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 available_events: 688 689 A list of events that can be enabled in tracing. 690 691 See events.rst for more information. 692 693 timestamp_mode: 694 695 Certain tracers may change the timestamp mode used when 696 logging trace events into the event buffer. Events with 697 different modes can coexist within a buffer but the mode in 698 effect when an event is logged determines which timestamp mode 699 is used for that event. The default timestamp mode is 700 'delta'. 701 702 Usual timestamp modes for tracing: 703 704 # cat timestamp_mode 705 [delta] absolute 706 707 The timestamp mode with the square brackets around it is the 708 one in effect. 709 710 delta: Default timestamp mode - timestamp is a delta against 711 a per-buffer timestamp. 712 713 absolute: The timestamp is a full timestamp, not a delta 714 against some other value. As such it takes up more 715 space and is less efficient. 716 717 hwlat_detector: 718 719 Directory for the Hardware Latency Detector. 720 See "Hardware Latency Detector" section below. 721 722 per_cpu: 723 724 This is a directory that contains the trace per_cpu information. 725 726 per_cpu/cpu0/buffer_size_kb: 727 728 The ftrace buffer is defined per_cpu. That is, there's a separate 729 buffer for each CPU to allow writes to be done atomically, 730 and free from cache bouncing. These buffers may have different 731 size buffers. This file is similar to the buffer_size_kb 732 file, but it only displays or sets the buffer size for the 733 specific CPU. (here cpu0). 734 735 per_cpu/cpu0/trace: 736 737 This is similar to the "trace" file, but it will only display 738 the data specific for the CPU. If written to, it only clears 739 the specific CPU buffer. 740 741 per_cpu/cpu0/trace_pipe 742 743 This is similar to the "trace_pipe" file, and is a consuming 744 read, but it will only display (and consume) the data specific 745 for the CPU. 746 747 per_cpu/cpu0/trace_pipe_raw 748 749 For tools that can parse the ftrace ring buffer binary format, 750 the trace_pipe_raw file can be used to extract the data 751 from the ring buffer directly. With the use of the splice() 752 system call, the buffer data can be quickly transferred to 753 a file or to the network where a server is collecting the 754 data. 755 756 Like trace_pipe, this is a consuming reader, where multiple 757 reads will always produce different data. 758 759 per_cpu/cpu0/snapshot: 760 761 This is similar to the main "snapshot" file, but will only 762 snapshot the current CPU (if supported). It only displays 763 the content of the snapshot for a given CPU, and if 764 written to, only clears this CPU buffer. 765 766 per_cpu/cpu0/snapshot_raw: 767 768 Similar to the trace_pipe_raw, but will read the binary format 769 from the snapshot buffer for the given CPU. 770 771 per_cpu/cpu0/stats: 772 773 This displays certain stats about the ring buffer: 774 775 entries: 776 The number of events that are still in the buffer. 777 778 overrun: 779 The number of lost events due to overwriting when 780 the buffer was full. 781 782 commit overrun: 783 Should always be zero. 784 This gets set if so many events happened within a nested 785 event (ring buffer is re-entrant), that it fills the 786 buffer and starts dropping events. 787 788 bytes: 789 Bytes actually read (not overwritten). 790 791 oldest event ts: 792 The oldest timestamp in the buffer 793 794 now ts: 795 The current timestamp 796 797 dropped events: 798 Events lost due to overwrite option being off. 799 800 read events: 801 The number of events read. 802 803The Tracers 804----------- 805 806Here is the list of current tracers that may be configured. 807 808 "function" 809 810 Function call tracer to trace all kernel functions. 811 812 "function_graph" 813 814 Similar to the function tracer except that the 815 function tracer probes the functions on their entry 816 whereas the function graph tracer traces on both entry 817 and exit of the functions. It then provides the ability 818 to draw a graph of function calls similar to C code 819 source. 820 821 Note that the function graph calculates the timings of when the 822 function starts and returns internally and for each instance. If 823 there are two instances that run function graph tracer and traces 824 the same functions, the length of the timings may be slightly off as 825 each read the timestamp separately and not at the same time. 826 827 "blk" 828 829 The block tracer. The tracer used by the blktrace user 830 application. 831 832 "hwlat" 833 834 The Hardware Latency tracer is used to detect if the hardware 835 produces any latency. See "Hardware Latency Detector" section 836 below. 837 838 "irqsoff" 839 840 Traces the areas that disable interrupts and saves 841 the trace with the longest max latency. 842 See tracing_max_latency. When a new max is recorded, 843 it replaces the old trace. It is best to view this 844 trace with the latency-format option enabled, which 845 happens automatically when the tracer is selected. 846 847 "preemptoff" 848 849 Similar to irqsoff but traces and records the amount of 850 time for which preemption is disabled. 851 852 "preemptirqsoff" 853 854 Similar to irqsoff and preemptoff, but traces and 855 records the largest time for which irqs and/or preemption 856 is disabled. 857 858 "wakeup" 859 860 Traces and records the max latency that it takes for 861 the highest priority task to get scheduled after 862 it has been woken up. 863 Traces all tasks as an average developer would expect. 864 865 "wakeup_rt" 866 867 Traces and records the max latency that it takes for just 868 RT tasks (as the current "wakeup" does). This is useful 869 for those interested in wake up timings of RT tasks. 870 871 "wakeup_dl" 872 873 Traces and records the max latency that it takes for 874 a SCHED_DEADLINE task to be woken (as the "wakeup" and 875 "wakeup_rt" does). 876 877 "mmiotrace" 878 879 A special tracer that is used to trace binary modules. 880 It will trace all the calls that a module makes to the 881 hardware. Everything it writes and reads from the I/O 882 as well. 883 884 "branch" 885 886 This tracer can be configured when tracing likely/unlikely 887 calls within the kernel. It will trace when a likely and 888 unlikely branch is hit and if it was correct in its prediction 889 of being correct. 890 891 "nop" 892 893 This is the "trace nothing" tracer. To remove all 894 tracers from tracing simply echo "nop" into 895 current_tracer. 896 897Error conditions 898---------------- 899 900 For most ftrace commands, failure modes are obvious and communicated 901 using standard return codes. 902 903 For other more involved commands, extended error information may be 904 available via the tracing/error_log file. For the commands that 905 support it, reading the tracing/error_log file after an error will 906 display more detailed information about what went wrong, if 907 information is available. The tracing/error_log file is a circular 908 error log displaying a small number (currently, 8) of ftrace errors 909 for the last (8) failed commands. 910 911 The extended error information and usage takes the form shown in 912 this example:: 913 914 # echo xxx > /sys/kernel/tracing/events/sched/sched_wakeup/trigger 915 echo: write error: Invalid argument 916 917 # cat /sys/kernel/tracing/error_log 918 [ 5348.887237] location: error: Couldn't yyy: zzz 919 Command: xxx 920 ^ 921 [ 7517.023364] location: error: Bad rrr: sss 922 Command: ppp qqq 923 ^ 924 925 To clear the error log, echo the empty string into it:: 926 927 # echo > /sys/kernel/tracing/error_log 928 929Examples of using the tracer 930---------------------------- 931 932Here are typical examples of using the tracers when controlling 933them only with the tracefs interface (without using any 934user-land utilities). 935 936Output format: 937-------------- 938 939Here is an example of the output format of the file "trace":: 940 941 # tracer: function 942 # 943 # entries-in-buffer/entries-written: 140080/250280 #P:4 944 # 945 # _-----=> irqs-off 946 # / _----=> need-resched 947 # | / _---=> hardirq/softirq 948 # || / _--=> preempt-depth 949 # ||| / delay 950 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 951 # | | | |||| | | 952 bash-1977 [000] .... 17284.993652: sys_close <-system_call_fastpath 953 bash-1977 [000] .... 17284.993653: __close_fd <-sys_close 954 bash-1977 [000] .... 17284.993653: _raw_spin_lock <-__close_fd 955 sshd-1974 [003] .... 17284.993653: __srcu_read_unlock <-fsnotify 956 bash-1977 [000] .... 17284.993654: add_preempt_count <-_raw_spin_lock 957 bash-1977 [000] ...1 17284.993655: _raw_spin_unlock <-__close_fd 958 bash-1977 [000] ...1 17284.993656: sub_preempt_count <-_raw_spin_unlock 959 bash-1977 [000] .... 17284.993657: filp_close <-__close_fd 960 bash-1977 [000] .... 17284.993657: dnotify_flush <-filp_close 961 sshd-1974 [003] .... 17284.993658: sys_select <-system_call_fastpath 962 .... 963 964A header is printed with the tracer name that is represented by 965the trace. In this case the tracer is "function". Then it shows the 966number of events in the buffer as well as the total number of entries 967that were written. The difference is the number of entries that were 968lost due to the buffer filling up (250280 - 140080 = 110200 events 969lost). 970 971The header explains the content of the events. Task name "bash", the task 972PID "1977", the CPU that it was running on "000", the latency format 973(explained below), the timestamp in <secs>.<usecs> format, the 974function name that was traced "sys_close" and the parent function that 975called this function "system_call_fastpath". The timestamp is the time 976at which the function was entered. 977 978Latency trace format 979-------------------- 980 981When the latency-format option is enabled or when one of the latency 982tracers is set, the trace file gives somewhat more information to see 983why a latency happened. Here is a typical trace:: 984 985 # tracer: irqsoff 986 # 987 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 988 # -------------------------------------------------------------------- 989 # latency: 259 us, #4/4, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 990 # ----------------- 991 # | task: ps-6143 (uid:0 nice:0 policy:0 rt_prio:0) 992 # ----------------- 993 # => started at: __lock_task_sighand 994 # => ended at: _raw_spin_unlock_irqrestore 995 # 996 # 997 # _------=> CPU# 998 # / _-----=> irqs-off 999 # | / _----=> need-resched 1000 # || / _---=> hardirq/softirq 1001 # ||| / _--=> preempt-depth 1002 # |||| / delay 1003 # cmd pid ||||| time | caller 1004 # \ / ||||| \ | / 1005 ps-6143 2d... 0us!: trace_hardirqs_off <-__lock_task_sighand 1006 ps-6143 2d..1 259us+: trace_hardirqs_on <-_raw_spin_unlock_irqrestore 1007 ps-6143 2d..1 263us+: time_hardirqs_on <-_raw_spin_unlock_irqrestore 1008 ps-6143 2d..1 306us : <stack trace> 1009 => trace_hardirqs_on_caller 1010 => trace_hardirqs_on 1011 => _raw_spin_unlock_irqrestore 1012 => do_task_stat 1013 => proc_tgid_stat 1014 => proc_single_show 1015 => seq_read 1016 => vfs_read 1017 => sys_read 1018 => system_call_fastpath 1019 1020 1021This shows that the current tracer is "irqsoff" tracing the time 1022for which interrupts were disabled. It gives the trace version (which 1023never changes) and the version of the kernel upon which this was executed on 1024(3.8). Then it displays the max latency in microseconds (259 us). The number 1025of trace entries displayed and the total number (both are four: #4/4). 1026VP, KP, SP, and HP are always zero and are reserved for later use. 1027#P is the number of online CPUs (#P:4). 1028 1029The task is the process that was running when the latency 1030occurred. (ps pid: 6143). 1031 1032The start and stop (the functions in which the interrupts were 1033disabled and enabled respectively) that caused the latencies: 1034 1035 - __lock_task_sighand is where the interrupts were disabled. 1036 - _raw_spin_unlock_irqrestore is where they were enabled again. 1037 1038The next lines after the header are the trace itself. The header 1039explains which is which. 1040 1041 cmd: The name of the process in the trace. 1042 1043 pid: The PID of that process. 1044 1045 CPU#: The CPU which the process was running on. 1046 1047 irqs-off: 'd' interrupts are disabled. '.' otherwise. 1048 1049 need-resched: 1050 - 'B' all, TIF_NEED_RESCHED, PREEMPT_NEED_RESCHED and TIF_RESCHED_LAZY is set, 1051 - 'N' both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED is set, 1052 - 'n' only TIF_NEED_RESCHED is set, 1053 - 'p' only PREEMPT_NEED_RESCHED is set, 1054 - 'L' both PREEMPT_NEED_RESCHED and TIF_RESCHED_LAZY is set, 1055 - 'b' both TIF_NEED_RESCHED and TIF_RESCHED_LAZY is set, 1056 - 'l' only TIF_RESCHED_LAZY is set 1057 - '.' otherwise. 1058 1059 hardirq/softirq: 1060 - 'Z' - NMI occurred inside a hardirq 1061 - 'z' - NMI is running 1062 - 'H' - hard irq occurred inside a softirq. 1063 - 'h' - hard irq is running 1064 - 's' - soft irq is running 1065 - '.' - normal context. 1066 1067 preempt-depth: The level of preempt_disabled 1068 1069The above is mostly meaningful for kernel developers. 1070 1071 time: 1072 When the latency-format option is enabled, the trace file 1073 output includes a timestamp relative to the start of the 1074 trace. This differs from the output when latency-format 1075 is disabled, which includes an absolute timestamp. 1076 1077 delay: 1078 This is just to help catch your eye a bit better. And 1079 needs to be fixed to be only relative to the same CPU. 1080 The marks are determined by the difference between this 1081 current trace and the next trace. 1082 1083 - '$' - greater than 1 second 1084 - '@' - greater than 100 millisecond 1085 - '*' - greater than 10 millisecond 1086 - '#' - greater than 1000 microsecond 1087 - '!' - greater than 100 microsecond 1088 - '+' - greater than 10 microsecond 1089 - ' ' - less than or equal to 10 microsecond. 1090 1091 The rest is the same as the 'trace' file. 1092 1093 Note, the latency tracers will usually end with a back trace 1094 to easily find where the latency occurred. 1095 1096trace_options 1097------------- 1098 1099The trace_options file (or the options directory) is used to control 1100what gets printed in the trace output, or manipulate the tracers. 1101To see what is available, simply cat the file:: 1102 1103 cat trace_options 1104 print-parent 1105 nosym-offset 1106 nosym-addr 1107 noverbose 1108 noraw 1109 nohex 1110 nobin 1111 noblock 1112 nofields 1113 trace_printk 1114 annotate 1115 nouserstacktrace 1116 nosym-userobj 1117 noprintk-msg-only 1118 context-info 1119 nolatency-format 1120 record-cmd 1121 norecord-tgid 1122 overwrite 1123 nodisable_on_free 1124 irq-info 1125 markers 1126 noevent-fork 1127 function-trace 1128 nofunction-fork 1129 nodisplay-graph 1130 nostacktrace 1131 nobranch 1132 1133To disable one of the options, echo in the option prepended with 1134"no":: 1135 1136 echo noprint-parent > trace_options 1137 1138To enable an option, leave off the "no":: 1139 1140 echo sym-offset > trace_options 1141 1142Here are the available options: 1143 1144 print-parent 1145 On function traces, display the calling (parent) 1146 function as well as the function being traced. 1147 :: 1148 1149 print-parent: 1150 bash-4000 [01] 1477.606694: simple_strtoul <-kstrtoul 1151 1152 noprint-parent: 1153 bash-4000 [01] 1477.606694: simple_strtoul 1154 1155 1156 sym-offset 1157 Display not only the function name, but also the 1158 offset in the function. For example, instead of 1159 seeing just "ktime_get", you will see 1160 "ktime_get+0xb/0x20". 1161 :: 1162 1163 sym-offset: 1164 bash-4000 [01] 1477.606694: simple_strtoul+0x6/0xa0 1165 1166 sym-addr 1167 This will also display the function address as well 1168 as the function name. 1169 :: 1170 1171 sym-addr: 1172 bash-4000 [01] 1477.606694: simple_strtoul <c0339346> 1173 1174 verbose 1175 This deals with the trace file when the 1176 latency-format option is enabled. 1177 :: 1178 1179 bash 4000 1 0 00000000 00010a95 [58127d26] 1720.415ms \ 1180 (+0.000ms): simple_strtoul (kstrtoul) 1181 1182 raw 1183 This will display raw numbers. This option is best for 1184 use with user applications that can translate the raw 1185 numbers better than having it done in the kernel. 1186 1187 hex 1188 Similar to raw, but the numbers will be in a hexadecimal format. 1189 1190 bin 1191 This will print out the formats in raw binary. 1192 1193 block 1194 When set, reading trace_pipe will not block when polled. 1195 1196 fields 1197 Print the fields as described by their types. This is a better 1198 option than using hex, bin or raw, as it gives a better parsing 1199 of the content of the event. 1200 1201 trace_printk 1202 Can disable trace_printk() from writing into the buffer. 1203 1204 trace_printk_dest 1205 Set to have trace_printk() and similar internal tracing functions 1206 write into this instance. Note, only one trace instance can have 1207 this set. By setting this flag, it clears the trace_printk_dest flag 1208 of the instance that had it set previously. By default, the top 1209 level trace has this set, and will get it set again if another 1210 instance has it set then clears it. 1211 1212 This flag cannot be cleared by the top level instance, as it is the 1213 default instance. The only way the top level instance has this flag 1214 cleared, is by it being set in another instance. 1215 1216 copy_trace_marker 1217 If there are applications that hard code writing into the top level 1218 trace_marker file (/sys/kernel/tracing/trace_marker or trace_marker_raw), 1219 and the tooling would like it to go into an instance, this option can 1220 be used. Create an instance and set this option, and then all writes 1221 into the top level trace_marker file will also be redirected into this 1222 instance. 1223 1224 Note, by default this option is set for the top level instance. If it 1225 is disabled, then writes to the trace_marker or trace_marker_raw files 1226 will not be written into the top level file. If no instance has this 1227 option set, then a write will error with the errno of ENODEV. 1228 1229 annotate 1230 It is sometimes confusing when the CPU buffers are full 1231 and one CPU buffer had a lot of events recently, thus 1232 a shorter time frame, were another CPU may have only had 1233 a few events, which lets it have older events. When 1234 the trace is reported, it shows the oldest events first, 1235 and it may look like only one CPU ran (the one with the 1236 oldest events). When the annotate option is set, it will 1237 display when a new CPU buffer started:: 1238 1239 <idle>-0 [001] dNs4 21169.031481: wake_up_idle_cpu <-add_timer_on 1240 <idle>-0 [001] dNs4 21169.031482: _raw_spin_unlock_irqrestore <-add_timer_on 1241 <idle>-0 [001] .Ns4 21169.031484: sub_preempt_count <-_raw_spin_unlock_irqrestore 1242 ##### CPU 2 buffer started #### 1243 <idle>-0 [002] .N.1 21169.031484: rcu_idle_exit <-cpu_idle 1244 <idle>-0 [001] .Ns3 21169.031484: _raw_spin_unlock <-clocksource_watchdog 1245 <idle>-0 [001] .Ns3 21169.031485: sub_preempt_count <-_raw_spin_unlock 1246 1247 userstacktrace 1248 This option changes the trace. It records a 1249 stacktrace of the current user space thread after 1250 each trace event. 1251 1252 sym-userobj 1253 when user stacktrace are enabled, look up which 1254 object the address belongs to, and print a 1255 relative address. This is especially useful when 1256 ASLR is on, otherwise you don't get a chance to 1257 resolve the address to object/file/line after 1258 the app is no longer running 1259 1260 The lookup is performed when you read 1261 trace,trace_pipe. Example:: 1262 1263 a.out-1623 [000] 40874.465068: /root/a.out[+0x480] <-/root/a.out[+0 1264 x494] <- /root/a.out[+0x4a8] <- /lib/libc-2.7.so[+0x1e1a6] 1265 1266 1267 printk-msg-only 1268 When set, trace_printk()s will only show the format 1269 and not their parameters (if trace_bprintk() or 1270 trace_bputs() was used to save the trace_printk()). 1271 1272 context-info 1273 Show only the event data. Hides the comm, PID, 1274 timestamp, CPU, and other useful data. 1275 1276 latency-format 1277 This option changes the trace output. When it is enabled, 1278 the trace displays additional information about the 1279 latency, as described in "Latency trace format". 1280 1281 pause-on-trace 1282 When set, opening the trace file for read, will pause 1283 writing to the ring buffer (as if tracing_on was set to zero). 1284 This simulates the original behavior of the trace file. 1285 When the file is closed, tracing will be enabled again. 1286 1287 hash-ptr 1288 When set, "%p" in the event printk format displays the 1289 hashed pointer value instead of real address. 1290 This will be useful if you want to find out which hashed 1291 value is corresponding to the real value in trace log. 1292 1293 bitmask-list 1294 When enabled, bitmasks are displayed as a human-readable list of 1295 ranges (e.g., 0,2-5,7) using the printk "%*pbl" format specifier. 1296 When disabled (the default), bitmasks are displayed in the 1297 traditional hexadecimal bitmap representation. The list format is 1298 particularly useful for tracing CPU masks and other large bitmasks 1299 where individual bit positions are more meaningful than their 1300 hexadecimal encoding. 1301 1302 record-cmd 1303 When any event or tracer is enabled, a hook is enabled 1304 in the sched_switch trace point to fill comm cache 1305 with mapped pids and comms. But this may cause some 1306 overhead, and if you only care about pids, and not the 1307 name of the task, disabling this option can lower the 1308 impact of tracing. See "saved_cmdlines". 1309 1310 record-tgid 1311 When any event or tracer is enabled, a hook is enabled 1312 in the sched_switch trace point to fill the cache of 1313 mapped Thread Group IDs (TGID) mapping to pids. See 1314 "saved_tgids". 1315 1316 overwrite 1317 This controls what happens when the trace buffer is 1318 full. If "1" (default), the oldest events are 1319 discarded and overwritten. If "0", then the newest 1320 events are discarded. 1321 (see per_cpu/cpu0/stats for overrun and dropped) 1322 1323 disable_on_free 1324 When the free_buffer is closed, tracing will 1325 stop (tracing_on set to 0). 1326 1327 irq-info 1328 Shows the interrupt, preempt count, need resched data. 1329 When disabled, the trace looks like:: 1330 1331 # tracer: function 1332 # 1333 # entries-in-buffer/entries-written: 144405/9452052 #P:4 1334 # 1335 # TASK-PID CPU# TIMESTAMP FUNCTION 1336 # | | | | | 1337 <idle>-0 [002] 23636.756054: ttwu_do_activate.constprop.89 <-try_to_wake_up 1338 <idle>-0 [002] 23636.756054: activate_task <-ttwu_do_activate.constprop.89 1339 <idle>-0 [002] 23636.756055: enqueue_task <-activate_task 1340 1341 1342 markers 1343 When set, the trace_marker is writable (only by root). 1344 When disabled, the trace_marker will error with EINVAL 1345 on write. 1346 1347 event-fork 1348 When set, tasks with PIDs listed in set_event_pid will have 1349 the PIDs of their children added to set_event_pid when those 1350 tasks fork. Also, when tasks with PIDs in set_event_pid exit, 1351 their PIDs will be removed from the file. 1352 1353 This affects PIDs listed in set_event_notrace_pid as well. 1354 1355 function-trace 1356 The latency tracers will enable function tracing 1357 if this option is enabled (default it is). When 1358 it is disabled, the latency tracers do not trace 1359 functions. This keeps the overhead of the tracer down 1360 when performing latency tests. 1361 1362 function-fork 1363 When set, tasks with PIDs listed in set_ftrace_pid will 1364 have the PIDs of their children added to set_ftrace_pid 1365 when those tasks fork. Also, when tasks with PIDs in 1366 set_ftrace_pid exit, their PIDs will be removed from the 1367 file. 1368 1369 This affects PIDs in set_ftrace_notrace_pid as well. 1370 1371 display-graph 1372 When set, the latency tracers (irqsoff, wakeup, etc) will 1373 use function graph tracing instead of function tracing. 1374 1375 stacktrace 1376 When set, a stack trace is recorded after any trace event 1377 is recorded. 1378 1379 branch 1380 Enable branch tracing with the tracer. This enables branch 1381 tracer along with the currently set tracer. Enabling this 1382 with the "nop" tracer is the same as just enabling the 1383 "branch" tracer. 1384 1385.. tip:: Some tracers have their own options. They only appear in this 1386 file when the tracer is active. They always appear in the 1387 options directory. 1388 1389 1390Here are the per tracer options: 1391 1392Options for function tracer: 1393 1394 func_stack_trace 1395 When set, a stack trace is recorded after every 1396 function that is recorded. NOTE! Limit the functions 1397 that are recorded before enabling this, with 1398 "set_ftrace_filter" otherwise the system performance 1399 will be critically degraded. Remember to disable 1400 this option before clearing the function filter. 1401 1402Options for function_graph tracer: 1403 1404 Since the function_graph tracer has a slightly different output 1405 it has its own options to control what is displayed. 1406 1407 funcgraph-overrun 1408 When set, the "overrun" of the graph stack is 1409 displayed after each function traced. The 1410 overrun, is when the stack depth of the calls 1411 is greater than what is reserved for each task. 1412 Each task has a fixed array of functions to 1413 trace in the call graph. If the depth of the 1414 calls exceeds that, the function is not traced. 1415 The overrun is the number of functions missed 1416 due to exceeding this array. 1417 1418 funcgraph-cpu 1419 When set, the CPU number of the CPU where the trace 1420 occurred is displayed. 1421 1422 funcgraph-overhead 1423 When set, if the function takes longer than 1424 A certain amount, then a delay marker is 1425 displayed. See "delay" above, under the 1426 header description. 1427 1428 funcgraph-proc 1429 Unlike other tracers, the process' command line 1430 is not displayed by default, but instead only 1431 when a task is traced in and out during a context 1432 switch. Enabling this options has the command 1433 of each process displayed at every line. 1434 1435 funcgraph-duration 1436 At the end of each function (the return) 1437 the duration of the amount of time in the 1438 function is displayed in microseconds. 1439 1440 funcgraph-abstime 1441 When set, the timestamp is displayed at each line. 1442 1443 funcgraph-irqs 1444 When disabled, functions that happen inside an 1445 interrupt will not be traced. 1446 1447 funcgraph-tail 1448 When set, the return event will include the function 1449 that it represents. By default this is off, and 1450 only a closing curly bracket "}" is displayed for 1451 the return of a function. 1452 1453 funcgraph-retval 1454 When set, the return value of each traced function 1455 will be printed after an equal sign "=". By default 1456 this is off. 1457 1458 funcgraph-retval-hex 1459 When set, the return value will always be printed 1460 in hexadecimal format. If the option is not set and 1461 the return value is an error code, it will be printed 1462 in signed decimal format; otherwise it will also be 1463 printed in hexadecimal format. By default, this option 1464 is off. 1465 1466 sleep-time 1467 When running function graph tracer, to include 1468 the time a task schedules out in its function. 1469 When enabled, it will account time the task has been 1470 scheduled out as part of the function call. 1471 1472 graph-time 1473 When running function profiler with function graph tracer, 1474 to include the time to call nested functions. When this is 1475 not set, the time reported for the function will only 1476 include the time the function itself executed for, not the 1477 time for functions that it called. 1478 1479Options for blk tracer: 1480 1481 blk_classic 1482 Shows a more minimalistic output. 1483 1484 1485irqsoff 1486------- 1487 1488When interrupts are disabled, the CPU can not react to any other 1489external event (besides NMIs and SMIs). This prevents the timer 1490interrupt from triggering or the mouse interrupt from letting 1491the kernel know of a new mouse event. The result is a latency 1492with the reaction time. 1493 1494The irqsoff tracer tracks the time for which interrupts are 1495disabled. When a new maximum latency is hit, the tracer saves 1496the trace leading up to that latency point so that every time a 1497new maximum is reached, the old saved trace is discarded and the 1498new trace is saved. 1499 1500To reset the maximum, echo 0 into tracing_max_latency. Here is 1501an example:: 1502 1503 # echo 0 > options/function-trace 1504 # echo irqsoff > current_tracer 1505 # echo 1 > tracing_on 1506 # echo 0 > tracing_max_latency 1507 # ls -ltr 1508 [...] 1509 # echo 0 > tracing_on 1510 # cat trace 1511 # tracer: irqsoff 1512 # 1513 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1514 # -------------------------------------------------------------------- 1515 # latency: 16 us, #4/4, CPU#0 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1516 # ----------------- 1517 # | task: swapper/0-0 (uid:0 nice:0 policy:0 rt_prio:0) 1518 # ----------------- 1519 # => started at: run_timer_softirq 1520 # => ended at: run_timer_softirq 1521 # 1522 # 1523 # _------=> CPU# 1524 # / _-----=> irqs-off 1525 # | / _----=> need-resched 1526 # || / _---=> hardirq/softirq 1527 # ||| / _--=> preempt-depth 1528 # |||| / delay 1529 # cmd pid ||||| time | caller 1530 # \ / ||||| \ | / 1531 <idle>-0 0d.s2 0us+: _raw_spin_lock_irq <-run_timer_softirq 1532 <idle>-0 0dNs3 17us : _raw_spin_unlock_irq <-run_timer_softirq 1533 <idle>-0 0dNs3 17us+: trace_hardirqs_on <-run_timer_softirq 1534 <idle>-0 0dNs3 25us : <stack trace> 1535 => _raw_spin_unlock_irq 1536 => run_timer_softirq 1537 => __do_softirq 1538 => call_softirq 1539 => do_softirq 1540 => irq_exit 1541 => smp_apic_timer_interrupt 1542 => apic_timer_interrupt 1543 => rcu_idle_exit 1544 => cpu_idle 1545 => rest_init 1546 => start_kernel 1547 => x86_64_start_reservations 1548 => x86_64_start_kernel 1549 1550Here we see that we had a latency of 16 microseconds (which is 1551very good). The _raw_spin_lock_irq in run_timer_softirq disabled 1552interrupts. The difference between the 16 and the displayed 1553timestamp 25us occurred because the clock was incremented 1554between the time of recording the max latency and the time of 1555recording the function that had that latency. 1556 1557Note the above example had function-trace not set. If we set 1558function-trace, we get a much larger output:: 1559 1560 with echo 1 > options/function-trace 1561 1562 # tracer: irqsoff 1563 # 1564 # irqsoff latency trace v1.1.5 on 3.8.0-test+ 1565 # -------------------------------------------------------------------- 1566 # latency: 71 us, #168/168, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1567 # ----------------- 1568 # | task: bash-2042 (uid:0 nice:0 policy:0 rt_prio:0) 1569 # ----------------- 1570 # => started at: ata_scsi_queuecmd 1571 # => ended at: ata_scsi_queuecmd 1572 # 1573 # 1574 # _------=> CPU# 1575 # / _-----=> irqs-off 1576 # | / _----=> need-resched 1577 # || / _---=> hardirq/softirq 1578 # ||| / _--=> preempt-depth 1579 # |||| / delay 1580 # cmd pid ||||| time | caller 1581 # \ / ||||| \ | / 1582 bash-2042 3d... 0us : _raw_spin_lock_irqsave <-ata_scsi_queuecmd 1583 bash-2042 3d... 0us : add_preempt_count <-_raw_spin_lock_irqsave 1584 bash-2042 3d..1 1us : ata_scsi_find_dev <-ata_scsi_queuecmd 1585 bash-2042 3d..1 1us : __ata_scsi_find_dev <-ata_scsi_find_dev 1586 bash-2042 3d..1 2us : ata_find_dev.part.14 <-__ata_scsi_find_dev 1587 bash-2042 3d..1 2us : ata_qc_new_init <-__ata_scsi_queuecmd 1588 bash-2042 3d..1 3us : ata_sg_init <-__ata_scsi_queuecmd 1589 bash-2042 3d..1 4us : ata_scsi_rw_xlat <-__ata_scsi_queuecmd 1590 bash-2042 3d..1 4us : ata_build_rw_tf <-ata_scsi_rw_xlat 1591 [...] 1592 bash-2042 3d..1 67us : delay_tsc <-__delay 1593 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc 1594 bash-2042 3d..2 67us : sub_preempt_count <-delay_tsc 1595 bash-2042 3d..1 67us : add_preempt_count <-delay_tsc 1596 bash-2042 3d..2 68us : sub_preempt_count <-delay_tsc 1597 bash-2042 3d..1 68us+: ata_bmdma_start <-ata_bmdma_qc_issue 1598 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1599 bash-2042 3d..1 71us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1600 bash-2042 3d..1 72us+: trace_hardirqs_on <-ata_scsi_queuecmd 1601 bash-2042 3d..1 120us : <stack trace> 1602 => _raw_spin_unlock_irqrestore 1603 => ata_scsi_queuecmd 1604 => scsi_dispatch_cmd 1605 => scsi_request_fn 1606 => __blk_run_queue_uncond 1607 => __blk_run_queue 1608 => blk_queue_bio 1609 => submit_bio_noacct 1610 => submit_bio 1611 => submit_bh 1612 => __ext3_get_inode_loc 1613 => ext3_iget 1614 => ext3_lookup 1615 => lookup_real 1616 => __lookup_hash 1617 => walk_component 1618 => lookup_last 1619 => path_lookupat 1620 => filename_lookup 1621 => user_path_at_empty 1622 => user_path_at 1623 => vfs_fstatat 1624 => vfs_stat 1625 => sys_newstat 1626 => system_call_fastpath 1627 1628 1629Here we traced a 71 microsecond latency. But we also see all the 1630functions that were called during that time. Note that by 1631enabling function tracing, we incur an added overhead. This 1632overhead may extend the latency times. But nevertheless, this 1633trace has provided some very helpful debugging information. 1634 1635If we prefer function graph output instead of function, we can set 1636display-graph option:: 1637 1638 with echo 1 > options/display-graph 1639 1640 # tracer: irqsoff 1641 # 1642 # irqsoff latency trace v1.1.5 on 4.20.0-rc6+ 1643 # -------------------------------------------------------------------- 1644 # latency: 3751 us, #274/274, CPU#0 | (M:desktop VP:0, KP:0, SP:0 HP:0 #P:4) 1645 # ----------------- 1646 # | task: bash-1507 (uid:0 nice:0 policy:0 rt_prio:0) 1647 # ----------------- 1648 # => started at: free_debug_processing 1649 # => ended at: return_to_handler 1650 # 1651 # 1652 # _-----=> irqs-off 1653 # / _----=> need-resched 1654 # | / _---=> hardirq/softirq 1655 # || / _--=> preempt-depth 1656 # ||| / 1657 # REL TIME CPU TASK/PID |||| DURATION FUNCTION CALLS 1658 # | | | | |||| | | | | | | 1659 0 us | 0) bash-1507 | d... | 0.000 us | _raw_spin_lock_irqsave(); 1660 0 us | 0) bash-1507 | d..1 | 0.378 us | do_raw_spin_trylock(); 1661 1 us | 0) bash-1507 | d..2 | | set_track() { 1662 2 us | 0) bash-1507 | d..2 | | save_stack_trace() { 1663 2 us | 0) bash-1507 | d..2 | | __save_stack_trace() { 1664 3 us | 0) bash-1507 | d..2 | | __unwind_start() { 1665 3 us | 0) bash-1507 | d..2 | | get_stack_info() { 1666 3 us | 0) bash-1507 | d..2 | 0.351 us | in_task_stack(); 1667 4 us | 0) bash-1507 | d..2 | 1.107 us | } 1668 [...] 1669 3750 us | 0) bash-1507 | d..1 | 0.516 us | do_raw_spin_unlock(); 1670 3750 us | 0) bash-1507 | d..1 | 0.000 us | _raw_spin_unlock_irqrestore(); 1671 3764 us | 0) bash-1507 | d..1 | 0.000 us | tracer_hardirqs_on(); 1672 bash-1507 0d..1 3792us : <stack trace> 1673 => free_debug_processing 1674 => __slab_free 1675 => kmem_cache_free 1676 => vm_area_free 1677 => remove_vma 1678 => exit_mmap 1679 => mmput 1680 => begin_new_exec 1681 => load_elf_binary 1682 => search_binary_handler 1683 => __do_execve_file.isra.32 1684 => __x64_sys_execve 1685 => do_syscall_64 1686 => entry_SYSCALL_64_after_hwframe 1687 1688preemptoff 1689---------- 1690 1691When preemption is disabled, we may be able to receive 1692interrupts but the task cannot be preempted and a higher 1693priority task must wait for preemption to be enabled again 1694before it can preempt a lower priority task. 1695 1696The preemptoff tracer traces the places that disable preemption. 1697Like the irqsoff tracer, it records the maximum latency for 1698which preemption was disabled. The control of preemptoff tracer 1699is much like the irqsoff tracer. 1700:: 1701 1702 # echo 0 > options/function-trace 1703 # echo preemptoff > current_tracer 1704 # echo 1 > tracing_on 1705 # echo 0 > tracing_max_latency 1706 # ls -ltr 1707 [...] 1708 # echo 0 > tracing_on 1709 # cat trace 1710 # tracer: preemptoff 1711 # 1712 # preemptoff latency trace v1.1.5 on 3.8.0-test+ 1713 # -------------------------------------------------------------------- 1714 # latency: 46 us, #4/4, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1715 # ----------------- 1716 # | task: sshd-1991 (uid:0 nice:0 policy:0 rt_prio:0) 1717 # ----------------- 1718 # => started at: do_IRQ 1719 # => ended at: do_IRQ 1720 # 1721 # 1722 # _------=> CPU# 1723 # / _-----=> irqs-off 1724 # | / _----=> need-resched 1725 # || / _---=> hardirq/softirq 1726 # ||| / _--=> preempt-depth 1727 # |||| / delay 1728 # cmd pid ||||| time | caller 1729 # \ / ||||| \ | / 1730 sshd-1991 1d.h. 0us+: irq_enter <-do_IRQ 1731 sshd-1991 1d..1 46us : irq_exit <-do_IRQ 1732 sshd-1991 1d..1 47us+: trace_preempt_on <-do_IRQ 1733 sshd-1991 1d..1 52us : <stack trace> 1734 => sub_preempt_count 1735 => irq_exit 1736 => do_IRQ 1737 => ret_from_intr 1738 1739 1740This has some more changes. Preemption was disabled when an 1741interrupt came in (notice the 'h'), and was enabled on exit. 1742But we also see that interrupts have been disabled when entering 1743the preempt off section and leaving it (the 'd'). We do not know if 1744interrupts were enabled in the mean time or shortly after this 1745was over. 1746:: 1747 1748 # tracer: preemptoff 1749 # 1750 # preemptoff latency trace v1.1.5 on 3.8.0-test+ 1751 # -------------------------------------------------------------------- 1752 # latency: 83 us, #241/241, CPU#1 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1753 # ----------------- 1754 # | task: bash-1994 (uid:0 nice:0 policy:0 rt_prio:0) 1755 # ----------------- 1756 # => started at: wake_up_new_task 1757 # => ended at: task_rq_unlock 1758 # 1759 # 1760 # _------=> CPU# 1761 # / _-----=> irqs-off 1762 # | / _----=> need-resched 1763 # || / _---=> hardirq/softirq 1764 # ||| / _--=> preempt-depth 1765 # |||| / delay 1766 # cmd pid ||||| time | caller 1767 # \ / ||||| \ | / 1768 bash-1994 1d..1 0us : _raw_spin_lock_irqsave <-wake_up_new_task 1769 bash-1994 1d..1 0us : select_task_rq_fair <-select_task_rq 1770 bash-1994 1d..1 1us : __rcu_read_lock <-select_task_rq_fair 1771 bash-1994 1d..1 1us : source_load <-select_task_rq_fair 1772 bash-1994 1d..1 1us : source_load <-select_task_rq_fair 1773 [...] 1774 bash-1994 1d..1 12us : irq_enter <-smp_apic_timer_interrupt 1775 bash-1994 1d..1 12us : rcu_irq_enter <-irq_enter 1776 bash-1994 1d..1 13us : add_preempt_count <-irq_enter 1777 bash-1994 1d.h1 13us : exit_idle <-smp_apic_timer_interrupt 1778 bash-1994 1d.h1 13us : hrtimer_interrupt <-smp_apic_timer_interrupt 1779 bash-1994 1d.h1 13us : _raw_spin_lock <-hrtimer_interrupt 1780 bash-1994 1d.h1 14us : add_preempt_count <-_raw_spin_lock 1781 bash-1994 1d.h2 14us : ktime_get_update_offsets <-hrtimer_interrupt 1782 [...] 1783 bash-1994 1d.h1 35us : lapic_next_event <-clockevents_program_event 1784 bash-1994 1d.h1 35us : irq_exit <-smp_apic_timer_interrupt 1785 bash-1994 1d.h1 36us : sub_preempt_count <-irq_exit 1786 bash-1994 1d..2 36us : do_softirq <-irq_exit 1787 bash-1994 1d..2 36us : __do_softirq <-call_softirq 1788 bash-1994 1d..2 36us : __local_bh_disable <-__do_softirq 1789 bash-1994 1d.s2 37us : add_preempt_count <-_raw_spin_lock_irq 1790 bash-1994 1d.s3 38us : _raw_spin_unlock <-run_timer_softirq 1791 bash-1994 1d.s3 39us : sub_preempt_count <-_raw_spin_unlock 1792 bash-1994 1d.s2 39us : call_timer_fn <-run_timer_softirq 1793 [...] 1794 bash-1994 1dNs2 81us : cpu_needs_another_gp <-rcu_process_callbacks 1795 bash-1994 1dNs2 82us : __local_bh_enable <-__do_softirq 1796 bash-1994 1dNs2 82us : sub_preempt_count <-__local_bh_enable 1797 bash-1994 1dN.2 82us : idle_cpu <-irq_exit 1798 bash-1994 1dN.2 83us : rcu_irq_exit <-irq_exit 1799 bash-1994 1dN.2 83us : sub_preempt_count <-irq_exit 1800 bash-1994 1.N.1 84us : _raw_spin_unlock_irqrestore <-task_rq_unlock 1801 bash-1994 1.N.1 84us+: trace_preempt_on <-task_rq_unlock 1802 bash-1994 1.N.1 104us : <stack trace> 1803 => sub_preempt_count 1804 => _raw_spin_unlock_irqrestore 1805 => task_rq_unlock 1806 => wake_up_new_task 1807 => do_fork 1808 => sys_clone 1809 => stub_clone 1810 1811 1812The above is an example of the preemptoff trace with 1813function-trace set. Here we see that interrupts were not disabled 1814the entire time. The irq_enter code lets us know that we entered 1815an interrupt 'h'. Before that, the functions being traced still 1816show that it is not in an interrupt, but we can see from the 1817functions themselves that this is not the case. 1818 1819preemptirqsoff 1820-------------- 1821 1822Knowing the locations that have interrupts disabled or 1823preemption disabled for the longest times is helpful. But 1824sometimes we would like to know when either preemption and/or 1825interrupts are disabled. 1826 1827Consider the following code:: 1828 1829 local_irq_disable(); 1830 call_function_with_irqs_off(); 1831 preempt_disable(); 1832 call_function_with_irqs_and_preemption_off(); 1833 local_irq_enable(); 1834 call_function_with_preemption_off(); 1835 preempt_enable(); 1836 1837The irqsoff tracer will record the total length of 1838call_function_with_irqs_off() and 1839call_function_with_irqs_and_preemption_off(). 1840 1841The preemptoff tracer will record the total length of 1842call_function_with_irqs_and_preemption_off() and 1843call_function_with_preemption_off(). 1844 1845But neither will trace the time that interrupts and/or 1846preemption is disabled. This total time is the time that we can 1847not schedule. To record this time, use the preemptirqsoff 1848tracer. 1849 1850Again, using this trace is much like the irqsoff and preemptoff 1851tracers. 1852:: 1853 1854 # echo 0 > options/function-trace 1855 # echo preemptirqsoff > current_tracer 1856 # echo 1 > tracing_on 1857 # echo 0 > tracing_max_latency 1858 # ls -ltr 1859 [...] 1860 # echo 0 > tracing_on 1861 # cat trace 1862 # tracer: preemptirqsoff 1863 # 1864 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+ 1865 # -------------------------------------------------------------------- 1866 # latency: 100 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1867 # ----------------- 1868 # | task: ls-2230 (uid:0 nice:0 policy:0 rt_prio:0) 1869 # ----------------- 1870 # => started at: ata_scsi_queuecmd 1871 # => ended at: ata_scsi_queuecmd 1872 # 1873 # 1874 # _------=> CPU# 1875 # / _-----=> irqs-off 1876 # | / _----=> need-resched 1877 # || / _---=> hardirq/softirq 1878 # ||| / _--=> preempt-depth 1879 # |||| / delay 1880 # cmd pid ||||| time | caller 1881 # \ / ||||| \ | / 1882 ls-2230 3d... 0us+: _raw_spin_lock_irqsave <-ata_scsi_queuecmd 1883 ls-2230 3...1 100us : _raw_spin_unlock_irqrestore <-ata_scsi_queuecmd 1884 ls-2230 3...1 101us+: trace_preempt_on <-ata_scsi_queuecmd 1885 ls-2230 3...1 111us : <stack trace> 1886 => sub_preempt_count 1887 => _raw_spin_unlock_irqrestore 1888 => ata_scsi_queuecmd 1889 => scsi_dispatch_cmd 1890 => scsi_request_fn 1891 => __blk_run_queue_uncond 1892 => __blk_run_queue 1893 => blk_queue_bio 1894 => submit_bio_noacct 1895 => submit_bio 1896 => submit_bh 1897 => ext3_bread 1898 => ext3_dir_bread 1899 => htree_dirblock_to_tree 1900 => ext3_htree_fill_tree 1901 => ext3_readdir 1902 => vfs_readdir 1903 => sys_getdents 1904 => system_call_fastpath 1905 1906 1907The trace_hardirqs_off_thunk is called from assembly on x86 when 1908interrupts are disabled in the assembly code. Without the 1909function tracing, we do not know if interrupts were enabled 1910within the preemption points. We do see that it started with 1911preemption enabled. 1912 1913Here is a trace with function-trace set:: 1914 1915 # tracer: preemptirqsoff 1916 # 1917 # preemptirqsoff latency trace v1.1.5 on 3.8.0-test+ 1918 # -------------------------------------------------------------------- 1919 # latency: 161 us, #339/339, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 1920 # ----------------- 1921 # | task: ls-2269 (uid:0 nice:0 policy:0 rt_prio:0) 1922 # ----------------- 1923 # => started at: schedule 1924 # => ended at: mutex_unlock 1925 # 1926 # 1927 # _------=> CPU# 1928 # / _-----=> irqs-off 1929 # | / _----=> need-resched 1930 # || / _---=> hardirq/softirq 1931 # ||| / _--=> preempt-depth 1932 # |||| / delay 1933 # cmd pid ||||| time | caller 1934 # \ / ||||| \ | / 1935 kworker/-59 3...1 0us : __schedule <-schedule 1936 kworker/-59 3d..1 0us : rcu_preempt_qs <-rcu_note_context_switch 1937 kworker/-59 3d..1 1us : add_preempt_count <-_raw_spin_lock_irq 1938 kworker/-59 3d..2 1us : deactivate_task <-__schedule 1939 kworker/-59 3d..2 1us : dequeue_task <-deactivate_task 1940 kworker/-59 3d..2 2us : update_rq_clock <-dequeue_task 1941 kworker/-59 3d..2 2us : dequeue_task_fair <-dequeue_task 1942 kworker/-59 3d..2 2us : update_curr <-dequeue_task_fair 1943 kworker/-59 3d..2 2us : update_min_vruntime <-update_curr 1944 kworker/-59 3d..2 3us : cpuacct_charge <-update_curr 1945 kworker/-59 3d..2 3us : __rcu_read_lock <-cpuacct_charge 1946 kworker/-59 3d..2 3us : __rcu_read_unlock <-cpuacct_charge 1947 kworker/-59 3d..2 3us : update_cfs_rq_blocked_load <-dequeue_task_fair 1948 kworker/-59 3d..2 4us : clear_buddies <-dequeue_task_fair 1949 kworker/-59 3d..2 4us : account_entity_dequeue <-dequeue_task_fair 1950 kworker/-59 3d..2 4us : update_min_vruntime <-dequeue_task_fair 1951 kworker/-59 3d..2 4us : update_cfs_shares <-dequeue_task_fair 1952 kworker/-59 3d..2 5us : hrtick_update <-dequeue_task_fair 1953 kworker/-59 3d..2 5us : wq_worker_sleeping <-__schedule 1954 kworker/-59 3d..2 5us : kthread_data <-wq_worker_sleeping 1955 kworker/-59 3d..2 5us : put_prev_task_fair <-__schedule 1956 kworker/-59 3d..2 6us : pick_next_task_fair <-pick_next_task 1957 kworker/-59 3d..2 6us : clear_buddies <-pick_next_task_fair 1958 kworker/-59 3d..2 6us : set_next_entity <-pick_next_task_fair 1959 kworker/-59 3d..2 6us : update_stats_wait_end <-set_next_entity 1960 ls-2269 3d..2 7us : finish_task_switch <-__schedule 1961 ls-2269 3d..2 7us : _raw_spin_unlock_irq <-finish_task_switch 1962 ls-2269 3d..2 8us : do_IRQ <-ret_from_intr 1963 ls-2269 3d..2 8us : irq_enter <-do_IRQ 1964 ls-2269 3d..2 8us : rcu_irq_enter <-irq_enter 1965 ls-2269 3d..2 9us : add_preempt_count <-irq_enter 1966 ls-2269 3d.h2 9us : exit_idle <-do_IRQ 1967 [...] 1968 ls-2269 3d.h3 20us : sub_preempt_count <-_raw_spin_unlock 1969 ls-2269 3d.h2 20us : irq_exit <-do_IRQ 1970 ls-2269 3d.h2 21us : sub_preempt_count <-irq_exit 1971 ls-2269 3d..3 21us : do_softirq <-irq_exit 1972 ls-2269 3d..3 21us : __do_softirq <-call_softirq 1973 ls-2269 3d..3 21us+: __local_bh_disable <-__do_softirq 1974 ls-2269 3d.s4 29us : sub_preempt_count <-_local_bh_enable_ip 1975 ls-2269 3d.s5 29us : sub_preempt_count <-_local_bh_enable_ip 1976 ls-2269 3d.s5 31us : do_IRQ <-ret_from_intr 1977 ls-2269 3d.s5 31us : irq_enter <-do_IRQ 1978 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter 1979 [...] 1980 ls-2269 3d.s5 31us : rcu_irq_enter <-irq_enter 1981 ls-2269 3d.s5 32us : add_preempt_count <-irq_enter 1982 ls-2269 3d.H5 32us : exit_idle <-do_IRQ 1983 ls-2269 3d.H5 32us : handle_irq <-do_IRQ 1984 ls-2269 3d.H5 32us : irq_to_desc <-handle_irq 1985 ls-2269 3d.H5 33us : handle_fasteoi_irq <-handle_irq 1986 [...] 1987 ls-2269 3d.s5 158us : _raw_spin_unlock_irqrestore <-rtl8139_poll 1988 ls-2269 3d.s3 158us : net_rps_action_and_irq_enable.isra.65 <-net_rx_action 1989 ls-2269 3d.s3 159us : __local_bh_enable <-__do_softirq 1990 ls-2269 3d.s3 159us : sub_preempt_count <-__local_bh_enable 1991 ls-2269 3d..3 159us : idle_cpu <-irq_exit 1992 ls-2269 3d..3 159us : rcu_irq_exit <-irq_exit 1993 ls-2269 3d..3 160us : sub_preempt_count <-irq_exit 1994 ls-2269 3d... 161us : __mutex_unlock_slowpath <-mutex_unlock 1995 ls-2269 3d... 162us+: trace_hardirqs_on <-mutex_unlock 1996 ls-2269 3d... 186us : <stack trace> 1997 => __mutex_unlock_slowpath 1998 => mutex_unlock 1999 => process_output 2000 => n_tty_write 2001 => tty_write 2002 => vfs_write 2003 => sys_write 2004 => system_call_fastpath 2005 2006This is an interesting trace. It started with kworker running and 2007scheduling out and ls taking over. But as soon as ls released the 2008rq lock and enabled interrupts (but not preemption) an interrupt 2009triggered. When the interrupt finished, it started running softirqs. 2010But while the softirq was running, another interrupt triggered. 2011When an interrupt is running inside a softirq, the annotation is 'H'. 2012 2013 2014wakeup 2015------ 2016 2017One common case that people are interested in tracing is the 2018time it takes for a task that is woken to actually wake up. 2019Now for non Real-Time tasks, this can be arbitrary. But tracing 2020it nonetheless can be interesting. 2021 2022Without function tracing:: 2023 2024 # echo 0 > options/function-trace 2025 # echo wakeup > current_tracer 2026 # echo 1 > tracing_on 2027 # echo 0 > tracing_max_latency 2028 # chrt -f 5 sleep 1 2029 # echo 0 > tracing_on 2030 # cat trace 2031 # tracer: wakeup 2032 # 2033 # wakeup latency trace v1.1.5 on 3.8.0-test+ 2034 # -------------------------------------------------------------------- 2035 # latency: 15 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2036 # ----------------- 2037 # | task: kworker/3:1H-312 (uid:0 nice:-20 policy:0 rt_prio:0) 2038 # ----------------- 2039 # 2040 # _------=> CPU# 2041 # / _-----=> irqs-off 2042 # | / _----=> need-resched 2043 # || / _---=> hardirq/softirq 2044 # ||| / _--=> preempt-depth 2045 # |||| / delay 2046 # cmd pid ||||| time | caller 2047 # \ / ||||| \ | / 2048 <idle>-0 3dNs7 0us : 0:120:R + [003] 312:100:R kworker/3:1H 2049 <idle>-0 3dNs7 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up 2050 <idle>-0 3d..3 15us : __schedule <-schedule 2051 <idle>-0 3d..3 15us : 0:120:R ==> [003] 312:100:R kworker/3:1H 2052 2053The tracer only traces the highest priority task in the system 2054to avoid tracing the normal circumstances. Here we see that 2055the kworker with a nice priority of -20 (not very nice), took 2056just 15 microseconds from the time it woke up, to the time it 2057ran. 2058 2059Non Real-Time tasks are not that interesting. A more interesting 2060trace is to concentrate only on Real-Time tasks. 2061 2062wakeup_rt 2063--------- 2064 2065In a Real-Time environment it is very important to know the 2066wakeup time it takes for the highest priority task that is woken 2067up to the time that it executes. This is also known as "schedule 2068latency". I stress the point that this is about RT tasks. It is 2069also important to know the scheduling latency of non-RT tasks, 2070but the average schedule latency is better for non-RT tasks. 2071Tools like LatencyTop are more appropriate for such 2072measurements. 2073 2074Real-Time environments are interested in the worst case latency. 2075That is the longest latency it takes for something to happen, 2076and not the average. We can have a very fast scheduler that may 2077only have a large latency once in a while, but that would not 2078work well with Real-Time tasks. The wakeup_rt tracer was designed 2079to record the worst case wakeups of RT tasks. Non-RT tasks are 2080not recorded because the tracer only records one worst case and 2081tracing non-RT tasks that are unpredictable will overwrite the 2082worst case latency of RT tasks (just run the normal wakeup 2083tracer for a while to see that effect). 2084 2085Since this tracer only deals with RT tasks, we will run this 2086slightly differently than we did with the previous tracers. 2087Instead of performing an 'ls', we will run 'sleep 1' under 2088'chrt' which changes the priority of the task. 2089:: 2090 2091 # echo 0 > options/function-trace 2092 # echo wakeup_rt > current_tracer 2093 # echo 1 > tracing_on 2094 # echo 0 > tracing_max_latency 2095 # chrt -f 5 sleep 1 2096 # echo 0 > tracing_on 2097 # cat trace 2098 # tracer: wakeup 2099 # 2100 # tracer: wakeup_rt 2101 # 2102 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2103 # -------------------------------------------------------------------- 2104 # latency: 5 us, #4/4, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2105 # ----------------- 2106 # | task: sleep-2389 (uid:0 nice:0 policy:1 rt_prio:5) 2107 # ----------------- 2108 # 2109 # _------=> CPU# 2110 # / _-----=> irqs-off 2111 # | / _----=> need-resched 2112 # || / _---=> hardirq/softirq 2113 # ||| / _--=> preempt-depth 2114 # |||| / delay 2115 # cmd pid ||||| time | caller 2116 # \ / ||||| \ | / 2117 <idle>-0 3d.h4 0us : 0:120:R + [003] 2389: 94:R sleep 2118 <idle>-0 3d.h4 1us+: ttwu_do_activate.constprop.87 <-try_to_wake_up 2119 <idle>-0 3d..3 5us : __schedule <-schedule 2120 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep 2121 2122 2123Running this on an idle system, we see that it only took 5 microseconds 2124to perform the task switch. Note, since the trace point in the schedule 2125is before the actual "switch", we stop the tracing when the recorded task 2126is about to schedule in. This may change if we add a new marker at the 2127end of the scheduler. 2128 2129Notice that the recorded task is 'sleep' with the PID of 2389 2130and it has an rt_prio of 5. This priority is user-space priority 2131and not the internal kernel priority. The policy is 1 for 2132SCHED_FIFO and 2 for SCHED_RR. 2133 2134Note, that the trace data shows the internal priority (99 - rtprio). 2135:: 2136 2137 <idle>-0 3d..3 5us : 0:120:R ==> [003] 2389: 94:R sleep 2138 2139The 0:120:R means idle was running with a nice priority of 0 (120 - 120) 2140and in the running state 'R'. The sleep task was scheduled in with 21412389: 94:R. That is the priority is the kernel rtprio (99 - 5 = 94) 2142and it too is in the running state. 2143 2144Doing the same with chrt -r 5 and function-trace set. 2145:: 2146 2147 echo 1 > options/function-trace 2148 2149 # tracer: wakeup_rt 2150 # 2151 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2152 # -------------------------------------------------------------------- 2153 # latency: 29 us, #85/85, CPU#3 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2154 # ----------------- 2155 # | task: sleep-2448 (uid:0 nice:0 policy:1 rt_prio:5) 2156 # ----------------- 2157 # 2158 # _------=> CPU# 2159 # / _-----=> irqs-off 2160 # | / _----=> need-resched 2161 # || / _---=> hardirq/softirq 2162 # ||| / _--=> preempt-depth 2163 # |||| / delay 2164 # cmd pid ||||| time | caller 2165 # \ / ||||| \ | / 2166 <idle>-0 3d.h4 1us+: 0:120:R + [003] 2448: 94:R sleep 2167 <idle>-0 3d.h4 2us : ttwu_do_activate.constprop.87 <-try_to_wake_up 2168 <idle>-0 3d.h3 3us : check_preempt_curr <-ttwu_do_wakeup 2169 <idle>-0 3d.h3 3us : resched_curr <-check_preempt_curr 2170 <idle>-0 3dNh3 4us : task_woken_rt <-ttwu_do_wakeup 2171 <idle>-0 3dNh3 4us : _raw_spin_unlock <-try_to_wake_up 2172 <idle>-0 3dNh3 4us : sub_preempt_count <-_raw_spin_unlock 2173 <idle>-0 3dNh2 5us : ttwu_stat <-try_to_wake_up 2174 <idle>-0 3dNh2 5us : _raw_spin_unlock_irqrestore <-try_to_wake_up 2175 <idle>-0 3dNh2 6us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2176 <idle>-0 3dNh1 6us : _raw_spin_lock <-__run_hrtimer 2177 <idle>-0 3dNh1 6us : add_preempt_count <-_raw_spin_lock 2178 <idle>-0 3dNh2 7us : _raw_spin_unlock <-hrtimer_interrupt 2179 <idle>-0 3dNh2 7us : sub_preempt_count <-_raw_spin_unlock 2180 <idle>-0 3dNh1 7us : tick_program_event <-hrtimer_interrupt 2181 <idle>-0 3dNh1 7us : clockevents_program_event <-tick_program_event 2182 <idle>-0 3dNh1 8us : ktime_get <-clockevents_program_event 2183 <idle>-0 3dNh1 8us : lapic_next_event <-clockevents_program_event 2184 <idle>-0 3dNh1 8us : irq_exit <-smp_apic_timer_interrupt 2185 <idle>-0 3dNh1 9us : sub_preempt_count <-irq_exit 2186 <idle>-0 3dN.2 9us : idle_cpu <-irq_exit 2187 <idle>-0 3dN.2 9us : rcu_irq_exit <-irq_exit 2188 <idle>-0 3dN.2 10us : rcu_eqs_enter_common.isra.45 <-rcu_irq_exit 2189 <idle>-0 3dN.2 10us : sub_preempt_count <-irq_exit 2190 <idle>-0 3.N.1 11us : rcu_idle_exit <-cpu_idle 2191 <idle>-0 3dN.1 11us : rcu_eqs_exit_common.isra.43 <-rcu_idle_exit 2192 <idle>-0 3.N.1 11us : tick_nohz_idle_exit <-cpu_idle 2193 <idle>-0 3dN.1 12us : menu_hrtimer_cancel <-tick_nohz_idle_exit 2194 <idle>-0 3dN.1 12us : ktime_get <-tick_nohz_idle_exit 2195 <idle>-0 3dN.1 12us : tick_do_update_jiffies64 <-tick_nohz_idle_exit 2196 <idle>-0 3dN.1 13us : cpu_load_update_nohz <-tick_nohz_idle_exit 2197 <idle>-0 3dN.1 13us : _raw_spin_lock <-cpu_load_update_nohz 2198 <idle>-0 3dN.1 13us : add_preempt_count <-_raw_spin_lock 2199 <idle>-0 3dN.2 13us : __cpu_load_update <-cpu_load_update_nohz 2200 <idle>-0 3dN.2 14us : sched_avg_update <-__cpu_load_update 2201 <idle>-0 3dN.2 14us : _raw_spin_unlock <-cpu_load_update_nohz 2202 <idle>-0 3dN.2 14us : sub_preempt_count <-_raw_spin_unlock 2203 <idle>-0 3dN.1 15us : calc_load_nohz_stop <-tick_nohz_idle_exit 2204 <idle>-0 3dN.1 15us : touch_softlockup_watchdog <-tick_nohz_idle_exit 2205 <idle>-0 3dN.1 15us : hrtimer_cancel <-tick_nohz_idle_exit 2206 <idle>-0 3dN.1 15us : hrtimer_try_to_cancel <-hrtimer_cancel 2207 <idle>-0 3dN.1 16us : lock_hrtimer_base.isra.18 <-hrtimer_try_to_cancel 2208 <idle>-0 3dN.1 16us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18 2209 <idle>-0 3dN.1 16us : add_preempt_count <-_raw_spin_lock_irqsave 2210 <idle>-0 3dN.2 17us : __remove_hrtimer <-remove_hrtimer.part.16 2211 <idle>-0 3dN.2 17us : hrtimer_force_reprogram <-__remove_hrtimer 2212 <idle>-0 3dN.2 17us : tick_program_event <-hrtimer_force_reprogram 2213 <idle>-0 3dN.2 18us : clockevents_program_event <-tick_program_event 2214 <idle>-0 3dN.2 18us : ktime_get <-clockevents_program_event 2215 <idle>-0 3dN.2 18us : lapic_next_event <-clockevents_program_event 2216 <idle>-0 3dN.2 19us : _raw_spin_unlock_irqrestore <-hrtimer_try_to_cancel 2217 <idle>-0 3dN.2 19us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2218 <idle>-0 3dN.1 19us : hrtimer_forward <-tick_nohz_idle_exit 2219 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward 2220 <idle>-0 3dN.1 20us : ktime_add_safe <-hrtimer_forward 2221 <idle>-0 3dN.1 20us : hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11 2222 <idle>-0 3dN.1 20us : __hrtimer_start_range_ns <-hrtimer_start_range_ns 2223 <idle>-0 3dN.1 21us : lock_hrtimer_base.isra.18 <-__hrtimer_start_range_ns 2224 <idle>-0 3dN.1 21us : _raw_spin_lock_irqsave <-lock_hrtimer_base.isra.18 2225 <idle>-0 3dN.1 21us : add_preempt_count <-_raw_spin_lock_irqsave 2226 <idle>-0 3dN.2 22us : ktime_add_safe <-__hrtimer_start_range_ns 2227 <idle>-0 3dN.2 22us : enqueue_hrtimer <-__hrtimer_start_range_ns 2228 <idle>-0 3dN.2 22us : tick_program_event <-__hrtimer_start_range_ns 2229 <idle>-0 3dN.2 23us : clockevents_program_event <-tick_program_event 2230 <idle>-0 3dN.2 23us : ktime_get <-clockevents_program_event 2231 <idle>-0 3dN.2 23us : lapic_next_event <-clockevents_program_event 2232 <idle>-0 3dN.2 24us : _raw_spin_unlock_irqrestore <-__hrtimer_start_range_ns 2233 <idle>-0 3dN.2 24us : sub_preempt_count <-_raw_spin_unlock_irqrestore 2234 <idle>-0 3dN.1 24us : account_idle_ticks <-tick_nohz_idle_exit 2235 <idle>-0 3dN.1 24us : account_idle_time <-account_idle_ticks 2236 <idle>-0 3.N.1 25us : sub_preempt_count <-cpu_idle 2237 <idle>-0 3.N.. 25us : schedule <-cpu_idle 2238 <idle>-0 3.N.. 25us : __schedule <-preempt_schedule 2239 <idle>-0 3.N.. 26us : add_preempt_count <-__schedule 2240 <idle>-0 3.N.1 26us : rcu_note_context_switch <-__schedule 2241 <idle>-0 3.N.1 26us : rcu_sched_qs <-rcu_note_context_switch 2242 <idle>-0 3dN.1 27us : rcu_preempt_qs <-rcu_note_context_switch 2243 <idle>-0 3.N.1 27us : _raw_spin_lock_irq <-__schedule 2244 <idle>-0 3dN.1 27us : add_preempt_count <-_raw_spin_lock_irq 2245 <idle>-0 3dN.2 28us : put_prev_task_idle <-__schedule 2246 <idle>-0 3dN.2 28us : pick_next_task_stop <-pick_next_task 2247 <idle>-0 3dN.2 28us : pick_next_task_rt <-pick_next_task 2248 <idle>-0 3dN.2 29us : dequeue_pushable_task <-pick_next_task_rt 2249 <idle>-0 3d..3 29us : __schedule <-preempt_schedule 2250 <idle>-0 3d..3 30us : 0:120:R ==> [003] 2448: 94:R sleep 2251 2252This isn't that big of a trace, even with function tracing enabled, 2253so I included the entire trace. 2254 2255The interrupt went off while when the system was idle. Somewhere 2256before task_woken_rt() was called, the NEED_RESCHED flag was set, 2257this is indicated by the first occurrence of the 'N' flag. 2258 2259Latency tracing and events 2260-------------------------- 2261As function tracing can induce a much larger latency, but without 2262seeing what happens within the latency it is hard to know what 2263caused it. There is a middle ground, and that is with enabling 2264events. 2265:: 2266 2267 # echo 0 > options/function-trace 2268 # echo wakeup_rt > current_tracer 2269 # echo 1 > events/enable 2270 # echo 1 > tracing_on 2271 # echo 0 > tracing_max_latency 2272 # chrt -f 5 sleep 1 2273 # echo 0 > tracing_on 2274 # cat trace 2275 # tracer: wakeup_rt 2276 # 2277 # wakeup_rt latency trace v1.1.5 on 3.8.0-test+ 2278 # -------------------------------------------------------------------- 2279 # latency: 6 us, #12/12, CPU#2 | (M:preempt VP:0, KP:0, SP:0 HP:0 #P:4) 2280 # ----------------- 2281 # | task: sleep-5882 (uid:0 nice:0 policy:1 rt_prio:5) 2282 # ----------------- 2283 # 2284 # _------=> CPU# 2285 # / _-----=> irqs-off 2286 # | / _----=> need-resched 2287 # || / _---=> hardirq/softirq 2288 # ||| / _--=> preempt-depth 2289 # |||| / delay 2290 # cmd pid ||||| time | caller 2291 # \ / ||||| \ | / 2292 <idle>-0 2d.h4 0us : 0:120:R + [002] 5882: 94:R sleep 2293 <idle>-0 2d.h4 0us : ttwu_do_activate.constprop.87 <-try_to_wake_up 2294 <idle>-0 2d.h4 1us : sched_wakeup: comm=sleep pid=5882 prio=94 success=1 target_cpu=002 2295 <idle>-0 2dNh2 1us : hrtimer_expire_exit: hrtimer=ffff88007796feb8 2296 <idle>-0 2.N.2 2us : power_end: cpu_id=2 2297 <idle>-0 2.N.2 3us : cpu_idle: state=4294967295 cpu_id=2 2298 <idle>-0 2dN.3 4us : hrtimer_cancel: hrtimer=ffff88007d50d5e0 2299 <idle>-0 2dN.3 4us : hrtimer_start: hrtimer=ffff88007d50d5e0 function=tick_sched_timer expires=34311211000000 softexpires=34311211000000 2300 <idle>-0 2.N.2 5us : rcu_utilization: Start context switch 2301 <idle>-0 2.N.2 5us : rcu_utilization: End context switch 2302 <idle>-0 2d..3 6us : __schedule <-schedule 2303 <idle>-0 2d..3 6us : 0:120:R ==> [002] 5882: 94:R sleep 2304 2305 2306Hardware Latency Detector 2307------------------------- 2308 2309The hardware latency detector is executed by enabling the "hwlat" tracer. 2310 2311NOTE, this tracer will affect the performance of the system as it will 2312periodically make a CPU constantly busy with interrupts disabled. 2313:: 2314 2315 # echo hwlat > current_tracer 2316 # sleep 100 2317 # cat trace 2318 # tracer: hwlat 2319 # 2320 # entries-in-buffer/entries-written: 13/13 #P:8 2321 # 2322 # _-----=> irqs-off 2323 # / _----=> need-resched 2324 # | / _---=> hardirq/softirq 2325 # || / _--=> preempt-depth 2326 # ||| / delay 2327 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2328 # | | | |||| | | 2329 <...>-1729 [001] d... 678.473449: #1 inner/outer(us): 11/12 ts:1581527483.343962693 count:6 2330 <...>-1729 [004] d... 689.556542: #2 inner/outer(us): 16/9 ts:1581527494.889008092 count:1 2331 <...>-1729 [005] d... 714.756290: #3 inner/outer(us): 16/16 ts:1581527519.678961629 count:5 2332 <...>-1729 [001] d... 718.788247: #4 inner/outer(us): 9/17 ts:1581527523.889012713 count:1 2333 <...>-1729 [002] d... 719.796341: #5 inner/outer(us): 13/9 ts:1581527524.912872606 count:1 2334 <...>-1729 [006] d... 844.787091: #6 inner/outer(us): 9/12 ts:1581527649.889048502 count:2 2335 <...>-1729 [003] d... 849.827033: #7 inner/outer(us): 18/9 ts:1581527654.889013793 count:1 2336 <...>-1729 [007] d... 853.859002: #8 inner/outer(us): 9/12 ts:1581527658.889065736 count:1 2337 <...>-1729 [001] d... 855.874978: #9 inner/outer(us): 9/11 ts:1581527660.861991877 count:1 2338 <...>-1729 [001] d... 863.938932: #10 inner/outer(us): 9/11 ts:1581527668.970010500 count:1 nmi-total:7 nmi-count:1 2339 <...>-1729 [007] d... 878.050780: #11 inner/outer(us): 9/12 ts:1581527683.385002600 count:1 nmi-total:5 nmi-count:1 2340 <...>-1729 [007] d... 886.114702: #12 inner/outer(us): 9/12 ts:1581527691.385001600 count:1 2341 2342 2343The above output is somewhat the same in the header. All events will have 2344interrupts disabled 'd'. Under the FUNCTION title there is: 2345 2346 #1 2347 This is the count of events recorded that were greater than the 2348 tracing_threshold (See below). 2349 2350 inner/outer(us): 11/11 2351 2352 This shows two numbers as "inner latency" and "outer latency". The test 2353 runs in a loop checking a timestamp twice. The latency detected within 2354 the two timestamps is the "inner latency" and the latency detected 2355 after the previous timestamp and the next timestamp in the loop is 2356 the "outer latency". 2357 2358 ts:1581527483.343962693 2359 2360 The absolute timestamp that the first latency was recorded in the window. 2361 2362 count:6 2363 2364 The number of times a latency was detected during the window. 2365 2366 nmi-total:7 nmi-count:1 2367 2368 On architectures that support it, if an NMI comes in during the 2369 test, the time spent in NMI is reported in "nmi-total" (in 2370 microseconds). 2371 2372 All architectures that have NMIs will show the "nmi-count" if an 2373 NMI comes in during the test. 2374 2375hwlat files: 2376 2377 tracing_threshold 2378 This gets automatically set to "10" to represent 10 2379 microseconds. This is the threshold of latency that 2380 needs to be detected before the trace will be recorded. 2381 2382 Note, when hwlat tracer is finished (another tracer is 2383 written into "current_tracer"), the original value for 2384 tracing_threshold is placed back into this file. 2385 2386 hwlat_detector/width 2387 The length of time the test runs with interrupts disabled. 2388 2389 hwlat_detector/window 2390 The length of time of the window which the test 2391 runs. That is, the test will run for "width" 2392 microseconds per "window" microseconds 2393 2394 tracing_cpumask 2395 When the test is started. A kernel thread is created that 2396 runs the test. This thread will alternate between CPUs 2397 listed in the tracing_cpumask between each period 2398 (one "window"). To limit the test to specific CPUs 2399 set the mask in this file to only the CPUs that the test 2400 should run on. 2401 2402function 2403-------- 2404 2405This tracer is the function tracer. Enabling the function tracer 2406can be done from the debug file system. Make sure the 2407ftrace_enabled is set; otherwise this tracer is a nop. 2408See the "ftrace_enabled" section below. 2409:: 2410 2411 # sysctl kernel.ftrace_enabled=1 2412 # echo function > current_tracer 2413 # echo 1 > tracing_on 2414 # usleep 1 2415 # echo 0 > tracing_on 2416 # cat trace 2417 # tracer: function 2418 # 2419 # entries-in-buffer/entries-written: 24799/24799 #P:4 2420 # 2421 # _-----=> irqs-off 2422 # / _----=> need-resched 2423 # | / _---=> hardirq/softirq 2424 # || / _--=> preempt-depth 2425 # ||| / delay 2426 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 2427 # | | | |||| | | 2428 bash-1994 [002] .... 3082.063030: mutex_unlock <-rb_simple_write 2429 bash-1994 [002] .... 3082.063031: __mutex_unlock_slowpath <-mutex_unlock 2430 bash-1994 [002] .... 3082.063031: __fsnotify_parent <-fsnotify_modify 2431 bash-1994 [002] .... 3082.063032: fsnotify <-fsnotify_modify 2432 bash-1994 [002] .... 3082.063032: __srcu_read_lock <-fsnotify 2433 bash-1994 [002] .... 3082.063032: add_preempt_count <-__srcu_read_lock 2434 bash-1994 [002] ...1 3082.063032: sub_preempt_count <-__srcu_read_lock 2435 bash-1994 [002] .... 3082.063033: __srcu_read_unlock <-fsnotify 2436 [...] 2437 2438 2439Note: function tracer uses ring buffers to store the above 2440entries. The newest data may overwrite the oldest data. 2441Sometimes using echo to stop the trace is not sufficient because 2442the tracing could have overwritten the data that you wanted to 2443record. For this reason, it is sometimes better to disable 2444tracing directly from a program. This allows you to stop the 2445tracing at the point that you hit the part that you are 2446interested in. To disable the tracing directly from a C program, 2447something like following code snippet can be used:: 2448 2449 int trace_fd; 2450 [...] 2451 int main(int argc, char *argv[]) { 2452 [...] 2453 trace_fd = open(tracing_file("tracing_on"), O_WRONLY); 2454 [...] 2455 if (condition_hit()) { 2456 write(trace_fd, "0", 1); 2457 } 2458 [...] 2459 } 2460 2461 2462Single thread tracing 2463--------------------- 2464 2465By writing into set_ftrace_pid you can trace a 2466single thread. For example:: 2467 2468 # cat set_ftrace_pid 2469 no pid 2470 # echo 3111 > set_ftrace_pid 2471 # cat set_ftrace_pid 2472 3111 2473 # echo function > current_tracer 2474 # cat trace | head 2475 # tracer: function 2476 # 2477 # TASK-PID CPU# TIMESTAMP FUNCTION 2478 # | | | | | 2479 yum-updatesd-3111 [003] 1637.254676: finish_task_switch <-thread_return 2480 yum-updatesd-3111 [003] 1637.254681: hrtimer_cancel <-schedule_hrtimeout_range 2481 yum-updatesd-3111 [003] 1637.254682: hrtimer_try_to_cancel <-hrtimer_cancel 2482 yum-updatesd-3111 [003] 1637.254683: lock_hrtimer_base <-hrtimer_try_to_cancel 2483 yum-updatesd-3111 [003] 1637.254685: fget_light <-do_sys_poll 2484 yum-updatesd-3111 [003] 1637.254686: pipe_poll <-do_sys_poll 2485 # echo > set_ftrace_pid 2486 # cat trace |head 2487 # tracer: function 2488 # 2489 # TASK-PID CPU# TIMESTAMP FUNCTION 2490 # | | | | | 2491 ##### CPU 3 buffer started #### 2492 yum-updatesd-3111 [003] 1701.957688: free_poll_entry <-poll_freewait 2493 yum-updatesd-3111 [003] 1701.957689: remove_wait_queue <-free_poll_entry 2494 yum-updatesd-3111 [003] 1701.957691: fput <-free_poll_entry 2495 yum-updatesd-3111 [003] 1701.957692: audit_syscall_exit <-sysret_audit 2496 yum-updatesd-3111 [003] 1701.957693: path_put <-audit_syscall_exit 2497 2498If you want to trace a function when executing, you could use 2499something like this simple program. 2500:: 2501 2502 #include <stdio.h> 2503 #include <stdlib.h> 2504 #include <sys/types.h> 2505 #include <sys/stat.h> 2506 #include <fcntl.h> 2507 #include <unistd.h> 2508 #include <string.h> 2509 2510 #define _STR(x) #x 2511 #define STR(x) _STR(x) 2512 #define MAX_PATH 256 2513 2514 const char *find_tracefs(void) 2515 { 2516 static char tracefs[MAX_PATH+1]; 2517 static int tracefs_found; 2518 char type[100]; 2519 FILE *fp; 2520 2521 if (tracefs_found) 2522 return tracefs; 2523 2524 if ((fp = fopen("/proc/mounts","r")) == NULL) { 2525 perror("/proc/mounts"); 2526 return NULL; 2527 } 2528 2529 while (fscanf(fp, "%*s %" 2530 STR(MAX_PATH) 2531 "s %99s %*s %*d %*d\n", 2532 tracefs, type) == 2) { 2533 if (strcmp(type, "tracefs") == 0) 2534 break; 2535 } 2536 fclose(fp); 2537 2538 if (strcmp(type, "tracefs") != 0) { 2539 fprintf(stderr, "tracefs not mounted"); 2540 return NULL; 2541 } 2542 2543 strcat(tracefs, "/tracing/"); 2544 tracefs_found = 1; 2545 2546 return tracefs; 2547 } 2548 2549 const char *tracing_file(const char *file_name) 2550 { 2551 static char trace_file[MAX_PATH+1]; 2552 snprintf(trace_file, MAX_PATH, "%s/%s", find_tracefs(), file_name); 2553 return trace_file; 2554 } 2555 2556 int main (int argc, char **argv) 2557 { 2558 if (argc < 1) 2559 exit(-1); 2560 2561 if (fork() > 0) { 2562 int fd, ffd; 2563 char line[64]; 2564 int s; 2565 2566 ffd = open(tracing_file("current_tracer"), O_WRONLY); 2567 if (ffd < 0) 2568 exit(-1); 2569 write(ffd, "nop", 3); 2570 2571 fd = open(tracing_file("set_ftrace_pid"), O_WRONLY); 2572 s = sprintf(line, "%d\n", getpid()); 2573 write(fd, line, s); 2574 2575 write(ffd, "function", 8); 2576 2577 close(fd); 2578 close(ffd); 2579 2580 execvp(argv[1], argv+1); 2581 } 2582 2583 return 0; 2584 } 2585 2586Or this simple script! 2587:: 2588 2589 #!/bin/bash 2590 2591 tracefs=`sed -ne 's/^tracefs \(.*\) tracefs.*/\1/p' /proc/mounts` 2592 echo 0 > $tracefs/tracing_on 2593 echo $$ > $tracefs/set_ftrace_pid 2594 echo function > $tracefs/current_tracer 2595 echo 1 > $tracefs/tracing_on 2596 exec "$@" 2597 2598 2599function graph tracer 2600--------------------------- 2601 2602This tracer is similar to the function tracer except that it 2603probes a function on its entry and its exit. This is done by 2604using a dynamically allocated stack of return addresses in each 2605task_struct. On function entry the tracer overwrites the return 2606address of each function traced to set a custom probe. Thus the 2607original return address is stored on the stack of return address 2608in the task_struct. 2609 2610Probing on both ends of a function leads to special features 2611such as: 2612 2613- measure of a function's time execution 2614- having a reliable call stack to draw function calls graph 2615 2616This tracer is useful in several situations: 2617 2618- you want to find the reason of a strange kernel behavior and 2619 need to see what happens in detail on any areas (or specific 2620 ones). 2621 2622- you are experiencing weird latencies but it's difficult to 2623 find its origin. 2624 2625- you want to find quickly which path is taken by a specific 2626 function 2627 2628- you just want to peek inside a working kernel and want to see 2629 what happens there. 2630 2631:: 2632 2633 # tracer: function_graph 2634 # 2635 # CPU DURATION FUNCTION CALLS 2636 # | | | | | | | 2637 2638 0) | sys_open() { 2639 0) | do_sys_open() { 2640 0) | getname() { 2641 0) | kmem_cache_alloc() { 2642 0) 1.382 us | __might_sleep(); 2643 0) 2.478 us | } 2644 0) | strncpy_from_user() { 2645 0) | might_fault() { 2646 0) 1.389 us | __might_sleep(); 2647 0) 2.553 us | } 2648 0) 3.807 us | } 2649 0) 7.876 us | } 2650 0) | alloc_fd() { 2651 0) 0.668 us | _spin_lock(); 2652 0) 0.570 us | expand_files(); 2653 0) 0.586 us | _spin_unlock(); 2654 2655 2656There are several columns that can be dynamically 2657enabled/disabled. You can use every combination of options you 2658want, depending on your needs. 2659 2660- The cpu number on which the function executed is default 2661 enabled. It is sometimes better to only trace one cpu (see 2662 tracing_cpumask file) or you might sometimes see unordered 2663 function calls while cpu tracing switch. 2664 2665 - hide: echo nofuncgraph-cpu > trace_options 2666 - show: echo funcgraph-cpu > trace_options 2667 2668- The duration (function's time of execution) is displayed on 2669 the closing bracket line of a function or on the same line 2670 than the current function in case of a leaf one. It is default 2671 enabled. 2672 2673 - hide: echo nofuncgraph-duration > trace_options 2674 - show: echo funcgraph-duration > trace_options 2675 2676- The overhead field precedes the duration field in case of 2677 reached duration thresholds. 2678 2679 - hide: echo nofuncgraph-overhead > trace_options 2680 - show: echo funcgraph-overhead > trace_options 2681 - depends on: funcgraph-duration 2682 2683 ie:: 2684 2685 3) # 1837.709 us | } /* __switch_to */ 2686 3) | finish_task_switch() { 2687 3) 0.313 us | _raw_spin_unlock_irq(); 2688 3) 3.177 us | } 2689 3) # 1889.063 us | } /* __schedule */ 2690 3) ! 140.417 us | } /* __schedule */ 2691 3) # 2034.948 us | } /* schedule */ 2692 3) * 33998.59 us | } /* schedule_preempt_disabled */ 2693 2694 [...] 2695 2696 1) 0.260 us | msecs_to_jiffies(); 2697 1) 0.313 us | __rcu_read_unlock(); 2698 1) + 61.770 us | } 2699 1) + 64.479 us | } 2700 1) 0.313 us | rcu_bh_qs(); 2701 1) 0.313 us | __local_bh_enable(); 2702 1) ! 217.240 us | } 2703 1) 0.365 us | idle_cpu(); 2704 1) | rcu_irq_exit() { 2705 1) 0.417 us | rcu_eqs_enter_common.isra.47(); 2706 1) 3.125 us | } 2707 1) ! 227.812 us | } 2708 1) ! 457.395 us | } 2709 1) @ 119760.2 us | } 2710 2711 [...] 2712 2713 2) | handle_IPI() { 2714 1) 6.979 us | } 2715 2) 0.417 us | scheduler_ipi(); 2716 1) 9.791 us | } 2717 1) + 12.917 us | } 2718 2) 3.490 us | } 2719 1) + 15.729 us | } 2720 1) + 18.542 us | } 2721 2) $ 3594274 us | } 2722 2723Flags:: 2724 2725 + means that the function exceeded 10 usecs. 2726 ! means that the function exceeded 100 usecs. 2727 # means that the function exceeded 1000 usecs. 2728 * means that the function exceeded 10 msecs. 2729 @ means that the function exceeded 100 msecs. 2730 $ means that the function exceeded 1 sec. 2731 2732 2733- The task/pid field displays the thread cmdline and pid which 2734 executed the function. It is default disabled. 2735 2736 - hide: echo nofuncgraph-proc > trace_options 2737 - show: echo funcgraph-proc > trace_options 2738 2739 ie:: 2740 2741 # tracer: function_graph 2742 # 2743 # CPU TASK/PID DURATION FUNCTION CALLS 2744 # | | | | | | | | | 2745 0) sh-4802 | | d_free() { 2746 0) sh-4802 | | call_rcu() { 2747 0) sh-4802 | | __call_rcu() { 2748 0) sh-4802 | 0.616 us | rcu_process_gp_end(); 2749 0) sh-4802 | 0.586 us | check_for_new_grace_period(); 2750 0) sh-4802 | 2.899 us | } 2751 0) sh-4802 | 4.040 us | } 2752 0) sh-4802 | 5.151 us | } 2753 0) sh-4802 | + 49.370 us | } 2754 2755 2756- The absolute time field is an absolute timestamp given by the 2757 system clock since it started. A snapshot of this time is 2758 given on each entry/exit of functions 2759 2760 - hide: echo nofuncgraph-abstime > trace_options 2761 - show: echo funcgraph-abstime > trace_options 2762 2763 ie:: 2764 2765 # 2766 # TIME CPU DURATION FUNCTION CALLS 2767 # | | | | | | | | 2768 360.774522 | 1) 0.541 us | } 2769 360.774522 | 1) 4.663 us | } 2770 360.774523 | 1) 0.541 us | __wake_up_bit(); 2771 360.774524 | 1) 6.796 us | } 2772 360.774524 | 1) 7.952 us | } 2773 360.774525 | 1) 9.063 us | } 2774 360.774525 | 1) 0.615 us | journal_mark_dirty(); 2775 360.774527 | 1) 0.578 us | __brelse(); 2776 360.774528 | 1) | reiserfs_prepare_for_journal() { 2777 360.774528 | 1) | unlock_buffer() { 2778 360.774529 | 1) | wake_up_bit() { 2779 360.774529 | 1) | bit_waitqueue() { 2780 360.774530 | 1) 0.594 us | __phys_addr(); 2781 2782 2783The function name is always displayed after the closing bracket 2784for a function if the start of that function is not in the 2785trace buffer. 2786 2787Display of the function name after the closing bracket may be 2788enabled for functions whose start is in the trace buffer, 2789allowing easier searching with grep for function durations. 2790It is default disabled. 2791 2792 - hide: echo nofuncgraph-tail > trace_options 2793 - show: echo funcgraph-tail > trace_options 2794 2795 Example with nofuncgraph-tail (default):: 2796 2797 0) | putname() { 2798 0) | kmem_cache_free() { 2799 0) 0.518 us | __phys_addr(); 2800 0) 1.757 us | } 2801 0) 2.861 us | } 2802 2803 Example with funcgraph-tail:: 2804 2805 0) | putname() { 2806 0) | kmem_cache_free() { 2807 0) 0.518 us | __phys_addr(); 2808 0) 1.757 us | } /* kmem_cache_free() */ 2809 0) 2.861 us | } /* putname() */ 2810 2811The return value of each traced function can be displayed after 2812an equal sign "=". When encountering system call failures, it 2813can be very helpful to quickly locate the function that first 2814returns an error code. 2815 2816 - hide: echo nofuncgraph-retval > trace_options 2817 - show: echo funcgraph-retval > trace_options 2818 2819 Example with funcgraph-retval:: 2820 2821 1) | cgroup_migrate() { 2822 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */ 2823 1) | cgroup_migrate_execute() { 2824 1) | cpu_cgroup_can_attach() { 2825 1) | cgroup_taskset_first() { 2826 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */ 2827 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */ 2828 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */ 2829 1) 2.335 us | } /* cpu_cgroup_can_attach = -22 */ 2830 1) 4.369 us | } /* cgroup_migrate_execute = -22 */ 2831 1) 7.143 us | } /* cgroup_migrate = -22 */ 2832 2833The above example shows that the function cpu_cgroup_can_attach 2834returned the error code -22 firstly, then we can read the code 2835of this function to get the root cause. 2836 2837When the option funcgraph-retval-hex is not set, the return value can 2838be displayed in a smart way. Specifically, if it is an error code, 2839it will be printed in signed decimal format, otherwise it will 2840printed in hexadecimal format. 2841 2842 - smart: echo nofuncgraph-retval-hex > trace_options 2843 - hexadecimal: echo funcgraph-retval-hex > trace_options 2844 2845 Example with funcgraph-retval-hex:: 2846 2847 1) | cgroup_migrate() { 2848 1) 0.651 us | cgroup_migrate_add_task(); /* = 0xffff93fcfd346c00 */ 2849 1) | cgroup_migrate_execute() { 2850 1) | cpu_cgroup_can_attach() { 2851 1) | cgroup_taskset_first() { 2852 1) 0.732 us | cgroup_taskset_next(); /* = 0xffff93fc8fb20000 */ 2853 1) 1.232 us | } /* cgroup_taskset_first = 0xffff93fc8fb20000 */ 2854 1) 0.380 us | sched_rt_can_attach(); /* = 0x0 */ 2855 1) 2.335 us | } /* cpu_cgroup_can_attach = 0xffffffea */ 2856 1) 4.369 us | } /* cgroup_migrate_execute = 0xffffffea */ 2857 1) 7.143 us | } /* cgroup_migrate = 0xffffffea */ 2858 2859At present, there are some limitations when using the funcgraph-retval 2860option, and these limitations will be eliminated in the future: 2861 2862- Even if the function return type is void, a return value will still 2863 be printed, and you can just ignore it. 2864 2865- Even if return values are stored in multiple registers, only the 2866 value contained in the first register will be recorded and printed. 2867 To illustrate, in the x86 architecture, eax and edx are used to store 2868 a 64-bit return value, with the lower 32 bits saved in eax and the 2869 upper 32 bits saved in edx. However, only the value stored in eax 2870 will be recorded and printed. 2871 2872- In certain procedure call standards, such as arm64's AAPCS64, when a 2873 type is smaller than a GPR, it is the responsibility of the consumer 2874 to perform the narrowing, and the upper bits may contain UNKNOWN values. 2875 Therefore, it is advisable to check the code for such cases. For instance, 2876 when using a u8 in a 64-bit GPR, bits [63:8] may contain arbitrary values, 2877 especially when larger types are truncated, whether explicitly or implicitly. 2878 Here are some specific cases to illustrate this point: 2879 2880 **Case One**: 2881 2882 The function narrow_to_u8 is defined as follows:: 2883 2884 u8 narrow_to_u8(u64 val) 2885 { 2886 // implicitly truncated 2887 return val; 2888 } 2889 2890 It may be compiled to:: 2891 2892 narrow_to_u8: 2893 < ... ftrace instrumentation ... > 2894 RET 2895 2896 If you pass 0x123456789abcdef to this function and want to narrow it, 2897 it may be recorded as 0x123456789abcdef instead of 0xef. 2898 2899 **Case Two**: 2900 2901 The function error_if_not_4g_aligned is defined as follows:: 2902 2903 int error_if_not_4g_aligned(u64 val) 2904 { 2905 if (val & GENMASK(31, 0)) 2906 return -EINVAL; 2907 2908 return 0; 2909 } 2910 2911 It could be compiled to:: 2912 2913 error_if_not_4g_aligned: 2914 CBNZ w0, .Lnot_aligned 2915 RET // bits [31:0] are zero, bits 2916 // [63:32] are UNKNOWN 2917 .Lnot_aligned: 2918 MOV x0, #-EINVAL 2919 RET 2920 2921 When passing 0x2_0000_0000 to it, the return value may be recorded as 2922 0x2_0000_0000 instead of 0. 2923 2924You can put some comments on specific functions by using 2925trace_printk() For example, if you want to put a comment inside 2926the __might_sleep() function, you just have to include 2927<linux/ftrace.h> and call trace_printk() inside __might_sleep():: 2928 2929 trace_printk("I'm a comment!\n") 2930 2931will produce:: 2932 2933 1) | __might_sleep() { 2934 1) | /* I'm a comment! */ 2935 1) 1.449 us | } 2936 2937 2938You might find other useful features for this tracer in the 2939following "dynamic ftrace" section such as tracing only specific 2940functions or tasks. 2941 2942dynamic ftrace 2943-------------- 2944 2945If CONFIG_DYNAMIC_FTRACE is set, the system will run with 2946virtually no overhead when function tracing is disabled. The way 2947this works is the mcount function call (placed at the start of 2948every kernel function, produced by the -pg switch in gcc), 2949starts of pointing to a simple return. (Enabling FTRACE will 2950include the -pg switch in the compiling of the kernel.) 2951 2952At compile time every C file object is run through the 2953recordmcount program (located in the scripts directory). This 2954program will parse the ELF headers in the C object to find all 2955the locations in the .text section that call mcount. Starting 2956with gcc version 4.6, the -mfentry has been added for x86, which 2957calls "__fentry__" instead of "mcount". Which is called before 2958the creation of the stack frame. 2959 2960Note, not all sections are traced. They may be prevented by either 2961a notrace, or blocked another way and all inline functions are not 2962traced. Check the "available_filter_functions" file to see what functions 2963can be traced. 2964 2965A section called "__mcount_loc" is created that holds 2966references to all the mcount/fentry call sites in the .text section. 2967The recordmcount program re-links this section back into the 2968original object. The final linking stage of the kernel will add all these 2969references into a single table. 2970 2971On boot up, before SMP is initialized, the dynamic ftrace code 2972scans this table and updates all the locations into nops. It 2973also records the locations, which are added to the 2974available_filter_functions list. Modules are processed as they 2975are loaded and before they are executed. When a module is 2976unloaded, it also removes its functions from the ftrace function 2977list. This is automatic in the module unload code, and the 2978module author does not need to worry about it. 2979 2980When tracing is enabled, the process of modifying the function 2981tracepoints is dependent on architecture. The old method is to use 2982kstop_machine to prevent races with the CPUs executing code being 2983modified (which can cause the CPU to do undesirable things, especially 2984if the modified code crosses cache (or page) boundaries), and the nops are 2985patched back to calls. But this time, they do not call mcount 2986(which is just a function stub). They now call into the ftrace 2987infrastructure. 2988 2989The new method of modifying the function tracepoints is to place 2990a breakpoint at the location to be modified, sync all CPUs, modify 2991the rest of the instruction not covered by the breakpoint. Sync 2992all CPUs again, and then remove the breakpoint with the finished 2993version to the ftrace call site. 2994 2995Some archs do not even need to monkey around with the synchronization, 2996and can just slap the new code on top of the old without any 2997problems with other CPUs executing it at the same time. 2998 2999One special side-effect to the recording of the functions being 3000traced is that we can now selectively choose which functions we 3001wish to trace and which ones we want the mcount calls to remain 3002as nops. 3003 3004Two files are used, one for enabling and one for disabling the 3005tracing of specified functions. They are: 3006 3007 set_ftrace_filter 3008 3009and 3010 3011 set_ftrace_notrace 3012 3013A list of available functions that you can add to these files is 3014listed in: 3015 3016 available_filter_functions 3017 3018:: 3019 3020 # cat available_filter_functions 3021 put_prev_task_idle 3022 kmem_cache_create 3023 pick_next_task_rt 3024 cpus_read_lock 3025 pick_next_task_fair 3026 mutex_lock 3027 [...] 3028 3029If I am only interested in sys_nanosleep and hrtimer_interrupt:: 3030 3031 # echo sys_nanosleep hrtimer_interrupt > set_ftrace_filter 3032 # echo function > current_tracer 3033 # echo 1 > tracing_on 3034 # usleep 1 3035 # echo 0 > tracing_on 3036 # cat trace 3037 # tracer: function 3038 # 3039 # entries-in-buffer/entries-written: 5/5 #P:4 3040 # 3041 # _-----=> irqs-off 3042 # / _----=> need-resched 3043 # | / _---=> hardirq/softirq 3044 # || / _--=> preempt-depth 3045 # ||| / delay 3046 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3047 # | | | |||| | | 3048 usleep-2665 [001] .... 4186.475355: sys_nanosleep <-system_call_fastpath 3049 <idle>-0 [001] d.h1 4186.475409: hrtimer_interrupt <-smp_apic_timer_interrupt 3050 usleep-2665 [001] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt 3051 <idle>-0 [003] d.h1 4186.475426: hrtimer_interrupt <-smp_apic_timer_interrupt 3052 <idle>-0 [002] d.h1 4186.475427: hrtimer_interrupt <-smp_apic_timer_interrupt 3053 3054To see which functions are being traced, you can cat the file: 3055:: 3056 3057 # cat set_ftrace_filter 3058 hrtimer_interrupt 3059 sys_nanosleep 3060 3061 3062Perhaps this is not enough. The filters also allow glob(7) matching. 3063 3064 ``<match>*`` 3065 will match functions that begin with <match> 3066 ``*<match>`` 3067 will match functions that end with <match> 3068 ``*<match>*`` 3069 will match functions that have <match> in it 3070 ``<match1>*<match2>`` 3071 will match functions that begin with <match1> and end with <match2> 3072 3073.. note:: 3074 It is better to use quotes to enclose the wild cards, 3075 otherwise the shell may expand the parameters into names 3076 of files in the local directory. 3077 3078:: 3079 3080 # echo 'hrtimer_*' > set_ftrace_filter 3081 3082Produces:: 3083 3084 # tracer: function 3085 # 3086 # entries-in-buffer/entries-written: 897/897 #P:4 3087 # 3088 # _-----=> irqs-off 3089 # / _----=> need-resched 3090 # | / _---=> hardirq/softirq 3091 # || / _--=> preempt-depth 3092 # ||| / delay 3093 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3094 # | | | |||| | | 3095 <idle>-0 [003] dN.1 4228.547803: hrtimer_cancel <-tick_nohz_idle_exit 3096 <idle>-0 [003] dN.1 4228.547804: hrtimer_try_to_cancel <-hrtimer_cancel 3097 <idle>-0 [003] dN.2 4228.547805: hrtimer_force_reprogram <-__remove_hrtimer 3098 <idle>-0 [003] dN.1 4228.547805: hrtimer_forward <-tick_nohz_idle_exit 3099 <idle>-0 [003] dN.1 4228.547805: hrtimer_start_range_ns <-hrtimer_start_expires.constprop.11 3100 <idle>-0 [003] d..1 4228.547858: hrtimer_get_next_event <-get_next_timer_interrupt 3101 <idle>-0 [003] d..1 4228.547859: hrtimer_start <-__tick_nohz_idle_enter 3102 <idle>-0 [003] d..2 4228.547860: hrtimer_force_reprogram <-__rem 3103 3104Notice that we lost the sys_nanosleep. 3105:: 3106 3107 # cat set_ftrace_filter 3108 hrtimer_run_queues 3109 hrtimer_run_pending 3110 hrtimer_setup 3111 hrtimer_cancel 3112 hrtimer_try_to_cancel 3113 hrtimer_forward 3114 hrtimer_start 3115 hrtimer_reprogram 3116 hrtimer_force_reprogram 3117 hrtimer_get_next_event 3118 hrtimer_interrupt 3119 hrtimer_nanosleep 3120 hrtimer_wakeup 3121 hrtimer_get_remaining 3122 hrtimer_get_res 3123 hrtimer_init_sleeper 3124 3125 3126This is because the '>' and '>>' act just like they do in bash. 3127To rewrite the filters, use '>' 3128To append to the filters, use '>>' 3129 3130To clear out a filter so that all functions will be recorded 3131again:: 3132 3133 # echo > set_ftrace_filter 3134 # cat set_ftrace_filter 3135 # 3136 3137Again, now we want to append. 3138 3139:: 3140 3141 # echo sys_nanosleep > set_ftrace_filter 3142 # cat set_ftrace_filter 3143 sys_nanosleep 3144 # echo 'hrtimer_*' >> set_ftrace_filter 3145 # cat set_ftrace_filter 3146 hrtimer_run_queues 3147 hrtimer_run_pending 3148 hrtimer_setup 3149 hrtimer_cancel 3150 hrtimer_try_to_cancel 3151 hrtimer_forward 3152 hrtimer_start 3153 hrtimer_reprogram 3154 hrtimer_force_reprogram 3155 hrtimer_get_next_event 3156 hrtimer_interrupt 3157 sys_nanosleep 3158 hrtimer_nanosleep 3159 hrtimer_wakeup 3160 hrtimer_get_remaining 3161 hrtimer_get_res 3162 hrtimer_init_sleeper 3163 3164 3165The set_ftrace_notrace prevents those functions from being 3166traced. 3167:: 3168 3169 # echo '*preempt*' '*lock*' > set_ftrace_notrace 3170 3171Produces:: 3172 3173 # tracer: function 3174 # 3175 # entries-in-buffer/entries-written: 39608/39608 #P:4 3176 # 3177 # _-----=> irqs-off 3178 # / _----=> need-resched 3179 # | / _---=> hardirq/softirq 3180 # || / _--=> preempt-depth 3181 # ||| / delay 3182 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3183 # | | | |||| | | 3184 bash-1994 [000] .... 4342.324896: file_ra_state_init <-do_dentry_open 3185 bash-1994 [000] .... 4342.324897: open_check_o_direct <-do_last 3186 bash-1994 [000] .... 4342.324897: ima_file_check <-do_last 3187 bash-1994 [000] .... 4342.324898: process_measurement <-ima_file_check 3188 bash-1994 [000] .... 4342.324898: ima_get_action <-process_measurement 3189 bash-1994 [000] .... 4342.324898: ima_match_policy <-ima_get_action 3190 bash-1994 [000] .... 4342.324899: do_truncate <-do_last 3191 bash-1994 [000] .... 4342.324899: setattr_should_drop_suidgid <-do_truncate 3192 bash-1994 [000] .... 4342.324899: notify_change <-do_truncate 3193 bash-1994 [000] .... 4342.324900: current_fs_time <-notify_change 3194 bash-1994 [000] .... 4342.324900: current_kernel_time <-current_fs_time 3195 bash-1994 [000] .... 4342.324900: timespec_trunc <-current_fs_time 3196 3197We can see that there's no more lock or preempt tracing. 3198 3199Selecting function filters via index 3200------------------------------------ 3201 3202Because processing of strings is expensive (the address of the function 3203needs to be looked up before comparing to the string being passed in), 3204an index can be used as well to enable functions. This is useful in the 3205case of setting thousands of specific functions at a time. By passing 3206in a list of numbers, no string processing will occur. Instead, the function 3207at the specific location in the internal array (which corresponds to the 3208functions in the "available_filter_functions" file), is selected. 3209 3210:: 3211 3212 # echo 1 > set_ftrace_filter 3213 3214Will select the first function listed in "available_filter_functions" 3215 3216:: 3217 3218 # head -1 available_filter_functions 3219 trace_initcall_finish_cb 3220 3221 # cat set_ftrace_filter 3222 trace_initcall_finish_cb 3223 3224 # head -50 available_filter_functions | tail -1 3225 x86_pmu_commit_txn 3226 3227 # echo 1 50 > set_ftrace_filter 3228 # cat set_ftrace_filter 3229 trace_initcall_finish_cb 3230 x86_pmu_commit_txn 3231 3232Dynamic ftrace with the function graph tracer 3233--------------------------------------------- 3234 3235Although what has been explained above concerns both the 3236function tracer and the function-graph-tracer, there are some 3237special features only available in the function-graph tracer. 3238 3239If you want to trace only one function and all of its children, 3240you just have to echo its name into set_graph_function:: 3241 3242 echo __do_fault > set_graph_function 3243 3244will produce the following "expanded" trace of the __do_fault() 3245function:: 3246 3247 0) | __do_fault() { 3248 0) | filemap_fault() { 3249 0) | find_lock_page() { 3250 0) 0.804 us | find_get_page(); 3251 0) | __might_sleep() { 3252 0) 1.329 us | } 3253 0) 3.904 us | } 3254 0) 4.979 us | } 3255 0) 0.653 us | _spin_lock(); 3256 0) 0.578 us | page_add_file_rmap(); 3257 0) 0.525 us | native_set_pte_at(); 3258 0) 0.585 us | _spin_unlock(); 3259 0) | unlock_page() { 3260 0) 0.541 us | page_waitqueue(); 3261 0) 0.639 us | __wake_up_bit(); 3262 0) 2.786 us | } 3263 0) + 14.237 us | } 3264 0) | __do_fault() { 3265 0) | filemap_fault() { 3266 0) | find_lock_page() { 3267 0) 0.698 us | find_get_page(); 3268 0) | __might_sleep() { 3269 0) 1.412 us | } 3270 0) 3.950 us | } 3271 0) 5.098 us | } 3272 0) 0.631 us | _spin_lock(); 3273 0) 0.571 us | page_add_file_rmap(); 3274 0) 0.526 us | native_set_pte_at(); 3275 0) 0.586 us | _spin_unlock(); 3276 0) | unlock_page() { 3277 0) 0.533 us | page_waitqueue(); 3278 0) 0.638 us | __wake_up_bit(); 3279 0) 2.793 us | } 3280 0) + 14.012 us | } 3281 3282You can also expand several functions at once:: 3283 3284 echo sys_open > set_graph_function 3285 echo sys_close >> set_graph_function 3286 3287Now if you want to go back to trace all functions you can clear 3288this special filter via:: 3289 3290 echo > set_graph_function 3291 3292 3293ftrace_enabled 3294-------------- 3295 3296Note, the proc sysctl ftrace_enable is a big on/off switch for the 3297function tracer. By default it is enabled (when function tracing is 3298enabled in the kernel). If it is disabled, all function tracing is 3299disabled. This includes not only the function tracers for ftrace, but 3300also for any other uses (perf, kprobes, stack tracing, profiling, etc). It 3301cannot be disabled if there is a callback with FTRACE_OPS_FL_PERMANENT set 3302registered. 3303 3304Please disable this with care. 3305 3306This can be disable (and enabled) with:: 3307 3308 sysctl kernel.ftrace_enabled=0 3309 sysctl kernel.ftrace_enabled=1 3310 3311 or 3312 3313 echo 0 > /proc/sys/kernel/ftrace_enabled 3314 echo 1 > /proc/sys/kernel/ftrace_enabled 3315 3316 3317Filter commands 3318--------------- 3319 3320A few commands are supported by the set_ftrace_filter interface. 3321Trace commands have the following format:: 3322 3323 <function>:<command>:<parameter> 3324 3325The following commands are supported: 3326 3327- mod: 3328 This command enables function filtering per module. The 3329 parameter defines the module. For example, if only the write* 3330 functions in the ext3 module are desired, run: 3331 3332 echo 'write*:mod:ext3' > set_ftrace_filter 3333 3334 This command interacts with the filter in the same way as 3335 filtering based on function names. Thus, adding more functions 3336 in a different module is accomplished by appending (>>) to the 3337 filter file. Remove specific module functions by prepending 3338 '!':: 3339 3340 echo '!writeback*:mod:ext3' >> set_ftrace_filter 3341 3342 Mod command supports module globbing. Disable tracing for all 3343 functions except a specific module:: 3344 3345 echo '!*:mod:!ext3' >> set_ftrace_filter 3346 3347 Disable tracing for all modules, but still trace kernel:: 3348 3349 echo '!*:mod:*' >> set_ftrace_filter 3350 3351 Enable filter only for kernel:: 3352 3353 echo '*write*:mod:!*' >> set_ftrace_filter 3354 3355 Enable filter for module globbing:: 3356 3357 echo '*write*:mod:*snd*' >> set_ftrace_filter 3358 3359- traceon/traceoff: 3360 These commands turn tracing on and off when the specified 3361 functions are hit. The parameter determines how many times the 3362 tracing system is turned on and off. If unspecified, there is 3363 no limit. For example, to disable tracing when a schedule bug 3364 is hit the first 5 times, run:: 3365 3366 echo '__schedule_bug:traceoff:5' > set_ftrace_filter 3367 3368 To always disable tracing when __schedule_bug is hit:: 3369 3370 echo '__schedule_bug:traceoff' > set_ftrace_filter 3371 3372 These commands are cumulative whether or not they are appended 3373 to set_ftrace_filter. To remove a command, prepend it by '!' 3374 and drop the parameter:: 3375 3376 echo '!__schedule_bug:traceoff:0' > set_ftrace_filter 3377 3378 The above removes the traceoff command for __schedule_bug 3379 that have a counter. To remove commands without counters:: 3380 3381 echo '!__schedule_bug:traceoff' > set_ftrace_filter 3382 3383- snapshot: 3384 Will cause a snapshot to be triggered when the function is hit. 3385 :: 3386 3387 echo 'native_flush_tlb_others:snapshot' > set_ftrace_filter 3388 3389 To only snapshot once: 3390 :: 3391 3392 echo 'native_flush_tlb_others:snapshot:1' > set_ftrace_filter 3393 3394 To remove the above commands:: 3395 3396 echo '!native_flush_tlb_others:snapshot' > set_ftrace_filter 3397 echo '!native_flush_tlb_others:snapshot:0' > set_ftrace_filter 3398 3399- enable_event/disable_event: 3400 These commands can enable or disable a trace event. Note, because 3401 function tracing callbacks are very sensitive, when these commands 3402 are registered, the trace point is activated, but disabled in 3403 a "soft" mode. That is, the tracepoint will be called, but 3404 just will not be traced. The event tracepoint stays in this mode 3405 as long as there's a command that triggers it. 3406 :: 3407 3408 echo 'try_to_wake_up:enable_event:sched:sched_switch:2' > \ 3409 set_ftrace_filter 3410 3411 The format is:: 3412 3413 <function>:enable_event:<system>:<event>[:count] 3414 <function>:disable_event:<system>:<event>[:count] 3415 3416 To remove the events commands:: 3417 3418 echo '!try_to_wake_up:enable_event:sched:sched_switch:0' > \ 3419 set_ftrace_filter 3420 echo '!schedule:disable_event:sched:sched_switch' > \ 3421 set_ftrace_filter 3422 3423- dump: 3424 When the function is hit, it will dump the contents of the ftrace 3425 ring buffer to the console. This is useful if you need to debug 3426 something, and want to dump the trace when a certain function 3427 is hit. Perhaps it's a function that is called before a triple 3428 fault happens and does not allow you to get a regular dump. 3429 3430- cpudump: 3431 When the function is hit, it will dump the contents of the ftrace 3432 ring buffer for the current CPU to the console. Unlike the "dump" 3433 command, it only prints out the contents of the ring buffer for the 3434 CPU that executed the function that triggered the dump. 3435 3436- stacktrace: 3437 When the function is hit, a stack trace is recorded. 3438 3439trace_pipe 3440---------- 3441 3442The trace_pipe outputs the same content as the trace file, but 3443the effect on the tracing is different. Every read from 3444trace_pipe is consumed. This means that subsequent reads will be 3445different. The trace is live. 3446:: 3447 3448 # echo function > current_tracer 3449 # cat trace_pipe > /tmp/trace.out & 3450 [1] 4153 3451 # echo 1 > tracing_on 3452 # usleep 1 3453 # echo 0 > tracing_on 3454 # cat trace 3455 # tracer: function 3456 # 3457 # entries-in-buffer/entries-written: 0/0 #P:4 3458 # 3459 # _-----=> irqs-off 3460 # / _----=> need-resched 3461 # | / _---=> hardirq/softirq 3462 # || / _--=> preempt-depth 3463 # ||| / delay 3464 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3465 # | | | |||| | | 3466 3467 # 3468 # cat /tmp/trace.out 3469 bash-1994 [000] .... 5281.568961: mutex_unlock <-rb_simple_write 3470 bash-1994 [000] .... 5281.568963: __mutex_unlock_slowpath <-mutex_unlock 3471 bash-1994 [000] .... 5281.568963: __fsnotify_parent <-fsnotify_modify 3472 bash-1994 [000] .... 5281.568964: fsnotify <-fsnotify_modify 3473 bash-1994 [000] .... 5281.568964: __srcu_read_lock <-fsnotify 3474 bash-1994 [000] .... 5281.568964: add_preempt_count <-__srcu_read_lock 3475 bash-1994 [000] ...1 5281.568965: sub_preempt_count <-__srcu_read_lock 3476 bash-1994 [000] .... 5281.568965: __srcu_read_unlock <-fsnotify 3477 bash-1994 [000] .... 5281.568967: sys_dup2 <-system_call_fastpath 3478 3479 3480Note, reading the trace_pipe file will block until more input is 3481added. This is contrary to the trace file. If any process opened 3482the trace file for reading, it will actually disable tracing and 3483prevent new entries from being added. The trace_pipe file does 3484not have this limitation. 3485 3486trace entries 3487------------- 3488 3489Having too much or not enough data can be troublesome in 3490diagnosing an issue in the kernel. The file buffer_size_kb is 3491used to modify the size of the internal trace buffers. The 3492number listed is the number of entries that can be recorded per 3493CPU. To know the full size, multiply the number of possible CPUs 3494with the number of entries. 3495:: 3496 3497 # cat buffer_size_kb 3498 1408 (units kilobytes) 3499 3500Or simply read buffer_total_size_kb 3501:: 3502 3503 # cat buffer_total_size_kb 3504 5632 3505 3506To modify the buffer, simple echo in a number (in 1024 byte segments). 3507:: 3508 3509 # echo 10000 > buffer_size_kb 3510 # cat buffer_size_kb 3511 10000 (units kilobytes) 3512 3513It will try to allocate as much as possible. If you allocate too 3514much, it can cause Out-Of-Memory to trigger. 3515:: 3516 3517 # echo 1000000000000 > buffer_size_kb 3518 -bash: echo: write error: Cannot allocate memory 3519 # cat buffer_size_kb 3520 85 3521 3522The per_cpu buffers can be changed individually as well: 3523:: 3524 3525 # echo 10000 > per_cpu/cpu0/buffer_size_kb 3526 # echo 100 > per_cpu/cpu1/buffer_size_kb 3527 3528When the per_cpu buffers are not the same, the buffer_size_kb 3529at the top level will just show an X 3530:: 3531 3532 # cat buffer_size_kb 3533 X 3534 3535This is where the buffer_total_size_kb is useful: 3536:: 3537 3538 # cat buffer_total_size_kb 3539 12916 3540 3541Writing to the top level buffer_size_kb will reset all the buffers 3542to be the same again. 3543 3544Snapshot 3545-------- 3546CONFIG_TRACER_SNAPSHOT makes a generic snapshot feature 3547available to all non latency tracers. (Latency tracers which 3548record max latency, such as "irqsoff" or "wakeup", can't use 3549this feature, since those are already using the snapshot 3550mechanism internally.) 3551 3552Snapshot preserves a current trace buffer at a particular point 3553in time without stopping tracing. Ftrace swaps the current 3554buffer with a spare buffer, and tracing continues in the new 3555current (=previous spare) buffer. 3556 3557The following tracefs files in "tracing" are related to this 3558feature: 3559 3560 snapshot: 3561 3562 This is used to take a snapshot and to read the output 3563 of the snapshot. Echo 1 into this file to allocate a 3564 spare buffer and to take a snapshot (swap), then read 3565 the snapshot from this file in the same format as 3566 "trace" (described above in the section "The File 3567 System"). Both reads snapshot and tracing are executable 3568 in parallel. When the spare buffer is allocated, echoing 3569 0 frees it, and echoing else (positive) values clear the 3570 snapshot contents. 3571 More details are shown in the table below. 3572 3573 +--------------+------------+------------+------------+ 3574 |status\\input | 0 | 1 | else | 3575 +==============+============+============+============+ 3576 |not allocated |(do nothing)| alloc+swap |(do nothing)| 3577 +--------------+------------+------------+------------+ 3578 |allocated | free | swap | clear | 3579 +--------------+------------+------------+------------+ 3580 3581Here is an example of using the snapshot feature. 3582:: 3583 3584 # echo 1 > events/sched/enable 3585 # echo 1 > snapshot 3586 # cat snapshot 3587 # tracer: nop 3588 # 3589 # entries-in-buffer/entries-written: 71/71 #P:8 3590 # 3591 # _-----=> irqs-off 3592 # / _----=> need-resched 3593 # | / _---=> hardirq/softirq 3594 # || / _--=> preempt-depth 3595 # ||| / delay 3596 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3597 # | | | |||| | | 3598 <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 3599 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 3600 [...] 3601 <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 3602 3603 # cat trace 3604 # tracer: nop 3605 # 3606 # entries-in-buffer/entries-written: 77/77 #P:8 3607 # 3608 # _-----=> irqs-off 3609 # / _----=> need-resched 3610 # | / _---=> hardirq/softirq 3611 # || / _--=> preempt-depth 3612 # ||| / delay 3613 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3614 # | | | |||| | | 3615 <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 3616 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 3617 [...] 3618 3619 3620If you try to use this snapshot feature when current tracer is 3621one of the latency tracers, you will get the following results. 3622:: 3623 3624 # echo wakeup > current_tracer 3625 # echo 1 > snapshot 3626 bash: echo: write error: Device or resource busy 3627 # cat snapshot 3628 cat: snapshot: Device or resource busy 3629 3630 3631Instances 3632--------- 3633In the tracefs tracing directory, there is a directory called "instances". 3634This directory can have new directories created inside of it using 3635mkdir, and removing directories with rmdir. The directory created 3636with mkdir in this directory will already contain files and other 3637directories after it is created. 3638:: 3639 3640 # mkdir instances/foo 3641 # ls instances/foo 3642 buffer_size_kb buffer_total_size_kb events free_buffer per_cpu 3643 set_event snapshot trace trace_clock trace_marker trace_options 3644 trace_pipe tracing_on 3645 3646As you can see, the new directory looks similar to the tracing directory 3647itself. In fact, it is very similar, except that the buffer and 3648events are agnostic from the main directory, or from any other 3649instances that are created. 3650 3651The files in the new directory work just like the files with the 3652same name in the tracing directory except the buffer that is used 3653is a separate and new buffer. The files affect that buffer but do not 3654affect the main buffer with the exception of trace_options. Currently, 3655the trace_options affect all instances and the top level buffer 3656the same, but this may change in future releases. That is, options 3657may become specific to the instance they reside in. 3658 3659Notice that none of the function tracer files are there, nor is 3660current_tracer and available_tracers. This is because the buffers 3661can currently only have events enabled for them. 3662:: 3663 3664 # mkdir instances/foo 3665 # mkdir instances/bar 3666 # mkdir instances/zoot 3667 # echo 100000 > buffer_size_kb 3668 # echo 1000 > instances/foo/buffer_size_kb 3669 # echo 5000 > instances/bar/per_cpu/cpu1/buffer_size_kb 3670 # echo function > current_trace 3671 # echo 1 > instances/foo/events/sched/sched_wakeup/enable 3672 # echo 1 > instances/foo/events/sched/sched_wakeup_new/enable 3673 # echo 1 > instances/foo/events/sched/sched_switch/enable 3674 # echo 1 > instances/bar/events/irq/enable 3675 # echo 1 > instances/zoot/events/syscalls/enable 3676 # cat trace_pipe 3677 CPU:2 [LOST 11745 EVENTS] 3678 bash-2044 [002] .... 10594.481032: _raw_spin_lock_irqsave <-get_page_from_freelist 3679 bash-2044 [002] d... 10594.481032: add_preempt_count <-_raw_spin_lock_irqsave 3680 bash-2044 [002] d..1 10594.481032: __rmqueue <-get_page_from_freelist 3681 bash-2044 [002] d..1 10594.481033: _raw_spin_unlock <-get_page_from_freelist 3682 bash-2044 [002] d..1 10594.481033: sub_preempt_count <-_raw_spin_unlock 3683 bash-2044 [002] d... 10594.481033: get_pageblock_flags_group <-get_pageblock_migratetype 3684 bash-2044 [002] d... 10594.481034: __mod_zone_page_state <-get_page_from_freelist 3685 bash-2044 [002] d... 10594.481034: zone_statistics <-get_page_from_freelist 3686 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics 3687 bash-2044 [002] d... 10594.481034: __inc_zone_state <-zone_statistics 3688 bash-2044 [002] .... 10594.481035: arch_dup_task_struct <-copy_process 3689 [...] 3690 3691 # cat instances/foo/trace_pipe 3692 bash-1998 [000] d..4 136.676759: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000 3693 bash-1998 [000] dN.4 136.676760: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000 3694 <idle>-0 [003] d.h3 136.676906: sched_wakeup: comm=rcu_preempt pid=9 prio=120 success=1 target_cpu=003 3695 <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 3696 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 3697 bash-1998 [000] d..4 136.677014: sched_wakeup: comm=kworker/0:1 pid=59 prio=120 success=1 target_cpu=000 3698 bash-1998 [000] dN.4 136.677016: sched_wakeup: comm=bash pid=1998 prio=120 success=1 target_cpu=000 3699 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 3700 kworker/0:1-59 [000] d..4 136.677022: sched_wakeup: comm=sshd pid=1995 prio=120 success=1 target_cpu=001 3701 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 3702 [...] 3703 3704 # cat instances/bar/trace_pipe 3705 migration/1-14 [001] d.h3 138.732674: softirq_raise: vec=3 [action=NET_RX] 3706 <idle>-0 [001] dNh3 138.732725: softirq_raise: vec=3 [action=NET_RX] 3707 bash-1998 [000] d.h1 138.733101: softirq_raise: vec=1 [action=TIMER] 3708 bash-1998 [000] d.h1 138.733102: softirq_raise: vec=9 [action=RCU] 3709 bash-1998 [000] ..s2 138.733105: softirq_entry: vec=1 [action=TIMER] 3710 bash-1998 [000] ..s2 138.733106: softirq_exit: vec=1 [action=TIMER] 3711 bash-1998 [000] ..s2 138.733106: softirq_entry: vec=9 [action=RCU] 3712 bash-1998 [000] ..s2 138.733109: softirq_exit: vec=9 [action=RCU] 3713 sshd-1995 [001] d.h1 138.733278: irq_handler_entry: irq=21 name=uhci_hcd:usb4 3714 sshd-1995 [001] d.h1 138.733280: irq_handler_exit: irq=21 ret=unhandled 3715 sshd-1995 [001] d.h1 138.733281: irq_handler_entry: irq=21 name=eth0 3716 sshd-1995 [001] d.h1 138.733283: irq_handler_exit: irq=21 ret=handled 3717 [...] 3718 3719 # cat instances/zoot/trace 3720 # tracer: nop 3721 # 3722 # entries-in-buffer/entries-written: 18996/18996 #P:4 3723 # 3724 # _-----=> irqs-off 3725 # / _----=> need-resched 3726 # | / _---=> hardirq/softirq 3727 # || / _--=> preempt-depth 3728 # ||| / delay 3729 # TASK-PID CPU# |||| TIMESTAMP FUNCTION 3730 # | | | |||| | | 3731 bash-1998 [000] d... 140.733501: sys_write -> 0x2 3732 bash-1998 [000] d... 140.733504: sys_dup2(oldfd: a, newfd: 1) 3733 bash-1998 [000] d... 140.733506: sys_dup2 -> 0x1 3734 bash-1998 [000] d... 140.733508: sys_fcntl(fd: a, cmd: 1, arg: 0) 3735 bash-1998 [000] d... 140.733509: sys_fcntl -> 0x1 3736 bash-1998 [000] d... 140.733510: sys_close(fd: a) 3737 bash-1998 [000] d... 140.733510: sys_close -> 0x0 3738 bash-1998 [000] d... 140.733514: sys_rt_sigprocmask(how: 0, nset: 0, oset: 6e2768, sigsetsize: 8) 3739 bash-1998 [000] d... 140.733515: sys_rt_sigprocmask -> 0x0 3740 bash-1998 [000] d... 140.733516: sys_rt_sigaction(sig: 2, act: 7fff718846f0, oact: 7fff71884650, sigsetsize: 8) 3741 bash-1998 [000] d... 140.733516: sys_rt_sigaction -> 0x0 3742 3743You can see that the trace of the top most trace buffer shows only 3744the function tracing. The foo instance displays wakeups and task 3745switches. 3746 3747To remove the instances, simply delete their directories: 3748:: 3749 3750 # rmdir instances/foo 3751 # rmdir instances/bar 3752 # rmdir instances/zoot 3753 3754Note, if a process has a trace file open in one of the instance 3755directories, the rmdir will fail with EBUSY. 3756 3757 3758Stack trace 3759----------- 3760Since the kernel has a fixed sized stack, it is important not to 3761waste it in functions. A kernel developer must be conscious of 3762what they allocate on the stack. If they add too much, the system 3763can be in danger of a stack overflow, and corruption will occur, 3764usually leading to a system panic. 3765 3766There are some tools that check this, usually with interrupts 3767periodically checking usage. But if you can perform a check 3768at every function call that will become very useful. As ftrace provides 3769a function tracer, it makes it convenient to check the stack size 3770at every function call. This is enabled via the stack tracer. 3771 3772CONFIG_STACK_TRACER enables the ftrace stack tracing functionality. 3773To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled. 3774:: 3775 3776 # echo 1 > /proc/sys/kernel/stack_tracer_enabled 3777 3778You can also enable it from the kernel command line to trace 3779the stack size of the kernel during boot up, by adding "stacktrace" 3780to the kernel command line parameter. 3781 3782After running it for a few minutes, the output looks like: 3783:: 3784 3785 # cat stack_max_size 3786 2928 3787 3788 # cat stack_trace 3789 Depth Size Location (18 entries) 3790 ----- ---- -------- 3791 0) 2928 224 update_sd_lb_stats+0xbc/0x4ac 3792 1) 2704 160 find_busiest_group+0x31/0x1f1 3793 2) 2544 256 load_balance+0xd9/0x662 3794 3) 2288 80 idle_balance+0xbb/0x130 3795 4) 2208 128 __schedule+0x26e/0x5b9 3796 5) 2080 16 schedule+0x64/0x66 3797 6) 2064 128 schedule_timeout+0x34/0xe0 3798 7) 1936 112 wait_for_common+0x97/0xf1 3799 8) 1824 16 wait_for_completion+0x1d/0x1f 3800 9) 1808 128 flush_work+0xfe/0x119 3801 10) 1680 16 tty_flush_to_ldisc+0x1e/0x20 3802 11) 1664 48 input_available_p+0x1d/0x5c 3803 12) 1616 48 n_tty_poll+0x6d/0x134 3804 13) 1568 64 tty_poll+0x64/0x7f 3805 14) 1504 880 do_select+0x31e/0x511 3806 15) 624 400 core_sys_select+0x177/0x216 3807 16) 224 96 sys_select+0x91/0xb9 3808 17) 128 128 system_call_fastpath+0x16/0x1b 3809 3810Note, if -mfentry is being used by gcc, functions get traced before 3811they set up the stack frame. This means that leaf level functions 3812are not tested by the stack tracer when -mfentry is used. 3813 3814Currently, -mfentry is used by gcc 4.6.0 and above on x86 only. 3815 3816More 3817---- 3818More details can be found in the source code, in the `kernel/trace/*.c` files. 3819