atomic 499 counter. It counts up one by one, but is in sync 500 with all CPUs. This is useful when you need to 501 know exactly the order events occurred with respect to 502 each other on different CPUs. 503 504 uptime: 505 This uses the jiffies counter and the time stamp 506 is relative to the time since boot up. 507 508 perf: 509 This makes ftrace use the same clock that perf uses. 510 Eventually perf will be able to read ftrace buffers 511 and this will help out in interleaving the data. 512 513 x86-tsc: 514 Architectures may define their own clocks. For 515 example, x86 uses its own TSC cycle clock here. 516 517 ppc-tb: 518 This uses the powerpc timebase register value. 519 This is in sync across CPUs and can also be used 520 to correlate events across hypervisor/guest if 521 tb_offset is known. 522 523 mono: 524 This uses the fast monotonic clock (CLOCK_MONOTONIC) 525 which is monotonic and is subject to NTP rate adjustments. 526 527 mono_raw: 528 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW) 529 which is monotonic but is not subject to any rate adjustments 530 and ticks at the same rate as the hardware clocksource. 531 532 boot: 533 This is the boot clock (CLOCK_BOOTTIME) and is based on the 534 fast monotonic clock, but also accounts for time spent in 535 suspend. Since the clock access is designed for use in 536 tracing in the suspend path, some side effects are possible 537 if clock is accessed after the suspend time is accounted before 538 the fast mono clock is updated. In this case, the clock update 539 appears to happen slightly sooner than it normally would have. 540 Also on 32-bit systems, it's possible that the 64-bit boot offset 541 sees a partial update. These effects are rare and post 542 processing should be able to handle them. See comments in the 543 ktime_get_boot_fast_ns() function for more information. 544 545 tai: 546 This is the tai clock (CLOCK_TAI) and is derived from the wall- 547 clock time. However, this clock does not experience 548 discontinuities and backwards jumps caused by NTP inserting leap 549 seconds. Since the clock access is designed for use in tracing, 550 side effects are possible. The clock access may yield wrong 551 readouts in case the internal TAI offset is updated e.g., caused 552 by setting the system time or using adjtimex() with an offset. 553 These effects are rare and post processing should be able to 554 handle them. See comments in the ktime_get_tai_fast_ns() 555 function for more information. 556 557 To set a clock, simply echo the clock name into this file:: 558 559 # echo global > trace_clock 560 561 Setting a clock clears the ring buffer content as well as the 562 "snapshot" buffer. 563 564 trace_marker: 565 566 This is a very useful file for synchronizing user space 567 with events happening in the kernel. Writing strings into 568 this file will be written into the ftrace buffer. 569 570 It is useful in applications to open this file at the start 571 of the application and just reference the file descriptor 572 for the file:: 573 574 void trace_write(const char *fmt, ...) 575 { 576 va_list ap; 577 char buf[256]; 578 int n; 579 580 if (trace_fd < 0) 581 return; 582 583 va_start(ap, fmt); 584 n = vsnprintf(buf, 256, fmt, ap); 585 va_end(ap); 586 587 write(trace_fd, buf, n); 588 } 589 590 start:: 591 592 trace_fd = open("trace_marker", O_WRONLY); 593 594 Note: Writing into the trace_marker file can also initiate triggers 595 that are written into /sys/kernel/tracing/events/ftrace/print/trigger 596 See "Event triggers" in Documentation/trace/events.rst and an 597 example in Documentation/trace/histogram.rst (Section 3.) 598 599 trace_marker_raw: 600 601 This is similar to trace_marker above, but is meant for binary data 602 to be written to it, where a tool can be used to parse the data 603 from trace_pipe_raw. 604 605 uprobe_events: 606 607 Add dynamic tracepoints in programs. 608 See uprobetracer.rst 609 610 uprobe_profile: 611 612 Uprobe statistics. See uprobetrace.txt 613 614 instances: 615 616 This is a way to make multiple trace buffers where different 617 events can be recorded in different buffers. 618 See "Instances" section below. 619 620 events: 621 622 This is the trace event directory. It holds event tracepoints 623 (also known as static tracepoints) that have been compiled 624 into the kernel. It shows what event tracepoints exist 625 and how they are grouped by system. There are "enable" 626 files at various levels that can enable the tracepoints 627 when a "1" is written to them. 628 629 See events.rst for more information. 630 631 set_event: 632 633 By echoing in the event into this file, will enable that event. 634 635 See events.rst for more information. 636 637 available_events: 638 639 A list of events that can be enabled in tracing. 640 641 See events.rst for more information. 642 643 timestamp_mode: 644 645 Certain tracers may change the timestamp mode used when 646 logging trace events into the event buffer. Events with 647 different modes can coexist within a buffer but the mode in 648 effect when an event is logged determines which timestamp mode 649 is used for that event. The default timestamp mode is 650 'delta'. 651 652 Usual timestamp modes for tracing: 653 654 # cat timestamp_mode 655 [delta] absolute 656 657 The timestamp mode with the square brackets around it is the 658 one in effect. 659 660 delta: Default timestamp mode - timestamp is a delta against 661 a per-buffer timestamp. 662 663 absolute: The timestamp is a full timestamp, not a delta 664 against some other value. As such it takes up more 665 space and is less efficient. 666 667 hwlat_detector: 668 669 Directory for the Hardware Latency Detector. 670 See "Hardware Latency Detector" section below. 671 672 per_cpu: 673 674 This is a directory that contains the trace per_cpu information. 675 676 per_cpu/cpu0/buffer_size_kb: 677 678 The ftrace buffer is defined per_cpu. That is, there's a separate 679 buffer for each CPU to allow writes to be done atomically, 680 and free from cache bouncing. These buffers may have different 681 size buffers. This file is similar to the buffer_size_kb 682 file, but it only displays or sets the buffer size for the 683 specific CPU. (here cpu0). 684 685 per_cpu/cpu0/trace: 686 687 This is similar to the "trace" file, but it will only display 688 the data specific for the CPU. If written to, it only clears 689 the specific CPU buffer. 690 691 per_cpu/cpu0/trace_pipe 692 693 This is similar to the "trace_pipe" file, and is a consuming 694 read, but it will only display (and consume) the data specific 695 for the CPU. 696 697 per_cpu/cpu0/trace_pipe_raw 698 699 For tools that can parse the ftrace ring buffer binary format, 700 the trace_pipe_raw file can be used to extract the data 701 from the ring buffer directly. With the use of the splice() 702 system call, the buffer data can be quickly transferred to 703 a file or to the network where a server is collecting the 704 data. 705 706 Like trace_pipe, this is a consuming reader, where multiple 707 reads will always produce different data. 708 709 per_cpu/cpu0/snapshot: 710 711 This is similar to the main "snapshot" file, but will only 712 snapshot the current CPU (if supported). It only displays 713 the content of the snapshot for a given CPU, and if 714 written to, only clears this CPU buffer. 715 716 per_cpu/cpu0/snapshot_raw: 717 718 Similar to the trace_pipe_raw, but will read the binary format 719 from the snapshot buffer for the given CPU. 720 721 per_cpu/cpu0/stats: 722 723 This displays certain stats about the ring buffer: 724 725 entries: 726 The number of events that are still in the buffer. 727 728 overrun: 729 The number of lost events due to overwriting when 730 the buffer was full. 731 732 commit overrun: 733 Should always be zero. 734 This gets set if so many events happened within a nested 735 event (ring buffer is re-entrant), that it fills the 736 buffer and starts dropping events. 737 738 bytes: 739 Bytes actually read (not overwritten). 740 741 oldest event ts: 742 The oldest timestamp in the buffer 743 744 now ts: 745 The current timestamp 746 747 dropped events: 748 Events lost due to overwrite option being off. 749 750 read events: 751 The number of events read. 752 753The Tracers 754----------- 755 756Here is the list of current tracers that may be configured. 757 758 "function" 759 760 Function call tracer to trace all kernel functions. 761 762 "function_graph" 763 764 Similar to the function tracer except that the 765 function tracer probes the functions on their entry 766 whereas the function graph tracer traces on both entry 767 and exit of the functions. It then provides the ability 768 to draw a graph of function calls similar to C code 769 source. 770 771 "blk" 772 773 The block tracer. The tracer used by the blktrace user 774 application. 775 776 "hwlat" 777 778 The Hardware Latency tracer is used to detect if the hardware 779 produces any latency. See "Hardware Latency Detector" section 780 below. 781 782 "irqsoff" 783 784 Traces the areas that disable interrupts and saves 785 the trace with the longest max latency. 786 See tracing_max_latency. When a new max is recorded, 787 it replaces the old trace. It is best to view this 788 trace with the latency-format option enabled, which 789 happens automatically when the tracer is selected. 790 791 "preemptoff" 792 793 Similar to irqsoff but traces and records the amount of 794 time for which preemption is disabled. 795 796 "preemptirqsoff" 797 798 Similar to irqsoff and preemptoff, but traces and 799 records the largest time for which irqs and/or preemption 800 is disabled. 801 802 "wakeup" 803 804 Traces and records the max latency that it takes for 805 the highest priority task to get scheduled after 806 it has been woken up. 807 Traces all tasks as an average developer would expect. 808 809 "wakeup_rt" 810 811 Traces and records the max latency that it takes for just 812 RT tasks (as the current "wakeup" does). This is useful 813 for those interested in wake up timings of RT tasks. 814 815 "wakeup_dl" 816 817 Traces and records the max latency that it takes for 818 a SCHED_DEADLINE task to be woken (as the "wakeup" and 819 "wakeup_rt" does). 820 821 "mmiotrace" 822 823 A special tracer that is used to trace binary module. 824 It will trace all the calls that a module makes to the 825 hardware. Everything it writes and reads from the I/O 826 as well. 827 828 "branch" 829 830 This tracer can be configured when tracing likely/unlikely 831 calls within the kernel. It will trace when a likely and 832 unlikely branch is hit and if it was correct in its prediction 833 of being correct. 834 835 "nop" 836 837 This is the "trace nothing" tracer. To remove all 838 tracers from tracing simply echo "nop" into 839 current_tracer. 840 841Error conditions 842---------------- 843 844 For most ftrace commands, failure modes are obvious and communicated 845 using standard return codes. 846 847 For other more involved commands, extended error information may be 848 available via the tracing/error_log file. For the commands that 849 support it, reading the tracing/error_log file after an error will 850 display more detailed information about what went wrong, if 851 information is available. The tracing/error_log file is a circular 852 error log displaying a small number (currently, 8) of ftrace errors 853 for the last (8) failed commands. 854 855 The extended error information and usage takes the form shown in 856 this example:: 857 858 # echo xxx > /sys/kernel/tracing/events/sched/sched_wakeup/trigger 859 echo: write error: Invalid argument 860 861 # cat /sys/kernel/tracing/error_log 862 [ 5348.887237] location: error: Couldn't yyy: zzz 863 Command: xxx 864 ^ 865 [ 7517.023364] location: error: Bad rrr: sss 866 Command: ppp qqq 867 ^ 868 869 To clear the error log, echo the empty string into it:: 870 871 # echo > /sys/kernel/tracing/error_log 872 873Examples of using the tracer 874---------------------------- 875 876Here are typical examples of using the tracers when controlling 877them only with the tracefs interface (without using any 878user-land utilities). 879 880Output format: 881-------------- 882 883Here is an example of the output format of the file "trace":: 884 885 # tracer: function 886 # 887 # entries-in-buffer/entries-written: 140080/250280 #P:4 888 # 889 # _-----=> irqs-off 890 # / _----=> need-resched 891 # | / _---=> hardirq/softirq 892 # || / _--=> preempt-depth 893 # ||| / delay 894 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 895 # | | | |||| | | 896 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath 897 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close 898 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd 899 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify 900 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock 901 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd 902 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock 903 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd 904 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close 905 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath 906 .... 907 908A header is printed with the tracer name that is represented by 909the trace. In this case the tracer is "function". Then it shows the 910number of events in the buffer as well as the total number of entries 911that were written. The difference is the number of entries that were 912lost due to the buffer filling up (250280 - 140080 = 110200 events 913lost). 914 915The header explains the content of the events. Task name "bash", the task 916PID "1977", the CPU that it was running on "000", the latency format 917(explained below), the timestamp in <secs>.<usecs> format, the 918function name that was traced "sys_close" and the parent function that 919called this function "system_call_fastpath". The timestamp is the time 920at which the function was entered. 921 922Latency trace format 923-------------------- 924 925When the latency-format option is enabled or when one of the latency 926tracers is set, the trace file gives somewhat more information to see 927why a latency happened. Here is a typical trace:: 928 929 # tracer: irqsoff 930 # 931 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 932 # -------------------------------------------------------------------- 933 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 934 # ----------------- 935 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0) 936 # ----------------- 937 # => started at: __lock_task_sighand 938 # => ended at: _raw_spin_unlock_irqrestore 939 # 940 # 941 # _------=> CPU# 942 # / _-----=> irqs-off 943 # | / _----=> need-resched 944 # || / _---=> hardirq/softirq 945 # ||| / _--=> preempt-depth 946 # |||| / delay 947 # cmd pid ||||| time | caller 948 # \ / ||||| \ | / 949 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand 950 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore 951 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore 952 ps-6143 2d..1 306us : <stack trace> 953 => trace_hardirqs_on_caller 954 => trace_hardirqs_on 955 => _raw_spin_unlock_irqrestore 956 => do_task_stat 957 => proc_tgid_stat 958 => proc_single_show 959 => seq_read 960 => vfs_read 961 => sys_read 962 => system_call_fastpath 963 964 965This shows that the current tracer is "irqsoff" tracing the time 966for which interrupts were disabled. It gives the trace version (which 967never changes) and the version of the kernel upon which this was executed on 968(3.8). Then it displays the max latency in microseconds (259 us). The number 969of trace entries displayed and the total number (both are four: #4/4). 970VP, KP, SP, and HP are always zero and are reserved for later use. 971#P is the number of online CPUs (#P:4). 972 973The task is the process that was running when the latency 974occurred. (ps pid: 6143). 975 976The start and stop (the functions in which the interrupts were 977disabled and enabled respectively) that caused the latencies: 978 979 - __lock_task_sighand is where the interrupts were disabled. 980 - _raw_spin_unlock_irqrestore is where they were enabled again. 981 982The next lines after the header are the trace itself. The header 983explains which is which. 984 985 cmd: The name of the process in the trace. 986 987 pid: The PID of that process. 988 989 CPU#: The CPU which the process was running on. 990 991 irqs-off: 'd' interrupts are disabled. '.' otherwise. 992 .. caution:: If the architecture does not support a way to 993 read the irq flags variable, an 'X' will always 994 be printed here. 995 996 need-resched: 997 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set, 998 - 'n' only TIF_NEED_RESCHED is set, 999 - 'p' only PREEMPT_NEED_RESCHED is set, 1000 - '.' otherwise. 1001 1002 hardirq/softirq: 1003 - 'Z' - NMI occurred inside a hardirq 1004 - 'z' - NMI is running 1005 - 'H' - hard irq occurred inside a softirq. 1006 - 'h' - hard irq is running 1007 - 's' - soft irq is running 1008 - '.' - normal context. 1009 1010 preempt-depth: The level of preempt_disabled 1011 1012The above is mostly meaningful for kernel developers. 1013 1014 time: 1015 When the latency-format option is enabled, the trace file 1016 output includes a timestamp relative to the start of the 1017 trace. This differs from the output when latency-format 1018 is disabled, which includes an absolute timestamp. 1019 1020 delay: 1021 This is just to help catch your eye a bit better. And 1022 needs to be fixed to be only relative to the same CPU. 1023 The marks are determined by the difference between this 1024 current trace and the next trace. 1025 1026 - '$' - greater than 1 second 1027 - '@' - greater than 100 millisecond 1028 - '*' - greater than 10 millisecond 1029 - '#' - greater than 1000 microsecond 1030 - '!' - greater than 100 microsecond 1031 - '+' - greater than 10 microsecond 1032 - ' ' - less than or equal to 10 microsecond. 1033 1034 The rest is the same as the 'trace' file. 1035 1036 Note, the latency tracers will usually end with a back trace 1037 to easily find where the latency occurred. 1038 1039trace_options 1040------------- 1041 1042The trace_options file (or the options directory) is used to control 1043what gets printed in the trace output, or manipulate the tracers. 1044To see what is available, simply cat the file:: 1045 1046 cat trace_options 1047 print-parent 1048 nosym-offset 1049 nosym-addr 1050 noverbose 1051 noraw 1052 nohex 1053 nobin 1054 noblock 1055 nofields 1056 trace_printk 1057 annotate 1058 nouserstacktrace 1059 nosym-userobj 1060 noprintk-msg-only 1061 context-info 1062 nolatency-format 1063 record-cmd 1064 norecord-tgid 1065 overwrite 1066 nodisable_on_free 1067 irq-info 1068 markers 1069 noevent-fork 1070 function-trace 1071 nofunction-fork 1072 nodisplay-graph 1073 nostacktrace 1074 nobranch 1075 1076To disable one of the options, echo in the option prepended with 1077"no":: 1078 1079 echo noprint-parent > trace_options 1080 1081To enable an option, leave off the "no":: 1082 1083 echo sym-offset > trace_options 1084 1085Here are the available options: 1086 1087 print-parent 1088 On function traces, display the calling (parent) 1089 function as well as the function being traced. 1090 :: 1091 1092 print-parent: 1093 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul 1094 1095 noprint-parent: 1096 bash-4000 [01] 1477.606694: simple_strtoul 1097 1098 1099 sym-offset 1100 Display not only the function name, but also the 1101 offset in the function. For example, instead of 1102 seeing just "ktime_get", you will see 1103 "ktime_get+0xb/0x20". 1104 :: 1105 1106 sym-offset: 1107 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0 1108 1109 sym-addr 1110 This will also display the function address as well 1111 as the function name. 1112 :: 1113 1114 sym-addr: 1115 bash-4000 [01] 1477.606694: simple_strtoul <c0339346> 1116 1117 verbose 1118 This deals with the trace file when the 1119 latency-format option is enabled. 1120 :: 1121 1122 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \ 1123 (+0.000ms): simple_strtoul (kstrtoul) 1124 1125 raw 1126 This will display raw numbers. This option is best for 1127 use with user applications that can translate the raw 1128 numbers better than having it done in the kernel. 1129 1130 hex 1131 Similar to raw, but the numbers will be in a hexadecimal format. 1132 1133 bin 1134 This will print out the formats in raw binary. 1135 1136 block 1137 When set, reading trace_pipe will not block when polled. 1138 1139 fields 1140 Print the fields as described by their types. This is a better 1141 option than using hex, bin or raw, as it gives a better parsing 1142 of the content of the event. 1143 1144 trace_printk 1145 Can disable trace_printk() from writing into the buffer. 1146 1147 annotate 1148 It is sometimes confusing when the CPU buffers are full 1149 and one CPU buffer had a lot of events recently, thus 1150 a shorter time frame, were another CPU may have only had 1151 a few events, which lets it have older events. When 1152 the trace is reported, it shows the oldest events first, 1153 and it may look like only one CPU ran (the one with the 1154 oldest events). When the annotate option is set, it will 1155 display when a new CPU buffer started:: 1156 1157 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on 1158 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on 1159 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore 1160 ##### CPU 2 buffer started #### 1161 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle 1162 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog 1163 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock 1164 1165 userstacktrace 1166 This option changes the trace. It records a 1167 stacktrace of the current user space thread after 1168 each trace event. 1169 1170 sym-userobj 1171 when user stacktrace are enabled, look up which 1172 object the address belongs to, and print a 1173 relative address. This is especially useful when 1174 ASLR is on, otherwise you don't get a chance to 1175 resolve the address to object/file/line after 1176 the app is no longer running 1177 1178 The lookup is performed when you read 1179 trace,trace_pipe. Example:: 1180 1181 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0 1182 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6] 1183 1184 1185 printk-msg-only 1186 When set, trace_printk()s will only show the format 1187 and not their parameters (if trace_bprintk() or 1188 trace_bputs() was used to save the trace_printk()). 1189 1190 context-info 1191 Show only the event data. Hides the comm, PID, 1192 timestamp, CPU, and other useful data. 1193 1194 latency-format 1195 This option changes the trace output. When it is enabled, 1196 the trace displays additional information about the 1197 latency, as described in "Latency trace format". 1198 1199 pause-on-trace 1200 When set, opening the trace file for read, will pause 1201 writing to the ring buffer (as if tracing_on was set to zero). 1202 This simulates the original behavior of the trace file. 1203 When the file is closed, tracing will be enabled again. 1204 1205 hash-ptr 1206 When set, "%p" in the event printk format displays the 1207 hashed pointer value instead of real address. 1208 This will be useful if you want to find out which hashed 1209 value is corresponding to the real value in trace log. 1210 1211 record-cmd 1212 When any event or tracer is enabled, a hook is enabled 1213 in the sched_switch trace point to fill comm cache 1214 with mapped pids and comms. But this may cause some 1215 overhead, and if you only care about pids, and not the 1216 name of the task, disabling this option can lower the 1217 impact of tracing. See "saved_cmdlines". 1218 1219 record-tgid 1220 When any event or tracer is enabled, a hook is enabled 1221 in the sched_switch trace point to fill the cache of 1222 mapped Thread Group IDs (TGID) mapping to pids. See 1223 "saved_tgids". 1224 1225 overwrite 1226 This controls what happens when the trace buffer is 1227 full. If "1" (default), the oldest events are 1228 discarded and overwritten. If "0", then the newest 1229 events are discarded. 1230 (see per_cpu/cpu0/stats for overrun and dropped) 1231 1232 disable_on_free 1233 When the free_buffer is closed, tracing will 1234 stop (tracing_on set to 0). 1235 1236 irq-info 1237 Shows the interrupt, preempt count, need resched data. 1238 When disabled, the trace looks like:: 1239 1240 # tracer: function 1241 # 1242 # entries-in-buffer/entries-written: 144405/9452052 #P:4 1243 # 1244 # TASK-PID CPU# TIMESTAMP FUNCTION 1245 # | | | | | 1246 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up 1247 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89 1248 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task 1249 1250 1251 markers 1252 When set, the trace_marker is writable (only by root). 1253 When disabled, the trace_marker will error with EINVAL 1254 on write. 1255 1256 event-fork 1257 When set, tasks with PIDs listed in set_event_pid will have 1258 the PIDs of their children added to set_event_pid when those 1259 tasks fork. Also, when tasks with PIDs in set_event_pid exit, 1260 their PIDs will be removed from the file. 1261 1262 This affects PIDs listed in set_event_notrace_pid as well. 1263 1264 function-trace 1265 The latency tracers will enable function tracing 1266 if this option is enabled (default it is). When 1267 it is disabled, the latency tracers do not trace 1268 functions. This keeps the overhead of the tracer down 1269 when performing latency tests. 1270 1271 function-fork 1272 When set, tasks with PIDs listed in set_ftrace_pid will 1273 have the PIDs of their children added to set_ftrace_pid 1274 when those tasks fork. Also, when tasks with PIDs in 1275 set_ftrace_pid exit, their PIDs will be removed from the 1276 file. 1277 1278 This affects PIDs in set_ftrace_notrace_pid as well. 1279 1280 display-graph 1281 When set, the latency tracers (irqsoff, wakeup, etc) will 1282 use function graph tracing instead of function tracing. 1283 1284 stacktrace 1285 When set, a stack trace is recorded after any trace event 1286 is recorded. 1287 1288 branch 1289 Enable branch tracing with the tracer. This enables branch 1290 tracer along with the currently set tracer. Enabling this 1291 with the "nop" tracer is the same as just enabling the 1292 "branch" tracer. 1293 1294.. tip:: Some tracers have their own options. They only appear in this 1295 file when the tracer is active. They always appear in the 1296 options directory. 1297 1298 1299Here are the per tracer options: 1300 1301Options for function tracer: 1302 1303 func_stack_trace 1304 When set, a stack trace is recorded after every 1305 function that is recorded. NOTE! Limit the functions 1306 that are recorded before enabling this, with 1307 "set_ftrace_filter" otherwise the system performance 1308 will be critically degraded. Remember to disable 1309 this option before clearing the function filter. 1310 1311Options for function_graph tracer: 1312 1313 Since the function_graph tracer has a slightly different output 1314 it has its own options to control what is displayed. 1315 1316 funcgraph-overrun 1317 When set, the "overrun" of the graph stack is 1318 displayed after each function traced. The 1319 overrun, is when the stack depth of the calls 1320 is greater than what is reserved for each task. 1321 Each task has a fixed array of functions to 1322 trace in the call graph. If the depth of the 1323 calls exceeds that, the function is not traced. 1324 The overrun is the number of functions missed 1325 due to exceeding this array. 1326 1327 funcgraph-cpu 1328 When set, the CPU number of the CPU where the trace 1329 occurred is displayed. 1330 1331 funcgraph-overhead 1332 When set, if the function takes longer than 1333 A certain amount, then a delay marker is 1334 displayed. See "delay" above, under the 1335 header description. 1336 1337 funcgraph-proc 1338 Unlike other tracers, the process' command line 1339 is not displayed by default, but instead only 1340 when a task is traced in and out during a context 1341 switch. Enabling this options has the command 1342 of each process displayed at every line. 1343 1344 funcgraph-duration 1345 At the end of each function (the return) 1346 the duration of the amount of time in the 1347 function is displayed in microseconds. 1348 1349 funcgraph-abstime 1350 When set, the timestamp is displayed at each line. 1351 1352 funcgraph-irqs 1353 When disabled, functions that happen inside an 1354 interrupt will not be traced. 1355 1356 funcgraph-tail 1357 When set, the return event will include the function 1358 that it represents. By default this is off, and 1359 only a closing curly bracket "}" is displayed for 1360 the return of a function. 1361 1362 sleep-time 1363 When running function graph tracer, to include 1364 the time a task schedules out in its function. 1365 When enabled, it will account time the task has been 1366 scheduled out as part of the function call. 1367 1368 graph-time 1369 When running function profiler with function graph tracer, 1370 to include the time to call nested functions. When this is 1371 not set, the time reported for the function will only 1372 include the time the function itself executed for, not the 1373 time for functions that it called. 1374 1375Options for blk tracer: 1376 1377 blk_classic 1378 Shows a more minimalistic output. 1379 1380 1381irqsoff 1382------- 1383 1384When interrupts are disabled, the CPU can not react to any other 1385external event (besides NMIs and SMIs). This prevents the timer 1386interrupt from triggering or the mouse interrupt from letting 1387the kernel know of a new mouse event. The result is a latency 1388with the reaction time. 1389 1390The irqsoff tracer tracks the time for which interrupts are 1391disabled. When a new maximum latency is hit, the tracer saves 1392the trace leading up to that latency point so that every time a 1393new maximum is reached, the old saved trace is discarded and the 1394new trace is saved. 1395 1396To reset the maximum, echo 0 into tracing_max_latency. Here is 1397an example:: 1398 1399 # echo 0 > options/function-trace 1400 # echo irqsoff > current_tracer 1401 # echo 1 > tracing_on 1402 # echo 0 > tracing_max_latency 1403 # ls -ltr 1404 [...] 1405 # echo 0 > tracing_on 1406 # cat trace 1407 # tracer: irqsoff 1408 # 1409 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1410 # -------------------------------------------------------------------- 1411 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1412 # ----------------- 1413 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0) 1414 # ----------------- 1415 # => started at: run_timer_softirq 1416 # => ended at: run_timer_softirq 1417 # 1418 # 1419 # _------=> CPU# 1420 # / _-----=> irqs-off 1421 # | / _----=> need-resched 1422 # || / _---=> hardirq/softirq 1423 # ||| / _--=> preempt-depth 1424 # |||| / delay 1425 # cmd pid ||||| time | caller 1426 # \ / ||||| \ | / 1427 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq 1428 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq 1429 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq 1430 <idle>-0 0dNs3 25us : <stack trace> 1431 => _raw_spin_unlock_irq 1432 => run_timer_softirq 1433 => __do_softirq 1434 => call_softirq 1435 => do_softirq 1436 => irq_exit 1437 => smp_apic_timer_interrupt 1438 => apic_timer_interrupt 1439 => rcu_idle_exit 1440 => cpu_idle 1441 => rest_init 1442 => start_kernel 1443 => x86_64_start_reservations 1444 => x86_64_start_kernel 1445 1446Here we see that we had a latency of 16 microseconds (which is 1447very good). The _raw_spin_lock_irq in run_timer_softirq disabled 1448interrupts. The difference between the 16 and the displayed 1449timestamp 25us occurred because the clock was incremented 1450between the time of recording the max latency and the time of 1451recording the function that had that latency. 1452 1453Note the above example had function-trace not set. If we set 1454function-trace, we get a much larger output:: 1455 1456 with echo 1 > options/function-trace 1457 1458 # tracer: irqsoff 1459 # 1460 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1461 # -------------------------------------------------------------------- 1462 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1463 # ----------------- 1464 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0) 1465 # ----------------- 1466 # => started at: ata_scsi_queuecmd 1467 # => ended at: ata_scsi_queuecmd 1468 # 1469 # 1470 # _------=> CPU# 1471 # / _-----=> irqs-off 1472 # | / _----=> need-resched 1473 # || / _---=> hardirq/softirq 1474 # ||| / _--=> preempt-depth 1475 # |||| / delay 1476 # cmd pid ||||| time | caller 1477 # \ / ||||| \ | / 1478 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd 1479 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave 1480 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd 1481 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev 1482 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev 1483 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd 1484 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd 1485 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd 1486 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat 1487 [...] 1488 bash-2042 3d..1 67us : delay_tsc <-__delay 1489 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc 1490 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc 1491 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc 1492 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc 1493 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue 1494 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1495 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1496 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd 1497 bash-2042 3d..1 120us : <stack trace> 1498 => _raw_spin_unlock_irqrestore 1499 => ata_scsi_queuecmd 1500 => scsi_dispatch_cmd 1501 => scsi_request_fn 1502 => __blk_run_queue_uncond 1503 => __blk_run_queue 1504 => blk_queue_bio 1505 => submit_bio_noacct 1506 => submit_bio 1507 => submit_bh 1508 => __ext3_get_inode_loc 1509 => ext3_iget 1510 => ext3_lookup 1511 => lookup_real 1512 => __lookup_hash 1513 => walk_component 1514 => lookup_last 1515 => path_lookupat 1516 => filename_lookup 1517 => user_path_at_empty 1518 => user_path_at 1519 => vfs_fstatat 1520 => vfs_stat 1521 => sys_newstat 1522 => system_call_fastpath 1523 1524 1525Here we traced a 71 microsecond latency. But we also see all the 1526functions that were called during that time. Note that by 1527enabling function tracing, we incur an added overhead. This 1528overhead may extend the latency times. But nevertheless, this 1529trace has provided some very helpful debugging information. 1530 1531If we prefer function graph output instead of function, we can set 1532display-graph option:: 1533 1534 with echo 1 > options/display-graph 1535 1536 # tracer: irqsoff 1537 # 1538 # irqsoff latency trace v1.1.5 on 4.20.0-rc6+ 1539 # -------------------------------------------------------------------- 1540 # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4) 1541 # ----------------- 1542 # | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0) 1543 # ----------------- 1544 # => started at: free_debug_processing 1545 # => ended at: return_to_handler 1546 # 1547 # 1548 # _-----=> irqs-off 1549 # / _----=> need-resched 1550 # | / _---=> hardirq/softirq 1551 # || / _--=> preempt-depth 1552 # ||| / 1553 # REL TIME CPU TASK/PID |||| DURATION FUNCTION CALLS 1554 # | | | | |||| | | | | | | 1555 0 us | 0) bash-1507 | d... | 0.000 us | _raw_spin_lock_irqsave(); 1556 0 us | 0) bash-1507 | d..1 | 0.378 us | do_raw_spin_trylock(); 1557 1 us | 0) bash-1507 | d..2 | | set_track() { 1558 2 us | 0) bash-1507 | d..2 | | save_stack_trace() { 1559 2 us | 0) bash-1507 | d..2 | | __save_stack_trace() { 1560 3 us | 0) bash-1507 | d..2 | | __unwind_start() { 1561 3 us | 0) bash-1507 | d..2 | | get_stack_info() { 1562 3 us | 0) bash-1507 | d..2 | 0.351 us | in_task_stack(); 1563 4 us | 0) bash-1507 | d..2 | 1.107 us | } 1564 [...] 1565 3750 us | 0) bash-1507 | d..1 | 0.516 us | do_raw_spin_unlock(); 1566 3750 us | 0) bash-1507 | d..1 | 0.000 us | _raw_spin_unlock_irqrestore(); 1567 3764 us | 0) bash-1507 | d..1 | 0.000 us | tracer_hardirqs_on(); 1568 bash-1507 0d..1 3792us : <stack trace> 1569 => free_debug_processing 1570 => __slab_free 1571 => kmem_cache_free 1572 => vm_area_free 1573 => remove_vma 1574 => exit_mmap 1575 => mmput 1576 => begin_new_exec 1577 => load_elf_binary 1578 => search_binary_handler 1579 => __do_execve_file.isra.32 1580 => __x64_sys_execve 1581 => do_syscall_64 1582 => entry_SYSCALL_64_after_hwframe 1583 1584preemptoff 1585---------- 1586 1587When preemption is disabled, we may be able to receive 1588interrupts but the task cannot be preempted and a higher 1589priority task must wait for preemption to be enabled again 1590before it can preempt a lower priority task. 1591 1592The preemptoff tracer traces the places that disable preemption. 1593Like the irqsoff tracer, it records the maximum latency for 1594which preemption was disabled. The control of preemptoff tracer 1595is much like the irqsoff tracer. 1596:: 1597 1598 # echo 0 > options/function-trace 1599 # echo preemptoff > current_tracer 1600 # echo 1 > tracing_on 1601 # echo 0 > tracing_max_latency 1602 # ls -ltr 1603 [...] 1604 # echo 0 > tracing_on 1605 # cat trace 1606 # tracer: preemptoff 1607 # 1608 # preemptoff latency trace v1.1.5 on 3.8.0-test+ 1609 # -------------------------------------------------------------------- 1610 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1611 # ----------------- 1612 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0) 1613 # ----------------- 1614 # => started at: do_IRQ 1615 # => ended at: do_IRQ 1616 # 1617 # 1618 # _------=> CPU# 1619 # / _-----=> irqs-off 1620 # | / _----=> need-resched 1621 # || / _---=> hardirq/softirq 1622 # ||| / _--=> preempt-depth 1623 # |||| / delay 1624 # cmd pid ||||| time | caller 1625 # \ / ||||| \ | / 1626 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ 1627 sshd-1991 1d..1 46us : irq_exit <-do_IRQ 1628 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ 1629 sshd-1991 1d..1 52us : <stack trace> 1630 => sub_preempt_count 1631 => irq_exit 1632 => do_IRQ 1633 => ret_from_intr 1634 1635 1636This has some more changes. Preemption was disabled when an 1637interrupt came in (notice the 'h'), and was enabled on exit. 1638But we also see that interrupts have been disabled when entering 1639the preempt off section and leaving it (the 'd'). We do not know if 1640interrupts were enabled in the mean time or shortly after this 1641was over. 1642:: 1643 1644 # tracer: preemptoff 1645 # 1646 # preemptoff latency trace v1.1.5 on 3.8.0-test+ 1647 # -------------------------------------------------------------------- 1648 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1649 # ----------------- 1650 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0) 1651 # ----------------- 1652 # => started at: wake_up_new_task 1653 # => ended at: task_rq_unlock 1654 # 1655 # 1656 # _------=> CPU# 1657 # / _-----=> irqs-off 1658 # | / _----=> need-resched 1659 # || / _---=> hardirq/softirq 1660 # ||| / _--=> preempt-depth 1661 # |||| / delay 1662 # cmd pid ||||| time | caller 1663 # \ / ||||| \ | / 1664 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task 1665 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq 1666 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair 1667 bash-1994 1d..1 1us : source_load <-select_task_rq_fair 1668 bash-1994 1d..1 1us : source_load <-select_task_rq_fair 1669 [...] 1670 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt 1671 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter 1672 bash-1994 1d..1 13us : add_preempt_count <-irq_enter 1673 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt 1674 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt 1675 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt 1676 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock 1677 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt 1678 [...] 1679 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event 1680 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt 1681 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit 1682 bash-1994 1d..2 36us : do_softirq <-irq_exit 1683 bash-1994 1d..2 36us : __do_softirq <-call_softirq 1684 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq 1685 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq 1686 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq 1687 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock 1688 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq 1689 [...] 1690 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks 1691 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq 1692 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable 1693 bash-1994 1dN.2 82us : idle_cpu <-irq_exit 1694 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit 1695 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit 1696 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock 1697 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock 1698 bash-1994 1.N.1 104us : <stack trace> 1699 => sub_preempt_count 1700 => _raw_spin_unlock_irqrestore 1701 => task_rq_unlock 1702 => wake_up_new_task 1703 => do_fork 1704 => sys_clone 1705 => stub_clone 1706 1707 1708The above is an example of the preemptoff trace with 1709function-trace set. Here we see that interrupts were not disabled 1710the entire time. The irq_enter code lets us know that we entered 1711an interrupt 'h'. Before that, the functions being traced still 1712show that it is not in an interrupt, but we can see from the 1713functions themselves that this is not the case. 1714 1715preemptirqsoff 1716-------------- 1717 1718Knowing the locations that have interrupts disabled or 1719preemption disabled for the longest times is helpful. But 1720sometimes we would like to know when either preemption and/or 1721interrupts are disabled. 1722 1723Consider the following code:: 1724 1725 local_irq_disable(); 1726 call_function_with_irqs_off(); 1727 preempt_disable(); 1728 call_function_with_irqs_and_preemption_off(); 1729 local_irq_enable(); 1730 call_function_with_preemption_off(); 1731 preempt_enable(); 1732 1733The irqsoff tracer will record the total length of 1734call_function_with_irqs_off() and 1735call_function_with_irqs_and_preemption_off(). 1736 1737The preemptoff tracer will record the total length of 1738call_function_with_irqs_and_preemption_off() and 1739call_function_with_preemption_off(). 1740 1741But neither will trace the time that interrupts and/or 1742preemption is disabled. This total time is the time that we can 1743not schedule. To record this time, use the preemptirqsoff 1744tracer. 1745 1746Again, using this trace is much like the irqsoff and preemptoff 1747tracers. 1748:: 1749 1750 # echo 0 > options/function-trace 1751 # echo preemptirqsoff > current_tracer 1752 # echo 1 > tracing_on 1753 # echo 0 > tracing_max_latency 1754 # ls -ltr 1755 [...] 1756 # echo 0 > tracing_on 1757 # cat trace 1758 # tracer: preemptirqsoff 1759 # 1760 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+ 1761 # -------------------------------------------------------------------- 1762 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1763 # ----------------- 1764 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0) 1765 # ----------------- 1766 # => started at: ata_scsi_queuecmd 1767 # => ended at: ata_scsi_queuecmd 1768 # 1769 # 1770 # _------=> CPU# 1771 # / _-----=> irqs-off 1772 # | / _----=> need-resched 1773 # || / _---=> hardirq/softirq 1774 # ||| / _--=> preempt-depth 1775 # |||| / delay 1776 # cmd pid ||||| time | caller 1777 # \ / ||||| \ | / 1778 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd 1779 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1780 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd 1781 ls-2230 3...1 111us : <stack trace> 1782 => sub_preempt_count 1783 => _raw_spin_unlock_irqrestore 1784 => ata_scsi_queuecmd 1785 => scsi_dispatch_cmd 1786 => scsi_request_fn 1787 => __blk_run_queue_uncond 1788 => __blk_run_queue 1789 => blk_queue_bio 1790 => submit_bio_noacct 1791 => submit_bio 1792 => submit_bh 1793 => ext3_bread 1794 => ext3_dir_bread 1795 => htree_dirblock_to_tree 1796 => ext3_htree_fill_tree 1797 => ext3_readdir 1798 => vfs_readdir 1799 => sys_getdents 1800 => system_call_fastpath 1801 1802 1803The trace_hardirqs_off_thunk is called from assembly on x86 when 1804interrupts are disabled in the assembly code. Without the 1805function tracing, we do not know if interrupts were enabled 1806within the preemption points. We do see that it started with 1807preemption enabled. 1808 1809Here is a trace with function-trace set:: 1810 1811 # tracer: preemptirqsoff 1812 # 1813 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+ 1814 # -------------------------------------------------------------------- 1815 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1816 # ----------------- 1817 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0) 1818 # ----------------- 1819 # => started at: schedule 1820 # => ended at: mutex_unlock 1821 # 1822 # 1823 # _------=> CPU# 1824 # / _-----=> irqs-off 1825 # | / _----=> need-resched 1826 # || / _---=> hardirq/softirq 1827 # ||| / _--=> preempt-depth 1828 # |||| / delay 1829 # cmd pid ||||| time | caller 1830 # \ / ||||| \ | / 1831 kworker/-59 3...1 0us : __schedule <-schedule 1832 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch 1833 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq 1834 kworker/-59 3d..2 1us : deactivate_task <-__schedule 1835 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task 1836 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task 1837 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task 1838 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair 1839 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr 1840 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr 1841 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge 1842 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge 1843 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair 1844 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair 1845 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair 1846 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair 1847 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair 1848 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair 1849 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule 1850 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping 1851 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule 1852 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task 1853 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair 1854 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair 1855 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity 1856 ls-2269 3d..2 7us : finish_task_switch <-__schedule 1857 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch 1858 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr 1859 ls-2269 3d..2 8us : irq_enter <-do_IRQ 1860 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter 1861 ls-2269 3d..2 9us : add_preempt_count <-irq_enter 1862 ls-2269 3d.h2 9us : exit_idle <-do_IRQ 1863 [...] 1864 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock 1865 ls-2269 3d.h2 20us : irq_exit <-do_IRQ 1866 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit 1867 ls-2269 3d..3 21us : do_softirq <-irq_exit 1868 ls-2269 3d..3 21us : __do_softirq <-call_softirq 1869 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq 1870 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip 1871 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip 1872 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr 1873 ls-2269 3d.s5 31us : irq_enter <-do_IRQ 1874 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter 1875 [...] 1876 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter 1877 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter 1878 ls-2269 3d.H5 32us : exit_idle <-do_IRQ 1879 ls-2269 3d.H5 32us : handle_irq <-do_IRQ 1880 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq 1881 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq 1882 [...] 1883 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll 1884 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action 1885 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq 1886 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable 1887 ls-2269 3d..3 159us : idle_cpu <-irq_exit 1888 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit 1889 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit 1890 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock 1891 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock 1892 ls-2269 3d... 186us : <stack trace> 1893 => __mutex_unlock_slowpath 1894 => mutex_unlock 1895 => process_output 1896 => n_tty_write 1897 => tty_write 1898 => vfs_write 1899 => sys_write 1900 => system_call_fastpath 1901 1902This is an interesting trace. It started with kworker running and 1903scheduling out and ls taking over. But as soon as ls released the 1904rq lock and enabled interrupts (but not preemption) an interrupt 1905triggered. When the interrupt finished, it started running softirqs. 1906But while the softirq was running, another interrupt triggered. 1907When an interrupt is running inside a softirq, the annotation is 'H'. 1908 1909 1910wakeup 1911------ 1912 1913One common case that people are interested in tracing is the 1914time it takes for a task that is woken to actually wake up. 1915Now for non Real-Time tasks, this can be arbitrary. But tracing 1916it none the less can be interesting. 1917 1918Without function tracing:: 1919 1920 # echo 0 > options/function-trace 1921 # echo wakeup > current_tracer 1922 # echo 1 > tracing_on 1923 # echo 0 > tracing_max_latency 1924 # chrt -f 5 sleep 1 1925 # echo 0 > tracing_on 1926 # cat trace 1927 # tracer: wakeup 1928 # 1929 # wakeup latency trace v1.1.5 on 3.8.0-test+ 1930 # -------------------------------------------------------------------- 1931 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1932 # ----------------- 1933 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0) 1934 # ----------------- 1935 # 1936 # _------=> CPU# 1937 # / _-----=> irqs-off 1938 # | / _----=> need-resched 1939 # || / _---=> hardirq/softirq 1940 # ||| / _--=> preempt-depth 1941 # |||| / delay 1942 # cmd pid ||||| time | caller 1943 # \ / ||||| \ | / 1944 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H 1945 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up 1946 <idle>-0 3d..3 15us : __schedule <-schedule 1947 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H 1948 1949The tracer only traces the highest priority task in the system 1950to avoid tracing the normal circumstances. Here we see that 1951the kworker with a nice priority of -20 (not very nice), took 1952just 15 microseconds from the time it woke up, to the time it 1953ran. 1954 1955Non Real-Time tasks are not that interesting. A more interesting 1956trace is to concentrate only on Real-Time tasks. 1957 1958wakeup_rt 1959--------- 1960 1961In a Real-Time environment it is very important to know the 1962wakeup time it takes for the highest priority task that is woken 1963up to the time that it executes. This is also known as "schedule 1964latency". I stress the point that this is about RT tasks. It is 1965also important to know the scheduling latency of non-RT tasks, 1966but the average schedule latency is better for non-RT tasks. 1967Tools like LatencyTop are more appropriate for such 1968measurements. 1969 1970Real-Time environments are interested in the worst case latency. 1971That is the longest latency it takes for something to happen, 1972and not the average. We can have a very fast scheduler that may 1973only have a large latency once in a while, but that would not 1974work well with Real-Time tasks. The wakeup_rt tracer was designed 1975to record the worst case wakeups of RT tasks. Non-RT tasks are 1976not recorded because the tracer only records one worst case and 1977tracing non-RT tasks that are unpredictable will overwrite the 1978worst case latency of RT tasks (just run the normal wakeup 1979tracer for a while to see that effect). 1980 1981Since this tracer only deals with RT tasks, we will run this 1982slightly differently than we did with the previous tracers. 1983Instead of performing an 'ls', we will run 'sleep 1' under 1984'chrt' which changes the priority of the task. 1985:: 1986 1987 # echo 0 > options/function-trace 1988 # echo wakeup_rt > current_tracer 1989 # echo 1 > tracing_on 1990 # echo 0 > tracing_max_latency 1991 # chrt -f 5 sleep 1 1992 # echo 0 > tracing_on 1993 # cat trace 1994 # tracer: wakeup 1995 # 1996 # tracer: wakeup_rt 1997 # 1998 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 1999 # -------------------------------------------------------------------- 2000 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2001 # ----------------- 2002 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5) 2003 # ----------------- 2004 # 2005 # _------=> CPU# 2006 # / _-----=> irqs-off 2007 # | / _----=> need-resched 2008 # || / _---=> hardirq/softirq 2009 # ||| / _--=> preempt-depth 2010 # |||| / delay 2011 # cmd pid ||||| time | caller 2012 # \ / ||||| \ | / 2013 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep 2014 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up 2015 <idle>-0 3d..3 5us : __schedule <-schedule 2016 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep 2017 2018 2019Running this on an idle system, we see that it only took 5 microseconds 2020to perform the task switch. Note, since the trace point in the schedule 2021is before the actual "switch", we stop the tracing when the recorded task 2022is about to schedule in. This may change if we add a new marker at the 2023end of the scheduler. 2024 2025Notice that the recorded task is 'sleep' with the PID of 2389 2026and it has an rt_prio of 5. This priority is user-space priority 2027and not the internal kernel priority. The policy is 1 for 2028SCHED_FIFO and 2 for SCHED_RR. 2029 2030Note, that the trace data shows the internal priority (99 - rtprio). 2031:: 2032 2033 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep 2034 2035The 0:120:R means idle was running with a nice priority of 0 (120 - 120) 2036and in the running state 'R'. The sleep task was scheduled in with 20372389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94) 2038and it too is in the running state. 2039 2040Doing the same with chrt -r 5 and function-trace set. 2041:: 2042 2043 echo 1 > options/function-trace 2044 2045 # tracer: wakeup_rt 2046 # 2047 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2048 # -------------------------------------------------------------------- 2049 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2050 # ----------------- 2051 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5) 2052 # ----------------- 2053 # 2054 # _------=> CPU# 2055 # / _-----=> irqs-off 2056 # | / _----=> need-resched 2057 # || / _---=> hardirq/softirq 2058 # ||| / _--=> preempt-depth 2059 # |||| / delay 2060 # cmd pid ||||| time | caller 2061 # \ / ||||| \ | / 2062 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep 2063 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up 2064 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup 2065 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr 2066 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup 2067 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up 2068 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock 2069 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up 2070 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up 2071 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2072 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer 2073 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock 2074 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt 2075 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock 2076 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt 2077 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event 2078 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event 2079 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event 2080 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt 2081 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit 2082 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit 2083 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit 2084 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit 2085 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit 2086 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle 2087 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit 2088 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle 2089 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit 2090 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit 2091 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit 2092 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit 2093 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz 2094 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock 2095 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz 2096 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update 2097 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz 2098 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock 2099 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit 2100 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit 2101 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit 2102 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel 2103 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel 2104 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18 2105 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave 2106 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16 2107 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer 2108 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram 2109 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event 2110 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event 2111 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event 2112 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel 2113 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2114 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit 2115 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward 2116 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward 2117 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11 2118 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns 2119 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns 2120 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18 2121 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave 2122 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns 2123 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns 2124 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns 2125 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event 2126 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event 2127 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event 2128 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns 2129 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2130 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit 2131 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks 2132 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle 2133 <idle>-0 3.N.. 25us : schedule <-cpu_idle 2134 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule 2135 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule 2136 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule 2137 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch 2138 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch 2139 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule 2140 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq 2141 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule 2142 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task 2143 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task 2144 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt 2145 <idle>-0 3d..3 29us : __schedule <-preempt_schedule 2146 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep 2147 2148This isn't that big of a trace, even with function tracing enabled, 2149so I included the entire trace. 2150 2151The interrupt went off while when the system was idle. Somewhere 2152before task_woken_rt() was called, the NEED_RESCHED flag was set, 2153this is indicated by the first occurrence of the 'N' flag. 2154 2155Latency tracing and events 2156-------------------------- 2157As function tracing can induce a much larger latency, but without 2158seeing what happens within the latency it is hard to know what 2159caused it. There is a middle ground, and that is with enabling 2160events. 2161:: 2162 2163 # echo 0 > options/function-trace 2164 # echo wakeup_rt > current_tracer 2165 # echo 1 > events/enable 2166 # echo 1 > tracing_on 2167 # echo 0 > tracing_max_latency 2168 # chrt -f 5 sleep 1 2169 # echo 0 > tracing_on 2170 # cat trace 2171 # tracer: wakeup_rt 2172 # 2173 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2174 # -------------------------------------------------------------------- 2175 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2176 # ----------------- 2177 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5) 2178 # ----------------- 2179 # 2180 # _------=> CPU# 2181 # / _-----=> irqs-off 2182 # | / _----=> need-resched 2183 # || / _---=> hardirq/softirq 2184 # ||| / _--=> preempt-depth 2185 # |||| / delay 2186 # cmd pid ||||| time | caller 2187 # \ / ||||| \ | / 2188 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep 2189 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up 2190 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002 2191 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8 2192 <idle>-0 2.N.2 2us : power_end: cpu_id=2 2193 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2 2194 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0 2195 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000 2196 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch 2197 <idle>-0 2.N.2 5us : rcu_utilization: End context switch 2198 <idle>-0 2d..3 6us : __schedule <-schedule 2199 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep 2200 2201 2202Hardware Latency Detector 2203------------------------- 2204 2205The hardware latency detector is executed by enabling the "hwlat" tracer. 2206 2207NOTE, this tracer will affect the performance of the system as it will 2208periodically make a CPU constantly busy with interrupts disabled. 2209:: 2210 2211 # echo hwlat > current_tracer 2212 # sleep 100 2213 # cat trace 2214 # tracer: hwlat 2215 # 2216 # entries-in-buffer/entries-written: 13/13 #P:8 2217 # 2218 # _-----=> irqs-off 2219 # / _----=> need-resched 2220 # | / _---=> hardirq/softirq 2221 # || / _--=> preempt-depth 2222 # ||| / delay 2223 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2224 # | | | |||| | | 2225 <...>-1729 [001] d... 678.473449: #1 inner/outer(us): 11/12 ts:1581527483.343962693 count:6 2226 <...>-1729 [004] d... 689.556542: #2 inner/outer(us): 16/9 ts:1581527494.889008092 count:1 2227 <...>-1729 [005] d... 714.756290: #3 inner/outer(us): 16/16 ts:1581527519.678961629 count:5 2228 <...>-1729 [001] d... 718.788247: #4 inner/outer(us): 9/17 ts:1581527523.889012713 count:1 2229 <...>-1729 [002] d... 719.796341: #5 inner/outer(us): 13/9 ts:1581527524.912872606 count:1 2230 <...>-1729 [006] d... 844.787091: #6 inner/outer(us): 9/12 ts:1581527649.889048502 count:2 2231 <...>-1729 [003] d... 849.827033: #7 inner/outer(us): 18/9 ts:1581527654.889013793 count:1 2232 <...>-1729 [007] d... 853.859002: #8 inner/outer(us): 9/12 ts:1581527658.889065736 count:1 2233 <...>-1729 [001] d... 855.874978: #9 inner/outer(us): 9/11 ts:1581527660.861991877 count:1 2234 <...>-1729 [001] d... 863.938932: #10 inner/outer(us): 9/11 ts:1581527668.970010500 count:1 nmi-total:7 nmi-count:1 2235 <...>-1729 [007] d... 878.050780: #11 inner/outer(us): 9/12 ts:1581527683.385002600 count:1 nmi-total:5 nmi-count:1 2236 <...>-1729 [007] d... 886.114702: #12 inner/outer(us): 9/12 ts:1581527691.385001600 count:1 2237 2238 2239The above output is somewhat the same in the header. All events will have 2240interrupts disabled 'd'. Under the FUNCTION title there is: 2241 2242 #1 2243 This is the count of events recorded that were greater than the 2244 tracing_threshold (See below). 2245 2246 inner/outer(us): 11/11 2247 2248 This shows two numbers as "inner latency" and "outer latency". The test 2249 runs in a loop checking a timestamp twice. The latency detected within 2250 the two timestamps is the "inner latency" and the latency detected 2251 after the previous timestamp and the next timestamp in the loop is 2252 the "outer latency". 2253 2254 ts:1581527483.343962693 2255 2256 The absolute timestamp that the first latency was recorded in the window. 2257 2258 count:6 2259 2260 The number of times a latency was detected during the window. 2261 2262 nmi-total:7 nmi-count:1 2263 2264 On architectures that support it, if an NMI comes in during the 2265 test, the time spent in NMI is reported in "nmi-total" (in 2266 microseconds). 2267 2268 All architectures that have NMIs will show the "nmi-count" if an 2269 NMI comes in during the test. 2270 2271hwlat files: 2272 2273 tracing_threshold 2274 This gets automatically set to "10" to represent 10 2275 microseconds. This is the threshold of latency that 2276 needs to be detected before the trace will be recorded. 2277 2278 Note, when hwlat tracer is finished (another tracer is 2279 written into "current_tracer"), the original value for 2280 tracing_threshold is placed back into this file. 2281 2282 hwlat_detector/width 2283 The length of time the test runs with interrupts disabled. 2284 2285 hwlat_detector/window 2286 The length of time of the window which the test 2287 runs. That is, the test will run for "width" 2288 microseconds per "window" microseconds 2289 2290 tracing_cpumask 2291 When the test is started. A kernel thread is created that 2292 runs the test. This thread will alternate between CPUs 2293 listed in the tracing_cpumask between each period 2294 (one "window"). To limit the test to specific CPUs 2295 set the mask in this file to only the CPUs that the test 2296 should run on. 2297 2298function 2299-------- 2300 2301This tracer is the function tracer. Enabling the function tracer 2302can be done from the debug file system. Make sure the 2303ftrace_enabled is set; otherwise this tracer is a nop. 2304See the "ftrace_enabled" section below. 2305:: 2306 2307 # sysctl kernel.ftrace_enabled=1 2308 # echo function > current_tracer 2309 # echo 1 > tracing_on 2310 # usleep 1 2311 # echo 0 > tracing_on 2312 # cat trace 2313 # tracer: function 2314 # 2315 # entries-in-buffer/entries-written: 24799/24799 #P:4 2316 # 2317 # _-----=> irqs-off 2318 # / _----=> need-resched 2319 # | / _---=> hardirq/softirq 2320 # || / _--=> preempt-depth 2321 # ||| / delay 2322 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2323 # | | | |||| | | 2324 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write 2325 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock 2326 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify 2327 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify 2328 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify 2329 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock 2330 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock 2331 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify 2332 [...] 2333 2334 2335Note: function tracer uses ring buffers to store the above 2336entries. The newest data may overwrite the oldest data. 2337Sometimes using echo to stop the trace is not sufficient because 2338the tracing could have overwritten the data that you wanted to 2339record. For this reason, it is sometimes better to disable 2340tracing directly from a program. This allows you to stop the 2341tracing at the point that you hit the part that you are 2342interested in. To disable the tracing directly from a C program, 2343something like following code snippet can be used:: 2344 2345 int trace_fd; 2346 [...] 2347 int main(int argc, char *argv[]) { 2348 [...] 2349 trace_fd = open(tracing_file("tracing_on"), O_WRONLY); 2350 [...] 2351 if (condition_hit()) { 2352 write(trace_fd, "0", 1); 2353 } 2354 [...] 2355 } 2356 2357 2358Single thread tracing 2359--------------------- 2360 2361By writing into set_ftrace_pid you can trace a 2362single thread. For example:: 2363 2364 # cat set_ftrace_pid 2365 no pid 2366 # echo 3111 > set_ftrace_pid 2367 # cat set_ftrace_pid 2368 3111 2369 # echo function > current_tracer 2370 # cat trace | head 2371 # tracer: function 2372 # 2373 # TASK-PID CPU# TIMESTAMP FUNCTION 2374 # | | | | | 2375 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return 2376 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range 2377 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel 2378 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel 2379 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll 2380 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll 2381 # echo > set_ftrace_pid 2382 # cat trace |head 2383 # tracer: function 2384 # 2385 # TASK-PID CPU# TIMESTAMP FUNCTION 2386 # | | | | | 2387 ##### CPU 3 buffer started #### 2388 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait 2389 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry 2390 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry 2391 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit 2392 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit 2393 2394If you want to trace a function when executing, you could use 2395something like this simple program. 2396:: 2397 2398 #include <stdio.h> 2399 #include <stdlib.h> 2400 #include <sys/types.h> 2401 #include <sys/stat.h> 2402 #include <fcntl.h> 2403 #include <unistd.h> 2404 #include <string.h> 2405 2406 #define _STR(x) #x 2407 #define STR(x) _STR(x) 2408 #define MAX_PATH 256 2409 2410 const char *find_tracefs(void) 2411 { 2412 static char tracefs[MAX_PATH+1]; 2413 static int tracefs_found; 2414 char type[100]; 2415 FILE *fp; 2416 2417 if (tracefs_found) 2418 return tracefs; 2419 2420 if ((fp = fopen("/proc/mounts","r")) == NULL) { 2421 perror("/proc/mounts"); 2422 return NULL; 2423 } 2424 2425 while (fscanf(fp, "%*s %" 2426 STR(MAX_PATH) 2427 "s %99s %*s %*d %*d\n", 2428 tracefs, type) == 2) { 2429 if (strcmp(type, "tracefs") == 0) 2430 break; 2431 } 2432 fclose(fp); 2433 2434 if (strcmp(type, "tracefs") != 0) { 2435 fprintf(stderr, "tracefs not mounted"); 2436 return NULL; 2437 } 2438 2439 strcat(tracefs, "/tracing/"); 2440 tracefs_found = 1; 2441 2442 return tracefs; 2443 } 2444 2445 const char *tracing_file(const char *file_name) 2446 { 2447 static char trace_file[MAX_PATH+1]; 2448 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name); 2449 return trace_file; 2450 } 2451 2452 int main (int argc, char **argv) 2453 { 2454 if (argc < 1) 2455 exit(-1); 2456 2457 if (fork() > 0) { 2458 int fd, ffd; 2459 char line[64]; 2460 int s; 2461 2462 ffd = open(tracing_file("current_tracer"), O_WRONLY); 2463 if (ffd < 0) 2464 exit(-1); 2465 write(ffd, "nop", 3); 2466 2467 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY); 2468 s = sprintf(line, "%d\n", getpid()); 2469 write(fd, line, s); 2470 2471 write(ffd, "function", 8); 2472 2473 close(fd); 2474 close(ffd); 2475 2476 execvp(argv[1], argv+1); 2477 } 2478 2479 return 0; 2480 } 2481 2482Or this simple script! 2483:: 2484 2485 #!/bin/bash 2486 2487 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts` 2488 echo 0 > $tracefs/tracing_on 2489 echo $$ > $tracefs/set_ftrace_pid 2490 echo function > $tracefs/current_tracer 2491 echo 1 > $tracefs/tracing_on 2492 exec "$@" 2493 2494 2495function graph tracer 2496--------------------------- 2497 2498This tracer is similar to the function tracer except that it 2499probes a function on its entry and its exit. This is done by 2500using a dynamically allocated stack of return addresses in each 2501task_struct. On function entry the tracer overwrites the return 2502address of each function traced to set a custom probe. Thus the 2503original return address is stored on the stack of return address 2504in the task_struct. 2505 2506Probing on both ends of a function leads to special features 2507such as: 2508 2509- measure of a function's time execution 2510- having a reliable call stack to draw function calls graph 2511 2512This tracer is useful in several situations: 2513 2514- you want to find the reason of a strange kernel behavior and 2515 need to see what happens in detail on any areas (or specific 2516 ones). 2517 2518- you are experiencing weird latencies but it's difficult to 2519 find its origin. 2520 2521- you want to find quickly which path is taken by a specific 2522 function 2523 2524- you just want to peek inside a working kernel and want to see 2525 what happens there. 2526 2527:: 2528 2529 # tracer: function_graph 2530 # 2531 # CPU DURATION FUNCTION CALLS 2532 # | | | | | | | 2533 2534 0) | sys_open() { 2535 0) | do_sys_open() { 2536 0) | getname() { 2537 0) | kmem_cache_alloc() { 2538 0) 1.382 us | __might_sleep(); 2539 0) 2.478 us | } 2540 0) | strncpy_from_user() { 2541 0) | might_fault() { 2542 0) 1.389 us | __might_sleep(); 2543 0) 2.553 us | } 2544 0) 3.807 us | } 2545 0) 7.876 us | } 2546 0) | alloc_fd() { 2547 0) 0.668 us | _spin_lock(); 2548 0) 0.570 us | expand_files(); 2549 0) 0.586 us | _spin_unlock(); 2550 2551 2552There are several columns that can be dynamically 2553enabled/disabled. You can use every combination of options you 2554want, depending on your needs. 2555 2556- The cpu number on which the function executed is default 2557 enabled. It is sometimes better to only trace one cpu (see 2558 tracing_cpu_mask file) or you might sometimes see unordered 2559 function calls while cpu tracing switch. 2560 2561 - hide: echo nofuncgraph-cpu > trace_options 2562 - show: echo funcgraph-cpu > trace_options 2563 2564- The duration (function's time of execution) is displayed on 2565 the closing bracket line of a function or on the same line 2566 than the current function in case of a leaf one. It is default 2567 enabled. 2568 2569 - hide: echo nofuncgraph-duration > trace_options 2570 - show: echo funcgraph-duration > trace_options 2571 2572- The overhead field precedes the duration field in case of 2573 reached duration thresholds. 2574 2575 - hide: echo nofuncgraph-overhead > trace_options 2576 - show: echo funcgraph-overhead > trace_options 2577 - depends on: funcgraph-duration 2578 2579 ie:: 2580 2581 3) # 1837.709 us | } /* __switch_to */ 2582 3) | finish_task_switch() { 2583 3) 0.313 us | _raw_spin_unlock_irq(); 2584 3) 3.177 us | } 2585 3) # 1889.063 us | } /* __schedule */ 2586 3) ! 140.417 us | } /* __schedule */ 2587 3) # 2034.948 us | } /* schedule */ 2588 3) * 33998.59 us | } /* schedule_preempt_disabled */ 2589 2590 [...] 2591 2592 1) 0.260 us | msecs_to_jiffies(); 2593 1) 0.313 us | __rcu_read_unlock(); 2594 1) + 61.770 us | } 2595 1) + 64.479 us | } 2596 1) 0.313 us | rcu_bh_qs(); 2597 1) 0.313 us | __local_bh_enable(); 2598 1) ! 217.240 us | } 2599 1) 0.365 us | idle_cpu(); 2600 1) | rcu_irq_exit() { 2601 1) 0.417 us | rcu_eqs_enter_common.isra.47(); 2602 1) 3.125 us | } 2603 1) ! 227.812 us | } 2604 1) ! 457.395 us | } 2605 1) @ 119760.2 us | } 2606 2607 [...] 2608 2609 2) | handle_IPI() { 2610 1) 6.979 us | } 2611 2) 0.417 us | scheduler_ipi(); 2612 1) 9.791 us | } 2613 1) + 12.917 us | } 2614 2) 3.490 us | } 2615 1) + 15.729 us | } 2616 1) + 18.542 us | } 2617 2) $ 3594274 us | } 2618 2619Flags:: 2620 2621 + means that the function exceeded 10 usecs. 2622 ! means that the function exceeded 100 usecs. 2623 # means that the function exceeded 1000 usecs. 2624 * means that the function exceeded 10 msecs. 2625 @ means that the function exceeded 100 msecs. 2626 $ means that the function exceeded 1 sec. 2627 2628 2629- The task/pid field displays the thread cmdline and pid which 2630 executed the function. It is default disabled. 2631 2632 - hide: echo nofuncgraph-proc > trace_options 2633 - show: echo funcgraph-proc > trace_options 2634 2635 ie:: 2636 2637 # tracer: function_graph 2638 # 2639 # CPU TASK/PID DURATION FUNCTION CALLS 2640 # | | | | | | | | | 2641 0) sh-4802 | | d_free() { 2642 0) sh-4802 | | call_rcu() { 2643 0) sh-4802 | | __call_rcu() { 2644 0) sh-4802 | 0.616 us | rcu_process_gp_end(); 2645 0) sh-4802 | 0.586 us | check_for_new_grace_period(); 2646 0) sh-4802 | 2.899 us | } 2647 0) sh-4802 | 4.040 us | } 2648 0) sh-4802 | 5.151 us | } 2649 0) sh-4802 | + 49.370 us | } 2650 2651 2652- The absolute time field is an absolute timestamp given by the 2653 system clock since it started. A snapshot of this time is 2654 given on each entry/exit of functions 2655 2656 - hide: echo nofuncgraph-abstime > trace_options 2657 - show: echo funcgraph-abstime > trace_options 2658 2659 ie:: 2660 2661 # 2662 # TIME CPU DURATION FUNCTION CALLS 2663 # | | | | | | | | 2664 360.774522 | 1) 0.541 us | } 2665 360.774522 | 1) 4.663 us | } 2666 360.774523 | 1) 0.541 us | __wake_up_bit(); 2667 360.774524 | 1) 6.796 us | } 2668 360.774524 | 1) 7.952 us | } 2669 360.774525 | 1) 9.063 us | } 2670 360.774525 | 1) 0.615 us | journal_mark_dirty(); 2671 360.774527 | 1) 0.578 us | __brelse(); 2672 360.774528 | 1) | reiserfs_prepare_for_journal() { 2673 360.774528 | 1) | unlock_buffer() { 2674 360.774529 | 1) | wake_up_bit() { 2675 360.774529 | 1) | bit_waitqueue() { 2676 360.774530 | 1) 0.594 us | __phys_addr(); 2677 2678 2679The function name is always displayed after the closing bracket 2680for a function if the start of that function is not in the 2681trace buffer. 2682 2683Display of the function name after the closing bracket may be 2684enabled for functions whose start is in the trace buffer, 2685allowing easier searching with grep for function durations. 2686It is default disabled. 2687 2688 - hide: echo nofuncgraph-tail > trace_options 2689 - show: echo funcgraph-tail > trace_options 2690 2691 Example with nofuncgraph-tail (default):: 2692 2693 0) | putname() { 2694 0) | kmem_cache_free() { 2695 0) 0.518 us | __phys_addr(); 2696 0) 1.757 us | } 2697 0) 2.861 us | } 2698 2699 Example with funcgraph-tail:: 2700 2701 0) | putname() { 2702 0) | kmem_cache_free() { 2703 0) 0.518 us | __phys_addr(); 2704 0) 1.757 us | } /* kmem_cache_free() */ 2705 0) 2.861 us | } /* putname() */ 2706 2707You can put some comments on specific functions by using 2708trace_printk() For example, if you want to put a comment inside 2709the __might_sleep() function, you just have to include 2710<linux/ftrace.h> and call trace_printk() inside __might_sleep():: 2711 2712 trace_printk("I'm a comment!\n") 2713 2714will produce:: 2715 2716 1) | __might_sleep() { 2717 1) | /* I'm a comment! */ 2718 1) 1.449 us | } 2719 2720 2721You might find other useful features for this tracer in the 2722following "dynamic ftrace" section such as tracing only specific 2723functions or tasks. 2724 2725dynamic ftrace 2726-------------- 2727 2728If CONFIG_DYNAMIC_FTRACE is set, the system will run with 2729virtually no overhead when function tracing is disabled. The way 2730this works is the mcount function call (placed at the start of 2731every kernel function, produced by the -pg switch in gcc), 2732starts of pointing to a simple return. (Enabling FTRACE will 2733include the -pg switch in the compiling of the kernel.) 2734 2735At compile time every C file object is run through the 2736recordmcount program (located in the scripts directory). This 2737program will parse the ELF headers in the C object to find all 2738the locations in the .text section that call mcount. Starting 2739with gcc version 4.6, the -mfentry has been added for x86, which 2740calls "__fentry__" instead of "mcount". Which is called before 2741the creation of the stack frame. 2742 2743Note, not all sections are traced. They may be prevented by either 2744a notrace, or blocked another way and all inline functions are not 2745traced. Check the "available_filter_functions" file to see what functions 2746can be traced. 2747 2748A section called "__mcount_loc" is created that holds 2749references to all the mcount/fentry call sites in the .text section. 2750The recordmcount program re-links this section back into the 2751original object. The final linking stage of the kernel will add all these 2752references into a single table. 2753 2754On boot up, before SMP is initialized, the dynamic ftrace code 2755scans this table and updates all the locations into nops. It 2756also records the locations, which are added to the 2757available_filter_functions list. Modules are processed as they 2758are loaded and before they are executed. When a module is 2759unloaded, it also removes its functions from the ftrace function 2760list. This is automatic in the module unload code, and the 2761module author does not need to worry about it. 2762 2763When tracing is enabled, the process of modifying the function 2764tracepoints is dependent on architecture. The old method is to use 2765kstop_machine to prevent races with the CPUs executing code being 2766modified (which can cause the CPU to do undesirable things, especially 2767if the modified code crosses cache (or page) boundaries), and the nops are 2768patched back to calls. But this time, they do not call mcount 2769(which is just a function stub). They now call into the ftrace 2770infrastructure. 2771 2772The new method of modifying the function tracepoints is to place 2773a breakpoint at the location to be modified, sync all CPUs, modify 2774the rest of the instruction not covered by the breakpoint. Sync 2775all CPUs again, and then remove the breakpoint with the finished 2776version to the ftrace call site. 2777 2778Some archs do not even need to monkey around with the synchronization, 2779and can just slap the new code on top of the old without any 2780problems with other CPUs executing it at the same time. 2781 2782One special side-effect to the recording of the functions being 2783traced is that we can now selectively choose which functions we 2784wish to trace and which ones we want the mcount calls to remain 2785as nops. 2786 2787Two files are used, one for enabling and one for disabling the 2788tracing of specified functions. They are: 2789 2790 set_ftrace_filter 2791 2792and 2793 2794 set_ftrace_notrace 2795 2796A list of available functions that you can add to these files is 2797listed in: 2798 2799 available_filter_functions 2800 2801:: 2802 2803 # cat available_filter_functions 2804 put_prev_task_idle 2805 kmem_cache_create 2806 pick_next_task_rt 2807 cpus_read_lock 2808 pick_next_task_fair 2809 mutex_lock 2810 [...] 2811 2812If I am only interested in sys_nanosleep and hrtimer_interrupt:: 2813 2814 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter 2815 # echo function > current_tracer 2816 # echo 1 > tracing_on 2817 # usleep 1 2818 # echo 0 > tracing_on 2819 # cat trace 2820 # tracer: function 2821 # 2822 # entries-in-buffer/entries-written: 5/5 #P:4 2823 # 2824 # _-----=> irqs-off 2825 # / _----=> need-resched 2826 # | / _---=> hardirq/softirq 2827 # || / _--=> preempt-depth 2828 # ||| / delay 2829 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2830 # | | | |||| | | 2831 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath 2832 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt 2833 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt 2834 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt 2835 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt 2836 2837To see which functions are being traced, you can cat the file: 2838:: 2839 2840 # cat set_ftrace_filter 2841 hrtimer_interrupt 2842 sys_nanosleep 2843 2844 2845Perhaps this is not enough. The filters also allow glob(7) matching. 2846 2847 ``<match>*`` 2848 will match functions that begin with <match> 2849 ``*<match>`` 2850 will match functions that end with <match> 2851 ``*<match>*`` 2852 will match functions that have <match> in it 2853 ``<match1>*<match2>`` 2854 will match functions that begin with <match1> and end with <match2> 2855 2856.. note:: 2857 It is better to use quotes to enclose the wild cards, 2858 otherwise the shell may expand the parameters into names 2859 of files in the local directory. 2860 2861:: 2862 2863 # echo 'hrtimer_*' > set_ftrace_filter 2864 2865Produces:: 2866 2867 # tracer: function 2868 # 2869 # entries-in-buffer/entries-written: 897/897 #P:4 2870 # 2871 # _-----=> irqs-off 2872 # / _----=> need-resched 2873 # | / _---=> hardirq/softirq 2874 # || / _--=> preempt-depth 2875 # ||| / delay 2876 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2877 # | | | |||| | | 2878 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit 2879 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel 2880 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer 2881 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit 2882 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11 2883 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt 2884 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter 2885 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem 2886 2887Notice that we lost the sys_nanosleep. 2888:: 2889 2890 # cat set_ftrace_filter 2891 hrtimer_run_queues 2892 hrtimer_run_pending 2893 hrtimer_init 2894 hrtimer_cancel 2895 hrtimer_try_to_cancel 2896 hrtimer_forward 2897 hrtimer_start 2898 hrtimer_reprogram 2899 hrtimer_force_reprogram 2900 hrtimer_get_next_event 2901 hrtimer_interrupt 2902 hrtimer_nanosleep 2903 hrtimer_wakeup 2904 hrtimer_get_remaining 2905 hrtimer_get_res 2906 hrtimer_init_sleeper 2907 2908 2909This is because the '>' and '>>' act just like they do in bash. 2910To rewrite the filters, use '>' 2911To append to the filters, use '>>' 2912 2913To clear out a filter so that all functions will be recorded 2914again:: 2915 2916 # echo > set_ftrace_filter 2917 # cat set_ftrace_filter 2918 # 2919 2920Again, now we want to append. 2921 2922:: 2923 2924 # echo sys_nanosleep > set_ftrace_filter 2925 # cat set_ftrace_filter 2926 sys_nanosleep 2927 # echo 'hrtimer_*' >> set_ftrace_filter 2928 # cat set_ftrace_filter 2929 hrtimer_run_queues 2930 hrtimer_run_pending 2931 hrtimer_init 2932 hrtimer_cancel 2933 hrtimer_try_to_cancel 2934 hrtimer_forward 2935 hrtimer_start 2936 hrtimer_reprogram 2937 hrtimer_force_reprogram 2938 hrtimer_get_next_event 2939 hrtimer_interrupt 2940 sys_nanosleep 2941 hrtimer_nanosleep 2942 hrtimer_wakeup 2943 hrtimer_get_remaining 2944 hrtimer_get_res 2945 hrtimer_init_sleeper 2946 2947 2948The set_ftrace_notrace prevents those functions from being 2949traced. 2950:: 2951 2952 # echo '*preempt*' '*lock*' > set_ftrace_notrace 2953 2954Produces:: 2955 2956 # tracer: function 2957 # 2958 # entries-in-buffer/entries-written: 39608/39608 #P:4 2959 # 2960 # _-----=> irqs-off 2961 # / _----=> need-resched 2962 # | / _---=> hardirq/softirq 2963 # || / _--=> preempt-depth 2964 # ||| / delay 2965 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2966 # | | | |||| | | 2967 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open 2968 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last 2969 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last 2970 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check 2971 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement 2972 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action 2973 bash-1994 [000] .... 4342.324899: do_truncate <-do_last 2974 bash-1994 [000] .... 4342.324899: setattr_should_drop_suidgid <-do_truncate 2975 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate 2976 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change 2977 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time 2978 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time 2979 2980We can see that there's no more lock or preempt tracing. 2981 2982Selecting function filters via index 2983------------------------------------ 2984 2985Because processing of strings is expensive (the address of the function 2986needs to be looked up before comparing to the string being passed in), 2987an index can be used as well to enable functions. This is useful in the 2988case of setting thousands of specific functions at a time. By passing 2989in a list of numbers, no string processing will occur. Instead, the function 2990at the specific location in the internal array (which corresponds to the 2991functions in the "available_filter_functions" file), is selected. 2992 2993:: 2994 2995 # echo 1 > set_ftrace_filter 2996 2997Will select the first function listed in "available_filter_functions" 2998 2999:: 3000 3001 # head -1 available_filter_functions 3002 trace_initcall_finish_cb 3003 3004 # cat set_ftrace_filter 3005 trace_initcall_finish_cb 3006 3007 # head -50 available_filter_functions | tail -1 3008 x86_pmu_commit_txn 3009 3010 # echo 1 50 > set_ftrace_filter 3011 # cat set_ftrace_filter 3012 trace_initcall_finish_cb 3013 x86_pmu_commit_txn 3014 3015Dynamic ftrace with the function graph tracer 3016--------------------------------------------- 3017 3018Although what has been explained above concerns both the 3019function tracer and the function-graph-tracer, there are some 3020special features only available in the function-graph tracer. 3021 3022If you want to trace only one function and all of its children, 3023you just have to echo its name into set_graph_function:: 3024 3025 echo __do_fault > set_graph_function 3026 3027will produce the following "expanded" trace of the __do_fault() 3028function:: 3029 3030 0) | __do_fault() { 3031 0) | filemap_fault() { 3032 0) | find_lock_page() { 3033 0) 0.804 us | find_get_page(); 3034 0) | __might_sleep() { 3035 0) 1.329 us | } 3036 0) 3.904 us | } 3037 0) 4.979 us | } 3038 0) 0.653 us | _spin_lock(); 3039 0) 0.578 us | page_add_file_rmap(); 3040 0) 0.525 us | native_set_pte_at(); 3041 0) 0.585 us | _spin_unlock(); 3042 0) | unlock_page() { 3043 0) 0.541 us | page_waitqueue(); 3044 0) 0.639 us | __wake_up_bit(); 3045 0) 2.786 us | } 3046 0) + 14.237 us | } 3047 0) | __do_fault() { 3048 0) | filemap_fault() { 3049 0) | find_lock_page() { 3050 0) 0.698 us | find_get_page(); 3051 0) | __might_sleep() { 3052 0) 1.412 us | } 3053 0) 3.950 us | } 3054 0) 5.098 us | } 3055 0) 0.631 us | _spin_lock(); 3056 0) 0.571 us | page_add_file_rmap(); 3057 0) 0.526 us | native_set_pte_at(); 3058 0) 0.586 us | _spin_unlock(); 3059 0) | unlock_page() { 3060 0) 0.533 us | page_waitqueue(); 3061 0) 0.638 us | __wake_up_bit(); 3062 0) 2.793 us | } 3063 0) + 14.012 us | } 3064 3065You can also expand several functions at once:: 3066 3067 echo sys_open > set_graph_function 3068 echo sys_close >> set_graph_function 3069 3070Now if you want to go back to trace all functions you can clear 3071this special filter via:: 3072 3073 echo > set_graph_function 3074 3075 3076ftrace_enabled 3077-------------- 3078 3079Note, the proc sysctl ftrace_enable is a big on/off switch for the 3080function tracer. By default it is enabled (when function tracing is 3081enabled in the kernel). If it is disabled, all function tracing is 3082disabled. This includes not only the function tracers for ftrace, but 3083also for any other uses (perf, kprobes, stack tracing, profiling, etc). It 3084cannot be disabled if there is a callback with FTRACE_OPS_FL_PERMANENT set 3085registered. 3086 3087Please disable this with care. 3088 3089This can be disable (and enabled) with:: 3090 3091 sysctl kernel.ftrace_enabled=0 3092 sysctl kernel.ftrace_enabled=1 3093 3094 or 3095 3096 echo 0 > /proc/sys/kernel/ftrace_enabled 3097 echo 1 > /proc/sys/kernel/ftrace_enabled 3098 3099 3100Filter commands 3101--------------- 3102 3103A few commands are supported by the set_ftrace_filter interface. 3104Trace commands have the following format:: 3105 3106 <function>:<command>:<parameter> 3107 3108The following commands are supported: 3109 3110- mod: 3111 This command enables function filtering per module. The 3112 parameter defines the module. For example, if only the write* 3113 functions in the ext3 module are desired, run: 3114 3115 echo 'write*:mod:ext3' > set_ftrace_filter 3116 3117 This command interacts with the filter in the same way as 3118 filtering based on function names. Thus, adding more functions 3119 in a different module is accomplished by appending (>>) to the 3120 filter file. Remove specific module functions by prepending 3121 '!':: 3122 3123 echo '!writeback*:mod:ext3' >> set_ftrace_filter 3124 3125 Mod command supports module globbing. Disable tracing for all 3126 functions except a specific module:: 3127 3128 echo '!*:mod:!ext3' >> set_ftrace_filter 3129 3130 Disable tracing for all modules, but still trace kernel:: 3131 3132 echo '!*:mod:*' >> set_ftrace_filter 3133 3134 Enable filter only for kernel:: 3135 3136 echo '*write*:mod:!*' >> set_ftrace_filter 3137 3138 Enable filter for module globbing:: 3139 3140 echo '*write*:mod:*snd*' >> set_ftrace_filter 3141 3142- traceon/traceoff: 3143 These commands turn tracing on and off when the specified 3144 functions are hit. The parameter determines how many times the 3145 tracing system is turned on and off. If unspecified, there is 3146 no limit. For example, to disable tracing when a schedule bug 3147 is hit the first 5 times, run:: 3148 3149 echo '__schedule_bug:traceoff:5' > set_ftrace_filter 3150 3151 To always disable tracing when __schedule_bug is hit:: 3152 3153 echo '__schedule_bug:traceoff' > set_ftrace_filter 3154 3155 These commands are cumulative whether or not they are appended 3156 to set_ftrace_filter. To remove a command, prepend it by '!' 3157 and drop the parameter:: 3158 3159 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter 3160 3161 The above removes the traceoff command for __schedule_bug 3162 that have a counter. To remove commands without counters:: 3163 3164 echo '!__schedule_bug:traceoff' > set_ftrace_filter 3165 3166- snapshot: 3167 Will cause a snapshot to be triggered when the function is hit. 3168 :: 3169 3170 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter 3171 3172 To only snapshot once: 3173 :: 3174 3175 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter 3176 3177 To remove the above commands:: 3178 3179 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter 3180 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter 3181 3182- enable_event/disable_event: 3183 These commands can enable or disable a trace event. Note, because 3184 function tracing callbacks are very sensitive, when these commands 3185 are registered, the trace point is activated, but disabled in 3186 a "soft" mode. That is, the tracepoint will be called, but 3187 just will not be traced. The event tracepoint stays in this mode 3188 as long as there's a command that triggers it. 3189 :: 3190 3191 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \ 3192 set_ftrace_filter 3193 3194 The format is:: 3195 3196 <function>:enable_event:<system>:<event>[:count] 3197 <function>:disable_event:<system>:<event>[:count] 3198 3199 To remove the events commands:: 3200 3201 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \ 3202 set_ftrace_filter 3203 echo '!schedule:disable_event:sched:sched_switch' > \ 3204 set_ftrace_filter 3205 3206- dump: 3207 When the function is hit, it will dump the contents of the ftrace 3208 ring buffer to the console. This is useful if you need to debug 3209 something, and want to dump the trace when a certain function 3210 is hit. Perhaps it's a function that is called before a triple 3211 fault happens and does not allow you to get a regular dump. 3212 3213- cpudump: 3214 When the function is hit, it will dump the contents of the ftrace 3215 ring buffer for the current CPU to the console. Unlike the "dump" 3216 command, it only prints out the contents of the ring buffer for the 3217 CPU that executed the function that triggered the dump. 3218 3219- stacktrace: 3220 When the function is hit, a stack trace is recorded. 3221 3222trace_pipe 3223---------- 3224 3225The trace_pipe outputs the same content as the trace file, but 3226the effect on the tracing is different. Every read from 3227trace_pipe is consumed. This means that subsequent reads will be 3228different. The trace is live. 3229:: 3230 3231 # echo function > current_tracer 3232 # cat trace_pipe > /tmp/trace.out & 3233 [1] 4153 3234 # echo 1 > tracing_on 3235 # usleep 1 3236 # echo 0 > tracing_on 3237 # cat trace 3238 # tracer: function 3239 # 3240 # entries-in-buffer/entries-written: 0/0 #P:4 3241 # 3242 # _-----=> irqs-off 3243 # / _----=> need-resched 3244 # | / _---=> hardirq/softirq 3245 # || / _--=> preempt-depth 3246 # ||| / delay 3247 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3248 # | | | |||| | | 3249 3250 # 3251 # cat /tmp/trace.out 3252 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write 3253 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock 3254 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify 3255 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify 3256 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify 3257 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock 3258 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock 3259 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify 3260 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath 3261 3262 3263Note, reading the trace_pipe file will block until more input is 3264added. This is contrary to the trace file. If any process opened 3265the trace file for reading, it will actually disable tracing and 3266prevent new entries from being added. The trace_pipe file does 3267not have this limitation. 3268 3269trace entries 3270------------- 3271 3272Having too much or not enough data can be troublesome in 3273diagnosing an issue in the kernel. The file buffer_size_kb is 3274used to modify the size of the internal trace buffers. The 3275number listed is the number of entries that can be recorded per 3276CPU. To know the full size, multiply the number of possible CPUs 3277with the number of entries. 3278:: 3279 3280 # cat buffer_size_kb 3281 1408 (units kilobytes) 3282 3283Or simply read buffer_total_size_kb 3284:: 3285 3286 # cat buffer_total_size_kb 3287 5632 3288 3289To modify the buffer, simple echo in a number (in 1024 byte segments). 3290:: 3291 3292 # echo 10000 > buffer_size_kb 3293 # cat buffer_size_kb 3294 10000 (units kilobytes) 3295 3296It will try to allocate as much as possible. If you allocate too 3297much, it can cause Out-Of-Memory to trigger. 3298:: 3299 3300 # echo 1000000000000 > buffer_size_kb 3301 -bash: echo: write error: Cannot allocate memory 3302 # cat buffer_size_kb 3303 85 3304 3305The per_cpu buffers can be changed individually as well: 3306:: 3307 3308 # echo 10000 > per_cpu/cpu0/buffer_size_kb 3309 # echo 100 > per_cpu/cpu1/buffer_size_kb 3310 3311When the per_cpu buffers are not the same, the buffer_size_kb 3312at the top level will just show an X 3313:: 3314 3315 # cat buffer_size_kb 3316 X 3317 3318This is where the buffer_total_size_kb is useful: 3319:: 3320 3321 # cat buffer_total_size_kb 3322 12916 3323 3324Writing to the top level buffer_size_kb will reset all the buffers 3325to be the same again. 3326 3327Snapshot 3328-------- 3329CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature 3330available to all non latency tracers. (Latency tracers which 3331record max latency, such as "irqsoff" or "wakeup", can't use 3332this feature, since those are already using the snapshot 3333mechanism internally.) 3334 3335Snapshot preserves a current trace buffer at a particular point 3336in time without stopping tracing. Ftrace swaps the current 3337buffer with a spare buffer, and tracing continues in the new 3338current (=previous spare) buffer. 3339 3340The following tracefs files in "tracing" are related to this 3341feature: 3342 3343 snapshot: 3344 3345 This is used to take a snapshot and to read the output 3346 of the snapshot. Echo 1 into this file to allocate a 3347 spare buffer and to take a snapshot (swap), then read 3348 the snapshot from this file in the same format as 3349 "trace" (described above in the section "The File 3350 System"). Both reads snapshot and tracing are executable 3351 in parallel. When the spare buffer is allocated, echoing 3352 0 frees it, and echoing else (positive) values clear the 3353 snapshot contents. 3354 More details are shown in the table below. 3355 3356 +--------------+------------+------------+------------+ 3357 |status\\input | 0 | 1 | else | 3358 +==============+============+============+============+ 3359 |not allocated |(do nothing)| alloc+swap |(do nothing)| 3360 +--------------+------------+------------+------------+ 3361 |allocated | free | swap | clear | 3362 +--------------+------------+------------+------------+ 3363 3364Here is an example of using the snapshot feature. 3365:: 3366 3367 # echo 1 > events/sched/enable 3368 # echo 1 > snapshot 3369 # cat snapshot 3370 # tracer: nop 3371 # 3372 # entries-in-buffer/entries-written: 71/71 #P:8 3373 # 3374 # _-----=> irqs-off 3375 # / _----=> need-resched 3376 # | / _---=> hardirq/softirq 3377 # || / _--=> preempt-depth 3378 # ||| / delay 3379 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3380 # | | | |||| | | 3381 <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 3382 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 3383 [...] 3384 <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 3385 3386 # cat trace 3387 # tracer: nop 3388 # 3389 # entries-in-buffer/entries-written: 77/77 #P:8 3390 # 3391 # _-----=> irqs-off 3392 # / _----=> need-resched 3393 # | / _---=> hardirq/softirq 3394 # || / _--=> preempt-depth 3395 # ||| / delay 3396 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3397 # | | | |||| | | 3398 <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 3399 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 3400 [...] 3401 3402 3403If you try to use this snapshot feature when current tracer is 3404one of the latency tracers, you will get the following results. 3405:: 3406 3407 # echo wakeup > current_tracer 3408 # echo 1 > snapshot 3409 bash: echo: write error: Device or resource busy 3410 # cat snapshot 3411 cat: snapshot: Device or resource busy 3412 3413 3414Instances 3415--------- 3416In the tracefs tracing directory, there is a directory called "instances". 3417This directory can have new directories created inside of it using 3418mkdir, and removing directories with rmdir. The directory created 3419with mkdir in this directory will already contain files and other 3420directories after it is created. 3421:: 3422 3423 # mkdir instances/foo 3424 # ls instances/foo 3425 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu 3426 set_event snapshot trace trace_clock trace_marker trace_options 3427 trace_pipe tracing_on 3428 3429As you can see, the new directory looks similar to the tracing directory 3430itself. In fact, it is very similar, except that the buffer and 3431events are agnostic from the main directory, or from any other 3432instances that are created. 3433 3434The files in the new directory work just like the files with the 3435same name in the tracing directory except the buffer that is used 3436is a separate and new buffer. The files affect that buffer but do not 3437affect the main buffer with the exception of trace_options. Currently, 3438the trace_options affect all instances and the top level buffer 3439the same, but this may change in future releases. That is, options 3440may become specific to the instance they reside in. 3441 3442Notice that none of the function tracer files are there, nor is 3443current_tracer and available_tracers. This is because the buffers 3444can currently only have events enabled for them. 3445:: 3446 3447 # mkdir instances/foo 3448 # mkdir instances/bar 3449 # mkdir instances/zoot 3450 # echo 100000 > buffer_size_kb 3451 # echo 1000 > instances/foo/buffer_size_kb 3452 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb 3453 # echo function > current_trace 3454 # echo 1 > instances/foo/events/sched/sched_wakeup/enable 3455 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable 3456 # echo 1 > instances/foo/events/sched/sched_switch/enable 3457 # echo 1 > instances/bar/events/irq/enable 3458 # echo 1 > instances/zoot/events/syscalls/enable 3459 # cat trace_pipe 3460 CPU:2 [LOST 11745 EVENTS] 3461 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist 3462 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave 3463 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist 3464 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist 3465 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock 3466 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype 3467 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist 3468 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist 3469 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics 3470 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics 3471 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process 3472 [...] 3473 3474 # cat instances/foo/trace_pipe 3475 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000 3476 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000 3477 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003 3478 <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 3479 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 3480 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000 3481 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000 3482 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 3483 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001 3484 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 3485 [...] 3486 3487 # cat instances/bar/trace_pipe 3488 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX] 3489 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX] 3490 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER] 3491 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU] 3492 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER] 3493 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER] 3494 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU] 3495 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU] 3496 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4 3497 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled 3498 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0 3499 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled 3500 [...] 3501 3502 # cat instances/zoot/trace 3503 # tracer: nop 3504 # 3505 # entries-in-buffer/entries-written: 18996/18996 #P:4 3506 # 3507 # _-----=> irqs-off 3508 # / _----=> need-resched 3509 # | / _---=> hardirq/softirq 3510 # || / _--=> preempt-depth 3511 # ||| / delay 3512 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3513 # | | | |||| | | 3514 bash-1998 [000] d... 140.733501: sys_write -> 0x2 3515 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1) 3516 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1 3517 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0) 3518 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1 3519 bash-1998 [000] d... 140.733510: sys_close(fd: a) 3520 bash-1998 [000] d... 140.733510: sys_close -> 0x0 3521 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8) 3522 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0 3523 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8) 3524 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0 3525 3526You can see that the trace of the top most trace buffer shows only 3527the function tracing. The foo instance displays wakeups and task 3528switches. 3529 3530To remove the instances, simply delete their directories: 3531:: 3532 3533 # rmdir instances/foo 3534 # rmdir instances/bar 3535 # rmdir instances/zoot 3536 3537Note, if a process has a trace file open in one of the instance 3538directories, the rmdir will fail with EBUSY. 3539 3540 3541Stack trace 3542----------- 3543Since the kernel has a fixed sized stack, it is important not to 3544waste it in functions. A kernel developer must be conscious of 3545what they allocate on the stack. If they add too much, the system 3546can be in danger of a stack overflow, and corruption will occur, 3547usually leading to a system panic. 3548 3549There are some tools that check this, usually with interrupts 3550periodically checking usage. But if you can perform a check 3551at every function call that will become very useful. As ftrace provides 3552a function tracer, it makes it convenient to check the stack size 3553at every function call. This is enabled via the stack tracer. 3554 3555CONFIG_STACK_TRACER enables the ftrace stack tracing functionality. 3556To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled. 3557:: 3558 3559 # echo 1 > /proc/sys/kernel/stack_tracer_enabled 3560 3561You can also enable it from the kernel command line to trace 3562the stack size of the kernel during boot up, by adding "stacktrace" 3563to the kernel command line parameter. 3564 3565After running it for a few minutes, the output looks like: 3566:: 3567 3568 # cat stack_max_size 3569 2928 3570 3571 # cat stack_trace 3572 Depth Size Location (18 entries) 3573 ----- ---- -------- 3574 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac 3575 1) 2704 160 find_busiest_group+0x31/0x1f1 3576 2) 2544 256 load_balance+0xd9/0x662 3577 3) 2288 80 idle_balance+0xbb/0x130 3578 4) 2208 128 __schedule+0x26e/0x5b9 3579 5) 2080 16 schedule+0x64/0x66 3580 6) 2064 128 schedule_timeout+0x34/0xe0 3581 7) 1936 112 wait_for_common+0x97/0xf1 3582 8) 1824 16 wait_for_completion+0x1d/0x1f 3583 9) 1808 128 flush_work+0xfe/0x119 3584 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20 3585 11) 1664 48 input_available_p+0x1d/0x5c 3586 12) 1616 48 n_tty_poll+0x6d/0x134 3587 13) 1568 64 tty_poll+0x64/0x7f 3588 14) 1504 880 do_select+0x31e/0x511 3589 15) 624 400 core_sys_select+0x177/0x216 3590 16) 224 96 sys_select+0x91/0xb9 3591 17) 128 128 system_call_fastpath+0x16/0x1b 3592 3593Note, if -mfentry is being used by gcc, functions get traced before 3594they set up the stack frame. This means that leaf level functions 3595are not tested by the stack tracer when -mfentry is used. 3596 3597Currently, -mfentry is used by gcc 4.6.0 and above on x86 only. 3598 3599More 3600---- 3601More details can be found in the source code, in the `kernel/trace/*.c` files. 3602