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 available_filter_functions_addrs: 328 329 Similar to available_filter_functions, but with address displayed 330 for each function. The displayed address is the patch-site address 331 and can differ from /proc/kallsyms address. 332 333 dyn_ftrace_total_info: 334 335 This file is for debugging purposes. The number of functions that 336 have been converted to nops and are available to be traced. 337 338 enabled_functions: 339 340 This file is more for debugging ftrace, but can also be useful 341 in seeing if any function has a callback attached to it. 342 Not only does the trace infrastructure use ftrace function 343 trace utility, but other subsystems might too. This file 344 displays all functions that have a callback attached to them 345 as well as the number of callbacks that have been attached. 346 Note, a callback may also call multiple functions which will 347 not be listed in this count. 348 349 If the callback registered to be traced by a function with 350 the "save regs" attribute (thus even more overhead), a 'R' 351 will be displayed on the same line as the function that 352 is returning registers. 353 354 If the callback registered to be traced by a function with 355 the "ip modify" attribute (thus the regs->ip can be changed), 356 an 'I' will be displayed on the same line as the function that 357 can be overridden. 358 359 If a non ftrace trampoline is attached (BPF) a 'D' will be displayed. 360 Note, normal ftrace trampolines can also be attached, but only one 361 "direct" trampoline can be attached to a given function at a time. 362 363 Some architectures can not call direct trampolines, but instead have 364 the ftrace ops function located above the function entry point. In 365 such cases an 'O' will be displayed. 366 367 If a function had either the "ip modify" or a "direct" call attached to 368 it in the past, a 'M' will be shown. This flag is never cleared. It is 369 used to know if a function was every modified by the ftrace infrastructure, 370 and can be used for debugging. 371 372 If the architecture supports it, it will also show what callback 373 is being directly called by the function. If the count is greater 374 than 1 it most likely will be ftrace_ops_list_func(). 375 376 If the callback of a function jumps to a trampoline that is 377 specific to the callback and which is not the standard trampoline, 378 its address will be printed as well as the function that the 379 trampoline calls. 380 381 touched_functions: 382 383 This file contains all the functions that ever had a function callback 384 to it via the ftrace infrastructure. It has the same format as 385 enabled_functions but shows all functions that have every been 386 traced. 387 388 To see any function that has every been modified by "ip modify" or a 389 direct trampoline, one can perform the following command: 390 391 grep ' M ' /sys/kernel/tracing/touched_functions 392 393 function_profile_enabled: 394 395 When set it will enable all functions with either the function 396 tracer, or if configured, the function graph tracer. It will 397 keep a histogram of the number of functions that were called 398 and if the function graph tracer was configured, it will also keep 399 track of the time spent in those functions. The histogram 400 content can be displayed in the files: 401 402 trace_stat/function<cpu> ( function0, function1, etc). 403 404 trace_stat: 405 406 A directory that holds different tracing stats. 407 408 kprobe_events: 409 410 Enable dynamic trace points. See kprobetrace.rst. 411 412 kprobe_profile: 413 414 Dynamic trace points stats. See kprobetrace.rst. 415 416 max_graph_depth: 417 418 Used with the function graph tracer. This is the max depth 419 it will trace into a function. Setting this to a value of 420 one will show only the first kernel function that is called 421 from user space. 422 423 printk_formats: 424 425 This is for tools that read the raw format files. If an event in 426 the ring buffer references a string, only a pointer to the string 427 is recorded into the buffer and not the string itself. This prevents 428 tools from knowing what that string was. This file displays the string 429 and address for the string allowing tools to map the pointers to what 430 the strings were. 431 432 saved_cmdlines: 433 434 Only the pid of the task is recorded in a trace event unless 435 the event specifically saves the task comm as well. Ftrace 436 makes a cache of pid mappings to comms to try to display 437 comms for events. If a pid for a comm is not listed, then 438 "<...>" is displayed in the output. 439 440 If the option "record-cmd" is set to "0", then comms of tasks 441 will not be saved during recording. By default, it is enabled. 442 443 saved_cmdlines_size: 444 445 By default, 128 comms are saved (see "saved_cmdlines" above). To 446 increase or decrease the amount of comms that are cached, echo 447 the number of comms to cache into this file. 448 449 saved_tgids: 450 451 If the option "record-tgid" is set, on each scheduling context switch 452 the Task Group ID of a task is saved in a table mapping the PID of 453 the thread to its TGID. By default, the "record-tgid" option is 454 disabled. 455 456 snapshot: 457 458 This displays the "snapshot" buffer and also lets the user 459 take a snapshot of the current running trace. 460 See the "Snapshot" section below for more details. 461 462 stack_max_size: 463 464 When the stack tracer is activated, this will display the 465 maximum stack size it has encountered. 466 See the "Stack Trace" section below. 467 468 stack_trace: 469 470 This displays the stack back trace of the largest stack 471 that was encountered when the stack tracer is activated. 472 See the "Stack Trace" section below. 473 474 stack_trace_filter: 475 476 This is similar to "set_ftrace_filter" but it limits what 477 functions the stack tracer will check. 478 479 trace_clock: 480 481 Whenever an event is recorded into the ring buffer, a 482 "timestamp" is added. This stamp comes from a specified 483 clock. By default, ftrace uses the "local" clock. This 484 clock is very fast and strictly per cpu, but on some 485 systems it may not be monotonic with respect to other 486 CPUs. In other words, the local clocks may not be in sync 487 with local clocks on other CPUs. 488 489 Usual clocks for tracing:: 490 491 # cat trace_clock 492 [local] global counter x86-tsc 493 494 The clock with the square brackets around it is the one in effect. 495 496 local: 497 Default clock, but may not be in sync across CPUs 498 499 global: 500 This clock is in sync with all CPUs but may 501 be a bit slower than the local clock. 502 503 counter: 504 This is not a clock at all, but literally an atomic 505 counter. It counts up one by one, but is in sync 506 with all CPUs. This is useful when you need to 507 know exactly the order events occurred with respect to 508 each other on different CPUs. 509 510 uptime: 511 This uses the jiffies counter and the time stamp 512 is relative to the time since boot up. 513 514 perf: 515 This makes ftrace use the same clock that perf uses. 516 Eventually perf will be able to read ftrace buffers 517 and this will help out in interleaving the data. 518 519 x86-tsc: 520 Architectures may define their own clocks. For 521 example, x86 uses its own TSC cycle clock here. 522 523 ppc-tb: 524 This uses the powerpc timebase register value. 525 This is in sync across CPUs and can also be used 526 to correlate events across hypervisor/guest if 527 tb_offset is known. 528 529 mono: 530 This uses the fast monotonic clock (CLOCK_MONOTONIC) 531 which is monotonic and is subject to NTP rate adjustments. 532 533 mono_raw: 534 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW) 535 which is monotonic but is not subject to any rate adjustments 536 and ticks at the same rate as the hardware clocksource. 537 538 boot: 539 This is the boot clock (CLOCK_BOOTTIME) and is based on the 540 fast monotonic clock, but also accounts for time spent in 541 suspend. Since the clock access is designed for use in 542 tracing in the suspend path, some side effects are possible 543 if clock is accessed after the suspend time is accounted before 544 the fast mono clock is updated. In this case, the clock update 545 appears to happen slightly sooner than it normally would have. 546 Also on 32-bit systems, it's possible that the 64-bit boot offset 547 sees a partial update. These effects are rare and post 548 processing should be able to handle them. See comments in the 549 ktime_get_boot_fast_ns() function for more information. 550 551 tai: 552 This is the tai clock (CLOCK_TAI) and is derived from the wall- 553 clock time. However, this clock does not experience 554 discontinuities and backwards jumps caused by NTP inserting leap 555 seconds. Since the clock access is designed for use in tracing, 556 side effects are possible. The clock access may yield wrong 557 readouts in case the internal TAI offset is updated e.g., caused 558 by setting the system time or using adjtimex() with an offset. 559 These effects are rare and post processing should be able to 560 handle them. See comments in the ktime_get_tai_fast_ns() 561 function for more information. 562 563 To set a clock, simply echo the clock name into this file:: 564 565 # echo global > trace_clock 566 567 Setting a clock clears the ring buffer content as well as the 568 "snapshot" buffer. 569 570 trace_marker: 571 572 This is a very useful file for synchronizing user space 573 with events happening in the kernel. Writing strings into 574 this file will be written into the ftrace buffer. 575 576 It is useful in applications to open this file at the start 577 of the application and just reference the file descriptor 578 for the file:: 579 580 void trace_write(const char *fmt, ...) 581 { 582 va_list ap; 583 char buf[256]; 584 int n; 585 586 if (trace_fd < 0) 587 return; 588 589 va_start(ap, fmt); 590 n = vsnprintf(buf, 256, fmt, ap); 591 va_end(ap); 592 593 write(trace_fd, buf, n); 594 } 595 596 start:: 597 598 trace_fd = open("trace_marker", O_WRONLY); 599 600 Note: Writing into the trace_marker file can also initiate triggers 601 that are written into /sys/kernel/tracing/events/ftrace/print/trigger 602 See "Event triggers" in Documentation/trace/events.rst and an 603 example in Documentation/trace/histogram.rst (Section 3.) 604 605 trace_marker_raw: 606 607 This is similar to trace_marker above, but is meant for binary data 608 to be written to it, where a tool can be used to parse the data 609 from trace_pipe_raw. 610 611 uprobe_events: 612 613 Add dynamic tracepoints in programs. 614 See uprobetracer.rst 615 616 uprobe_profile: 617 618 Uprobe statistics. See uprobetrace.txt 619 620 instances: 621 622 This is a way to make multiple trace buffers where different 623 events can be recorded in different buffers. 624 See "Instances" section below. 625 626 events: 627 628 This is the trace event directory. It holds event tracepoints 629 (also known as static tracepoints) that have been compiled 630 into the kernel. It shows what event tracepoints exist 631 and how they are grouped by system. There are "enable" 632 files at various levels that can enable the tracepoints 633 when a "1" is written to them. 634 635 See events.rst for more information. 636 637 set_event: 638 639 By echoing in the event into this file, will enable that event. 640 641 See events.rst for more information. 642 643 available_events: 644 645 A list of events that can be enabled in tracing. 646 647 See events.rst for more information. 648 649 timestamp_mode: 650 651 Certain tracers may change the timestamp mode used when 652 logging trace events into the event buffer. Events with 653 different modes can coexist within a buffer but the mode in 654 effect when an event is logged determines which timestamp mode 655 is used for that event. The default timestamp mode is 656 'delta'. 657 658 Usual timestamp modes for tracing: 659 660 # cat timestamp_mode 661 [delta] absolute 662 663 The timestamp mode with the square brackets around it is the 664 one in effect. 665 666 delta: Default timestamp mode - timestamp is a delta against 667 a per-buffer timestamp. 668 669 absolute: The timestamp is a full timestamp, not a delta 670 against some other value. As such it takes up more 671 space and is less efficient. 672 673 hwlat_detector: 674 675 Directory for the Hardware Latency Detector. 676 See "Hardware Latency Detector" section below. 677 678 per_cpu: 679 680 This is a directory that contains the trace per_cpu information. 681 682 per_cpu/cpu0/buffer_size_kb: 683 684 The ftrace buffer is defined per_cpu. That is, there's a separate 685 buffer for each CPU to allow writes to be done atomically, 686 and free from cache bouncing. These buffers may have different 687 size buffers. This file is similar to the buffer_size_kb 688 file, but it only displays or sets the buffer size for the 689 specific CPU. (here cpu0). 690 691 per_cpu/cpu0/trace: 692 693 This is similar to the "trace" file, but it will only display 694 the data specific for the CPU. If written to, it only clears 695 the specific CPU buffer. 696 697 per_cpu/cpu0/trace_pipe 698 699 This is similar to the "trace_pipe" file, and is a consuming 700 read, but it will only display (and consume) the data specific 701 for the CPU. 702 703 per_cpu/cpu0/trace_pipe_raw 704 705 For tools that can parse the ftrace ring buffer binary format, 706 the trace_pipe_raw file can be used to extract the data 707 from the ring buffer directly. With the use of the splice() 708 system call, the buffer data can be quickly transferred to 709 a file or to the network where a server is collecting the 710 data. 711 712 Like trace_pipe, this is a consuming reader, where multiple 713 reads will always produce different data. 714 715 per_cpu/cpu0/snapshot: 716 717 This is similar to the main "snapshot" file, but will only 718 snapshot the current CPU (if supported). It only displays 719 the content of the snapshot for a given CPU, and if 720 written to, only clears this CPU buffer. 721 722 per_cpu/cpu0/snapshot_raw: 723 724 Similar to the trace_pipe_raw, but will read the binary format 725 from the snapshot buffer for the given CPU. 726 727 per_cpu/cpu0/stats: 728 729 This displays certain stats about the ring buffer: 730 731 entries: 732 The number of events that are still in the buffer. 733 734 overrun: 735 The number of lost events due to overwriting when 736 the buffer was full. 737 738 commit overrun: 739 Should always be zero. 740 This gets set if so many events happened within a nested 741 event (ring buffer is re-entrant), that it fills the 742 buffer and starts dropping events. 743 744 bytes: 745 Bytes actually read (not overwritten). 746 747 oldest event ts: 748 The oldest timestamp in the buffer 749 750 now ts: 751 The current timestamp 752 753 dropped events: 754 Events lost due to overwrite option being off. 755 756 read events: 757 The number of events read. 758 759The Tracers 760----------- 761 762Here is the list of current tracers that may be configured. 763 764 "function" 765 766 Function call tracer to trace all kernel functions. 767 768 "function_graph" 769 770 Similar to the function tracer except that the 771 function tracer probes the functions on their entry 772 whereas the function graph tracer traces on both entry 773 and exit of the functions. It then provides the ability 774 to draw a graph of function calls similar to C code 775 source. 776 777 "blk" 778 779 The block tracer. The tracer used by the blktrace user 780 application. 781 782 "hwlat" 783 784 The Hardware Latency tracer is used to detect if the hardware 785 produces any latency. See "Hardware Latency Detector" section 786 below. 787 788 "irqsoff" 789 790 Traces the areas that disable interrupts and saves 791 the trace with the longest max latency. 792 See tracing_max_latency. When a new max is recorded, 793 it replaces the old trace. It is best to view this 794 trace with the latency-format option enabled, which 795 happens automatically when the tracer is selected. 796 797 "preemptoff" 798 799 Similar to irqsoff but traces and records the amount of 800 time for which preemption is disabled. 801 802 "preemptirqsoff" 803 804 Similar to irqsoff and preemptoff, but traces and 805 records the largest time for which irqs and/or preemption 806 is disabled. 807 808 "wakeup" 809 810 Traces and records the max latency that it takes for 811 the highest priority task to get scheduled after 812 it has been woken up. 813 Traces all tasks as an average developer would expect. 814 815 "wakeup_rt" 816 817 Traces and records the max latency that it takes for just 818 RT tasks (as the current "wakeup" does). This is useful 819 for those interested in wake up timings of RT tasks. 820 821 "wakeup_dl" 822 823 Traces and records the max latency that it takes for 824 a SCHED_DEADLINE task to be woken (as the "wakeup" and 825 "wakeup_rt" does). 826 827 "mmiotrace" 828 829 A special tracer that is used to trace binary module. 830 It will trace all the calls that a module makes to the 831 hardware. Everything it writes and reads from the I/O 832 as well. 833 834 "branch" 835 836 This tracer can be configured when tracing likely/unlikely 837 calls within the kernel. It will trace when a likely and 838 unlikely branch is hit and if it was correct in its prediction 839 of being correct. 840 841 "nop" 842 843 This is the "trace nothing" tracer. To remove all 844 tracers from tracing simply echo "nop" into 845 current_tracer. 846 847Error conditions 848---------------- 849 850 For most ftrace commands, failure modes are obvious and communicated 851 using standard return codes. 852 853 For other more involved commands, extended error information may be 854 available via the tracing/error_log file. For the commands that 855 support it, reading the tracing/error_log file after an error will 856 display more detailed information about what went wrong, if 857 information is available. The tracing/error_log file is a circular 858 error log displaying a small number (currently, 8) of ftrace errors 859 for the last (8) failed commands. 860 861 The extended error information and usage takes the form shown in 862 this example:: 863 864 # echo xxx > /sys/kernel/tracing/events/sched/sched_wakeup/trigger 865 echo: write error: Invalid argument 866 867 # cat /sys/kernel/tracing/error_log 868 [ 5348.887237] location: error: Couldn't yyy: zzz 869 Command: xxx 870 ^ 871 [ 7517.023364] location: error: Bad rrr: sss 872 Command: ppp qqq 873 ^ 874 875 To clear the error log, echo the empty string into it:: 876 877 # echo > /sys/kernel/tracing/error_log 878 879Examples of using the tracer 880---------------------------- 881 882Here are typical examples of using the tracers when controlling 883them only with the tracefs interface (without using any 884user-land utilities). 885 886Output format: 887-------------- 888 889Here is an example of the output format of the file "trace":: 890 891 # tracer: function 892 # 893 # entries-in-buffer/entries-written: 140080/250280 #P:4 894 # 895 # _-----=> irqs-off 896 # / _----=> need-resched 897 # | / _---=> hardirq/softirq 898 # || / _--=> preempt-depth 899 # ||| / delay 900 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 901 # | | | |||| | | 902 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath 903 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close 904 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd 905 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify 906 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock 907 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd 908 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock 909 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd 910 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close 911 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath 912 .... 913 914A header is printed with the tracer name that is represented by 915the trace. In this case the tracer is "function". Then it shows the 916number of events in the buffer as well as the total number of entries 917that were written. The difference is the number of entries that were 918lost due to the buffer filling up (250280 - 140080 = 110200 events 919lost). 920 921The header explains the content of the events. Task name "bash", the task 922PID "1977", the CPU that it was running on "000", the latency format 923(explained below), the timestamp in <secs>.<usecs> format, the 924function name that was traced "sys_close" and the parent function that 925called this function "system_call_fastpath". The timestamp is the time 926at which the function was entered. 927 928Latency trace format 929-------------------- 930 931When the latency-format option is enabled or when one of the latency 932tracers is set, the trace file gives somewhat more information to see 933why a latency happened. Here is a typical trace:: 934 935 # tracer: irqsoff 936 # 937 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 938 # -------------------------------------------------------------------- 939 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 940 # ----------------- 941 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0) 942 # ----------------- 943 # => started at: __lock_task_sighand 944 # => ended at: _raw_spin_unlock_irqrestore 945 # 946 # 947 # _------=> CPU# 948 # / _-----=> irqs-off 949 # | / _----=> need-resched 950 # || / _---=> hardirq/softirq 951 # ||| / _--=> preempt-depth 952 # |||| / delay 953 # cmd pid ||||| time | caller 954 # \ / ||||| \ | / 955 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand 956 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore 957 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore 958 ps-6143 2d..1 306us : <stack trace> 959 => trace_hardirqs_on_caller 960 => trace_hardirqs_on 961 => _raw_spin_unlock_irqrestore 962 => do_task_stat 963 => proc_tgid_stat 964 => proc_single_show 965 => seq_read 966 => vfs_read 967 => sys_read 968 => system_call_fastpath 969 970 971This shows that the current tracer is "irqsoff" tracing the time 972for which interrupts were disabled. It gives the trace version (which 973never changes) and the version of the kernel upon which this was executed on 974(3.8). Then it displays the max latency in microseconds (259 us). The number 975of trace entries displayed and the total number (both are four: #4/4). 976VP, KP, SP, and HP are always zero and are reserved for later use. 977#P is the number of online CPUs (#P:4). 978 979The task is the process that was running when the latency 980occurred. (ps pid: 6143). 981 982The start and stop (the functions in which the interrupts were 983disabled and enabled respectively) that caused the latencies: 984 985 - __lock_task_sighand is where the interrupts were disabled. 986 - _raw_spin_unlock_irqrestore is where they were enabled again. 987 988The next lines after the header are the trace itself. The header 989explains which is which. 990 991 cmd: The name of the process in the trace. 992 993 pid: The PID of that process. 994 995 CPU#: The CPU which the process was running on. 996 997 irqs-off: 'd' interrupts are disabled. '.' otherwise. 998 .. caution:: If the architecture does not support a way to 999 read the irq flags variable, an 'X' will always 1000 be printed here. 1001 1002 need-resched: 1003 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set, 1004 - 'n' only TIF_NEED_RESCHED is set, 1005 - 'p' only PREEMPT_NEED_RESCHED is set, 1006 - '.' otherwise. 1007 1008 hardirq/softirq: 1009 - 'Z' - NMI occurred inside a hardirq 1010 - 'z' - NMI is running 1011 - 'H' - hard irq occurred inside a softirq. 1012 - 'h' - hard irq is running 1013 - 's' - soft irq is running 1014 - '.' - normal context. 1015 1016 preempt-depth: The level of preempt_disabled 1017 1018The above is mostly meaningful for kernel developers. 1019 1020 time: 1021 When the latency-format option is enabled, the trace file 1022 output includes a timestamp relative to the start of the 1023 trace. This differs from the output when latency-format 1024 is disabled, which includes an absolute timestamp. 1025 1026 delay: 1027 This is just to help catch your eye a bit better. And 1028 needs to be fixed to be only relative to the same CPU. 1029 The marks are determined by the difference between this 1030 current trace and the next trace. 1031 1032 - '$' - greater than 1 second 1033 - '@' - greater than 100 millisecond 1034 - '*' - greater than 10 millisecond 1035 - '#' - greater than 1000 microsecond 1036 - '!' - greater than 100 microsecond 1037 - '+' - greater than 10 microsecond 1038 - ' ' - less than or equal to 10 microsecond. 1039 1040 The rest is the same as the 'trace' file. 1041 1042 Note, the latency tracers will usually end with a back trace 1043 to easily find where the latency occurred. 1044 1045trace_options 1046------------- 1047 1048The trace_options file (or the options directory) is used to control 1049what gets printed in the trace output, or manipulate the tracers. 1050To see what is available, simply cat the file:: 1051 1052 cat trace_options 1053 print-parent 1054 nosym-offset 1055 nosym-addr 1056 noverbose 1057 noraw 1058 nohex 1059 nobin 1060 noblock 1061 nofields 1062 trace_printk 1063 annotate 1064 nouserstacktrace 1065 nosym-userobj 1066 noprintk-msg-only 1067 context-info 1068 nolatency-format 1069 record-cmd 1070 norecord-tgid 1071 overwrite 1072 nodisable_on_free 1073 irq-info 1074 markers 1075 noevent-fork 1076 function-trace 1077 nofunction-fork 1078 nodisplay-graph 1079 nostacktrace 1080 nobranch 1081 1082To disable one of the options, echo in the option prepended with 1083"no":: 1084 1085 echo noprint-parent > trace_options 1086 1087To enable an option, leave off the "no":: 1088 1089 echo sym-offset > trace_options 1090 1091Here are the available options: 1092 1093 print-parent 1094 On function traces, display the calling (parent) 1095 function as well as the function being traced. 1096 :: 1097 1098 print-parent: 1099 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul 1100 1101 noprint-parent: 1102 bash-4000 [01] 1477.606694: simple_strtoul 1103 1104 1105 sym-offset 1106 Display not only the function name, but also the 1107 offset in the function. For example, instead of 1108 seeing just "ktime_get", you will see 1109 "ktime_get+0xb/0x20". 1110 :: 1111 1112 sym-offset: 1113 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0 1114 1115 sym-addr 1116 This will also display the function address as well 1117 as the function name. 1118 :: 1119 1120 sym-addr: 1121 bash-4000 [01] 1477.606694: simple_strtoul <c0339346> 1122 1123 verbose 1124 This deals with the trace file when the 1125 latency-format option is enabled. 1126 :: 1127 1128 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \ 1129 (+0.000ms): simple_strtoul (kstrtoul) 1130 1131 raw 1132 This will display raw numbers. This option is best for 1133 use with user applications that can translate the raw 1134 numbers better than having it done in the kernel. 1135 1136 hex 1137 Similar to raw, but the numbers will be in a hexadecimal format. 1138 1139 bin 1140 This will print out the formats in raw binary. 1141 1142 block 1143 When set, reading trace_pipe will not block when polled. 1144 1145 fields 1146 Print the fields as described by their types. This is a better 1147 option than using hex, bin or raw, as it gives a better parsing 1148 of the content of the event. 1149 1150 trace_printk 1151 Can disable trace_printk() from writing into the buffer. 1152 1153 annotate 1154 It is sometimes confusing when the CPU buffers are full 1155 and one CPU buffer had a lot of events recently, thus 1156 a shorter time frame, were another CPU may have only had 1157 a few events, which lets it have older events. When 1158 the trace is reported, it shows the oldest events first, 1159 and it may look like only one CPU ran (the one with the 1160 oldest events). When the annotate option is set, it will 1161 display when a new CPU buffer started:: 1162 1163 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on 1164 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on 1165 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore 1166 ##### CPU 2 buffer started #### 1167 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle 1168 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog 1169 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock 1170 1171 userstacktrace 1172 This option changes the trace. It records a 1173 stacktrace of the current user space thread after 1174 each trace event. 1175 1176 sym-userobj 1177 when user stacktrace are enabled, look up which 1178 object the address belongs to, and print a 1179 relative address. This is especially useful when 1180 ASLR is on, otherwise you don't get a chance to 1181 resolve the address to object/file/line after 1182 the app is no longer running 1183 1184 The lookup is performed when you read 1185 trace,trace_pipe. Example:: 1186 1187 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0 1188 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6] 1189 1190 1191 printk-msg-only 1192 When set, trace_printk()s will only show the format 1193 and not their parameters (if trace_bprintk() or 1194 trace_bputs() was used to save the trace_printk()). 1195 1196 context-info 1197 Show only the event data. Hides the comm, PID, 1198 timestamp, CPU, and other useful data. 1199 1200 latency-format 1201 This option changes the trace output. When it is enabled, 1202 the trace displays additional information about the 1203 latency, as described in "Latency trace format". 1204 1205 pause-on-trace 1206 When set, opening the trace file for read, will pause 1207 writing to the ring buffer (as if tracing_on was set to zero). 1208 This simulates the original behavior of the trace file. 1209 When the file is closed, tracing will be enabled again. 1210 1211 hash-ptr 1212 When set, "%p" in the event printk format displays the 1213 hashed pointer value instead of real address. 1214 This will be useful if you want to find out which hashed 1215 value is corresponding to the real value in trace log. 1216 1217 record-cmd 1218 When any event or tracer is enabled, a hook is enabled 1219 in the sched_switch trace point to fill comm cache 1220 with mapped pids and comms. But this may cause some 1221 overhead, and if you only care about pids, and not the 1222 name of the task, disabling this option can lower the 1223 impact of tracing. See "saved_cmdlines". 1224 1225 record-tgid 1226 When any event or tracer is enabled, a hook is enabled 1227 in the sched_switch trace point to fill the cache of 1228 mapped Thread Group IDs (TGID) mapping to pids. See 1229 "saved_tgids". 1230 1231 overwrite 1232 This controls what happens when the trace buffer is 1233 full. If "1" (default), the oldest events are 1234 discarded and overwritten. If "0", then the newest 1235 events are discarded. 1236 (see per_cpu/cpu0/stats for overrun and dropped) 1237 1238 disable_on_free 1239 When the free_buffer is closed, tracing will 1240 stop (tracing_on set to 0). 1241 1242 irq-info 1243 Shows the interrupt, preempt count, need resched data. 1244 When disabled, the trace looks like:: 1245 1246 # tracer: function 1247 # 1248 # entries-in-buffer/entries-written: 144405/9452052 #P:4 1249 # 1250 # TASK-PID CPU# TIMESTAMP FUNCTION 1251 # | | | | | 1252 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up 1253 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89 1254 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task 1255 1256 1257 markers 1258 When set, the trace_marker is writable (only by root). 1259 When disabled, the trace_marker will error with EINVAL 1260 on write. 1261 1262 event-fork 1263 When set, tasks with PIDs listed in set_event_pid will have 1264 the PIDs of their children added to set_event_pid when those 1265 tasks fork. Also, when tasks with PIDs in set_event_pid exit, 1266 their PIDs will be removed from the file. 1267 1268 This affects PIDs listed in set_event_notrace_pid as well. 1269 1270 function-trace 1271 The latency tracers will enable function tracing 1272 if this option is enabled (default it is). When 1273 it is disabled, the latency tracers do not trace 1274 functions. This keeps the overhead of the tracer down 1275 when performing latency tests. 1276 1277 function-fork 1278 When set, tasks with PIDs listed in set_ftrace_pid will 1279 have the PIDs of their children added to set_ftrace_pid 1280 when those tasks fork. Also, when tasks with PIDs in 1281 set_ftrace_pid exit, their PIDs will be removed from the 1282 file. 1283 1284 This affects PIDs in set_ftrace_notrace_pid as well. 1285 1286 display-graph 1287 When set, the latency tracers (irqsoff, wakeup, etc) will 1288 use function graph tracing instead of function tracing. 1289 1290 stacktrace 1291 When set, a stack trace is recorded after any trace event 1292 is recorded. 1293 1294 branch 1295 Enable branch tracing with the tracer. This enables branch 1296 tracer along with the currently set tracer. Enabling this 1297 with the "nop" tracer is the same as just enabling the 1298 "branch" tracer. 1299 1300.. tip:: Some tracers have their own options. They only appear in this 1301 file when the tracer is active. They always appear in the 1302 options directory. 1303 1304 1305Here are the per tracer options: 1306 1307Options for function tracer: 1308 1309 func_stack_trace 1310 When set, a stack trace is recorded after every 1311 function that is recorded. NOTE! Limit the functions 1312 that are recorded before enabling this, with 1313 "set_ftrace_filter" otherwise the system performance 1314 will be critically degraded. Remember to disable 1315 this option before clearing the function filter. 1316 1317Options for function_graph tracer: 1318 1319 Since the function_graph tracer has a slightly different output 1320 it has its own options to control what is displayed. 1321 1322 funcgraph-overrun 1323 When set, the "overrun" of the graph stack is 1324 displayed after each function traced. The 1325 overrun, is when the stack depth of the calls 1326 is greater than what is reserved for each task. 1327 Each task has a fixed array of functions to 1328 trace in the call graph. If the depth of the 1329 calls exceeds that, the function is not traced. 1330 The overrun is the number of functions missed 1331 due to exceeding this array. 1332 1333 funcgraph-cpu 1334 When set, the CPU number of the CPU where the trace 1335 occurred is displayed. 1336 1337 funcgraph-overhead 1338 When set, if the function takes longer than 1339 A certain amount, then a delay marker is 1340 displayed. See "delay" above, under the 1341 header description. 1342 1343 funcgraph-proc 1344 Unlike other tracers, the process' command line 1345 is not displayed by default, but instead only 1346 when a task is traced in and out during a context 1347 switch. Enabling this options has the command 1348 of each process displayed at every line. 1349 1350 funcgraph-duration 1351 At the end of each function (the return) 1352 the duration of the amount of time in the 1353 function is displayed in microseconds. 1354 1355 funcgraph-abstime 1356 When set, the timestamp is displayed at each line. 1357 1358 funcgraph-irqs 1359 When disabled, functions that happen inside an 1360 interrupt will not be traced. 1361 1362 funcgraph-tail 1363 When set, the return event will include the function 1364 that it represents. By default this is off, and 1365 only a closing curly bracket "}" is displayed for 1366 the return of a function. 1367 1368 funcgraph-retval 1369 When set, the return value of each traced function 1370 will be printed after an equal sign "=". By default 1371 this is off. 1372 1373 funcgraph-retval-hex 1374 When set, the return value will always be printed 1375 in hexadecimal format. If the option is not set and 1376 the return value is an error code, it will be printed 1377 in signed decimal format; otherwise it will also be 1378 printed in hexadecimal format. By default, this option 1379 is off. 1380 1381 sleep-time 1382 When running function graph tracer, to include 1383 the time a task schedules out in its function. 1384 When enabled, it will account time the task has been 1385 scheduled out as part of the function call. 1386 1387 graph-time 1388 When running function profiler with function graph tracer, 1389 to include the time to call nested functions. When this is 1390 not set, the time reported for the function will only 1391 include the time the function itself executed for, not the 1392 time for functions that it called. 1393 1394Options for blk tracer: 1395 1396 blk_classic 1397 Shows a more minimalistic output. 1398 1399 1400irqsoff 1401------- 1402 1403When interrupts are disabled, the CPU can not react to any other 1404external event (besides NMIs and SMIs). This prevents the timer 1405interrupt from triggering or the mouse interrupt from letting 1406the kernel know of a new mouse event. The result is a latency 1407with the reaction time. 1408 1409The irqsoff tracer tracks the time for which interrupts are 1410disabled. When a new maximum latency is hit, the tracer saves 1411the trace leading up to that latency point so that every time a 1412new maximum is reached, the old saved trace is discarded and the 1413new trace is saved. 1414 1415To reset the maximum, echo 0 into tracing_max_latency. Here is 1416an example:: 1417 1418 # echo 0 > options/function-trace 1419 # echo irqsoff > current_tracer 1420 # echo 1 > tracing_on 1421 # echo 0 > tracing_max_latency 1422 # ls -ltr 1423 [...] 1424 # echo 0 > tracing_on 1425 # cat trace 1426 # tracer: irqsoff 1427 # 1428 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1429 # -------------------------------------------------------------------- 1430 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1431 # ----------------- 1432 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0) 1433 # ----------------- 1434 # => started at: run_timer_softirq 1435 # => ended at: run_timer_softirq 1436 # 1437 # 1438 # _------=> CPU# 1439 # / _-----=> irqs-off 1440 # | / _----=> need-resched 1441 # || / _---=> hardirq/softirq 1442 # ||| / _--=> preempt-depth 1443 # |||| / delay 1444 # cmd pid ||||| time | caller 1445 # \ / ||||| \ | / 1446 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq 1447 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq 1448 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq 1449 <idle>-0 0dNs3 25us : <stack trace> 1450 => _raw_spin_unlock_irq 1451 => run_timer_softirq 1452 => __do_softirq 1453 => call_softirq 1454 => do_softirq 1455 => irq_exit 1456 => smp_apic_timer_interrupt 1457 => apic_timer_interrupt 1458 => rcu_idle_exit 1459 => cpu_idle 1460 => rest_init 1461 => start_kernel 1462 => x86_64_start_reservations 1463 => x86_64_start_kernel 1464 1465Here we see that we had a latency of 16 microseconds (which is 1466very good). The _raw_spin_lock_irq in run_timer_softirq disabled 1467interrupts. The difference between the 16 and the displayed 1468timestamp 25us occurred because the clock was incremented 1469between the time of recording the max latency and the time of 1470recording the function that had that latency. 1471 1472Note the above example had function-trace not set. If we set 1473function-trace, we get a much larger output:: 1474 1475 with echo 1 > options/function-trace 1476 1477 # tracer: irqsoff 1478 # 1479 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1480 # -------------------------------------------------------------------- 1481 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1482 # ----------------- 1483 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0) 1484 # ----------------- 1485 # => started at: ata_scsi_queuecmd 1486 # => ended at: ata_scsi_queuecmd 1487 # 1488 # 1489 # _------=> CPU# 1490 # / _-----=> irqs-off 1491 # | / _----=> need-resched 1492 # || / _---=> hardirq/softirq 1493 # ||| / _--=> preempt-depth 1494 # |||| / delay 1495 # cmd pid ||||| time | caller 1496 # \ / ||||| \ | / 1497 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd 1498 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave 1499 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd 1500 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev 1501 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev 1502 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd 1503 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd 1504 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd 1505 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat 1506 [...] 1507 bash-2042 3d..1 67us : delay_tsc <-__delay 1508 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc 1509 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc 1510 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc 1511 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc 1512 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue 1513 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1514 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1515 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd 1516 bash-2042 3d..1 120us : <stack trace> 1517 => _raw_spin_unlock_irqrestore 1518 => ata_scsi_queuecmd 1519 => scsi_dispatch_cmd 1520 => scsi_request_fn 1521 => __blk_run_queue_uncond 1522 => __blk_run_queue 1523 => blk_queue_bio 1524 => submit_bio_noacct 1525 => submit_bio 1526 => submit_bh 1527 => __ext3_get_inode_loc 1528 => ext3_iget 1529 => ext3_lookup 1530 => lookup_real 1531 => __lookup_hash 1532 => walk_component 1533 => lookup_last 1534 => path_lookupat 1535 => filename_lookup 1536 => user_path_at_empty 1537 => user_path_at 1538 => vfs_fstatat 1539 => vfs_stat 1540 => sys_newstat 1541 => system_call_fastpath 1542 1543 1544Here we traced a 71 microsecond latency. But we also see all the 1545functions that were called during that time. Note that by 1546enabling function tracing, we incur an added overhead. This 1547overhead may extend the latency times. But nevertheless, this 1548trace has provided some very helpful debugging information. 1549 1550If we prefer function graph output instead of function, we can set 1551display-graph option:: 1552 1553 with echo 1 > options/display-graph 1554 1555 # tracer: irqsoff 1556 # 1557 # irqsoff latency trace v1.1.5 on 4.20.0-rc6+ 1558 # -------------------------------------------------------------------- 1559 # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4) 1560 # ----------------- 1561 # | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0) 1562 # ----------------- 1563 # => started at: free_debug_processing 1564 # => ended at: return_to_handler 1565 # 1566 # 1567 # _-----=> irqs-off 1568 # / _----=> need-resched 1569 # | / _---=> hardirq/softirq 1570 # || / _--=> preempt-depth 1571 # ||| / 1572 # REL TIME CPU TASK/PID |||| DURATION FUNCTION CALLS 1573 # | | | | |||| | | | | | | 1574 0 us | 0) bash-1507 | d... | 0.000 us | _raw_spin_lock_irqsave(); 1575 0 us | 0) bash-1507 | d..1 | 0.378 us | do_raw_spin_trylock(); 1576 1 us | 0) bash-1507 | d..2 | | set_track() { 1577 2 us | 0) bash-1507 | d..2 | | save_stack_trace() { 1578 2 us | 0) bash-1507 | d..2 | | __save_stack_trace() { 1579 3 us | 0) bash-1507 | d..2 | | __unwind_start() { 1580 3 us | 0) bash-1507 | d..2 | | get_stack_info() { 1581 3 us | 0) bash-1507 | d..2 | 0.351 us | in_task_stack(); 1582 4 us | 0) bash-1507 | d..2 | 1.107 us | } 1583 [...] 1584 3750 us | 0) bash-1507 | d..1 | 0.516 us | do_raw_spin_unlock(); 1585 3750 us | 0) bash-1507 | d..1 | 0.000 us | _raw_spin_unlock_irqrestore(); 1586 3764 us | 0) bash-1507 | d..1 | 0.000 us | tracer_hardirqs_on(); 1587 bash-1507 0d..1 3792us : <stack trace> 1588 => free_debug_processing 1589 => __slab_free 1590 => kmem_cache_free 1591 => vm_area_free 1592 => remove_vma 1593 => exit_mmap 1594 => mmput 1595 => begin_new_exec 1596 => load_elf_binary 1597 => search_binary_handler 1598 => __do_execve_file.isra.32 1599 => __x64_sys_execve 1600 => do_syscall_64 1601 => entry_SYSCALL_64_after_hwframe 1602 1603preemptoff 1604---------- 1605 1606When preemption is disabled, we may be able to receive 1607interrupts but the task cannot be preempted and a higher 1608priority task must wait for preemption to be enabled again 1609before it can preempt a lower priority task. 1610 1611The preemptoff tracer traces the places that disable preemption. 1612Like the irqsoff tracer, it records the maximum latency for 1613which preemption was disabled. The control of preemptoff tracer 1614is much like the irqsoff tracer. 1615:: 1616 1617 # echo 0 > options/function-trace 1618 # echo preemptoff > current_tracer 1619 # echo 1 > tracing_on 1620 # echo 0 > tracing_max_latency 1621 # ls -ltr 1622 [...] 1623 # echo 0 > tracing_on 1624 # cat trace 1625 # tracer: preemptoff 1626 # 1627 # preemptoff latency trace v1.1.5 on 3.8.0-test+ 1628 # -------------------------------------------------------------------- 1629 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1630 # ----------------- 1631 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0) 1632 # ----------------- 1633 # => started at: do_IRQ 1634 # => ended at: do_IRQ 1635 # 1636 # 1637 # _------=> CPU# 1638 # / _-----=> irqs-off 1639 # | / _----=> need-resched 1640 # || / _---=> hardirq/softirq 1641 # ||| / _--=> preempt-depth 1642 # |||| / delay 1643 # cmd pid ||||| time | caller 1644 # \ / ||||| \ | / 1645 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ 1646 sshd-1991 1d..1 46us : irq_exit <-do_IRQ 1647 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ 1648 sshd-1991 1d..1 52us : <stack trace> 1649 => sub_preempt_count 1650 => irq_exit 1651 => do_IRQ 1652 => ret_from_intr 1653 1654 1655This has some more changes. Preemption was disabled when an 1656interrupt came in (notice the 'h'), and was enabled on exit. 1657But we also see that interrupts have been disabled when entering 1658the preempt off section and leaving it (the 'd'). We do not know if 1659interrupts were enabled in the mean time or shortly after this 1660was over. 1661:: 1662 1663 # tracer: preemptoff 1664 # 1665 # preemptoff latency trace v1.1.5 on 3.8.0-test+ 1666 # -------------------------------------------------------------------- 1667 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1668 # ----------------- 1669 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0) 1670 # ----------------- 1671 # => started at: wake_up_new_task 1672 # => ended at: task_rq_unlock 1673 # 1674 # 1675 # _------=> CPU# 1676 # / _-----=> irqs-off 1677 # | / _----=> need-resched 1678 # || / _---=> hardirq/softirq 1679 # ||| / _--=> preempt-depth 1680 # |||| / delay 1681 # cmd pid ||||| time | caller 1682 # \ / ||||| \ | / 1683 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task 1684 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq 1685 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair 1686 bash-1994 1d..1 1us : source_load <-select_task_rq_fair 1687 bash-1994 1d..1 1us : source_load <-select_task_rq_fair 1688 [...] 1689 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt 1690 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter 1691 bash-1994 1d..1 13us : add_preempt_count <-irq_enter 1692 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt 1693 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt 1694 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt 1695 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock 1696 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt 1697 [...] 1698 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event 1699 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt 1700 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit 1701 bash-1994 1d..2 36us : do_softirq <-irq_exit 1702 bash-1994 1d..2 36us : __do_softirq <-call_softirq 1703 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq 1704 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq 1705 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq 1706 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock 1707 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq 1708 [...] 1709 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks 1710 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq 1711 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable 1712 bash-1994 1dN.2 82us : idle_cpu <-irq_exit 1713 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit 1714 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit 1715 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock 1716 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock 1717 bash-1994 1.N.1 104us : <stack trace> 1718 => sub_preempt_count 1719 => _raw_spin_unlock_irqrestore 1720 => task_rq_unlock 1721 => wake_up_new_task 1722 => do_fork 1723 => sys_clone 1724 => stub_clone 1725 1726 1727The above is an example of the preemptoff trace with 1728function-trace set. Here we see that interrupts were not disabled 1729the entire time. The irq_enter code lets us know that we entered 1730an interrupt 'h'. Before that, the functions being traced still 1731show that it is not in an interrupt, but we can see from the 1732functions themselves that this is not the case. 1733 1734preemptirqsoff 1735-------------- 1736 1737Knowing the locations that have interrupts disabled or 1738preemption disabled for the longest times is helpful. But 1739sometimes we would like to know when either preemption and/or 1740interrupts are disabled. 1741 1742Consider the following code:: 1743 1744 local_irq_disable(); 1745 call_function_with_irqs_off(); 1746 preempt_disable(); 1747 call_function_with_irqs_and_preemption_off(); 1748 local_irq_enable(); 1749 call_function_with_preemption_off(); 1750 preempt_enable(); 1751 1752The irqsoff tracer will record the total length of 1753call_function_with_irqs_off() and 1754call_function_with_irqs_and_preemption_off(). 1755 1756The preemptoff tracer will record the total length of 1757call_function_with_irqs_and_preemption_off() and 1758call_function_with_preemption_off(). 1759 1760But neither will trace the time that interrupts and/or 1761preemption is disabled. This total time is the time that we can 1762not schedule. To record this time, use the preemptirqsoff 1763tracer. 1764 1765Again, using this trace is much like the irqsoff and preemptoff 1766tracers. 1767:: 1768 1769 # echo 0 > options/function-trace 1770 # echo preemptirqsoff > current_tracer 1771 # echo 1 > tracing_on 1772 # echo 0 > tracing_max_latency 1773 # ls -ltr 1774 [...] 1775 # echo 0 > tracing_on 1776 # cat trace 1777 # tracer: preemptirqsoff 1778 # 1779 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+ 1780 # -------------------------------------------------------------------- 1781 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1782 # ----------------- 1783 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0) 1784 # ----------------- 1785 # => started at: ata_scsi_queuecmd 1786 # => ended at: ata_scsi_queuecmd 1787 # 1788 # 1789 # _------=> CPU# 1790 # / _-----=> irqs-off 1791 # | / _----=> need-resched 1792 # || / _---=> hardirq/softirq 1793 # ||| / _--=> preempt-depth 1794 # |||| / delay 1795 # cmd pid ||||| time | caller 1796 # \ / ||||| \ | / 1797 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd 1798 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1799 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd 1800 ls-2230 3...1 111us : <stack trace> 1801 => sub_preempt_count 1802 => _raw_spin_unlock_irqrestore 1803 => ata_scsi_queuecmd 1804 => scsi_dispatch_cmd 1805 => scsi_request_fn 1806 => __blk_run_queue_uncond 1807 => __blk_run_queue 1808 => blk_queue_bio 1809 => submit_bio_noacct 1810 => submit_bio 1811 => submit_bh 1812 => ext3_bread 1813 => ext3_dir_bread 1814 => htree_dirblock_to_tree 1815 => ext3_htree_fill_tree 1816 => ext3_readdir 1817 => vfs_readdir 1818 => sys_getdents 1819 => system_call_fastpath 1820 1821 1822The trace_hardirqs_off_thunk is called from assembly on x86 when 1823interrupts are disabled in the assembly code. Without the 1824function tracing, we do not know if interrupts were enabled 1825within the preemption points. We do see that it started with 1826preemption enabled. 1827 1828Here is a trace with function-trace set:: 1829 1830 # tracer: preemptirqsoff 1831 # 1832 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+ 1833 # -------------------------------------------------------------------- 1834 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1835 # ----------------- 1836 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0) 1837 # ----------------- 1838 # => started at: schedule 1839 # => ended at: mutex_unlock 1840 # 1841 # 1842 # _------=> CPU# 1843 # / _-----=> irqs-off 1844 # | / _----=> need-resched 1845 # || / _---=> hardirq/softirq 1846 # ||| / _--=> preempt-depth 1847 # |||| / delay 1848 # cmd pid ||||| time | caller 1849 # \ / ||||| \ | / 1850 kworker/-59 3...1 0us : __schedule <-schedule 1851 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch 1852 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq 1853 kworker/-59 3d..2 1us : deactivate_task <-__schedule 1854 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task 1855 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task 1856 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task 1857 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair 1858 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr 1859 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr 1860 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge 1861 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge 1862 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair 1863 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair 1864 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair 1865 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair 1866 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair 1867 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair 1868 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule 1869 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping 1870 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule 1871 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task 1872 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair 1873 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair 1874 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity 1875 ls-2269 3d..2 7us : finish_task_switch <-__schedule 1876 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch 1877 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr 1878 ls-2269 3d..2 8us : irq_enter <-do_IRQ 1879 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter 1880 ls-2269 3d..2 9us : add_preempt_count <-irq_enter 1881 ls-2269 3d.h2 9us : exit_idle <-do_IRQ 1882 [...] 1883 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock 1884 ls-2269 3d.h2 20us : irq_exit <-do_IRQ 1885 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit 1886 ls-2269 3d..3 21us : do_softirq <-irq_exit 1887 ls-2269 3d..3 21us : __do_softirq <-call_softirq 1888 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq 1889 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip 1890 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip 1891 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr 1892 ls-2269 3d.s5 31us : irq_enter <-do_IRQ 1893 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter 1894 [...] 1895 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter 1896 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter 1897 ls-2269 3d.H5 32us : exit_idle <-do_IRQ 1898 ls-2269 3d.H5 32us : handle_irq <-do_IRQ 1899 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq 1900 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq 1901 [...] 1902 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll 1903 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action 1904 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq 1905 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable 1906 ls-2269 3d..3 159us : idle_cpu <-irq_exit 1907 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit 1908 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit 1909 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock 1910 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock 1911 ls-2269 3d... 186us : <stack trace> 1912 => __mutex_unlock_slowpath 1913 => mutex_unlock 1914 => process_output 1915 => n_tty_write 1916 => tty_write 1917 => vfs_write 1918 => sys_write 1919 => system_call_fastpath 1920 1921This is an interesting trace. It started with kworker running and 1922scheduling out and ls taking over. But as soon as ls released the 1923rq lock and enabled interrupts (but not preemption) an interrupt 1924triggered. When the interrupt finished, it started running softirqs. 1925But while the softirq was running, another interrupt triggered. 1926When an interrupt is running inside a softirq, the annotation is 'H'. 1927 1928 1929wakeup 1930------ 1931 1932One common case that people are interested in tracing is the 1933time it takes for a task that is woken to actually wake up. 1934Now for non Real-Time tasks, this can be arbitrary. But tracing 1935it none the less can be interesting. 1936 1937Without function tracing:: 1938 1939 # echo 0 > options/function-trace 1940 # echo wakeup > current_tracer 1941 # echo 1 > tracing_on 1942 # echo 0 > tracing_max_latency 1943 # chrt -f 5 sleep 1 1944 # echo 0 > tracing_on 1945 # cat trace 1946 # tracer: wakeup 1947 # 1948 # wakeup latency trace v1.1.5 on 3.8.0-test+ 1949 # -------------------------------------------------------------------- 1950 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1951 # ----------------- 1952 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0) 1953 # ----------------- 1954 # 1955 # _------=> CPU# 1956 # / _-----=> irqs-off 1957 # | / _----=> need-resched 1958 # || / _---=> hardirq/softirq 1959 # ||| / _--=> preempt-depth 1960 # |||| / delay 1961 # cmd pid ||||| time | caller 1962 # \ / ||||| \ | / 1963 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H 1964 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up 1965 <idle>-0 3d..3 15us : __schedule <-schedule 1966 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H 1967 1968The tracer only traces the highest priority task in the system 1969to avoid tracing the normal circumstances. Here we see that 1970the kworker with a nice priority of -20 (not very nice), took 1971just 15 microseconds from the time it woke up, to the time it 1972ran. 1973 1974Non Real-Time tasks are not that interesting. A more interesting 1975trace is to concentrate only on Real-Time tasks. 1976 1977wakeup_rt 1978--------- 1979 1980In a Real-Time environment it is very important to know the 1981wakeup time it takes for the highest priority task that is woken 1982up to the time that it executes. This is also known as "schedule 1983latency". I stress the point that this is about RT tasks. It is 1984also important to know the scheduling latency of non-RT tasks, 1985but the average schedule latency is better for non-RT tasks. 1986Tools like LatencyTop are more appropriate for such 1987measurements. 1988 1989Real-Time environments are interested in the worst case latency. 1990That is the longest latency it takes for something to happen, 1991and not the average. We can have a very fast scheduler that may 1992only have a large latency once in a while, but that would not 1993work well with Real-Time tasks. The wakeup_rt tracer was designed 1994to record the worst case wakeups of RT tasks. Non-RT tasks are 1995not recorded because the tracer only records one worst case and 1996tracing non-RT tasks that are unpredictable will overwrite the 1997worst case latency of RT tasks (just run the normal wakeup 1998tracer for a while to see that effect). 1999 2000Since this tracer only deals with RT tasks, we will run this 2001slightly differently than we did with the previous tracers. 2002Instead of performing an 'ls', we will run 'sleep 1' under 2003'chrt' which changes the priority of the task. 2004:: 2005 2006 # echo 0 > options/function-trace 2007 # echo wakeup_rt > current_tracer 2008 # echo 1 > tracing_on 2009 # echo 0 > tracing_max_latency 2010 # chrt -f 5 sleep 1 2011 # echo 0 > tracing_on 2012 # cat trace 2013 # tracer: wakeup 2014 # 2015 # tracer: wakeup_rt 2016 # 2017 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2018 # -------------------------------------------------------------------- 2019 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2020 # ----------------- 2021 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5) 2022 # ----------------- 2023 # 2024 # _------=> CPU# 2025 # / _-----=> irqs-off 2026 # | / _----=> need-resched 2027 # || / _---=> hardirq/softirq 2028 # ||| / _--=> preempt-depth 2029 # |||| / delay 2030 # cmd pid ||||| time | caller 2031 # \ / ||||| \ | / 2032 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep 2033 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up 2034 <idle>-0 3d..3 5us : __schedule <-schedule 2035 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep 2036 2037 2038Running this on an idle system, we see that it only took 5 microseconds 2039to perform the task switch. Note, since the trace point in the schedule 2040is before the actual "switch", we stop the tracing when the recorded task 2041is about to schedule in. This may change if we add a new marker at the 2042end of the scheduler. 2043 2044Notice that the recorded task is 'sleep' with the PID of 2389 2045and it has an rt_prio of 5. This priority is user-space priority 2046and not the internal kernel priority. The policy is 1 for 2047SCHED_FIFO and 2 for SCHED_RR. 2048 2049Note, that the trace data shows the internal priority (99 - rtprio). 2050:: 2051 2052 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep 2053 2054The 0:120:R means idle was running with a nice priority of 0 (120 - 120) 2055and in the running state 'R'. The sleep task was scheduled in with 20562389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94) 2057and it too is in the running state. 2058 2059Doing the same with chrt -r 5 and function-trace set. 2060:: 2061 2062 echo 1 > options/function-trace 2063 2064 # tracer: wakeup_rt 2065 # 2066 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2067 # -------------------------------------------------------------------- 2068 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2069 # ----------------- 2070 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5) 2071 # ----------------- 2072 # 2073 # _------=> CPU# 2074 # / _-----=> irqs-off 2075 # | / _----=> need-resched 2076 # || / _---=> hardirq/softirq 2077 # ||| / _--=> preempt-depth 2078 # |||| / delay 2079 # cmd pid ||||| time | caller 2080 # \ / ||||| \ | / 2081 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep 2082 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up 2083 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup 2084 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr 2085 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup 2086 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up 2087 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock 2088 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up 2089 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up 2090 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2091 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer 2092 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock 2093 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt 2094 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock 2095 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt 2096 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event 2097 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event 2098 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event 2099 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt 2100 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit 2101 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit 2102 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit 2103 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit 2104 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit 2105 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle 2106 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit 2107 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle 2108 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit 2109 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit 2110 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit 2111 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit 2112 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz 2113 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock 2114 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz 2115 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update 2116 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz 2117 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock 2118 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit 2119 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit 2120 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit 2121 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel 2122 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel 2123 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18 2124 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave 2125 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16 2126 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer 2127 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram 2128 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event 2129 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event 2130 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event 2131 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel 2132 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2133 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit 2134 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward 2135 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward 2136 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11 2137 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns 2138 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns 2139 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18 2140 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave 2141 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns 2142 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns 2143 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns 2144 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event 2145 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event 2146 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event 2147 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns 2148 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2149 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit 2150 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks 2151 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle 2152 <idle>-0 3.N.. 25us : schedule <-cpu_idle 2153 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule 2154 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule 2155 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule 2156 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch 2157 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch 2158 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule 2159 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq 2160 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule 2161 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task 2162 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task 2163 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt 2164 <idle>-0 3d..3 29us : __schedule <-preempt_schedule 2165 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep 2166 2167This isn't that big of a trace, even with function tracing enabled, 2168so I included the entire trace. 2169 2170The interrupt went off while when the system was idle. Somewhere 2171before task_woken_rt() was called, the NEED_RESCHED flag was set, 2172this is indicated by the first occurrence of the 'N' flag. 2173 2174Latency tracing and events 2175-------------------------- 2176As function tracing can induce a much larger latency, but without 2177seeing what happens within the latency it is hard to know what 2178caused it. There is a middle ground, and that is with enabling 2179events. 2180:: 2181 2182 # echo 0 > options/function-trace 2183 # echo wakeup_rt > current_tracer 2184 # echo 1 > events/enable 2185 # echo 1 > tracing_on 2186 # echo 0 > tracing_max_latency 2187 # chrt -f 5 sleep 1 2188 # echo 0 > tracing_on 2189 # cat trace 2190 # tracer: wakeup_rt 2191 # 2192 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2193 # -------------------------------------------------------------------- 2194 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2195 # ----------------- 2196 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5) 2197 # ----------------- 2198 # 2199 # _------=> CPU# 2200 # / _-----=> irqs-off 2201 # | / _----=> need-resched 2202 # || / _---=> hardirq/softirq 2203 # ||| / _--=> preempt-depth 2204 # |||| / delay 2205 # cmd pid ||||| time | caller 2206 # \ / ||||| \ | / 2207 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep 2208 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up 2209 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002 2210 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8 2211 <idle>-0 2.N.2 2us : power_end: cpu_id=2 2212 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2 2213 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0 2214 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000 2215 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch 2216 <idle>-0 2.N.2 5us : rcu_utilization: End context switch 2217 <idle>-0 2d..3 6us : __schedule <-schedule 2218 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep 2219 2220 2221Hardware Latency Detector 2222------------------------- 2223 2224The hardware latency detector is executed by enabling the "hwlat" tracer. 2225 2226NOTE, this tracer will affect the performance of the system as it will 2227periodically make a CPU constantly busy with interrupts disabled. 2228:: 2229 2230 # echo hwlat > current_tracer 2231 # sleep 100 2232 # cat trace 2233 # tracer: hwlat 2234 # 2235 # entries-in-buffer/entries-written: 13/13 #P:8 2236 # 2237 # _-----=> irqs-off 2238 # / _----=> need-resched 2239 # | / _---=> hardirq/softirq 2240 # || / _--=> preempt-depth 2241 # ||| / delay 2242 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2243 # | | | |||| | | 2244 <...>-1729 [001] d... 678.473449: #1 inner/outer(us): 11/12 ts:1581527483.343962693 count:6 2245 <...>-1729 [004] d... 689.556542: #2 inner/outer(us): 16/9 ts:1581527494.889008092 count:1 2246 <...>-1729 [005] d... 714.756290: #3 inner/outer(us): 16/16 ts:1581527519.678961629 count:5 2247 <...>-1729 [001] d... 718.788247: #4 inner/outer(us): 9/17 ts:1581527523.889012713 count:1 2248 <...>-1729 [002] d... 719.796341: #5 inner/outer(us): 13/9 ts:1581527524.912872606 count:1 2249 <...>-1729 [006] d... 844.787091: #6 inner/outer(us): 9/12 ts:1581527649.889048502 count:2 2250 <...>-1729 [003] d... 849.827033: #7 inner/outer(us): 18/9 ts:1581527654.889013793 count:1 2251 <...>-1729 [007] d... 853.859002: #8 inner/outer(us): 9/12 ts:1581527658.889065736 count:1 2252 <...>-1729 [001] d... 855.874978: #9 inner/outer(us): 9/11 ts:1581527660.861991877 count:1 2253 <...>-1729 [001] d... 863.938932: #10 inner/outer(us): 9/11 ts:1581527668.970010500 count:1 nmi-total:7 nmi-count:1 2254 <...>-1729 [007] d... 878.050780: #11 inner/outer(us): 9/12 ts:1581527683.385002600 count:1 nmi-total:5 nmi-count:1 2255 <...>-1729 [007] d... 886.114702: #12 inner/outer(us): 9/12 ts:1581527691.385001600 count:1 2256 2257 2258The above output is somewhat the same in the header. All events will have 2259interrupts disabled 'd'. Under the FUNCTION title there is: 2260 2261 #1 2262 This is the count of events recorded that were greater than the 2263 tracing_threshold (See below). 2264 2265 inner/outer(us): 11/11 2266 2267 This shows two numbers as "inner latency" and "outer latency". The test 2268 runs in a loop checking a timestamp twice. The latency detected within 2269 the two timestamps is the "inner latency" and the latency detected 2270 after the previous timestamp and the next timestamp in the loop is 2271 the "outer latency". 2272 2273 ts:1581527483.343962693 2274 2275 The absolute timestamp that the first latency was recorded in the window. 2276 2277 count:6 2278 2279 The number of times a latency was detected during the window. 2280 2281 nmi-total:7 nmi-count:1 2282 2283 On architectures that support it, if an NMI comes in during the 2284 test, the time spent in NMI is reported in "nmi-total" (in 2285 microseconds). 2286 2287 All architectures that have NMIs will show the "nmi-count" if an 2288 NMI comes in during the test. 2289 2290hwlat files: 2291 2292 tracing_threshold 2293 This gets automatically set to "10" to represent 10 2294 microseconds. This is the threshold of latency that 2295 needs to be detected before the trace will be recorded. 2296 2297 Note, when hwlat tracer is finished (another tracer is 2298 written into "current_tracer"), the original value for 2299 tracing_threshold is placed back into this file. 2300 2301 hwlat_detector/width 2302 The length of time the test runs with interrupts disabled. 2303 2304 hwlat_detector/window 2305 The length of time of the window which the test 2306 runs. That is, the test will run for "width" 2307 microseconds per "window" microseconds 2308 2309 tracing_cpumask 2310 When the test is started. A kernel thread is created that 2311 runs the test. This thread will alternate between CPUs 2312 listed in the tracing_cpumask between each period 2313 (one "window"). To limit the test to specific CPUs 2314 set the mask in this file to only the CPUs that the test 2315 should run on. 2316 2317function 2318-------- 2319 2320This tracer is the function tracer. Enabling the function tracer 2321can be done from the debug file system. Make sure the 2322ftrace_enabled is set; otherwise this tracer is a nop. 2323See the "ftrace_enabled" section below. 2324:: 2325 2326 # sysctl kernel.ftrace_enabled=1 2327 # echo function > current_tracer 2328 # echo 1 > tracing_on 2329 # usleep 1 2330 # echo 0 > tracing_on 2331 # cat trace 2332 # tracer: function 2333 # 2334 # entries-in-buffer/entries-written: 24799/24799 #P:4 2335 # 2336 # _-----=> irqs-off 2337 # / _----=> need-resched 2338 # | / _---=> hardirq/softirq 2339 # || / _--=> preempt-depth 2340 # ||| / delay 2341 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2342 # | | | |||| | | 2343 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write 2344 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock 2345 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify 2346 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify 2347 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify 2348 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock 2349 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock 2350 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify 2351 [...] 2352 2353 2354Note: function tracer uses ring buffers to store the above 2355entries. The newest data may overwrite the oldest data. 2356Sometimes using echo to stop the trace is not sufficient because 2357the tracing could have overwritten the data that you wanted to 2358record. For this reason, it is sometimes better to disable 2359tracing directly from a program. This allows you to stop the 2360tracing at the point that you hit the part that you are 2361interested in. To disable the tracing directly from a C program, 2362something like following code snippet can be used:: 2363 2364 int trace_fd; 2365 [...] 2366 int main(int argc, char *argv[]) { 2367 [...] 2368 trace_fd = open(tracing_file("tracing_on"), O_WRONLY); 2369 [...] 2370 if (condition_hit()) { 2371 write(trace_fd, "0", 1); 2372 } 2373 [...] 2374 } 2375 2376 2377Single thread tracing 2378--------------------- 2379 2380By writing into set_ftrace_pid you can trace a 2381single thread. For example:: 2382 2383 # cat set_ftrace_pid 2384 no pid 2385 # echo 3111 > set_ftrace_pid 2386 # cat set_ftrace_pid 2387 3111 2388 # echo function > current_tracer 2389 # cat trace | head 2390 # tracer: function 2391 # 2392 # TASK-PID CPU# TIMESTAMP FUNCTION 2393 # | | | | | 2394 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return 2395 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range 2396 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel 2397 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel 2398 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll 2399 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll 2400 # echo > set_ftrace_pid 2401 # cat trace |head 2402 # tracer: function 2403 # 2404 # TASK-PID CPU# TIMESTAMP FUNCTION 2405 # | | | | | 2406 ##### CPU 3 buffer started #### 2407 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait 2408 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry 2409 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry 2410 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit 2411 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit 2412 2413If you want to trace a function when executing, you could use 2414something like this simple program. 2415:: 2416 2417 #include <stdio.h> 2418 #include <stdlib.h> 2419 #include <sys/types.h> 2420 #include <sys/stat.h> 2421 #include <fcntl.h> 2422 #include <unistd.h> 2423 #include <string.h> 2424 2425 #define _STR(x) #x 2426 #define STR(x) _STR(x) 2427 #define MAX_PATH 256 2428 2429 const char *find_tracefs(void) 2430 { 2431 static char tracefs[MAX_PATH+1]; 2432 static int tracefs_found; 2433 char type[100]; 2434 FILE *fp; 2435 2436 if (tracefs_found) 2437 return tracefs; 2438 2439 if ((fp = fopen("/proc/mounts","r")) == NULL) { 2440 perror("/proc/mounts"); 2441 return NULL; 2442 } 2443 2444 while (fscanf(fp, "%*s %" 2445 STR(MAX_PATH) 2446 "s %99s %*s %*d %*d\n", 2447 tracefs, type) == 2) { 2448 if (strcmp(type, "tracefs") == 0) 2449 break; 2450 } 2451 fclose(fp); 2452 2453 if (strcmp(type, "tracefs") != 0) { 2454 fprintf(stderr, "tracefs not mounted"); 2455 return NULL; 2456 } 2457 2458 strcat(tracefs, "/tracing/"); 2459 tracefs_found = 1; 2460 2461 return tracefs; 2462 } 2463 2464 const char *tracing_file(const char *file_name) 2465 { 2466 static char trace_file[MAX_PATH+1]; 2467 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name); 2468 return trace_file; 2469 } 2470 2471 int main (int argc, char **argv) 2472 { 2473 if (argc < 1) 2474 exit(-1); 2475 2476 if (fork() > 0) { 2477 int fd, ffd; 2478 char line[64]; 2479 int s; 2480 2481 ffd = open(tracing_file("current_tracer"), O_WRONLY); 2482 if (ffd < 0) 2483 exit(-1); 2484 write(ffd, "nop", 3); 2485 2486 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY); 2487 s = sprintf(line, "%d\n", getpid()); 2488 write(fd, line, s); 2489 2490 write(ffd, "function", 8); 2491 2492 close(fd); 2493 close(ffd); 2494 2495 execvp(argv[1], argv+1); 2496 } 2497 2498 return 0; 2499 } 2500 2501Or this simple script! 2502:: 2503 2504 #!/bin/bash 2505 2506 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts` 2507 echo 0 > $tracefs/tracing_on 2508 echo $$ > $tracefs/set_ftrace_pid 2509 echo function > $tracefs/current_tracer 2510 echo 1 > $tracefs/tracing_on 2511 exec "$@" 2512 2513 2514function graph tracer 2515--------------------------- 2516 2517This tracer is similar to the function tracer except that it 2518probes a function on its entry and its exit. This is done by 2519using a dynamically allocated stack of return addresses in each 2520task_struct. On function entry the tracer overwrites the return 2521address of each function traced to set a custom probe. Thus the 2522original return address is stored on the stack of return address 2523in the task_struct. 2524 2525Probing on both ends of a function leads to special features 2526such as: 2527 2528- measure of a function's time execution 2529- having a reliable call stack to draw function calls graph 2530 2531This tracer is useful in several situations: 2532 2533- you want to find the reason of a strange kernel behavior and 2534 need to see what happens in detail on any areas (or specific 2535 ones). 2536 2537- you are experiencing weird latencies but it's difficult to 2538 find its origin. 2539 2540- you want to find quickly which path is taken by a specific 2541 function 2542 2543- you just want to peek inside a working kernel and want to see 2544 what happens there. 2545 2546:: 2547 2548 # tracer: function_graph 2549 # 2550 # CPU DURATION FUNCTION CALLS 2551 # | | | | | | | 2552 2553 0) | sys_open() { 2554 0) | do_sys_open() { 2555 0) | getname() { 2556 0) | kmem_cache_alloc() { 2557 0) 1.382 us | __might_sleep(); 2558 0) 2.478 us | } 2559 0) | strncpy_from_user() { 2560 0) | might_fault() { 2561 0) 1.389 us | __might_sleep(); 2562 0) 2.553 us | } 2563 0) 3.807 us | } 2564 0) 7.876 us | } 2565 0) | alloc_fd() { 2566 0) 0.668 us | _spin_lock(); 2567 0) 0.570 us | expand_files(); 2568 0) 0.586 us | _spin_unlock(); 2569 2570 2571There are several columns that can be dynamically 2572enabled/disabled. You can use every combination of options you 2573want, depending on your needs. 2574 2575- The cpu number on which the function executed is default 2576 enabled. It is sometimes better to only trace one cpu (see 2577 tracing_cpu_mask file) or you might sometimes see unordered 2578 function calls while cpu tracing switch. 2579 2580 - hide: echo nofuncgraph-cpu > trace_options 2581 - show: echo funcgraph-cpu > trace_options 2582 2583- The duration (function's time of execution) is displayed on 2584 the closing bracket line of a function or on the same line 2585 than the current function in case of a leaf one. It is default 2586 enabled. 2587 2588 - hide: echo nofuncgraph-duration > trace_options 2589 - show: echo funcgraph-duration > trace_options 2590 2591- The overhead field precedes the duration field in case of 2592 reached duration thresholds. 2593 2594 - hide: echo nofuncgraph-overhead > trace_options 2595 - show: echo funcgraph-overhead > trace_options 2596 - depends on: funcgraph-duration 2597 2598 ie:: 2599 2600 3) # 1837.709 us | } /* __switch_to */ 2601 3) | finish_task_switch() { 2602 3) 0.313 us | _raw_spin_unlock_irq(); 2603 3) 3.177 us | } 2604 3) # 1889.063 us | } /* __schedule */ 2605 3) ! 140.417 us | } /* __schedule */ 2606 3) # 2034.948 us | } /* schedule */ 2607 3) * 33998.59 us | } /* schedule_preempt_disabled */ 2608 2609 [...] 2610 2611 1) 0.260 us | msecs_to_jiffies(); 2612 1) 0.313 us | __rcu_read_unlock(); 2613 1) + 61.770 us | } 2614 1) + 64.479 us | } 2615 1) 0.313 us | rcu_bh_qs(); 2616 1) 0.313 us | __local_bh_enable(); 2617 1) ! 217.240 us | } 2618 1) 0.365 us | idle_cpu(); 2619 1) | rcu_irq_exit() { 2620 1) 0.417 us | rcu_eqs_enter_common.isra.47(); 2621 1) 3.125 us | } 2622 1) ! 227.812 us | } 2623 1) ! 457.395 us | } 2624 1) @ 119760.2 us | } 2625 2626 [...] 2627 2628 2) | handle_IPI() { 2629 1) 6.979 us | } 2630 2) 0.417 us | scheduler_ipi(); 2631 1) 9.791 us | } 2632 1) + 12.917 us | } 2633 2) 3.490 us | } 2634 1) + 15.729 us | } 2635 1) + 18.542 us | } 2636 2) $ 3594274 us | } 2637 2638Flags:: 2639 2640 + means that the function exceeded 10 usecs. 2641 ! means that the function exceeded 100 usecs. 2642 # means that the function exceeded 1000 usecs. 2643 * means that the function exceeded 10 msecs. 2644 @ means that the function exceeded 100 msecs. 2645 $ means that the function exceeded 1 sec. 2646 2647 2648- The task/pid field displays the thread cmdline and pid which 2649 executed the function. It is default disabled. 2650 2651 - hide: echo nofuncgraph-proc > trace_options 2652 - show: echo funcgraph-proc > trace_options 2653 2654 ie:: 2655 2656 # tracer: function_graph 2657 # 2658 # CPU TASK/PID DURATION FUNCTION CALLS 2659 # | | | | | | | | | 2660 0) sh-4802 | | d_free() { 2661 0) sh-4802 | | call_rcu() { 2662 0) sh-4802 | | __call_rcu() { 2663 0) sh-4802 | 0.616 us | rcu_process_gp_end(); 2664 0) sh-4802 | 0.586 us | check_for_new_grace_period(); 2665 0) sh-4802 | 2.899 us | } 2666 0) sh-4802 | 4.040 us | } 2667 0) sh-4802 | 5.151 us | } 2668 0) sh-4802 | + 49.370 us | } 2669 2670 2671- The absolute time field is an absolute timestamp given by the 2672 system clock since it started. A snapshot of this time is 2673 given on each entry/exit of functions 2674 2675 - hide: echo nofuncgraph-abstime > trace_options 2676 - show: echo funcgraph-abstime > trace_options 2677 2678 ie:: 2679 2680 # 2681 # TIME CPU DURATION FUNCTION CALLS 2682 # | | | | | | | | 2683 360.774522 | 1) 0.541 us | } 2684 360.774522 | 1) 4.663 us | } 2685 360.774523 | 1) 0.541 us | __wake_up_bit(); 2686 360.774524 | 1) 6.796 us | } 2687 360.774524 | 1) 7.952 us | } 2688 360.774525 | 1) 9.063 us | } 2689 360.774525 | 1) 0.615 us | journal_mark_dirty(); 2690 360.774527 | 1) 0.578 us | __brelse(); 2691 360.774528 | 1) | reiserfs_prepare_for_journal() { 2692 360.774528 | 1) | unlock_buffer() { 2693 360.774529 | 1) | wake_up_bit() { 2694 360.774529 | 1) | bit_waitqueue() { 2695 360.774530 | 1) 0.594 us | __phys_addr(); 2696 2697 2698The function name is always displayed after the closing bracket 2699for a function if the start of that function is not in the 2700trace buffer. 2701 2702Display of the function name after the closing bracket may be 2703enabled for functions whose start is in the trace buffer, 2704allowing easier searching with grep for function durations. 2705It is default disabled. 2706 2707 - hide: echo nofuncgraph-tail > trace_options 2708 - show: echo funcgraph-tail > trace_options 2709 2710 Example with nofuncgraph-tail (default):: 2711 2712 0) | putname() { 2713 0) | kmem_cache_free() { 2714 0) 0.518 us | __phys_addr(); 2715 0) 1.757 us | } 2716 0) 2.861 us | } 2717 2718 Example with funcgraph-tail:: 2719 2720 0) | putname() { 2721 0) | kmem_cache_free() { 2722 0) 0.518 us | __phys_addr(); 2723 0) 1.757 us | } /* kmem_cache_free() */ 2724 0) 2.861 us | } /* putname() */ 2725 2726The return value of each traced function can be displayed after 2727an equal sign "=". When encountering system call failures, it 2728can be very helpful to quickly locate the function that first 2729returns an error code. 2730 2731 - hide: echo nofuncgraph-retval > trace_options 2732 - show: echo funcgraph-retval > trace_options 2733 2734 Example with funcgraph-retval:: 2735 2736 1) | cgroup_migrate() { 2737 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */ 2738 1) | cgroup_migrate_execute() { 2739 1) | cpu_cgroup_can_attach() { 2740 1) | cgroup_taskset_first() { 2741 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */ 2742 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */ 2743 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */ 2744 1) 2.335 us | } /* cpu_cgroup_can_attach = -22 */ 2745 1) 4.369 us | } /* cgroup_migrate_execute = -22 */ 2746 1) 7.143 us | } /* cgroup_migrate = -22 */ 2747 2748The above example shows that the function cpu_cgroup_can_attach 2749returned the error code -22 firstly, then we can read the code 2750of this function to get the root cause. 2751 2752When the option funcgraph-retval-hex is not set, the return value can 2753be displayed in a smart way. Specifically, if it is an error code, 2754it will be printed in signed decimal format, otherwise it will 2755printed in hexadecimal format. 2756 2757 - smart: echo nofuncgraph-retval-hex > trace_options 2758 - hexadecimal: echo funcgraph-retval-hex > trace_options 2759 2760 Example with funcgraph-retval-hex:: 2761 2762 1) | cgroup_migrate() { 2763 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */ 2764 1) | cgroup_migrate_execute() { 2765 1) | cpu_cgroup_can_attach() { 2766 1) | cgroup_taskset_first() { 2767 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */ 2768 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */ 2769 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */ 2770 1) 2.335 us | } /* cpu_cgroup_can_attach = 0xffffffea */ 2771 1) 4.369 us | } /* cgroup_migrate_execute = 0xffffffea */ 2772 1) 7.143 us | } /* cgroup_migrate = 0xffffffea */ 2773 2774At present, there are some limitations when using the funcgraph-retval 2775option, and these limitations will be eliminated in the future: 2776 2777- Even if the function return type is void, a return value will still 2778 be printed, and you can just ignore it. 2779 2780- Even if return values are stored in multiple registers, only the 2781 value contained in the first register will be recorded and printed. 2782 To illustrate, in the x86 architecture, eax and edx are used to store 2783 a 64-bit return value, with the lower 32 bits saved in eax and the 2784 upper 32 bits saved in edx. However, only the value stored in eax 2785 will be recorded and printed. 2786 2787- In certain procedure call standards, such as arm64's AAPCS64, when a 2788 type is smaller than a GPR, it is the responsibility of the consumer 2789 to perform the narrowing, and the upper bits may contain UNKNOWN values. 2790 Therefore, it is advisable to check the code for such cases. For instance, 2791 when using a u8 in a 64-bit GPR, bits [63:8] may contain arbitrary values, 2792 especially when larger types are truncated, whether explicitly or implicitly. 2793 Here are some specific cases to illustrate this point: 2794 2795 **Case One**: 2796 2797 The function narrow_to_u8 is defined as follows:: 2798 2799 u8 narrow_to_u8(u64 val) 2800 { 2801 // implicitly truncated 2802 return val; 2803 } 2804 2805 It may be compiled to:: 2806 2807 narrow_to_u8: 2808 < ... ftrace instrumentation ... > 2809 RET 2810 2811 If you pass 0x123456789abcdef to this function and want to narrow it, 2812 it may be recorded as 0x123456789abcdef instead of 0xef. 2813 2814 **Case Two**: 2815 2816 The function error_if_not_4g_aligned is defined as follows:: 2817 2818 int error_if_not_4g_aligned(u64 val) 2819 { 2820 if (val & GENMASK(31, 0)) 2821 return -EINVAL; 2822 2823 return 0; 2824 } 2825 2826 It could be compiled to:: 2827 2828 error_if_not_4g_aligned: 2829 CBNZ w0, .Lnot_aligned 2830 RET // bits [31:0] are zero, bits 2831 // [63:32] are UNKNOWN 2832 .Lnot_aligned: 2833 MOV x0, #-EINVAL 2834 RET 2835 2836 When passing 0x2_0000_0000 to it, the return value may be recorded as 2837 0x2_0000_0000 instead of 0. 2838 2839You can put some comments on specific functions by using 2840trace_printk() For example, if you want to put a comment inside 2841the __might_sleep() function, you just have to include 2842<linux/ftrace.h> and call trace_printk() inside __might_sleep():: 2843 2844 trace_printk("I'm a comment!\n") 2845 2846will produce:: 2847 2848 1) | __might_sleep() { 2849 1) | /* I'm a comment! */ 2850 1) 1.449 us | } 2851 2852 2853You might find other useful features for this tracer in the 2854following "dynamic ftrace" section such as tracing only specific 2855functions or tasks. 2856 2857dynamic ftrace 2858-------------- 2859 2860If CONFIG_DYNAMIC_FTRACE is set, the system will run with 2861virtually no overhead when function tracing is disabled. The way 2862this works is the mcount function call (placed at the start of 2863every kernel function, produced by the -pg switch in gcc), 2864starts of pointing to a simple return. (Enabling FTRACE will 2865include the -pg switch in the compiling of the kernel.) 2866 2867At compile time every C file object is run through the 2868recordmcount program (located in the scripts directory). This 2869program will parse the ELF headers in the C object to find all 2870the locations in the .text section that call mcount. Starting 2871with gcc version 4.6, the -mfentry has been added for x86, which 2872calls "__fentry__" instead of "mcount". Which is called before 2873the creation of the stack frame. 2874 2875Note, not all sections are traced. They may be prevented by either 2876a notrace, or blocked another way and all inline functions are not 2877traced. Check the "available_filter_functions" file to see what functions 2878can be traced. 2879 2880A section called "__mcount_loc" is created that holds 2881references to all the mcount/fentry call sites in the .text section. 2882The recordmcount program re-links this section back into the 2883original object. The final linking stage of the kernel will add all these 2884references into a single table. 2885 2886On boot up, before SMP is initialized, the dynamic ftrace code 2887scans this table and updates all the locations into nops. It 2888also records the locations, which are added to the 2889available_filter_functions list. Modules are processed as they 2890are loaded and before they are executed. When a module is 2891unloaded, it also removes its functions from the ftrace function 2892list. This is automatic in the module unload code, and the 2893module author does not need to worry about it. 2894 2895When tracing is enabled, the process of modifying the function 2896tracepoints is dependent on architecture. The old method is to use 2897kstop_machine to prevent races with the CPUs executing code being 2898modified (which can cause the CPU to do undesirable things, especially 2899if the modified code crosses cache (or page) boundaries), and the nops are 2900patched back to calls. But this time, they do not call mcount 2901(which is just a function stub). They now call into the ftrace 2902infrastructure. 2903 2904The new method of modifying the function tracepoints is to place 2905a breakpoint at the location to be modified, sync all CPUs, modify 2906the rest of the instruction not covered by the breakpoint. Sync 2907all CPUs again, and then remove the breakpoint with the finished 2908version to the ftrace call site. 2909 2910Some archs do not even need to monkey around with the synchronization, 2911and can just slap the new code on top of the old without any 2912problems with other CPUs executing it at the same time. 2913 2914One special side-effect to the recording of the functions being 2915traced is that we can now selectively choose which functions we 2916wish to trace and which ones we want the mcount calls to remain 2917as nops. 2918 2919Two files are used, one for enabling and one for disabling the 2920tracing of specified functions. They are: 2921 2922 set_ftrace_filter 2923 2924and 2925 2926 set_ftrace_notrace 2927 2928A list of available functions that you can add to these files is 2929listed in: 2930 2931 available_filter_functions 2932 2933:: 2934 2935 # cat available_filter_functions 2936 put_prev_task_idle 2937 kmem_cache_create 2938 pick_next_task_rt 2939 cpus_read_lock 2940 pick_next_task_fair 2941 mutex_lock 2942 [...] 2943 2944If I am only interested in sys_nanosleep and hrtimer_interrupt:: 2945 2946 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter 2947 # echo function > current_tracer 2948 # echo 1 > tracing_on 2949 # usleep 1 2950 # echo 0 > tracing_on 2951 # cat trace 2952 # tracer: function 2953 # 2954 # entries-in-buffer/entries-written: 5/5 #P:4 2955 # 2956 # _-----=> irqs-off 2957 # / _----=> need-resched 2958 # | / _---=> hardirq/softirq 2959 # || / _--=> preempt-depth 2960 # ||| / delay 2961 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2962 # | | | |||| | | 2963 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath 2964 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt 2965 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt 2966 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt 2967 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt 2968 2969To see which functions are being traced, you can cat the file: 2970:: 2971 2972 # cat set_ftrace_filter 2973 hrtimer_interrupt 2974 sys_nanosleep 2975 2976 2977Perhaps this is not enough. The filters also allow glob(7) matching. 2978 2979 ``<match>*`` 2980 will match functions that begin with <match> 2981 ``*<match>`` 2982 will match functions that end with <match> 2983 ``*<match>*`` 2984 will match functions that have <match> in it 2985 ``<match1>*<match2>`` 2986 will match functions that begin with <match1> and end with <match2> 2987 2988.. note:: 2989 It is better to use quotes to enclose the wild cards, 2990 otherwise the shell may expand the parameters into names 2991 of files in the local directory. 2992 2993:: 2994 2995 # echo 'hrtimer_*' > set_ftrace_filter 2996 2997Produces:: 2998 2999 # tracer: function 3000 # 3001 # entries-in-buffer/entries-written: 897/897 #P:4 3002 # 3003 # _-----=> irqs-off 3004 # / _----=> need-resched 3005 # | / _---=> hardirq/softirq 3006 # || / _--=> preempt-depth 3007 # ||| / delay 3008 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3009 # | | | |||| | | 3010 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit 3011 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel 3012 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer 3013 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit 3014 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11 3015 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt 3016 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter 3017 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem 3018 3019Notice that we lost the sys_nanosleep. 3020:: 3021 3022 # cat set_ftrace_filter 3023 hrtimer_run_queues 3024 hrtimer_run_pending 3025 hrtimer_init 3026 hrtimer_cancel 3027 hrtimer_try_to_cancel 3028 hrtimer_forward 3029 hrtimer_start 3030 hrtimer_reprogram 3031 hrtimer_force_reprogram 3032 hrtimer_get_next_event 3033 hrtimer_interrupt 3034 hrtimer_nanosleep 3035 hrtimer_wakeup 3036 hrtimer_get_remaining 3037 hrtimer_get_res 3038 hrtimer_init_sleeper 3039 3040 3041This is because the '>' and '>>' act just like they do in bash. 3042To rewrite the filters, use '>' 3043To append to the filters, use '>>' 3044 3045To clear out a filter so that all functions will be recorded 3046again:: 3047 3048 # echo > set_ftrace_filter 3049 # cat set_ftrace_filter 3050 # 3051 3052Again, now we want to append. 3053 3054:: 3055 3056 # echo sys_nanosleep > set_ftrace_filter 3057 # cat set_ftrace_filter 3058 sys_nanosleep 3059 # echo 'hrtimer_*' >> set_ftrace_filter 3060 # cat set_ftrace_filter 3061 hrtimer_run_queues 3062 hrtimer_run_pending 3063 hrtimer_init 3064 hrtimer_cancel 3065 hrtimer_try_to_cancel 3066 hrtimer_forward 3067 hrtimer_start 3068 hrtimer_reprogram 3069 hrtimer_force_reprogram 3070 hrtimer_get_next_event 3071 hrtimer_interrupt 3072 sys_nanosleep 3073 hrtimer_nanosleep 3074 hrtimer_wakeup 3075 hrtimer_get_remaining 3076 hrtimer_get_res 3077 hrtimer_init_sleeper 3078 3079 3080The set_ftrace_notrace prevents those functions from being 3081traced. 3082:: 3083 3084 # echo '*preempt*' '*lock*' > set_ftrace_notrace 3085 3086Produces:: 3087 3088 # tracer: function 3089 # 3090 # entries-in-buffer/entries-written: 39608/39608 #P:4 3091 # 3092 # _-----=> irqs-off 3093 # / _----=> need-resched 3094 # | / _---=> hardirq/softirq 3095 # || / _--=> preempt-depth 3096 # ||| / delay 3097 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3098 # | | | |||| | | 3099 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open 3100 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last 3101 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last 3102 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check 3103 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement 3104 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action 3105 bash-1994 [000] .... 4342.324899: do_truncate <-do_last 3106 bash-1994 [000] .... 4342.324899: setattr_should_drop_suidgid <-do_truncate 3107 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate 3108 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change 3109 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time 3110 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time 3111 3112We can see that there's no more lock or preempt tracing. 3113 3114Selecting function filters via index 3115------------------------------------ 3116 3117Because processing of strings is expensive (the address of the function 3118needs to be looked up before comparing to the string being passed in), 3119an index can be used as well to enable functions. This is useful in the 3120case of setting thousands of specific functions at a time. By passing 3121in a list of numbers, no string processing will occur. Instead, the function 3122at the specific location in the internal array (which corresponds to the 3123functions in the "available_filter_functions" file), is selected. 3124 3125:: 3126 3127 # echo 1 > set_ftrace_filter 3128 3129Will select the first function listed in "available_filter_functions" 3130 3131:: 3132 3133 # head -1 available_filter_functions 3134 trace_initcall_finish_cb 3135 3136 # cat set_ftrace_filter 3137 trace_initcall_finish_cb 3138 3139 # head -50 available_filter_functions | tail -1 3140 x86_pmu_commit_txn 3141 3142 # echo 1 50 > set_ftrace_filter 3143 # cat set_ftrace_filter 3144 trace_initcall_finish_cb 3145 x86_pmu_commit_txn 3146 3147Dynamic ftrace with the function graph tracer 3148--------------------------------------------- 3149 3150Although what has been explained above concerns both the 3151function tracer and the function-graph-tracer, there are some 3152special features only available in the function-graph tracer. 3153 3154If you want to trace only one function and all of its children, 3155you just have to echo its name into set_graph_function:: 3156 3157 echo __do_fault > set_graph_function 3158 3159will produce the following "expanded" trace of the __do_fault() 3160function:: 3161 3162 0) | __do_fault() { 3163 0) | filemap_fault() { 3164 0) | find_lock_page() { 3165 0) 0.804 us | find_get_page(); 3166 0) | __might_sleep() { 3167 0) 1.329 us | } 3168 0) 3.904 us | } 3169 0) 4.979 us | } 3170 0) 0.653 us | _spin_lock(); 3171 0) 0.578 us | page_add_file_rmap(); 3172 0) 0.525 us | native_set_pte_at(); 3173 0) 0.585 us | _spin_unlock(); 3174 0) | unlock_page() { 3175 0) 0.541 us | page_waitqueue(); 3176 0) 0.639 us | __wake_up_bit(); 3177 0) 2.786 us | } 3178 0) + 14.237 us | } 3179 0) | __do_fault() { 3180 0) | filemap_fault() { 3181 0) | find_lock_page() { 3182 0) 0.698 us | find_get_page(); 3183 0) | __might_sleep() { 3184 0) 1.412 us | } 3185 0) 3.950 us | } 3186 0) 5.098 us | } 3187 0) 0.631 us | _spin_lock(); 3188 0) 0.571 us | page_add_file_rmap(); 3189 0) 0.526 us | native_set_pte_at(); 3190 0) 0.586 us | _spin_unlock(); 3191 0) | unlock_page() { 3192 0) 0.533 us | page_waitqueue(); 3193 0) 0.638 us | __wake_up_bit(); 3194 0) 2.793 us | } 3195 0) + 14.012 us | } 3196 3197You can also expand several functions at once:: 3198 3199 echo sys_open > set_graph_function 3200 echo sys_close >> set_graph_function 3201 3202Now if you want to go back to trace all functions you can clear 3203this special filter via:: 3204 3205 echo > set_graph_function 3206 3207 3208ftrace_enabled 3209-------------- 3210 3211Note, the proc sysctl ftrace_enable is a big on/off switch for the 3212function tracer. By default it is enabled (when function tracing is 3213enabled in the kernel). If it is disabled, all function tracing is 3214disabled. This includes not only the function tracers for ftrace, but 3215also for any other uses (perf, kprobes, stack tracing, profiling, etc). It 3216cannot be disabled if there is a callback with FTRACE_OPS_FL_PERMANENT set 3217registered. 3218 3219Please disable this with care. 3220 3221This can be disable (and enabled) with:: 3222 3223 sysctl kernel.ftrace_enabled=0 3224 sysctl kernel.ftrace_enabled=1 3225 3226 or 3227 3228 echo 0 > /proc/sys/kernel/ftrace_enabled 3229 echo 1 > /proc/sys/kernel/ftrace_enabled 3230 3231 3232Filter commands 3233--------------- 3234 3235A few commands are supported by the set_ftrace_filter interface. 3236Trace commands have the following format:: 3237 3238 <function>:<command>:<parameter> 3239 3240The following commands are supported: 3241 3242- mod: 3243 This command enables function filtering per module. The 3244 parameter defines the module. For example, if only the write* 3245 functions in the ext3 module are desired, run: 3246 3247 echo 'write*:mod:ext3' > set_ftrace_filter 3248 3249 This command interacts with the filter in the same way as 3250 filtering based on function names. Thus, adding more functions 3251 in a different module is accomplished by appending (>>) to the 3252 filter file. Remove specific module functions by prepending 3253 '!':: 3254 3255 echo '!writeback*:mod:ext3' >> set_ftrace_filter 3256 3257 Mod command supports module globbing. Disable tracing for all 3258 functions except a specific module:: 3259 3260 echo '!*:mod:!ext3' >> set_ftrace_filter 3261 3262 Disable tracing for all modules, but still trace kernel:: 3263 3264 echo '!*:mod:*' >> set_ftrace_filter 3265 3266 Enable filter only for kernel:: 3267 3268 echo '*write*:mod:!*' >> set_ftrace_filter 3269 3270 Enable filter for module globbing:: 3271 3272 echo '*write*:mod:*snd*' >> set_ftrace_filter 3273 3274- traceon/traceoff: 3275 These commands turn tracing on and off when the specified 3276 functions are hit. The parameter determines how many times the 3277 tracing system is turned on and off. If unspecified, there is 3278 no limit. For example, to disable tracing when a schedule bug 3279 is hit the first 5 times, run:: 3280 3281 echo '__schedule_bug:traceoff:5' > set_ftrace_filter 3282 3283 To always disable tracing when __schedule_bug is hit:: 3284 3285 echo '__schedule_bug:traceoff' > set_ftrace_filter 3286 3287 These commands are cumulative whether or not they are appended 3288 to set_ftrace_filter. To remove a command, prepend it by '!' 3289 and drop the parameter:: 3290 3291 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter 3292 3293 The above removes the traceoff command for __schedule_bug 3294 that have a counter. To remove commands without counters:: 3295 3296 echo '!__schedule_bug:traceoff' > set_ftrace_filter 3297 3298- snapshot: 3299 Will cause a snapshot to be triggered when the function is hit. 3300 :: 3301 3302 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter 3303 3304 To only snapshot once: 3305 :: 3306 3307 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter 3308 3309 To remove the above commands:: 3310 3311 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter 3312 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter 3313 3314- enable_event/disable_event: 3315 These commands can enable or disable a trace event. Note, because 3316 function tracing callbacks are very sensitive, when these commands 3317 are registered, the trace point is activated, but disabled in 3318 a "soft" mode. That is, the tracepoint will be called, but 3319 just will not be traced. The event tracepoint stays in this mode 3320 as long as there's a command that triggers it. 3321 :: 3322 3323 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \ 3324 set_ftrace_filter 3325 3326 The format is:: 3327 3328 <function>:enable_event:<system>:<event>[:count] 3329 <function>:disable_event:<system>:<event>[:count] 3330 3331 To remove the events commands:: 3332 3333 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \ 3334 set_ftrace_filter 3335 echo '!schedule:disable_event:sched:sched_switch' > \ 3336 set_ftrace_filter 3337 3338- dump: 3339 When the function is hit, it will dump the contents of the ftrace 3340 ring buffer to the console. This is useful if you need to debug 3341 something, and want to dump the trace when a certain function 3342 is hit. Perhaps it's a function that is called before a triple 3343 fault happens and does not allow you to get a regular dump. 3344 3345- cpudump: 3346 When the function is hit, it will dump the contents of the ftrace 3347 ring buffer for the current CPU to the console. Unlike the "dump" 3348 command, it only prints out the contents of the ring buffer for the 3349 CPU that executed the function that triggered the dump. 3350 3351- stacktrace: 3352 When the function is hit, a stack trace is recorded. 3353 3354trace_pipe 3355---------- 3356 3357The trace_pipe outputs the same content as the trace file, but 3358the effect on the tracing is different. Every read from 3359trace_pipe is consumed. This means that subsequent reads will be 3360different. The trace is live. 3361:: 3362 3363 # echo function > current_tracer 3364 # cat trace_pipe > /tmp/trace.out & 3365 [1] 4153 3366 # echo 1 > tracing_on 3367 # usleep 1 3368 # echo 0 > tracing_on 3369 # cat trace 3370 # tracer: function 3371 # 3372 # entries-in-buffer/entries-written: 0/0 #P:4 3373 # 3374 # _-----=> irqs-off 3375 # / _----=> need-resched 3376 # | / _---=> hardirq/softirq 3377 # || / _--=> preempt-depth 3378 # ||| / delay 3379 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3380 # | | | |||| | | 3381 3382 # 3383 # cat /tmp/trace.out 3384 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write 3385 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock 3386 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify 3387 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify 3388 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify 3389 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock 3390 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock 3391 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify 3392 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath 3393 3394 3395Note, reading the trace_pipe file will block until more input is 3396added. This is contrary to the trace file. If any process opened 3397the trace file for reading, it will actually disable tracing and 3398prevent new entries from being added. The trace_pipe file does 3399not have this limitation. 3400 3401trace entries 3402------------- 3403 3404Having too much or not enough data can be troublesome in 3405diagnosing an issue in the kernel. The file buffer_size_kb is 3406used to modify the size of the internal trace buffers. The 3407number listed is the number of entries that can be recorded per 3408CPU. To know the full size, multiply the number of possible CPUs 3409with the number of entries. 3410:: 3411 3412 # cat buffer_size_kb 3413 1408 (units kilobytes) 3414 3415Or simply read buffer_total_size_kb 3416:: 3417 3418 # cat buffer_total_size_kb 3419 5632 3420 3421To modify the buffer, simple echo in a number (in 1024 byte segments). 3422:: 3423 3424 # echo 10000 > buffer_size_kb 3425 # cat buffer_size_kb 3426 10000 (units kilobytes) 3427 3428It will try to allocate as much as possible. If you allocate too 3429much, it can cause Out-Of-Memory to trigger. 3430:: 3431 3432 # echo 1000000000000 > buffer_size_kb 3433 -bash: echo: write error: Cannot allocate memory 3434 # cat buffer_size_kb 3435 85 3436 3437The per_cpu buffers can be changed individually as well: 3438:: 3439 3440 # echo 10000 > per_cpu/cpu0/buffer_size_kb 3441 # echo 100 > per_cpu/cpu1/buffer_size_kb 3442 3443When the per_cpu buffers are not the same, the buffer_size_kb 3444at the top level will just show an X 3445:: 3446 3447 # cat buffer_size_kb 3448 X 3449 3450This is where the buffer_total_size_kb is useful: 3451:: 3452 3453 # cat buffer_total_size_kb 3454 12916 3455 3456Writing to the top level buffer_size_kb will reset all the buffers 3457to be the same again. 3458 3459Snapshot 3460-------- 3461CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature 3462available to all non latency tracers. (Latency tracers which 3463record max latency, such as "irqsoff" or "wakeup", can't use 3464this feature, since those are already using the snapshot 3465mechanism internally.) 3466 3467Snapshot preserves a current trace buffer at a particular point 3468in time without stopping tracing. Ftrace swaps the current 3469buffer with a spare buffer, and tracing continues in the new 3470current (=previous spare) buffer. 3471 3472The following tracefs files in "tracing" are related to this 3473feature: 3474 3475 snapshot: 3476 3477 This is used to take a snapshot and to read the output 3478 of the snapshot. Echo 1 into this file to allocate a 3479 spare buffer and to take a snapshot (swap), then read 3480 the snapshot from this file in the same format as 3481 "trace" (described above in the section "The File 3482 System"). Both reads snapshot and tracing are executable 3483 in parallel. When the spare buffer is allocated, echoing 3484 0 frees it, and echoing else (positive) values clear the 3485 snapshot contents. 3486 More details are shown in the table below. 3487 3488 +--------------+------------+------------+------------+ 3489 |status\\input | 0 | 1 | else | 3490 +==============+============+============+============+ 3491 |not allocated |(do nothing)| alloc+swap |(do nothing)| 3492 +--------------+------------+------------+------------+ 3493 |allocated | free | swap | clear | 3494 +--------------+------------+------------+------------+ 3495 3496Here is an example of using the snapshot feature. 3497:: 3498 3499 # echo 1 > events/sched/enable 3500 # echo 1 > snapshot 3501 # cat snapshot 3502 # tracer: nop 3503 # 3504 # entries-in-buffer/entries-written: 71/71 #P:8 3505 # 3506 # _-----=> irqs-off 3507 # / _----=> need-resched 3508 # | / _---=> hardirq/softirq 3509 # || / _--=> preempt-depth 3510 # ||| / delay 3511 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3512 # | | | |||| | | 3513 <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 3514 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 3515 [...] 3516 <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 3517 3518 # cat trace 3519 # tracer: nop 3520 # 3521 # entries-in-buffer/entries-written: 77/77 #P:8 3522 # 3523 # _-----=> irqs-off 3524 # / _----=> need-resched 3525 # | / _---=> hardirq/softirq 3526 # || / _--=> preempt-depth 3527 # ||| / delay 3528 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3529 # | | | |||| | | 3530 <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 3531 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 3532 [...] 3533 3534 3535If you try to use this snapshot feature when current tracer is 3536one of the latency tracers, you will get the following results. 3537:: 3538 3539 # echo wakeup > current_tracer 3540 # echo 1 > snapshot 3541 bash: echo: write error: Device or resource busy 3542 # cat snapshot 3543 cat: snapshot: Device or resource busy 3544 3545 3546Instances 3547--------- 3548In the tracefs tracing directory, there is a directory called "instances". 3549This directory can have new directories created inside of it using 3550mkdir, and removing directories with rmdir. The directory created 3551with mkdir in this directory will already contain files and other 3552directories after it is created. 3553:: 3554 3555 # mkdir instances/foo 3556 # ls instances/foo 3557 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu 3558 set_event snapshot trace trace_clock trace_marker trace_options 3559 trace_pipe tracing_on 3560 3561As you can see, the new directory looks similar to the tracing directory 3562itself. In fact, it is very similar, except that the buffer and 3563events are agnostic from the main directory, or from any other 3564instances that are created. 3565 3566The files in the new directory work just like the files with the 3567same name in the tracing directory except the buffer that is used 3568is a separate and new buffer. The files affect that buffer but do not 3569affect the main buffer with the exception of trace_options. Currently, 3570the trace_options affect all instances and the top level buffer 3571the same, but this may change in future releases. That is, options 3572may become specific to the instance they reside in. 3573 3574Notice that none of the function tracer files are there, nor is 3575current_tracer and available_tracers. This is because the buffers 3576can currently only have events enabled for them. 3577:: 3578 3579 # mkdir instances/foo 3580 # mkdir instances/bar 3581 # mkdir instances/zoot 3582 # echo 100000 > buffer_size_kb 3583 # echo 1000 > instances/foo/buffer_size_kb 3584 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb 3585 # echo function > current_trace 3586 # echo 1 > instances/foo/events/sched/sched_wakeup/enable 3587 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable 3588 # echo 1 > instances/foo/events/sched/sched_switch/enable 3589 # echo 1 > instances/bar/events/irq/enable 3590 # echo 1 > instances/zoot/events/syscalls/enable 3591 # cat trace_pipe 3592 CPU:2 [LOST 11745 EVENTS] 3593 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist 3594 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave 3595 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist 3596 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist 3597 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock 3598 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype 3599 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist 3600 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist 3601 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics 3602 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics 3603 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process 3604 [...] 3605 3606 # cat instances/foo/trace_pipe 3607 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000 3608 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000 3609 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003 3610 <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 3611 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 3612 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000 3613 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000 3614 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 3615 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001 3616 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 3617 [...] 3618 3619 # cat instances/bar/trace_pipe 3620 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX] 3621 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX] 3622 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER] 3623 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU] 3624 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER] 3625 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER] 3626 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU] 3627 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU] 3628 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4 3629 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled 3630 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0 3631 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled 3632 [...] 3633 3634 # cat instances/zoot/trace 3635 # tracer: nop 3636 # 3637 # entries-in-buffer/entries-written: 18996/18996 #P:4 3638 # 3639 # _-----=> irqs-off 3640 # / _----=> need-resched 3641 # | / _---=> hardirq/softirq 3642 # || / _--=> preempt-depth 3643 # ||| / delay 3644 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3645 # | | | |||| | | 3646 bash-1998 [000] d... 140.733501: sys_write -> 0x2 3647 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1) 3648 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1 3649 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0) 3650 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1 3651 bash-1998 [000] d... 140.733510: sys_close(fd: a) 3652 bash-1998 [000] d... 140.733510: sys_close -> 0x0 3653 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8) 3654 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0 3655 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8) 3656 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0 3657 3658You can see that the trace of the top most trace buffer shows only 3659the function tracing. The foo instance displays wakeups and task 3660switches. 3661 3662To remove the instances, simply delete their directories: 3663:: 3664 3665 # rmdir instances/foo 3666 # rmdir instances/bar 3667 # rmdir instances/zoot 3668 3669Note, if a process has a trace file open in one of the instance 3670directories, the rmdir will fail with EBUSY. 3671 3672 3673Stack trace 3674----------- 3675Since the kernel has a fixed sized stack, it is important not to 3676waste it in functions. A kernel developer must be conscious of 3677what they allocate on the stack. If they add too much, the system 3678can be in danger of a stack overflow, and corruption will occur, 3679usually leading to a system panic. 3680 3681There are some tools that check this, usually with interrupts 3682periodically checking usage. But if you can perform a check 3683at every function call that will become very useful. As ftrace provides 3684a function tracer, it makes it convenient to check the stack size 3685at every function call. This is enabled via the stack tracer. 3686 3687CONFIG_STACK_TRACER enables the ftrace stack tracing functionality. 3688To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled. 3689:: 3690 3691 # echo 1 > /proc/sys/kernel/stack_tracer_enabled 3692 3693You can also enable it from the kernel command line to trace 3694the stack size of the kernel during boot up, by adding "stacktrace" 3695to the kernel command line parameter. 3696 3697After running it for a few minutes, the output looks like: 3698:: 3699 3700 # cat stack_max_size 3701 2928 3702 3703 # cat stack_trace 3704 Depth Size Location (18 entries) 3705 ----- ---- -------- 3706 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac 3707 1) 2704 160 find_busiest_group+0x31/0x1f1 3708 2) 2544 256 load_balance+0xd9/0x662 3709 3) 2288 80 idle_balance+0xbb/0x130 3710 4) 2208 128 __schedule+0x26e/0x5b9 3711 5) 2080 16 schedule+0x64/0x66 3712 6) 2064 128 schedule_timeout+0x34/0xe0 3713 7) 1936 112 wait_for_common+0x97/0xf1 3714 8) 1824 16 wait_for_completion+0x1d/0x1f 3715 9) 1808 128 flush_work+0xfe/0x119 3716 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20 3717 11) 1664 48 input_available_p+0x1d/0x5c 3718 12) 1616 48 n_tty_poll+0x6d/0x134 3719 13) 1568 64 tty_poll+0x64/0x7f 3720 14) 1504 880 do_select+0x31e/0x511 3721 15) 624 400 core_sys_select+0x177/0x216 3722 16) 224 96 sys_select+0x91/0xb9 3723 17) 128 128 system_call_fastpath+0x16/0x1b 3724 3725Note, if -mfentry is being used by gcc, functions get traced before 3726they set up the stack frame. This means that leaf level functions 3727are not tested by the stack tracer when -mfentry is used. 3728 3729Currently, -mfentry is used by gcc 4.6.0 and above on x86 only. 3730 3731More 3732---- 3733More details can be found in the source code, in the `kernel/trace/*.c` files. 3734