1======================== 2ftrace - Function Tracer 3======================== 4 5Copyright 2008 Red Hat Inc. 6 7:Author: Steven Rostedt <srostedt@redhat.com> 8:License: The GNU Free Documentation License, Version 1.2 9 (dual licensed under the GPL v2) 10:Original Reviewers: Elias Oltmanns, Randy Dunlap, Andrew Morton, 11 John Kacur, and David Teigland. 12 13- Written for: 2.6.28-rc2 14- Updated for: 3.10 15- Updated for: 4.13 - Copyright 2017 VMware Inc. Steven Rostedt 16- Converted to rst format - Changbin Du <changbin.du@intel.com> 17 18Introduction 19------------ 20 21Ftrace is an internal tracer designed to help out developers and 22designers of systems to find what is going on inside the kernel. 23It can be used for debugging or analyzing latencies and 24performance issues that take place outside of user-space. 25 26Although ftrace is typically considered the function tracer, it 27is really a framework of several assorted tracing utilities. 28There's latency tracing to examine what occurs between interrupts 29disabled and enabled, as well as for preemption and from a time 30a task is woken to the task is actually scheduled in. 31 32One of the most common uses of ftrace is the event tracing. 33Throughout the kernel are hundreds of static event points that 34can be enabled via the tracefs file system to see what is 35going on in certain parts of the kernel. 36 37See events.rst for more information. 38 39 40Implementation Details 41---------------------- 42 43See Documentation/trace/ftrace-design.rst for details for arch porters and such. 44 45 46The File System 47--------------- 48 49Ftrace uses the tracefs file system to hold the control files as 50well as the files to display output. 51 52When tracefs is configured into the kernel (which selecting any ftrace 53option will do) the directory /sys/kernel/tracing will be created. To mount 54this directory, you can add to your /etc/fstab file:: 55 56 tracefs /sys/kernel/tracing tracefs defaults 0 0 57 58Or you can mount it at run time with:: 59 60 mount -t tracefs nodev /sys/kernel/tracing 61 62For quicker access to that directory you may want to make a soft link to 63it:: 64 65 ln -s /sys/kernel/tracing /tracing 66 67.. attention:: 68 69 Before 4.1, all ftrace tracing control files were within the debugfs 70 file system, which is typically located at /sys/kernel/debug/tracing. 71 For backward compatibility, when mounting the debugfs file system, 72 the tracefs file system will be automatically mounted at: 73 74 /sys/kernel/debug/tracing 75 76 All files located in the tracefs file system will be located in that 77 debugfs file system directory as well. 78 79.. attention:: 80 81 Any selected ftrace option will also create the tracefs file system. 82 The rest of the document will assume that you are in the ftrace directory 83 (cd /sys/kernel/tracing) and will only concentrate on the files within that 84 directory and not distract from the content with the extended 85 "/sys/kernel/tracing" path name. 86 87That's it! (assuming that you have ftrace configured into your kernel) 88 89After mounting tracefs you will have access to the control and output files 90of ftrace. Here is a list of some of the key files: 91 92 93 Note: all time values are in microseconds. 94 95 current_tracer: 96 97 This is used to set or display the current tracer 98 that is configured. Changing the current tracer clears 99 the ring buffer content as well as the "snapshot" buffer. 100 101 available_tracers: 102 103 This holds the different types of tracers that 104 have been compiled into the kernel. The 105 tracers listed here can be configured by 106 echoing their name into current_tracer. 107 108 tracing_on: 109 110 This sets or displays whether writing to the trace 111 ring buffer is enabled. Echo 0 into this file to disable 112 the tracer or 1 to enable it. Note, this only disables 113 writing to the ring buffer, the tracing overhead may 114 still be occurring. 115 116 The kernel function tracing_off() can be used within the 117 kernel to disable writing to the ring buffer, which will 118 set this file to "0". User space can re-enable tracing by 119 echoing "1" into the file. 120 121 Note, the function and event trigger "traceoff" will also 122 set this file to zero and stop tracing. Which can also 123 be re-enabled by user space using this file. 124 125 trace: 126 127 This file holds the output of the trace in a human 128 readable format (described below). Opening this file for 129 writing with the O_TRUNC flag clears the ring buffer content. 130 Note, this file is not a consumer. If tracing is off 131 (no tracer running, or tracing_on is zero), it will produce 132 the same output each time it is read. When tracing is on, 133 it may produce inconsistent results as it tries to read 134 the entire buffer without consuming it. 135 136 trace_pipe: 137 138 The output is the same as the "trace" file but this 139 file is meant to be streamed with live tracing. 140 Reads from this file will block until new data is 141 retrieved. Unlike the "trace" file, this file is a 142 consumer. This means reading from this file causes 143 sequential reads to display more current data. Once 144 data is read from this file, it is consumed, and 145 will not be read again with a sequential read. The 146 "trace" file is static, and if the tracer is not 147 adding more data, it will display the same 148 information every time it is read. 149 150 trace_options: 151 152 This file lets the user control the amount of data 153 that is displayed in one of the above output 154 files. Options also exist to modify how a tracer 155 or events work (stack traces, timestamps, etc). 156 157 options: 158 159 This is a directory that has a file for every available 160 trace option (also in trace_options). Options may also be set 161 or cleared by writing a "1" or "0" respectively into the 162 corresponding file with the option name. 163 164 tracing_max_latency: 165 166 Some of the tracers record the max latency. 167 For example, the maximum time that interrupts are disabled. 168 The maximum time is saved in this file. The max trace will also be 169 stored, and displayed by "trace". A new max trace will only be 170 recorded if the latency is greater than the value in this file 171 (in microseconds). 172 173 By echoing in a time into this file, no latency will be recorded 174 unless it is greater than the time in this file. 175 176 tracing_thresh: 177 178 Some latency tracers will record a trace whenever the 179 latency is greater than the number in this file. 180 Only active when the file contains a number greater than 0. 181 (in microseconds) 182 183 buffer_percent: 184 185 This is the watermark for how much the ring buffer needs to be filled 186 before a waiter is woken up. That is, if an application calls a 187 blocking read syscall on one of the per_cpu trace_pipe_raw files, it 188 will block until the given amount of data specified by buffer_percent 189 is in the ring buffer before it wakes the reader up. This also 190 controls how the splice system calls are blocked on this file:: 191 192 0 - means to wake up as soon as there is any data in the ring buffer. 193 50 - means to wake up when roughly half of the ring buffer sub-buffers 194 are full. 195 100 - means to block until the ring buffer is totally full and is 196 about to start overwriting the older data. 197 198 buffer_size_kb: 199 200 This sets or displays the number of kilobytes each CPU 201 buffer holds. By default, the trace buffers are the same size 202 for each CPU. The displayed number is the size of the 203 CPU buffer and not total size of all buffers. The 204 trace buffers are allocated in pages (blocks of memory 205 that the kernel uses for allocation, usually 4 KB in size). 206 A few extra pages may be allocated to accommodate buffer management 207 meta-data. If the last page allocated has room for more bytes 208 than requested, the rest of the page will be used, 209 making the actual allocation bigger than requested or shown. 210 ( Note, the size may not be a multiple of the page size 211 due to buffer management meta-data. ) 212 213 Buffer sizes for individual CPUs may vary 214 (see "per_cpu/cpu0/buffer_size_kb" below), and if they do 215 this file will show "X". 216 217 buffer_total_size_kb: 218 219 This displays the total combined size of all the trace buffers. 220 221 buffer_subbuf_size_kb: 222 223 This sets or displays the sub buffer size. The ring buffer is broken up 224 into several same size "sub buffers". An event can not be bigger than 225 the size of the sub buffer. Normally, the sub buffer is the size of the 226 architecture's page (4K on x86). The sub buffer also contains meta data 227 at the start which also limits the size of an event. That means when 228 the sub buffer is a page size, no event can be larger than the page 229 size minus the sub buffer meta data. 230 231 Note, the buffer_subbuf_size_kb is a way for the user to specify the 232 minimum size of the subbuffer. The kernel may make it bigger due to the 233 implementation details, or simply fail the operation if the kernel can 234 not handle the request. 235 236 Changing the sub buffer size allows for events to be larger than the 237 page size. 238 239 Note: When changing the sub-buffer size, tracing is stopped and any 240 data in the ring buffer and the snapshot buffer will be discarded. 241 242 free_buffer: 243 244 If a process is performing tracing, and the ring buffer should be 245 shrunk "freed" when the process is finished, even if it were to be 246 killed by a signal, this file can be used for that purpose. On close 247 of this file, the ring buffer will be resized to its minimum size. 248 Having a process that is tracing also open this file, when the process 249 exits its file descriptor for this file will be closed, and in doing so, 250 the ring buffer will be "freed". 251 252 It may also stop tracing if disable_on_free option is set. 253 254 tracing_cpumask: 255 256 This is a mask that lets the user only trace on specified CPUs. 257 The format is a hex string representing the CPUs. 258 259 set_ftrace_filter: 260 261 When dynamic ftrace is configured in (see the 262 section below "dynamic ftrace"), the code is dynamically 263 modified (code text rewrite) to disable calling of the 264 function profiler (mcount). This lets tracing be configured 265 in with practically no overhead in performance. This also 266 has a side effect of enabling or disabling specific functions 267 to be traced. Echoing names of functions into this file 268 will limit the trace to only those functions. 269 This influences the tracers "function" and "function_graph" 270 and thus also function profiling (see "function_profile_enabled"). 271 272 The functions listed in "available_filter_functions" are what 273 can be written into this file. 274 275 This interface also allows for commands to be used. See the 276 "Filter commands" section for more details. 277 278 As a speed up, since processing strings can be quite expensive 279 and requires a check of all functions registered to tracing, instead 280 an index can be written into this file. A number (starting with "1") 281 written will instead select the same corresponding at the line position 282 of the "available_filter_functions" file. 283 284 set_ftrace_notrace: 285 286 This has an effect opposite to that of 287 set_ftrace_filter. Any function that is added here will not 288 be traced. If a function exists in both set_ftrace_filter 289 and set_ftrace_notrace, the function will _not_ be traced. 290 291 set_ftrace_pid: 292 293 Have the function tracer only trace the threads whose PID are 294 listed in this file. 295 296 If the "function-fork" option is set, then when a task whose 297 PID is listed in this file forks, the child's PID will 298 automatically be added to this file, and the child will be 299 traced by the function tracer as well. This option will also 300 cause PIDs of tasks that exit to be removed from the file. 301 302 set_ftrace_notrace_pid: 303 304 Have the function tracer ignore threads whose PID are listed in 305 this file. 306 307 If the "function-fork" option is set, then when a task whose 308 PID is listed in this file forks, the child's PID will 309 automatically be added to this file, and the child will not be 310 traced by the function tracer as well. This option will also 311 cause PIDs of tasks that exit to be removed from the file. 312 313 If a PID is in both this file and "set_ftrace_pid", then this 314 file takes precedence, and the thread will not be traced. 315 316 set_event_pid: 317 318 Have the events only trace a task with a PID listed in this file. 319 Note, sched_switch and sched_wake_up will also trace events 320 listed in this file. 321 322 To have the PIDs of children of tasks with their PID in this file 323 added on fork, enable the "event-fork" option. That option will also 324 cause the PIDs of tasks to be removed from this file when the task 325 exits. 326 327 set_event_notrace_pid: 328 329 Have the events not trace a task with a PID listed in this file. 330 Note, sched_switch and sched_wakeup will trace threads not listed 331 in this file, even if a thread's PID is in the file if the 332 sched_switch or sched_wakeup events also trace a thread that should 333 be traced. 334 335 To have the PIDs of children of tasks with their PID in this file 336 added on fork, enable the "event-fork" option. That option will also 337 cause the PIDs of tasks to be removed from this file when the task 338 exits. 339 340 set_graph_function: 341 342 Functions listed in this file will cause the function graph 343 tracer to only trace these functions and the functions that 344 they call. (See the section "dynamic ftrace" for more details). 345 Note, set_ftrace_filter and set_ftrace_notrace still affects 346 what functions are being traced. 347 348 set_graph_notrace: 349 350 Similar to set_graph_function, but will disable function graph 351 tracing when the function is hit until it exits the function. 352 This makes it possible to ignore tracing functions that are called 353 by a specific function. 354 355 available_filter_functions: 356 357 This lists the functions that ftrace has processed and can trace. 358 These are the function names that you can pass to 359 "set_ftrace_filter", "set_ftrace_notrace", 360 "set_graph_function", or "set_graph_notrace". 361 (See the section "dynamic ftrace" below for more details.) 362 363 available_filter_functions_addrs: 364 365 Similar to available_filter_functions, but with address displayed 366 for each function. The displayed address is the patch-site address 367 and can differ from /proc/kallsyms address. 368 369 syscall_user_buf_size: 370 371 Some system call trace events will record the data from a user 372 space address that one of the parameters point to. The amount of 373 data per event is limited. This file holds the max number of bytes 374 that will be recorded into the ring buffer to hold this data. 375 The max value is currently 165. 376 377 dyn_ftrace_total_info: 378 379 This file is for debugging purposes. The number of functions that 380 have been converted to nops and are available to be traced. 381 382 enabled_functions: 383 384 This file is more for debugging ftrace, but can also be useful 385 in seeing if any function has a callback attached to it. 386 Not only does the trace infrastructure use ftrace function 387 trace utility, but other subsystems might too. This file 388 displays all functions that have a callback attached to them 389 as well as the number of callbacks that have been attached. 390 Note, a callback may also call multiple functions which will 391 not be listed in this count. 392 393 If the callback registered to be traced by a function with 394 the "save regs" attribute (thus even more overhead), an 'R' 395 will be displayed on the same line as the function that 396 is returning registers. 397 398 If the callback registered to be traced by a function with 399 the "ip modify" attribute (thus the regs->ip can be changed), 400 an 'I' will be displayed on the same line as the function that 401 can be overridden. 402 403 If a non-ftrace trampoline is attached (BPF) a 'D' will be displayed. 404 Note, normal ftrace trampolines can also be attached, but only one 405 "direct" trampoline can be attached to a given function at a time. 406 407 Some architectures can not call direct trampolines, but instead have 408 the ftrace ops function located above the function entry point. In 409 such cases an 'O' will be displayed. 410 411 If a function had either the "ip modify" or a "direct" call attached to 412 it in the past, a 'M' will be shown. This flag is never cleared. It is 413 used to know if a function was ever modified by the ftrace infrastructure, 414 and can be used for debugging. 415 416 If the architecture supports it, it will also show what callback 417 is being directly called by the function. If the count is greater 418 than 1 it most likely will be ftrace_ops_list_func(). 419 420 If the callback of a function jumps to a trampoline that is 421 specific to the callback and which is not the standard trampoline, 422 its address will be printed as well as the function that the 423 trampoline calls. 424 425 touched_functions: 426 427 This file contains all the functions that ever had a function callback 428 to it via the ftrace infrastructure. It has the same format as 429 enabled_functions but shows all functions that have ever been 430 traced. 431 432 To see any function that has every been modified by "ip modify" or a 433 direct trampoline, one can perform the following command: 434 435 grep ' M ' /sys/kernel/tracing/touched_functions 436 437 function_profile_enabled: 438 439 When set it will enable all functions with either the function 440 tracer, or if configured, the function graph tracer. It will 441 keep a histogram of the number of functions that were called 442 and if the function graph tracer was configured, it will also keep 443 track of the time spent in those functions. The histogram 444 content can be displayed in the files: 445 446 trace_stat/function<cpu> ( function0, function1, etc). 447 448 trace_stat: 449 450 A directory that holds different tracing stats. 451 452 kprobe_events: 453 454 Enable dynamic trace points. See kprobetrace.rst. 455 456 kprobe_profile: 457 458 Dynamic trace points stats. See kprobetrace.rst. 459 460 max_graph_depth: 461 462 Used with the function graph tracer. This is the max depth 463 it will trace into a function. Setting this to a value of 464 one will show only the first kernel function that is called 465 from user space. 466 467 printk_formats: 468 469 This is for tools that read the raw format files. If an event in 470 the ring buffer references a string, only a pointer to the string 471 is recorded into the buffer and not the string itself. This prevents 472 tools from knowing what that string was. This file displays the string 473 and address for the string allowing tools to map the pointers to what 474 the strings were. 475 476 saved_cmdlines: 477 478 Only the pid of the task is recorded in a trace event unless 479 the event specifically saves the task comm as well. Ftrace 480 makes a cache of pid mappings to comms to try to display 481 comms for events. If a pid for a comm is not listed, then 482 "<...>" is displayed in the output. 483 484 If the option "record-cmd" is set to "0", then comms of tasks 485 will not be saved during recording. By default, it is enabled. 486 487 saved_cmdlines_size: 488 489 By default, 128 comms are saved (see "saved_cmdlines" above). To 490 increase or decrease the amount of comms that are cached, echo 491 the number of comms to cache into this file. 492 493 saved_tgids: 494 495 If the option "record-tgid" is set, on each scheduling context switch 496 the Task Group ID of a task is saved in a table mapping the PID of 497 the thread to its TGID. By default, the "record-tgid" option is 498 disabled. 499 500 snapshot: 501 502 This displays the "snapshot" buffer and also lets the user 503 take a snapshot of the current running trace. 504 See the "Snapshot" section below for more details. 505 506 stack_max_size: 507 508 When the stack tracer is activated, this will display the 509 maximum stack size it has encountered. 510 See the "Stack Trace" section below. 511 512 stack_trace: 513 514 This displays the stack back trace of the largest stack 515 that was encountered when the stack tracer is activated. 516 See the "Stack Trace" section below. 517 518 stack_trace_filter: 519 520 This is similar to "set_ftrace_filter" but it limits what 521 functions the stack tracer will check. 522 523 trace_clock: 524 525 Whenever an event is recorded into the ring buffer, a 526 "timestamp" is added. This stamp comes from a specified 527 clock. By default, ftrace uses the "local" clock. This 528 clock is very fast and strictly per CPU, but on some 529 systems it may not be monotonic with respect to other 530 CPUs. In other words, the local clocks may not be in sync 531 with local clocks on other CPUs. 532 533 Usual clocks for tracing:: 534 535 # cat trace_clock 536 [local] global counter x86-tsc 537 538 The clock with the square brackets around it is the one in effect. 539 540 local: 541 Default clock, but may not be in sync across CPUs 542 543 global: 544 This clock is in sync with all CPUs but may 545 be a bit slower than the local clock. 546 547 counter: 548 This is not a clock at all, but literally an atomic 549 counter. It counts up one by one, but is in sync 550 with all CPUs. This is useful when you need to 551 know exactly the order events occurred with respect to 552 each other on different CPUs. 553 554 uptime: 555 This uses the jiffies counter and the time stamp 556 is relative to the time since boot up. 557 558 perf: 559 This makes ftrace use the same clock that perf uses. 560 Eventually perf will be able to read ftrace buffers 561 and this will help out in interleaving the data. 562 563 x86-tsc: 564 Architectures may define their own clocks. For 565 example, x86 uses its own TSC cycle clock here. 566 567 ppc-tb: 568 This uses the powerpc timebase register value. 569 This is in sync across CPUs and can also be used 570 to correlate events across hypervisor/guest if 571 tb_offset is known. 572 573 mono: 574 This uses the fast monotonic clock (CLOCK_MONOTONIC) 575 which is monotonic and is subject to NTP rate adjustments. 576 577 mono_raw: 578 This is the raw monotonic clock (CLOCK_MONOTONIC_RAW) 579 which is monotonic but is not subject to any rate adjustments 580 and ticks at the same rate as the hardware clocksource. 581 582 boot: 583 This is the boot clock (CLOCK_BOOTTIME) and is based on the 584 fast monotonic clock, but also accounts for time spent in 585 suspend. Since the clock access is designed for use in 586 tracing in the suspend path, some side effects are possible 587 if clock is accessed after the suspend time is accounted before 588 the fast mono clock is updated. In this case, the clock update 589 appears to happen slightly sooner than it normally would have. 590 Also on 32-bit systems, it's possible that the 64-bit boot offset 591 sees a partial update. These effects are rare and post 592 processing should be able to handle them. See comments in the 593 ktime_get_boot_fast_ns() function for more information. 594 595 tai: 596 This is the tai clock (CLOCK_TAI) and is derived from the wall- 597 clock time. However, this clock does not experience 598 discontinuities and backwards jumps caused by NTP inserting leap 599 seconds. Since the clock access is designed for use in tracing, 600 side effects are possible. The clock access may yield wrong 601 readouts in case the internal TAI offset is updated e.g., caused 602 by setting the system time or using adjtimex() with an offset. 603 These effects are rare and post processing should be able to 604 handle them. See comments in the ktime_get_tai_fast_ns() 605 function for more information. 606 607 To set a clock, simply echo the clock name into this file:: 608 609 # echo global > trace_clock 610 611 Setting a clock clears the ring buffer content as well as the 612 "snapshot" buffer. 613 614 trace_marker: 615 616 This is a very useful file for synchronizing user space 617 with events happening in the kernel. Writing strings into 618 this file will be written into the ftrace buffer. 619 620 It is useful in applications to open this file at the start 621 of the application and just reference the file descriptor 622 for the file:: 623 624 void trace_write(const char *fmt, ...) 625 { 626 va_list ap; 627 char buf[256]; 628 int n; 629 630 if (trace_fd < 0) 631 return; 632 633 va_start(ap, fmt); 634 n = vsnprintf(buf, 256, fmt, ap); 635 va_end(ap); 636 637 write(trace_fd, buf, n); 638 } 639 640 start:: 641 642 trace_fd = open("trace_marker", O_WRONLY); 643 644 Note: Writing into the trace_marker file can also initiate triggers 645 that are written into /sys/kernel/tracing/events/ftrace/print/trigger 646 See "Event triggers" in Documentation/trace/events.rst and an 647 example in Documentation/trace/histogram.rst (Section 3.) 648 649 trace_marker_raw: 650 651 This is similar to trace_marker above, but is meant for binary data 652 to be written to it, where a tool can be used to parse the data 653 from trace_pipe_raw. 654 655 uprobe_events: 656 657 Add dynamic tracepoints in programs. 658 See uprobetracer.rst 659 660 uprobe_profile: 661 662 Uprobe statistics. See uprobetrace.txt 663 664 instances: 665 666 This is a way to make multiple trace buffers where different 667 events can be recorded in different buffers. 668 See "Instances" section below. 669 670 events: 671 672 This is the trace event directory. It holds event tracepoints 673 (also known as static tracepoints) that have been compiled 674 into the kernel. It shows what event tracepoints exist 675 and how they are grouped by system. There are "enable" 676 files at various levels that can enable the tracepoints 677 when a "1" is written to them. 678 679 See events.rst for more information. 680 681 set_event: 682 683 By echoing in the event into this file, will enable that event. 684 685 See events.rst for more information. 686 687 show_event_filters: 688 689 A list of events that have filters. This shows the 690 system/event pair along with the filter that is attached to 691 the event. 692 693 See events.rst for more information. 694 695 show_event_triggers: 696 697 A list of events that have triggers. This shows the 698 system/event pair along with the trigger that is attached to 699 the event. 700 701 See events.rst for more information. 702 703 available_events: 704 705 A list of events that can be enabled in tracing. 706 707 See events.rst for more information. 708 709 timestamp_mode: 710 711 Certain tracers may change the timestamp mode used when 712 logging trace events into the event buffer. Events with 713 different modes can coexist within a buffer but the mode in 714 effect when an event is logged determines which timestamp mode 715 is used for that event. The default timestamp mode is 716 'delta'. 717 718 Usual timestamp modes for tracing: 719 720 # cat timestamp_mode 721 [delta] absolute 722 723 The timestamp mode with the square brackets around it is the 724 one in effect. 725 726 delta: Default timestamp mode - timestamp is a delta against 727 a per-buffer timestamp. 728 729 absolute: The timestamp is a full timestamp, not a delta 730 against some other value. As such it takes up more 731 space and is less efficient. 732 733 hwlat_detector: 734 735 Directory for the Hardware Latency Detector. 736 See "Hardware Latency Detector" section below. 737 738 per_cpu: 739 740 This is a directory that contains the trace per_cpu information. 741 742 per_cpu/cpu0/buffer_size_kb: 743 744 The ftrace buffer is defined per_cpu. That is, there's a separate 745 buffer for each CPU to allow writes to be done atomically, 746 and free from cache bouncing. These buffers may have different 747 size buffers. This file is similar to the buffer_size_kb 748 file, but it only displays or sets the buffer size for the 749 specific CPU. (here cpu0). 750 751 per_cpu/cpu0/trace: 752 753 This is similar to the "trace" file, but it will only display 754 the data specific for the CPU. If written to, it only clears 755 the specific CPU buffer. 756 757 per_cpu/cpu0/trace_pipe 758 759 This is similar to the "trace_pipe" file, and is a consuming 760 read, but it will only display (and consume) the data specific 761 for the CPU. 762 763 per_cpu/cpu0/trace_pipe_raw 764 765 For tools that can parse the ftrace ring buffer binary format, 766 the trace_pipe_raw file can be used to extract the data 767 from the ring buffer directly. With the use of the splice() 768 system call, the buffer data can be quickly transferred to 769 a file or to the network where a server is collecting the 770 data. 771 772 Like trace_pipe, this is a consuming reader, where multiple 773 reads will always produce different data. 774 775 per_cpu/cpu0/snapshot: 776 777 This is similar to the main "snapshot" file, but will only 778 snapshot the current CPU (if supported). It only displays 779 the content of the snapshot for a given CPU, and if 780 written to, only clears this CPU buffer. 781 782 per_cpu/cpu0/snapshot_raw: 783 784 Similar to the trace_pipe_raw, but will read the binary format 785 from the snapshot buffer for the given CPU. 786 787 per_cpu/cpu0/stats: 788 789 This displays certain stats about the ring buffer: 790 791 entries: 792 The number of events that are still in the buffer. 793 794 overrun: 795 The number of lost events due to overwriting when 796 the buffer was full. 797 798 commit overrun: 799 Should always be zero. 800 This gets set if so many events happened within a nested 801 event (ring buffer is re-entrant), that it fills the 802 buffer and starts dropping events. 803 804 bytes: 805 Bytes actually read (not overwritten). 806 807 oldest event ts: 808 The oldest timestamp in the buffer 809 810 now ts: 811 The current timestamp 812 813 dropped events: 814 Events lost due to overwrite option being off. 815 816 read events: 817 The number of events read. 818 819The Tracers 820----------- 821 822Here is the list of current tracers that may be configured. 823 824 "function" 825 826 Function call tracer to trace all kernel functions. 827 828 "function_graph" 829 830 Similar to the function tracer except that the 831 function tracer probes the functions on their entry 832 whereas the function graph tracer traces on both entry 833 and exit of the functions. It then provides the ability 834 to draw a graph of function calls similar to C code 835 source. 836 837 Note that the function graph calculates the timings of when the 838 function starts and returns internally and for each instance. If 839 there are two instances that run function graph tracer and traces 840 the same functions, the length of the timings may be slightly off as 841 each read the timestamp separately and not at the same time. 842 843 "blk" 844 845 The block tracer. The tracer used by the blktrace user 846 application. 847 848 "hwlat" 849 850 The Hardware Latency tracer is used to detect if the hardware 851 produces any latency. See "Hardware Latency Detector" section 852 below. 853 854 "irqsoff" 855 856 Traces the areas that disable interrupts and saves 857 the trace with the longest max latency. 858 See tracing_max_latency. When a new max is recorded, 859 it replaces the old trace. It is best to view this 860 trace with the latency-format option enabled, which 861 happens automatically when the tracer is selected. 862 863 "preemptoff" 864 865 Similar to irqsoff but traces and records the amount of 866 time for which preemption is disabled. 867 868 "preemptirqsoff" 869 870 Similar to irqsoff and preemptoff, but traces and 871 records the largest time for which irqs and/or preemption 872 is disabled. 873 874 "wakeup" 875 876 Traces and records the max latency that it takes for 877 the highest priority task to get scheduled after 878 it has been woken up. 879 Traces all tasks as an average developer would expect. 880 881 "wakeup_rt" 882 883 Traces and records the max latency that it takes for just 884 RT tasks (as the current "wakeup" does). This is useful 885 for those interested in wake up timings of RT tasks. 886 887 "wakeup_dl" 888 889 Traces and records the max latency that it takes for 890 a SCHED_DEADLINE task to be woken (as the "wakeup" and 891 "wakeup_rt" does). 892 893 "mmiotrace" 894 895 A special tracer that is used to trace binary modules. 896 It will trace all the calls that a module makes to the 897 hardware. Everything it writes and reads from the I/O 898 as well. 899 900 "branch" 901 902 This tracer can be configured when tracing likely/unlikely 903 calls within the kernel. It will trace when a likely and 904 unlikely branch is hit and if it was correct in its prediction 905 of being correct. 906 907 "nop" 908 909 This is the "trace nothing" tracer. To remove all 910 tracers from tracing simply echo "nop" into 911 current_tracer. 912 913Error conditions 914---------------- 915 916 For most ftrace commands, failure modes are obvious and communicated 917 using standard return codes. 918 919 For other more involved commands, extended error information may be 920 available via the tracing/error_log file. For the commands that 921 support it, reading the tracing/error_log file after an error will 922 display more detailed information about what went wrong, if 923 information is available. The tracing/error_log file is a circular 924 error log displaying a small number (currently, 8) of ftrace errors 925 for the last (8) failed commands. 926 927 The extended error information and usage takes the form shown in 928 this example:: 929 930 # echo xxx > /sys/kernel/tracing/events/sched/sched_wakeup/trigger 931 echo: write error: Invalid argument 932 933 # cat /sys/kernel/tracing/error_log 934 [ 5348.887237] location: error: Couldn't yyy: zzz 935 Command: xxx 936 ^ 937 [ 7517.023364] location: error: Bad rrr: sss 938 Command: ppp qqq 939 ^ 940 941 To clear the error log, echo the empty string into it:: 942 943 # echo > /sys/kernel/tracing/error_log 944 945Examples of using the tracer 946---------------------------- 947 948Here are typical examples of using the tracers when controlling 949them only with the tracefs interface (without using any 950user-land utilities). 951 952Output format: 953-------------- 954 955Here is an example of the output format of the file "trace":: 956 957 # tracer: function 958 # 959 # entries-in-buffer/entries-written: 140080/250280 #P:4 960 # 961 # _-----=> irqs-off 962 # / _----=> need-resched 963 # | / _---=> hardirq/softirq 964 # || / _--=> preempt-depth 965 # ||| / delay 966 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 967 # | | | |||| | | 968 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath 969 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close 970 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd 971 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify 972 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock 973 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd 974 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock 975 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd 976 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close 977 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath 978 .... 979 980A header is printed with the tracer name that is represented by 981the trace. In this case the tracer is "function". Then it shows the 982number of events in the buffer as well as the total number of entries 983that were written. The difference is the number of entries that were 984lost due to the buffer filling up (250280 - 140080 = 110200 events 985lost). 986 987The header explains the content of the events. Task name "bash", the task 988PID "1977", the CPU that it was running on "000", the latency format 989(explained below), the timestamp in <secs>.<usecs> format, the 990function name that was traced "sys_close" and the parent function that 991called this function "system_call_fastpath". The timestamp is the time 992at which the function was entered. 993 994Latency trace format 995-------------------- 996 997When the latency-format option is enabled or when one of the latency 998tracers is set, the trace file gives somewhat more information to see 999why a latency happened. Here is a typical trace:: 1000 1001 # tracer: irqsoff 1002 # 1003 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1004 # -------------------------------------------------------------------- 1005 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1006 # ----------------- 1007 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0) 1008 # ----------------- 1009 # => started at: __lock_task_sighand 1010 # => ended at: _raw_spin_unlock_irqrestore 1011 # 1012 # 1013 # _------=> CPU# 1014 # / _-----=> irqs-off 1015 # | / _----=> need-resched 1016 # || / _---=> hardirq/softirq 1017 # ||| / _--=> preempt-depth 1018 # |||| / delay 1019 # cmd pid ||||| time | caller 1020 # \ / ||||| \ | / 1021 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand 1022 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore 1023 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore 1024 ps-6143 2d..1 306us : <stack trace> 1025 => trace_hardirqs_on_caller 1026 => trace_hardirqs_on 1027 => _raw_spin_unlock_irqrestore 1028 => do_task_stat 1029 => proc_tgid_stat 1030 => proc_single_show 1031 => seq_read 1032 => vfs_read 1033 => sys_read 1034 => system_call_fastpath 1035 1036 1037This shows that the current tracer is "irqsoff" tracing the time 1038for which interrupts were disabled. It gives the trace version (which 1039never changes) and the version of the kernel upon which this was executed on 1040(3.8). Then it displays the max latency in microseconds (259 us). The number 1041of trace entries displayed and the total number (both are four: #4/4). 1042VP, KP, SP, and HP are always zero and are reserved for later use. 1043#P is the number of online CPUs (#P:4). 1044 1045The task is the process that was running when the latency 1046occurred. (ps pid: 6143). 1047 1048The start and stop (the functions in which the interrupts were 1049disabled and enabled respectively) that caused the latencies: 1050 1051 - __lock_task_sighand is where the interrupts were disabled. 1052 - _raw_spin_unlock_irqrestore is where they were enabled again. 1053 1054The next lines after the header are the trace itself. The header 1055explains which is which. 1056 1057 cmd: The name of the process in the trace. 1058 1059 pid: The PID of that process. 1060 1061 CPU#: The CPU which the process was running on. 1062 1063 irqs-off: 'd' interrupts are disabled. '.' otherwise. 1064 1065 need-resched: 1066 - 'B' all, TIF_NEED_RESCHED, PREEMPT_NEED_RESCHED and TIF_RESCHED_LAZY is set, 1067 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set, 1068 - 'n' only TIF_NEED_RESCHED is set, 1069 - 'p' only PREEMPT_NEED_RESCHED is set, 1070 - 'L' both PREEMPT_NEED_RESCHED and TIF_RESCHED_LAZY is set, 1071 - 'b' both TIF_NEED_RESCHED and TIF_RESCHED_LAZY is set, 1072 - 'l' only TIF_RESCHED_LAZY is set 1073 - '.' otherwise. 1074 1075 hardirq/softirq: 1076 - 'Z' - NMI occurred inside a hardirq 1077 - 'z' - NMI is running 1078 - 'H' - hard irq occurred inside a softirq. 1079 - 'h' - hard irq is running 1080 - 's' - soft irq is running 1081 - '.' - normal context. 1082 1083 preempt-depth: The level of preempt_disabled 1084 1085The above is mostly meaningful for kernel developers. 1086 1087 time: 1088 When the latency-format option is enabled, the trace file 1089 output includes a timestamp relative to the start of the 1090 trace. This differs from the output when latency-format 1091 is disabled, which includes an absolute timestamp. 1092 1093 delay: 1094 This is just to help catch your eye a bit better. And 1095 needs to be fixed to be only relative to the same CPU. 1096 The marks are determined by the difference between this 1097 current trace and the next trace. 1098 1099 - '$' - greater than 1 second 1100 - '@' - greater than 100 millisecond 1101 - '*' - greater than 10 millisecond 1102 - '#' - greater than 1000 microsecond 1103 - '!' - greater than 100 microsecond 1104 - '+' - greater than 10 microsecond 1105 - ' ' - less than or equal to 10 microsecond. 1106 1107 The rest is the same as the 'trace' file. 1108 1109 Note, the latency tracers will usually end with a back trace 1110 to easily find where the latency occurred. 1111 1112trace_options 1113------------- 1114 1115The trace_options file (or the options directory) is used to control 1116what gets printed in the trace output, or manipulate the tracers. 1117To see what is available, simply cat the file:: 1118 1119 cat trace_options 1120 print-parent 1121 nosym-offset 1122 nosym-addr 1123 noverbose 1124 noraw 1125 nohex 1126 nobin 1127 noblock 1128 nofields 1129 trace_printk 1130 annotate 1131 nouserstacktrace 1132 nosym-userobj 1133 noprintk-msg-only 1134 context-info 1135 nolatency-format 1136 record-cmd 1137 norecord-tgid 1138 overwrite 1139 nodisable_on_free 1140 irq-info 1141 markers 1142 noevent-fork 1143 function-trace 1144 nofunction-fork 1145 nodisplay-graph 1146 nostacktrace 1147 nobranch 1148 1149To disable one of the options, echo in the option prepended with 1150"no":: 1151 1152 echo noprint-parent > trace_options 1153 1154To enable an option, leave off the "no":: 1155 1156 echo sym-offset > trace_options 1157 1158Here are the available options: 1159 1160 print-parent 1161 On function traces, display the calling (parent) 1162 function as well as the function being traced. 1163 :: 1164 1165 print-parent: 1166 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul 1167 1168 noprint-parent: 1169 bash-4000 [01] 1477.606694: simple_strtoul 1170 1171 1172 sym-offset 1173 Display not only the function name, but also the 1174 offset in the function. For example, instead of 1175 seeing just "ktime_get", you will see 1176 "ktime_get+0xb/0x20". 1177 :: 1178 1179 sym-offset: 1180 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0 1181 1182 sym-addr 1183 This will also display the function address as well 1184 as the function name. 1185 :: 1186 1187 sym-addr: 1188 bash-4000 [01] 1477.606694: simple_strtoul <c0339346> 1189 1190 verbose 1191 This deals with the trace file when the 1192 latency-format option is enabled. 1193 :: 1194 1195 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \ 1196 (+0.000ms): simple_strtoul (kstrtoul) 1197 1198 raw 1199 This will display raw numbers. This option is best for 1200 use with user applications that can translate the raw 1201 numbers better than having it done in the kernel. 1202 1203 hex 1204 Similar to raw, but the numbers will be in a hexadecimal format. 1205 1206 bin 1207 This will print out the formats in raw binary. 1208 1209 block 1210 When set, reading trace_pipe will not block when polled. 1211 1212 fields 1213 Print the fields as described by their types. This is a better 1214 option than using hex, bin or raw, as it gives a better parsing 1215 of the content of the event. 1216 1217 trace_printk 1218 Can disable trace_printk() from writing into the buffer. 1219 1220 trace_printk_dest 1221 Set to have trace_printk() and similar internal tracing functions 1222 write into this instance. Note, only one trace instance can have 1223 this set. By setting this flag, it clears the trace_printk_dest flag 1224 of the instance that had it set previously. By default, the top 1225 level trace has this set, and will get it set again if another 1226 instance has it set then clears it. 1227 1228 This flag cannot be cleared by the top level instance, as it is the 1229 default instance. The only way the top level instance has this flag 1230 cleared, is by it being set in another instance. 1231 1232 copy_trace_marker 1233 If there are applications that hard code writing into the top level 1234 trace_marker file (/sys/kernel/tracing/trace_marker or trace_marker_raw), 1235 and the tooling would like it to go into an instance, this option can 1236 be used. Create an instance and set this option, and then all writes 1237 into the top level trace_marker file will also be redirected into this 1238 instance. 1239 1240 Note, by default this option is set for the top level instance. If it 1241 is disabled, then writes to the trace_marker or trace_marker_raw files 1242 will not be written into the top level file. If no instance has this 1243 option set, then a write will error with the errno of ENODEV. 1244 1245 annotate 1246 It is sometimes confusing when the CPU buffers are full 1247 and one CPU buffer had a lot of events recently, thus 1248 a shorter time frame, were another CPU may have only had 1249 a few events, which lets it have older events. When 1250 the trace is reported, it shows the oldest events first, 1251 and it may look like only one CPU ran (the one with the 1252 oldest events). When the annotate option is set, it will 1253 display when a new CPU buffer started:: 1254 1255 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on 1256 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on 1257 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore 1258 ##### CPU 2 buffer started #### 1259 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle 1260 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog 1261 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock 1262 1263 userstacktrace 1264 This option changes the trace. It records a 1265 stacktrace of the current user space thread after 1266 each trace event. 1267 1268 sym-userobj 1269 when user stacktrace are enabled, look up which 1270 object the address belongs to, and print a 1271 relative address. This is especially useful when 1272 ASLR is on, otherwise you don't get a chance to 1273 resolve the address to object/file/line after 1274 the app is no longer running 1275 1276 The lookup is performed when you read 1277 trace,trace_pipe. Example:: 1278 1279 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0 1280 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6] 1281 1282 1283 printk-msg-only 1284 When set, trace_printk()s will only show the format 1285 and not their parameters (if trace_bprintk() or 1286 trace_bputs() was used to save the trace_printk()). 1287 1288 context-info 1289 Show only the event data. Hides the comm, PID, 1290 timestamp, CPU, and other useful data. 1291 1292 latency-format 1293 This option changes the trace output. When it is enabled, 1294 the trace displays additional information about the 1295 latency, as described in "Latency trace format". 1296 1297 pause-on-trace 1298 When set, opening the trace file for read, will pause 1299 writing to the ring buffer (as if tracing_on was set to zero). 1300 This simulates the original behavior of the trace file. 1301 When the file is closed, tracing will be enabled again. 1302 1303 hash-ptr 1304 When set, "%p" in the event printk format displays the 1305 hashed pointer value instead of real address. 1306 This will be useful if you want to find out which hashed 1307 value is corresponding to the real value in trace log. 1308 1309 bitmask-list 1310 When enabled, bitmasks are displayed as a human-readable list of 1311 ranges (e.g., 0,2-5,7) using the printk "%*pbl" format specifier. 1312 When disabled (the default), bitmasks are displayed in the 1313 traditional hexadecimal bitmap representation. The list format is 1314 particularly useful for tracing CPU masks and other large bitmasks 1315 where individual bit positions are more meaningful than their 1316 hexadecimal encoding. 1317 1318 record-cmd 1319 When any event or tracer is enabled, a hook is enabled 1320 in the sched_switch trace point to fill comm cache 1321 with mapped pids and comms. But this may cause some 1322 overhead, and if you only care about pids, and not the 1323 name of the task, disabling this option can lower the 1324 impact of tracing. See "saved_cmdlines". 1325 1326 record-tgid 1327 When any event or tracer is enabled, a hook is enabled 1328 in the sched_switch trace point to fill the cache of 1329 mapped Thread Group IDs (TGID) mapping to pids. See 1330 "saved_tgids". 1331 1332 overwrite 1333 This controls what happens when the trace buffer is 1334 full. If "1" (default), the oldest events are 1335 discarded and overwritten. If "0", then the newest 1336 events are discarded. 1337 (see per_cpu/cpu0/stats for overrun and dropped) 1338 1339 disable_on_free 1340 When the free_buffer is closed, tracing will 1341 stop (tracing_on set to 0). 1342 1343 irq-info 1344 Shows the interrupt, preempt count, need resched data. 1345 When disabled, the trace looks like:: 1346 1347 # tracer: function 1348 # 1349 # entries-in-buffer/entries-written: 144405/9452052 #P:4 1350 # 1351 # TASK-PID CPU# TIMESTAMP FUNCTION 1352 # | | | | | 1353 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up 1354 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89 1355 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task 1356 1357 1358 markers 1359 When set, the trace_marker is writable (only by root). 1360 When disabled, the trace_marker will error with EINVAL 1361 on write. 1362 1363 event-fork 1364 When set, tasks with PIDs listed in set_event_pid will have 1365 the PIDs of their children added to set_event_pid when those 1366 tasks fork. Also, when tasks with PIDs in set_event_pid exit, 1367 their PIDs will be removed from the file. 1368 1369 This affects PIDs listed in set_event_notrace_pid as well. 1370 1371 function-trace 1372 The latency tracers will enable function tracing 1373 if this option is enabled (default it is). When 1374 it is disabled, the latency tracers do not trace 1375 functions. This keeps the overhead of the tracer down 1376 when performing latency tests. 1377 1378 function-fork 1379 When set, tasks with PIDs listed in set_ftrace_pid will 1380 have the PIDs of their children added to set_ftrace_pid 1381 when those tasks fork. Also, when tasks with PIDs in 1382 set_ftrace_pid exit, their PIDs will be removed from the 1383 file. 1384 1385 This affects PIDs in set_ftrace_notrace_pid as well. 1386 1387 display-graph 1388 When set, the latency tracers (irqsoff, wakeup, etc) will 1389 use function graph tracing instead of function tracing. 1390 1391 stacktrace 1392 When set, a stack trace is recorded after any trace event 1393 is recorded. 1394 1395 branch 1396 Enable branch tracing with the tracer. This enables branch 1397 tracer along with the currently set tracer. Enabling this 1398 with the "nop" tracer is the same as just enabling the 1399 "branch" tracer. 1400 1401.. tip:: Some tracers have their own options. They only appear in this 1402 file when the tracer is active. They always appear in the 1403 options directory. 1404 1405 1406Here are the per tracer options: 1407 1408Options for function tracer: 1409 1410 func_stack_trace 1411 When set, a stack trace is recorded after every 1412 function that is recorded. NOTE! Limit the functions 1413 that are recorded before enabling this, with 1414 "set_ftrace_filter" otherwise the system performance 1415 will be critically degraded. Remember to disable 1416 this option before clearing the function filter. 1417 1418Options for function_graph tracer: 1419 1420 Since the function_graph tracer has a slightly different output 1421 it has its own options to control what is displayed. 1422 1423 funcgraph-overrun 1424 When set, the "overrun" of the graph stack is 1425 displayed after each function traced. The 1426 overrun, is when the stack depth of the calls 1427 is greater than what is reserved for each task. 1428 Each task has a fixed array of functions to 1429 trace in the call graph. If the depth of the 1430 calls exceeds that, the function is not traced. 1431 The overrun is the number of functions missed 1432 due to exceeding this array. 1433 1434 funcgraph-cpu 1435 When set, the CPU number of the CPU where the trace 1436 occurred is displayed. 1437 1438 funcgraph-overhead 1439 When set, if the function takes longer than 1440 A certain amount, then a delay marker is 1441 displayed. See "delay" above, under the 1442 header description. 1443 1444 funcgraph-proc 1445 Unlike other tracers, the process' command line 1446 is not displayed by default, but instead only 1447 when a task is traced in and out during a context 1448 switch. Enabling this options has the command 1449 of each process displayed at every line. 1450 1451 funcgraph-duration 1452 At the end of each function (the return) 1453 the duration of the amount of time in the 1454 function is displayed in microseconds. 1455 1456 funcgraph-abstime 1457 When set, the timestamp is displayed at each line. 1458 1459 funcgraph-irqs 1460 When disabled, functions that happen inside an 1461 interrupt will not be traced. 1462 1463 funcgraph-tail 1464 When set, the return event will include the function 1465 that it represents. By default this is off, and 1466 only a closing curly bracket "}" is displayed for 1467 the return of a function. 1468 1469 funcgraph-retval 1470 When set, the return value of each traced function 1471 will be printed after an equal sign "=". By default 1472 this is off. 1473 1474 funcgraph-retval-hex 1475 When set, the return value will always be printed 1476 in hexadecimal format. If the option is not set and 1477 the return value is an error code, it will be printed 1478 in signed decimal format; otherwise it will also be 1479 printed in hexadecimal format. By default, this option 1480 is off. 1481 1482 sleep-time 1483 When running function graph tracer, to include 1484 the time a task schedules out in its function. 1485 When enabled, it will account time the task has been 1486 scheduled out as part of the function call. 1487 1488 graph-time 1489 When running function profiler with function graph tracer, 1490 to include the time to call nested functions. When this is 1491 not set, the time reported for the function will only 1492 include the time the function itself executed for, not the 1493 time for functions that it called. 1494 1495Options for blk tracer: 1496 1497 blk_classic 1498 Shows a more minimalistic output. 1499 1500 1501irqsoff 1502------- 1503 1504When interrupts are disabled, the CPU can not react to any other 1505external event (besides NMIs and SMIs). This prevents the timer 1506interrupt from triggering or the mouse interrupt from letting 1507the kernel know of a new mouse event. The result is a latency 1508with the reaction time. 1509 1510The irqsoff tracer tracks the time for which interrupts are 1511disabled. When a new maximum latency is hit, the tracer saves 1512the trace leading up to that latency point so that every time a 1513new maximum is reached, the old saved trace is discarded and the 1514new trace is saved. 1515 1516To reset the maximum, echo 0 into tracing_max_latency. Here is 1517an example:: 1518 1519 # echo 0 > options/function-trace 1520 # echo irqsoff > current_tracer 1521 # echo 1 > tracing_on 1522 # echo 0 > tracing_max_latency 1523 # ls -ltr 1524 [...] 1525 # echo 0 > tracing_on 1526 # cat trace 1527 # tracer: irqsoff 1528 # 1529 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1530 # -------------------------------------------------------------------- 1531 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1532 # ----------------- 1533 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0) 1534 # ----------------- 1535 # => started at: run_timer_softirq 1536 # => ended at: run_timer_softirq 1537 # 1538 # 1539 # _------=> CPU# 1540 # / _-----=> irqs-off 1541 # | / _----=> need-resched 1542 # || / _---=> hardirq/softirq 1543 # ||| / _--=> preempt-depth 1544 # |||| / delay 1545 # cmd pid ||||| time | caller 1546 # \ / ||||| \ | / 1547 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq 1548 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq 1549 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq 1550 <idle>-0 0dNs3 25us : <stack trace> 1551 => _raw_spin_unlock_irq 1552 => run_timer_softirq 1553 => __do_softirq 1554 => call_softirq 1555 => do_softirq 1556 => irq_exit 1557 => smp_apic_timer_interrupt 1558 => apic_timer_interrupt 1559 => rcu_idle_exit 1560 => cpu_idle 1561 => rest_init 1562 => start_kernel 1563 => x86_64_start_reservations 1564 => x86_64_start_kernel 1565 1566Here we see that we had a latency of 16 microseconds (which is 1567very good). The _raw_spin_lock_irq in run_timer_softirq disabled 1568interrupts. The difference between the 16 and the displayed 1569timestamp 25us occurred because the clock was incremented 1570between the time of recording the max latency and the time of 1571recording the function that had that latency. 1572 1573Note the above example had function-trace not set. If we set 1574function-trace, we get a much larger output:: 1575 1576 with echo 1 > options/function-trace 1577 1578 # tracer: irqsoff 1579 # 1580 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1581 # -------------------------------------------------------------------- 1582 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1583 # ----------------- 1584 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0) 1585 # ----------------- 1586 # => started at: ata_scsi_queuecmd 1587 # => ended at: ata_scsi_queuecmd 1588 # 1589 # 1590 # _------=> CPU# 1591 # / _-----=> irqs-off 1592 # | / _----=> need-resched 1593 # || / _---=> hardirq/softirq 1594 # ||| / _--=> preempt-depth 1595 # |||| / delay 1596 # cmd pid ||||| time | caller 1597 # \ / ||||| \ | / 1598 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd 1599 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave 1600 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd 1601 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev 1602 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev 1603 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd 1604 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd 1605 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd 1606 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat 1607 [...] 1608 bash-2042 3d..1 67us : delay_tsc <-__delay 1609 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc 1610 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc 1611 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc 1612 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc 1613 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue 1614 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1615 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1616 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd 1617 bash-2042 3d..1 120us : <stack trace> 1618 => _raw_spin_unlock_irqrestore 1619 => ata_scsi_queuecmd 1620 => scsi_dispatch_cmd 1621 => scsi_request_fn 1622 => __blk_run_queue_uncond 1623 => __blk_run_queue 1624 => blk_queue_bio 1625 => submit_bio_noacct 1626 => submit_bio 1627 => submit_bh 1628 => __ext3_get_inode_loc 1629 => ext3_iget 1630 => ext3_lookup 1631 => lookup_real 1632 => __lookup_hash 1633 => walk_component 1634 => lookup_last 1635 => path_lookupat 1636 => filename_lookup 1637 => user_path_at_empty 1638 => user_path_at 1639 => vfs_fstatat 1640 => vfs_stat 1641 => sys_newstat 1642 => system_call_fastpath 1643 1644 1645Here we traced a 71 microsecond latency. But we also see all the 1646functions that were called during that time. Note that by 1647enabling function tracing, we incur an added overhead. This 1648overhead may extend the latency times. But nevertheless, this 1649trace has provided some very helpful debugging information. 1650 1651If we prefer function graph output instead of function, we can set 1652display-graph option:: 1653 1654 with echo 1 > options/display-graph 1655 1656 # tracer: irqsoff 1657 # 1658 # irqsoff latency trace v1.1.5 on 4.20.0-rc6+ 1659 # -------------------------------------------------------------------- 1660 # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4) 1661 # ----------------- 1662 # | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0) 1663 # ----------------- 1664 # => started at: free_debug_processing 1665 # => ended at: return_to_handler 1666 # 1667 # 1668 # _-----=> irqs-off 1669 # / _----=> need-resched 1670 # | / _---=> hardirq/softirq 1671 # || / _--=> preempt-depth 1672 # ||| / 1673 # REL TIME CPU TASK/PID |||| DURATION FUNCTION CALLS 1674 # | | | | |||| | | | | | | 1675 0 us | 0) bash-1507 | d... | 0.000 us | _raw_spin_lock_irqsave(); 1676 0 us | 0) bash-1507 | d..1 | 0.378 us | do_raw_spin_trylock(); 1677 1 us | 0) bash-1507 | d..2 | | set_track() { 1678 2 us | 0) bash-1507 | d..2 | | save_stack_trace() { 1679 2 us | 0) bash-1507 | d..2 | | __save_stack_trace() { 1680 3 us | 0) bash-1507 | d..2 | | __unwind_start() { 1681 3 us | 0) bash-1507 | d..2 | | get_stack_info() { 1682 3 us | 0) bash-1507 | d..2 | 0.351 us | in_task_stack(); 1683 4 us | 0) bash-1507 | d..2 | 1.107 us | } 1684 [...] 1685 3750 us | 0) bash-1507 | d..1 | 0.516 us | do_raw_spin_unlock(); 1686 3750 us | 0) bash-1507 | d..1 | 0.000 us | _raw_spin_unlock_irqrestore(); 1687 3764 us | 0) bash-1507 | d..1 | 0.000 us | tracer_hardirqs_on(); 1688 bash-1507 0d..1 3792us : <stack trace> 1689 => free_debug_processing 1690 => __slab_free 1691 => kmem_cache_free 1692 => vm_area_free 1693 => remove_vma 1694 => exit_mmap 1695 => mmput 1696 => begin_new_exec 1697 => load_elf_binary 1698 => search_binary_handler 1699 => __do_execve_file.isra.32 1700 => __x64_sys_execve 1701 => do_syscall_64 1702 => entry_SYSCALL_64_after_hwframe 1703 1704preemptoff 1705---------- 1706 1707When preemption is disabled, we may be able to receive 1708interrupts but the task cannot be preempted and a higher 1709priority task must wait for preemption to be enabled again 1710before it can preempt a lower priority task. 1711 1712The preemptoff tracer traces the places that disable preemption. 1713Like the irqsoff tracer, it records the maximum latency for 1714which preemption was disabled. The control of preemptoff tracer 1715is much like the irqsoff tracer. 1716:: 1717 1718 # echo 0 > options/function-trace 1719 # echo preemptoff > current_tracer 1720 # echo 1 > tracing_on 1721 # echo 0 > tracing_max_latency 1722 # ls -ltr 1723 [...] 1724 # echo 0 > tracing_on 1725 # cat trace 1726 # tracer: preemptoff 1727 # 1728 # preemptoff latency trace v1.1.5 on 3.8.0-test+ 1729 # -------------------------------------------------------------------- 1730 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1731 # ----------------- 1732 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0) 1733 # ----------------- 1734 # => started at: do_IRQ 1735 # => ended at: do_IRQ 1736 # 1737 # 1738 # _------=> CPU# 1739 # / _-----=> irqs-off 1740 # | / _----=> need-resched 1741 # || / _---=> hardirq/softirq 1742 # ||| / _--=> preempt-depth 1743 # |||| / delay 1744 # cmd pid ||||| time | caller 1745 # \ / ||||| \ | / 1746 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ 1747 sshd-1991 1d..1 46us : irq_exit <-do_IRQ 1748 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ 1749 sshd-1991 1d..1 52us : <stack trace> 1750 => sub_preempt_count 1751 => irq_exit 1752 => do_IRQ 1753 => ret_from_intr 1754 1755 1756This has some more changes. Preemption was disabled when an 1757interrupt came in (notice the 'h'), and was enabled on exit. 1758But we also see that interrupts have been disabled when entering 1759the preempt off section and leaving it (the 'd'). We do not know if 1760interrupts were enabled in the mean time or shortly after this 1761was over. 1762:: 1763 1764 # tracer: preemptoff 1765 # 1766 # preemptoff latency trace v1.1.5 on 3.8.0-test+ 1767 # -------------------------------------------------------------------- 1768 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1769 # ----------------- 1770 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0) 1771 # ----------------- 1772 # => started at: wake_up_new_task 1773 # => ended at: task_rq_unlock 1774 # 1775 # 1776 # _------=> CPU# 1777 # / _-----=> irqs-off 1778 # | / _----=> need-resched 1779 # || / _---=> hardirq/softirq 1780 # ||| / _--=> preempt-depth 1781 # |||| / delay 1782 # cmd pid ||||| time | caller 1783 # \ / ||||| \ | / 1784 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task 1785 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq 1786 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair 1787 bash-1994 1d..1 1us : source_load <-select_task_rq_fair 1788 bash-1994 1d..1 1us : source_load <-select_task_rq_fair 1789 [...] 1790 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt 1791 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter 1792 bash-1994 1d..1 13us : add_preempt_count <-irq_enter 1793 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt 1794 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt 1795 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt 1796 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock 1797 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt 1798 [...] 1799 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event 1800 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt 1801 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit 1802 bash-1994 1d..2 36us : do_softirq <-irq_exit 1803 bash-1994 1d..2 36us : __do_softirq <-call_softirq 1804 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq 1805 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq 1806 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq 1807 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock 1808 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq 1809 [...] 1810 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks 1811 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq 1812 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable 1813 bash-1994 1dN.2 82us : idle_cpu <-irq_exit 1814 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit 1815 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit 1816 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock 1817 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock 1818 bash-1994 1.N.1 104us : <stack trace> 1819 => sub_preempt_count 1820 => _raw_spin_unlock_irqrestore 1821 => task_rq_unlock 1822 => wake_up_new_task 1823 => do_fork 1824 => sys_clone 1825 => stub_clone 1826 1827 1828The above is an example of the preemptoff trace with 1829function-trace set. Here we see that interrupts were not disabled 1830the entire time. The irq_enter code lets us know that we entered 1831an interrupt 'h'. Before that, the functions being traced still 1832show that it is not in an interrupt, but we can see from the 1833functions themselves that this is not the case. 1834 1835preemptirqsoff 1836-------------- 1837 1838Knowing the locations that have interrupts disabled or 1839preemption disabled for the longest times is helpful. But 1840sometimes we would like to know when either preemption and/or 1841interrupts are disabled. 1842 1843Consider the following code:: 1844 1845 local_irq_disable(); 1846 call_function_with_irqs_off(); 1847 preempt_disable(); 1848 call_function_with_irqs_and_preemption_off(); 1849 local_irq_enable(); 1850 call_function_with_preemption_off(); 1851 preempt_enable(); 1852 1853The irqsoff tracer will record the total length of 1854call_function_with_irqs_off() and 1855call_function_with_irqs_and_preemption_off(). 1856 1857The preemptoff tracer will record the total length of 1858call_function_with_irqs_and_preemption_off() and 1859call_function_with_preemption_off(). 1860 1861But neither will trace the time that interrupts and/or 1862preemption is disabled. This total time is the time that we can 1863not schedule. To record this time, use the preemptirqsoff 1864tracer. 1865 1866Again, using this trace is much like the irqsoff and preemptoff 1867tracers. 1868:: 1869 1870 # echo 0 > options/function-trace 1871 # echo preemptirqsoff > current_tracer 1872 # echo 1 > tracing_on 1873 # echo 0 > tracing_max_latency 1874 # ls -ltr 1875 [...] 1876 # echo 0 > tracing_on 1877 # cat trace 1878 # tracer: preemptirqsoff 1879 # 1880 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+ 1881 # -------------------------------------------------------------------- 1882 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1883 # ----------------- 1884 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0) 1885 # ----------------- 1886 # => started at: ata_scsi_queuecmd 1887 # => ended at: ata_scsi_queuecmd 1888 # 1889 # 1890 # _------=> CPU# 1891 # / _-----=> irqs-off 1892 # | / _----=> need-resched 1893 # || / _---=> hardirq/softirq 1894 # ||| / _--=> preempt-depth 1895 # |||| / delay 1896 # cmd pid ||||| time | caller 1897 # \ / ||||| \ | / 1898 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd 1899 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1900 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd 1901 ls-2230 3...1 111us : <stack trace> 1902 => sub_preempt_count 1903 => _raw_spin_unlock_irqrestore 1904 => ata_scsi_queuecmd 1905 => scsi_dispatch_cmd 1906 => scsi_request_fn 1907 => __blk_run_queue_uncond 1908 => __blk_run_queue 1909 => blk_queue_bio 1910 => submit_bio_noacct 1911 => submit_bio 1912 => submit_bh 1913 => ext3_bread 1914 => ext3_dir_bread 1915 => htree_dirblock_to_tree 1916 => ext3_htree_fill_tree 1917 => ext3_readdir 1918 => vfs_readdir 1919 => sys_getdents 1920 => system_call_fastpath 1921 1922 1923The trace_hardirqs_off_thunk is called from assembly on x86 when 1924interrupts are disabled in the assembly code. Without the 1925function tracing, we do not know if interrupts were enabled 1926within the preemption points. We do see that it started with 1927preemption enabled. 1928 1929Here is a trace with function-trace set:: 1930 1931 # tracer: preemptirqsoff 1932 # 1933 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+ 1934 # -------------------------------------------------------------------- 1935 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1936 # ----------------- 1937 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0) 1938 # ----------------- 1939 # => started at: schedule 1940 # => ended at: mutex_unlock 1941 # 1942 # 1943 # _------=> CPU# 1944 # / _-----=> irqs-off 1945 # | / _----=> need-resched 1946 # || / _---=> hardirq/softirq 1947 # ||| / _--=> preempt-depth 1948 # |||| / delay 1949 # cmd pid ||||| time | caller 1950 # \ / ||||| \ | / 1951 kworker/-59 3...1 0us : __schedule <-schedule 1952 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch 1953 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq 1954 kworker/-59 3d..2 1us : deactivate_task <-__schedule 1955 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task 1956 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task 1957 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task 1958 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair 1959 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr 1960 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr 1961 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge 1962 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge 1963 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair 1964 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair 1965 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair 1966 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair 1967 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair 1968 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair 1969 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule 1970 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping 1971 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule 1972 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task 1973 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair 1974 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair 1975 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity 1976 ls-2269 3d..2 7us : finish_task_switch <-__schedule 1977 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch 1978 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr 1979 ls-2269 3d..2 8us : irq_enter <-do_IRQ 1980 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter 1981 ls-2269 3d..2 9us : add_preempt_count <-irq_enter 1982 ls-2269 3d.h2 9us : exit_idle <-do_IRQ 1983 [...] 1984 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock 1985 ls-2269 3d.h2 20us : irq_exit <-do_IRQ 1986 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit 1987 ls-2269 3d..3 21us : do_softirq <-irq_exit 1988 ls-2269 3d..3 21us : __do_softirq <-call_softirq 1989 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq 1990 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip 1991 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip 1992 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr 1993 ls-2269 3d.s5 31us : irq_enter <-do_IRQ 1994 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter 1995 [...] 1996 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter 1997 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter 1998 ls-2269 3d.H5 32us : exit_idle <-do_IRQ 1999 ls-2269 3d.H5 32us : handle_irq <-do_IRQ 2000 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq 2001 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq 2002 [...] 2003 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll 2004 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action 2005 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq 2006 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable 2007 ls-2269 3d..3 159us : idle_cpu <-irq_exit 2008 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit 2009 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit 2010 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock 2011 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock 2012 ls-2269 3d... 186us : <stack trace> 2013 => __mutex_unlock_slowpath 2014 => mutex_unlock 2015 => process_output 2016 => n_tty_write 2017 => tty_write 2018 => vfs_write 2019 => sys_write 2020 => system_call_fastpath 2021 2022This is an interesting trace. It started with kworker running and 2023scheduling out and ls taking over. But as soon as ls released the 2024rq lock and enabled interrupts (but not preemption) an interrupt 2025triggered. When the interrupt finished, it started running softirqs. 2026But while the softirq was running, another interrupt triggered. 2027When an interrupt is running inside a softirq, the annotation is 'H'. 2028 2029 2030wakeup 2031------ 2032 2033One common case that people are interested in tracing is the 2034time it takes for a task that is woken to actually wake up. 2035Now for non Real-Time tasks, this can be arbitrary. But tracing 2036it nonetheless can be interesting. 2037 2038Without function tracing:: 2039 2040 # echo 0 > options/function-trace 2041 # echo wakeup > current_tracer 2042 # echo 1 > tracing_on 2043 # echo 0 > tracing_max_latency 2044 # chrt -f 5 sleep 1 2045 # echo 0 > tracing_on 2046 # cat trace 2047 # tracer: wakeup 2048 # 2049 # wakeup latency trace v1.1.5 on 3.8.0-test+ 2050 # -------------------------------------------------------------------- 2051 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2052 # ----------------- 2053 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0) 2054 # ----------------- 2055 # 2056 # _------=> CPU# 2057 # / _-----=> irqs-off 2058 # | / _----=> need-resched 2059 # || / _---=> hardirq/softirq 2060 # ||| / _--=> preempt-depth 2061 # |||| / delay 2062 # cmd pid ||||| time | caller 2063 # \ / ||||| \ | / 2064 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H 2065 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up 2066 <idle>-0 3d..3 15us : __schedule <-schedule 2067 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H 2068 2069The tracer only traces the highest priority task in the system 2070to avoid tracing the normal circumstances. Here we see that 2071the kworker with a nice priority of -20 (not very nice), took 2072just 15 microseconds from the time it woke up, to the time it 2073ran. 2074 2075Non Real-Time tasks are not that interesting. A more interesting 2076trace is to concentrate only on Real-Time tasks. 2077 2078wakeup_rt 2079--------- 2080 2081In a Real-Time environment it is very important to know the 2082wakeup time it takes for the highest priority task that is woken 2083up to the time that it executes. This is also known as "schedule 2084latency". I stress the point that this is about RT tasks. It is 2085also important to know the scheduling latency of non-RT tasks, 2086but the average schedule latency is better for non-RT tasks. 2087Tools like LatencyTop are more appropriate for such 2088measurements. 2089 2090Real-Time environments are interested in the worst case latency. 2091That is the longest latency it takes for something to happen, 2092and not the average. We can have a very fast scheduler that may 2093only have a large latency once in a while, but that would not 2094work well with Real-Time tasks. The wakeup_rt tracer was designed 2095to record the worst case wakeups of RT tasks. Non-RT tasks are 2096not recorded because the tracer only records one worst case and 2097tracing non-RT tasks that are unpredictable will overwrite the 2098worst case latency of RT tasks (just run the normal wakeup 2099tracer for a while to see that effect). 2100 2101Since this tracer only deals with RT tasks, we will run this 2102slightly differently than we did with the previous tracers. 2103Instead of performing an 'ls', we will run 'sleep 1' under 2104'chrt' which changes the priority of the task. 2105:: 2106 2107 # echo 0 > options/function-trace 2108 # echo wakeup_rt > current_tracer 2109 # echo 1 > tracing_on 2110 # echo 0 > tracing_max_latency 2111 # chrt -f 5 sleep 1 2112 # echo 0 > tracing_on 2113 # cat trace 2114 # tracer: wakeup 2115 # 2116 # tracer: wakeup_rt 2117 # 2118 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2119 # -------------------------------------------------------------------- 2120 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2121 # ----------------- 2122 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5) 2123 # ----------------- 2124 # 2125 # _------=> CPU# 2126 # / _-----=> irqs-off 2127 # | / _----=> need-resched 2128 # || / _---=> hardirq/softirq 2129 # ||| / _--=> preempt-depth 2130 # |||| / delay 2131 # cmd pid ||||| time | caller 2132 # \ / ||||| \ | / 2133 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep 2134 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up 2135 <idle>-0 3d..3 5us : __schedule <-schedule 2136 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep 2137 2138 2139Running this on an idle system, we see that it only took 5 microseconds 2140to perform the task switch. Note, since the trace point in the schedule 2141is before the actual "switch", we stop the tracing when the recorded task 2142is about to schedule in. This may change if we add a new marker at the 2143end of the scheduler. 2144 2145Notice that the recorded task is 'sleep' with the PID of 2389 2146and it has an rt_prio of 5. This priority is user-space priority 2147and not the internal kernel priority. The policy is 1 for 2148SCHED_FIFO and 2 for SCHED_RR. 2149 2150Note, that the trace data shows the internal priority (99 - rtprio). 2151:: 2152 2153 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep 2154 2155The 0:120:R means idle was running with a nice priority of 0 (120 - 120) 2156and in the running state 'R'. The sleep task was scheduled in with 21572389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94) 2158and it too is in the running state. 2159 2160Doing the same with chrt -r 5 and function-trace set. 2161:: 2162 2163 echo 1 > options/function-trace 2164 2165 # tracer: wakeup_rt 2166 # 2167 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2168 # -------------------------------------------------------------------- 2169 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2170 # ----------------- 2171 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5) 2172 # ----------------- 2173 # 2174 # _------=> CPU# 2175 # / _-----=> irqs-off 2176 # | / _----=> need-resched 2177 # || / _---=> hardirq/softirq 2178 # ||| / _--=> preempt-depth 2179 # |||| / delay 2180 # cmd pid ||||| time | caller 2181 # \ / ||||| \ | / 2182 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep 2183 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up 2184 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup 2185 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr 2186 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup 2187 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up 2188 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock 2189 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up 2190 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up 2191 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2192 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer 2193 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock 2194 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt 2195 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock 2196 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt 2197 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event 2198 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event 2199 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event 2200 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt 2201 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit 2202 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit 2203 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit 2204 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit 2205 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit 2206 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle 2207 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit 2208 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle 2209 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit 2210 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit 2211 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit 2212 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit 2213 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz 2214 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock 2215 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz 2216 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update 2217 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz 2218 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock 2219 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit 2220 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit 2221 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit 2222 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel 2223 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel 2224 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18 2225 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave 2226 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16 2227 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer 2228 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram 2229 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event 2230 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event 2231 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event 2232 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel 2233 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2234 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit 2235 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward 2236 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward 2237 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11 2238 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns 2239 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns 2240 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18 2241 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave 2242 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns 2243 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns 2244 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns 2245 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event 2246 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event 2247 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event 2248 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns 2249 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2250 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit 2251 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks 2252 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle 2253 <idle>-0 3.N.. 25us : schedule <-cpu_idle 2254 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule 2255 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule 2256 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule 2257 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch 2258 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch 2259 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule 2260 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq 2261 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule 2262 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task 2263 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task 2264 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt 2265 <idle>-0 3d..3 29us : __schedule <-preempt_schedule 2266 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep 2267 2268This isn't that big of a trace, even with function tracing enabled, 2269so I included the entire trace. 2270 2271The interrupt went off while when the system was idle. Somewhere 2272before task_woken_rt() was called, the NEED_RESCHED flag was set, 2273this is indicated by the first occurrence of the 'N' flag. 2274 2275Latency tracing and events 2276-------------------------- 2277As function tracing can induce a much larger latency, but without 2278seeing what happens within the latency it is hard to know what 2279caused it. There is a middle ground, and that is with enabling 2280events. 2281:: 2282 2283 # echo 0 > options/function-trace 2284 # echo wakeup_rt > current_tracer 2285 # echo 1 > events/enable 2286 # echo 1 > tracing_on 2287 # echo 0 > tracing_max_latency 2288 # chrt -f 5 sleep 1 2289 # echo 0 > tracing_on 2290 # cat trace 2291 # tracer: wakeup_rt 2292 # 2293 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2294 # -------------------------------------------------------------------- 2295 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2296 # ----------------- 2297 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5) 2298 # ----------------- 2299 # 2300 # _------=> CPU# 2301 # / _-----=> irqs-off 2302 # | / _----=> need-resched 2303 # || / _---=> hardirq/softirq 2304 # ||| / _--=> preempt-depth 2305 # |||| / delay 2306 # cmd pid ||||| time | caller 2307 # \ / ||||| \ | / 2308 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep 2309 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up 2310 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002 2311 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8 2312 <idle>-0 2.N.2 2us : power_end: cpu_id=2 2313 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2 2314 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0 2315 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000 2316 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch 2317 <idle>-0 2.N.2 5us : rcu_utilization: End context switch 2318 <idle>-0 2d..3 6us : __schedule <-schedule 2319 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep 2320 2321 2322Hardware Latency Detector 2323------------------------- 2324 2325The hardware latency detector is executed by enabling the "hwlat" tracer. 2326 2327NOTE, this tracer will affect the performance of the system as it will 2328periodically make a CPU constantly busy with interrupts disabled. 2329:: 2330 2331 # echo hwlat > current_tracer 2332 # sleep 100 2333 # cat trace 2334 # tracer: hwlat 2335 # 2336 # entries-in-buffer/entries-written: 13/13 #P:8 2337 # 2338 # _-----=> irqs-off 2339 # / _----=> need-resched 2340 # | / _---=> hardirq/softirq 2341 # || / _--=> preempt-depth 2342 # ||| / delay 2343 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2344 # | | | |||| | | 2345 <...>-1729 [001] d... 678.473449: #1 inner/outer(us): 11/12 ts:1581527483.343962693 count:6 2346 <...>-1729 [004] d... 689.556542: #2 inner/outer(us): 16/9 ts:1581527494.889008092 count:1 2347 <...>-1729 [005] d... 714.756290: #3 inner/outer(us): 16/16 ts:1581527519.678961629 count:5 2348 <...>-1729 [001] d... 718.788247: #4 inner/outer(us): 9/17 ts:1581527523.889012713 count:1 2349 <...>-1729 [002] d... 719.796341: #5 inner/outer(us): 13/9 ts:1581527524.912872606 count:1 2350 <...>-1729 [006] d... 844.787091: #6 inner/outer(us): 9/12 ts:1581527649.889048502 count:2 2351 <...>-1729 [003] d... 849.827033: #7 inner/outer(us): 18/9 ts:1581527654.889013793 count:1 2352 <...>-1729 [007] d... 853.859002: #8 inner/outer(us): 9/12 ts:1581527658.889065736 count:1 2353 <...>-1729 [001] d... 855.874978: #9 inner/outer(us): 9/11 ts:1581527660.861991877 count:1 2354 <...>-1729 [001] d... 863.938932: #10 inner/outer(us): 9/11 ts:1581527668.970010500 count:1 nmi-total:7 nmi-count:1 2355 <...>-1729 [007] d... 878.050780: #11 inner/outer(us): 9/12 ts:1581527683.385002600 count:1 nmi-total:5 nmi-count:1 2356 <...>-1729 [007] d... 886.114702: #12 inner/outer(us): 9/12 ts:1581527691.385001600 count:1 2357 2358 2359The above output is somewhat the same in the header. All events will have 2360interrupts disabled 'd'. Under the FUNCTION title there is: 2361 2362 #1 2363 This is the count of events recorded that were greater than the 2364 tracing_threshold (See below). 2365 2366 inner/outer(us): 11/11 2367 2368 This shows two numbers as "inner latency" and "outer latency". The test 2369 runs in a loop checking a timestamp twice. The latency detected within 2370 the two timestamps is the "inner latency" and the latency detected 2371 after the previous timestamp and the next timestamp in the loop is 2372 the "outer latency". 2373 2374 ts:1581527483.343962693 2375 2376 The absolute timestamp that the first latency was recorded in the window. 2377 2378 count:6 2379 2380 The number of times a latency was detected during the window. 2381 2382 nmi-total:7 nmi-count:1 2383 2384 On architectures that support it, if an NMI comes in during the 2385 test, the time spent in NMI is reported in "nmi-total" (in 2386 microseconds). 2387 2388 All architectures that have NMIs will show the "nmi-count" if an 2389 NMI comes in during the test. 2390 2391hwlat files: 2392 2393 tracing_threshold 2394 This gets automatically set to "10" to represent 10 2395 microseconds. This is the threshold of latency that 2396 needs to be detected before the trace will be recorded. 2397 2398 Note, when hwlat tracer is finished (another tracer is 2399 written into "current_tracer"), the original value for 2400 tracing_threshold is placed back into this file. 2401 2402 hwlat_detector/width 2403 The length of time the test runs with interrupts disabled. 2404 2405 hwlat_detector/window 2406 The length of time of the window which the test 2407 runs. That is, the test will run for "width" 2408 microseconds per "window" microseconds 2409 2410 tracing_cpumask 2411 When the test is started. A kernel thread is created that 2412 runs the test. This thread will alternate between CPUs 2413 listed in the tracing_cpumask between each period 2414 (one "window"). To limit the test to specific CPUs 2415 set the mask in this file to only the CPUs that the test 2416 should run on. 2417 2418function 2419-------- 2420 2421This tracer is the function tracer. Enabling the function tracer 2422can be done from the debug file system. Make sure the 2423ftrace_enabled is set; otherwise this tracer is a nop. 2424See the "ftrace_enabled" section below. 2425:: 2426 2427 # sysctl kernel.ftrace_enabled=1 2428 # echo function > current_tracer 2429 # echo 1 > tracing_on 2430 # usleep 1 2431 # echo 0 > tracing_on 2432 # cat trace 2433 # tracer: function 2434 # 2435 # entries-in-buffer/entries-written: 24799/24799 #P:4 2436 # 2437 # _-----=> irqs-off 2438 # / _----=> need-resched 2439 # | / _---=> hardirq/softirq 2440 # || / _--=> preempt-depth 2441 # ||| / delay 2442 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2443 # | | | |||| | | 2444 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write 2445 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock 2446 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify 2447 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify 2448 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify 2449 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock 2450 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock 2451 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify 2452 [...] 2453 2454 2455Note: function tracer uses ring buffers to store the above 2456entries. The newest data may overwrite the oldest data. 2457Sometimes using echo to stop the trace is not sufficient because 2458the tracing could have overwritten the data that you wanted to 2459record. For this reason, it is sometimes better to disable 2460tracing directly from a program. This allows you to stop the 2461tracing at the point that you hit the part that you are 2462interested in. To disable the tracing directly from a C program, 2463something like following code snippet can be used:: 2464 2465 int trace_fd; 2466 [...] 2467 int main(int argc, char *argv[]) { 2468 [...] 2469 trace_fd = open(tracing_file("tracing_on"), O_WRONLY); 2470 [...] 2471 if (condition_hit()) { 2472 write(trace_fd, "0", 1); 2473 } 2474 [...] 2475 } 2476 2477 2478Single thread tracing 2479--------------------- 2480 2481By writing into set_ftrace_pid you can trace a 2482single thread. For example:: 2483 2484 # cat set_ftrace_pid 2485 no pid 2486 # echo 3111 > set_ftrace_pid 2487 # cat set_ftrace_pid 2488 3111 2489 # echo function > current_tracer 2490 # cat trace | head 2491 # tracer: function 2492 # 2493 # TASK-PID CPU# TIMESTAMP FUNCTION 2494 # | | | | | 2495 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return 2496 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range 2497 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel 2498 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel 2499 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll 2500 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll 2501 # echo > set_ftrace_pid 2502 # cat trace |head 2503 # tracer: function 2504 # 2505 # TASK-PID CPU# TIMESTAMP FUNCTION 2506 # | | | | | 2507 ##### CPU 3 buffer started #### 2508 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait 2509 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry 2510 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry 2511 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit 2512 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit 2513 2514If you want to trace a function when executing, you could use 2515something like this simple program. 2516:: 2517 2518 #include <stdio.h> 2519 #include <stdlib.h> 2520 #include <sys/types.h> 2521 #include <sys/stat.h> 2522 #include <fcntl.h> 2523 #include <unistd.h> 2524 #include <string.h> 2525 2526 #define _STR(x) #x 2527 #define STR(x) _STR(x) 2528 #define MAX_PATH 256 2529 2530 const char *find_tracefs(void) 2531 { 2532 static char tracefs[MAX_PATH+1]; 2533 static int tracefs_found; 2534 char type[100]; 2535 FILE *fp; 2536 2537 if (tracefs_found) 2538 return tracefs; 2539 2540 if ((fp = fopen("/proc/mounts","r")) == NULL) { 2541 perror("/proc/mounts"); 2542 return NULL; 2543 } 2544 2545 while (fscanf(fp, "%*s %" 2546 STR(MAX_PATH) 2547 "s %99s %*s %*d %*d\n", 2548 tracefs, type) == 2) { 2549 if (strcmp(type, "tracefs") == 0) 2550 break; 2551 } 2552 fclose(fp); 2553 2554 if (strcmp(type, "tracefs") != 0) { 2555 fprintf(stderr, "tracefs not mounted"); 2556 return NULL; 2557 } 2558 2559 strcat(tracefs, "/tracing/"); 2560 tracefs_found = 1; 2561 2562 return tracefs; 2563 } 2564 2565 const char *tracing_file(const char *file_name) 2566 { 2567 static char trace_file[MAX_PATH+1]; 2568 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name); 2569 return trace_file; 2570 } 2571 2572 int main (int argc, char **argv) 2573 { 2574 if (argc < 1) 2575 exit(-1); 2576 2577 if (fork() > 0) { 2578 int fd, ffd; 2579 char line[64]; 2580 int s; 2581 2582 ffd = open(tracing_file("current_tracer"), O_WRONLY); 2583 if (ffd < 0) 2584 exit(-1); 2585 write(ffd, "nop", 3); 2586 2587 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY); 2588 s = sprintf(line, "%d\n", getpid()); 2589 write(fd, line, s); 2590 2591 write(ffd, "function", 8); 2592 2593 close(fd); 2594 close(ffd); 2595 2596 execvp(argv[1], argv+1); 2597 } 2598 2599 return 0; 2600 } 2601 2602Or this simple script! 2603:: 2604 2605 #!/bin/bash 2606 2607 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts` 2608 echo 0 > $tracefs/tracing_on 2609 echo $$ > $tracefs/set_ftrace_pid 2610 echo function > $tracefs/current_tracer 2611 echo 1 > $tracefs/tracing_on 2612 exec "$@" 2613 2614 2615function graph tracer 2616--------------------------- 2617 2618This tracer is similar to the function tracer except that it 2619probes a function on its entry and its exit. This is done by 2620using a dynamically allocated stack of return addresses in each 2621task_struct. On function entry the tracer overwrites the return 2622address of each function traced to set a custom probe. Thus the 2623original return address is stored on the stack of return address 2624in the task_struct. 2625 2626Probing on both ends of a function leads to special features 2627such as: 2628 2629- measure of a function's time execution 2630- having a reliable call stack to draw function calls graph 2631 2632This tracer is useful in several situations: 2633 2634- you want to find the reason of a strange kernel behavior and 2635 need to see what happens in detail on any areas (or specific 2636 ones). 2637 2638- you are experiencing weird latencies but it's difficult to 2639 find its origin. 2640 2641- you want to find quickly which path is taken by a specific 2642 function 2643 2644- you just want to peek inside a working kernel and want to see 2645 what happens there. 2646 2647:: 2648 2649 # tracer: function_graph 2650 # 2651 # CPU DURATION FUNCTION CALLS 2652 # | | | | | | | 2653 2654 0) | sys_open() { 2655 0) | do_sys_open() { 2656 0) | getname() { 2657 0) | kmem_cache_alloc() { 2658 0) 1.382 us | __might_sleep(); 2659 0) 2.478 us | } 2660 0) | strncpy_from_user() { 2661 0) | might_fault() { 2662 0) 1.389 us | __might_sleep(); 2663 0) 2.553 us | } 2664 0) 3.807 us | } 2665 0) 7.876 us | } 2666 0) | alloc_fd() { 2667 0) 0.668 us | _spin_lock(); 2668 0) 0.570 us | expand_files(); 2669 0) 0.586 us | _spin_unlock(); 2670 2671 2672There are several columns that can be dynamically 2673enabled/disabled. You can use every combination of options you 2674want, depending on your needs. 2675 2676- The cpu number on which the function executed is default 2677 enabled. It is sometimes better to only trace one cpu (see 2678 tracing_cpumask file) or you might sometimes see unordered 2679 function calls while cpu tracing switch. 2680 2681 - hide: echo nofuncgraph-cpu > trace_options 2682 - show: echo funcgraph-cpu > trace_options 2683 2684- The duration (function's time of execution) is displayed on 2685 the closing bracket line of a function or on the same line 2686 than the current function in case of a leaf one. It is default 2687 enabled. 2688 2689 - hide: echo nofuncgraph-duration > trace_options 2690 - show: echo funcgraph-duration > trace_options 2691 2692- The overhead field precedes the duration field in case of 2693 reached duration thresholds. 2694 2695 - hide: echo nofuncgraph-overhead > trace_options 2696 - show: echo funcgraph-overhead > trace_options 2697 - depends on: funcgraph-duration 2698 2699 ie:: 2700 2701 3) # 1837.709 us | } /* __switch_to */ 2702 3) | finish_task_switch() { 2703 3) 0.313 us | _raw_spin_unlock_irq(); 2704 3) 3.177 us | } 2705 3) # 1889.063 us | } /* __schedule */ 2706 3) ! 140.417 us | } /* __schedule */ 2707 3) # 2034.948 us | } /* schedule */ 2708 3) * 33998.59 us | } /* schedule_preempt_disabled */ 2709 2710 [...] 2711 2712 1) 0.260 us | msecs_to_jiffies(); 2713 1) 0.313 us | __rcu_read_unlock(); 2714 1) + 61.770 us | } 2715 1) + 64.479 us | } 2716 1) 0.313 us | rcu_bh_qs(); 2717 1) 0.313 us | __local_bh_enable(); 2718 1) ! 217.240 us | } 2719 1) 0.365 us | idle_cpu(); 2720 1) | rcu_irq_exit() { 2721 1) 0.417 us | rcu_eqs_enter_common.isra.47(); 2722 1) 3.125 us | } 2723 1) ! 227.812 us | } 2724 1) ! 457.395 us | } 2725 1) @ 119760.2 us | } 2726 2727 [...] 2728 2729 2) | handle_IPI() { 2730 1) 6.979 us | } 2731 2) 0.417 us | scheduler_ipi(); 2732 1) 9.791 us | } 2733 1) + 12.917 us | } 2734 2) 3.490 us | } 2735 1) + 15.729 us | } 2736 1) + 18.542 us | } 2737 2) $ 3594274 us | } 2738 2739Flags:: 2740 2741 + means that the function exceeded 10 usecs. 2742 ! means that the function exceeded 100 usecs. 2743 # means that the function exceeded 1000 usecs. 2744 * means that the function exceeded 10 msecs. 2745 @ means that the function exceeded 100 msecs. 2746 $ means that the function exceeded 1 sec. 2747 2748 2749- The task/pid field displays the thread cmdline and pid which 2750 executed the function. It is default disabled. 2751 2752 - hide: echo nofuncgraph-proc > trace_options 2753 - show: echo funcgraph-proc > trace_options 2754 2755 ie:: 2756 2757 # tracer: function_graph 2758 # 2759 # CPU TASK/PID DURATION FUNCTION CALLS 2760 # | | | | | | | | | 2761 0) sh-4802 | | d_free() { 2762 0) sh-4802 | | call_rcu() { 2763 0) sh-4802 | | __call_rcu() { 2764 0) sh-4802 | 0.616 us | rcu_process_gp_end(); 2765 0) sh-4802 | 0.586 us | check_for_new_grace_period(); 2766 0) sh-4802 | 2.899 us | } 2767 0) sh-4802 | 4.040 us | } 2768 0) sh-4802 | 5.151 us | } 2769 0) sh-4802 | + 49.370 us | } 2770 2771 2772- The absolute time field is an absolute timestamp given by the 2773 system clock since it started. A snapshot of this time is 2774 given on each entry/exit of functions 2775 2776 - hide: echo nofuncgraph-abstime > trace_options 2777 - show: echo funcgraph-abstime > trace_options 2778 2779 ie:: 2780 2781 # 2782 # TIME CPU DURATION FUNCTION CALLS 2783 # | | | | | | | | 2784 360.774522 | 1) 0.541 us | } 2785 360.774522 | 1) 4.663 us | } 2786 360.774523 | 1) 0.541 us | __wake_up_bit(); 2787 360.774524 | 1) 6.796 us | } 2788 360.774524 | 1) 7.952 us | } 2789 360.774525 | 1) 9.063 us | } 2790 360.774525 | 1) 0.615 us | journal_mark_dirty(); 2791 360.774527 | 1) 0.578 us | __brelse(); 2792 360.774528 | 1) | reiserfs_prepare_for_journal() { 2793 360.774528 | 1) | unlock_buffer() { 2794 360.774529 | 1) | wake_up_bit() { 2795 360.774529 | 1) | bit_waitqueue() { 2796 360.774530 | 1) 0.594 us | __phys_addr(); 2797 2798 2799The function name is always displayed after the closing bracket 2800for a function if the start of that function is not in the 2801trace buffer. 2802 2803Display of the function name after the closing bracket may be 2804enabled for functions whose start is in the trace buffer, 2805allowing easier searching with grep for function durations. 2806It is default disabled. 2807 2808 - hide: echo nofuncgraph-tail > trace_options 2809 - show: echo funcgraph-tail > trace_options 2810 2811 Example with nofuncgraph-tail (default):: 2812 2813 0) | putname() { 2814 0) | kmem_cache_free() { 2815 0) 0.518 us | __phys_addr(); 2816 0) 1.757 us | } 2817 0) 2.861 us | } 2818 2819 Example with funcgraph-tail:: 2820 2821 0) | putname() { 2822 0) | kmem_cache_free() { 2823 0) 0.518 us | __phys_addr(); 2824 0) 1.757 us | } /* kmem_cache_free() */ 2825 0) 2.861 us | } /* putname() */ 2826 2827The return value of each traced function can be displayed after 2828an equal sign "=". When encountering system call failures, it 2829can be very helpful to quickly locate the function that first 2830returns an error code. 2831 2832 - hide: echo nofuncgraph-retval > trace_options 2833 - show: echo funcgraph-retval > trace_options 2834 2835 Example with funcgraph-retval:: 2836 2837 1) | cgroup_migrate() { 2838 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */ 2839 1) | cgroup_migrate_execute() { 2840 1) | cpu_cgroup_can_attach() { 2841 1) | cgroup_taskset_first() { 2842 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */ 2843 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */ 2844 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */ 2845 1) 2.335 us | } /* cpu_cgroup_can_attach = -22 */ 2846 1) 4.369 us | } /* cgroup_migrate_execute = -22 */ 2847 1) 7.143 us | } /* cgroup_migrate = -22 */ 2848 2849The above example shows that the function cpu_cgroup_can_attach 2850returned the error code -22 firstly, then we can read the code 2851of this function to get the root cause. 2852 2853When the option funcgraph-retval-hex is not set, the return value can 2854be displayed in a smart way. Specifically, if it is an error code, 2855it will be printed in signed decimal format, otherwise it will 2856printed in hexadecimal format. 2857 2858 - smart: echo nofuncgraph-retval-hex > trace_options 2859 - hexadecimal: echo funcgraph-retval-hex > trace_options 2860 2861 Example with funcgraph-retval-hex:: 2862 2863 1) | cgroup_migrate() { 2864 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */ 2865 1) | cgroup_migrate_execute() { 2866 1) | cpu_cgroup_can_attach() { 2867 1) | cgroup_taskset_first() { 2868 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */ 2869 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */ 2870 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */ 2871 1) 2.335 us | } /* cpu_cgroup_can_attach = 0xffffffea */ 2872 1) 4.369 us | } /* cgroup_migrate_execute = 0xffffffea */ 2873 1) 7.143 us | } /* cgroup_migrate = 0xffffffea */ 2874 2875At present, there are some limitations when using the funcgraph-retval 2876option, and these limitations will be eliminated in the future: 2877 2878- Even if the function return type is void, a return value will still 2879 be printed, and you can just ignore it. 2880 2881- Even if return values are stored in multiple registers, only the 2882 value contained in the first register will be recorded and printed. 2883 To illustrate, in the x86 architecture, eax and edx are used to store 2884 a 64-bit return value, with the lower 32 bits saved in eax and the 2885 upper 32 bits saved in edx. However, only the value stored in eax 2886 will be recorded and printed. 2887 2888- In certain procedure call standards, such as arm64's AAPCS64, when a 2889 type is smaller than a GPR, it is the responsibility of the consumer 2890 to perform the narrowing, and the upper bits may contain UNKNOWN values. 2891 Therefore, it is advisable to check the code for such cases. For instance, 2892 when using a u8 in a 64-bit GPR, bits [63:8] may contain arbitrary values, 2893 especially when larger types are truncated, whether explicitly or implicitly. 2894 Here are some specific cases to illustrate this point: 2895 2896 **Case One**: 2897 2898 The function narrow_to_u8 is defined as follows:: 2899 2900 u8 narrow_to_u8(u64 val) 2901 { 2902 // implicitly truncated 2903 return val; 2904 } 2905 2906 It may be compiled to:: 2907 2908 narrow_to_u8: 2909 < ... ftrace instrumentation ... > 2910 RET 2911 2912 If you pass 0x123456789abcdef to this function and want to narrow it, 2913 it may be recorded as 0x123456789abcdef instead of 0xef. 2914 2915 **Case Two**: 2916 2917 The function error_if_not_4g_aligned is defined as follows:: 2918 2919 int error_if_not_4g_aligned(u64 val) 2920 { 2921 if (val & GENMASK(31, 0)) 2922 return -EINVAL; 2923 2924 return 0; 2925 } 2926 2927 It could be compiled to:: 2928 2929 error_if_not_4g_aligned: 2930 CBNZ w0, .Lnot_aligned 2931 RET // bits [31:0] are zero, bits 2932 // [63:32] are UNKNOWN 2933 .Lnot_aligned: 2934 MOV x0, #-EINVAL 2935 RET 2936 2937 When passing 0x2_0000_0000 to it, the return value may be recorded as 2938 0x2_0000_0000 instead of 0. 2939 2940You can put some comments on specific functions by using 2941trace_printk() For example, if you want to put a comment inside 2942the __might_sleep() function, you just have to include 2943<linux/ftrace.h> and call trace_printk() inside __might_sleep():: 2944 2945 trace_printk("I'm a comment!\n") 2946 2947will produce:: 2948 2949 1) | __might_sleep() { 2950 1) | /* I'm a comment! */ 2951 1) 1.449 us | } 2952 2953 2954You might find other useful features for this tracer in the 2955following "dynamic ftrace" section such as tracing only specific 2956functions or tasks. 2957 2958dynamic ftrace 2959-------------- 2960 2961If CONFIG_DYNAMIC_FTRACE is set, the system will run with 2962virtually no overhead when function tracing is disabled. The way 2963this works is the mcount function call (placed at the start of 2964every kernel function, produced by the -pg switch in gcc), 2965starts of pointing to a simple return. (Enabling FTRACE will 2966include the -pg switch in the compiling of the kernel.) 2967 2968At compile time every C file object is run through the 2969recordmcount program (located in the scripts directory). This 2970program will parse the ELF headers in the C object to find all 2971the locations in the .text section that call mcount. Starting 2972with gcc version 4.6, the -mfentry has been added for x86, which 2973calls "__fentry__" instead of "mcount". Which is called before 2974the creation of the stack frame. 2975 2976Note, not all sections are traced. They may be prevented by either 2977a notrace, or blocked another way and all inline functions are not 2978traced. Check the "available_filter_functions" file to see what functions 2979can be traced. 2980 2981A section called "__mcount_loc" is created that holds 2982references to all the mcount/fentry call sites in the .text section. 2983The recordmcount program re-links this section back into the 2984original object. The final linking stage of the kernel will add all these 2985references into a single table. 2986 2987On boot up, before SMP is initialized, the dynamic ftrace code 2988scans this table and updates all the locations into nops. It 2989also records the locations, which are added to the 2990available_filter_functions list. Modules are processed as they 2991are loaded and before they are executed. When a module is 2992unloaded, it also removes its functions from the ftrace function 2993list. This is automatic in the module unload code, and the 2994module author does not need to worry about it. 2995 2996When tracing is enabled, the process of modifying the function 2997tracepoints is dependent on architecture. The old method is to use 2998kstop_machine to prevent races with the CPUs executing code being 2999modified (which can cause the CPU to do undesirable things, especially 3000if the modified code crosses cache (or page) boundaries), and the nops are 3001patched back to calls. But this time, they do not call mcount 3002(which is just a function stub). They now call into the ftrace 3003infrastructure. 3004 3005The new method of modifying the function tracepoints is to place 3006a breakpoint at the location to be modified, sync all CPUs, modify 3007the rest of the instruction not covered by the breakpoint. Sync 3008all CPUs again, and then remove the breakpoint with the finished 3009version to the ftrace call site. 3010 3011Some archs do not even need to monkey around with the synchronization, 3012and can just slap the new code on top of the old without any 3013problems with other CPUs executing it at the same time. 3014 3015One special side-effect to the recording of the functions being 3016traced is that we can now selectively choose which functions we 3017wish to trace and which ones we want the mcount calls to remain 3018as nops. 3019 3020Two files are used, one for enabling and one for disabling the 3021tracing of specified functions. They are: 3022 3023 set_ftrace_filter 3024 3025and 3026 3027 set_ftrace_notrace 3028 3029A list of available functions that you can add to these files is 3030listed in: 3031 3032 available_filter_functions 3033 3034:: 3035 3036 # cat available_filter_functions 3037 put_prev_task_idle 3038 kmem_cache_create 3039 pick_next_task_rt 3040 cpus_read_lock 3041 pick_next_task_fair 3042 mutex_lock 3043 [...] 3044 3045If I am only interested in sys_nanosleep and hrtimer_interrupt:: 3046 3047 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter 3048 # echo function > current_tracer 3049 # echo 1 > tracing_on 3050 # usleep 1 3051 # echo 0 > tracing_on 3052 # cat trace 3053 # tracer: function 3054 # 3055 # entries-in-buffer/entries-written: 5/5 #P:4 3056 # 3057 # _-----=> irqs-off 3058 # / _----=> need-resched 3059 # | / _---=> hardirq/softirq 3060 # || / _--=> preempt-depth 3061 # ||| / delay 3062 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3063 # | | | |||| | | 3064 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath 3065 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt 3066 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt 3067 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt 3068 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt 3069 3070To see which functions are being traced, you can cat the file: 3071:: 3072 3073 # cat set_ftrace_filter 3074 hrtimer_interrupt 3075 sys_nanosleep 3076 3077 3078Perhaps this is not enough. The filters also allow glob(7) matching. 3079 3080 ``<match>*`` 3081 will match functions that begin with <match> 3082 ``*<match>`` 3083 will match functions that end with <match> 3084 ``*<match>*`` 3085 will match functions that have <match> in it 3086 ``<match1>*<match2>`` 3087 will match functions that begin with <match1> and end with <match2> 3088 3089.. note:: 3090 It is better to use quotes to enclose the wild cards, 3091 otherwise the shell may expand the parameters into names 3092 of files in the local directory. 3093 3094:: 3095 3096 # echo 'hrtimer_*' > set_ftrace_filter 3097 3098Produces:: 3099 3100 # tracer: function 3101 # 3102 # entries-in-buffer/entries-written: 897/897 #P:4 3103 # 3104 # _-----=> irqs-off 3105 # / _----=> need-resched 3106 # | / _---=> hardirq/softirq 3107 # || / _--=> preempt-depth 3108 # ||| / delay 3109 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3110 # | | | |||| | | 3111 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit 3112 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel 3113 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer 3114 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit 3115 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11 3116 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt 3117 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter 3118 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem 3119 3120Notice that we lost the sys_nanosleep. 3121:: 3122 3123 # cat set_ftrace_filter 3124 hrtimer_run_queues 3125 hrtimer_run_pending 3126 hrtimer_setup 3127 hrtimer_cancel 3128 hrtimer_try_to_cancel 3129 hrtimer_forward 3130 hrtimer_start 3131 hrtimer_reprogram 3132 hrtimer_force_reprogram 3133 hrtimer_get_next_event 3134 hrtimer_interrupt 3135 hrtimer_nanosleep 3136 hrtimer_wakeup 3137 hrtimer_get_remaining 3138 hrtimer_get_res 3139 hrtimer_init_sleeper 3140 3141 3142This is because the '>' and '>>' act just like they do in bash. 3143To rewrite the filters, use '>' 3144To append to the filters, use '>>' 3145 3146To clear out a filter so that all functions will be recorded 3147again:: 3148 3149 # echo > set_ftrace_filter 3150 # cat set_ftrace_filter 3151 # 3152 3153Again, now we want to append. 3154 3155:: 3156 3157 # echo sys_nanosleep > set_ftrace_filter 3158 # cat set_ftrace_filter 3159 sys_nanosleep 3160 # echo 'hrtimer_*' >> set_ftrace_filter 3161 # cat set_ftrace_filter 3162 hrtimer_run_queues 3163 hrtimer_run_pending 3164 hrtimer_setup 3165 hrtimer_cancel 3166 hrtimer_try_to_cancel 3167 hrtimer_forward 3168 hrtimer_start 3169 hrtimer_reprogram 3170 hrtimer_force_reprogram 3171 hrtimer_get_next_event 3172 hrtimer_interrupt 3173 sys_nanosleep 3174 hrtimer_nanosleep 3175 hrtimer_wakeup 3176 hrtimer_get_remaining 3177 hrtimer_get_res 3178 hrtimer_init_sleeper 3179 3180 3181The set_ftrace_notrace prevents those functions from being 3182traced. 3183:: 3184 3185 # echo '*preempt*' '*lock*' > set_ftrace_notrace 3186 3187Produces:: 3188 3189 # tracer: function 3190 # 3191 # entries-in-buffer/entries-written: 39608/39608 #P:4 3192 # 3193 # _-----=> irqs-off 3194 # / _----=> need-resched 3195 # | / _---=> hardirq/softirq 3196 # || / _--=> preempt-depth 3197 # ||| / delay 3198 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3199 # | | | |||| | | 3200 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open 3201 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last 3202 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last 3203 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check 3204 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement 3205 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action 3206 bash-1994 [000] .... 4342.324899: do_truncate <-do_last 3207 bash-1994 [000] .... 4342.324899: setattr_should_drop_suidgid <-do_truncate 3208 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate 3209 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change 3210 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time 3211 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time 3212 3213We can see that there's no more lock or preempt tracing. 3214 3215Selecting function filters via index 3216------------------------------------ 3217 3218Because processing of strings is expensive (the address of the function 3219needs to be looked up before comparing to the string being passed in), 3220an index can be used as well to enable functions. This is useful in the 3221case of setting thousands of specific functions at a time. By passing 3222in a list of numbers, no string processing will occur. Instead, the function 3223at the specific location in the internal array (which corresponds to the 3224functions in the "available_filter_functions" file), is selected. 3225 3226:: 3227 3228 # echo 1 > set_ftrace_filter 3229 3230Will select the first function listed in "available_filter_functions" 3231 3232:: 3233 3234 # head -1 available_filter_functions 3235 trace_initcall_finish_cb 3236 3237 # cat set_ftrace_filter 3238 trace_initcall_finish_cb 3239 3240 # head -50 available_filter_functions | tail -1 3241 x86_pmu_commit_txn 3242 3243 # echo 1 50 > set_ftrace_filter 3244 # cat set_ftrace_filter 3245 trace_initcall_finish_cb 3246 x86_pmu_commit_txn 3247 3248Dynamic ftrace with the function graph tracer 3249--------------------------------------------- 3250 3251Although what has been explained above concerns both the 3252function tracer and the function-graph-tracer, there are some 3253special features only available in the function-graph tracer. 3254 3255If you want to trace only one function and all of its children, 3256you just have to echo its name into set_graph_function:: 3257 3258 echo __do_fault > set_graph_function 3259 3260will produce the following "expanded" trace of the __do_fault() 3261function:: 3262 3263 0) | __do_fault() { 3264 0) | filemap_fault() { 3265 0) | find_lock_page() { 3266 0) 0.804 us | find_get_page(); 3267 0) | __might_sleep() { 3268 0) 1.329 us | } 3269 0) 3.904 us | } 3270 0) 4.979 us | } 3271 0) 0.653 us | _spin_lock(); 3272 0) 0.578 us | page_add_file_rmap(); 3273 0) 0.525 us | native_set_pte_at(); 3274 0) 0.585 us | _spin_unlock(); 3275 0) | unlock_page() { 3276 0) 0.541 us | page_waitqueue(); 3277 0) 0.639 us | __wake_up_bit(); 3278 0) 2.786 us | } 3279 0) + 14.237 us | } 3280 0) | __do_fault() { 3281 0) | filemap_fault() { 3282 0) | find_lock_page() { 3283 0) 0.698 us | find_get_page(); 3284 0) | __might_sleep() { 3285 0) 1.412 us | } 3286 0) 3.950 us | } 3287 0) 5.098 us | } 3288 0) 0.631 us | _spin_lock(); 3289 0) 0.571 us | page_add_file_rmap(); 3290 0) 0.526 us | native_set_pte_at(); 3291 0) 0.586 us | _spin_unlock(); 3292 0) | unlock_page() { 3293 0) 0.533 us | page_waitqueue(); 3294 0) 0.638 us | __wake_up_bit(); 3295 0) 2.793 us | } 3296 0) + 14.012 us | } 3297 3298You can also expand several functions at once:: 3299 3300 echo sys_open > set_graph_function 3301 echo sys_close >> set_graph_function 3302 3303Now if you want to go back to trace all functions you can clear 3304this special filter via:: 3305 3306 echo > set_graph_function 3307 3308 3309ftrace_enabled 3310-------------- 3311 3312Note, the proc sysctl ftrace_enable is a big on/off switch for the 3313function tracer. By default it is enabled (when function tracing is 3314enabled in the kernel). If it is disabled, all function tracing is 3315disabled. This includes not only the function tracers for ftrace, but 3316also for any other uses (perf, kprobes, stack tracing, profiling, etc). It 3317cannot be disabled if there is a callback with FTRACE_OPS_FL_PERMANENT set 3318registered. 3319 3320Please disable this with care. 3321 3322This can be disable (and enabled) with:: 3323 3324 sysctl kernel.ftrace_enabled=0 3325 sysctl kernel.ftrace_enabled=1 3326 3327 or 3328 3329 echo 0 > /proc/sys/kernel/ftrace_enabled 3330 echo 1 > /proc/sys/kernel/ftrace_enabled 3331 3332 3333Filter commands 3334--------------- 3335 3336A few commands are supported by the set_ftrace_filter interface. 3337Trace commands have the following format:: 3338 3339 <function>:<command>:<parameter> 3340 3341The following commands are supported: 3342 3343- mod: 3344 This command enables function filtering per module. The 3345 parameter defines the module. For example, if only the write* 3346 functions in the ext3 module are desired, run: 3347 3348 echo 'write*:mod:ext3' > set_ftrace_filter 3349 3350 This command interacts with the filter in the same way as 3351 filtering based on function names. Thus, adding more functions 3352 in a different module is accomplished by appending (>>) to the 3353 filter file. Remove specific module functions by prepending 3354 '!':: 3355 3356 echo '!writeback*:mod:ext3' >> set_ftrace_filter 3357 3358 Mod command supports module globbing. Disable tracing for all 3359 functions except a specific module:: 3360 3361 echo '!*:mod:!ext3' >> set_ftrace_filter 3362 3363 Disable tracing for all modules, but still trace kernel:: 3364 3365 echo '!*:mod:*' >> set_ftrace_filter 3366 3367 Enable filter only for kernel:: 3368 3369 echo '*write*:mod:!*' >> set_ftrace_filter 3370 3371 Enable filter for module globbing:: 3372 3373 echo '*write*:mod:*snd*' >> set_ftrace_filter 3374 3375- traceon/traceoff: 3376 These commands turn tracing on and off when the specified 3377 functions are hit. The parameter determines how many times the 3378 tracing system is turned on and off. If unspecified, there is 3379 no limit. For example, to disable tracing when a schedule bug 3380 is hit the first 5 times, run:: 3381 3382 echo '__schedule_bug:traceoff:5' > set_ftrace_filter 3383 3384 To always disable tracing when __schedule_bug is hit:: 3385 3386 echo '__schedule_bug:traceoff' > set_ftrace_filter 3387 3388 These commands are cumulative whether or not they are appended 3389 to set_ftrace_filter. To remove a command, prepend it by '!' 3390 and drop the parameter:: 3391 3392 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter 3393 3394 The above removes the traceoff command for __schedule_bug 3395 that have a counter. To remove commands without counters:: 3396 3397 echo '!__schedule_bug:traceoff' > set_ftrace_filter 3398 3399- snapshot: 3400 Will cause a snapshot to be triggered when the function is hit. 3401 :: 3402 3403 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter 3404 3405 To only snapshot once: 3406 :: 3407 3408 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter 3409 3410 To remove the above commands:: 3411 3412 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter 3413 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter 3414 3415- enable_event/disable_event: 3416 These commands can enable or disable a trace event. Note, because 3417 function tracing callbacks are very sensitive, when these commands 3418 are registered, the trace point is activated, but disabled in 3419 a "soft" mode. That is, the tracepoint will be called, but 3420 just will not be traced. The event tracepoint stays in this mode 3421 as long as there's a command that triggers it. 3422 :: 3423 3424 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \ 3425 set_ftrace_filter 3426 3427 The format is:: 3428 3429 <function>:enable_event:<system>:<event>[:count] 3430 <function>:disable_event:<system>:<event>[:count] 3431 3432 To remove the events commands:: 3433 3434 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \ 3435 set_ftrace_filter 3436 echo '!schedule:disable_event:sched:sched_switch' > \ 3437 set_ftrace_filter 3438 3439- dump: 3440 When the function is hit, it will dump the contents of the ftrace 3441 ring buffer to the console. This is useful if you need to debug 3442 something, and want to dump the trace when a certain function 3443 is hit. Perhaps it's a function that is called before a triple 3444 fault happens and does not allow you to get a regular dump. 3445 3446- cpudump: 3447 When the function is hit, it will dump the contents of the ftrace 3448 ring buffer for the current CPU to the console. Unlike the "dump" 3449 command, it only prints out the contents of the ring buffer for the 3450 CPU that executed the function that triggered the dump. 3451 3452- stacktrace: 3453 When the function is hit, a stack trace is recorded. 3454 3455trace_pipe 3456---------- 3457 3458The trace_pipe outputs the same content as the trace file, but 3459the effect on the tracing is different. Every read from 3460trace_pipe is consumed. This means that subsequent reads will be 3461different. The trace is live. 3462:: 3463 3464 # echo function > current_tracer 3465 # cat trace_pipe > /tmp/trace.out & 3466 [1] 4153 3467 # echo 1 > tracing_on 3468 # usleep 1 3469 # echo 0 > tracing_on 3470 # cat trace 3471 # tracer: function 3472 # 3473 # entries-in-buffer/entries-written: 0/0 #P:4 3474 # 3475 # _-----=> irqs-off 3476 # / _----=> need-resched 3477 # | / _---=> hardirq/softirq 3478 # || / _--=> preempt-depth 3479 # ||| / delay 3480 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3481 # | | | |||| | | 3482 3483 # 3484 # cat /tmp/trace.out 3485 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write 3486 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock 3487 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify 3488 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify 3489 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify 3490 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock 3491 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock 3492 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify 3493 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath 3494 3495 3496Note, reading the trace_pipe file will block until more input is 3497added. This is contrary to the trace file. If any process opened 3498the trace file for reading, it will actually disable tracing and 3499prevent new entries from being added. The trace_pipe file does 3500not have this limitation. 3501 3502trace entries 3503------------- 3504 3505Having too much or not enough data can be troublesome in 3506diagnosing an issue in the kernel. The file buffer_size_kb is 3507used to modify the size of the internal trace buffers. The 3508number listed is the number of entries that can be recorded per 3509CPU. To know the full size, multiply the number of possible CPUs 3510with the number of entries. 3511:: 3512 3513 # cat buffer_size_kb 3514 1408 (units kilobytes) 3515 3516Or simply read buffer_total_size_kb 3517:: 3518 3519 # cat buffer_total_size_kb 3520 5632 3521 3522To modify the buffer, simple echo in a number (in 1024 byte segments). 3523:: 3524 3525 # echo 10000 > buffer_size_kb 3526 # cat buffer_size_kb 3527 10000 (units kilobytes) 3528 3529It will try to allocate as much as possible. If you allocate too 3530much, it can cause Out-Of-Memory to trigger. 3531:: 3532 3533 # echo 1000000000000 > buffer_size_kb 3534 -bash: echo: write error: Cannot allocate memory 3535 # cat buffer_size_kb 3536 85 3537 3538The per_cpu buffers can be changed individually as well: 3539:: 3540 3541 # echo 10000 > per_cpu/cpu0/buffer_size_kb 3542 # echo 100 > per_cpu/cpu1/buffer_size_kb 3543 3544When the per_cpu buffers are not the same, the buffer_size_kb 3545at the top level will just show an X 3546:: 3547 3548 # cat buffer_size_kb 3549 X 3550 3551This is where the buffer_total_size_kb is useful: 3552:: 3553 3554 # cat buffer_total_size_kb 3555 12916 3556 3557Writing to the top level buffer_size_kb will reset all the buffers 3558to be the same again. 3559 3560Snapshot 3561-------- 3562CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature 3563available to all non latency tracers. (Latency tracers which 3564record max latency, such as "irqsoff" or "wakeup", can't use 3565this feature, since those are already using the snapshot 3566mechanism internally.) 3567 3568Snapshot preserves a current trace buffer at a particular point 3569in time without stopping tracing. Ftrace swaps the current 3570buffer with a spare buffer, and tracing continues in the new 3571current (=previous spare) buffer. 3572 3573The following tracefs files in "tracing" are related to this 3574feature: 3575 3576 snapshot: 3577 3578 This is used to take a snapshot and to read the output 3579 of the snapshot. Echo 1 into this file to allocate a 3580 spare buffer and to take a snapshot (swap), then read 3581 the snapshot from this file in the same format as 3582 "trace" (described above in the section "The File 3583 System"). Both reads snapshot and tracing are executable 3584 in parallel. When the spare buffer is allocated, echoing 3585 0 frees it, and echoing else (positive) values clear the 3586 snapshot contents. 3587 More details are shown in the table below. 3588 3589 +--------------+------------+------------+------------+ 3590 |status\\input | 0 | 1 | else | 3591 +==============+============+============+============+ 3592 |not allocated |(do nothing)| alloc+swap |(do nothing)| 3593 +--------------+------------+------------+------------+ 3594 |allocated | free | swap | clear | 3595 +--------------+------------+------------+------------+ 3596 3597Here is an example of using the snapshot feature. 3598:: 3599 3600 # echo 1 > events/sched/enable 3601 # echo 1 > snapshot 3602 # cat snapshot 3603 # tracer: nop 3604 # 3605 # entries-in-buffer/entries-written: 71/71 #P:8 3606 # 3607 # _-----=> irqs-off 3608 # / _----=> need-resched 3609 # | / _---=> hardirq/softirq 3610 # || / _--=> preempt-depth 3611 # ||| / delay 3612 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3613 # | | | |||| | | 3614 <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 3615 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 3616 [...] 3617 <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 3618 3619 # cat trace 3620 # tracer: nop 3621 # 3622 # entries-in-buffer/entries-written: 77/77 #P:8 3623 # 3624 # _-----=> irqs-off 3625 # / _----=> need-resched 3626 # | / _---=> hardirq/softirq 3627 # || / _--=> preempt-depth 3628 # ||| / delay 3629 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3630 # | | | |||| | | 3631 <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 3632 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 3633 [...] 3634 3635 3636If you try to use this snapshot feature when current tracer is 3637one of the latency tracers, you will get the following results. 3638:: 3639 3640 # echo wakeup > current_tracer 3641 # echo 1 > snapshot 3642 bash: echo: write error: Device or resource busy 3643 # cat snapshot 3644 cat: snapshot: Device or resource busy 3645 3646 3647Instances 3648--------- 3649In the tracefs tracing directory, there is a directory called "instances". 3650This directory can have new directories created inside of it using 3651mkdir, and removing directories with rmdir. The directory created 3652with mkdir in this directory will already contain files and other 3653directories after it is created. 3654:: 3655 3656 # mkdir instances/foo 3657 # ls instances/foo 3658 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu 3659 set_event snapshot trace trace_clock trace_marker trace_options 3660 trace_pipe tracing_on 3661 3662As you can see, the new directory looks similar to the tracing directory 3663itself. In fact, it is very similar, except that the buffer and 3664events are agnostic from the main directory, or from any other 3665instances that are created. 3666 3667The files in the new directory work just like the files with the 3668same name in the tracing directory except the buffer that is used 3669is a separate and new buffer. The files affect that buffer but do not 3670affect the main buffer with the exception of trace_options. Currently, 3671the trace_options affect all instances and the top level buffer 3672the same, but this may change in future releases. That is, options 3673may become specific to the instance they reside in. 3674 3675Notice that none of the function tracer files are there, nor is 3676current_tracer and available_tracers. This is because the buffers 3677can currently only have events enabled for them. 3678:: 3679 3680 # mkdir instances/foo 3681 # mkdir instances/bar 3682 # mkdir instances/zoot 3683 # echo 100000 > buffer_size_kb 3684 # echo 1000 > instances/foo/buffer_size_kb 3685 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb 3686 # echo function > current_trace 3687 # echo 1 > instances/foo/events/sched/sched_wakeup/enable 3688 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable 3689 # echo 1 > instances/foo/events/sched/sched_switch/enable 3690 # echo 1 > instances/bar/events/irq/enable 3691 # echo 1 > instances/zoot/events/syscalls/enable 3692 # cat trace_pipe 3693 CPU:2 [LOST 11745 EVENTS] 3694 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist 3695 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave 3696 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist 3697 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist 3698 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock 3699 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype 3700 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist 3701 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist 3702 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics 3703 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics 3704 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process 3705 [...] 3706 3707 # cat instances/foo/trace_pipe 3708 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000 3709 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000 3710 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003 3711 <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 3712 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 3713 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000 3714 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000 3715 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 3716 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001 3717 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 3718 [...] 3719 3720 # cat instances/bar/trace_pipe 3721 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX] 3722 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX] 3723 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER] 3724 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU] 3725 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER] 3726 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER] 3727 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU] 3728 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU] 3729 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4 3730 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled 3731 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0 3732 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled 3733 [...] 3734 3735 # cat instances/zoot/trace 3736 # tracer: nop 3737 # 3738 # entries-in-buffer/entries-written: 18996/18996 #P:4 3739 # 3740 # _-----=> irqs-off 3741 # / _----=> need-resched 3742 # | / _---=> hardirq/softirq 3743 # || / _--=> preempt-depth 3744 # ||| / delay 3745 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3746 # | | | |||| | | 3747 bash-1998 [000] d... 140.733501: sys_write -> 0x2 3748 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1) 3749 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1 3750 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0) 3751 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1 3752 bash-1998 [000] d... 140.733510: sys_close(fd: a) 3753 bash-1998 [000] d... 140.733510: sys_close -> 0x0 3754 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8) 3755 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0 3756 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8) 3757 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0 3758 3759You can see that the trace of the top most trace buffer shows only 3760the function tracing. The foo instance displays wakeups and task 3761switches. 3762 3763To remove the instances, simply delete their directories: 3764:: 3765 3766 # rmdir instances/foo 3767 # rmdir instances/bar 3768 # rmdir instances/zoot 3769 3770Note, if a process has a trace file open in one of the instance 3771directories, the rmdir will fail with EBUSY. 3772 3773 3774Stack trace 3775----------- 3776Since the kernel has a fixed sized stack, it is important not to 3777waste it in functions. A kernel developer must be conscious of 3778what they allocate on the stack. If they add too much, the system 3779can be in danger of a stack overflow, and corruption will occur, 3780usually leading to a system panic. 3781 3782There are some tools that check this, usually with interrupts 3783periodically checking usage. But if you can perform a check 3784at every function call that will become very useful. As ftrace provides 3785a function tracer, it makes it convenient to check the stack size 3786at every function call. This is enabled via the stack tracer. 3787 3788CONFIG_STACK_TRACER enables the ftrace stack tracing functionality. 3789To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled. 3790:: 3791 3792 # echo 1 > /proc/sys/kernel/stack_tracer_enabled 3793 3794You can also enable it from the kernel command line to trace 3795the stack size of the kernel during boot up, by adding "stacktrace" 3796to the kernel command line parameter. 3797 3798After running it for a few minutes, the output looks like: 3799:: 3800 3801 # cat stack_max_size 3802 2928 3803 3804 # cat stack_trace 3805 Depth Size Location (18 entries) 3806 ----- ---- -------- 3807 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac 3808 1) 2704 160 find_busiest_group+0x31/0x1f1 3809 2) 2544 256 load_balance+0xd9/0x662 3810 3) 2288 80 idle_balance+0xbb/0x130 3811 4) 2208 128 __schedule+0x26e/0x5b9 3812 5) 2080 16 schedule+0x64/0x66 3813 6) 2064 128 schedule_timeout+0x34/0xe0 3814 7) 1936 112 wait_for_common+0x97/0xf1 3815 8) 1824 16 wait_for_completion+0x1d/0x1f 3816 9) 1808 128 flush_work+0xfe/0x119 3817 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20 3818 11) 1664 48 input_available_p+0x1d/0x5c 3819 12) 1616 48 n_tty_poll+0x6d/0x134 3820 13) 1568 64 tty_poll+0x64/0x7f 3821 14) 1504 880 do_select+0x31e/0x511 3822 15) 624 400 core_sys_select+0x177/0x216 3823 16) 224 96 sys_select+0x91/0xb9 3824 17) 128 128 system_call_fastpath+0x16/0x1b 3825 3826Note, if -mfentry is being used by gcc, functions get traced before 3827they set up the stack frame. This means that leaf level functions 3828are not tested by the stack tracer when -mfentry is used. 3829 3830Currently, -mfentry is used by gcc 4.6.0 and above on x86 only. 3831 3832More 3833---- 3834More details can be found in the source code, in the `kernel/trace/*.c` files. 3835